Nondestructive Testing NDT PDF

Nondestructive
Testing (NDT)
Edited by
Giuseppe Lacidogna
Printed Edition of the Special Issue Published in Applied Sciences
www.mdpi.com/journal/applsci
Nondestructive Testing (NDT)
Nondestructive Testing (NDT)
Editor
Giuseppe Lacidogna
MDPI • Basel • Beijing • Wuhan • Barcelona • Belgrade • Manchester • Tokyo • Cluj • Tianjin
Editor
Giuseppe Lacidogna
Structural, Geotechnical and
Building Engineering
Politecnico di Torino
Torino
Italy
Editorial Office
MDPI
St. Alban-Anlage 66
4052 Basel, Switzerland
This is a reprint of articles from the Special Issue published online in the open access journal
Applied Sciences (ISSN 2076-3417) (available at: www.mdpi.com/journal/applsci/special issues/
nondestructive testing).
For citation purposes, cite each article independently as indicated on the article page online and as
indicated below:
LastName, A.A.; LastName, B.B.; LastName, C.C. Article Title. Journal Name Year, Volume Number,
Page Range.
ISBN 978-3-0365-2292-0 (Hbk)

ISBN 978-3-0365-2291-3 (PDF)
© 2021 by the authors. Articles in this book are Open Access and distributed under the Creative
Commons Attribution (CC BY) license, which allows users to download, copy and build upon
published articles, as long as the author and publisher are properly credited, which ensures maximum
dissemination and a wider impact of our publications.
The book as a whole is distributed by MDPI under the terms and conditions of the Creative Commons
license CC BY-NC-ND.
Contents
About the Editor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Preface to ”Nondestructive Testing (NDT)” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Marco Domaneschi, Gianni Niccolini, Giuseppe Lacidogna and Gian Paolo Cimellaro
Nondestructive Monitoring Techniques for Crack Detection and Localization in RC Elements
Reprinted from: Appl. Sci. 2020, 10, 3248, doi:10.3390/app10093248 . . . . . . . . . . . . . . . . . 1
Katsufumi Hashimoto, Tomoki Shiotani and Masayasu Ohtsu

Application of Impact-Echo Method to 3D SIBIE Procedure for Damage Detection in Concrete
Alessandro Grazzini, Sara Fasana, Marco Zerbinatti and Giuseppe Lacidogna

Non-Destructive Tests for Damage Evaluation of Stone Columns: The Case Study of Sacro
Monte in Ghiffa (Italy)
Elsa Garavaglia, Anna Anzani, Fabio Maroldi and Fabio Vanerio

Non-Invasive Identification of Vulnerability Elements in Existing Buildings and Their
Visualization in the BIM Model for Better Project Management: The Case Study of Cuccagna
Farmhouse
Zaifu Zhan, Shen Wang, Fuping Wang, Songling Huang, Wei Zhao and Zhe Wang
Simulation of Three Constitutive Behaviors Based on Nonlinear Ultrasound
Claudia Barile, Caterina Casavola, Vincenzo Moramarco, Carmine Pappalettere and

Paramsamy Kannan Vimalathithan
Bonding Characteristics of Single- and Joggled-Lap CFRP Specimens: Mechanical and Acoustic
Investigations
Huiting Huan, Lixian Liu, Andreas Mandelis, Cuiling Peng, Xiaolong Chen and Jinsong
Zhan
Mechanical Strength Evaluation of Elastic Materials by Multiphysical Nondestructive Methods:
A Review
Dimitrios G. Dimogianopoulos, Panagiotis J. Charitidis and Dionysios E. Mouzakis

Inducing Damage Diagnosis Capabilities in Carbon Fiber Reinforced Polymer Composites by
Magnetoelastic Sensor Integration via 3D Printing
Jihyun Jun, Young-Dae Shim and Kyung-Young Jhang

Stress Estimation Using the Acoustoelastic Effect of Surface Waves in Weak Anisotropic
Materials
Reprinted from: Appl. Sci. 2019, 10, 169, doi:10.3390/app10010169 . . . . . . . . . . . . . . . . . . 137
Michael Stamm, Helge Pfeiffer, Johan Reynaert and Martine Wevers

Using Acoustic Emission Measurements for Ice-Melting Detection
Reprinted from: Appl. Sci. 2019, 9, 5387, doi:10.3390/app9245387 . . . . . . . . . . . . . . . . . . . 153
v
Marco Civera, Matteo Ferraris, Rosario Ceravolo, Cecilia Surace and Raimondo Betti
The Teager-Kaiser Energy Cepstral Coefficients as an Effective Structural Health Monitoring
Tool
Shunmin Yang, Mingquan Wang and Lu Yang

The Rapid Detection Technology of Lamb Wave for Microcracks in Thin-Walled Tubes
Shuai Han, Heng Li, Mingchao Li and Timothy Rose

A Deep Learning Based Method for the Non-Destructive Measuring of Rock Strength through
Hammering Sound
Rongning Cao, Meng Ma, Ruihua Liang and Chao Niu

Detecting the Void behind the Tunnel Lining by Impact-Echo Methods with Different Signal
Analysis Approaches
Dongdong Wen, Mengbao Fan, Binghua Cao, Zhian Xue and Ping Wang
A PEC Thrice Subtraction Method for Obtaining Permeability Invariance Feature in
Conductivity Measurement of Ferromagnetic Samples
vi
About the Editor
Giuseppe Lacidogna
Giuseppe Lacidogna achieved the National Academic Qualification in Italy as Full Professor of
Structural Mechanics in 2018. He is Fellow of the European Academy of Sciences, Brussels, Belgium
and Officer of the Engineering Division of the same Academy. He has been Associate Professor in
Structural Mechanics at Politecnico di Torino, Italy, since January 2011.
Professor Lacidogna is Member of the Academic Board of the Doctorate Course in Civil and
Environmental Engineering of Politecnico di Torino, and has been Director of the Doctorate Course
in Structural Engineering (2016-2018).
He is author of more than 300 publications, including: eight monographs, more than 120 papers
in refereed international journals, and 24 book chapters. He received a Certificate Merit Award from
the European Society for Experimental Mechanics (EuraSEM) and has been inserted in the Italian
Scientists List, Engineering-Mechanics Area, Macro Area Engineering.
vii
Preface to ”Nondestructive Testing (NDT)”
Because of the unavoidable aging and deterioration of materials, the efficient maintenance of
civil structures and infrastructures is becoming a major issue for public and private institutions. As
a matter of fact, the structural conditions of built heritage affect both safety and economic aspects.
In this context, structural health monitoring (SHM) is emerging as a crucial research field, able to
provide vital information regarding the damage levels of structures and materials. In particular, by
exploiting the most advanced technologies and techniques, nondestructive testing (NDT) is the ideal
candidate for detecting defects and structural issues, both at the laboratory and full-scale levels, in a
non-invasive way. Among others, acoustic emission, vibration-based identification methods, digital
image correlation, tomography techniques, sonic-ultrasonic tests, Raman and terahertz spectroscopy,
electromagnetic analysis, etc., allow us to evaluate the state of damage and its evolution over
time. The aim of this book is to collect the newest contributions by eminent authors in the field
of NDT-SHM, both at the material and structure scale. It therefore provides novel insight at
experimental and numerical levels on the application of NDT to a wide variety of materials (concrete,
steel, masonry, composites, etc.) in the field of Civil Engineering and Architecture.
Giuseppe Lacidogna
Editor
ix
applied
sciences
Article
Nondestructive Monitoring Techniques for Crack
Detection and Localization in RC Elements
Marco Domaneschi , Gianni Niccolini , Giuseppe Lacidogna * and Gian Paolo Cimellaro
Department of Structural, Geotechnical and Building Engineering, Politecnico di Torino, 24,
Corso Duca degli Abruzzi, 10129 Torino, Italy; [email protected] (M.D.);
[email protected] (G.N.); [email protected] (G.P.C.)
* Correspondence: [email protected]; Tel.: +39-011-090-4871

Received: 28 March 2020; Accepted: 2 May 2020; Published: 7 May 2020
Featured Application: Damage assessment of a reinforced concrete (RC) by means of different

nondestructive testing (NDT) techniques. Joined application of a PZT sensor network for
AE-based local damage detection and a FOS setup for global strain mapping. Coupling of local and
global methods through DIC-based strain localization to reduce the uncertainty on the strain field
description. Potential for AE- and strain-based damage detection in large structures of interest to
civil engineering.
Abstract: This paper presents the structural and damage assessment of a reinforced concrete (RC)
beam subjected to a four-point bending test until yielding of reinforcing steel. The deterioration
progress was monitored using different nondestructive testing (NDT) techniques. The strain was
measured by distributed fiber optic sensors (FOSs), embedded prior to concrete pouring. The initiation
and propagation of cracks were monitored by acoustic emission (AE) sensors attached to the surface
of the material. The recorded AE activity results in good agreement with FOS strain measurements.
The results of the integrated monitoring system are confirmed by visual observation of the actual
crack pattern. At different loading steps, digital image correlation (DIC) analysis was also conducted.
Keywords: reinforced concrete; four-point bending test; structural health monitoring; nondestructive
testing techniques; fiber optic sensors; acoustic emission monitoring; digital image correlation
1. Introduction
During their service life structural elements can experience variable and increasing demand
conditions (e.g., traffic increase in transportation infrastructures). They are also exposed to environmental
effects, that can also degrade their capacity and affect their performance. In particular, reinforced
concrete (RC) components can develop cracking due to tensile conditions that are normally absorbed
by steel reinforcements. However, cracking can lead to the exposure of the steel bars to the aggression
of external agents, such as chlorides, triggering corrosion and strength losses. In this incremental risk
scenario for existing infrastructures, the role of the structural health monitoring (SHM) becomes crucial
in order to detect unusual behaviors and damage [1–3].
The use of fiber optic sensors (FOSs) for SHM is proposed in several research works, including
the seminal one that dates to 1990 by Glossop et al. [4]. FOSs can be used in aggressive environments,
they show geometric adaptability, independence from electrical and magnetic field interference,
and high resolution [5]. These characteristics make them excellent for implementation in the civil
engineering field to detect anomalies and cracks in static conditions, while, in dynamic conditions,
they can also be used to assess modal parameters [6–8].
1
Appl. Sci. 2020, 10, 3248
Acoustic emission (AE) sensors have been widely used to detect early stage damage before it
results in failure. One of the earliest applications of AE was related to monitoring rotating machinery
in the late 1960s. A comprehensive and critical review on the field of application of AE to condition
monitoring and diagnostics of different mechanical components can be found in Mba and Rao (2006) [9].
Behnia et al. [10] present a comprehensive review of the acoustic emission (AE) technique for its
applications in concrete structure health monitoring. Methods of AE are also developed for large
structures in field application, e.g., for walls, bridge decks and reinforced concrete multi-story buildings
(Carpinteri et al., 2007, 2011; McLaskey et al., 2010; Shiotani et al., 2009) [11–14].
Although AE sensors are normally developed using piezoelectric technology, there are many
examples where optical fibers have been introduced as an alternative to piezoelectric sensors
(e.g., Liang et al., 2009 [15]). However, there are a few applications where AE sensors are implemented
in addition to distributed FOS sensors for strain detection, in order to integrate the monitoring and
damage detection results obtained separately from the two systems. A representative example is
one by Li et al. [16], where the results of an experimental investigation on corrosion monitoring of
a steel reinforced mortar block through combined acoustic emission and fiber Bragg grating strain
measurement are presented. Ansari [17] provided a short review of long gage interferometric and
acoustic sensors with representative examples on the implementation of serially multiplexed long
gage interferometric sensors and multiplexing of optical fiber acoustic emission sensors. Moving to
composite fiber/epoxy materials, Park et al. [18] present an application for micro-failure evaluation
using embedded fiber-optic sensors and acoustic emission piezoelectric sensors.
Detection of crack-induced AEs by a piezoelectric (PZT) sensor network focuses on individual
cracks and displacements providing a very detailed damage description, but it suffers from limited
detection range due to high attenuation of ultrasonic waves. Vice versa, global sensors act as a fully
distributed sensor network to be interrogated at any point along the fiber length or integrating the
response along the FOS length so as to cover a larger area. The disadvantage of global damage detection
systems is their poor damage location capability. Therefore, the concurrent application of local and
global methods combines damage detection possibilities and advantages of both systems.
Verstrynge et al. [19] applied a global FOS setup for AE-based damage detection in concrete
elements. This line-integrating technique did not allow AE source location—carried out in that
experiment through concurrent application of PZT transducers—but it has the potential for AE
detection in large structures of interest to civil engineering.
Here, AE-based local damage detection through a PZT sensor network is coupled with a FOS
setup for global strain mapping. FOSs act as a distributed sensor network to be interrogated at different
points along the fiber length. Local and global methods are herein coupled through a digital image
correlation (DIC)-based strain localization covering an intermediate-sized area. Hence, DIC-based
strain data is herein used to reduce uncertainty on the FOS-based strain field description.
The integrated SHM system has been implemented in an RC beam specimen subjected to a
four-points loading test to monitor the state of cracking in terms of initiation and the thickness
propagation. The outcomes of the integrated AE and FOS monitoring systems are confirmed by
comparison with the final crack pattern. At different loading steps, digital image correlation (DIC)
analysis was also conducted in specific areas of the beam specimen to reconstruct the strain field.
A numerical model has been implemented to investigate the mechanical behavior of the beam up to
collapse and to confirm the monitoring outcomes.
The following sections describe the methodology and the SHM systems; then, the laboratory
specimen and setup are discussed. Finally, the results and the conclusive remarks are presented.
2. Experimental Setup and Monitoring Systems

The laboratory test was conducted on a reinforced concrete beam under gradually increasing
loading in the four-point bending configuration, as shown in Figure 1. The reinforced concrete beam is
4 m long and was designed to have a ductile behavior to be able to follow a progressive and gradual
2
Appl. Sci. 2020, 10, 3248
propagation of cracks. The cross section of the beam is 150 mm wide and 300 mm high. The longitudinal
reinforcements are composed of two bars of 14 mm in diameter at the bottom and two bars of 18 mm in
diameter at the top of the section, with a concrete cover of 30 mm. Stirrups of 8 mm in diameter as shear
reinforcement have been installed with 70 mm spacing at the bearings, while in the remaining beam
shear reinforcements are spaced every 140 mm. According to the Italian ministerial decree, Updating
of the Technical standards for construction (Gazzetta Ufficiale, D.M. 17 Gennaio 2018), C20/25 and
B450C have been selected for the concrete and steel materials. The maximum aggregate size has been
fixed to 10 mm with a water–cement weight ratio of about 0.5 and 300 kg/m3 of cement.
The load frame and the front side of the beam specimen are shown in Figure 2 (top), with a servo
hydraulic actuator fixed at the center of the frame. A stepwise increasing load was applied downwards
in displacement control by means of steel roller bearings in a four-point bending configuration.
The load was measured with a 1000 kN load cell at the bottom of the actuator, while the mid span
vertical displacement was measured through a linear variable differential transformer (LVDT) device,
compensated with two LVDTs measurement points at the beam supports. Figure 2 (bottom) shows
the beam specimen at the back side with the speckle pattern applied to a central surface of about
0.8 m length.
Figure 1. Specimen geometry and acoustic emission (AE) sensor locations. The eight AE sensors are
numbered from 0 to 7. Dimensions are in cm.
3
Appl. Sci. 2020, 10, 3248
Figure 2. Front (top) and back (bottom) sides of the test setup showing the servo hydraulic actuator,
the linear variable differential transformer (LVDT) sensors, the location of some AE sensors, and the
speckle pattern on the central beam surface for digital image correlation (DIC) applications.
3. AE Sensors
Different damage mechanisms and levels can be verified during the four-point bending test of the
RC beam, such as tensile cracking followed by shear cracking, yielding of reinforced steel, and crushing.
The fracture propagation mainly includes two stages: accumulation of uncorrelated tensile cracks
between the central loads, and appearance of diagonal shear cracks, starting on the bottom surface
and developing to the top, interconnecting the previously formed tensile cracks. Statistical analysis of
the AE signal features can reveal trends that can be ascribed to different damage stages. For example,
4
Appl. Sci. 2020, 10, 3248
shear events originate signals with longer rise time (the time delay between onset and peak amplitude)
and higher amplitude.
Eight 60 kHz resonant AE sensors were attached at one-fourth of the span’s length (refer to
Figures 1 and 2 respectively for beam design and application). When the sensors are hit by a wave
(due to crack growth or another reason), the pressure on their surface is converted to electric voltage,
exploiting the piezoelectric effect. If the voltage was beyond a given threshold (60 dB to filter out the
signal noise in the specific case), the signal was digitized and stored in an 8 channel Lunitek AEmission
system with 5 MHz sampling rate per channel. The software associated to the acquisition board permits
automatic extraction and storage of some basic parameters for each individual AE signal waveform
(feature-based AE analysis), such as amplitude, duration and ring-down counts (RDC), i.e., the total
number of times the signal exceeds the threshold level. RDC divided by duration gives the average
signal frequency.
4. FOS Sensors
Different FOS distributed sensors have been embedded in the concrete beam to collect strain and
temperature measurements. In particular, strain sensors have been applied to the steel reinforcements
at
ε the top (FOS #1) and the bottom (FOS #2) of the cross section along the whole beam length (Figure 3).
When the optical fiber is strained in the longitudinal direction, the backscattered light of Brillouin
is subjected to a frequency shift proportional υ ε −υ
= the strain. Equation (1) expresses the strain ε as function
ε to
of the Brillouin frequency shift:
υB (ε) − υB (0)
υ ε ε=
c
,υ (1)
where υB (ε) is the Brillouin frequency with strain, while υB (0) is the Brillouin frequency without
strain. The constant c is the proportional coefficient of strain that depends on the sensor characteristics.
Basic principles of Brillouin systems are detailed in Bao et al. [20].
Although sensitive along the entire length, the fiber optic sensors measure at discrete points,
that are spaced by constant value called the sampling interval. The measured parameter is an average
over a certain length called spatial resolution and corresponds to the pulse width used to interrogate
the fiber. For the present research, a SHM FOS system usually employed in full scale real structures
has been used, thus the sampling interval and the spatial resolution have been assumed to be 0.2 and
0.5 m, respectively.
2.2
Figure 3. Fiber optic sensor (FOS) installation (dimensions in cm).
5. DIC Technique
DIC photogrammetry is a non-contact, optical measuring method adapted for extracting surface
displacement and geometry profiles at different stages or times from images acquired through a
5
Appl. Sci. 2020, 10, 3248
camera (or multi-camera). It has been proposed for structural inspection and monitoring, and has been
effectively applied previously to analyze different structures at different scales [21].
To perform a DIC measurement, a stochastic pattern (e.g., black dots on a white background or
white dots on a black background), must be applied to the surface of interest and the relative position
of each of them is identified as the surface deforms over time. Each image can be considered as a
matrix of natural integers where white pixels have a 0 grayscale level, while level 100 is associated to
black pixels. Since a single value is not a unique point, a neighborhood of pixels is used (i.e., facets or
subsets). These facets include several dots of the pattern used and are typically squares with sides of
10–50 pixels. The main principle of DIC is to match the same physical point between a reference image
and several deformed images based on gray-scale variations of continuous patterns [22].
The DIC computation is herein performed through NCORR software [23], which is an open-source
subset-based 2D DIC package that uses modern DIC algorithms with additional improvements.
6. Test Results
The global behavior in terms of load and mid-span vertical displacement is illustrated in Figure 4.
Vertical load has been applied in multiple steps up to the yielding limit of reinforcing bars, and then
removed. The first cracking point can be identified between steps #1 and #2. Then, the ductile beam
behavior allowed crack propagation up to step #7 when the longitudinal reinforcing steel bars at
the bottom of the cross section start yielding. The loading phase of the four-points bending test was
stopped at step #8 and loads removed from the beam (unloading phase).
Figure 4. Load versus mid-span deflection at the different load steps, from #0 to #8.
7. AE Measurements
7.1. Total AE Activity

The applied load history along with the accumulated AE events are shown in Figure 5. The number
of AE events started to increase rapidly during 4th and 7th load steps at approximately 2100 and 4500 s,
meaning intensive cracking phenomena occurred at the time of increased central deflection. The major
cluster of events comes at the unloading stage, from approximately 6900 s, where a possible AE source
mechanism is the friction in aggregates and reinforcing bars due to cracks reclosing.
The AE activity is first investigated considering the individual sensor recordings and the related
statistical distributions of the signal parameters, i.e., peak amplitude, duration and average frequency.
Figure 6a shows the distributions of the received signals by sensors AE0 and AE1, and Figure 6b
shows those received by sensors AE4 and AE5. Considering the mean values of the distributions,
6
Appl. Sci. 2020, 10, 3248
recordings of AE0 and AE1 (located on the top surface of the beam) are characterized Δt by lower
frequency, lower amplitude and longer signal duration than AE4 and AE5 (located next to the bottom
surface), namely <f > = 17 kHz vs. 31 kHz, <A> = 3.3 mV vs. 4.5 mV, and <∆t> = 0.8 ms vs. 0.4 ms.
These findings can be explained in terms of amplitude loss (signal attenuation)—stronger for high
frequencies—and, possibly, of spreading of AE wave packets traveling in a dispersive medium [24],
such as the concrete to some extent. Hence, the measured effects appear to be descriptive of AE activity
predominantly originating near AE4 and AE5 sensors, consistent with a crack pattern developing from
the bottom surface of the beam.
Stages 1-3 Stages 4-6 Stages 7-8 Δt
AE
Load
Figure 5. Step load (blue line) and cumulated AE signals time history (red line). Dashed lines correspond
to the 3rd, 6th and 8th load stages.
7.2. The b-Value Analysis

It is reasonable that as the load increases, larger-scale fracture events occur giving rise to AE
signals of larger amplitude. The signal amplitudes are converted in a magnitude scale by using the
equation M = log10 (V⁄V 0 ), where V is the peak of voltage signal expressed in microvolts, and V 0 = 1 µV
is the reference voltage (conversion to decibels is possible through M = AdB ⁄20). The signal magnitudes
are studied by their cumulative distribution, using the b-value analysis. For a population of N events,
the b-value is calculated as:
log10 N (M) = a − bM, (2)
where N(M) is the number of signals with magnitude higher than M, and a and b are fitting constants.
Figure 6. Cont.
7
Appl. Sci. 2020, 10, 3248
(a)
(b)
Figure 6. (a) AE features relationships and distributions of received signals by sensors AE0 and AE1.
The mean values of distributions represented by histograms from left to right are: <A> = 3.3 mV,
Δt
<f > = 17 kHz, and <∆t> = 0.8 ms. (b) AE features relationships and distributions of received signals by
sensors AE4 and AE5. The mean values of distributions represented by histograms from left to right
are: <A> = 4.5 mV, <f > = 31 kHz, and <∆t>Δt = 0.4 ms.
The b-value expresses the absolute gradient of the cumulative magnitude distribution as illustrated
in Figure 7. As the fracture process becomes more intense, the percentage of high-amplitude events
increases relative to the low-amplitude ones in the total population, resulting in drops of the b-value [25].
⁄
⁄
= − ,
8
Appl. Sci. 2020, 10, 3248
Figure 7. AE magnitude distribution for different damage conditions: low damage (high b-value) and
high damage (low b-value).
Taking into account the number N of recent events to calculate every single b-value, possible
decreasing trends can be identified and associated with damage severity and change in structural
performance state, especially during laboratory loading tests where a single minimum is observed just
prior to specimen failure. The recommended value number of events for computation (see e.g., [24])
is typically N = 100. Smaller values would yield too strong oscillations, masking the actual trend.
Instead, using too large N, any small occurrence producing a small number of AE signals would be
undermined in the average of the large population. Here, four partitioning criteria are followed to
calculate the evolving b-values during the test, namely using disjoint subsets of N = 100 and 150 events
(Figure 8 (left)), or overlapping subsets of N = 100 events with step 25 and a loading stage-based
partition (Figure 8 (right)) [25,26]. The most important parameters to be considered in a b-value graph
are the general trend and the minima of the curve. Comparing all the time series, nothing more than an
oscillating trend can be identified, with b-values mainly ranging between 1.2 and 1.6 (with an isolated
minimum at 1.1 reported in two cases) as descriptive of minor to moderate damage levels reached
during the test. As herein illustrated—and widely recognized—the b-value, although being sensitive
to damage, does not necessarily decrease monotonically. That is due to the presence of both healthy
material encountered during crack propagation and steel reinforcements that limit crack growth in
concrete matrix. Namely, fracture propagates during steel deformation or concrete sliding with respect
to steel. Fracture generates signals related to early damage, which are mixed with those of the already
heavily damaged part, as stated in [27–30].
Figure 8. b-value time histories with dashed lines referring to the load stages of Figure 5. Using disjoint
subsets of N = 100 (red dots) and N = 150 signals (cyan dots) (left). Ib-value with overlapping subsets
of N = 100 events with step 25 (green dots) and b-value by a stage-based partition (blue dots) (right).
9
Appl. Sci. 2020, 10, 3248
7.3. AE Source Location

Due to the configuration of the experiment, the tensile surface is identified at the bottom of the
beam cross sections. Therefore, crack initiation reasonably occurs close to the bottom surface. This is
revealed by Figure 9, where the height of localized AE sources is depicted for different loading steps.
−
The fitting line identifies the average height of localized events, exhibiting a slightly ascending trend.
At the lowest load level, the average location is close to the bottom surface (−0.06 m). As the cracks
propagate towards the top, the AE sources naturally move to higher y-locations, reaching 0.1 m.
Figure 9. Height of localized AE events for different loading stages. Adopted reference system (x, y, z)
as in Figure 1. Green color refers to load stages #1–#4, blue to #5 and #6, yellow to #7 and #8, and red
to #9. –
The determined locations of AE events concerning different loading stages are depicted together
with sensor positions in Figure 10. Location data are only available in 2D, along the beam’s lateral axis
due to the almost 2D sensor setup, and therefore, information regarding depth positionbeam’s la
of AE sources
cannot be provided. Typically, AE source location is carried out along lines between detecting sensors,
i.e., AE0, AE1, AE4 and AE5. Hence, location data do not exactly match the crack initiation from the
bottom surface of the beam. One can see localized AE sources starting from the left and accumulating
vertically towards the top surface, as the signature of coalescence of uncorrelated tensile cracks and the
appearance of diagonal shear cracks. Red circles, representing localized events at the unloading stage,
are probably due to friction between aggregates and reinforcing bars during reclosing of the cracks.
0.15
0.05
y(m)
-1.6 -1.4 -1.2 -1

-0.05
-0.15
x(m)
Figure 10. Location of AE sources and AE sensors (numbered gray squares). Adopted reference system
(x, y, z) as in Figure 1 and color map as in Figure 9.
8. FOS Strain Measurements

In line with the configuration of the experiment, the tensile state at the bottom of the beam cross
sections is revealed by Figure 11, which depicts the history of the internal positive strain measured
10
Appl. Sci. 2020, 10, 3248
by the embedded FOS #2 running about 30 mm from the bottom of the external surface. Vice versa,
the expected negative strain values in the compressed zone are measured by the FOS #1, running 30 mm
from the top external surface (Figure 3).
Figure 11. FOS strain measurements at different loading stages along the beam length: FOS #1 (up),
FOS #2 (down).
8.1. AE vs. FOS Measurements

Figure 12a,b compares the strain history (depicted by a blue line) measured close to the bottom
− −
surface by FOS #2, respectively at the horizontal locations of −1600 and −1200 mm, with the accumulated
AE activity (red line) recorded by the nearest AE sensor, respectively AE4 and AE5.
1-3 4-6 7-8 1-3 4-6 7-8
(a) (b)
Figure 12. Cumulative AE activity (red) and FOS strain (blue) time histories for: (a) fiber optic n.9 at
x = −1.6 m and AE sensor n.4; (b) fiber optic n.9 at x = −1.2 m and AE sensor n.5. The dashed lines
− −
identify the different loading stages.
−
−
11
− −
Appl. Sci. 2020, 10, 3248
−
Figure 13a,b compares the strain history measured close to the top surface by FOS #1, at x = −1600
−
and −1200 mm, with the accumulated AE activity recorded by the nearest AE sensors, namely AE0
and AE1. Both Figures 12 and 13 illustrate noticeable similarity between AE and FOS graphs.
1-3 4-6 7-8 1-3 4-6 7-8
(a) (b)
Figure 13. Cumulative AE activity and FOS strain time histories for: (a) fiber optic n.7 at x = −1.6 m
−
and AE sensor n.0; (b) fiber optic n.7 at x = −1.2 m and AE sensor n.1. The dashed lines identify the
−
different loading stages.
8.2. Cracks Pattern

The results of FOS and AE monitoring are compared with the actual pattern of cracks directly
observable in the laboratory. Figure 14 shows the cracks at loading step #7 where the crack development
at the bottom side of the beam can be noticed. Furthermore, the transition of crack development from
the vertical shape (due to pure bending moment at the center of the beam) to the inclined shape at the
extremity (due to shear effects at the support) is highlighted.
Figure 14. Crack pattern at loading step #7 and numbered AE sensors.
12
Appl. Sci. 2020, 10, 3248
9. DIC Strain Assessment

A comparison between DIC results in terms of the strain field on the central span of the beam
specimen and the FOS #2 measurements has also been presented in this section. Figure 15 (top)
presents the strain field computed through DIC [22] at loading steps #4 (left) and #8 (right), while the
FOS strains at the bottom side of the specimen are reported for the same loading steps at Figure 15
(bottom). Focusing on the 0.8 m central length of interest, the FOS #2 sensor highlights a reasonable
constant tensile strain of about 1700 µm/m at step #4 and 7000 µm/m at step #8. These results have
been computed considering a sampling interval and spatial resolution of 0.2 and 0.5 m, respectively,
typically used for large real structures.
The DIC strain field at the bottom side results in the range 1000–3500 µm/m at step #4,
and 4000–12,000 µm/m at step #8. Therefore, a satisfactory comparison between the two adopted
techniques can be verified.
Figure 15. DIC (top) and FOS #2 (bottom) outcomes at loading steps #4 (top-left) and #8 (top-right) in
the central area of the beam specimen (between the vertical loads—see Figure 2).
10. Numerical Analysis

A numerical model of the beam has been prepared through the applied element method (AEM) to
perform nonlinear static analysis and reproduce the laboratory observations up to collapse. The AEM
is a rather new method, akin to the discrete element method, which is capable of predicting with a
reasonable degree of accuracy the continuum and with a high degree of accuracy the discrete behavior
of structures during the collapse. The AEM has been proved to track the structural collapse behavior
passing through all the application load stages. Within the AEM, the structure is modeled as an assembly
of 8-points hexahedral elements. Therefore, each element is assumed rigid (6 degrees-of-freedom)
and has a 3-D physical solid shape. Two adjacent elements are assumed to be connected by one
normal and two shear springs distributed around the elements’ edges on the interface. Each group
of springs represents the entire stresses and deformations of a certain volume. More details on the
theoretical aspects related to AEM and its comparison with finite element method can be found in
Grunwald et al. [31].
13
Appl. Sci. 2020, 10, 3248
The four-points bending test on the reinforced concrete beam has been reproduced through
nonlinear static analysis in AEM. The results have been used to anticipate the structural behavior and
to compare with the measurements [32]. Different AEM models have been developed by considering
a sensitivity analysis to identify a reasonable discretization for the beam. Accordingly, materials
have been assumed to be C20/25 and B450C for concrete and steel respectively, with nonlinear
constitutive laws following the models by Menegotto and Pinto for steel and Maekawa and Okamura
for concrete [33]. A 5 cm side-length element mesh has been selected for the comparison with SHM
systems as the reasonable compromise between computational efforts and accuracy.
Considering the global behavior (Figure 16a), the AEM model performs satisfactorily by reasonably
anticipating the force–displacement response. Moving to the local behavior, the selected mesh results
also are able to correctly trace the bending and shear cracks. Figure 16b, in particular, reports the
development of a crack at step #8. The two regions where the AE and DIC techniques have been
applied can be observed. The strain concentration due to crack opening, as detected by the DIC
technique (Figure 15, top-right) in the central region and by the AE sensors at the extremity, is correctly
reproduced by the AEM simulation. It is worth underlining how the screen effect of voids, generated
by micro-fractures and distributed damage in the concrete matrix, does not allow the AE sensors to
detect the cracks generated at the mid-span of the beam.
(a) (b)
Figure 16. (a) Force–displacement response up to collapse, (b) development of cracks at step #8.
11. Conclusions
The implementation of integrated SHM schemes of nondestructive testing techniques is presented
in this paper for a reinforced concrete beam tested in laboratory through a four-points bending
arrangement. The beam is subjected to loading stages to induce characteristic concrete cracking and
steel reinforcements yielding. All of these phases have been identified in a global behavior curve in
terms of force–displacement measurements.
The strain measures of the embedded fiber optic system have been compared to the acoustic
emission sensors’ outcomes. Furthermore, visual observation of the actual cracks pattern has been also
considered in parallel with DIC analysis in the central region of the beam specimen.
The recorded AE activity resulted in good agreement with FOS strain measurements. Furthermore,
the results of the AE–FOS integrated SHM system are confirmed by the actual cracks pattern.
Focusing on the pure bending moment region of the specimen, the strain field computed by DIC also
resulted in good agreement with the FOS tensile strains at the bottom side of the specimen.
A numerical model has also been prepared to investigate the mechanical behavior of the beam up
to collapse. The analysis has been focused on both the global behavior in terms of force–displacement
response, and also the local characterization of cracks. A satisfactory match between the numerical
outcomes and the monitoring data has been observed.
14
Appl. Sci. 2020, 10, 3248
Therefore, the integrated use of different nondestructive testing techniques as proposed in this
paper is a reasonable and reliable SHM strategy for damage detection and localization in RC elements.
This study confirms how the application of different nondestructive testing techniques can be useful for
specific civil engineering applications, e.g., for large-sized structural elements where visual inspection is
not always possible and accurate. Thus, the simultaneous use of nondestructive testing techniques such
as acoustic emission, embedded fiber optic sensors, and digital image correlation, possibly aided by
visual inspection, seems to be a step toward the realization of a reliable real-time structural alert system.
Author Contributions: Conceptualization, M.D., G.N., G.L., G.P.C.; investigation and data curation, M.D., G.N.;
writing—review and editing, M.D., G.N., G.L.; supervision and project administration, G.P.C. All authors have
read and agreed to the published version of the manuscript.
Funding: The research is partially supported by the European Research Council under the Grant Agreement
n◦ ERC_IDEalreSCUE_637842 of the project IDEAL RESCUE-Integrated Design and control of Sustainable
Communities during Emergencies.
Acknowledgments: Maurizio Morgese assisted the laboratory analyses in partial fulfillment for the requirements
of the Bachelor’s Degree in Civil Engineering at Politecnico di Torino, under the guidance of the Authors.
His contribution is gratefully acknowledged. The technical support of SMARTEC (CH) and MastrLab at Politecnico
di Torino-DISEG is gratefully acknowledged.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. Farrar, C.R.; Czarnecki, J.J.; Sohn, H.; Hemez, F.M. A Review of Structural Health Monitoring Literature 1996–2001;
Report Number: LA-13976-MS; Los Alamos National Laboratory: Los Alamos, NM, USA, 2002; pp. 1–301.
2. Farrar, C.R.; Worden, K. An introduction to structural health monitoring. Philos. Trans. R. Soc. A 2006,
365, 303–317. [CrossRef]
3. Mechbal, N.; Uribe, J.S.; Rébillat, M. A probabilistic multi-class classifier for structural health monitoring.
Mech. Syst. Signal 2015, 60–61, 106–123. [CrossRef]
4. Glossop, N.D.W.; Dubois, S.; Tsaw, W.; Leblanc, M.; Lymer, J.; Measures, R.M.; Tennyson, R.C. Optical fibre
damage detection for an aircraft composite leading edge. Composites 1990, 21, 71–80. [CrossRef]
5. Glišić, B.; Inaudi, D. Fibre Optic Methods for Structural Health Monitoring; Wiley Online Library: Hoboken,
NJ, USA, 2007; pp. 1–262.
6. Ansari, F. Practical implementation of optical fiber sensors in civil structural health monitoring. J. Intell.
Mater. Syst. Struct. 2007, 18, 879–889. [CrossRef]
7. Casciati, S.; Domaneschi, M.; Inaudi, D. Damage assessment from SOFO dynamic measurements.
In Proceedings of the SPIE 17th International Conference on Optical Fibre Sensors, Bruges, Belgium,
23–27 May 2005; Volume 5855, pp. 1048–1051.
8. Domaneschi, M.; Sigurdardottir, D.; Glišić, B. Damage detection based on output-only monitoring of dynamic
curvature in concrete-steel composite bridge decks. Struct. Monit. Maint. Int. J. 2017, 4, 1–15.
9. Mba, D.; Rao, R.B.K.N. Development of acoustic emission technology for condition monitoring and diagnosis
of rotating machines: Bearings, pumps, gearboxes, engines, and rotating structures. Shock Vib. Dig. 2006,
38, 3–16. [CrossRef]
10. Behnia, A.; Chai, H.K.; Shiotani, T. Advanced structural health monitoring of concrete structures with the aid
of acoustic emission. Constr. Build. Mater. 2014, 65, 282–302. [CrossRef]
11. Carpinteri, A.; Lacidogna, G.; Pugno, N. Structural damage diagnosis and life-time assessment by acoustic
emission monitoring. Eng. Fract. Mech. 2007, 74, 273–289. [CrossRef]
12. McLaskey, G.C.; Glaser, S.D.; Grosse, C.U. Beamforming array techniques for acoustic emission monitoring
of large concrete structures. J. Sound Vib. 2010, 329, 2384–2394. [CrossRef]
13. Shiotani, T.; Aggelis, D.G.; Makishima, O. Global monitoring of large concrete structures using acoustic
emission and ultrasonic techniques: Case study. J. Bridge Eng. 2009, 14, 188–192. [CrossRef]
14. Carpinteri, A.; Lacidogna, G.; Niccolini, G. Damage analysis of reinforced concrete buildings by the acoustic
emission technique. Struct. Control Health 2011, 18, 660–673. [CrossRef]
15. Liang, S.; Zhang, C.; Lin, W.; Li, L.; Li, C.; Feng, X.; Lin, B. Fiber-optic intrinsic distributed acoustic emission
sensor for large structure health monitoring. Opt. Lett. 2009, 34, 1858–1860. [CrossRef] [PubMed]
15
Appl. Sci. 2020, 10, 3248
16. Li, W.; Xu, C.; Ho, S.C.; Wang, B.; Song, G. Monitoring concrete deterioration due to reinforcement corrosion
by integrating acoustic emission and FBG strain measurements. Sensors 2017, 17, 657. [CrossRef]
17. Ansari, A. Fiber optic health monitoring of civil structures using long gage and acoustic sensors. Smart Mater.
Struct. 2005, 14, S1–S7. [CrossRef]
18. Park, J.M.; Lee, S.I.; Kwon, O.Y.; Choi, H.S.; Lee, J.H. Comparison of nondestructive micro-failure evaluation
of fiber-optic Bragg grating and acoustic emission piezoelectric sensors using fragmentation test. Compos. Part
A Appl. Sci. Manuf. 2003, 34, 203–216. [CrossRef]
19. Verstrynge, E.; Pfeiffer, H.; Wevers, M. A novel technique for acoustic emission monitoring in civil structures
with global fiber optic sensors. Smart Mater. Struct. 2014, 23, 065022. [CrossRef]
20. Bao, X.; Chen, L. Recent progress in Brillouin scattering based fiber sensors. Sensors 2011, 11, 4152–4187.
[CrossRef]
21. Nonis, C.; Niezrecki, C.; Yu, T.Y.; Ahmed, S.; Su, C.F.; Schmidt, T. Structural health monitoring of bridges
using digital image correlation. In Proceedings of the SPIE Smart Structures and Materials + Nondestructive
Evaluation and Health Monitoring, San Diego, CA, USA, 17 April 2013; Volume 869507, pp. 1–13.
22. Chu, T.; Ranson, W.; Sutton, M.A. Applications of digital-image-correlation techniques to experimental
mechanics. Exp. Mech. 1985, 25, 232–244. [CrossRef]
23. Blaber, J.; Adair, B.; Antoniou, A. Ncorr: Open-Source 2D Digital Image Correlation Matlab Software.
Exp. Mech. 2015, 55, 1105–1122. [CrossRef]
24. Lympertos, E.M.; Dermatas, E.S. Acoustic emission source location in dispersive media. Signal Process. 2007,
87, 3218–3225. [CrossRef]
25. Shiotani, T.; Yuyama, S.; Li, Z.W.; Ohtsu, M. Application of AE improved b-value to quantitative evaluation
of fracture process in concrete materials. J. Acoust. Emiss. 2001, 19, 118–132.
26. Rao, M.V.M.S.; Prasanna-Lakshmi, K.J. Analysis of b-value and improved b-value of acoustic emissions
accompanying rock fracture. Curr. Sci. 2005, 89, 1577–1582.
27. Colombo, S.; Main, I.; Forde, M. Assessing damage of reinforced concrete beam using “b–value” analysis of
acoustic emission signals. J. Mater. Civ. Eng. 2003, 15, 280–286. [CrossRef]
28. Carpinteri, A.; Lacidogna, G.; Manuello, A.; Niccolini, G. A study on the structural stability of the Asinelli
Tower in Bologna. Struct. Control Health 2016, 23, 659–667. [CrossRef]
29. Carpinteri, A.; Lacidogna, G.; Corrado, M.; Di Battista, E. Cracking and crackling in concrete-like materials:
A dynamic energy balance. Eng. Fract. Mech. 2016, 155, 130–144. [CrossRef]
30. Niccolini, G.; Borla, O.; Accornero, F.; Lacidogna, G.; Carpinteri, A. Scaling in damage by electrical resistance
measurements: An application to the terracotta statues of the Sacred Mountain of Varallo Renaissance
Complex (Italy). Rend. Lincei Sci. Fis. Nat. 2015, 26, 203–209. [CrossRef]
31. Grunwald, C.; Khalil, A.A.; Schaufelberger, B.; Ricciardi, E.M.; Pellecchia, C.; De Iuliis, E.; Riedel, W.
Reliability of collapse simulation—Comparing Finite and Applied Element Method at different levels.
Eng. Struct. 2018, 176, 265–278. [CrossRef]
32. Domaneschi, M.; Cimellaro, G.P.; Marano, G.C.; Morgese, M.; Pellecchia, C.; Khalil, A.A. Numerical
simulations of collapse tests on RC beams. In Proceedings of the 10th International Conference on Bridge
Maintenance, Safety and Management—IABMAS 2020, Sapporo, Japan, 11–15 April 2020.
33. Applied Science International LLC. Extreme Loading for Structures 2018; Applied Science International
LLC.: Durham, NC, USA, 2018; Available online: https://www.appliedscienceint.com/extreme-loading-for-
structures/ (accessed on 30 March 2020).
© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
16
applied
sciences
Article
Application of Impact-Echo Method to 3D SIBIE
Procedure for Damage Detection in Concrete
Katsufumi Hashimoto 1, *, Tomoki Shiotani 1 and Masayasu Ohtsu 2
1 Department of Civil and Earth Resources Engineering, Graduate School of Engineering, Kyoto University,
Kyoto 615-8540, Japan; [email protected]
2 Center for the Promotion of Interdisciplinary Education and Research, Graduate School of Engineering,
Kyoto University, Kyoto 615-8540, Japan; [email protected]

Received: 25 February 2020; Accepted: 12 April 2020; Published: 15 April 2020
Abstract: In this study, to visualize damage and defects, such as cracks and voids in concrete, the SIBIE
(stack imaging of spectral amplitudes based on impact echo) procedure is applied and numerically
improved to construct a three-dimensional (3D) model of elastic wave propagation behavior. A unit of
arrayed accelerometers is installed to detect multi-channel signal waveforms in the frequency domain.
The resonant frequencies due to reflections at each node in 3D lattice nodes are computed by using
the distances from elastic wave input to multiple output locations. The amplitudes corresponding
to the resonant frequencies in the spectrum are summed up as the reflection intensity of elastic
wave at each node. The reflection intensity distribution is visualized finally in the targeted area
three-dimensionally. Case studies are carried out on the proposal of the improved 3D-SIBIE procedure,
applied to a concrete specimen with simulated-damage as well as in-situ highway RC (Reinforced
Concrete) slabs in service. As for the signal detection, a non-contact elastic wave detecting system
using a laser doppler vibrometer is also introduced to consider and validate the promising remote
sensing and inspection technique for damage evaluation in concrete with the 3D SIBIE procedure.
Keywords: non-destructive testing; concrete; impact-echo; elastic wave; SIBIE; 3D visualization;

accelerometer; laser doppler vibrometer; remote sensing
1. Introduction
1.1. Non-Destructive Testing for Concrete

It has been generally recognized that preventive and proactive maintenance ideas are necessary for
such social infrastructure as bridges and tunnels to develop rational life-cycle scenarios. For reinforced
concrete (RC) members, essential issues involve the establishment of a maintenance system with
appropriate measurements before evolving the serious damage and failure. For non-destructive testing
(NDT), to qualify the materials’ property or structural integrity of concrete, several NDT methods have
been applied.
Since sufficient and effective maintenance of infrastructure is desired under the restriction of
budget, appropriate inspections by means of NDT must be applied adequately. As for the damage
assessment and estimation of repair and retrofit recovery in concrete structures, in addition to
developing effective NDTs, innovative methods corresponding to the damage characteristics must be
established in parallel. Also, it is strongly demanded to propose appropriate management systems for
ageing concrete infrastructures, adapting the aforementioned accurate and ready-to-use inspection
techniques for the damage in concrete.
In concrete structures of bridges such as decks, the most problematic type of damage is fatigue
damage due to numerous repetitions of excessive traffic overloads [1,2]. The final form of damage due
17
Appl. Sci. 2020, 10, 2729
to cyclic loads is well known as horizontally induced large cracks in steel–concrete composite slabs [3].
As it is difficult to detect interior damage, which potentially affects structural performance, by visual
inspection [4], this becomes a very serious problem for highway structures composed of reinforced
concrete (RC) [5]. For the RC bridge slabs in service, inspection work is generally allowed to implement
the bottom side, not including the top side, to avoid controlling the traffic. Therefore, a one-side access
NDT method is strongly demanded to quantify or assess the damage in RC members. When the
special permission to access both sides is given for in-depth inspections or experimental purposes,
tomographic approaches with elastic waves have been applied to visualize the internal damages.
Specifically, AE (acoustic emission) sensors or accelerometer arrays are installed on the bottom side and
artificial excitations are made on the top side so the internal damage can be visualized with such elastic
wave parameters as P-wave velocity and energy attenuation [6,7]. One-side measurements with surface
wave techniques have thus been employed to investigate the interior damage of concrete [8]. However,
specific depths could not so far be identified with this technique. Meanwhile, the impact-echo method
is the most well-used non-destructive testing for concrete structures to characterize the surface area of
interest [9,10]. The SIBIE (Stack Imaging of spectral amplitudes Based on Impact-Echo) procedure is an
advanced visualization technique using the data of impact echo testing [11,12]. SIBIE demonstrates a
two-dimensional (2D) internal image of concrete simply with one receiver while leaving such issues to
study such as 3D visualization and more accurate damage identification.
1.2. Impact-Echo Method

Impact-echo is an acoustic method for the nondestructive evaluation of concrete, which is based
on the use of impact-generated elastic waves that propagate through concrete, and they are reflected by
internal flaws or external surfaces [9]. Impact-echo can be used to not only determine the location and
extent of flaws, such as cracks, delamination, and voids in the targeted concrete, but also to provide
thickness measurements of concrete slabs as well [10].
As shown in Figure 1, in the impact-echo system, an induced elastic wave at a surface of concrete
propagates and reflected by defects or the surfaces in concrete. The received signal is monitored
by a transducer coupled to the surface near the elastic wave source. The time-domain waveform is
transformed into the frequency-domain one using the fast Fourier transform (FFT). The frequency
spectrum contains those resonance frequencies due to reflections at such interfaces as the internal
defects and the surfaces of the material of interest. When the P-wave velocity in the concrete is known,
the detected peak frequency can be expressed as in Equations (1) and (2), respectively.
Cp
fT = (1)
2T
Cp
fd = (2)
2d
where fT is resonance frequency due to material thickness, fd is resonance frequency due to internal
damage (defect), T is the thickness of targeted member, d is defect depth, and Cp is P-wave velocity
in concrete.
18
Appl. Sci. 2020, 10, 2729
Figure 1. Schematic of impact-echo method applied to the testing interior concrete.
1.3. SIBIE Procedure

In order to visualize such defect as crack and void in concrete, SIBIE (stack imaging of spectrum
amplitudes based on impact-echo) procedure has been developed [11,12]. In the procedure, a targeted
cross-section is modeled and divided into square elements as shown in Figure 2. Then, resonant
frequencies obtained from FFT analysis due to elastic wave reflections at each element are computed.
The travel distance from the input, which is excited by hammering (tapping) concrete surface with
steel sphere ball of 10 mm diameter, to the output through the elements is calculated by Equation (3).
R = r1 + r2 (3)
where r1 is the distance from the input (impact) point to the nodal point of an element and r2 is the
distance from the nodal point of an element to the output (signal detection) point. The resonant
frequencies due to reflections at each element fR are calculated by Equation (4).
Cp
fR = (4)
R
Figure 2. Mesh division and results of SIBIE (Stack Imaging of spectral amplitudes Based on
Impact-Echo) procedure [3].
In the frequency spectrum, amplitudes corresponding to these resonances are summed up at each
nodal point. Thus, the reflection intensity as the data stacking result at all the analytical elements is
prepared for the visualized contour image. After the SIBIE analysis, a 2D view of the contour image
is displayed, and the high reflected zone is visually identified as the estimated location of anomaly
namely damage.
19
Appl. Sci. 2020, 10, 2729
The SIBIE procedure has been applied for the case studies in the author’s previous research.
The results for detecting grouted and un-grouted PC (pre-stressed concrete) sheaths is illustrated as in
Figure 2 [11]. The size of each mesh is of dimensions 10 × 10 mm. In the figures, it can be observed that
both grouted and un-grouted sheaths were clearly identified with the values of reflection intensities.
The contour values are given as the relative amplitude, which is obtained by normalizing data with a
maximum amplitude provided from the FFT result, of the P-wave intensity computed from frequency
spectrum of received signal.
In the above-mentioned results, after the SIBIE procedure, it is possible to identify the grouting
condition only when the sheath location is preliminarily known. When the sum of distance R in
Equation (3) from input point to a nodal point and the nodal point to output point is the same
(Figure 2), it is understandable that the same reflection intensities are given. As the location is based
on Equation (3), there are many candidates of nodes with the same travel distance, and therefore
the conventional the SIBIE procedure with only one receiver cannot occasionally provide correct
geometrical information for asymmetrically distributed damage to the vertical axis e.g., in Figure 2.
To upgrade the procedure from 2D to 3D with multiple receivers, in this study, an array with multiple
sensors over a targeted area is employed. Specifically, four accelerometers and laser doppler vibrometer
system, which is innovatively introduced for remote sensing technique, are used to improve the
identification of damage in concrete. A concrete slab of existing structures and an experimentally casted
slabs embedded by artificial damage with styrofoam are subjected to the expanded SIBIE procedure.
1.4. Remote Sensing for Elastic Wave Detection with Laser Doppler Vibrometer
As for remotely detecting elastic waves propagated through concrete, non-contact acoustic systems
with a laser doppler vibrometer (LDV) for detecting signals have been reported [13–15]. These research
approaches provide remote and rapid operations for inspection works based on the elastic wave
method. In this study, a laser doppler vibrometer is also introduced, in addition to conventional
accelerometers, to detect the elastic wave in concrete. The signal output from LDV is generally a
continuous analog voltage that is directly proportional to the velocity component at targeted surface
along the laser beam direction. The system specification of LDV used in this study is given as digital
√
velocity decoder with 0.5 Hz–22 kHz frequency band and velocity resolution is <0.02 (µm/sec)/ Hz.
The signal detectable distance of LDV is from 0.1 to 30 m. This enables the non-contact vibration
measurements of concrete surface and obtain the signal information required for the SIBIE analysis in
addition to the conventional contact signal acquisition system with accelerometers.
2. 3D SIBIE Procedure
2.1. Sensor Array with Accelerometers and Laser Doppler Vibrometer

To develop an improvement for the SIBIE procedure in this study, the measurement and the
computation steps in the procedure are extended to three-dimensional (3D) model, using a multi-
sensor (accelerometer) array unit for on-site measurement as shown in Figure 3. The unit is developed
to access only underneath an RC deck in service with 30 × 30 cm grid composing of 4 sensors, arranging
each apex of 300 mm square. The thickness of the deck is approximately 400 mm, composed of 180 mm
RC (reinforced concrete), 70 mm FRC (fiber reinforced concrete) overlay, and 50 mm asphalt pavement.
The multi-sensor inputs and outputs provide the spectral amplitudes of resonance frequencies
due to 3D propagation of elastic wave reflections at each element (voxels) in depth (thickness of the
targeted concrete) direction of 2D plane (300 × 300 mm), which is the volume of interest for 3D SIBIE
analysis. Additionally, to achieve remote-controlled non-destructive testing, the center of the array
unit is targeted by laser doppler vibrometer to obtain the signal as well as the accelerometers in the
laboratory test. Figure 4 shows the application overview and the received signal when introducing the
laser doppler vibrometer to detect the signals. As the waveform signal detected by LDV demonstrates
sufficient amplitude, this can be used as the received signal through the concrete by the SIBIE procedure.
20
Appl. Sci. 2020, 10, 2729
Figure 3. (a) Accelerometer array unit; (b) Overview of on-site implementation.
Figure 4. (a) Laser doppler vibrometer; (b) Detected waveform with application of laser doppler
vibrometer for laboratory test.
2.2. Computing Procedure for 3D SIBIE

The following steps show one-side access measurement and the computation procedure, developed
in this study for the 3D SIBIE procedure. Figure 5 displays the targeted volume of interest and cross
sections employed for the computation.
1. Place the sensor array on one surface of targeted concrete and hammer with steel sphere ball of
10 mm diameter near each sensor (n = 4) for 10 times respectively.
2. Divide 300 (x-direction) × 300 mm (z-direction) in depth (y-direction) for volume of interest into
10 mm meshes.
3. Employ the nodal points on the cross sections of x = 0, x = 300, z = 0, z = 300, x = z, x + z = 300 to
mitigate computational load.
4. Calculate R (travel length of elastic wave) from each nodal point to input and output points (R is
the shortest linear path as explained).
5. Search for fR frequencies due to reflection of propagating P-wave with velocity of Cp .
6. Normalize the amplitude from 0 to 1 based on the highest peak value in the frequency spectrum
to cancel the variability of each manual hammering impact.
7. Stack amplitudes according to fR and quantify reflection intensity values at each nodal point.
8. Visualize 3D geometry with the stacking data based on local polynomial regression method.
Figure 6 shows the example of frequency spectrum successfully obtained from the received
signal followed by FFT analysis. Sampling rate is set at 100 kHz in the following measurement.
Those sampling rates are appropriately determined to discuss the FFT (Fast Fourier Transform) results
in this study since the waveform is sampled at 10000 points for FFT to achieve the sufficient resolution
less than 50 Hz.
Meanwhile, some recorded waveform due to an impact of low energy may cause the low S/N
(signal-to-noise ratio) signal and FFT indeterminate peaks. The procedure of FFT signal data stacking
with/without including low S/N waveforms are shown in Figure 7. Tapping 10 times according to
the above-mentioned steps for 3D SIBIE method, FFT peak stacking results in giving an incorrect
21
Appl. Sci. 2020, 10, 2729
data with unclear frequency spectrum when some signals into a sensor are not entirely high S/N,
wherein n = 3, 6, 8, 9 for the example. As a consequence, between the process of 5 and 6 for the
development of the 3D SIBIE procedure, low S/N signals are removed from the computation to
guarantee data quality. The FFT signal data stacking result without signal normalization procedure is
indicated numerically as in Figure 8, where the process of 6 is excluded. It is obvious that the spectrum
of n:3 is different from others in 5 kHz implying that this is attributed to some noises. As shown in
Figure 8a, the stacked amplitude of 10 raw data without the signal normalization is influenced by the
noise signal of n:10 but now found in Figure 8b. The normalization for each frequency spectrum before
stacking is thus very effective on the data acquisition process of the 3D SIBIE procedure.
Figure 5. Volume of interest and cross section for 3D SIBIE procedure.
Figure 6. An example of FFT (Fast Fourier Transform) result of low S/N signal.
Figure 7. FFT signal data stacking procedure.
22
Appl. Sci. 2020, 10, 2729
Figure 8. FFT signal data stacking procedure with/without each signal normalization procedure.
3. Examination of 3D SIBIE Procedure
3.1. Application of Accelerometer Array for Detecting Signals
3.1.1. Reinforced Concrete Bridge Deck

The 3D SIBIE procedure has been applied for existing structures. A RC bridge deck in service
was targeted. Two different areas were selected for the case study after the crack inspection work.
The crack inspection procedure is schematically described in Figure 9 with 4 steps as follows.
1. Drilling a 5 mm diameter hole into the concrete deck from the surface of asphalt pavement
2. Injecting epoxy resin (in orange-colored to clearly see the crack)
3. Re-Drilling a 10 mm diameter hole at the same location after the resin is hardened
4. Inserting a flexible fiber optic borescope into concrete to take photos
Figure 9. Visual rack inspection procedure.
Figure 10 shows an example of the view from the fiberscope when it caught the crack filled
with resin. In this way, damaged and non-damaged namely intact parts are chosen for the 3D SIBIE
procedure. Figure 11 shows the 3D SIBIE result for the RC deck in the intact part and Figure 12 shows
23
Appl. Sci. 2020, 10, 2729
that for RC deck in the damaged part where a lateral crack in concrete and also delamination between
an asphalt pavement layer and a concrete layer was found. The sensor array was placed on the bottom
surface of the deck and measurement was carried out for the 3D SIBIE procedure. The top surface
denotes that of asphalt pavement and the bottom shows the bottom of the concrete layer both in
Figures 11 and 12. In these figures, the thickness of the concrete layer is 0–180 mm, 180–250 mm in the
fiber reinforced concrete (FRC) overlay, and the asphalt pavement is 250–300 mm from the bottom.
In the intact part as shown in Figure 11, no sharply defined reflection intensity is found in the volume of
interest while, for the damaged part as shown in Figure 12, the color gradations show strong reflections
of elastic waves due to anomaly such as cracks or voids around the bottom of the RC deck (see the
bottom area in the right side sectional view). Also, three layers consisting of concrete, FRC and asphalt
can be classified by the 3D visualized image. As the strong reflections are obtained from the large
difference of acoustic impedance of which the individual materials have, the delamination between the
asphalt and the concrete layers, which can be visually confirmed, and sub-critical cracks, which cannot
be visually confirmed, are possibly attributed to this fact.
Figure 10. Cracks filled with resin (a,b).
Figure 11. 3D SIBIE result at intact part.
Figure 12. 3D SIBIE result at damaged part.
3.1.2. Concrete Specimen with Simulated Defect

Concrete specimen using high-early strength cement was prepared for 30 MPa design strength.
The size is 300 (x-direction) × 400 (y-direction) × 300 mm (z-direction). The thickness is defined as
24
Appl. Sci. 2020, 10, 2729
y-direction in this study. Figure 13 shows the geometrical information of concrete specimens and
sensing (accelerometers and LDV) locations. As for experimentally simulating damage in concrete,
a cubic styrofoam of 300 (x-direction) × 50 (y-direction for thickness) × 150 mm (z-direction) was
embedded at 175 mm depth in the y-direction in the center of specimen.
Figure 13. Concrete specimens and sensor locations (with defect and with no defect (intact)).
As shown in Figures 14 and 15, when the accelerometer array unit is employed for detecting
signals, in the visualization results with the 3D SIBIE procedure for interior concrete, the location
of defect (simulated damage) is clearly identified from 175 to 225 mm in the depth (y) direction in
Figure 14, where higher reflection intensity is given, while no remarkable changes is not found in other
depths as well as in the intact specimen as shown in Figure 15. Note, the values of reflection intensities
are provided as color gradations relatively from 0 to 1 as normalized data based on the maximum
value of each contour displayed. The reflection intensity at the embedded styrofoam is even lower
than that of the thickness reflection at 400 mm depth in the y-direction, since the acoustic impedance
of air is obviously lower than that of the styrofoam. As for the result of intact concrete in Figure 15,
only the thickness reflection is demonstrably observed at 400 mm depth in the y-direction.
Figure 14. Result of 3D SIBIE procedure for damage-simulated concrete specimen.
25
Appl. Sci. 2020, 10, 2729
Figure 15. Result of 3D SIBIE procedure for intact concrete specimen.
3.2. Application of Laser Doppler Vibrometer for Detecting Signals

A laser doppler vibrometer for detecting signals has been introduced for the 3D SIBIE procedure
to visualize the simulated damage in the concrete specimen. When LDV signal detection is applied
and the data are introduced to the 3D SIBIE procedure, the visualized images in Figure 16 successfully
indicate the defect position with higher reflection intensity at the middle depth of the specimen than
that at the other area in the targeted volume of interest. On the other hand, when it is applied to the
intact specimen as shown in Figure 17, an intensive reflection is only obtained from the thickness at
400 mm depth in the y-direction, as is the case of the accelerometers.
Figure 16. 3D SIBIE result with LDV (Laser Doppler Vibrometer) signal detection for damage-simulated
concrete specimen.
26
Appl. Sci. 2020, 10, 2729
Figure 17. 3D SIBIE result with LDV signal detection for intact concrete specimen.
3.3. 3D SIBIE Procedure Improved with Signal Amplification

As explained in this study, the 3D SIBIE procedure is developed based on 3D model with the
use of the multiple-sensor array. The procedure is successfully applied for the case studies of the
targeted concrete specimen for the laboratory test by employing the accelerometer array unit and the
laser doppler vibrometer for detecting signals as well as the case of the existing RC deck in service.
As the results, the extension of the 2D computation process for the conventional SIBIE procedure to 3D
coordinates enables to visualize the interior damage. The 3D SIBIE procedure has a great promise
for on-site and non-contact measurement and potentially could be used to operate remote and rapid
damage inspection and material assessment for concrete structures.
The 3D SIBIE procedure do not consider the attenuation of propagation in elastic waves so that
the results described above in Figures 14–17, relatively high reflection intensity is found even in the
intact area of interest. Since the amplitude in frequency domain of the signal detected at the receiving
point with the accelerometers or LDV by reflection of elastic wave becomes smaller as its propagation
distance becomes longer, it is considered that the deeper the detection depth in the targeted volume,
the smaller the reflection intensity mathematically in the result of signal computing process of the
SIBIE procedure.
Pile Integrity Test (PIT) utilizes a simple principle of elastic wave characters to assure the quality of
the concrete pile. When elastic waves are excited at the pile head, it is assumed that elastic waves travel
at the speed of C inside the concrete pile shaft, and the pile length can be determined by measuring
the time lapse of T. Based on the detected signal waveform between the time of hammering and the
time of receiving reflections on pile head, as shown in Figure 18, the amplitude of the reflected wave is
attenuated due to the effect of geometrical or internal damping of the materials during the propagation.
Thus, in the applied technique for in-situ measurement, an exponential amplification function is
applied across the recorded waveform to amplify the low energy reflection peaks from the pile toe and
other internal defects such as voids and cracks [16–18].
In the same manner as in the signal amplification method of PIT principle, the method is also
introduced for the 3D SIBIE procedure to correct the reflection intensity in the visualization results.
Figure 19 shows some examples of actually recorded signals with the accelerometer sensor array,
explained in Figure 13, followed by hammering at concrete surface. In order to define an exponential
function for signal amplification in consideration with the signal attenuation through the targeted
concrete, an exponential function is derived for each data set from the obtained waveforms as in
Figure 20.
27
Appl. Sci. 2020, 10, 2729
Figure 18. Principle of PIT (Pile Integrity Test).
Figure 19. Detected signals at accelerometers with a hammer strike.
Figure 20. Signal attenuation along wave propagation in sensor distance.
The respective functions for each recorded data can be used to calibrate the attenuation of the
detected signals. By using the calibrated data according to the signal amplification function, as shown
in Figures 21–24, the visualization images of reflection intensity are modified by the 3D SIBIE procedure.
The results indicate that the visualized images are changed due to the improved signal processing.
28
Appl. Sci. 2020, 10, 2729
This means that reflection intensities at the damage-simulated part with styrofoam is sufficiently high
compared to those at an intact part without the consideration of the attenuation effect during elastic
waves propagation. On the other hand, as for intact cases, the relatively high reflection intensities
expressed in the intact part are perfectly removed and only the high reflection at 400-mm depth
corresponding to the boundary of the specimen is obtained.
Figure 21. Improved 3D SIBIE results for damage-simulated concrete specimen.
Figure 22. Improved 3D SIBIE results for intact concrete specimen.
In order to get even clearer visualized image, more elastic wave ray paths by increasing the
number of hammering locations and sensors would work effectively. In addition, it is necessary to take
into account the fact that the elastic wave velocity is not constant in real situation. It is assumed in
the above-mentioned computations of this study that the elastic velocity is the same throughout the
targeted concrete matrix. The further investigation needs to consider that the velocity could be varied
according to the material property and treated as the variable number in the 3D-SIBIE procedure.
29
Appl. Sci. 2020, 10, 2729
Figure 23. 3D SIBIE result with LDV signal detection for damage-simulated concrete specimen.
Figure 24. 3D SIBIE result with LDV signal detection for intact concrete specimen.
4. Conclusions
There is great demand for simplified and innovative techniques for the maintenance and assessment
of concrete structures. Inspection works need to be ready-to-use as well as save time and costs related
to damage evaluation. Developments in this area will support damage visualization to diagnose social
infrastructures practically, i.e. highways, tunnels, and dams built using concrete.
In order to visualize damage, such as cracks and voids in concrete, the SIBIE (stack imaging
of spectral amplitudes based on impact echo) procedure was applied and improved numerically.
SIBIE was proposed in this study to be upgraded from a 2D to 3D model of elastic wave propagation
behavior. Using the multiple sensor array composed of a handy mechanical device with accelerometers,
the 3D SIBIE procedure was developed as a simple and ready to use measuring technique. It was also
found that the laser doppler vibration sensing method could work for the remote data acquisition in
the detection of elastic wave reflection for the procedure.
The procedure was successfully applied for the case studies of targeting RC deck in service
as well as a concrete specimen for the laboratory test. In terms of results, the extension of the 2D
computation process for the conventional SIBIE procedure to 3D coordinates enabled to estimate and
display interior damages and geometrical information of materials constituting the layers of RC deck,
30
Appl. Sci. 2020, 10, 2729
such as asphalt pavement, fiber-reinforced overlay, and concrete. Additionally, the data computation
process was improved considering the signal attenuation during the propagation, resulting in a clearer
finding in terms of the damage assessment in the visualized images than that of the conventional
calculation method.
It could be concluded that the 3D SIBIE procedure has a great promise for on-site measurement
and can be employed to operate deterioration diagnosis for invisible damage in concrete and material
assessment for concrete infrastructures accordingly.
Author Contributions: Conceptualization, K.H., T.S. and M.O.; Formal analysis, K.H.; Funding acquisition, T.S.;
Investigation, K.H.; Methodology, K.H.; Visualization, K.H.; Validation, K.H. and T.S.; Supervision, T.S. and
M.O.; Writing—Original Draft, K.H.; Writing—Review and Editing, T.S. All authors have read and agreed to the
published version of the manuscript.
Funding: This research received no external funding.
References
1. ACI Committee 215. Consideration for design of concrete subjected to fatigue loading. ACI J. 1974, 71, 97–121.
2. Zhang, J.; Peng, H.; Cai, C.S. Field study of overload behavior of an existing reinforced concrete bridge under
simulated vehicle loads. J. Bridge Eng. 2010, 16, 226–237. [CrossRef]
3. Fujiyama, C.; Maekawa, K. A computational simulation for the damage mechanism of steel-concrete
composite slabs under high cycle fatigue loads. J. Adv. Concr. Technol. 2011, 9, 193–204. [CrossRef]
4. Fathalla, E.; Tanaka, Y.; Maekawa, K. Effect of crack orientation on fatigue life of reinforced concrete bridge
decks. Appl. Sci. 2019, 9, 1644. [CrossRef]
5. Toutlemonde, F.; Ranc, G. Fatigue tests of cracked reinforced concrete slabs for estimating the service life of
composite bridge decks. Revue Française de Génie Civil 2001, 5, 483–494. [CrossRef]
6. Hashimoto, K.; Shiotani, T.; Nishida, T.; Miyagawa, T. Application of elastic-wave tomography to repair
inspection in deteriorated concrete structures. J. Disaster Res. 2017, 12, 2017. [CrossRef]
7. Chai, H.K.; Momoki, S.; Kobayashi, Y.; Aggelis, D.G.; Shiotani, T. Tomographic reconstruction for concrete
using attenuation of ultrasound. NDT E Int. 2011, 44, 206–211. [CrossRef]
8. Chai, H.K.; Aggelis, D.G.; Momoki, S.; Kobayashi, Y.; Shiotani, T. Single-side access tomography for
evaluating interior defect of concrete. Constr. Build. Mater. 2010, 24, 2411–2418. [CrossRef]
9. Carino, N.J.; Sansalone, M. Detecting voids in metal tendon ducts using the impact-echo method. Mater. J.
Am. Concr. Inst. 1992, 89, 296–303.
10. Sansalone, M.; Lin, J.M.; Streett, W.B. A New Procedure for determining the thickness of concrete highway
pavements using surface wave speed measurements and the impact-echo method. In Innovations of
Nondestructive Testing; Pessiki, S., Olson, L., Eds.; American Concrete Institute: Farmington Hills, MI, USA,
1997; p. 167.
11. Ohtsu, M.; Watanabe, T. Stack Imaging of Spectral Amplitudes based on Impact-Echo for Flaw Detection.
NDT E Int. 2002, 35, 189–196. [CrossRef]
12. Ohtsu, M.; Yamada, M. and Sonoda, T. Quantitative evaluation of SIBIE procedure and case studies. Constr. Build.
Mater. 2013, 48, 1248–1254. [CrossRef]
13. Akamatsu, R.; Sugimoto, T.; Utagawa, N.; Katakura, K. Proposal of non-contact inspection method for
concrete structures using high-power directional sound source and scanning laser doppler vibrometer. Jpn. J.
Appl. Phys. 2013, 52, 07HC12. [CrossRef]
14. Kurahashi, S.; Mikami, K.; Kitamura, T.; Hasegawa, N.; Okada, H.; Kondo, S.; Nishikino, M.; Kawauchi, T.;
Shimada, Y. Demonstration of 25-Hz-inspection speed laser remote sensing for internal concrete defects.
J. Appl. Remote Sens. 2018, 12, 015009. [CrossRef]
15. Furusawa, A.; Takenaka, Y.; Nishimura, A. Proposal of laser-induced ultrasonic guided wave for corrosion
detection of reinforced concrete structures in Fukushima daiichi nuclear power plant decommissioning site.
Appl. Sci. 2019, 9, 3544. [CrossRef]
16. Rausche, F.; Kung, S.R.; Linkins, G. Comparison of pulse echo and transient response pile integrity test
methods. Transp. Res. Rec. 1991, 1331, 21–27.
31
Appl. Sci. 2020, 10, 2729
17. Rausche, F.; Linkins, G.; Kung, S.R. Pile integrity testing and analysis. In Proceedings of the International
Conference on the Application of Stress-Wave Theory to Piles, The Hague, The Netherlands, 21–24 September
1992; pp. 613–617.
18. Ni, S.H.; Lehmann, L.; Charng, J.J.; Lo, K.F. Low-strain integrity testing of drilled piles with high slenderness
ratio. Comput. Geotech. 2006, 33, 283–293. [CrossRef]
32
applied
sciences
Article
Non-Destructive Tests for Damage Evaluation of
Stone Columns: The Case Study of Sacro Monte in
Ghiffa (Italy)
Alessandro Grazzini * , Sara Fasana, Marco Zerbinatti and Giuseppe Lacidogna
Department of Structural Geotechnical and Building Engineering, Politecnico di Torino, 10138 Torino, Italy;
[email protected] (S.F.); [email protected] (M.Z.); [email protected] (G.L.)
* Correspondence: [email protected]

Received: 29 February 2020; Accepted: 10 April 2020; Published: 13 April 2020
Abstract: The Italian Sacri Monti are heritage sites with some unique characteristics; they are
a successful symbiosis between nature and art and are unconfined structures, therefore always being
accessible but exposed to atmospheric agents, with many relevant consequences with regard to
conservation problems. The paper discusses some aspects related to the application of non-destructive
techniques (NDT) for the interpretation of degradation phenomena occurring in stone structural
elements. Ultrasonic and impact tests were used to evaluate the structural properties of the stone
columns in the Via Crucis portico, within the monumental complex of the Sacro Monte in Ghiffa
(Piedmont, Italy), in order to determine their conditions of maintenance and to evaluate the portico
static stability. Ultrasonic tests made it possible to obtain the value of the dynamic elastic modulus,
which was variable at different points of the columns due to the diversified level of material damage.
The impact test, performed with an instrumented hammer in the same points of the ultrasonic test,
enables, by comparison, a deeper knowledge of the surface resistance of damaged columns. These
results are the first step in a research path that will require further laboratory tests to better calibrate
the diagnostic techniques applied to different levels of damage to surface materials.
Keywords: non-destructive techniques (NDT); historical buildings; maintenance; stone columns;

ultrasonic test; impact test
1. Introduction
Historical architectural heritage requires specific maintenance and restoration interventions,
especially in the presence of prolonged deterioration over time. Restoration works may control
deteriorations, which can affect monuments, without secondary effects on treated surfaces of the
historical elements [1–5]. Similarly, diagnostics, which is a fundamental phase of knowledge in
order to deepen the critical aspects of structure assessments and degradation causes, must operate
with technologies that can be used in a non-destructive and non-invasive way for the historical
buildings [6–13]. The main purposes related to the general framework of the material maintenance
conditions are those to refine methodological paths for such contexts, to have well-founded analytical
support and to reduce the error range in the qualitative and quantitative evaluation of the damage
process. Ultrasonic tests are one of the non-destructive techniques most often used in the field
of diagnostics on historical buildings [14–16]. They measure the wave speed through the material
thickness, which is in relation to the dynamic elastic modulus. The results obtained by tests on
architectural heritage are representative, if compared with known physical characteristics of similar
materials, of the degradation inside the structural section of historical elements. In addition, the impact
tests, performed with an instrumented hammer in the same points of the ultrasonic test, enable the
evaluation of the surface degradation level [17].
33
Appl. Sci. 2020, 10, 2673
This research concerns the mechanical characterization, through non-destructive techniques (NDT),
of the 13 stone columns (granitoid rock) that support the Via Crucis portico inside the monumental
complex of the Sacro Monte in Ghiffa. The Sacri Monti of northern Italy are a UNESCO heritage site
that includes nine complexes that constitute groups of chapels and other buildings with many artistic
masterpieces. All buildings are part of a high-quality natural environment. The first complex was
founded in the late 15th century in Varallo Sesia, and the others were constructed up to the end of
17th century and dedicated to different aspects of the Christian faith [18]. In addition to their symbolic
spiritual meaning, they are of great beauty by virtue of the skill with which they have been integrated
into the surrounding natural landscape of hills, forests and lakes. They also house much important
artistic material in the form of wall paintings and statuary.
Many buildings within the Sacri Monti monumental complexes suffer from different conditions
of degradation, linked in particular to the climatic conditions of moisture, rain, damp action from
capillary rising, freeze–thaw cycles that cause the detachment of plaster and the degradation of the
stone elements [19]. A research group of the Politecnico di Torino has been interested for years in
the study of non-destructive diagnostic techniques for the evaluation of degradation levels of the
architectural and structural elements of historical buildings in the Sacri Monti complexes [1,20–23].
In the specific case of the Sacro Monte at Ghiffa, the Via Crucis portico (Figure 1) is supported by
13 stone columns with evident intense superficial material degradation. The portico was built between
1752 and 1761. It is marked by 14 spans and has columns and capitals made by stones derived both
from local “granitic stones”. It is possible that the origin of the stone of columns is not from historical
quarries but from erratic boulders found near the construction site (aplite is the current interpretation,
but other investigations are in progress), referred to in slang as “trovanti”, while for the capitals there
is a strong similarity with a serizzo quarried near Locarno (Valle Verzasca, Switzerland) (Figure 2).
Considering a macroscopic mineralogical analysis with a 10x magnifying glass, the rock is mainly made
up of more than 90% of crystalline clear silica minerals (quartz and feldspar), with little biotitic dark
mica and rare light mica. The main accessory mineral is a small red-vined garnet that characterizes the
stone. The veins present confirm the low quality of the stone, almost certainly extracted from makeshift
material, greatly influencing the variability of mechanical characteristics. Perhaps, also for the lower
quality, it is very prone to degradation phenomena, worsened by atmospheric moisture, the presence of
salts, freeze–thaw cycles and thermal expansion, which is also due to strong summer radiation, which
can produce erosion and exfoliation on the stone surface [24,25] (Figure 3). The material degradation
is increasing from the inside to the outside of the structural section of the columns, and the result is
a decay of their compressive strength, the residual value of which needs to be investigated.
The aim of the ongoing research is to deepen the knowledge of the mechanical characteristics of
the columns of the Via Crucis portico resulting from degradation, to estimate their residual strength.
In order to improve the maintenance and conservation planning of the historical buildings belonging
to the Sacri Monti monumental complex, some non-destructive tests were carried out on the columns
of the portico to investigate the elastic modulus of the stone. Specifically, the ultrasonic technique was
used, associated with that of impacts for a more precise comparison of the stone surface degradation.
From the value of the dynamic elastic modulus, it was possible to estimate the compressive strength
of each single column using experimental reports. As a matter of fact, the ultrasonic and impact
tests showed a great variability of the elastic modulus already within the same column and between
columns. This variability was the demonstration of a material degradation distributed unevenly within
the same stone element and between the columns.
34
Appl. Sci. 2020, 10, 2673
Figure 1. The Via Crucis portico inside the Sacro Monte of Ghiffa (Piedmont, Italy).
Figure 2. The stone columns of the Via Crucis portico in the Sacro Monte at Ghiffa.
35
Appl. Sci. 2020, 10, 2673
Appl. Sci. 2020, 10, x FOR PEER REVIEW 4 of 16
(a) (b)
Figure 3.
Figure 3. Examples
Examples ofof erosion
erosion (a),
(a), exfoliation
exfoliation and
and detachment
detachment (b)
(b) problems
problems in
in stone
stone columns
columns of
of the
the
Via Crucis
Via Crucis portico.
portico. The reference
reference lexicon
lexicon is that of the "Ente Nazionale Italiano di Unificazione" (UNI)
Standard rules
Standard rules 11182:2006.
11182:2006.
These
The aim results
of thecould
ongoingbe used in forthcoming
research is to deepen finite element numerical
the knowledge modeling
of the mechanical of the portico,
characteristics of
in order to evaluate its stability in the hypothesis of replacing the current roof covering,
the columns of the Via Crucis portico resulting from degradation, to estimate their residual strength. made with
lighter
In ordermodern stones
to improve theslabs, by traditional
maintenance roof coveringplanning
and conservation made byofmore heavy ones.
the historical The assessments
buildings belonging
that emerged
to the frommonumental
Sacri Monti these diagnostic tests will
complex, some also be useful for future
non-destructive maintenance
tests were carried outwork planned
on the for
columns
the conservation of the Via Crucis portico; the diagnostic procedure proposed could then
of the portico to investigate the elastic modulus of the stone. Specifically, the ultrasonic technique be extended
to
wasinvestigate for the various
used, associated degradation
with that of impacts problems afflicting
for a more the historical
precise comparison buildings
of the of Sacrisurface
stone Monti
monumental
degradation. complexes.
From the value of the dynamic elastic modulus, it was possible to estimate the
The experimental
compressive strength ofresults described
each single column represent a first phase reports.
using experimental of the research,
As a matterwhich will the
of fact, be
subsequently
ultrasonic andextended
impact tests to showed
similar atypes
greatof stones obtained
variability frommodulus
of the elastic quarriesalready
on which to the
within perform
same
compression tests in the laboratory and related non-destructive tests (ultrasonic
column and between columns. This variability was the demonstration of a material degradation and impact). In this
way it will be
distributed possiblewithin
unevenly to calibrate the non-destructive
the same stone element and techniques,
betweenwhich will then be more frequently
the columns.
used These
in situresults
on valuable
could architectural elements. finite element numerical modeling of the portico, in
be used in forthcoming
order to evaluate its stability in the hypothesis of replacing the current roof covering, made with
2. Materials and Methods
lighter modern stones slabs, by traditional roof covering made by more heavy ones. The assessments
that emerged
Each columnfromwas these diagnosticattests
investigated will alsocenter
the bottom, be useful for future
and top maintenance
by an ultrasonic test work planned for
and subsequently
thethe
at conservation
same pointsof bythean Via Crucis
impact test.portico; the of
The choice diagnostic
the pointsprocedure
was basedproposed could
above all on thenexamination,
a visual be extended
to investigate
comparing for the
points where various degradation
the material problems
degradation wasafflicting
evident the historical
with those in buildings
which theof Sacriseemed
stone Monti
monumental
in complexes.
better condition (the columns were also struck with light plastic hammers to hear their “sound”).
Both The
wereexperimental
non-destructive results describedtechniques
investigation representthata first
adaptphase
very of
wellthe
to research,
analyze the which will be
architectural
subsequently extended to similar types of stones obtained from quarries on
element under investigation. The goal was to obtain the value of the elastic modulus and to evaluate which to perform
compression
the quality oftests in the laboratory
the resistant and related
stone section subject non-destructive
to degradation. tests (ultrasonic
The test was alsoandan impact). In this
opportunity to
way it will
compare theberesults
possible to calibrate
of the ultrasonicthetest,
non-destructive techniques,the
useful for investigating which will
entire thenof
length bethe
more frequently
stone section,
used those
with in situobtained
on valuablefromarchitectural
the impact test,elements.
which, on the contrary, estimates the material characteristic
deformation by analyzing only the external surface on which the hit is performed.
2.1. Ultrasonic Test
Each column was investigated at the bottom, center and top by an ultrasonic test and
The ultrasonic
subsequently at the test
samedevice
pointsconsisted
by an impactof a Proceq
test. ThePundit
choicePL-200
of the points
® pulse generator (0.1–7930 µs
was based above all on a
visual resolution
range; examination, of 0.1comparing points
µs for pulse where
speed <793theµs,
material
1 µs fordegradation
pulse speedwas >793evident with those
µs; impulse in
voltage
which the stone seemed in better condition (the columns were also struck with light plastic hammers
100–450 Vpp); two transducers of 54 kHz, one emitting and the other receiving; a pulse amplifier and
to hear their “sound”). Both were non-destructive investigation techniques that adapt very well to
analyze the architectural element under investigation. The goal was to obtain the value of the elastic
36
Appl. Sci. 2020, 10, 2673
an electronic device for measuring the time interval. On the basis of the time taken by the wave to
travel the distance between the two transducers, the propagation speed of the elastic waves through
the stone material of the column is calculated. The speed propagation of the waves in a homogeneous
material depends on its density and elastic characteristics (elastic modulus, Poisson’s ratio), closely
related to the quality and compressive strength of the material itself. The speed of the wave is in fact
low if the material has voids, cracks or detachments that slow down the path; vice versa, it is high
if the material is compact and homogeneous, demonstrating good quality. The measurement of the
propagation time of the ultrasonic pulses was carried out with the use of two transducers, applied on
the surface of the elements involved in the control.
Direct tests were carried out by positioning the transducers on the two opposite faces of the element
under investigation, as indicated in Figure 4. The coupling of the transducer on the stone surface was
facilitated through the application of a thin layer of modeling clay in order to avoid air interpositions
and facilitate the transmission of the wave in the material. The measurements were conducted in
accordance with the methods proposed in American Society for Testing and Materials (ASTM) rules
D2845 1995 and in Ente Nazionale Italiano di Unificazione, Comité Européen de Normalisation,
(UNI EN) rules 12504-4: 2005 [26,27]. The tests are based on the determination of the propagation
speed (v) of the pulses of the ultrasonic longitudinal waves in the material, the values of which, in the
first approximation, are related to the dynamic modulus (Ed ) of the material, which consequently is
related to its mechanical strength [16]:
(1 + ν) (1 − 2ν)
Ed = v2 ρ (1)
(1 − ν)
Figure 4. Ultrasonic test carried out on columns of Via Crucis, Ghiffa.
In Equation (1), ρ is the density of material, and ν is Poisson’s ratio. In this case, from scientific
literature sources, a ρ value equal to 2630 daN/m3 and a ν value equal to 0.18 were considered [28].
37
Appl. Sci. 2020, 10, 2673
2.2. Impact Test

The impact test was performed using an instrumented hammer capable of generating an impact,
with a known mass and predetermined energy, against the surface of the stone. The software that
analyzes the data is able to return the force–time diagram of each impact. This type of test, unlike
other impulsive methods, allows the evaluation of the evolution over time of the forces acting in the
impact. For example, the sclerometer test detects only the elastic energy returned by the concrete
after the impact. On the other hand, the impact test also evaluates the given energy, the dissipated
energy, the impact time and the maximum force. It therefore represents a non-destructive technique for
evaluating the elastic and inelastic properties of materials. [17,29]. The analysis of the diagram (F, t) can
obtain the value of the homogeneous material elastic modulus through Hertz’s impact theory [29–31]
(see Appendix A).
The electric impact hammer (Figure 5) is a device produced by Piezotronic PCB Inc Company
(Depew, NY, USA). It is characterized by a variable force from 44.48 N to 4448.26 N, and a force sensor
sensitivity of 2.47 mV/N. It performs the impact of a 10 mm diameter steel ball with a total mass of
207 grams, connected to a piezoelectric pulse transducer. The impact is generated by the hammer
by means of a predetermined energy, characterized by a level of amplifier and impedance adapter.
The acquisition is performed by a Pimento model multichannel LMS signal analyzer (Figure 5).
Figure 5. Setup of the impact test.
The hammer was positioned perpendicular to the surface to be investigated. Perpendicularity

alignment was achieved thanks to four fixed metal brackets. The impact of the small impact mass
was generated by pressing a trigger on the electric hammer. The intensity of the impact is very low
and therefore not invasive even for surfaces of artistic value. The elastic modulus of the stone was
–
𝐸 ∗ 𝐸0 (1 − ν2 )
𝐸𝑠 = 38
𝐸0 + 𝐸 ∗ (ν20 − 1)
Appl. Sci. 2020, 10, 2673
determined according to the classical theory of the elastic collision between two bodies elaborated by
Hertz [17,29–31], using the following equations:

E∗ E0 1 − ν2
Es = (2)
E0 + E∗ ν20 − 1
where E* was obtained by:

s
5
2, 87 m3
E∗ = (3)
T RA1
The parameters in Equations (2) and (3) are the following: A1 is the area beneath the diagram
(F, t) (see Figure 6), T represents the time in which the contact occurs between the two surfaces; R is
the radius of curvature (10 mm) of the impact mass m (0.207 Kg), E0 is the mass elastic modulus
(222,000 MPa) and ν0 is its Poisson’s ratio (0.2), while, with regard to the stone material of the columns,
a Poisson’s
Appl. Sci. 2020,ratio ν ofPEER
10, x FOR 0.18 REVIEW
was considered [28]. 8 of 16
(a) (b)
Figure 6. (a) Generic
Figure 6. Generic force–time diagram of an impact test; (b) test results by impact test on column 1
of the portico.
3. Results
3. Results
The diagnostic
The diagnostic study
studyofofthe
thestructural
structuralcharacteristics
characteristics of of
thethe
columns
columnsof the Via Via
of the Crucis portico
Crucis was
portico
mainly
was based
mainly on the
based onexecution of ultrasonic
the execution tests, tests,
of ultrasonic whichwhich
were were
followed by impact
followed tests at
by impact theat
tests same
the
points. While the ultrasonic tests enable the evaluation of the dynamic elastic modulus of
same points. While the ultrasonic tests enable the evaluation of the dynamic elastic modulus of the the columns
by measuring
columns the speedthe
by measuring pulses,
speedthe impactthe
pulses, tests investigate
impact the material
tests investigate thedegradation on the surface
material degradation of
on the
the columns (the detachment points of material, for example).
surface of the columns (the detachment points of material, for example).
3.1. Ultrasonic Tests
3.1. Ultrasonic Tests
The ultrasonic tests were carried out on all the columns of the Via Crucis portico on 1–3 points
The ultrasonic tests were carried out on all the columns of the Via Crucis portico on 1–3 points
each (depending on the visual level of degradation), in order to have an average of the speed pulses
each (depending on the visual level of degradation), in order to have an average of the speed pulses
and the related dynamic modulus over the entire height of the column. As can be seen from the results
and the related dynamic modulus over the entire height of the column. As can be seen from the results
of Table 1, the columns showed highly variable values of the dynamic modulus, sometimes even on
of Table 1, the columns showed highly variable values of the dynamic modulus, sometimes even on
the same column. This variability confirmed the diversified state of material degradation to which the
the same column. This variability confirmed the diversified state of material degradation to which
columns were subjected in relation with the alteration of the original natural stone used (Figure 7).
the columns were subjected in relation with the alteration of the original natural stone used (Figure
Through Correlation Curve (4), experimentally obtained by Vasconcelos et al. [16], it was possible to
7). Through Correlation Curve (4), experimentally obtained by Vasconcelos et al. [16], it was possible
estimate the compressive strength, fc , of the granitoid rock from the value of the dynamic modulus.
to estimate the compressive strength, fc, of the granitoid rock from the value of the dynamic modulus.
The fc value obtained by the experimental correlation laws represents an estimation (the values of
The fc value obtained by the experimental correlation laws represents an estimation (the values of
which fall within the ranges present in the scientific literature), pending a better confirmation from
which fall within the ranges present in the scientific literature), pending a better confirmation from
subsequent tests that will be carried out in the laboratory. These tests will provide the compression
subsequent tests that will be carried out in the laboratory. These tests will provide the compression
stress results on samples taken from erratic stones present in the area, made of the same material as
that of the columns. 39
𝑓 = 0.0407 𝑣 − 36.31 (4)

Appl. Sci. 2020, 10, 2673
stress results on samples taken from erratic stones present in the area, made of the same material as
that of the columns.
fc = 0.0407 v − 36.31 (4)
Table 1. Values of dynamic modulus by ultrasonic tests on Via Crucis portico columns.
Column Test Point Height (cm) v (m/s) Ed (MPa) fc (MPa) 2

1 57 3269 25,878 97
1 2 98 2503 15,170 66
3 230 2805 19064 78
1 65 3411 28,183 103
2 2 123 2466 14,728 64
3 223 2494 15,067 65
1 70 2910 20,506 82
3
2 230 2939 25,406 96
1 60 4024 39,219 127
4
2 232 3908 36,985 123
1 68 2724 17,972 75
5
2 228 2738 18,156 75
6 1 55 2766 18,525 76
1 86 3402 28,033 102
7
2 206 2948 21,046 84
1 86 2774 18,644 77
8
2 229 3223 25,158 95
9 1 146 2495 15,080 65
1 100 2522 15,406 66
10
2 228 2756 18,402 76
1 101 2698 17,627 73
11
2 204 2773 18,629 77
1 90 2965 21,300 84
12
2 190 3178 24,468 93
1 35 4638 52,111 1 152 1
13 2 130 2840 19,532 79
3 224 3149 24,024 92
1 Basic element of the column made of different stone materials of better quality. 2 Estimated according to the
correlation curve of Vasconcelos et al. [13].
3.2. Impact Tests

The purpose of the impact test was to verify the surface resistance of the stone elements, considering
their surface degradation. The impact test, when performed on homogeneous material such as concrete
or stone, can provide through Equation (2) values of the elastic modulus completely comparable to
those obtained by the ultrasonic technique [29]. It should be remembered that the value obtained with
the impact test represents a static elastic modulus (Es ), the conversion of which into a dynamic value
deserves separate research considering the many different formulas present in scientific literature for
the different types of stones and rocks [32]. To obtain a correct and reliable correlation formula for
each type of stone, a large number of tests must be performed in the laboratory and correlated with
non-destructive tests; this correlation could then be calibrated as a suitable diagnostic technology to be
applied on specific architectural elements. However, since the test is based on an energetic evaluation
relative to the impact on the surface of the element (Figure 8), in presence of a surface degradation,
like the one found in situ, it does not provide elastic modulus values that can be correlated with those
obtained by ultrasonic tests.
40
Appl. Sci. 2020, 10, 2673
(a) (b)
Figure 7. Different types of surface material degradation on the columns of the Via Crucis portico:
(a) erosion; (b) exfoliation.
Figure 8. Impact test carried out on a column of Via Crucis portico.
Figure 9 shows, for example, the time–force curves between two different points of the same test
on column 3. In presence of surface degradation, the curves are characterized by lower peak force and
wider response times. The ratio between the returned A2 and supplied energy A1 (Figure 6a) provided
indications regarding the detachment of the degraded matter crust [17,33].
41
Appl. Sci. 2020, 10, 2673
–
Figure 9. Time–force diagrams of third point of column 3 (I = point inside of the portico; O = outside of
Figure 9. Time–force diagrams of third point of column 3 (I = point inside of the portico; O = outside
the portico).
of the portico).
With reference to the values of the dynamic elastic modulus, Figure 10 reports, for example,
the dispersion of the data collected in the impact tests on the columns at Ghiffa and the same comparison
obtained from the similar type of homogeneous and non-degraded stone tested in the laboratory.
(a) (b)
Figure
Figure 10.
10. Gaussian distribution E
Gaussian distribution Ess of
ofnon-degraded
non-degraded stone
stone (a)
(a) and
and of
of the
the same
same type
type of
of stone
stone degraded
degraded
in the columns of the Via Crucis portico at Ghiffa
Ghiffa (b).
(b).
From Figure
From Figure 10,
10, aa significant
significant difference
difference can
can be
be seen
seen between
between the
the two
two Gaussian
Gaussian curves. The one
curves. The one
relating to laboratory tests on a sample of new
relating to laboratory tests on a sample of new stone stone similar to that used in the Ghiffa columns revealed
 similar to that used in the Ghiffa columns
a significantly
revealed lower standard
a significantly lowerdeviation
standardσ than that detected
deviation  thanwith
thatthe impact with
detected tests performed
the impactintests
situ
on the columns
performed in situofon
the Via
the Crucis of
columns portico.
the ViaThe differences
Crucis portico. are
Thedue to the considerable
differences are due to thediversification
considerable
of the material degradation affecting the surface of the columns. Similarly, Table
diversification of the material degradation affecting the surface of the columns. Similarly, 2 shows theTable
results2
provided by impact tests on the new stone in the laboratory and on the old in situ columns
shows the results provided by impact tests on the new stone in the laboratory and on the old in situ of the Via
Crucis portico.
columns of the Via Crucis portico.
Table 2. Impact tests on new and old granitic stones.


Stone Es average (MPa)  (MPa)
New stone in the laboratory 34,711 843
Old stone of the columns 6394 5100
4. Discussion 42
Appl. Sci. 2020, 10, 2673
Table 2. Impact tests on new and old granitic stones.
Stone Es average (MPa) σ (MPa)

New stone in the laboratory 34,711 843
Old stone of the columns 6394 5100
4. Discussion
The in situ diagnostic tests and their comparison in the laboratory show how dispersive the results
of the ultrasonic and impact tests can be when applied to architectural elements in a degraded state.
The surface roughness of ancient in situ elements affects the impact test results, not allowing a good
correlation with the values of the dynamic elastic modulus obtained from the ultrasonic tests. On the
contrary, the same tests performed in the laboratory on a sample of new and unaltered stone show
Appl.
a verySci.good
2020, 10, x FOR PEER
correlation REVIEW the dynamic elastic modulus obtained from the ultrasonic tests
between 12 of
and16
the static modulus obtained from the impact test, both with a Gaussian curve characterized by very
The in situ diagnostic tests and their comparison in the laboratory show how dispersive the
small standard deviations.
results of the ultrasonic and impact tests can be when applied to architectural elements in a degraded
The ultrasonic tests recorded speed waves average values of about 3000 m/s (Table 3), which the
state. The surface roughness of ancient in situ elements affects the impact test results, not allowing a
scientific and experimental literature [16,28] classifies as medium quality material. However, there is
good correlation with the values of the dynamic elastic modulus obtained from the ultrasonic tests.
a high variability of the values derived from the material degradation level (not excluding the original
On the contrary, the same tests performed in the laboratory on a sample of new and unaltered stone
alteration of the quarried material, from upper part of the extraction site, or by single big rocks like
show a very good correlation between the dynamic elastic modulus obtained from the ultrasonic tests
the “trovanti”), from possible detachments or exfoliation of surface material. The speeds measured
and the static modulus obtained from the impact test, both with a Gaussian curve characterized by
on the same type of not degraded stone obtained from the quarry detected higher values (albeit with
very small standard deviations.
little difference) and statistically more stable speed values (Table 3). The standard deviation value was
The ultrasonic tests recorded speed waves average values of about 3000 m/s (Table 3), which the
representative of the material degradation of the stone surface on which the ultrasonic and impact
scientific and experimental literature [16,28] classifies as medium quality material. However, there is
tests were performed. In this case, both were characterized by a high value of the standard deviation
a high variability of the values derived from the material degradation level (not excluding the original
due to the diversified level of material degradation that characterizes the columns of the Via Crucis
alteration of the quarried material, from upper part of the extraction site, or by single big rocks like
portico (Figures 10b and 11b). Vice versa, as is visible in Tables 2 and 3 and in Figures 10a and 11a,
the “trovanti”), from possible detachments or exfoliation of surface material. The speeds measured
corresponding to the same stone, though intact and without any material alteration, there was a very
on the same type of not degraded stone obtained from the quarry detected higher values (albeit with
low value of the standard deviation and a good correlation between the values of the elastic modulus
little difference) and statistically more stable speed values (Table 3). The standard deviation value
obtained from both types of non-destructive tests.
was representative of the material degradation of the stone surface on which the ultrasonic and
impact tests were performed.Table In this case, bothtests
3. Ultrasonic were characterized
on new by a high value of the standard
and old stones.
deviation d
Stone Vaverage (m/s) Ed average (MPa) σ (MPa)
New stone in the laboratory 3498 29,640 611
Old stone of the columns 2961 21,623 6183
(a) (b)
Figure
Figure 11.
11. Gaussian
Gaussian distribution Edd of
ofnon-degraded
non-degraded stone
stone (a)
(a) and
and of
of the
the same
same type
type of
of stone
stone degraded
degraded
in the columns of the Via Crucis portico
portico at
at Ghiffa
Ghiffa (b).
(b).
Table 3. Ultrasonic tests on new and old stones.
Stone 43 (m/s)
Vaverage Ed average (MPa)  (MPa)
New stone in the laboratory 3498 29,640 611
Old stone of the columns 2961 21,623 6183
Appl. Sci. 2020, 10, 2673
5. Conclusions
The experimental research carried out in situ to determine the residual strength of the degraded
stone columns of the Via Crucis portico at the Sacro Monte of Ghiffa is a first step in a research path that
aims to calibrating non-destructive techniques for the diagnosis of heritage monument degradation.
Ultrasonic and impact techniques were used, critically analyzing their correlation in the presence of
surface material degradation. The ultrasonic test highlighted the variability of the dynamic elastic
modulus within the same columns, while the impact test highlighted the local points of surface material
degradation. Many other tests in the laboratory and in situ will be needed, on different types of stones,
to correlate the values of dynamic elastic modulus obtained by the ultrasonic technique with the static
one that resulted from the impact tests. Subsequent laboratory tests will be also planned to calibrate
the impact test applied to different levels of surface material damage, in order to find correlations with
the values of dynamic elastic modulus obtained from ultrasonic tests.
Author Contributions: Conceptualization, A.G., M.Z., S.F. and G.L.; In situ tests, S.F., M.Z. and A.G.; Writing A.G.,
S.F.; Supervision, M.Z. and G.L.; Project administration, M.Z. All authors have read and agreed to the published
version of the manuscript.
Funding: This research was funded by MAIN10ANCE Interreg Project.
Acknowledgments: The authors would like to thank Giovanni Bricca and Franco Grindatto for their valuable
technical support during the execution of the in situ tests. The authors also wish to thank Josè Delgado Rodrigues
for his precious visit and petrographic consultancy at the Sacro Monte of Ghiffa.
Appendix A
’
Hertz’s theory of the elastic collision between two bodies in the absence of friction is modeled
according tothe scheme in Figure A1, where a sphere of radius R impacts with velocity v0 penetrating
with a depth δ in an elastic semi-infinite space [27]:
Figure A1. Impact between the moving mass and the elastic semi-infinite surface.
The depth of indentation is:

!1
9F2 3
δ= (A1)
16RE∗21⁄3
9𝐹 2
The applied force F is related to the δdisplacement
= ( )
16𝑅𝐸 ∗2 δ by:
4 1 ∗ 3 3
F = R2 E δ 2 = Kδ 2 (A2)
3
4 1 3 3
𝐹 = 𝑅 ⁄2 𝐸 ∗ δ ⁄2 = 𝐾δ ⁄2
where 3
1 1 − ν20 1 − ν21
= + (A3)
E∗ E0 E1
1 1 − ν20 1 − ν12
= +
𝐸∗ 𝐸0 𝐸1
  Poisson’s
44
Appl. Sci. 2020, 10, 2673
and ν0 , ν1 , E0 and E1 are, respectively, Poisson’s ratios and the elastic moduli associated with the sphere
(0) and the semi-infinite space (1).
Considering that the sphere impacts on the surface in the initial time with a speed v0 , and that v1
is the speed in the next instant, the following equation can be written:
dv d2 δ
m = m 2 = −F, (A4)
dt dt
and from (A1) and from (A4):
d2 δ 4 1 3
2
= − R 2 E∗ δ 2 .
m (A5)
dt 3
Through integration, the following can be achieved:
 !2  5
1  2 dδ  2Kδ 2
v − = (A6)
2 0 dt  5m
where K is the value indicated in (A2).

dδ
The maximum depth of indentation δ* occurs in the case of dt = 0, for which:
 2   25
 5mv20  5  15mv20 
δ = 
∗  =   . (A7)
4K   1
16E∗ R 2
The diagram (F, t) in the compression phase can be obtained by means of a second integration:
Z
δ∗ d(δ/δ∗ )
t=
v0 5
1 . (A8)
∗ 2 2
1 − (δ/δ )
The instant t*, corresponding to the maximum depth of the deformation δ∗ during the impact,
referred to as time t1 in Figure 6a, is given by the following equation:
Z 1
δ∗ d(δ/δ∗ )
t∗ = 1
= 1, 435 m2 /RE∗2 v0 5 . (A9)
v0 5 2
1 − (δ/δ∗ ) 2
The overall time T of the whole impact is given by:

1
T = 2t∗ = 2.87 m2 /RE∗2 v0 5 . (A10)
From the curve (F, t) the experimental value of T is obtained, which, when inserted into (A10),
provides the value of E* (see Equation (3)). Finally, using Equation (2), the value of the elastic modulus
Es can be calculated.
References
1. Formia, A.; Serra, C.L.; Zerbinatti, M.; Tulliani, J.M. The plasters of the Sacro Monte of Varallo Sesia. From the
characterization to the proposition of a restorative mix. Case Stud. Constr. Mater. 2014, 1, 46–52. [CrossRef]
2. Cizer, Ö.; Schueremans, L.; Serre, G.; Janssens, E.; Van Balen, K. Assessment of the compatibility of repair
mortars in restoration projects. Adv. Mater. Res. 2010, 133–134, 1071–1076. [CrossRef]
3. Wiggins, D.E.; Klemm, A.J. Compatible repairs to stone built structures. Proc. Inst. Civ. Eng. Eng. Hist. Herit.
2014, 167, 196–207. [CrossRef]
45
Appl. Sci. 2020, 10, 2673
4. Grazzini, A.; Zerbinatti, M.; Fasana, S. Mechanical characterization of mortars used in the restoration of
historical buildings: An operative atlas for maintenance and conservation. IOP Conf. Ser. Mater. Sci. Eng.
2019, 629, 1–7. [CrossRef]
5. Isebaert, A.; Van Parys, L.; Cnudde, V. Composition and compatibility requirements of mineral repair mortars
for stone—A review. Const. Build. Mater. 2014, 59, 39–50. [CrossRef]
6. Binda, L.; Gambarotta, L.; Lagomarsino, S.; Modena, C. A multilevel approach to the damage assessment and
the seismic improvement of masonry buildings in Italy. In Seismic Damage to Masonry Buildings; Bernardini, A.,
Ed.; Balkema: Rotterdam, Holland, 1999; pp. 179–194.
7. Binda, L.; Cantini, L.; Tedeschi, C. Diagnosis of Historic Masonry Structures Using Non-Destructive
Techniques. In Nondestructive Testing of Materials and Structures; Buyukozturk, O., Tasdemir, M.A., Gunes, O.,
Akkaya, Y., Eds.; Springer Netherlands: Heidelberg, Germany, 2013; pp. 1089–1102. [CrossRef]
8. Grazzini, A. The important role of diagnostics for the structural recovery of historic buildings. Int. J. Arch.
Eng. Technol. 2019, 6, 7–16. [CrossRef]
9. Invernizzi, S.; Lacidogna, G.; Carpinteri, A. Structural monitoring and assessment of an ancient masonry
tower. Eng. Fract. Mech. 2019, 210, 429–443. [CrossRef]
10. Carpinteri, A.; Lacidogna, G. Damage monitoring of an historical masonry building by the acoustic emission
technique. Mater. Struct. 2006, 39, 161–167. [CrossRef]
11. Carpinteri, A.; Lacidogna, G.; Pugno, N. Structural damage diagnosis and life-time assessment by acoustic
emission monitoring. Eng. Fract. Mech. 2007, 74, 273–289. [CrossRef]
12. Carpinteri, A.; Invernizzi, S.; Lacidogna, G. Historical brick-masonry subjected to double flat-jack test:
Acoustic emissions and scale effects on cracking density. Constr. Build. Mater. 2009, 23, 2813–2820. [CrossRef]
13. Cantini, L.; Bonavita, A.; Parisi, M.A.; Tardini, C. Historical analysis and diagnostic investigations in the
knowledge acquisition path for architectural heritage. In Proceedings of the Congress in Structural Analysis
of Historical Constructions—Anamnesis, Diagnosis, Therapy, Controls, Leuven, Belgium, 13–15 September 2016;
Van Balen, V., Ed.; Taylor & Francis Group: London, UK, 2016.
14. Grazzini, A. Sonic and Impact Test for Structural Assessment of Historical Masonry. Appl. Sci. 2019, 9, 5148.
[CrossRef]
15. Chastre, C.; Ludovico-Marques, M. Nondestructive testing methodology to assess the conservation of
historic stone buildings and monuments. In Handbook of Materials Failure Analysis with Case Studies from the
Construction Industries; Hamdy Makhlouf, A.S., Aliofkhazraei, M., Eds.; Elsevier: New York, NY, USA, 2018;
pp. 255–294, ISBN 978-0-08-101928-3. [CrossRef]
16. Vasconcelos, G.; Lourenço, P.B.; Alves, C.S.A.; Pamplona, G. Prediction of the mechanical properties of
granites by ultrasonic pulse velocity and Schmidt hammer hardness. In Proceedings of the 10th North American
Masonry Conference, St. Louis, MO, USA, 3–6 June 2007; Myers, B., Ed.; Masonry Society: Longmont,
CO, USA, 2007.
17. Grazzini, A. In Situ Analysis of Plaster Detachment by Impact Tests. Appl. Sci. 2019, 9, 258. [CrossRef]
18. The Sacred Mounts of Piedmont and of Lombardy. Available online: www.sacrimonti.net (accessed on
10 February 2020).
19. Gonçalves, T.D.; Pel, L.; Delgado Rodrigues, J. Influence of paints on drying and salt distribution processes
in porous building materials. Constr. Build. Mater. 2009, 23, 1751–1759. [CrossRef]
20. Bocca, P.; Grazzini, A.; Masera, D.; Alberto, A.; Valente, S. Mechanical interaction between historical brick
and repair mortar: Experimental and numerical tests. J. Phys. Conf. Ser. 2011, 305, 1–10. [CrossRef]
21. Bocca, P.; Valente, S.; Grazzini, A.; Alberto, A. Detachment analysis of dehumidified repair mortars applied
to historical masonry walls. Int. J. Arch. Herit. 2014, 8, 336–348. [CrossRef]
22. Grazzini, A.; Lacidogna, G.; Accornero, F. Delamination of plasters applied to historical masonry walls:
Analysis by acoustic emission technique and numerical model. IPO Conf. Ser. Mater. Sci. Eng. 2018, 372, 1–7.
[CrossRef]
23. Accornero, F.; Invernizzi, S.; Lacidogna, G.; Carpinteri, A. The Sacred Mountain of Varallo renaissance
complex in Italy: Damage analysis of decorated surfaces and structural supports. In Acoustic, Electromagnetic,
Neutron Emissions from Fracture and Earthquakes, 1st ed.; Carpinteri, A., Lacidogna, G., Manuello, A., Eds.;
Springer: Heidelberg, Germany, 2015; pp. 249–264. [CrossRef]
24. Delgado Rodrigues, J. Conservation of stone monuments. From diagnostic to practice. In Proceedings of the
MINBAR AL JAMIAA n◦ 7—RIPAM 2005, Mekés, Maroc, 26–28 September 2005; pp. 287–295.
46
Appl. Sci. 2020, 10, 2673
25. Delgado Rodrigues, J.; Ferreira Pinto, A.; Rodrigues da Costa, D. Tracing of decay profiles and evaluation of
stone treatments by means of microdrilling techniques. J. Cult. Herit. 2002, 3, 117–125. [CrossRef]
26. ASTM D2845 1995. Standard Test Method for Laboratory Determination of Pulse Velocities and Ultrasonic Elastic
Constants of Rock; American Society for Testing Materials: West Conshohocken, PA, USA, 1995.
27. UNI EN 12504-4: 2005. Prove sul Calcestruzzo nelle Strutture—Parte 4: Determinazione della Velocità di
Propagazione Degli Impulsi Ultrasonici (in Italian); Ente Nazionale Italiano di Unificazione: Milan, Italy, 2005.
28. Domede, N.; Parent, T.; Sellier, A. Mechanical behaviour of granite: A compilation, analysis and correlation
of data from around the world. Eur. J. Environ. Civ. Eng. 2019, 23, 193–211. [CrossRef]
29. Bocca, P.; Scavia, C. The impulse method for the evaluation of concrete elastic characteristics. In Proceedings
of the 9th International Conference on Experimental Mechanics, Copenhagen, Denmark, 20–24 August 1990.
30. Johnson, K.L. Contact Mechanics; Cambridge University Press: Cambridge, UK, 1985.
31. Deresiewicz, H. A note on Hertz Impact. Acta Mech. 1968, 6, 110–112. [CrossRef]
32. Dhawan, K.R.; Muralidhar, B. Relationship between static and two types of dynamic moduli for different
rocks. Ind. Getech. J. 2015, 45, 341–348. [CrossRef]
33. Bocca, P.; Carpinteri, A.; Valente, S. On the applicability of fracture mechanics to masonry. In Proceedings of
the 8th International Brick/Block Masonry Conference, Dublin, Ireland, 19–21 September 1988.
47
applied
sciences
Article
Non-Invasive Identification of Vulnerability Elements
in Existing Buildings and Their Visualization in the
BIM Model for Better Project Management: The Case
Study of Cuccagna Farmhouse
Elsa Garavaglia 1, * , Anna Anzani 2 , Fabio Maroldi 1 and Fabio Vanerio 1
1 Department of Civil and Environmental Engineering, Politecnico di Milano, 20133 Milano, Italy;
[email protected] (F.M.); [email protected] (F.V.)
2 Department of Design, Politecnico di Milano, 20158 Milano, Italy; [email protected]

Received: 21 February 2020; Accepted: 15 March 2020; Published: 20 March 2020
Featured Application: The procedure could be applied, on a municipal or territorial scale, in order
to proceed with a rapid assessment of the vulnerability of the existing historical heritage.
Abstract: Due to the conjunction between the European and African plates, complex tectonic
phenomena take place in the Mediterranean basin. These phenomena cause more or less violent
seismic resentments in the countries facing the basin itself. The diffused built historical heritage,
characteristic of villages in the Mediterranean countries, is the most vulnerable toward seismic action,
and in case of a catastrophic event can cause the loss of human lives. In Italy, the protection of
historic buildings is a significant issue, and many regions promoted policies to ensure the safety of the
diffused built heritage. Research groups work in synergy to develop procedures for the vulnerability
assessment of existing buildings and to define appropriate action plans. This research presents a little
or not at all invasive procedure for investigating vulnerability. This procedure is easily replicable and
able to support techniques already in use with innovative aspects such as laser scanning of the entire
complex and visual identification of vulnerable elements through the BIM (building information
modeling) methodology. The procedure applicability is shown in the study of a Milanese farmhouse
that has been financed by Fondazione CARIPLO, Bandi 2017 Arte e Cultura-Beni culturali a rischio,
Project PRE.CU.R.S.OR.
Keywords: existing buildings; seismic vulnerability assessment; non-invasive experimental technics;

BIM models
1. Introduction
The Mediterranean basin, a cradle of many civilizations and a repository of a remarkable built
heritage, is located at the junction between the European and African tectonic plates. This conjunction
constitutes the so-called Faglia Giulia that runs along the entire Italian Apennine ridge (INGV Database
of individual seismogenetic sources DISS_v3, http://diss.rm.ingv.it/diss/), a set of more or less long
faults, often almost parallel to each other and juxtaposed, sometimes transversal, whose movements
are the cause of seismic resentments in the peninsula. In recent years, the land movements recorded on
the Italian Apennine ridge have caused considerable losses in the historical heritage of the affected
regions, not to mention the consequent loss of human lives due to the induced collapses. The reason
for these disasters is linked to the high vulnerability that characterizes the Italian historical heritage, as
49
Appl. Sci. 2020, 10, 2119
well as the historical heritage of all countries in the Mediterranean basin, which is not exempt from
vulnerability to seismic actions.
A campaign to secure the diffused built heritage requires a careful assessment of the buildings’
vulnerability and an economic commitment on the part of their owners. Since 2014, the ReLUIS
Network of Seismic Engineering University Laboratories (http://www.reluis.it/) has been working to
survey the existing heritage and achieve a reliable vulnerability classification. With the creation of
the Casa Italia department (http://www.casaitalia.governo.it/it/) in 2016, a campaign was launched
to raise public and administrative awareness of the need to secure public and private buildings, also
encouraged by the introduction of possible tax benefits.
In recent years, to support and integrate government initiatives, seismic research has developed
numerous investigation methods aimed at the census of aggregate buildings and the subsequent
identification of minimum units and structural units on which base the seismic behavior can be analyzed.
In complex urban situations, such as those present in historical centers, this census is not easy.
In recent decades, numerous studies have adopted Earth geo-referencing (GIS) technology. Such
tool allows association of an object not only to a geo-referenced location that is easy to browse,
but also to some of its peculiar characteristics. In the seismic field, this methodology has made it
possible to associate each geo-referenced object with the evaluated level of vulnerability. Consequently,
vulnerability maps of public and private buildings of entire municipalities could be drawn up [1–12].
To achieve both more detailed investigations on buildings seismic behavior and 3D representation
of the level of vulnerability characterizing some construction elements, recent research focuses on the
use of Building Information Modeling (BIM).
The BIM methodology is used in the identification phase of vulnerable structural elements; as
a possible self-diagnosis process in the pre- and post-earthquake phase if associated with a health
monitoring system; as a tool for the creation of an emergency hub to control and shutdown particular
facilities (e.g., gas pipes) in the event of an earthquake [13].
As for the use of BIM in the field of seismic vulnerability, the structural analysis, seismic response
assessment, and fragility curves associated with different levels of damage are implemented in a BIM
environment in order to obtain a 3D graphical display of vulnerable or damaged elements as well
as their associated level of vulnerability or damage. A cost analysis also allows the development of
clearer intervention strategies for both the designers and the clients [13–16].
The latter also includes the research proposal presented in this paper. The investigation procedure,
replicable on different buildings, is based on a quick assessment of the vulnerability obtained with:
• The application of reliable methodologies already known in the literature,

• The use of laser scanner surveys,
• The implementation of the results in a BIM environment.
The implementation of the results in a BIM environment aims to obtain an effective graphical
representation of the different aspects of building vulnerability, their severity and possible co-presence,
thanks to the opportunity offered by the BIM methodology to perform queries on probable occurrence
scenarios (in this case vulnerability scenarios), including the level of vulnerability found on each
interrogated element. This methodology is a first approach to the graphical representation of historical
buildings vulnerability. It can already be useful for planning a possible targeted diagnostic campaign
and for designing suitable interventions of seismic adaptation or improvement addressed to solve the
problems that have emerged. The implementation of the economic evaluation of the diagnostic and
seismic improvement interventions associated with every possible investigation scenario will be the
future objective of this research.
The proposed procedure has been applied to safeguard Cascina Cuccagna, a historic farmhouse
in the urban fabric of Milan city. The safety works have been partially financed by Fondazione Cariplo
in the framework of the calls for proposals Art and Culture-Cultural Heritage at risk.
50
Appl. Sci. 2020, 10, 2119
The paper is made of three sections: the first section presents the investigation procedure for the
vulnerability assessment of historic buildings. The second section introduces Building Information
Modeling (BIM) as a method of graphic detection of building vulnerabilities and of their impact on
the buildings’ overall ability to withstand an earthquake of a certain magnitude. The third section
presents an application of the proposed method to Cascina Cuccagna, a historic building located in the
city of Milan, Italy.
2. The Procedure
The knowledge of a building and of its history is an essential prerequisite for any safeguarding
action; therefore, all possible historical documentation summarizing the different construction phases
of the building should be retrieved and, if possible, their typical construction techniques.
Another form of knowledge comes from a detailed survey of the building in its current state, to be
compared with a previous survey, if available.
The next phase of knowledge will require visual and photographic investigations of the building
and the outline of its possible crack pattern. The identification of the loads acting on the structural
elements will also be important, as well as the evaluation of the mechanical qualities of the masonry,
wooden elements, and floors.
The results of this first phase of investigation provide enough elements to face an initial and rapid
assessment of possible vulnerabilities.
2.1. Historical Survey of the Building

In order for an intervention to be structurally effective, compatible with the existing built
patrimony, and respectful of its material, constructive and documentary characters, an articulated and
interdisciplinary survey has to be carried out. A fundamental contribution to the study of an existing
building or a constructed centre is the historical survey, which should take account of indirect and
direct sources of information.
Analysis of indirect sources include cataloguing of the data collected through documentaries,
archives, and bibliographic information, with the aim to reconstruct the evolution of the complex,
its building, and transformation phases and its load history as an history of the earthquakes that
occurred in the past. When known, the seismic history of the building is particularly relevant with
respect to the safety and intervention aspects. Of great relevance are also the iconographic elements
(views, ancient documents) giving information on the historical housing, allowing the evaluation of
the transformations of a territory and the comparison with its present configuration.
A compared study is fundamental, especially when the research objects are minor centers or
rural architectures. If the documental (indirect) sources do not allow a complete reconstruction of
the complex evolution, the historical information coming from the documents should be verified in
situ and accompanied by a direct analysis of the buildings. Direct sources, which sometimes are
the sole available if historic documents have been lost, come from the observation of the building
itself as a data source and include various methods for approaching the knowledge of the building
and its chronological phases. They can take advantage from a multidisciplinary approach which
critically combines the information coming from the detection of the adopted constructive techniques
and materials, traces of previous interventions, epigraphic and heraldic documentations if present,
decorative apparatus, and a stratigraphic survey when possible.
2.2. Geometry and Material Knowledge: Help Comes from the Laser Scanner Survey
In the vulnerability assessment, the laser scanner is certainly a useful type of survey. This
methodology is able to provide a 3D survey of the entire building, its precision depending on the finesse
of the instrument and the density of the measured points. The output is a point cloud that represents
an operational guide for computer aided design (CAD) and BIM graphic rendering programs capable
51
Appl. Sci. 2020, 10, 2119
ofAppl.
transforming
Sci. 2020, 10, it intoPEER
x FOR moreREVIEW
classic, very detailed, reproductions such as plans, elevations and sections 4 of 24
(Figure 1).
Aswe
As we will
will see, thethe laser
laserscanner
scannersurvey
surveyalsoalsoallows thethe
allows construction
construction of aof
3Da model
3D modelof theofentire
the
building
entire that that
building serves as aasbasis
serves forfor
a basis thetheBuilding
BuildingInformation
InformationModeling
Modeling(BIM),
(BIM), here
here used in inthethe
identification of vulnerable elements and their associated degree of vulnerability.
identification of vulnerable elements and their associated degree of vulnerability.
Theknowledge
The knowledgeprocess processofofananexisting
existingbuilding
buildingshould
shouldnecessarily
necessarilyinvolve
involvethe thedetection
detectionofofits its
constructivematerials
constructive materials and techniques,
techniques, aimedaimedatatthe thereduction
reduction and
and control
controlof physical,
of physical, chemical,
chemical,and
mechanical
and mechanical material
material decay.
decay.AsAs mentioned,
mentioned,graphicgraphicrepresentation
representationisis aa fundamental
fundamental tool in inthethe
preliminaryphase
preliminary phase of of knowledge
knowledge as well
as well as anas an accurate
accurate surveysurvey of the building
of the building in all its
in all its formal, formal,
material,
material,
and and degradation
degradation aspects, as aspects,
well as in asthe
well as in the subsequent
subsequent preservationpreservation
project. project.
Thematerial
The materialsurvey
surveyisisaafundamental
fundamentaltool toolofofinquiry
inquiryinto
intothe
thebuilding
buildingandandshould
shouldbe becoupled
coupled
withthe
with thehistorical
historicalresearch
researchandandthethegeometric
geometricsurvey.
survey.ItItshould
shouldthoroughly
thoroughlystudystudythe theconstruction
construction
consistency,the
consistency, thematerial
materialorigin
originand
andthe theproduction
productiontechnique,
technique,thetheopening
openingfeatures
featureswhich
whichcancanhelp
help
datingthe
dating theconstruction
constructionphases,phases,and
andshould
shouldnotenoteits
itsfeatures
featuresand
andpeculiarities
peculiaritiessuch
suchasasalignments,
alignments,wallwall
thicknessdiversities,
thickness diversities, discontinuities,
discontinuities, and and variations
variations of the of
wallthe wall apparatus,
apparatus, which arewhichalwaysare always
indicative
ofindicative of somehistorical-artistic
some significant significant historical-artistic
or simply ofor simply of technical-constructive
technical-constructive choice. choice.
(a)
(b)
Figure 1. CAD
Figure (Computer-Aided
1. CAD Design)
(Computer-Aided geometric
Design) survey
geometric obtained
survey fromfrom
obtained a points cloud:
a points (a) cloud
cloud: resolution;
(a) cloud
resolution; (b) CAD re-drawing of the transversal section.
(b) CAD re-drawing of the transversal section.
2.3. Redrawing the Crack and Deformation Pattern
2.3. Redrawing the Crack and Deformation Pattern
The building assessment requires the architectural redrawing of the complex starting from a
The building
traditional survey orassessment requires
from the points cloud theobtained
architectural redrawing
from the of theinstrumental
laser scanner complex starting
survey.from
Thisa
traditional survey or from
redrawing could highlight: the points cloud obtained from the laser scanner instrumental survey. This
redrawing could highlight:
• Possible deformation anomalies in the horizontal elements, symptom of high stress, decrease in
Possible deformation
 load-bearing anomalies in
capacity, constraints the horizontal
failure, etc.; elements, symptom of high stress, decrease in
load-bearing capacity, constraints failure, etc.;
 Possible out of plane of vertical elements, symptom of not contrasted horizontal thrusts, lack of
interlocking between walls, local instability of unconnected walls, etc.;
 Presence of not contrasted thrusts in the roof 52 or incidence of vaults on load-bearing walls;
 Misalignments between walls and between inter-stories.
Appl. Sci. 2020, 10, 2119
• Possible out of plane of vertical elements, symptom of not contrasted horizontal thrusts, lack of
interlocking between walls, local instability of unconnected walls, etc.;
• Presence of not contrasted thrusts in the roof or incidence of vaults on load-bearing walls;
• Misalignments between walls and between inter-stories.
These anomalies could threaten the building safety and require a visual check and a more accurate
photographic survey.
For all the different parts of the building, the survey should include drawings of the crack patterns,
a necessary step to formulate hypotheses on its possible causes and to suggest possible diagnostic
investigations to be carried out for a more thorough verification of the problems found.
In the case of a historical asset, a material and pathological survey should also be carried out in
order to plan preservation interventions compatible with the historicity of the building.
2.4. Mechanical Characteristics of Structural Elements and Loads Analysis

Certainly, the knowledge level reached with rapid investigations and non-destructive tests can
only be limited, but it already provides the professional with information on where and how to obtain
more precise knowledge levels.
The loads analysis should be carried out taking into account the actions to which the structure
is currently subjected, considering any change in its intended use and the replacement of structural
elements, floors, roofing membranes, or other transformations undergone by the building during its
service life.
If a detailed examination of the flooring system is not possible, the acting loads can be estimated
according to the constructive typologies used at the building time or due to whatever change
has occurred.
For better knowledge of the mechanical characteristics of the structural elements, more or less
invasive experimental tests both in the laboratory and on site will be necessary. The connection
between floors and walls, and between walls perpendicular to each other should be investigated
on-site. However, the damage level of wooden elements can already be detected through a visual
investigation of their crack and deformation patterns, of the constraints quality (if detectable), and of
any rottenness, biological, and/or parasitic attack. This could be part of the qualitative vulnerability
evaluation and provide useful indications for a selective and targeted diagnosis.
The masonry mechanical qualities will also have to be evaluated experimentally, but reasonable
hypotheses on the type of masonry texture present in the building can help to establish its quality
and behavior toward in plane and out of plane actions. The historical investigation, associated with
the knowledge of the construction techniques typical of the area, will allow achievement of truthful
assumptions about the type of masonry, while a possible geo-radar investigation could help understand
if the masonry is made of a single or multiple leaves.
A rapid method of characterizing the masonry quality has been developed by Borri et al. [17–20].
Having at their disposal a rich catalogue of masonry typologies typical of Italian regions [21,22]
(https://www.abacomurature.it/), the authors have developed a methodology that, on the basis of a
sample visual survey of the masonry surface, can provide an index of masonry quality. This classifies
masonry into three categories against its possible response to vertical or horizontal—in plane or out of
plane—forces: good (A), medium (B), very bad (C) (Table 1 and Figure 2).
The masonry quality index (IQM) calculation is based on the rules of good building. It takes into
account the units’ quality (type of brick/stone and its geometry), the mortar quality, the presence of
staggered joints, in prospect and section (if visible), and the presence of headers or other forms of
connection of the wall leaves. Depending on their recorded quality, each of these parameters is assigned
a weight. This weight is established on an empirical basis (expert judgment) and on a statistical basis
where possible (Figure 2).
The final evaluation associates the masonry with a certain IQM value as shown in Table 1.
53
Appl. Sci. 2020, 10, 2119
Table 1. Attribution of masonry category.
Actions on the Masonry Class A Class B Class C

Vertical 5 ≤ IQ ≤ 10 2.5 ≤ IQ < 5 0 ≤ IQ < 2.5
Horizontal
7 ≤ IQ ≤ 10 4 < IQ < 7 0 ≤ IQ ≤ 4
perpendicular
Horizontal coplanar 5 < IQ ≤ 10 3 < IQ ≤ 5 0 ≤ IQ ≤ 3
2.5. Initial Vulnerability Assessment

The redrawing of the building in plan and in elevation allows us to: define the surface of each
floor and each inter-storey average height; identify the resistant elements against horizontal actions in
two orthogonal directions (x and y) and their dimensions in plan (total length and average thickness of
the resistant elements in both directions); identify the points of transmission of vertical and horizontal
loads (thrusts of vaults and coverage).
Figure 2. Identification of the masonry quality.
These data are the basis for an initial assessment of the building vulnerability.
Having identified the seismic zone and the parameters that characterize its response spectrum for
the reference limit state, Borri et al. [19,20] propose a methodology that allows control of the overall
building vulnerability on the basis of:
• An expert judgement on the vulnerability level of the entire building formulated on the basis of
the presence, or absence, of 10 possible vulnerability aspects and the severity level found (Table 2).
• Three simplified numerical checks: simplified global check (VG), which allows the definition of
the ground acceleration that could lead to collapse; simplified local check (VL), which allows
the determination of the acceleration of activation of a given mechanism (overturning, wedge
overturning, buckling); simplified static check (VS), which allows the determination of vertical
loads only and the safety factor.
54
Appl. Sci. 2020, 10, 2119
Table 2. C/Q Classification/Quantification) seismic method; structural deficiencies to be considered for

vulnerability identification (synthesis) [20].
N Description of Possible Vulnerable Evidences

1 State of conservation. Extent and severity of significant cracks or serious decay.
Masonry quality. Assessment of compliance with the rule of art and number of walls with poor
2
workmanship.
3 Linkage. Connections or metal tying between outer and inner walls and with floors and roofs.
4 Floors and roofs. Span or insufficient support.
5 Structural regularity (only on stiff floors). Presence of plane irregularities.
Inclusion of a building in a block or a line. Corner or head position, presence of protruding or
6
elevating elements not contrasted by other buildings.
7 Elevation regularity. Mass or stiffness variation between two subsequent levels.
8 Foundation and soil. Evidence of foundation settlements.
Local structural defects. Thrusting elements, unsupported walls or pillars, superfetation,
9
hammering with adjacent buildings.
Vulnerable non-structural elements. Chimneys, parapets, balconies, eaves, veils, badly connected
10
significantly heavy ceilings.
The results of the three simplified checks are compared with the safety levels required by the Italian
NTC2018 standards for the protection of human life limit state (SLV) for the site under examination,
the considered type of building, and the envisaged class of use. These results are expressed in terms
of conventional safety factor (SF). In other words: for each of the three checks, VG, VL, VS, and for
each verified element, there will be a SF factor (e.g., SFVL,xi SFVL,yi factors related to the VL check on
element i in the direction x and y); the minimum safety factor obtained in the check under examination
will be divided by the safety factor present in the standard and referred to the SLV state.
SFV j,min
SFV j (%) = with j = G, L, S, (1)
SFV j, SLV−Law
SFV j (%) expressed in percentage, is the conventional safety factor for the check Vj.
SFV j,min indicates the minimum safety factor among all the safety factors calculated for the check
Vj and for all the elements of the structural system, SFV j, SLV−Law is the safety factor indicated by the
code for the check Vj.
The minimum among all the safety factors SFVj will determine the vulnerability class of the
system [20] n o
SF(%) = min SFV j (%) with j = G, L, S, (2)
The combination of the vulnerability assessments with the minimum safety factor derived from
the three simplified numerical checks results into the association of the building with a low, medium,
high or very high vulnerability class (Table 3).
Table 3. Anti-seismic classification [20].
Safety Assessment for the State of Life Preservation Limit (SLV) A+

Results of the Numerical Assessment
Vulnerability judgment >80% 80%–60% 60%–40% 40%–20% <20%
Low vulnerability A B1 B2 C1 D1
Medium vulnerability B1 B2 C1 C2 D2
High vulnerability B2 C1 C2 D1 D3
Very high vulnerability C1 C2 D1 D2 E
55
Appl. Sci. 2020, 10, 2119
The methodology has been developed within the framework of the POR-ESFR 2007–2013 for
the Umbria Region seismic certification/qualification of the existing asset (Seismic C/Q). Currently
the guidelines for the seismic C/Q have been formally acquired by the Umbria Region and the
Marche Region.
2.6. Non-Destructive Diagnostics

For interventions on buildings, the Italian Standard requires more or less severe safety factors
depending on the knowledge level reached about the typology and details and mechanical characteristics
of the building studied. In order to reduce the severity imposed by the regulatory coefficients, a
diagnostic study should be necessary and the investigations previously described can help not to
disperse resources, highlighting critical points on which possible experimental tests should focus.
High knowledge levels require in-depth and invasive diagnostic investigations, often not
compatible with the available economic resources and the intrinsic value of the asset. Setting
up a non-destructive or minimally invasive testing campaign (sonic and ultrasonic tests, superficial
and deep penetrometric tests, georadar, partial plaster removal, small inspections, etc.) focused on the
areas with the main problems can help contain costs but also reach a higher level of knowledge, as
required by regulations.
2.7. Identification of Vulnerable Elements and the Level of Damage

In the literature, damage classifications for different types of buildings are available: they are
defined from the study of the fragility curves that link the damage probability to the intensity of
the seismic event. However, these classifications mainly concern the probable response of a given
structural typology to a given seismic action [23–25], not the response of a single vulnerable element in
a single building. Therefore, the assessment of the vulnerability level of each element of the building
must follow other paths. In 2000 for the Marche Region, Francesco Doglioni proposed to associate
the forms of specific vulnerability, often found in existing buildings, with six well-defined causes
(Table 4) [26]. This classification can be extended to the entire historical heritage at risk and represents
a very important starting point on which the identification of specific vulnerabilities and an initial
estimate of the propensity to damage of vulnerable elements for an earthquake of a certain magnitude
can be based.
The propensity to damage can be assessed by associating weights to each detected criticality
connected with one of the specific vulnerabilities listed in Table 4. These weights can be established
according to the detected damage level.
Obviously, this step must be subjected to “expert” judgment. In fact, on the basis of an expert
judgment, a certainly subjective, but supported by the results of previous investigations, weight can
be assigned to each of the vulnerability forms found and a “judgment” can be formulated on their
influence on the overall vulnerable behavior of the building. In this way, each of them will be qualified
as absent (0), scarcely influential (1), averagely influential (2), significantly influential (3).
More than one of the specific vulnerabilities listed in Table 4 may be present in the building
elements. Now, if each element were affected by more than one form of specific vulnerability, it would
be interesting to have an immediate graphical feedback of its vulnerabilities [13–16].
The research presented here focuses on the possibility of using the BIM methodology also in this
field, evaluating the effectiveness of a three-dimensional building representation with the graphical
identification of elements affected by one or more forms of vulnerability and the classification of their
level of influence on the total vulnerability of the building through a given range of colors.
56
Appl. Sci. 2020, 10, 2119
Table 4. Classification of specific vulnerability forms in thematic groups attributed to six different
possible causes.
Thematic Groups of Specific Vulnerability Forms Attributed to:
Materials and constructive • Quality of supports and binders

techniques • Masonry adhesion and/or cohesion
• Related to the masonry section:

Geometric and dimensional positioning of the outer leaf
masonry characters • Related to the architectural-structural
elements: slenderness
• Presence of fuel pipes

Elements that reduce the • Interruptions due to plants ducts
masonry cross section • Eaves/rain pipes within the
Initial methods of
1 masonry section
construction
• Slenderness
Structural elements with • Poor thickness
inadequate section • Towering elements
Peculiar openings • Misaligned apertures

distribution • Apertures close to corners
Lack of connection between • Inadequate interlocking

stone elements • Lack of adhesion
Enlargement
• Non-interlocked
Super-elevation • Discontinuous
Closing/opening of apertures • Of heterogeneous material
Building • Demolition of partitions

Subtraction of elements or
2 transformation • Opening of large apertures
masonry parts
processes
Inadequate or dangerous • Unsupported walls

structural solutions • Pillars above vaults
Wall-wall connections • Discontinuities in wall node

Lack of structural
connections and Wall-roof connections
• Non-compensated local thrust
3 role of existing Wall-floor connections • Lack or poor effectiveness of connections
protection
devices • Lacking or ineffective ties
Inadequate or damaged
connection systems • Inadequate buttresses
57
Appl. Sci. 2020, 10, 2119
Table 4. Cont.
Thematic Groups of Specific Vulnerability Forms Attributed to:
• Lack of binder in the joints

• Masonry de-cohesion
Masonry decay • Presence of water
• Decay of stone elements
• Heads rotting
Structural • Decay of secondary elements and/or of
deterioration and Decay of timber elements in the tables
4 the roofs
lack of • General decay
maintenance • Elements deflection
Lack of maintenance of • Exposed outer leaf

masonry and plasters • Plaster efficiency
Lack of maintenance of the • Conditions of the covering layer

roof • Inefficient eaves and pipes
• Cracking
Previous Seismic • Deformations/out of plumb
structural
5 damage not
sufficiently
• Cracking
repaired Static • Deformations/out of plumb
• Reinforced concrete beams with

inadequate sections and/or detached from
the masonry
Recent structural Executed with “modern” • Non-distributed and
6
refurbishment techniques nonhomogeneous injections
• Reinforced plaster non suitably connected
with the masonry
3. Building Information Modeling Applied to Seismic Vulnerability

The 3D design in BIM environment is able to support multidisciplinary collaborative processes
useful in the management and control of the building process; in fact, the BIM modeling allows
optimization of the planning, implementation, and management of the entire building complex from
the architectural, structural, and plant engineering point of view. Each element of the building,
created in BIM environment, has parametric characteristics that can be modified: at each project
implementation, with additions or changes, the entire design system is updated according to the
changes made; this model potentiality allows a flexible management of the entire construction process
from design, to building, to maintenance.
BIM modeling can also be used in the interventions on existing assets. In this case, the graphic
reproduction of an existing building requires input from a base track; this base can be provided through
a points cloud obtained from a laser scanner survey, or a geometric survey file resulting from computer
aided design (CAD) programs.
The 3D model, created as a geometric survey of the building, is then parameterized and reproduced
as a BIM model. The BIM modeling allows association of to each building element (component) and
58
Appl. Sci. 2020, 10, 2119
all its characteristics (geometry, material, mechanical parameters, etc.) in parametric form. Taking
advantage of this BIM environment peculiarity, it seems interesting to associate to each element, in
parametric form, with the vulnerability characteristics resulting from previous investigations.
The first step required by the parameterization procedure is creating an association between the
specific vulnerabilities found and their assigned weights for each individual element of the building.
The purpose of having a graphical resolution of the vulnerabilities requires that each vulnerability
and each weight is associated with a field that identifies the type of specific vulnerability and the
weight it could have on the overall vulnerability of the building. The graphical representation offered
by the BIM modeling is certainly useful to have an immediate picture of the distribution of some
vulnerabilities, and their danger level, throughout the whole building. These graphic representations
can be obtained through targeted queries, a classic approach to work in a BIM environment.
These queries can concern the same specific vulnerability but with different weight: 0 = absent, 1
= light damage, 2 = moderate damage, 3 = severe damage.
• Query01_Querying the same vulnerability:

• “Presence of flue pipes” classified with weight 2.
• “Presence of flue pipes” classified with weight 3.
In the example shown, query01 gives the position of voids inside both load-bearing and
non-load-bearing walls, and the weight associated with this vulnerability according to the recorded
damage level.
Or they can concern two different vulnerabilities
• Query02_Querying two vulnerabilities:

• “Uncompensated localized roof thrust” ranked with weight 3;
• “Unconnected building expansion” classified with weight 2.
Query02, thus defined, gives the critical and significant points for the possible detachment or
partial overturning of the masonry structure.
Another possible query is for three or more vulnerabilities.
• Query03_Querying three or more vulnerabilities:

• “No or poor effectiveness of wall-roof connection” classified with weight 2.
• “Missing or ineffective tie rods” classified with weight 2.
• “Condition of the roof covering” classified with weight 1.
This type of query, for instance, is useful for understanding the state of connection and degradation
of the roof structures with respect to the underlying masonry.
The parametric modeling and the performed queries presented here seem to overcome the
difficulties of a synthetic and global reading of the vulnerabilities present in a building. In fact, the
classification of vulnerabilities is generally reported on spreadsheets and tables that are not easy
or quick to interpret, especially for non-professionals, as the property owners can be. A graphic
representation certainly helps the owner to better understand the problems of his building and to take
informed decisions on the interventions to be carried out.
4. Application of the Proposed Procedure to the Case Study of Cuccagna Farmhouse in Milan
(North Italy)
The methodology presented in this work has been applied in a project of redevelopment and
anti-seismic safety of a historical asset in the city of Milan, Italy.
In 2017, the pilot project PRE.CU.R.S.OR. was the winner of a grant from the Fondazione Cariplo
di Milano in the field of Art and Culture-Cultural Heritage at Risk, a funding line that supports
preservation interventions on assets protected by law. The involved asset is Cascina Cuccagna, one
59
The methodology presented in this work has been applied in a project of redevelopment and
anti-seismic safety of a historical asset in the city of Milan, Italy.
In 2017, the pilot project PRE.CU.R.S.OR. was the winner of a grant from the Fondazione Cariplo
Appl. Sci. 2020, 10, 2119
di Milano in the field of Art and Culture-Cultural Heritage at Risk, a funding line that supports
preservation interventions on assets protected by law. The involved asset is Cascina Cuccagna, one
ofof
thethe rural
rural buildings
buildings still
still existing
existing inin the
the city
city ofof Milan
Milan (Figure
(Figure 3).3). Until
Until about
about the
the middle
middle ofof
thethe 20th
20th
century,
century, farmsteads
farmsteads represented
represented the
the place
place ofof life
life and
and productivity
productivity forfor most
most ofof the
the inhabitants
inhabitants ofof the
the
Lombardplain.
Lombard plain.Still
Stilltoday,
today,they
theyare
area remarkable
a remarkablearchitectural
architecturalresource
resourceandandoften,
often,asasinin the
the case
case ofof
Cascina
Cascina Cuccagna,
Cuccagna, they
they areare used
used with
with different
different functions
functions from
from thethe original
original ones.
ones.
NORTH
(a) (b)
Figure
Figure 3. 3. Cascina
Cascina Cuccagna,(a)
Cuccagna, overallview
(a) overall view and
and (b)(b)ground
groundfloor
floor plan.
plan.
Cascina
CascinaCuccagna
Cuccagnarepresents
representsananideal
idealscenario
scenarioforfora areplicable
replicable“pilot
“pilotproject”
project”ofofprogressive
progressive
improvement
improvementofofthe theearthquake
earthquakeresistance
resistanceconditions
conditionsonon historical
historicalbuildings.
buildings.InInfact,fact,from
fromthe the
construction point of view, its model is highly repetitive in Lombardy and more
construction point of view, its model is highly repetitive in Lombardy and more generally in northerngenerally in northern
Italy.ItItisislocated
Italy. locatedinina acontext
contextofofmedium-low
medium-lowseismicity
seismicitythat
thatallows
allowstesting
testingofofinvestigation
investigationand and
intervention
intervention techniques
techniques inin safe
safe conditions.Cascina
conditions. CascinaCuccagna
Cuccagnaisisalso
alsothe
thesite
siteofofmany
manycultural,
cultural,social,
social,
and
and economic
economic activities
activities and
and a good
a good place
place forfor
thethe culture
culture dissemination
dissemination ofof risk
risk protection
protection through
through thethe
direct
direct observation
observation ofof the
the actions
actions toto
bebe taken,their
taken, theirinvasiveness
invasivenessand andinterference
interferencewith withthe
theusers’
users’lives,
lives,
the
the implementation
implementation timing,
timing, thethe possibility
possibility ofof extending
extending them
them over
over time,
time, andand thethe related
related costs.
costs.
4.1.
4.1.Historical
HistoricalInvestigation, Survey,
Investigation, Redrawing,
Survey, and
Redrawing, Crack
and Pattern
Crack Pattern
Based
Based onon
thethe
historical research,
historical the phases
research, of expansion
the phases and modification
of expansion of the original
and modification of thebuilding
original
over the years could be reconstructed (Figure 4).
building over the years could be reconstructed (Figure 4).
The
Thecontact
contactsurfaces
surfacesbetween
between portions
portions built
builtinin
different
differentperiods
periodsare delicate
are delicateand require
and require attention
attention
because, often, the extensions are made without sufficient connection to adjacent
because, often, the extensions are made without sufficient connection to adjacent constructions. constructions. Some
Some
ofof
these surfaces have also been highlighted by redrawing the crack pattern (Figure
these surfaces have also been highlighted by redrawing the crack pattern (Figure 5): the red cracks5): the red cracks
inin
Figure
Figure 5 are anan
5 are example
example ofof
this.
this.The
The crack
crackpattern
pattern survey
surveyperformed
performed room
room bybyroom
room (Figure 6)6)
(Figure has
has
highlighted other criticalities where a more in-depth investigation is required. Instead,
highlighted other criticalities where a more in-depth investigation is required. Instead, the building the building
redrawing
redrawinginin a BIM
a BIM environment
environment from
from the laser
the laserscanner
scanner survey
survey points
points cloud
cloudhas
hashighlighted
highlighted some
some
shortcomings at the deformation level: floor sags and possible lack of plomb
shortcomings at the deformation level: floor sags and possible lack of plomb (Figure 7). (Figure 7).
60
Appl. Sci. 2020, 10, 2119
Figure
Figure4.4.Historic
Historicsurvey of construction
survey of constructionphases.
phases.
61
Appl. Sci. 2020, 10, 2119
Section B-B
Section B-B
Prospective views from section B-B
Prospective views from section B-B

Figure 5. Example of crack pattern (in red a crack probably due to lack of connection between two
Figure 5. Example of crack pattern (in red a crack probably due to lack of connection between two
blocks
Figure 5.built in different
Example construction
of crack pattern (inperiods).
red a crack probably due to lack of connection between two
blocks built in different construction periods).
blocks built in different construction periods).
DECAY PHENOMENA INDEX
1.PULVERIZATION/SPALLING
2.BIOLOGICALINDEX
DECAY PHENOMENA DEPOSIT
3.STAINING
1.PULVERIZATION/SPALLING
4.SPUNTERING
2.BIOLOGICAL DEPOSIT
5.SURFACE EROSION
3.STAINING
6.WOOD DETERIOARTION
4.SPUNTERING
7.VERTICAL JOINTS
5.SURFACE
8.MORTAREROSION
DETACHMENT
6.WOOD DETERIOARTION
9.PLASTER DETACHMENT external
7.VERTICAL JOINTS
10.PLASTER DETACHMENT internal Original wooden floor New wood structure.
decay phenomena diffuse Absence of decay
8.MORTAR DETACHMENT
11.PATINA Presence of beams phenomena
9.PLASTER deformation, woodworms,
12.LOSSDETACHMENT
OF BOND external
crackswooden
and white patine.
10.PLASTER DETACHMENT internal Original floor New wood structure.
13.SCALING decay phenomena diffuse Absence of decay
11.PATINA Presence of beams phenomena
14.RUST deformation, woodworms,
12.LOSS OF BOND
15.HOLES cracks and white patine.
13.SCALING
16.HIGHER PLANTS
14.RUST
15.HOLES
NON-ACCESSIBLE ROOF
16.HIGHER PLANTS
NON-ACCESSIBLE ROOF
Figure 6. Crack pattern representation for each room of the building.
62
Appl. Sci. 2020, 10, 2119
(a) (b)
(a) (b)
Figure 7. Example of survey of a flexural anomaly in a floor beam (a) laser scanner survey; (b)
Figure 7. Example
Figure 7.ofExample
surveyofof a flexural
survey anomaly
of a flexural anomalyin in
a floor
a floor beam (a)laser
beam (a) laser scanner
scanner survey;
survey; (b) (b)
section redrawing.
section redrawing.
section redrawing.
The construction of theof3D

The construction model,
the 3D model,again
again inin a BIM
a BIM environment
environment (Figure 8),(Figure 8), step
was the next wasinthe next
The construction
planningthe
ofmanagement
the 3D model,
the management of the
again structural
in a BIMsystem
environment (Figure 8), was thespecific
next step in
step in planning of whole
the whole structural in the phase in
system of identification
the phase of of identification
planning thevulnerabilities
management of the whole structural system in the phase of identification
and in the phase of design and management of seismic consolidation and adaptation of specific
of specific vulnerabilities and in the phase of design and management of seismic consolidation and
vulnerabilities and in the phase of design and management of seismic consolidation and adaptation
interventions.
adaptation interventions.
interventions.

(a)
(a)
(b)
Figure 8. Figure 8. Reconstruction

Reconstruction of theof the parametric 3D
parametric 3Dmodel and photorealistic
model representation
and photorealistic of the white- of the
representation
model render: (a) point cloud, (b) re-drawing.
white-model render: (a) point cloud, (b) re-drawing.
4.2. Mechanical Characterization of Materials and Rapid Vulnerability Analysis
The survey on the masonry quality has shown 63 masonry with variable quality, depending above
all on the time of construction (Figure 4). In this case, the calculation of a safety factor SF (%) (Equation
(2)) is not required, but we can assume that: for class A masonry, SF can be considered greater than
80% of the value required by law; for class B masonry, it can be considered in the range of 50%–75%;
Appl. Sci. 2020, 10, 2119
4.2. Mechanical Characterization of Materials and Rapid Vulnerability Analysis

The survey on the masonry quality has shown masonry with variable quality, depending above
all on the time of construction (Figure 4). In this case, the calculation of a safety factor SF (%) (Equation
(2)) is not required, but we can assume that: for class A masonry, SF can be considered greater than 80%
of the value required by law; for class B masonry, it can be considered in the range of 50%–75%; while
for class C masonry, it can be considered between 25% and 50% of the value required by law. The walls
tested, except for a couple of cases, show a medium/high behavior to the transmission of vertical loads,
and some criticalities for the transmission of out of plane loads on the most deteriorated panels.
The rapid vulnerability assessment showed a good overall behavior for the whole building (SFVG
> 80%) and also a good behavior of the wall leaves which, in case of an earthquake, could be subject
to tilting mechanisms (SFVL > 80%). On the other hand, some roof and inter-storey floors showed
problems in transmitting vertical loads (60% < SFVS < 40%).
The presence of non-structural elements that could cause vulnerability, such as flue pipes or
balconies and walkways, together with poor floor-masonry anchoring, led to a vulnerability level of
type B2 (Table 3).
The surveys, crack pattern observation and the rapid evaluations which were carried out have
indicated some important shortcomings, worthy of diagnostic investigation. Therefore, the following
were carried out:
• Geo-radar surveys in some rooms of the Cascina to understand the direction of floor structural
elements when covered with a thatched ceiling, to verify the wall thickness as well as the
foundations depth and size. The foundations investigation has also confirmed the probable
difference in the construction period of some parts of the building [27,28];
• Sonic investigations on some walls, in those areas where the crack pattern showed fissures of
doubtful interpretation, in order to investigate the panel homogeneity and identify possible
detachments of internal leaves and/or surface plasters;
• Superficial and deep penetrometric tests for mortars characterization;
• Bricks characterization based on the behavior detected on bricks preserved on site and probably
contemporary to the ones in use. The results obtained were compared with values available in the
literature for materials of the same type and period [29–31];
• Penetrometric tests (©Resistograph) to characterize the timber floors and roofs [32–34].
Normally, a focused diagnostic phase such as the one described above affects less than 6% the cost
of the building’s safety intervention.
4.3. Identification of Specific Vulnerabilities

From the rapid evaluations and the experimental tests carried out, each vertical and horizontal
structural element and all the non-structural elements could be classified as indicated in Table 4.
Each type of specific vulnerability found on individual elements was assigned a weight, according
to the damage level obtained in the previous investigations: absent damage (0), scarcely influential
(1), averagely influential (2), significantly influential (3). The result of this operation is presented in
a matrix with a correspondence between structural elements and specific vulnerabilities, useful to
characterize the elements in a BIM environment. In the commercial software that was used, in addition
to the geometric, mechanical, and material characteristics, each structural element could be associated
with the table of specific vulnerabilities and weights assigned to each item referred to that element.
(Figure 9).
4.4. Queries on Vulnerability Levels Carried Out in the BIM Environment

Once the 3D model of the building had been prepared in a BIM environment and the matrix of
the correspondence between structural elements and specific vulnerability had been constructed, the
queries were carried out.
64
Appl. Sci. 2020, 10, 2119
After associating the vulnerabilities and their respective weights to the building elements, a
colored field (selected by the user for each query) was connected with each weight and queries
were carried out about the presence or absence of vulnerability elements and their influence on the
assessment of the overall vulnerability of the building (Figure 9).
Different types of queries can be performed. The system can be queried about the presence of a
certain type of specific vulnerability throughout the building and its degree of incidence on a given
element or on the overall vulnerability of the system (Figure 10). The presence of two or more specific
vulnerabilities with different degrees of incidence on the overall vulnerability can also be visualized
(Figures 10 and 11).
An interesting aspect is the possibility to know which elements are particularly vulnerable because
they are affected by two or more criticalities, each measured by its weight on the global vulnerability.
By carrying out a single query, the BIM system is able to respond to the question by highlighting the
elements that present all (and only) the queried criticalities and with the weights investigated. The
graphic result is the coloring of the elements affected by these vulnerabilities (Figures 12 and 13).
If effective structural improvements had been made, like an improvement of the connection
between the walls and the thrusting roof, the level of damage associated with the elements will be
decreased; so, by making new queries, the effectiveness of the improvements and their location will be
visualized in the model.
Masonry
Wall
Thickness 63
cm.
Damage
level
quality of supports and 2.0000
binders
masonry adhesion 2.0000
and/or cohesion
related to the masonry
section: positioning of
the outer leaf
related to the
architectural-structural
elements: slenderness
masonry adhesion
and/or cohesion
presence of chimneys
interruptions due to
plants ducts
eaves/rain pipes within
(a)
the masonry section
Figure 9. Cont.
65
g
Appl. Sci. 2020, 10, 2119
Roof Cover
Roof tile
Damage
level
lack of binder in the
joints
masonry de-cohesion
presence of water
decay of stone elements
masonry adhesion
and/or cohesion
heads rotting
decay of secondary
elements and/or of the
tables
general decay 1.0000
exposed outer leaf 1.0000
plaster efficiency
conditions of the
covering layer
inefficient eaves and 1.0000 (b)
pipes
Figure 9. Example of coupling between structural element, specific vulnerability, and associated
weights: (a) masonry wall thickness; (b) roof cover.
CHART_01
QUERIED VULNERABILITY INDEXES
Presence of chimneys
LEGEND
Label Rule
presence of chimneys = 2
presence of chimneys = 3
Figure10.
Figure 10.Simple
Simplequery:
query:presence
presenceofofflue
fluepipes
pipesand
andtheir
theirdegree
degreeofofincidence
incidenceon
onthe
thevulnerability
vulnerabilityofof
the wall panel.
the wall panel.
CHART_02
66
Lack or poor effectiveness of connections walls-roof
Lacking or ineffective ties
Figure
Appl. Sci. 10.2119
2020, 10, Simple query: presence of flue pipes and their degree of incidence on the vulnerability of
the wall panel.
CHART_02
Conditions of the covering layer
LEGEND
Label Rule
lack or poor effectiveness of
connections walls-roof = 2
lacking or ineffective ties = 2
conditions of the covering layer = 1
Figure 11. Integrated queries: display of three different types of specific vulnerabilities with an impact
of less than 2 on the overall vulnerability.
67
Figure
Appl. Sci. 2020, 11. Integrated queries: display of three different types of specific vulnerabilities with an impact
10, 2119
of less than 2 on the overall vulnerability.

General decay
Deformations
Lack or poor effectiveness of connections
LEGEND
Label Rule
General decay = 2
Deformations = 2
Lack or poor effectiveness of
connections = 1
Figure12.
Figure 12.Multiple
Multiplequeries:
queries: presence
presence of
of multiple
multiple vulnerabilities
vulnerabilities with
with different
different weights
weightson onthe
thesame
same
element of the system. The coloring identifies the elements affected by all 3 vulnerabilities in the in
element of the system. The coloring identifies the elements affected by all 3 vulnerabilities the
legend
legend with their
with their weights. weights.
68
Appl. Sci. 2020, 10, 2119

Conditions of the covering layer
LEGEND
Label Rule
Deformations = 3
Lack or poor effectiveness of

connections = 3
Figure
Figure 13. 13. Multiple
Multiple queries:
queries: presence
presence ofofmultiple
multiple vulnerabilities
vulnerabilities with different
with weights
different on theon
weights same
the same
element of the system. The coloring identifies the elements affected by all 2 vulnerabilities in the
element of the system. The coloring identifies the elements affected by all 2 vulnerabilities in the legend
legend with their weights.
with their weights.
5. Conclusions and Future Developments
5. Conclusions and Future Developments
The research issue of assessing the seismic vulnerability of existing buildings is still unresolved.
The research
For issue years,
about thirty of assessing
scholarstheand
seismic vulnerability
researchers have beenof existing
working buildings is stillthe
on quantifying unresolved.
seismic For
about thirty years, scholars
vulnerability and
of historical researchers
buildings, havedifferent
following been working on quantifying
approaches the seismic
at different levels of detail vulnerability
(rapid,
static,buildings,
of historical dynamic approaches,
followingetc.).
different approaches at different levels of detail (rapid, static, dynamic
This
approaches, etc.). open-ended problem is raising awareness among the public and owners of the need for
implementing safety measures even on buildings not hit by the earthquake in areas with medium-
This open-ended problem is raising awareness among the public and owners of the need
low seismicity.
for implementing safety measures even on buildings not hit by the earthquake in areas with
In this work, we presented the application of a procedure for assessing the seismic vulnerability
medium-low seismicity.
of an existing building based on survey tools already present in the guidelines of several Italian
In Regions.
this work, we presented the application of a procedure for assessing the seismic vulnerability of
The procedure
an existing building based hason
proved
surveyto be effective
tools andpresent
already easy to use
in for
theaguidelines
professional.ofThe innovative
several part
Italian Regions.
of the work carried out consisted of the graphic representation of the results obtained.
The procedure has proved to be effective and easy to use for a professional. The innovative part
of the workUntil now, the degree of vulnerability of individual elements or macro elements was only
carried out consisted of the graphic representation of the results obtained.
possible in tabular or diagrammatic form, not easy to understand for non-experts. In this work, the
Until now, the degree of vulnerability of individual elements or macro elements was only possible
use of design in a BIM environment has been experimented. The adopted investigation procedure
in tabular
and orthediagrammatic
non-destructive form, not tests
diagnostic easyused
to understand for non-experts.
allowed definition In this work,
of, for each structural thea use of
element,
design in a BIMclassification
detailed environment of itshas been experimented.
vulnerabilities The on
and their weight adopted investigation
the overall procedure and the
building vulnerability.
non-destructive diagnostic tests used allowed definition of, for each structural element, a detailed
classification of its vulnerabilities and their weight on the overall building vulnerability.
Therefore, in addition to the classic geometric, mechanical, and material characteristics, the use
of a BIM model allowed associating to each structural element the presence of vulnerability aspects
and their weight. By means of focused queries, specific vulnerabilities present in the building and the
reached level of damage (measured as a weight on the total vulnerability) could be highlighted. The
69
Appl. Sci. 2020, 10, 2119
graphic representation is very helpful to understand both the location of critical points and the problem
severity. The owner becomes aware that the building presents problems and that the interventions
have different degrees of urgency but are necessary; on the other hand, the designer has a tool to plan,
in terms of time, the maintenance and adaptation interventions and to control ex-post the effectiveness
of the interventions carried out.
The procedure presented here served as a basis for planning the consolidation and safety
interventions on Cascina Cuccagna in Milan. Critical issues in the floors of the South-West wing and of
the South-East wing have been highlighted that require immediate static safety interventions. The
floors in the corner area between the South-East and the South-West wings also turned out to be poorly
connected to the perimeter masonry and require seismic adaptation. In addition, historical analysis
and geo-radar survey of the foundations have shown that the South-West wing is posterior to the
South-East wing; redrawing of the cracked pattern has shown a possible poor connection between the
perimeter walls of the two wings. All these aspects have been reported in the BIM environment and
expressed in graphic form. Thus, the designers were able to plan the most urgent interventions.
Of course, designing in a BIM environment is not without difficulties. First of all, every element
of the project must be precisely defined; if this gives great advantages to managing the construction
site, certainly it is a not easy constraint in the architectural design phase. In fact, if the architectural
design criteria do not follow the BIM environment logic from the very beginning, then the subsequent
modeling in a BIM environment is not so easy. Difficulties have also been met due to the compatibility
between CAD graphics and structural modeling programs, used in professional practice, with modeling
in a BIM environment; this affects the easy transfer of project files in this environment.
These problems are mitigated when the application deals with existing assets, where geometries
and materials are given and the implementation in a BIM environment is easier. However, it is necessary
to underline that working on existing buildings also requires modeling of structural sections often out of
plan, variable, or deformed geometry not easy to shape with the simple elements present in the software.
Therefore, simplifications or devices are often used to facilitate the three-dimensional construction of
the building and the interchange with third-party software (structural, energy, computational, etc.).
This aspect must be solved as well as the improvement of graphic resolution; the architectural side is
still very poor and forces professionals to produce two or more types of design tables. Another aspect
in which the current BIM modeling is poor is its software compatibility to facilitate professional use.
As for the procedure introduced here, the authors believe that the proposed procedure can be
replicated with the same degree of effectiveness on other existing buildings and represents a real
way of involving the client in the safety processing of the existing historical heritage. However, in
order to improve the input of collected data and the graphic output, it will still require application
efforts concerning both different vulnerabilities and monitoring of the effectiveness of the implemented
actions for static and seismic adaptation.
Author Contributions: Conceptualization, E.G.; Methodology, E.G. and A.A.; Structural evaluation, Writing, E.G.;
Restoration, Reviewing and Editing, A.A.; Methodology, Validation, F.M.; Laser scanner survey, BIM application,
F.V. All authors have read and agreed to the published version of the manuscript.
Funding: The following research is funded by the CARIPLO Foundation, Milan, Italy, in the context of the 2017
Art and Culture-Cultural Heritage at Risk Calls, Project PRE.CU.R.S.OR.
Acknowledgments: Thanks to the ACCC association of Cascina Cuccagna, the design team of Hydea s.p.a. and
Eng. Luciano Ardito for the collaboration undertaken in this project.
References
1. Dell’Orto, C.; Guzzetti, F.; Maroldi, F.; Molina, C. G.I.S. for the post-earthquake damage assessment. The case
of the Fontecchio historic heritage. In Proceedings of the National Conference Sicurezza e Conservazione
nel Recupero dei Beni Culturali Colpiti da Sisma. Strategie e Tecniche di Ricostruzione ad un Anno dal
Terremoto Abruzzese, Venice, Italy, 8–9 April 2010; CD-ROM. pp. 1–10. (In Italian).
70
Appl. Sci. 2020, 10, 2119
2. Dell’Orto, C.; Maroldi, F. Post-Earthquake damage evaluation and reconstruction strategies: G.I.S. on
Fontecchio. In Proceedings of the 14th National Conference ASITA, Brescia, Italy, 9–12 November 2010;
CD-ROM. pp. 835–840. (In Italian).
3. Paudel, B.; Pradhan, P.K. Seismic Vulnerability Assessment of Buildings in GIS Environment: Dhankuta
Municipality, Nepal. Geogr. J. Nepal 2011, 9, 1–12. [CrossRef]
4. Ricci, P.; Verderame, G.M.; Manfredi, G.; Pollino, M.; Borfecchia, F.; De Cecco, L.; Martini, S.; Pascale, C.;
Ristoratore, E.; James, V. Seismic Vulnerability Assessment Using Field Survey and Remote Sensing
Techniques. In Proceedings of the International Conference on Computational Science and its Application,
ICCSA 2011, Santander, Spain, 20–23 June 2011; Murgante, B., Ed.; Springer: Berlin/Heidelberg, Germany,
2011. Part II, LNCS 6783. pp. 109–124. [CrossRef]
5. Ahmed, M.M.; Jahan, I.; Alam, J. Earthquake Vulnerability Assessment of Existing Buildings in
Cox’s-Bazar using Field Survey & GIS. IJERT Int. J. Eng. Res. Technol. 2014, 3, 1147–1156. Available
online: https://www.ijert.org/research/earthquake-vulnerability-assessment-of-existing-buildings-in-coxs-
bazar-using-field-survey-gis-IJERTV3IS080729.pdf (accessed on 8 July 2014).
6. Rahman, N.; Ansary, M.A.; Islam, I. GIS based mapping of vulnerability to earthquake and fire hazard in
Dhaka city, Bangladesh. Int. J. Disaster Risk Reduct. 2015, 13, 291–300. [CrossRef]
7. Rezaie, F.; Panahi, M. GIS modeling of seismic vulnerability of residential fabrics considering geotechnical,
structural, social and physical distance indicators in Tehran using multi-criteria decision-making techniques.
Nat. Hazards Earth Syst. Sci. 2015, 15, 461–474. [CrossRef]
8. Taffarel, S.; Marson, C.; Valotto, C.; Rovevrato, M.; Munari, M.; Da Porto, F.; Modena, C. Seismic vulnerability
maps of Timisoara historical centre based on fragility curves. In Proceedings of the 10th International
Conference on Structural Analysis of Historical Constructions, SAHC2016, Leuven, Belgium, 13–15 September
2016; Von Balen, K., Verstrynge, E., Eds.; CRC Press: Boca Raton, FL, USA, 2016.
9. Taffarel, S.; Caliman, M.; Valluzzi, M.R.; Modena, C. Seismic vulnerability assessment of clustered historical
centers: Fragility curves based on local collapse mechanisms analyses. In Proceedings of the IB2MAC, 16th
International Brick and Block Masonry Conference, Trends, Innovations and Challenges, Padua, Italy, 26–30
June 2016; Modena, C., da Porto, F., Valluzzi, M.R., Eds.; CRC Press, Taylor & Francis Group: London, UK,
2016. A Balkema Book, electronic support. pp. 2499–2506, ISBN 9781138029996. [CrossRef]
10. Sadrykia, M.; Delavar, M.R.; Zare, M. A GIS-Based Fuzzy Decision Making Model for Seismic Vulnerability
Assessment in Areas with Incomplete Data. ISPRS Int. J. Geo-Inf. 2017, 6, 119. [CrossRef]
11. Taffarel, S. Metodi Speditivi Per la Valutazione Della Vulnerabilità Sismica Del Costruito Storico: Approccio
All’incertezza Nelle Forme di Aggregazione Complessa a Diversa Scala. Ph.D. Thesis, dBC Department,
University of Padua, Padua, Italy, 2017. (In Italian).
12. Coccia, D.E. Interoperabilità BIM-GIS Nella Classificazione Della Vulnerabilità Sismica Degli Edifici. Master’s
Thesis, Politecnico di Torino, I School of Engineering, Turin, Italy, 2017. (In Italian).
13. Welch, D.P.; Sullivan, T.; Filiatrault, A. Potential of Building Information Modelling for seismic risk
mitigation in buildings. Bull. New Zealand Soc. Earthq. Eng. 2014, 47, 253–263. Available online:
https://www.nzsee.org.nz/library/nzsee-quarterly-bulletin/vol-41-50/ (accessed on 8 July 2014). [CrossRef]
14. Christodoulou, S.E.; Vamvatsikos, D.; Georgiou, C. A BIM-Based Framework for Forecasting and Visualizing
Seismic Damage, Cost and Time to Repair. In eWork and eBusiness in Architecture, Engineering and Construction;
Menzel, Scherer, R., Eds.; Taylor & Francis Group: London, UK, 2010; pp. 33–38. ISBN 978-0-415-60507-6.
15. Alirezaei, M.; Noori, M.; Tatari, O.; Mackie, K.R.; Elgamal, A. BIM-Based Damage Estimation of Buildings
under Earthquake Loading Condition. Procedia Eng. 2016, 145, 1051–1058. [CrossRef]
16. Xu, Z.; Lu, X.; Zeng, X.; Xu, Y.; Li, Y. Seismic loss assessment for buildings with various –LOD BIM data. Adv.
Eng. Inform. 2019, 39, 112–126. [CrossRef]
17. Borri, A.; De Maria, A.; Casaglia, S. The EAL-M method for the seismic classification of the existing masonry
buildings: A comparison between different methods and preliminary evaluations of other typologies. Progett.
Sismica 2014, 5, 11–29. Available online: http://dx.medra.org/10.7414/PS.5.2.11-29 (accessed on 8 July 2014).
(In Italian).
18. Borri, A.; De Maria, A. Masonry Quality Index (MQI): Correlation with the mechanical characteristics and
knowledge levels. Progett. Sismica 2015, 6, 45–63. (In Italian) [CrossRef]
19. Borri, A.; Corradi, M.; Castori, G.; De Maria, A. A method for the analysis and classification of historic
masonry. Bull. Earthq. Eng. 2015, 13, 2647–2665. [CrossRef]
71
Appl. Sci. 2020, 10, 2119
20. Borri, A.; De Maria, A. La Classificazione Sismica: Un Protocollo Metodologico Già Operativo
Applicabile Agli Edifici Esistenti in Murature, Ingegno-Web. 2016. Available online:
https://www.ingenio-web.it/6003-la-classificazione-sismica-un-protocollo-metodologico-gia-operativo-
applicabile-agli-edifici-esistenti-in-muratura (accessed on 8 July 2016).
21. Binda, L.; Saisi, A.; Tiraboschi, C. Investigation procedures for the diagnosis of historic masonries. J. Constr.
Build. Mater. 2000, 14, 199–233. Available online: https://www.sciencedirect.com/journal/construction-and-
building-materials/vol/14/issue/4 (accessed on 8 July 2000). [CrossRef]
22. Binda, L.; Borri, A.; Cardani, G.; Doglioni, F. Scheda Qualità Muraria: Relazione Finale e Linee Guida per la
Compilazione Della Scheda di Valutazione Della Qualità Muraria; Report ReLUIS 2005–2008; DPC Department Of
Civil Protection: Rome, Italy, 2009. (In Italian)
23. Erberik, M.A. Generation of fragility curves for Turkish masonry buildings considering in-plane failure
modes. Earthq. Eng. Struct. Dyn. 2008, 37, 387–405. [CrossRef]
24. Rota, M.; Penna, A.; Magenes, G. A methodology for deriving analytical fragility curves for masonry
buildings based on stochastic nonlinear analyses. Eng. Struct. 2010, 32, 1312–1323. [CrossRef]
25. Pagnini, L.C.; Vicente, R.; Lagomarsino, S.; Varum, H. A mechanical model for the seismic vulnerability
assessment of old masonry buildings. Earthq. Struct. 2011, 2, 25–42. [CrossRef]
26. Doglioni, F. Codice di Pratica (Linee Guida) per la Progettazione Degli Interventi di Riparazione, Miglioramento
Sismico e Restauro dei Beni Architettonici Danneggiati dal Terremoto Umbro-Marchigiano del 1997; BUR, 29
settembre 2000, bollettino ufficiale della regione Marche, anno XXXI • n. ed. s. 15; Regione Marche:
Ancona, Italy, 2000; Available online: http://www.ahrcos.it/newsite/terremoto/Codice_di_pratica_terremoto_
umbromarchigiano97.pdf (accessed on 8 July 2000). (In Italian)
27. Binda, L.; Zanzi, L.; Lualdi, M.; Condoleo, P. The use of georadar to assess damage to a masonry Bell Tower
in Cremona, Italy. NDT E Int. 2005, 38, 171–179. [CrossRef]
28. Binda, L.; Lualdi, M.; Saisi, A. Investigation strategies for the diagnosis of historic structures: On-site tests on
Avio Castle, Italy, and Pišece Castle, Slovenia. Can. J. Civ. Eng. 2008, 35, 555–566. [CrossRef]
29. Candela, M.; Cattari, S.; Lagomarsino, S.; Fonti, R. Prove in situ per la caratterizzazione della muratura
in pietrame grezzo alle azioni nel piano e fuori dal piano. In Proceedings of the Conference of the Italian
National Association of Earthquake Engineering (ANIDIS), Padua, Italy, 30 June–4 July 2013; pp. 1–10. (In
Italian).
30. Vintzileou, E. Testing Historic Masonry Elements and/or Building Models. In Perspectives on European
Earthquake Engineering and Seismology; Ansal, A., Ed.; Springer International Publishing: Cham, Switzerland,
2014; Volume 1, Chapter 8. [CrossRef]
31. Marastoni, D.; Benedetti, A.; Pelà, L.; Pignagnoli, G. Torque Penetrometric Test for the in-situ characterisation
of historical mortars: Fracture mechanics interpretation and experimental validation. Constr. Build. Mater.
2017, 157, 509–520. [CrossRef]
32. Ilharco, T.; Guedes, J.M.; Arêde, A.; Pauepério, E.; Costa, A.G. Analysis and diagnosis of timber structures in
Porto Historical Centre. In Proceedings of the 6th International Conference on Structural Analysis of Historic
Construction, Bath, UK, 2–4 July 2008; D’Ayala, D., Fodde, E., Eds.; Taylor and Francis Group: London, UK,
2008; pp. 653–661, ISBN 978-0-415-46872-5.
33. Piazza, M.P.; Riggio, M.R.; Brentari, G.B. Strengthening and control methods for old timber trusses: The
queen-post truss of the Trento theatre. In Proceedings of the 4th International Conference on Structural
Analysis of Historic Construction, Padua, Italy, 10–13 November 2004; Modena, C., Lourenço, P., Roca, P.,
Eds.; Taylor and Francis Group: London, UK, 2005; pp. 957–965.
34. Kloiber, M.; Tippner, J.; Praus, L.; Hrivinák, J. Experimental Verification of a New Tool for Wood Mechanical
resistance measurement. Wood Res. 2012, 57, 383–398. Available online: http://www.woodresearch.sk/wr/
201203/05.pdf (accessed on 8 July 2012).
72
applied
sciences
Article
Simulation of Three Constitutive Behaviors Based on
Nonlinear Ultrasound
Zaifu Zhan , Shen Wang * , Fuping Wang , Songling Huang , Wei Zhao and Zhe Wang
State Key Laboratory of Power System, Department of Electrical Engineering, Tsinghua University,
Beijing 100084, China; [email protected] (Z.Z.); [email protected] (F.W.);
[email protected] (S.H.); [email protected] (W.Z.); [email protected] (Z.W.)

Received: 12 February 2020; Accepted: 6 March 2020; Published: 13 March 2020
Abstract: Nonlinear ultrasound has attracted more and more attention. In classical acoustic nonlinear
theory, the source of nonlinearity is the change of constitutive relation of materials. Structure response
that distorts after a single tone ultrasound wave is important to detect imperfection. This is rarely
found in current simulations. The current simulation always introduces defects which do not match
to the classical acoustic nonlinear theory. In this manuscript, the recurrence expressions of three
kinds of imperfect materials for subroutine are given. The verifying simulation model that is used
for verifying recurrence equations and wave propagation model that are used for analysing the
process of ultrasonic propagation are established. The results show that the two constitutive models
are effective in the verifying simulation and the hysteresis material has some special characteristics.
Finally, ultrasonic propagation in two types of materials produce the expected harmonics, which build
foundations for simulations of nonlinear ultrasound.
Keywords: classical acoustic nonlinearity; constitutive behavior; ultrasonic testing
1. Introduction
Under cyclic loads or harsh environments, no matter what materials, parts and complete structures
will undergo material properties degradation. Continuous degradation leads to predictable and
unpredictable problems, sometimes with serious consequences. Therefore, it is necessary to do health
testing for structures or materials in practical engineering [1–5].
The early degradation of the mechanical properties takes up most of the life of the structures.
Generally in metals, the damage first appears in the form of dislocation substructures such as veins and
persistent slip bands (PSBs), and these accumulate at grain boundaries to produce strain localization
and then further produce micro-cracks. This process accounts for 80–90% [6,7] of the life of the whole
structure when the metal material structure is bearing a cyclic load or harsh environment. When the
cracks show, the life the of material is almost over, so the testing of the early properties of the materials
is particularly important.
With the development of ultrasonic non-destructive testing technology, linear ultrasound cannot
satisfy the current testing needs, because linear ultrasound technology has no high sensitivity in the
early stage of deterioration of material mechanical properties [8]. Studies show that the dislocation
substructures and resulting microplastic deformation do not cause large changes in the macroscopic
properties such as elastic modulus, sound speed and attenuation of a material [9,10], thus the changes
in the linear ultrasonic quantities are not large enough to be accurately measured with conventional
linear ultrasound.
The testing method of nonlinear ultrasound has attracted considerable attention in recent
years [3,11–16]. Ultrasonic waves are highly sensitive to material damage, propagate in all directions,
73
Appl. Sci. 2020, 10, 1982
spread quickly and have high penetration into materials [17]. The nonlinear phenomenon manifests
itself with second, third and even more harmonic generation: when a sinusoidal ultrasonic wave of
finite amplitude at the f frequency crosses a nonlinear elastic material, harmonics at the 2 f ,3 f (et al.)
frequency are generated [18]. In refs [8–10], experimental results and physical models show that the
harmonic amplitude is directly related to material damage. Therefore, it can be expected that the
degree of material degradation can be evaluated by measuring the nonlinearity of the ultrasonic wave
propagated through the target material.
The sources of nonlinearity can be divided into classical acoustic nonlinearity and contact acoustic
nonlinearity (CAN). CAN manifests the phenomenon of nonlinearity by introducing a micrometer-
or even nanometer-level small crack, which has been done in many simulations [17,19–23] and
experiments [17,24–26]. However, it is known that when the material is under a load or in a severe
environment, not only a small area of the material is damaged (e.g., micro-cracks), but also every
under-service part of a structure is damaged to different degrees, which is rarely discovered in the
current simulation. It is difficult to decide where to set micro-cracks in simulation for the structure
under uniform load, but the material or structure must be damaged in service, so it is necessary to
study the degradation of the mechanical properties of the material or structure.
The work of this manuscript is based on classical acoustic nonlinearity. When the material is
undamaged, it complies with the linear constitutive relation. The wave propagation in the material is
linear. However, when the material is damaged, the mechanical constitutive behavior no longer obeys
the linear relation and the relation is complex and nonlinear [16,27,28]. As a result, the distortion of the
wave form appears and the harmonics are derived because the constitutive relation is no longer linear.
Different materials exhibit different constitutive behaviors when damaged. The coefficient of
material constitutive relation is also different for different damage degrees. In current simulation
research, no one has carefully studied the constitutive relationship of materials. In addition,
few researchers has verified the constitutive behavior of material in the simulation. Almost all
simulations show the nonlinear results directly, but it is not known whether the nonlinearity is caused
by the nonlinear constitutive relationship. In this manuscript, subroutines are written to defined three
constitutive relationships which are first and second order perturbation of classic nonlinearity and
nonlinear hysteresis respectively. Ref [28] compares linear wave propagation of a monofrequency
signal with the results for the various nonlinear and hysteretic contributions. The results of the
comparison are shown in Figure 1.
Figure 1. Schematic overview of the nonlinear contributions to the constitutive equation and its
implication for dynamic one-dimensional wave propagation of a finite-amplitude monofrequency signal.
74
Appl. Sci. 2020, 10, 1982
The remainder of this manuscript is organized as follows: Section 2 introduces not only the basic
theoretical background of nonlinear ultrasound and higher harmonics generation in theory but also
three constitutive behaviors which are written by a subroutine in simulations. In Section 3, a finite
element model of the propagation of nonlinear longitudinal waves is described in detail. finite element
method (FEM) simulation results are presented and discussed in Section 4, and conclusions are drawn
and future studies proposed in Section 5.
2. Classical Nonlinear Elasticity

In this section, the derivation of nonlinear wave equations in classical nonlinearity will be
reviewed, then we will introduce the constitutive relation of general materials. The general constitutive
relation can be decomposed to three basic constitutive relations.
In order to save computing resources, the stress is calculated by strain thorough recurrence
expression of the stress-strain relationship in simulations. Therefore, the recurrence relation of each
constitutive relation will be given in the follow subsection.
In order to unify the symbols in equations and make it easy for readers, the parameters involved
in this paper and their meanings are shown in Table 1.
Table 1. Main symbols used in the equation.
Free energy F
Lamé coefficients λ,µ
Strain tensor ε ik
Displacement component ui
Cartesian coordinate xi
Quadratic nonlinear perturbation coefficient β
Cubic nonlinear perturbation coefficient ǫ
Hysteresis α
Simplified first perturbation coefficient C1 (= − 12 E0 β)
Simplified second perturbation coefficient C2 (= − 13 E0 ǫ)
Simplified nonlinear hysteretic coefficient C3 (= 12 E0 α)
Young’s modulus E
Kronecker symbol δij
Strain rate ε̇ = dε/dt
2.1. Nonlinear Wave Equation

This subsection will begin with Hooke’s law and simply study the source of classical acoustic
nonlinearity through theoretical analysis.
This is the general expression for the free energy of a deformed isotropic body [27]:
1
F = F0 + λε2ii + µε2ik (1)
2
where, ε ik is the strain tensor whose linear expression is [27]:
1 ∂ui ∂u
ε ik = ( + k) (2)
2 ∂xk ∂xi
When nonlinearity is considered, the strain tensor should be expanded [27]:
1 ∂ui ∂u ∂u ∂u
ε ik = ( + k + l l) (3)
2 ∂xk ∂xi ∂xk ∂xi
75
Appl. Sci. 2020, 10, 1982
If only a one-dimensional situation is considered, the linear wave equation:
∂2 u ∂2 u
− c 0 =0 (4)
∂t2 ∂x2
should be amended to:
∂2 u ∂u ∂u ∂2 u
2
− c0 (1 − β + ǫ ( )2 ) 2 = 0 (5)
∂t ∂x ∂x ∂x
There is another simpler theoretical explanation that when the material is undamaged,
the relationship between stress and strain is linear:
σ = Eε (6)
When the material is damaged, however, the relationship between stress and strain is no
longer linear:
σ = Eε(1 + βε + · · · ) (7)
Substituting this relation into the one-directional wave equation:
∂2 u ∂σ
ρ 2
= (8)
∂t ∂x
Then the nonlinear wave equation is obtained:
∂2 u ∂2 u ∂u ∂2 u
ρ 2
= A 2 +B +··· (9)
∂t ∂x ∂x ∂x2
Cantrell [7,29] et al. also established a similar second nonlinear ultrasonic wave equation:
∂2 u ∂2 u ∂u ∂2 u
ρ = K 2 + ( 3K 2 + K 3 ) (10)
∂t2 ∂x2 ∂x ∂x2
Therefore, it is found that when the material constitutive relation changes, the wave equation
is no longer linear. Based on this theory, when the material is damaged, the input of the system is
monochromatic, and the output must contain harmonics.
The derivation of wave equation above does not consider the nonlinear hysteresis for simplicity.
2.2. General Constitutive Relationship

To first approximation, the one-dimensional constitutive relation between the stress σ and the
strain ε used in simulations of the dynamic behavior of solids can be expressed as follows [28]:
Z
σ= E(ε, ε̇)dε (11)
with E the nonlinear and hysteretic modulus given by [28]
E(ε, ε̇) = E0 {1 − βε − ǫε2 − α[∆ε + ε(t)sign(ε̇)] + · · · } (12)
where E0 is the linear modulus, ∆ε is the local strain amplitude over the previous period, sign(ε̇) = 1
if ε̇ > 0 and sign(ε̇) = −1 if ε̇ < 0.
In simulation, the component of constitutive relation is needed:
σij = λδij ε kk + 2µε ij (13)
where δij is the Kronecker Symbol, δij = 1 if i = j and δij = 0 if i 6= j.
76
Appl. Sci. 2020, 10, 1982
2.3. Nonlinear Classic: 1st Perturbation (β)

If only the first perturbation is considered, the nonlinear modulus will be simplified to:
E(ε, ε̇) = E0 (1 − βε) (14)
The one-dimensional constitutive relation between the stress σ and the strain ε should be:
1
Z
σ= E(ε, ε̇)dε = E0 ε − βE0 ε2 (15)
2
In the simulations, the code of the recursive constitutive equations are programmed as follows:
 m +1 +1

 ∆ε ij = εm ij − εm ij

m + 1
ε ij + εm



 m +1 ij
 λij = C1 + λ0


2 (16)
+1

 m +1
 εmij + εm ij
 µ = C 1 + µ0
 ij 2



 m +1 m +1 +1 m +1 m +1 m +1 m +1 m +1 +1
= σijm + δij (λ11 ∆εm
11 + λ22 ∆ε 22 + λ33 ∆ε 33 ) + 2µij ∆εm

σij ij
where, λ0 and µ0 are Lamé coefficients when the material is linear.
2.4. Nonlinear Classic: 2nd Perturbation (ǫ)

If only the second perturbation is considered, the nonlinear modulus will be simplified to:
E(ε, ε̇) = E0 (1 − ǫε2 ) (17)
The one-dimensional constitutive relation between the stress σ and the strain ε should be:
1
Z
σ= E(ε, ε̇)dε = E0 ε − ǫE0 ε3 (18)
3
In the simulations, the code of the recursive constitutive equations are programmed as follows:
 m +1 +1

 ∆ε ij = εmij − εm ij

m +1
ε ij + εm



ij 2
 λijm+1 = C2 (

) + λ0


2 (19)
+1
 εmij + εmij 2
m +1

µij = C2 ( ) + µ0


2




 m +1 m +1 +1 m +1 m +1 m +1 m +1 m +1 +1
11 + λ22 ∆ε 22 + λ33 ∆ε 33 ) + 2µij ∆εm

σij ij
2.5. Nonlinear Hysteretic (α)

If only the hysteresis of material is considered, the nonlinear modulus will be simplified to:
E(ε, ε̇) = E0 {1 − α[∆ε + ε(t)sign(ε̇)])} (20)
In simulations, increment step and mesh elements are both set very small. Therefore, eliminating
the ∆ε can be simple and satisfy the accuracy at the same time. The one-dimensional constitutive
relation between the stress σ and the strain ε should be:
77
Appl. Sci. 2020, 10, 1982
1

 E0 ε − E0 αε2 ,
 ε̇ > 0
2
Z
σ= E(ε, ε̇)dε = (21)
 E ε + 1 E αε2 ,

ε̇ < 0
0 0
2
In the simulations, the code of recursive constitutive equations are programmed as follows:
 m +1 +1

 ∆ε ij = εm ij − εm ij

m +1
ε ij + εm



ij
 λijm+1 = c3′ C3 (c3′ = 1, i f ∆εm +1
c3′ = −1, i f ∆εm +1

+ λ0 <0 and > 0)


2 ij ij
(22)
+1

 m +1 εmij + εm
ij
′
µ = c C + µ0

3 3

 ij 2



 m +1 m +1 +1 m +1 m +1 m +1 m +1 m +1 +1
11 + λ22 ∆ε 22 + λ33 ∆ε 33 ) + 2µij ∆εm

σij ij
3. Numerical Simulation
Two two-dimensional (2D) finite element method (FEM) models were built with Abaqus/CAE
and dynamic simulations were conducted with Abaqus/Explicit software.
The Explicit solver of Abaqus specializes in solving complex problems. It does not solve the
coupled equation of systems or generate the total stiffness matrix in each increment step, but recurs the
result in a small increment step in the time domain. It only calculates the total energy in the simulation
to determine whether there is divergence in each increment step.
What is more, the solver provides an interface—VUMAT, which is used to define the mechanical
constitutive behavior of a material thorough FORTRAN language. Recursive expressions mentioned
in Section 2 are achieved by subroutine.
There are two 2D FEM models were build. The first model was built for verifying the correctness
of the codes. The propagation of the longitudinal wave was simulated in the second model.
3.1. Verifying the Model

In order to verify the reliability of the program, the following model was established.
The 2D verifying model was a square with 10 mm side length as shown in Figure 2. The boundary
condition set on the left side was encastré which means that the left edge could not move in the x and
y directions during the whole process. Displacement in the y direction of the top and bottom edges
was set to 0 so it could not move in the y direction. In the right side of the model, the displacement
boundary condition was used to produce deformation so as to analyze the relation between stress and
strain. The displacement signal in the right side of the verifying model was a triangular wave which is
shown in Figure 3. The amplitude of the boundary condition was set to 0.25 mm.
Figure 2. Geometry of the verifying model for simulations.
78
Appl. Sci. 2020, 10, 1982
Amplitude (mm)
0.5
-0.5
-1
0 0.2 0.4 0.6 0.8 1
Time (s) 10-5
Figure 3. Displacement signal for the right side boundary condition of the verifying model.
We chose steel as the simulation material because it is widely used. The material properties are
shown in Table 2. In this manuscript, all materials in the simulation were steel.
Table 2. Material properties in simulations.
Material Type Mass Density Young’s Modulus Poisson’s Ratio

Steel 7800 kg/m3 2 × 11 Pa 0.3
Abaqus software contains a wealth of mesh elements libraries. We chose CPE4R (i.e., linear reduction
integral plane strain type) as the element type. It should be noted that the linear reduced integral refers
to the linear interpolation used by the mesh elements in the calculation, which has nothing to do with the
nonlinear source mentioned in this manuscript, but only for saving calculation resources and meeting
the algorithm of Abaqus/Explicit solver.
The most important consideration is that only one mesh element was formed in the verifying
model which means the whole model was an element. It can be clearly known that the constitutive
relation does change in simulations by observing the stress and strain of an element. The whole
simulation can only be correct when the constitutive behavior of one element is correct.
3.2. Wave Propagation Model

After the code written for the different constitutive behavior was verified, the wave propagation
in a solid was simulated to study different nonlinear phenomena from different materials whose
constitutive behavior was different.
The 2D model was 40 mm long and 20 mm thick as shown in Figure 4 The material properties are
in Table 2. The boundary conditions of four sides in the model were free boundary conditions which
means the stress on the boundary was zero.
Figure 4. The 2D wave propagation model and the place of excitation.
79
Appl. Sci. 2020, 10, 1982
The excitation was applied in the left edge of model within the width of 6 mm and the excitation
amplitude was 90 nm. The excitation signal was a five cycle tone-burst, as shown in Figure 5.
The window function was used to prevent spectrum leakage. The excitation signal adopted a frequency
of 3 × 106 Hz, and its expression was as follows:
y = sin(2π f t)(0.08 + 0.46(1 − cos(2π f t/5))) (23)
1 0.3
0.5
Amplitude
0.2
0
y
0.1
-0.5
-1 0
0 0.5 1 1.5 2 0 2 4 6 8
Time (s) 10-6 Frequency (Hz) 106
(a) (b)
Figure 5. (a) Temporal waveform of the excited tone burst signal. (b) Frequency spectrum of the excited
tone burst signal.
Mesh generation is important in FEM. The mesh element size should be much smaller than the
wavelength, otherwise there is no wave propagation phenomenon. The accuracy requires that the
smallest wavelength λmin must be correctly sampled to describe the propagating mode in the frequency
range for analysis. The denser the mesh, the more accurate the result. However, more density means
more computing resource usage. The mesh element size should balance the accuracy of results with
computing resources usage. Therefore, the spatial discretization of δx1 and δx2 of each element of the
mesh satisfies the condition [30]:
λmin
> 10 (24)
max (δx1 , δx2 )
Due to the need to measure harmonics, the max (δx1 , δx2 ) of mesh elements was set to 0.05 mm.
The type of mesh element was CPE4R which specializes in plane strain problems.
Another stability condition is that the time step must be chosen so that no wave can propagate
across one mesh spacing in less than one time step. Typically, the time discretization must satisfy the
condition [30]:
min(δx1 , δx2 )
∆t ≤ 0.7 (25)
cl
where cl is the velocity of the longitudinal wave. Therefore, in order to be precise, the step time was
set as 1 × 10−9 s, and the total simulation time was 8 × 10−6 s.
4. Simulation Result and Discussion

This section firstly presents and discusses the simulation results generated from three group
verifying models. Simulation results from models of first and second order perturbation of classic
nonlinearity agreed with the nonlinear theory. However, the results of nonlinear hysteresis had
some differences from the ideal situation. Some analysis will be made based on the results. Finally,
two stable codes of classic nonlinearity were used in the simulation of wave propagation. The growth
of harmonics in different materials was obtained by multiple simulations.
80
Appl. Sci. 2020, 10, 1982
4.1. Verification Simulation
4.1.1. Nonlinear Classic: 1st Perturbation (β)

When the first order perturbation was added to the verifying model, three deformation results of
model are shown in Figure 6.
The solid line represents the deformation when the displacement was zero. The thick dotted line
and thin dotted line are deformation of maximum tension and compression respectively.
Figure 6. Deformation of verifying model.
From the deformation in Figure 6, it can be seen that verifying model only deformed in the x
direction. Owing to the whole model being just one element and none of the other three sides can
move, this element only produced strain in x direction, which was ε 11 .
In Figure 7, two groups of simulations show that the constitutive relation was changed by the
subroutine. Figure 7a,b is the change of strain in the simulation process. They were exactly the same
because of the boundary condition. The stress of the verifying model was very different as shown
in Figure 7c,d. We can see that stress-time curve did distort when C1 = −4 × 106 . It is obvious that
the upper part and the lower part of the stress were no longer symmetrical. Although it can be seen
that the stress changed obviously, it was not intuitive enough. Every point of stress and strain at every
moment can be found and connected, which is shown in Figure 7e,f. It is clear that the stress-strain
relationship was no longer linear. Therefore, it can be confirmed that the constitutive behavior of the
material was indeed changed to what we want.
0.03 0.03
0.02 0.02
0.01 0.01
Strain
Strain
0 0
-0.01 -0.01
-0.02 -0.02
-0.03 -0.03
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
Time (s) 10-5 Time (s) 10-5
(a) Strain ε 11 changes with time when C1 = 0. (b) Strain ε 11 changes with time when C1 = −4,000,000.
(c) Stress σ11 changes with time when C1 = 0. (d) Stress σ11 changes with time when C1 = −4,000,000.
Figure 7. Cont.
81
Appl. Sci. 2020, 10, 1982
(e) Relation of stress and strain when C1 = 0. (f) Relation of stress and strain when C1 = −4,000,000.
Figure 7. Comparative analysis of strain and stress when C1 is equal to 0 and −4,000,000 respectively.
A series of stress-strain curves could be obtained by changing the value of C1 as shown in Figure 8.
It can be seen that the degree of stress-strain nonlinearity increased with the increase of the first-order
perturbation. In the following simulation, several values of perturbation were selected to simulate the
propagation of ultrasound.
Figure 8. Stress-strain curves under different first order perturbations.
Simulation results of Figures 7 and 8 were consistent with the theory of Ref [28], which confirms
the correctness of code of the subroutine.
4.1.2. Nonlinear Classic: 2nd Perturbation (ǫ)

When only adding the second order perturbation to the verifying model, three comparison results
of verifying model are shown in Figure 9 like the last subsection. In the same way, the comparison
between two groups of simulations verified the code of the second perturbation.
It is clear that the constitutive relationship changed by comparison in Figure 9. Figure 9a,b is
the change of strain in the simulation process. The stress of the verifying model was very different
as shown in Figure 9c,d when C2 was different. It is certain that stress-time curve distorted when
C2 = −3 × 108 , which led to the change of constitutive behavior as shown in Figure 9e,f.
In addition, lots of simulations were conducted and the curve of relation between strain and stress
were obtained as shown in Figure 10. Figure 10 shows that the constitutive relation became more and
more complicated with the change of the perturbation coefficient C2 . Therefore, it can be confirmed
that the constitutive behavior of the material was indeed changed to meet the expected goal.
82
Appl. Sci. 2020, 10, 1982
0.03 0.03
0.02 0.02
0.01 0.01
Strain
Strain
0 0
-0.01 -0.01
-0.02 -0.02
-0.03 -0.03
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
Time (s) 10-5 Time (s) 10-5
(a) Strain ε 11 changes with time when C2 = 0. (b) Strain ε 11 changes with time when C2 = −3 × 108 .
(c) Stress σ11 changes with time when C2 = 0. (d) Stress σ11 changes with time when C2 = −3 × 108 .
(e) Relation of stress and strain when C2 = 0. (f) Relation of stress and strain when C2 = −3 × 108 .
Figure 9. Comparative analysis of strain and stress when C2 is equal to 0 and −3 × 108 respectively.
Figure 10. Stress-strain curves under different second order perturbations.
83
Appl. Sci. 2020, 10, 1982
In Figure 9d, waveform distortion was more serious by comparison with Figure 7d. However,
the stress upper and lower waveform was symmetrical in this simulation. This phenomenon also
explains why the second order perturbation only produced odd harmonics which was exhibited in
Figure 1 of the Introduction section. This was the main difference between the first order perturbation
and the second order perturbation. It is a reminder that the properties of materials can be judged by a
stress curve. For example, in Ref [21], a strain curve is used to measure the damage of materials.
Figure 1 is the result of theory from Ref [20] which is consistent with the result of simulation as
shown in Figures 9 and 10, which confirmed the correctness of the code of the subroutine.
4.1.3. Nonlinear Hysteresis (α)

In th everifying model, the nonlinear hysteresis was only added to a constitutive relationship.
The modified constitutive relationship is Equation (18).
Figure 11a is the result of stress of verifying model when C3 = 1 × 107 . The distortion of stress
curve was more severe compared with the previous two results. It can be found that, in the process
of tension and compression, the stress curve showed an upward trend. Similarly, the whole curve
of stress-strain relationship rose in the process of deformation which is shown in Figure 11b. This is
because of the recursive equations and the curve must start at the origin point (0,0).
Simulations were conducted and the curves of relation between strain and stress are shown
in Figure 12.
(a) Stress ε 11 changes with time. (b) Relation of stress and strain.
Figure 11. Simulation result of hysteretic material in the verifying model when C3 = 1 × 107 .
Figure 12. Stress-strain curves for different hysteresis.
84
Appl. Sci. 2020, 10, 1982
The result in Figures 11 and 12 also means that repeated tension and compression made the
materials reach the ideal hysteretic properties in the degradation process. In the simulation, if the
target is to study the constitutive relation of hysteretic materials, the best way is the look-up table
method. However, this method is limited because different damage degrees correspond to different
constitutive curves. The constitutive formula of hysteretic materials has been defined by Nazarov and
Ostrovsky [31] in 1988 but there are several coefficients involved in the formula. These coefficients
need to be measured by actual experiments. Therefore, it is difficult to involve several coefficients in
the simulation before conducting experiments.
4.2. Wave Propagation Simulation
4.2.1. Nonlinear Classic: 1st Perturbation (β)

The models were built as described in Section 3.2. A group of simulations were conducted by
changing the value of the simplified first perturbation coefficient C1 . C1 was respectively set to 0,
−2000, −20,000, −200,000, and −2,000,000. When C1 equals 0, the material was not damaged.
The propagation of wave in simulation was exhibited by the stress nephogram which is more
intuitive as shown in Figure 13. From this figure, it can be seen that the excitation produced not only
longitudinal waves but also transverse waves, and the longitudinal wave beam spread in propagation.
As a result, some longitudinal waves reflected in the upper and lower surfaces. It was difficult to
find the difference from the stress nephogram after material constitutive relation changed. Therefore,
the stress nephogram was not used to compare the result.
The resulting signal is shown in Figure 14 when C1 = −2,000,000. The second and third
harmonics are shown very clearly. Due to the phenomenon of diffraction in the process of propagation,
however, the harmonic magnitude can be not directly compared because of diffraction in the process
of wave propagation.
In order to characterize the accumulation of harmonics better in the process of ultrasonic
propagation, the nonlinear parameter β′ was introduced as nonlinear ultrasonic detection coefficients.
The expression of β′ is as follows:
A2
β′ = 2 (26)
A1
where, Ai is the amplitude of the ith harmonic.
(a) t = 2.001×10−6 (s)
(b) t = 4.001×10−6 (s)
Figure 13. Cont.
85
Appl. Sci. 2020, 10, 1982
(c) t = 6.001×10−6 (s)
(d) t = 8×10−6 (s)
Figure 13. Stress nephogram at different times during ultrasonic propagation.
40 4
Stress amplitude (MPa)

20 3 0.1
0 2
y
0.05
-20 1
-40 0 0
0 2 4 6 8 0 0.5 1 1.5 2 0 2 4 6 8 10
-6 7
Time (s) 10 Frequency (Hz) 10 Frequency (Hz) 106
(a) (b) (c)
Figure 14. Simulation result of the first order perturbation when C1 = −2,000,000: (a) Time domain
signal. (b) Spectrum of signal. (c) Harmonic of signal.
Figure 15a shows that, under different perturbation coefficients, the nonlinear ultrasonic detection
coefficient β′ changed with wave propagation distance. When the perturbation coefficient C1 was
larger than a certain value, the coefficient β′ increased with the propagation distance. Figure 15b is
the result of Ref [17] which also used the subroutine to define the material. However, it did not give a
definite value of C1 . The correctness of simulation result was verified by contrast, which also verified
the code that is clearly described in Section 2.
(a) (b)
Figure 15. Coefficient β′ changes with propagation distance. (a) result of wave propagation model.
(b) result from Ref [17].
86
Appl. Sci. 2020, 10, 1982
4.2.2. Nonlinear Classic: 2nd Perturbation (ǫ)

A group of simulations in which the second perturbation coefficient C2 was different were
conducted by Abaqus software.
The resulting signal is shown in the Figure 16 when C2 = −3 × 107 . It can be obviously seen that
the third and fifth harmonics were generated by the second perturbation.
4 0.4 10-3
2.5

2 0.3 2
1.5
0 0.2
y
1
-2 0.1
0.5
-4 0 0
0 2 4 6 8 0 0.5 1 1.5 2 0 5 10 15
Time (s) 10-6 Frequency (Hz) 107 Frequency (Hz) 106
(a) (b) (c)
Figure 16. Simulation result of second order perturbation when C2 = −3 × 107 : (a) Time domain signal.
(b) Spectrum of signal. (c) Harmonic of signal.
Current finite element simulation software requires that the deformation of mesh elements in
simulation should not be too large or it can not guarantee the correctness of result. Therefore, the mesh
elements should not be too small. On the other hand, if the objective is to measure the third or
fifth harmonics, mesh elements must be small enough as described in Equation (24). There is a
contradiction between the two theories, so there was no need to compare the amplitude of the third
and fifth harmonics in wave propagation.
5. Conclusions
In this manuscript, three types of constitutive relations are reviewed and their subroutines with
recursive expression are written. Two kinds of simulation models are established. One is to verify the
correctness of subroutines, and the other is to conduct ultrasonic simulation. From the result of the
verifying model, the subroutines of the first and second order perturbations satisfy the nonlinear theory.
Hysteretic material simulation shows the deterioration process of materials, but it cannot be directly
used to simulate the damaged materials and some analyses are made on it. Finally, the simulation
of the first and second order perturbation shows that the harmonics appear. Especially, nonlinear
coefficient β′ increases with propagation distance in first order perturbation simulations and odd
harmonics of the second order perturbation simulations are obvious. The work in this manuscript
provides a new idea for the simulation of nonlinear ultrasonic testing.
There is still some follow-up work to be studied. Firstly, although recurrence equations can
improve the calculation speed, it is not suitable to study the hysteresis material directly. Secondly,
the current FE simulation software cannot guarantee the accuracy when the mesh element deformation
is too large. Thirdly, the recurrence equations are not verified in three-dimensional simulation.
Author Contributions: Conceptualization, Z.Z., S.W. and F.W.; data curation, Z.Z.; formal analysis, Z.Z.; funding
acquisition, F.W. and S.H.; investigation, Z.Z. and S.W.; methodology, Z.Z.; project administration, S.W. and
F.W.; resources, Z.Z., S.W. and F.W.; supervision, S.W. and F.W.; validation, Z.Z.; writing—original draft, Z.Z.;
writing—review and editing, Z.Z., S.W., F.W., S.H., W.Z. and Z.W. All authors have read and agreed to the
published version of the manuscript.
Funding: This research was funded by National Key Research and Development Program grant
number 2018YFC0809002.
Acknowledgments: The author greatly appreciates the help of translator Mengyuan Cui.
Conflicts of Interest: The authors declare no conflicts of interest.
87
Appl. Sci. 2020, 10, 1982
References
1. Farrar, C.R.; Worden, K. An introduction to structural health monitoring. Philos. Trans. A Math. Phys.
Eng. Sci. 2007, 365, 303–315. [CrossRef] [PubMed]
2. Marcantonio, V.; Monarca, D.; Colantoni, A.; Cecchini, M. Ultrasonic waves for materials evaluation in
fatigue, thermal and corrosion damage: A review. Mech. Syst. Signal Process. 2019, 120, 32–42. [CrossRef]
3. Blanloeuil, P.; Rose, L.; Veidt, M.; Wang, C.H. Time reversal invariance for a nonlinear scatterer exhibiting
contact acoustic nonlinearity. J. Sound Vib. 2019, 417, 413–431. [CrossRef]
4. Aleshin, V.; Delrue, S.; Trifonov, A.; Matar, O.B.; Abeele, K.V.D. Two dimensional modeling of elastic wave
propagation in solids containing cracks with rough surfaces and friction—Part I: Theoretical background.
Ultrasonics 2017, 82, 11–17. [CrossRef]
5. Kim, J.; Park, S.J.; Yim, H. Nonlinear Resonance Vibration Assessment to Evaluate the Freezing and Thawing
Resistance of Concrete. Materials 2019, 12, 325. [CrossRef]
6. Meyendorf, N.G.; Rösner, H.; Kramb, V.; Sathish, S. Thermo-acoustic fatigue characterization. Ultrasonics
2002, 40, 427–434. [CrossRef]
7. Cantrell, J.H. Substructural organization, dislocation plasticity and harmonic generation in cyclically stressed
wavy slip metals. Proc. R. Soc. A Math. Phys. Eng. Sci. 2004, 460, 757–780. [CrossRef]
8. Nagy, P.B. Fatigue damage assessment by nonlinear ultrasonic materials characterization. Ultrasonics 1998,
36, 375–381. [CrossRef]
9. Herrmann, J.; Kim, J.Y.; Jacobs, L.J.; Qu, J.; Littles, J.W.; Savage, M.F. Assessment of material damage in a
nickel-base superalloy using nonlinear Rayleigh surface waves. J. Appl. Phys. 2006, 99, 124913. [CrossRef]
10. Kim, J.Y. Experimental characterization of fatigue damage in a nickel-base superalloy using nonlinear
ultrasonic waves. J. Acoust. Soc. Am. 2006, 120, 1266–1273. [CrossRef]
11. Raghavan, A. Guided-wave structural health monitoring. Diss. Abstr. Int. 2007, 975, 91–114. [CrossRef]
12. Solodov, I.Y.; Krohn, N.; Busse, G. CAN: An example of nonclassical acoustic nonlinearity in solids.
13. Zarembo, I.K.; Krasil’Nikov, V.A. Nonlinear Phenomena in the Propagation of Elastic Waves in Solids.
Sov. Phys. Uspekhi 1970, 102, 778. [CrossRef]
14. Jhang, K.Y.; Kim, K.C. Evaluation of material degradation using nonlinear acoustic effect. Ultrasonics 1999,
37, 39–44. [CrossRef]
15. Zheng, Y.; Maev, R.G.; Solodov, I.Y. Nonlinear acoustic applications for material characterization: A review.
Can. J. Phys. 1999, 77, 927–967. [CrossRef]
16. Broda, D.; Staszewski, W.; Martowicz, A.; Uhl, T.; Silberschmidt, V. Modelling of nonlinear crack–wave
interactions for damage detection based on ultrasound—A review. J. Sound Vib. 2014, 333, 1097–1118.
[CrossRef]
17. Hong, M.; Su, Z.; Wang, Q.; Cheng, L.; Qing, X. Modeling nonlinearities of ultrasonic waves for fatigue
damage characterization: Theory, simulation, and experimental validation. Ultrasonics 2014, 54, 770–778.
[CrossRef]
18. Ruiz, A.; Ortiz, N.; Medina, A.; Kim, J.Y.; Jacobs, L. Application of ultrasonic methods for early detection of
thermal damage in 2205 duplex stainless steel. NDT E Int. 2013, 54, 19–26. [CrossRef]
19. Fierro, G.M.; Ciampa, F.; Ginzburg, D.; Onder, E.; Meo, M. Nonlinear ultrasound modelling and validation
of fatigue damage. J. Sound Vib. 2015, 343, 121–130. [CrossRef]
20. Ding, X.; Feilong, L.; Youxuan, Z.; Yongmei, X.; Ning, H.; Peng, C.; Mingxi, D. Generation Mechanism of
Nonlinear Rayleigh Surface Waves for Randomly Distributed Surface Micro-Cracks. Materials 2019, 11, 644.
[CrossRef]
21. Wan, X.; Zhang, Q.; Xu, G.; Tse, P. Numerical Simulation of Nonlinear Lamb Waves Used in a Thin Plate for
Detecting Buried Micro-Cracks. Sensors 2014, 14, 8528–8546. [CrossRef] [PubMed]
22. Melchora, J.; Parnell, W.; Bochud, N.; Peralta, L.; Rusa, G. Damage prediction via nonlinear ultrasound:
A micro-mechanical approach. Ultrasonics 2019, 93, 145–155. [CrossRef] [PubMed]
23. Šofer, M.; Ferfecki, P.; Fusek, M.; Šofer, P.; Gnatowska, R.; Malujda, I.; Dudziak, M.; Krawiec, P.; Talaśka, K.;
Wilczyński, D.A. The numerical analysis of the interaction between the low-order Lamb wave modes and
surface breaking cracks. MATEC Web Conf. 2019, 254. [CrossRef]
88
Appl. Sci. 2020, 10, 1982
24. Cantrell, J.H.; Yost, W.T. Nonlinear ultrasonic characterization of fatigue microstructures. Int. J. Fatigue 2001,
23, 487–490. [CrossRef]
25. Shui, G.; Wang, Y.; Qu, J. Advances in nodestructive test and evaluation of material degradation using
nonlinear ultrasound. Adv. Mech. 2005, 35, 54–70.
26. Jiao, J.; Lv, H.; He, C.; Wu, B. Fatigue crack evaluation using the non-collinear wave mixing technique.
Smart Mater. Struct. 2017, 26, 65–71. [CrossRef]
27. Landau, L.D.; Pitaevskii, L.P.; Kosevich, A.M.; Lifshitz, E.M. Theory of Elasticity; Elsevier Ltd.: Oxford,
UK, 1986.
28. Van Den Abeele, K.A.; Johnson, P.A.; Sutin, A. Nonlinear Elastic Wave Spectroscopy (NEWS) Techniques to
Discern Material Damage, Part I: Nonlinear Wave Modulation Spectroscopy (NWMS). Res. Nondestruct. Eval.
2000, 12, 31–42. [CrossRef]
29. Van Den Abeele, K.; Breazeale, M.A. Theoretical model to describe dispersive nonlinear properties of lead
zirconate–titanate ceramics. J. Acoust. Soc. Am. 1996, 99, 1430–1437. [CrossRef]
30. Jingpin, J.; Xiangji, M.; Cunfu, H.; Bin, W. Nonlinear Lamb wave-mixing technique for micro-crack detection
in plates. NDT E Int. 2016, 85, 63–71. [CrossRef]
31. Ostrovsky, L.A.; Johnson, P.A. Dynamic nonlinear elasticity in geomaterials. Riv. Nuovo Cimento della Soc.
Ital. Fis. 2001, 24, 1–46.
89
applied
sciences
Article
Bonding Characteristics of Single- and Joggled-Lap
CFRP Specimens: Mechanical and
Acoustic Investigations
Claudia Barile * , Caterina Casavola , Vincenzo Moramarco , Carmine Pappalettere
and Paramsamy Kannan Vimalathithan
Dipartimento di Meccanica, Matematica e Management, Politecnico di Bari, Via E. Orabona 4, 70125 Bari, Italy;
[email protected] (C.C.); [email protected] (V.M.); [email protected] (C.P.);
[email protected] (P.K.V.)

Received: 30 January 2020; Accepted: 3 March 2020; Published: 5 March 2020
Abstract: Two different configurations of adhesive-bonded carbon fiber-reinforced plastic (CFRP)

specimens, joggled lap-joint specimens and single lap-joint specimens, are mechanically tested.
The mechanical tests show that the joggled lap specimens have lower strength than the single
lap specimens. The damage modes in both the specimens are analysed by the Acoustic Emission
descriptors recorded during the mechanical tests. The acoustic data as cumulative counts and
cumulative energy show the critical points of failure in both the specimen groups under loading.
Moreover, they also show that the damage modes in both the specimens are dissimilar. Finally, the data
provided by acoustic emission descriptors are verified by fractographic analysis on the failed surface.
Keywords: acoustic emission; acoustic energy; acoustic counts; CFRP; single lap joint; joggled lap
joints; fractography
1. Introduction
The advancements in designing composite materials, particularly carbon fiber-reinforced plastic
(CFRP), have increased their usage tremendously in aerospace applications [1]. CFRPs are not only
sought for their high specific strength and stiffness, but also for their feasibility in forming different
structures through adhesive bonding, fasteners or the combination of both [2,3]. This paves way for
easy application of composite panels for repairing, thereby largely reducing both the manufacturing
costs and the idle time for the flight. Each of these advancements in CFRPs has equally created
complications in predicting their strength and structural integrity, particularly in composite joints [4].
Adhesive bonds or the combination of adhesive and fasteners are preferred over the conventional
bolting and riveting of composite structures, as of late [5,6]. The stress concentration around the
circumference of the fastener holes and the non-circumferential stress distribution around the rivets/bolts
can accelerate structural failure in the composite materials [7–9]. However, in a bonded joint, uniform
stress distributions can be obtained. Their improved specific strength can be attributed to peel
stress and shear stress of the adhesive used. Nonetheless, the bonded joints have their limitations.
During shearing of the adhesive, stress energy is released in three different modes: during the
adhesive opening mode, sliding mode and tearing mode [10–12]. Moreover, the specific strength
of the composite structure is governed by the adhesive thickness. Monitoring the bonding quality,
prediction of damage propagation through the adhesives and identifying low impact damages have
become tedious. This compels the researchers to search for different damage characterization and
integrity monitoring techniques.
91
Appl. Sci. 2020, 10, 1782
Acoustic emission (AE) has become one of the most sought-after non-destructive evaluation
technique in recent years to characterize the damage progression in a composite structure. It is one
of the very few passive non-destructive evaluation techniques, which can provide details of the
damage progression during its entire loading history. When a material is strained, it releases sudden
energy in the form of elastic waves [13–15]. The AE technique records these elastic waves in terms
of both signal-based waveforms and parameter-based data such as peak amplitude, counts, energy,
duration and so on (which can be collectively called AE descriptors). Since the AE records the elastic
waves, this technique is very sensitive, even to transient signals [16–18]. This makes AE one of the
most efficient and reliable techniques for characterizing damage progression.
However, for a composite material under loading, it is challenging to associate different AE
descriptors recorded with different types of damage mode. Although researchers have been working
tirelessly over the past 20 years and have identified innovative ways to associate the recorded AE
signals with the damage modes, this still remain in debate [19,20]. Some researchers have associated
specific peak amplitudes and peak frequencies to the specific types of damage modes such as matrix
cracking, delamination and fiber breakage [21,22]. Others are not inclined towards this idea and
encompass the necessity of using different signal-based descriptors such as frequency centroid and
weighted peak frequency [23,24]. In terms of common ground, numerous researchers alike have agreed
upon the reliability of using cumulative counts and cumulative acoustic energy [14,17,21]. The radical
change in the cumulative counts/energy with respect to time or displacement can identify the critical
damage points and help predict larger damage to the structure.
The question that persists is whether the AE technique can identify the different damage modes
in a complicated material such as composite materials. In our previous works, the signal-based AE
parameter—the wavelets and parameter based function named as the ‘Sentry Function’—have been
used for characterizing the damage modes [16,17]. Moreover, numerical and finite element models
were also designed for mode I delamination tests [6]. This work differs in a way that the relative
change in cumulative counts and energy are directly related to the damage modes. In the present
research work, the cumulative counts and cumulative energy distribution have been used to identify
the different damage modes in a CFRP bonded with adhesives. For this purpose, two different types of
CFRP-bonded specimens are used: joggled-lap shear (JLS) and single-lap shear (SLS). The fracture
analysis of the specimens is performed by using an optical microscope. The main objective of this
study is to understand the critical points of failure and the differences in the damage modes in both
JLS and SLS specimens using the AE technique.
2.1. Materials
For preparing the specimens, a prepreg fabric laminate with ER450 epoxy matrix (SAATI CIT
CC206 ER450 43%) was used. The carbon fiber configuration is stitched with layers of fibers overlapping
one other. The nominal ply thickness is 0.244 mm. The composite laminates for both JLS and SLS
specimens (Figure 1) were cured by the autoclave method. Other details such as number of plies and
geometry of the adherend can be found in Tables 1 and 2, respectively. The specimens were prepared
by following the ASTM D5868–Standard Test Method for Lap Shear Adhesion for Fiber-Reinforced
Plastic (FRP) Bonding [25]. The adhesive used for this study has a shear strength of 25 MPa and a peel
strength of 65 MPa. The thickness of the adhesive coated and the area of the adhesive coated are also
provided in Tables 1 and 2.
92
Appl. Sci. 2020, 10, 1782
(a)
(b)
Figure 1. Single-lap shear (SLS) (a) and joggled-lap shear (JLS) (b) specimens’ ’nominal dimensions.
Table 1. JLS specimens’ geometry and characteristics.

’
Flat Adherend
Length (mm) Width (mm) Thickness (mm) No. of Plies Stacking sequence
101.6 ± 0.12 26.09 ± 0.07 2.0 ± 0.04 8 [+45/−45]3 /−45/+45
Curved Adherend
101.6 ± 0.17 26.09 ± 0.05 1.5 ± 0.02 6 [+45/+45/−45]S
Overlapping Region (Adhesive)
Length (mm) Width (mm) Thickness (mm)
26 ± 0.21 26.09 ± 0.06 3.67 ± 0.05
Table 2. SLS specimens’ geometry and characteristics.
Upper Adherend
101.6 ± 0.11 25.33 ± 0.12 1.3 ± 0.05 5 +45/+45/+45/−45/+45
Lower Adherend
101.6 ± 0.09 25.33 ± 0.14 6.4 ± 0.12 26 +45/[+45/−45]12 /+45
Overlapping Region (Adhesive)
Length (mm) Width (mm) Thickness (mm)
26 ± 0.12 25.33 ± 0.25 8.5 ± 0.11
93
Appl. Sci. 2020, 10, 1782
The adhesive was cured at temperature of 65 ◦ C for 1 hour, then the specimens were tested after
5 days. Three specimens in each type have been studied. The JLS and SLS bonded laminates are
provided in Figure 2.
b 3mm
d 3mm
Figure 2. (a) SLS specimen, (b) joint overlap of SLS specimen, (c) JLS specimen, (d) joint overlap of
JLS specimen.
The reason behind using large variation in thickness between the upper adherend and lower
adherend in the SLS specimens (Figure 1a) is to simulate the configuration of the composites used in
aerospace structures. SLS configuration, in particular, represents the bonding between the flange of the
lower spar of the fuselage and the front bulkhead. JLS configuration (Figure 1b) represents the bonding
between the fuselage and the half-wing. Although there is no standard procedure that has been
followed for setting the thickness and number of plies, the test procedures were carried out carefully as
per the ASTM D5868 standards. Moreover, it has been indicated by many researchers that the thickness
of the adherend does not play a significant role in load distribution under displacement-controlled
testing conditions [26]. The entire load is distributed through the thickness of the adhesive, which is
the crucial factor in this research work.
2.2. Testing Methods

The tension was applied to the specimens under displacement-controlled mode at a speed of
13 mm/min, in accordance with the ASTM D5868 standard and previous works presented in the
literature [27,28]. Although the value of the strain rate suggested from the standard method is quite
high, it does not significantly influence the mechanical response of the lap joints [29]. The test was
carried out in the INSTRON servo-hydraulic testing machine with a maximum load capacity of 100 kN.
To record the acoustic activities under loading, two narrow-band general purpose AE sensors were
used: R30a (Physical Acoustics, MISTRAS Group, NJ, USA). The sensor has an operating range of
150 kHz to 400 kHz and a resonant frequency of 300 kHz. The reason for selecting the narrow band
sensor is because in CFRPs, the recorded AE signals mostly are under the frequency band of 400 kHz.
To avoid unnecessary noise, the narrow-band sensor was selected. For the same purpose, the sampling
94
Appl. Sci. 2020, 10, 1782
rate was set as 1 MSPS (mega samples per second). The threshold of AE acquisition was set at 35 dB
and the recorded signals were amplified by 40 dB through a 2/4/6 AE (Acoustic Emission) preamplifier.
The sensors were placed at 40 mm either side from the centre of the adhesive overlapping region
(Figure 3). Silica gel was interposed between the surface of the sensor and the specimens, in order to
improve the coupling of the elements. This location was selected based on the length of the overlapping
region and to record the AE signals produced within the overlapping region. As indicated in the
previous section, the load is carried out only through the adhesives under displacement-controlled
mode. For this reason, the AE signals recorded beyond the adhesive overlapping region are discarded.
In the fractured specimen, not many damages were observed except in the area covering the overlapping
adhesive layer. The surface of the specimen post rupture has been analysed by means of an optical
microscope NIKON SMZ800. Moreover, the region of the adhesive and the fractured surfaces have
also been studied after the shearing failure.
Figure 3. Testing rig with acoustic emission (AE) sensors mounted on the specimen.
It is well known that the distance from the sensor to the acoustic source affects the propagation of
the AE signals. This can result in the attenuation of the recorded AE signals. To avoid this attenuation,
the AE sensors were calibrated for the attenuation in signal with respect to the distance from the
sensor [30]. A pencil break test was performed, which normally produces a peak amplitude of 95 dB to
99 dB at varying distances from the sensor on the SLS and JLS specimens. The table of this calibration
was then fed to the PAC PCI 2 data acquisition system, which automatically calibrated the attenuation
of the signal.
2.3. Acoustic Emission (AE) Characterization =

The recorded acoustic data used for this study are cumulative counts and cumulative energy.
Counts are the number of instances of the recorded acoustic signal crosses the set threshold.
95
Appl. Sci. 2020, 10, 1782
The acoustic energy is the total energy of the recorded acoustic event over the time period of the
signal t0 to ti . If Ui is the transient voltage of the recorded acoustic event, then the acoustic energy EAE
is given by Equation (1) [31]:
Zti
EAE = Ui2 (t)dt (1)
t0
3. Results and Discussions
3.1. Mechanical Test Results

The bonding characteristics of the JLS and SLS specimens were tested by applying tensile load
under displacement-controlled mode as per the ASTM D5868 standard. When the tensile load was
applied, the entire load was distributed through the adhesive region. It has been reported by several
researchers with solid evidences that the thickness of the adherend does not have a significant role in the
distribution of load under shearing [26]. There could also be more than one peak load value, with each
of them contributing to the damage characteristics or the bonding characteristics of the specimen.
Figure 4 shows the load vs. time curve for the specimens. In the research work presented,
more than one peak can be observed in the load vs. time curve for both specimens. It must be noted
that although there are several small peaks observed in both JLS and SLS specimens in Figure 4,
the initial peak load is considered as the initial rupture point and the final peak load is considered as
the final rupture point. Since these mechanical results provide only a limited amount of information
on the damage characteristics, the acoustic emission results have been used in the subsequent sections.
Table 3 shows the mechanical characteristics of both JLS and SLS specimens.
Figure 4. Load vs. time curve: (a) JLS Specimens; (b) SLS Specimens.
96
Appl. Sci. 2020, 10, 1782
Table 3. Damage characteristics in JLS and SLS specimens.
Peak Load (kN)

Specimen
Initial Rupture Final Rupture
JLS 1 6.02 4.09
JLS 2 2.67 3.89
JLS 3 3.86 5.57
Mean 4.18 4.52
Std. Dev. 1.70 0.92
SLS 1 7.05 7.75
SLS 2 7.06 5.96
SLS 3 5.93 7.36
Mean 6.78 7.02
Std. Dev. 0.75 0.94
The first major observation from the mechanical test results is the varying peak load values of
the same specimen group, both in JLS and SLS. This is common when it comes to the practice of
adhesive bonded specimens [27–29] because the bond quality is not only determined by the curing
characteristics, the thickness of the adhesive and other mechanical characteristics, but also on the
efficiency of the user who handles the adhesives. The adhesives are normally mixed using a hand
mixer rather than a mechanical controller owing to its high viscosity and requirements. Thus, it is
inevitable that a considerable amount of difference can be observed between the adhesive bonded
specimens of the same group. This is the reason why there is a difference in both the peak loads at
initial and final ruptures and they are not similar between the same specimen group (refer to Figure 4
and Table 3). Furthermore, under this view, an increase in the number of test specimens would increase
the accuracy of the evaluation of the scattering but it not will reduce the latter.
As indicated in the previous sections, the thickness of the adherend does not generally
affect the strength of the adhesive bonded shear specimens under displacement-controlled mode.
Nonetheless, the thickness of the adhesive carries the majority, if not all the load. In the SLS specimens,
owing to their larger thickness than JLS specimens, higher peak loads are accounted for. The average
peak load at the initial rupture in SLS specimens is 6.78 kN, while the JLS specimens have an average
peak load of 4.18 kN. Similarly, at the final rupture stage, SLS specimens have an average peak
load of 7.02 kN, while JLS specimens have 4.52 kN. In SLS specimens, the load is carried along the
cross-sectional area of the adhesive region. However, in the JLS specimens, the shear strength of the
specimen is also governed by the end curvature effect. This is explained by Taib et. al. [11] in their
research work. Single lap joints, when loaded, transfer most of the load in shearing. Joggled lap
joints, however, transfer most of the load in peeling. This is also the reason why the joggled lap joints
have lower displacement when compared to the single lap joint specimens. The joggled joints suffer
lateral deflections rather than shearing, which result in inducing high peel stress on the neutral axis.
This results in the crack initiating at the joggled knee of the curvature region. This will result in low
peak loads, which are observed in the present research work (see Table 3 and Figure 4). Similar results
have been presented by Kishore and Prasad [32], who also addressed the effect of the eccentricity of
the joggle in the strength of the joggled lap specimens. It has been addressed by both Taib et. al [32]
and Kishore and Prasad [33] that the joggled lap specimens will have lower strength in comparison
with the single lap joints. The same has been observed in the presented research work.
3.2. Acoustic Emission Results: Cumulative Counts and Cumulative Energy

The damage characteristics or the mode of failure in the adhesive bonded specimens cannot be
explained through the mechanical characteristics. Furthermore, predicting the failure at an early stage
is also not easy by studying only the mechanical characteristics. It can be explained by the acoustic
emission characteristics. The load vs. time curve of the JLS specimens is plotted over the cumulative
97
Appl. Sci. 2020, 10, 1782
counts and the cumulative energy of the AE recorded. It is presented in Figure 5. Similarly, the SLS
specimens are presented in Figure 6.
Figure 5. Load, cumulative counts, cumulative energy vs. time: (a) JLS 1; (b) JLS 2; (c) JLS 3.
98
Appl. Sci. 2020, 10, 1782
Figure 6. Load, cumulative counts, cumulative energy vs. time: (a) SLS 1; (b) SLS 2; (c) SLS 3.
In Figure 5, it can be observed that whenever there is a load peak in the load vs. time curve in any
of the JLS specimens, the cumulative counts and cumulative energy show a steep increase. The initial
load drops in all JLS 1, JLS 2 and JLS 3 at 1.1 s, 0.1 s and 0.35 s, respectively, have been associated with
the increase in cumulative counts and cumulative energy. Similarly, at the final stages of loading prior
to failure 1.8 s, 1.35 s and 2.25 s, respectively, for JLS 1, JLS 2 and JLS 3 are also associated with the
increase in cumulative counts and energy. However, there are some instances, particularly around 1.2 s
in the JLS 2 specimen, where the cumulative counts and energy increased unexpectedly without any
change in the load vs. time curve. This phenomenon could not be explained without utilizing another
characterizing technique or a deeper investigation like the frequency content of the acoustic data.
Nonetheless, analysing the waveform of those signals in their time-frequency domain can also provide
information about whether these signals are representatives of machine or surrounding noises. This is
not the only information that can be obtained from the AE results. In JLS 1 specimen, the cumulative
energy increases steeply around 1.1 s duration and it remains almost constant until the specimen
reaches its final rupture region. Nonetheless, the cumulative counts kept increasing gradually beyond
the 1.1 s duration and the slope took a steep increase as the material reaches failure. This means that
a large number of counts have been recorded with very low energy from 1.1 s duration to the final
failure. In a non-linear material, such as plastics or composite materials, the large counts with low
99
Appl. Sci. 2020, 10, 1782
energy represent the matrix cracking. The ratio between the cumulative energy and the cumulative
count varies at different damage modes in a material. When this ratio exhibits a very low value, it can
be said that the energy is low with large counts. On that basis, this can be attributed to the matrix
cracking. Brittle fractures such as interlaminar crack growth and fiber breakage release large energy
acoustic events, which are responsible for the initial and final ruptures. Between the stages of the initial
failure and the final failure in the JLS 1 specimen, the material has experienced only matrix cracking.
This indicates that shearing has occurred in the adhesive region rather than peeling which attributes to
the low-energy acoustic events. This indicates that the JLS 1 specimen has carried more load before
failure than the JLS 2 and JLS 3 specimens due to the shearing mode of failure. This could be one of the
reasons why the JLS 1 specimen has a higher peak load when compared to JLS 2 and JLS 3.
In the cases of JLS 2 and JLS 3, they have similar load vs. time, cumulative counts and energy
patterns. Both these specimens experienced the initial rupture at a very early stage, 0.1 s for JLS 2 and
0.35 s for JLS 3. The cumulative counts and energy have also increased steeply at these time periods.
Beyond that point, the cumulative counts and the cumulative energy increase almost linearly with
one another. This represents the delamination in the adhesive bonded region. The adhesive peeling
is occurring at this stage, which is the reason for the simultaneous increase in the counts and energy.
While reaching the final stage of failure, JLS 2 and JLS 3 have probably experienced fiber breakage
which is indicated by the sudden increase in counts and energy near the final rupture region.
While looking at Figure 6, all the three specimens have almost similar patterns in terms of load vs.
time, cumulative counts and energy. However, the simultaneous increase in the counts and energy of
the SLS specimens must not be compared or to be confused with the JLS specimens. Although in both
the cases the cumulative counts and cumulative energy increased linearly, the slope of the increment
is entirely different. The slope of the increment in SLS specimens is higher than the JLS specimens.
Therefore, they do not represent the same damage mode although it looks similar. In the SLS specimens,
the cumulative counts and energy are very low during the initial stages of loading until the specimens
have suffered their initial rupture. In the SLS specimens, most of the load is distributed in shearing
and these low-energy and low-count AE signals indicate the shearing occurring in the adhesive region.
AE signals connecting to fiber breakages and transverse cracks growth release acoustic data with higher
energy and lower counts, owing to their lower order symmetric nature. Since the energy and counts
recorded are low during these stages, these signals are higher order asymmetric in nature that could be
connected only to shearing phenomena [20]. Beyond the region of the initial rupture, in all three SLS
specimens, the cumulative counts and energy start to increase simultaneously until they reach the final
failure. This indicates that the cumulative counts and energy were very low until the crack opens in
the adhesive. Generally, the crack openFs at the adhesive and adherend interface, which is governed
by the geometry of the overlapping area. Once the crack had opened, due to the brittle nature of the
adhesive, it progressed faster through the thickness of the adhesive. This interlaminar crack growth
always generates higher counts and higher energy. This pattern continued until the specimens failed
at the final rupture.
3.3. Fracture Surface Analysis

To confirm the brittle failure of the adhesives in the SLS specimens and to identify the peeling and
fiber breakage failure in JLS specimens, the fractured surfaces were observed under optical microscope.
The fractographic results of specimen JLS is presented in Figure 7.
100
Appl. Sci. 2020, 10, 1782
Figure 7. Fractographic analysis of JLS specimens. (a) Peeling at the knee of adhesive; (b) Debonding
of laminates; (c) Ruptured Fiber; (d) Fiber/Matric Debonding.
In Figure 7a, the adhesive layer peeling off at the knee of the curvature in the joggled overlapping
region can be observed. The debonding of the laminates along the applied adhesive layers can be
viewed in Figure 7b. The adhesive not only peeled off from the lamina but also inflicted fiber/matrix
debonding in the same lamina. Figure 7c is the zoomed view of the circled region in Figure 7b.
From Figure 7c, the ruptured fiber at the end of the lamina can be observed. Figure 7d shows the
presence of more fiber/matrix debonding and fiber breakage near the knee of the adhesive joint.
This consolidates the explanations provided in Section 3.2 regarding the failure modes by acoustic
emission parameters.
Similarly, the optical images of SLS specimens are presented in Figure 8. Figure 8a shows the
complete lack of matrix/fiber debonding in the area where the adhesive is applied. The fibers are left
unharmed in the fractured region of the adhesive. The adhesive has delaminated from the outermost
ply, but it did not affect the fiber/matrix integrity. Figure 8b shows the brittle fracture and the direction
of the crack grew along the width of the adhesive under loading. Figure 8c shows the fiber breakage at
the end of the specimen. Figure 8d is the zoomed image of the ruptured fibers. So, the fiber breakage
has occurred in the SLS specimens, nonetheless, only at the edge of the lamina.
101
Appl. Sci. 2020, 10, 1782
Figure 8. Fractographic analysis of SLS specimens. (a) Lack of Fiber/Matrix debonding; (b) Direction of
cracking growth along the length of adhesive; (c) Fiber breakage; (d) Close in view of fiber breakage.
Therefore, by comparing Figures 7 and 8, it can be concluded that the crack growth initiated at the
knee of the joggled lap in the JLS specimens, which in turn results in the peeling. The peeling also
has induced the debonding between the fiber and matrix in the area where the adhesive is applied.
However, in SLS specimens, the crack opened in the adhesive region and travelled through the width
of the specimen resulting in a brittle failure.
The fractographic results are sufficient to prove the damage modes explained in the acoustic
emission result sections. Thus, the AE proves to be a powerful tool, not only in predicting the failure in
an adhesive bonded FRP but also in identifying the different damage modes in the failure. By setting
up a proper experimental campaign and understanding the different AE descriptors, it is efficient to
analyse the failure modes in an entire structure comprising adhesive-bonded FRP.
4. Conclusions
The bonding characteristics of adhesive-bonded FRP specimens were tested using the acoustic
emission technique. The differences in the failure mechanisms were observed by comparing the
cumulative counts and cumulative energy under loading. The slope of the cumulative counts and
energy were different in JLS and SLS specimens indicating that both the specimen groups failed
under different damage modes. Moreover, the cumulative energy and cumulative counts can provide
information on the critical points of failure. In the joggled-lap specimens, the initial failure was due to
the peeling of the adhesive layer at the knee of the curvature region and it induced the fiber/matrix
debonding. The final failure was due to the fiber breakage in the lamina. In the single-lap specimens,
the initial failure was due to the crack opening of the adhesive and the crack propagated through
the width of the adhesive resulting in the final failure. The explanations provided by the AE results
were evidentially proved by the fractographic analysis of the specimens. Acoustic emission proves to
be a powerful tool in analysing the damage characteristics of fiber-reinforced plastics.
Author Contributions: Conceptualization, P.K.V. and C.B.; methodology, P.K.V., C.B. and V.M.; validation, C.B.
and P.K.V.; formal analysis, P.K.V. and C.B.; investigation, P.K.V. and C.B.; data curation, C.B., V.M. and P.K.V.;
writing—original draft preparation, P.K.V.; writing—review and editing, P.K.V. and C.B.; supervision, C.C. and C.P.;
project administration, C.C. and C.P. All authors have read and agreed to the published version of the manuscript.
102
Appl. Sci. 2020, 10, 1782

References
1. Sugita, Y.; Winkelmann, C.; La Saponara, V. Environmental and chemical degradation of carbon/epoxy lap joints
for aerospace applications, and effects on their mechanical performance. Compos. Sci. Technol. 2010, 70, 829–839.
[CrossRef]
2. Campilho, R.D.S.G.; de Moura, M.F.S.F.; Domingues, J.J.M.S. Using a cohesive damage model to predict the
tensile behaviour of CFRP single-strap repairs. Int. J. Solids Struct. 2008, 45, 1497–1512. [CrossRef]
3. Yu, Q.; Gao, R.; Gu, X.; Zhao, X.; Chen, T. Bond behavior of CFRP-steel double-lap joints exposed to marine
atmosphere and fatigue loading. Eng. Struct. 2018, 15, 76–85. [CrossRef]
4. Kucher, N.K.; Zemtsov, M.P.; Zarazovskii, M.N. Deformation behavior and strength of unidirectional carbon
fiber laminates. Mech. Compos. Mater. 2006, 42, 407–418. [CrossRef]
5. Tong, L.; Steven, G.P. Analysis and Design of Structural Bonded Joints; Kluwer Academic Publishers:
Boston, MA, USA, 1999.
6. Barile, C.; Casavola, C.; Gambino, B.; Mellone, A.; Spagnolo, M. Mode-I fracture behavior of CFRPs:
Numerical model of the experimental results. Materials 2019, 12, 513. [CrossRef] [PubMed]
7. Benatar, A.; Gillespie, J., Jr.; Kedward, K. Joining of composites. In Advanced Composites Manufacturing;
Gutowski, T.G., Ed.; John Wiley: New York, NY, USA, 1997; pp. 487–512.
8. Jones, R.M. Mechanics of Composite Materials, 2nd ed.; Taylor and Francis Publishers: New York, NY, USA, 1999.
9. Barile, C.; Casavola, C.; Pappalettera, G.; Vimalathithan, P.K. Characterization of adhesive bonded CFRP
laminates using full-field digital image stereo-correlation and finite element analysis. Compos. Sci. Technol.
2019, 169, 16–25. [CrossRef]
10. Li, G.; Pang, S.S.; Woldesenbet, E.; Stubblefield, M.A.; Mensah, P.F.; Iberkwe, S.I. Investigation of prepreg
bonded composite single lap joint. Compos. Part B Eng. 2001, 32, 651–658. [CrossRef]
11. Taib, A.A.; Boukhili, R.; Achiou, S.; Gordon, S.; Boukehili, H. Bonded joints with composite adherends. Part
I. Effect of specimen configuration, adhesive thickness, spew filler and adherend stiffness on fracture. Int. J.
Adhes. Adhes. 2006, 26, 226–236. [CrossRef]
12. Yang, C.; Tomblin, J.S. Investigation of Adhesive Behavior in Aircraft Applications; DOT/FAA/AR-01/57;
US Department of Transportation Federal Aviation Administration, Office of Aviation Research:
Washington, DC, USA, 2001.
13. Grosse, C. Introduction. In Acoustic Emission Testing; Grosse, C., Ohtsu, M., Eds.; Springer:
Berlin/Heidelberg, Germany, 2008.
14. Balázs, G.L.; Grosse, C.; Koch, R.; Reinhardt, H.W. Damage accumulation on deformed steel bar to concrete
interaction detected by acoustic emission technique. Mag. Concr. Res. 1996, 48, 311–320. [CrossRef]
15. Grosse, C.; Linzer, L. Signal-based AE analysis. In Acoustic Emission Testing; Grosse, C., Ohtsu, M., Eds.;
Springer: Berlin/Heidelberg, Germany, 2008.
16. Barile, C.; Casavola, C.; Pappalettera, G. Acoustic emission waveform analysis in CFRP under Mode I test.
Eng. Fract. Mech. 2018, 210, 408–413. [CrossRef]
17. Barile, C.; Casavola, C.; Pappalettera, G.; Vimalathithan, P.K. Damage characterization in composite materials
using acoustic emission signal-based and parameter-based data. Compos. Part B-Eng. 2019. [CrossRef]
18. Barile, C.; Casavola, C.; Pappalettera, G.; Vimalathithan, P.K. Investigation of structural integrity of composite
materials using wavelet packet transform. Procedia Eng. 2019, 17, 582–588. [CrossRef]
19. Hamstad, M.A. Frequencies and Amplitudes of AE Signals in a Plate as a Function of Source Rise Time.
In Proceedings of the 29th European Conference on Acoustic Emission Testing; Available online: https:
//www.ndt.net/events/EWGAE%202010/proceedings/papers/20_Hamstad.pdf (accessed on 4 March 2020).
20. Oz, F.E.; Ersoy, N.; Lomov, S.V. Do high frequency acoustic emission events always represent fibre failure in
CFRP laminates? Compos. Part A Appl. Sci. 2017, 103, 230–235. [CrossRef]
21. Liu, P.F.; Chu, J.K.; Liu, Y.L.; Zheng, J.Y. A study on the failure mechanisms of carbon fiber/epoxy composite
laminates using acoustic emission. Mater. Design 2012, 37, 228–235. [CrossRef]
103
Appl. Sci. 2020, 10, 1782
22. De Groot, P.J.; Wijnen, P.A.M.; Janssen, R.B.F. Real-time frequency determination of acoustic emission
for different fracture mechanisms in carbon/epoxy composites. Compos. Sci. Technol. 1995, 55, 405–412.
[CrossRef]
23. Njuhovic, E.; Bräu, M.; Wolff-Fabris, F.; Starzynski, K.; Altstädt, V. Identification of failure mechanisms
of metallised glass fibre reinforced composites under tensile loading using acoustic emission analysis.
Compos. Part B-Eng. 2015, 81, 1–13. [CrossRef]
24. Alia, A.; Fantozzi, G.; Godin, N.; Osmani, H.; Reynaud, P. Mechanical behaviour of jute fibre-reinforced
polyester composite: Characterization of damage mechanisms using acoustic emission and microstructural
observations. J. Compos. Mater. 2019. [CrossRef]
25. ASTM D5868-01(2014). Standard Test Method for Lap Shear Adhesion for Fiber Reinforced Plastic (FRP)
Bonding; ASTM International: West Conshohocken, PA, USA, 2014; Available online: www.astm.
org (accessed on 4 March 2020).
26. Mangalgiri, P.D.; Johnson, W.S.; Everett, R.A., Jr. Effect of adherend thickness and mixed mode loading on
debond growth in adhesively bonded composite joints. J. Adhes. 1987, 23, 263–288. [CrossRef]
27. Hsiao, K.T.; Alms, J.; Advani, S.G. Use of epoxy/multiwalled carbon nanotubes as adhesives to join graphite
fibre reinforced polymer composites. Nanotechnology 2003, 14, 791–793. [CrossRef]
28. Avila, A.F.; Bueno, P.O. An experimental and numerical study on adhesive joints for composites.
Compos. Struct. 2004, 64, 531–537. [CrossRef]
29. Liu, X.; Shao, X.; Li, Q.; Sun, G. Experimental study on residual properties of carbon fibre reinforced
plastic (CFRP) and aluminum single-lap adhesive joints at different strain rates after transverse pre-impact.
Compos. Part A Appl. Sci. Manuf. 2019, 124, 105372. [CrossRef]
30. Finkel, P.; Mitchell, J.R.; Carlos, M.F. Experimental study of ‘Auto Sensor Test-Self Test Mode’ for acoustic
emission system performance verification. AIP. Conf. Proc. 2000, 509, 1995–2002.
31. Harris, D.O.; Bell, R.L. The measurement and significance of energy in acoustic-emission testing. Exp. Mech.
1977, 17, 347–353. [CrossRef]
32. Kishore, A.N.; Prasad, N.S. An experimental study of Flat-Joggle-Flat bonded joints in composite laminates.
Int. J. Adhes. Adhes. 2012, 35, 55–58. [CrossRef]
33. Taib, A.A.; Boukhili, R.; Achiou, S.; Boukhili, H. Bonded joints with composite adherends. Part II. Finite
element analysis of joggle lap joints. Int. J. Adhes. Adhes. 2006, 26, 237–248. [CrossRef]
104
applied
sciences
Review
Mechanical Strength Evaluation of Elastic Materials
by Multiphysical Nondestructive Methods: A Review
Huiting Huan 1,2 , Lixian Liu 1,2, *, Andreas Mandelis 2 , Cuiling Peng 1 , Xiaolong Chen 1 and
Jinsong Zhan 1, *
1 School of Mechano-Electronic Engineering, Xidian University, Xi’an 710071, China;
[email protected] (H.H.); [email protected] (C.P.); [email protected] (X.C.)
2 Center for Advanced Diffusion-Wave and Photoacoustic Technologies (CADIPT), Department of Mechanical
and Industrial Engineering, University of Toronto, Toronto, ON M5S 3G8, Canada;
[email protected]
* Correspondence: [email protected] (L.L.); [email protected] (J.Z.)

Received: 24 December 2019; Accepted: 19 February 2020; Published: 27 February 2020
Abstract: The main purpose of industrial nondestructive testing (NDT) is to diagnose the stability,
reliability and failure probability of materials, components and structures. Industrial component
mechanical strength is one of the most important properties NDT is used to characterize. Subtle
but perceptible changes in stress-strain behavior can be reliable indicators of defect formation. A
detailed review on the state-of-the-art NDT methods using optical-radiation, photoacoustic, and
photothermal techniques for mechanical strength evaluation and defect pre-diagnosis is presented in
this article. Mechanical strength is analyzed in terms of the deformation/strain field, the stress-strain
relation, and the residual stress in an elastic material subjected to tensile or compressive loading,
or impact. By introducing typical NDT experiments, the history and features of each methodology
are revisited and typical applications are discussed. This review also aims to be used as a reference
toward further research and development of NDT technologies characterizing mechanical strength of
materials and components.
Keywords: stress-strain relation; nondestructive testing; optical-radiation; photoacoustic; photothermal
1. Introduction
Nondestructive testing/evaluation (NDT/NDE) is a collection of scientific techniques which are
able to evaluate the properties and condition of materials without inflicting permanent damage on
them. They were originally designated for inspecting parts, products, and structures of high value.
The number of published research reports on NDT rapidly increased in the beginning of the 20th
century because of industrialization and the two World Wars. There have been wide applications
for NDT technologies in numerous fields such as industrial manufacturing [1,2], aerospace and
aeronautics [3], civil engineering [4,5], and material science and technology [6,7]. Practical NDT
techniques used for materials characterization can reduce maintenance costs and enforce lifetime
management [8]. From the viewpoint of NDT on components made from solid elastic materials such
as alloys, polymers and composites, quality is assessed by inspecting the possible existence of defects
and changes in mechanical strength [9,10]. Conventionally, the mechanical strength of a material
is measured with a tensile machine and the stress-strain relation is determined by dynamic tensile
testing from load-free to failure [11–13]. The stress-strain data are recorded on samples of cylindrical,
center-necked shape and contact stress/strain measurement modules such as adhesive transducers are
usually part of the machine. Such tensile tests are destructive because samples undergo irreversible
deformation during the tensile test. To resolve the problem associated with mechanical strength testing,
105
Appl. Sci. 2020, 10, 1588
several nondestructive methodologies have been developed for the purpose of stress-strain evaluation.
Some of them focus on noncontact approaches which characterize the change of stress-strain state
by other physical fields during tensile testing. Other methodologies do not involve tensile testing
but are devoted to measuring surface or internal residual stress distribution in an elastic sample
both qualitatively and quantitatively. Nevertheless, the ultimate goal always is pre-diagnosing the
mechanical strength of a component before actual cracks and failure occur.
This review article discusses typical NDT methodologies that measure stress or strain by means of
both contact and noncontact configurations. Noncontact methods can be used under extreme testing
conditions or for in-situ inspection. Features of these noncontact NDT modalities are discussed with
regard to their physical basis and state-of-the-art applications. No ranking of the various techniques is
made or conclusions drawn with respect to their relative advantages because they all have distinctive
features and limitations. It is expected that this review article will help interested readers to learn
about the underlying physical principles, development histories and modern applications of a number
of major NDT methods used for industrial quality control to-date.
2. Noncontact Mechanical Property Testing Methods
2.1. Optical/Radiation Techniques

The stress-strain relation of elastic materials is important because it is not only a way to characterize
the hardness and brittleness of raw materials, but also reveals the mechanical status of a sample under
test [14,15]. Normally, the strain value is measured with a strain gauge affixed on the sample surface.
Under extreme conditions such as high temperature, strain sensors along with special designs of
insulating and high-Curie-temperature materials, are used for stress-strain measurements in contact
configurations [16–19]. Such a scheme cannot satisfy the requirements of many in-field applications due
to difficulties with transducer bonding. To perform totally noncontact measurements, digital-image
correlation (DIC), a.k.a. digital laser speckle (DLS), a purely optical imaging technique, was devised.
This method was first proposed by Lyons et al. [20] with the basic principle shown in Figure 1.
To avoid blackbody radiation from a high temperature surface, a visible-range expanded-beam laser
is usually adopted in DIC measurement to generate diffuse-reflection-induced speckle patterns. A
camera (with optical filter) is placed in front of the sample to record speckle images during tensile tests.
By cross-correlating the digital images before and after tensile loading, the displacement field can be
determined using the following equation [21]:
! !
ux 1 ∆mPx
= (1)
uy Vi ∆nP y
where ux and uy represent the two orthogonal displacement fields in the camera, Vi is the image
size and ∆m, ∆n are the respective cross-correlation values. P stands for the pixel pitch which is
determined by the resolution of camera. The ultimate strain field of interest can be further derived
from the gradients of ux and uy . Using the proposed method, Anwander et al. achieved strain
measurements in tensile tests at 1200 Celsius degree with aluminum [21]. Four years later, Vǒlkl
et al. further increased the temperature to 3000 degree with an Ohmic heater and conducted strain
field measurements using a similar DIC schematic [22]. Zhu et al. developed a time-dependent
DIC system and dynamically recorded the change of strain field under a constant loading rate with
respect to non-metallic composites [23]. By using laser extensometry, the local strain rate value can
be characterized by time-dependent phase shift of correlated signal thus real-time measurement
is possible.
106
Appl. Sci. 2020, 10, 1588
Figure 1. The schematic of digital-image correlation (DIC) strain evaluation.
Another interesting topic is exploring the stress-strain behavior of alloys under high temperature
which can reveal the strength of materials under extreme conditions, and also investigating how they
react to thermal stress at such high temperatures. Such a thermal reactive character can be used to
determine thermal compatibility with other materials in a heterogeneous structure [24]. Pan et al.
introduced an optical bandpass imaging system shown in Figure 2 to evaluate surface thermal strain
without a surface irradiating laser source [25]. The bandpass filter allows only the violet-to-blue
light to pass and avoids interferences from blackbody radiation at high temperatures. The use of an
incoherent light source makes the system less costly but still sensitive [26]. The DIC of incoherent
speckle imaging technique can also be used for real-time strain recording, as proposed by Yang et al.,
to assist optimization of supersonic aircraft coating-material thermal shock processes [27].
Figure 2. Strain measurement configuration with blue light source, redrawn from [24].
Apart from industrial applications, noncontact optical measurements can also be used to perform
strength evaluation of concrete in civil engineering using
Ψ an alternative configuration of DLS. A
narrow collimated visible laser beam impinges perpendicularly on a sample surface, while the
camera monitors and captures scattered laser light from the incident spot at a non-normal angle
inΨΦa scanning patterns over the surface so as to determine roughness mappings. This method is
usually named laser triangulation. A series of tests with laser triangulation scanning were carried
out [28–30] to systematically quantify mechanical bonding strength between different concrete layers.
The correlation between adhesive performance of concrete and roughness, grain size, texture etc. was
investigated. Optical non-contact measurements represent an accurate and effective modality for
evaluating mechanical strength of inelastic building materials.
Ψ reconstruct
DIC is a successful approach to remotely detect small displacements and Ψ strain fields. It
relies heavily on the resolution, processing speed and aberration performance of the camera. However,
there are some limitations with respect to its applications. The processing algorithm requires input of
107
Appl. Sci. 2020, 10, 1588
speckle images, therefore, the signal-to-noise ratio from highly reflective surfaces can easily deteriorate.
Furthermore, the strain field inside the sample cannot be visualized with purely optical means.
The existence of local stress/strain should be justified with other techniques to yield a comprehensive
stress-strain evaluation. Ionizing radiative electromagnetic waves have deep penetration and thus can
be used for internal stress-strain evaluation.
Among the established methods, X-ray diffraction is regarded as a classic and powerful tool
for diagnosing subsurface or internal stress without contact. Having very small wavelength, X-ray
radiation penetrates metallic materials and diffracts from irregular-lattice area caused by internal
stress/strain, a common case in the production of composites consisting of non-metal and metal layers
with different thermal expansion/contraction properties Ψ [31,32].
X-rays have also been shown to be capable of evaluating the stress-strain relation from mechanical
tests by single-angle, two-angle and sin2 Ψ scattering modes [33–35]. The schematic of X-ray diffraction
and
ΨΦ the associated relevant coordinates is shown in Figure 3. The basic principle for evaluating strain
due to external load can be represented as a function of latitudinal lattice spacing dΨΦ as [32,33]:
dψφ − d0
εψφ = (2)
d0
where, d0 is the lattice spacing before deformation. The strain component Ψ can
Ψ be determined by
analyzing the X-ray radiation signal in different directions, i.e., at angles Ψ and Ψ. Based on this basic
principle, a numerical analysis was developed to obtain tomographic images of stressed materials in
recent reports [36–39]. X-ray computed tomography was further extended to evaluate mechanical
performance and defects. Youssef et al. [40] and Patterson et al. [36] investigated the stress-strain
property of polymers by X-ray tomography. They used a finite element method to model and in-situ test
to validate elastic and hyper-elastic deformation properties of polymers under compression; Fieres et al.
focused on the new technology of 3D printing and used X-ray to test the failure possibility of printed
parts [37]; Xing et al. inspected jointed rocks with X-ray and scanning electron microscope to find
evidence of cracks under mechanical compression [39]. Although X-ray diffraction provides sensitive,
reliable and quantitative stress-strain measurements and usually acts as a reference, such radiographic
NDT has the obvious disadvantage of complicated instrumentation and harmful ionizing radiation
which requires high level of operation and testing standards. The extremely high sensitivity of X-ray
diffraction to stress makes it more applicable to small- and micro-scale structures in laboratories. Other
types of NDT such as ultrasonic scanning and thermography should be explored to find the possibility
for stress-strain evaluation and treat macro-scale components.
Figure 3. Scheme of residual stress detection by X-ray diffraction.
2.2. Photoacoustic (PA) Techniques

Regarding optical NDT techniques, the term “photo” is used here to refer to the phenomenon of
photonic-to-acoustic (specifically ultrasonic) energy conversion. The photo-thermo-elastic or simply
photoacoustic (PA) effect was discovered by Alexander Graham Bell in 1880 [41], followed by a
number of theoretical and experimental research reports on the physical principle of ultrasound
108
Appl. Sci. 2020, 10, 1588
generation in solids by transient optical power absorption [42–44]. From the acoustic point of view,
the existence of stress in a sample can affect the propagation of elastic waves because of elastic
inhomogeneity or anisotropy. Such a phenomenon was first discussed theoretically by seismologist
Biot [45] and experimentally validated by Hughes and Kelly [46] and Bergman and Shahbender [47].
They concluded that the existence of static pre-stress can change the acoustic wave velocity inside
a medium, a phenomenon described as the acoustoelastic effect [48–50]. In the case of irreversible
residual stress, finite and irrecoverable deformation and strain should be considered with the result
that the stress-strain relation becomes nonlinear [51]. According to Murnaghan’s finite deformation
of elastic materials, the stress-strain relation should be determined with the help of the free-energy
function Ws which is defined as [46,52]:
1 1
Ws = (λ + 2µ)I12 − 2µI2 + (l + 2m)I13 − 2mI1 I2 + nI3 (3)
2 3
Here, λ and µ are the Lamé constants with respect to infinitesimal deformation, l, m and n are
Murnaghan’s constants determined by the type of sample material, I1 , I2 and I3 are the strain invariants
of first-, second- and third-order. Conservation of energy requires Hooke’s Law to be expressed as:
∂δui
ρδWs = σij (4)
∂x j
where δW and δui denote the finite increment in free-energy function and displacement field, respectively.
ρ is the density after deformation. Combining Equations (3) and (4) results in an acoustoelastic equation
which connects static loading with the elastic wave velocity under hydrostatic pressure P:
ρ0 v2c = λ + 2µ − P
3λ+2µ (6l + 4m + 7λ + 10µ)
(5)
ρ0 v2s = µ − P
3λ+2µ (3m − 0.5n + 3λ + 6µ)
The subscripts c and s imply compressive (longitudinal) and shear wave respectively. Subscript
0 denotes the undeformed state. The level of stress can thus be determined by measuring the wave
velocity based on Equation (5) [53].
A typical ultrasonic stress measurement setup is illustrated by Figure 4. The sample is fixed on a
tensile machine and subjected to uniaxial loading. Ultrasonic transducers are used to generate and
receive an acoustic pulse that propagates along the sample and measure the wave velocity from the two
ends. Chaki and Bourse extended this time-of-flight velocity measurement to metallic cable strands and
determined their stress-strain state [54]. Gennisson et al. focused on stressed soft solids and explored
shear modulus measurements [55]. Besides bulk waves, surface acoustic waves were also used to
evaluate the existence and level of surface residual stress. A nonlinear acoustoelastic theory with respect
to Rayleigh wave propagation was introduced by Iwashimizu and Kobori [56]. They concluded that the
propagation of a Rayleigh surface wave in finitely deformed solids satisfies the wave equation similar
to the linear-elastic material, only the elastic modulus tensor loses some of the symmetric conditions as
the regular stress-strain relation. More recently, Duquennoy et al. used an interdigitated transducer
and a laser interferometer to test superficial residual stress experimentally [57,58]. The transducer was
designated for surface wave generation using a contacting approach.
109
Appl. Sci. 2020, 10, 1588
Figure 4. Stress measurement scheme by pulsed ultrasound.
The advantage of acoustic methodologies is their multimode propagation which is applicable

for both internal and superficial stress-strain evaluation. However, the generation and detection of
ultrasonic waves invariably relies on contact transducers. Based on the PA effect investigated in the
1980s and 1990s, it is possible to realize noncontact acoustic stress-strain characterization. Kasai and
Sawada first discussed the possibility of stress distribution measurements using a PA microscope [59].
Because of the existence of photon-phonon energy conversion, the heat equation was modified as:
!
Q ∆T 1 ∂E
+ ρC = − α− 2 σ ∆σ (6)
T0 T0  E ∂T 
 
where, E is Young’s modulus, α is linear expansion  coefficient, σ and ∆σ are static stress and stress
change after deformation, Q is the heat flux induced by photon (laser) irradiation and T is the
α σ Δσ
temperature, the subscript 0 refers to stress-free state. Equation (6) shows the coupling of stress
and temperature change. Using piezoelectric devices, the amplitude of the detected PA signal is
proportional to the ∆σ for small signal approximation. However, Equation (6) does not seem to be
comprehensive enough and thus no experimental procedure was proposed in Ref. [59].
Δσ
Muratikov et al. published a series of theoretical and experimental reports [60–62] on PA residual
stress evaluation. They started from the basic form of the first law of thermodynamics and Murnaghan’s
nonlinear theory and wrote the modified wave equation as:
∂Pij ∂2 ui
= ρ0 (7)
∂x j ∂t2
where, ui is the displacement field and ρ0 is the density before deformation. Equation (7) can be
regarded as an alternative form to Equation (4) by introducing the Piola-Kirchhoff stress tensor Pij [63].
Repeated subscripts stand for summationρ and the equation is in Euler coordinates (x1 , x2 , x3 ). Pij is a
function of strain energy due to pre-stressing (Equation (3)) and photon-phonon conversion, the latter
being expressed as:
!
3 0 0 ∂ui
∂u j ∂ui ∂u j
W0 = λ + µ α (1 + β0 I1 )δij + β1 εij + + ∆T (8)
2 ∂x j ∂xi ∂xk ∂xk
   
 function
where, W 0 is the free-energy  
 of laser radiation, β0 and β1 are thermoelastic
 coupling coefficients
  
of static strain and repeated subscript summation is used. Further derivation shows  that the PA signal
is proportional to the vertical displacement component. In their β experiments,
β Muratikov et al. used
several levels of force to produce indentations on a silicon nitride sample and the PA signal was plotted
110
Appl. Sci. 2020, 10, 1588
as a function of laser scanning distance. Figure 5 shows the result using a certain level of indentation
with an obvious change of PA signal in the region of indentation-induced flaws. In their reports,
Muratikov et al. did not quantify the correlation between PA signal and the actual local residual
stress [60]. Moreover, the system used does not qualify as totally noncontact because of the use of
contacting piezoelectric crystals as detectors.
Figure 5. The scanning photoacoustic signal amplitude change. The visible defect is between 0
and 97 µm (“X” marking); reproduced from [60] with permission from The American Institute of
Physics © 2000.
Huan et al. developed a noncontact PA stress-strain measurement method using a

narrow-bandwidth immersion ultrasonic transducer [64]. The use of a tensile machine and adhesive
strain gauge enabled simultaneous measurements of actual strain and PA signal. As shown in Figure 6,
in the MHz range, the detected PA signal phase and amplitude show linear-to-nonlinear evolution of
local strain from elastic to plastic deformation. The results show that a change of elastic properties by
means of tensile stress exists even for linear deformation, although a corresponding explanation of the
stress effect was not given in the published report. Another limitation of the proposed setup was the
requirement for water coupling.
Figure 6. PA stress-strain test schematic with water coupling (left) and the stress-strain curve on the
basis of PA phase (right); reproduced from [64] with permission from Springer Nature © 2017.
If the ultrasonic wave is detected with an optical method, the PA stress-strain measurement can
be truly noncontact. As proposed by Sun and Zhou [65], a high sensitivity laser interferometric system
was developed to measure stress-induced delamination in carbon-fiber reinforced polymer. Based on
the principle of time-of-flight approach [65], Karabutov et al. measured superficial and subsurface
residual stresses in a metal with laser-induced ultrasound [66]. Not only lasers, but also incoherent
111
Appl. Sci. 2020, 10, 1588
sources such as a helium lamp, can be used in stress-strain analysis. McDonald et al. developed a
PA spectroscopic methodology to detect the stress level in some transparent materials such as a thin
polymer film [67]. A Fourier Transform Infrared (FTIR) spectrometer [68,69] was adopted in the test.
The presence of stress can slightly shift absorption peaks of the film, which lays a basis for quantitative
stress measurement by spectral analysis.
Photoacoustic techniques have inherent advantages for mechanical strength evaluation. Ultrasonic
wave mode propagation velocities exhibit a clear theoretical dependence on the stress-strain level [70],
which makes PA techniques sensitive and fast. Optical generation and detection schemes can produce
multi-physical and totally noncontact NDT methods. The main disadvantage of PA techniques is their
low energy conversion due to thermoelasticity and thus sophisticated apparatuses must be adopted.
In addition, they are not sensitive to stress accumulation and are hard to perform local tests with,
because ultrasonic waves travel fast in elastic media.
2.3. Photothermal (PT) Techniques

The discussion of PA signal dependence on the stress-strain state is based on the coupling
between elastic and thermophysical properties such as density and bulk modulus, the change of
which has perceptible impact on the acoustic signal. However, according to the thermodynamic
theory proposed by Landau and Lifshitz [71] and Love [72], a change of stress-strain energy can affect
another two groups of parameters apart from elasticity, namely, thermophysical properties [69] and
the thermoelastic coupling coefficient [73,74]. Therefore, it is possible to determine the mechanical
performance of a material remotely by examining its thermal property changes by means of, e.g.,
thermal infrared detectors.
The earliest implementation of this relationship is found in a patented instrumentation system
named Stress Pattern Analysis by measurement of Thermal Emission, or SPATE [75]. A sample under
test is fixed on a repetitive loading machine which provides a constant strain rate. The temperature
associated with the strain energy change is measured with an infrared detector. The measurement is
based on the relation between temperature and stress level described in Equation (6) and as a result a
quantitative measurement of stress can materialize [76]. SPATE is a passive infrared thermographic
(IT) system that does not require pumping external energy: the thermal radiation is the result of stress
work in the sample. In 1988, Wong et al. reported that IT method can also be used for characterizing
static loading in addition to cyclic loading state such as in a SPATE system [77]. They exerted a static
bending load on an aluminum plate and captured thermographic images with an infrared camera. By
analyzing the images before and after static deformation, they validated the possibility of detecting
residual stress in metallic materials [77]. Similar tests were carried out for thermographical residual
stress analysis in titanium alloys in NASA by Gyekenyesi and Baaklini [78]. Quinn et al. tested a series
of steel and aluminum alloys with a commercial tensile machine and an IT setup. The thermoelastic
signal even showed a qualitative stress distribution around holes on plates [79]. For hyper-elastic
materials such as polymers, the use of passive IT can also diagnose residual stress and fatigues in
them [80]. More recently, the development of NDT in reinforced composites drew the attention of a
number of researchers. When the laminated structure was subjected to high levels of tensile, bending
and torsional loading, the formation of internal delamination and micro cracks occurred along with
energy release. By observing thermal infrared emission with a camera, IT techniques enables inspection
of a whole piece of material and analysis of the inelastic stress-strain performance [81,82].
Generally, the heat equation in the presence of static stress can be derived from the first law of
thermodynamics, which has the following form [70,83–85]:
2
!
∂T 2∂ T ∂2 T
ρC − kii Jii δij + (1 − δij )(kij + k ji ) Jii J jj =g (9)
∂t ∂xi 2 ∂xi ∂x j
where, Jij is a Jacobian indicating the coordinate transformation due to loading (repeated subscripts
denote summation), g is the source term and kij is the thermal conductivity tensor. It can be shown that
112
Appl. Sci. 2020, 10, 1588
the thermal conductivity tensor has a direct connection with the static stress-strain state [45,60,71,83].
As a result, instead of the SPATE mode which is suitable for evaluating cyclic loading, the static
stress-strain performance can be determined based on thermal anisotropy analysis [85,86] with a
proper heat source, usually a laser, which brings photothermal (PT) techniques into this consideration.
Compared with PA techniques, PT detection decouples the stress-strain dependence of many physical
properties by focusing only on thermal fields. The PT testing modality is local, quantitative, and
absolutely noncontact.
Photothermal radiometry (PTR) is an active infrared emission testing modality. High power lasers
are typically used to generate transient or harmonic local heating in solid and liquid samples, followed
by free diffusion or convection. Monitoring the behavior of the diffusing field, the thermophysical
properties can be evaluated. Since the 1980s, Long et al. [87,88], Milner et al. [89,90], Busse et al. [91]
and Mandelis et al. [92–94] have reported on a series of PT-NDT techniques and applications,
mainly for measuring thermophysical properties. The first report on mechanical performance
monitoring published by Yarai et al. [95] used a pyroelectric device to capture the thermal field and
evaluate the residual stress. The system was simple but could not achieve fully noncontact detection.
Pron et al. [96,97] constructed a totally noncontact system using an Ar-ion laser and an infrared camera
to generate and record the PT field on a stressed sample as shown in Figure 7. The beam splitter
in that figure ensured most of the laser power was directed to the sample while an optical detector
provided a reference signal for amplitude and phase data acquisition. The relative change in thermal
conductivity was calculated based on the diffusion field amplitude and phase images. For a carbon
steel sample, a few percent change in its thermal conductivity was found in the direction paralleled to
the uniaxial loading within the elastic regime [96]. Based on a similar setup, Paoloni et al. examined
the PT signal of a plastically deformed metal sample [98]. The sample was stretched to fracture so that
a high level of residual stress was distributed non-uniformly around the failure surface. The diffusivity
changed as much as 65% between the fracture region and the surrounding intact part. Mzali et al. used
a halogen lamp instead of a laser to test the change in thermal properties of a plastically deformed
metal subjected to tensile loading [99]. Since the test was continuous, a thermocouple was attached to
the sample surface to measure the temperature. The contact measurement makes the system inflexible
and rigid. An elastic-to-plastic full-range tensile photo-thermo-mechanical radiometry (PTMR) test
was carried out by Huan et al. using both a Mercury Cadmium Telluride (MCT) detector (single point
measurement, shown in Figure 8 [100] and an infrared camera for imaging [84]. In those systems, a
fiber-coupled high power diode laser was used to provide more operational flexibility. The pump laser
was configured at oblique incidence and the tested data were processed with a normalization algorithm.
The results are shown in Figure 9. The stress-strain relation of full-range tensile loading was plotted as
a function of the thermal parameter κ which is defined as the sample thickness over square-root of
thermal diffusivity. The curves show a good analogy with the conventional stress-strain relation and
the relative change of thermal conductivity is at the same level as [97]. The aforementioned PTMR
development shows the possibility of using a PTR setup as a non-contact strain gauge.
113
κ
Appl. Sci. 2020, 10, 1588
Figure 7. Schematic of a lock-in photothermal technique for stress-strain evaluation, redrawn based
on [97].
Figure 8. Schematic of lock-in photo-thermo-mechanical radiometry (PTMR) testing with a single-point

detector and a confocal radiation collection system, redrawn based on [100].
Figure 9. Composite thermal parameter κ asκ a function of strain that shows good analogy with a typical
engineering stress-strain curve, reproduced from [100] with permission from Elsevier © 2016.
In addition to regular metallic materials, polymers and composites are also treated with PT
techniques to assess their mechanical performance, especially after plastic deformation occurs. Wang
and Wright studied the change of polymer thermal diffusivity in the plastically deformed range [101].
They derived the principal diffusivity tensor as a function of the Cauchy-Green deformation tensor. A
much more conspicuous change in diffusivity was observed in polymers compared with metals as
expected. However, such change is reversible because of the hyperelasticity of polymers. Huan et al.
further used a single point PTR configuration to test nano-coated aluminum composites [102]. A
114
Appl. Sci. 2020, 10, 1588
linear dependence of thermal diffusivities on the load level was found within the elastic regime and
only a minimal 0.6% diffusivity change occurred the in nano-coated samples. Based on these reports,
the PTMR research was extended to a number of complex materials and structures with different
elasticity. Mechanical properties of intact, defective and multi-structured samples were evaluated by
studying the stability of thermo-mechanical parameters under loading.
Photothermal techniques have advantages over other methods in characterizing stress-strain
relations. Compared with PA, PTMR detection usually evaluates mechanical performance through
diffusivity changes. Coupling between thermoelastic properties can be completely separated and the
mechanical stress-strain relation can be represented by a single parameter. In addition, dynamic PT
techniques generate thermal-wave fields inside a sample where the effective thermal diffusion length
can be controlled by the modulation frequency or the delay time of the photothermal source which can
be made to impinge on specific locations and thus can be more sensitive to regionally distributed stresses.
The use of continuous wave lasers and thermal infrared detectors and cameras makes the PTMR
methodology totally non-contact. Nevertheless, PT techniques analyze the thermophysical properties
by sensing thermal-wave field changes and thus are time consuming and optimally performing when
confocal optical systems are used which can be complicated and expensive.
3. Conclusions
Nondestructive stress-strain characterization is of fundamental importance in modern industry
for the early diagnosis of mechanical failure and prediction of safety issues. Many new nondestructive
testing methodologies for mechanical performance assessment of industrial manufactured parts have
been developed since the first application of X-ray tomography machines for industrial component
residual stress monitoring in the 1940s. This article has reviewed three categories of typical NDT/NDE
approaches which have their own unique features toward characterizing the stress-strain state.
The highlights and drawbacks of each method were discussed by reviewing the development history
and some results from state-of-the-art research. A brief summary is listed in Table 1 for each
testing method.
Table 1. Summary of popular stress-strain NDT/NDE methods.
System Measurement
Category Methodology Physical Field Contact/Non-Contact
Complexity Range
DIC Electromagnetic Low Non-contact Surface only
Optical/radiative
X-ray Electromagnetic High Non-contact Internal
Pure acoustic Elastic Low Contact Surface/internal
Acoustic
PA Thermoelastic Medium Semi-/total-noncontact Surface/internal
SPATE Electromagnetic Medium Non-contact Average
Thermal
PTMR Electromagnetic Low Non-contact Surface/internal
In future, the following points should be considered to further boost related technologies and
in-field applications:
Stress-strain imaging and distribution reconstruction. Until now most reports on mechanical
measurements are based on simple loaded state configurations such as hydrostatic or uniaxial
loading. Unfortunately, they are insufficient for dealing with real-world stress-strain conditions.
New mathematical modeling and numerical approaches should be sought to develop in-field and
quantitative stress distribution analyses.
Involvement of artificial intelligence (AI) and big data (BD) technologies in NDT. Multiphysical interfaces
allow inputs of data from different physical fields. This approach incorporates advantages of various
processes and can bring to bear testing of different properties such as thermal, optical, elastic etc.
More data may be generated from a single group of tests and provide more powerful assistance in
industrial applications. AI and BD technologies such as deep-learning and machine-vision enable
automatic and massive data analysis with reliable and quantitative outputs. They are also useful for
115
Appl. Sci. 2020, 10, 1588
diagnosing components of irregular shape and dimension. Pioneering works that treated cylinders,
spheres, corners and wedges etc. can be found in [103–107].
Author Contributions: Methodology, H.H. and A.M.; resources, L.L., C.P., and X.C.; writing—original draft
preparation, H.H.; writing—review and editing, J.Z.; supervision, A.M.; funding acquisition, L.L. All authors have
read and agreed to the published version of the manuscript.
Funding: This research was funded by National Natural Science Foundation of China (NNSFC), grant number
61801358, 61805187 and 61727804; the NSERC Discovery grants and the Canada Research Chairs Program;
the China Postdoctoral Science Foundation, grant number 2019M65346; the Fundamental Research Funds for the
Central Universities, grant number JB20190412 and XJS190505; the National Key R&D Program of China, grant
number 2017YFF0106705. The APC was funded by 61801358 and 2019M65346.
Acknowledgments: The authors want to acknowledge the School of Optoelectronic Science and Engineering,
University of Electronic Science and Technology of China for a supporting program.
References
1. Sposito, G.; Ward, C.; Cawley, P.; Nagy, P.B.; Scruby, C. A review of non-destructive techniques for the
detection of creep damage in power plant steels. NDT&E Int. 2010, 43, 555–567.
2. Breysse, D. Nondestructive evaluation of concrete strength: An historical review and a new perspective by
combining NDT methods. Constr. Build. Mater. 2012, 33, 139–163. [CrossRef]
3. Kamsu-Foguem, B. Knowledge-based support in Non-Destructive Testing for health monitoring of aircraft
structures. Adv. Eng. Inform. 2012, 26, 859–869. [CrossRef]
4. Hola, J.; Schabowicz, K. State-of-the-art non-destructive methods for diagnostic testing of building
structures—Anticipated development trends. Arch. Civ. Mech. Eng. 2010, 10, 5–18. [CrossRef]
5. Hola, J.; Beiń, J.; Sadowski, L.; Schabowicz, K. Non-destructive and semi-destructive diagnostics of concrete
structures in assessment of their durability. Civ. Eng. 2015, 63, 87–96.
6. Ibrahim, M.E. Nondestructive evaluation of thick-section composites and sandwich structures: A review.
Compos. Part A-Appl. Sci. Manuf. 2014, 64, 36–48. [CrossRef]
7. Duchene, P.; Chaki, S.; Ayadi, A.; Krawczak, P. A review of non-destructive techniques used for mechanical
damage assessment in polymer composites. J. Mater. Sci. 2018, 53, 7915–7938. [CrossRef]
8. Burte, H.M. A science base for NDE and its coupling to technology. In International Advances in Nondestructive
Testing; McGonnagle, W.J., Ed.; CRC Press: Boca Raton, FL, USA, 1979; Volume 6, pp. 19–38.
9. Adams, R.D.; Cawley, P. A review of defect types and nondestructive testing techniques for composites and
bonded joints. NDT&E Int. 1988, 21, 208–222.
10. Malhotra, V.M.; Carino, N.J. Handbook on Nondestructive Testing of Concrete; CRC Press: Boca Raton, FL, USA,
2003; pp. 1–13.
11. Harding, J.; Wood, E.O.; Campbell, J.D. Tensile testing of materials at impact rates of strain. J. Mech. Eng. Sci.
1960, 2, 88–96. [CrossRef]
12. Ramberg, W.; Osgood, W.R. Description of Stress-Strain Curves by Three Parameters; National Advisory
Committee for Aeronautics Technical Note; NACA-TN-902; National Advisory Committee for Aeronautics:
13. Hill, H.N. Determination of Stress-Strain Relations from “Offset” Yield Strength Values; National Advisory
Committee for Aeronautics Technical Note; NACA-TN-927; National Advisory Committee for Aeronautics:
14. McCullough, K.Y.G.; Fleck, N.A.; Ashby, M.F. Uniaxial stress-strain behaviour of aluminium alloy foams.
Acta Mater. 1999, 47, 2323–2330. [CrossRef]
15. Ling, Y. Uniaxial true stress-strain after necking. AMP J. Technol. 1996, 5, 37–48.
16. Turner, R.C.; Fuierer, P.A.; Newnham, R.E.; Shrout, T.R. Materials for high temperature acoustic and vibration
sensors: A review. Appl. Acoust. 1994, 41, 299–324. [CrossRef]
17. Cegla, F.B.; Cawley, P.; Allin, J.; Davies, J. High-temperature (>500 ◦ C) wall thickness monitoring using
dry-coupled ultrasonic waveguide transducers. IEEE Trans. Ultrason. Ferroelectr. Freq. Control 2011,
58, 156–167. [CrossRef] [PubMed]
116
Appl. Sci. 2020, 10, 1588
18. Kobayashi, M.; Jen, C.K.; Bussiere, J.F.; Wu, K.T. High-temperature integrated and flexible ultrasonic
transducers for nondestructive testing. NDT&E Int. 2009, 42, 157–161.
19. Liu, Z.; Wu, H.; Paterson, A.; Luo, Z.; Ren, W.; Ye, Z.G. High Curie-temperature (TC) piezo-/ferroelectric
single crystals with bismuth-based complex perovskites: Growth, structures and properties. Acta Mater.
2017, 136, 32–38. [CrossRef]
20. Lyons, J.S.; Liu, J.; Sutton, M.A. High-temperature deformation measurements using digital-image correlation.
Exp. Mech. 1996, 36, 64–70. [CrossRef]
21. Anwander, M.; Zagar, B.G.; Weiss, B.; Weiss, H. Noncontacting strain measurements at high temperature by
the digital laser speckle technique. Exp. Mech. 2000, 40, 98–105. [CrossRef]
22. Vǒlkl, R.; Fischer, B. Mechanical testing of ultra-high temperature alloys. Exp. Mech. 2004, 44, 121–127.
[CrossRef]
23. Zhu, D.; Mobasher, B.; Rajan, S.D. Non-contacting strain measurement for cement-based composites in
dynamic tensile testing. Cem. Concr. Compos. 2012, 34, 147–155. [CrossRef]
24. Wolverton, M.; Bhattacharyya, A.; Kannarpady, G.K. Efficient, flexible, noncontact deformation measurements
using video multi-extensometry. Exp. Tech. 2009, 33, 24–33. [CrossRef]
25. Pan, B.; Wu, D.; Wang, Z.; Xia, Y. High-temperature digital image correlation method for full-field deformation
measurement at 1200 ◦ C. Meas. Sci. Technol. 2011, 22, 015701. [CrossRef]
26. Pan, B.; Qian, K.; Xie, H.; Asundi, A. Two-dimensional digital image correlation for in-plane displacement
and strain measurement: A review. Meas. Sci. Technol. 2009, 20, 062001. [CrossRef]
27. Yang, X.; Liu, Z.; Xie, H. A real time deformation evaluation method for surface and interface of thermal
barrier coatings during 1100 ◦ C thermal shock. Meas. Sci. Technol. 2012, 23, 105604. [CrossRef]
28. Hoła, J.; Sadowski, Ł.; Reiner, J.; Stach, S. Usefulness of 3D surface roughness parameters for nondestructive
evaluation of pull-off adhesion of concrete layers. Constr. Build. Mater. 2015, 84, 111–120. [CrossRef]
29. Czarnecki, S.; Hoła, J. Evaluation of the height 3D roughness parameters of concrete substrate and the
adhesion to epoxy resin. Int. J. Adhes. Adhes. 2016, 67, 3–13.
30. Sadowski, Ł. Non-destructive identification of pull-off adhesion between concrete layers. Automat. Constr.
2015, 57, 146–155. [CrossRef]
31. Barrett, C.S.; Predecki, P. Stress measurement in graphite/epoxy uniaxial composites by X-ray. Polym. Compos.
1980, 1, 2–6. [CrossRef]
32. Ledbetter, H.M.; Austin, M.W. Internal strain (stress) in an SiC-Al particle-reinforced composite: An X-ray
diffraction study. Mater. Sci. Eng. 1987, 89, 53–61. [CrossRef]
33. Hemley, R.J.; Mao, H.; Shen, G.; Badro, J.; Gillet, P.; Hanfland, M.; Häusermann, D. X-ray imaging of stress
and strain of diamond, iron, and tungsten at megabar pressures. Science 1997, 276, 1242–1245. [CrossRef]
34. Prevey, P.S. X-ray diffraction residual stress techniques. ASM Handb. 1986, 10, 380–392.
35. Hughes, D.J.; Mahendrasingam, A.; Martin, C.; Oatway, M.B.; Heeley, E.L.; Bingham, S.J.; Fuller, W. An
instrument for the collection of simultaneous small and wide angle X-ray scattering and stress–strain data
during deformation of polymers at high strain rates using synchrotron radiation sources. Rev. Sci. Instrum.
1999, 70, 4051–4054. [CrossRef]
36. Patterson, B.M.; Cordes, N.L.; Henderson, K.; Williams, J.J.; Stannard, T.; Singh, S.S.; Ovejero, A.R.; Xiao, X.;
Robinson, M.; Chawla, N. In situ X-ray synchrotron tomographic imaging during the compression of
hyper-elastic polymeric materials. J. Mater. Sci. 2016, 51, 171–187. [CrossRef]
37. Fieres, J.; Schumann, P.; Reinhart, C. Predicting failure in additively manufactured parts using X-ray
computed tomography and simulation. Procedia Eng. 2018, 213, 69–78. [CrossRef]
38. Subramanian, J.; Seetharaman, S.; Gupta, M. Processing and properties of aluminum and magnesium based
composites containing amorphous reinforcement: A review. Metals 2015, 5, 743–762. [CrossRef]
39. Xing, J.; Zhao, C.; Yu, S.; Matsuda, H.; Ma, C. Experimental study on rock-like specimens with single flaw
under hydro-mechanical coupling. Appl. Sci. 2019, 9, 3234. [CrossRef]
40. Youssef, S.; Marie, E.; Gaertner, R. Finite element modelling of the actual structure of cellular materials
determined by X-ray tomography. Acta Mater. 2005, 53, 710–730. [CrossRef]
41. Bell, A.G. On the production and reproduction of sound by light. Am. J. Sci. 1880, 20, 305–324. [CrossRef]
42. Scruby, C.B.; Dewhurst, R.J.; Hutchins, D.A.; Palmer, B. Quantitative studies of thermally generated elastic
waves in laser-irradiated metals. J. Appl. Phys. 1981, 51, 6210–6216. [CrossRef]
117
Appl. Sci. 2020, 10, 1588
43. Rose, L.R.F. Point-source representation for laser-generated ultrasound. J. Acoust. Soc. Am. 1984, 75, 723–732.
[CrossRef]
44. Hutchins, D.A. Mechanisms of pulsed photoacoustic generation. Can. J. Phys. 1986, 64, 1247–1264. [CrossRef]
45. Biot, M.A. The influence of initial stress on elastic waves. J. Appl. Phys. 1940, 11, 522–530. [CrossRef]
46. Hughes, D.S.; Kelly, J.L. Second-order elastic deformation of solids. Phys. Rev. 1953, 92, 1145–1149. [CrossRef]
47. Bergman, R.H.; Shahbender, R.A. Effect of statically applied stresses on the velocity of propagation of
ultrasonic waves. J. Appl. Phys. 1958, 29, 1736–1738. [CrossRef]
48. Tokuoka, T.; Saito, M. Elastic wave propagations and acoustical birefringence in stressed crystals. J. Acoust.
Soc. Am. 1968, 45, 1241–1246. [CrossRef]
49. Husson, D.; Kino, G.S. A perturbation theory for acoustoelastic effects. J. Appl. Phys. 1982, 53, 7250–7258.
[CrossRef]
50. Shams, M.; Destrade, M.; Ogden, R.W. Initial stresses in elastic solids: Constitutive laws and acoustoelasticity.
Wave Motion 2011, 48, 552–567. [CrossRef]
51. Davies, G.F. Quansi-harmonic finite strain equations of state of solids. J. Phys. Chem. Solids 1973, 34, 1417–1429.
[CrossRef]
52. Murnaghan, F.D. Finite deformations of an elastic solid. Am. J. Math. 1937, 59, 235–260. [CrossRef]
53. Tylczyński, Z.; Mróz, B. The influence of uniaxial stress on ultrasonic wave propagation in ferroelastic
(NH4 )4 LiH3 (SO4 )4 . Solid State Commun. 1997, 101, 653–656. [CrossRef]
54. Chaki, S.; Bourse, G. Guided ultrasonic waves for non-destructive monitoring of the stress levels in prestressed
steel strands. Ultrasonics 2009, 49, 162–171. [CrossRef]
55. Gennisson, J.L.; Renier, M.; Catheline, S.; Barriere, C.; Bercoff, J.; Tanter, M.; Fink, M. Acoustoelasticity in soft
solids: Assessment of the nonlinear shear modulus with the acoustic radiation force. J. Acoust. Soc. Am.
2007, 122, 3211–3219. [CrossRef] [PubMed]
56. Iwashimizu, Y.; Kobori, O. The Rayleigh wave in a finitely deformed isotropic elastic material. J. Acoust.
Soc. Am. 1978, 64, 910–916. [CrossRef]
57. Duquennoy, M.; Ouaftouh, M.; Deboucq, J.; Lefebvre, J.E.; Jenot, F.; Ourak, M. Influence of a superficial
field of residual stress on the propagation of surface waves—Applied to the estimation of the depth of the
superficial stressed zone. Appl. Phys. Lett. 2012, 101, 234104. [CrossRef]
58. Duquennoy, M.; Ouaftouh, M.; Ourak, M.; Jenot, F. Theoretical determination of Rayleigh wave acoustoelastic
coefficients: Comparison with experimental values. Ultrasonics 2002, 39, 575–583. [CrossRef]
59. Kasai, M.; Sawada, T. Non-destructive evaluation of the distribution of stress by means of the photoacoustic
microscope. In Photoacoustic and Photothermal Phenomena II; Murphy, J.C., Maclachlan Spicer, J.W.,
Aamodt, L.C., Royce, B.S.H., Eds.; Springer: Berlin, Germany, 1990; pp. 33–36.
60. Muratikov, K.L.; Glazov, A.L.; Rose, D.N.; Dumar, J.E. Photoacoustic effect in stressed elastic solids.
J. Appl. Phys. 2000, 88, 1948–2955. [CrossRef]
61. Muratikov, K.L. Theory of stress influence on the photoacoustic thermoelastic signal near the vertical crack
tips. Rev. Sci. Instrum. 2003, 74, 722–724. [CrossRef]
62. Muratikov, K.L. Theory of the generation of mechanical vibrations by laser radiation in solids containing
internal stresses on the basis of the thermoelastic effect. Tech. Phys. 1999, 44, 792–796. [CrossRef]
63. Keller, J.B. Finite elastic deformation governed by linear equations. J. Appl. Mech. 1986, 53, 819–820.
[CrossRef]
64. Huan, H.; Mandelis, A.; Lashkari, B.; Liu, L. Frequency-domain laser ultrasound (FDLU) non-destructive
evaluation of stress-strain behavior in an aluminum alloy. Int. J. Thermophys. 2017, 38, 62. [CrossRef]
65. Sun, G.; Zhou, Z. Non-contact detection of delamination in layered anisotropic composite materials with
ultrasonic waves generated and detected by lasers. Optik 2016, 127, 6424–6433. [CrossRef]
66. Karabutov, A.; Devichensky, A.; Ivochkin, A.; Lyamshev, M.; Pelivanov, I.; Rohdgi, U.; Solomatin, V.;
Subudhi, M. Laser ultrasonic diagnostics of residual stress. Ultrasonics 2008, 48, 631–635. [CrossRef]
[PubMed]
67. McDonald, W.F.; Goettler, H.; Urban, M.W. A novel approach to photoacoustic FT-IR spectroscopy:
Rheo-photoacoustic measurements. Appl. Spectrosc. 1989, 43, 1387–1393. [CrossRef]
68. Liu, L.; Mandelis, A.; Huan, H.; Melnikov, A. Step-scan T cell-based differential Fourier transform infrared
photoacoustic spectroscopy (DFTIR-PAS) for detection of ambient air contaminants. Appl. Phys. B 2016,
122, 268. [CrossRef]
118
Appl. Sci. 2020, 10, 1588
69. Liu, L.; Mandelis, A.; Huan, H.; Michaelian, K.H. Step-scan differential Fourier transform infrared
photoacoustic spectroscopy (DFTIR-PAS): A spectral deconvolution method for weak absorber detection in
the presence of strongly overlapping background absorptions. Opt. Lett. 2017, 42, 1424–1427. [CrossRef]
[PubMed]
70. Qian, M. New thermoelastic technique for detection of residual stress distribution in solids. Acta Acust. 1995,
14, 97–106, (Chinese Version).
71. Landau, L.D.; Lifshitz, E.M. Theory of Elasticity; Pergamon: Oxford, UK, 1959; pp. 119–121.
72. Love, A.E.H. A Treatise on the Mathematical Theory of Elasticity, 2nd ed.; Cambridge University Press:
Cambridge, UK, 1906; pp. 90–107.
73. Wong, A.K.; Jones, R.; Sparrow, J.G. Thermoelastic constant or thermoelastic parameter? J. Phys. Chem. Solids
1987, 48, 749–753. [CrossRef]
74. Wong, A.K.; Sparrow, J.G.; Dunn, S.A. On the revised theory of the thermoelastic effect. J. Phys. Chem. Solids
1988, 49, 395–400. [CrossRef]
75. Belgen, M.H. Structural stress measurements with an infrared radiometer. ISA Trans. 1967, 6, 49–53.
76. Stanley, P.; Chan, W.K. Quantitative stress analysis by means of the thermoelastic effect. J. Strain Anal. Eng.
1985, 20, 129–137. [CrossRef]
77. Wong, A.K.; Dunn, S.A.; Sparrow, J.G. Residual stress measurement by means of the thermoelastic effect.
Nature 1988, 332, 613–615. [CrossRef]
78. Gyekenyesi, A.L.; Baaklini, G.Y. Thermoelastic stress analysis: The mean stress effect in metallic alloys.
In Proceedings of the Nondestructive Evaluation Techniques for Aging Infrastructures and Manufacturing,
Newport Beach, CA, USA, 3–5 March 1999. NASA/TM-1999-209376.
79. Quinn, S.; Dulieu-Barton, J.M.; Langlands, J.M. Progress in thermoelastic residual stress measurement. Strain
2004, 40, 127–133. [CrossRef]
80. Menczel, J.D.; Prime, R.B. Thermal Analysis of Polymers Fundamentals and Applications; John Wiley & Sons:
Hoboken, NJ, USA, 2009; pp. 384–495.
81. Zhang, H.; Sfarra, S.; Sarasini, F.; Santulli, C.; Fernandes, H.; Avdelidis, N.P.; Ibarra-Castanedo, C.;
Maldague, X.P.V. Thermographic non-destructive evaluation for natural fiber-reinforced composite laminates.
Appl. Sci. 2018, 8, 240. [CrossRef]
82. Liu, J.; Gong, J.; Liu, L.; Qin, L.; Wang, Y. Investigation on stress distribution of multilayered composite
structure (MCS) using infrared thermographic technique. Infrared Phys. Technol. 2013, 61, 134–143. [CrossRef]
83. Huan, H.; Mandelis, A.; Liu, L. Characterization of the mechanical stress–strain performance of aerospace
alloy materials using frequency-domain photoacoustic ultrasound and photothermal methods: An FEM
approach. Int. J. Thermophys. 2018, 39, 55. [CrossRef]
84. Huan, H.; Mandelis, A.; Liu, L.; Melnikov, A. Local-stress-induced thermal conductivity anisotropy analysis
using non-destructive photo-thermo-mechanical lock-in thermography (PTM-LIT) imaging. NDT&E Int.
2017, 91, 79–87.
85. Salazar, A.; Sanchez-Lavega, A.; Ocariz, A.; Guitonny, J.; Pandey, G.C.; Fournier, D.; Boccara, A.C. Thermal
diffusivity of anisotropic materials by photothermal methods. J. Appl. Phys. 1996, 79, 3984–3993. [CrossRef]
86. Iravani, M.V.; Nikoonahad, M. Photothermal waves in anisotropic media. J. Appl. Phys. 1987, 62, 4065–4071.
[CrossRef]
87. Long, F.H.; Anderson, R.R.; Deutsch, T.F. Pulsed photothermal radiometry for depth profiling of layered
media. Appl. Phys. Lett. 1987, 51, 2076–2078. [CrossRef]
88. Prahl, S.A.; Vitkin, I.A.; Bruggemann, U.; Wilson, B.C.; Aderson, R.R. Determination of optical properties of
turbid media using pulsed photothermal radiometry. Phys. Med. Biol. 1992, 37, 1203–1217. [CrossRef]
89. Milner, T.E.; Smithies, D.J.; Goodman, D.M.; Lau, A.; Nelson, J.S. Depth determination of chromophores in
human skin by pulsed photothermal radiometry. Appl. Opt. 1996, 35, 3379–3385. [CrossRef]
90. Milner, T.E.; Katzir, A.; Jacques, S.L. Pulsed photothermal radiometry of port-wine stains. Proc. SPIE 1993,
1882, 34–42.
91. Busse, G.; Wu, D.; Karpen, W. Thermal wave imaging with phase sensitive modulated thermography.
J. Appl. Phys. 1992, 71, 3962–3965. [CrossRef]
92. Mandelis, A.; Abrams, S.H.; Nicolaides, L.; Garcia, J.A. Method and Apparatus for Detection of Defects in
Teeth. U.S. Patent US6584341, 24 June 2003.
119
Appl. Sci. 2020, 10, 1588
93. Wang, C.; Mandelis, A.; Liu, Y. Thermal-wave nondestructive evaluation of cylindrical composite structures
using frequency-domain photothermal radiometry. J. Appl. Phys. 2005, 97, 014911. [CrossRef]
94. Balderas-López, J.A.; Mandelis, A.; Garcia, J.A. Thermal-wave resonator cavity design and measurements of
the thermal diffusivity of liquids. Rev. Sci. Instrum. 2000, 71, 2933–2937. [CrossRef]
95. Yarai, A.; Yokoyama, Y.; Nakanishi, T. New non-destructive photothermal measurement of anisotropically
distributed residual stress inside samples. In Proceedings of the IEEE Ultrasonics Symposium, Cannes,
France, 31 October–3 November 1994; pp. 683–686.
96. Pron, H.; Henry, J.F.; Offermann, S.; Bissieux, C.; Beaudoin, J.L. Analysis of stress influence on thermal
diffusivity by photothermal infrared thermography, Estimation of local thermophysical properties by
means of front-face photothermal infrared thermography: Application to mechanical stress analysis.
High Temp.-High Press. 2000, 32, 473–477. [CrossRef]
97. Pron, H.; Bissieux, C. 3-D thermal modelling applied to stress-induced anisotropy of thermal conductivity.
Int. J. Therm. Sci. 2004, 43, 1161–1169. [CrossRef]
98. Paoloni, S.; Tata, M.E.; Scudieri, F.; Mercuri, F.; Marinelli, M.; Zammit, U. IR thermography characterization
of residual stress in plastically deformed metallic components. Appl. Phys. A 2010, 98, 461–465. [CrossRef]
99. Mzali, F.; Albouchi, F.; Nasrallah, S.B.; Petit, D. Optimal experiment design and thermo-physical
characterization of a plastically deformed solid. Inverse Probl. Sci. Eng. 2009, 17, 335–345. [CrossRef]
100. Huan, H.; Mandelis, A.; Liu, L.; Melnikov, A. Non-destructive and non-contacting stress-strain
characterization of aerospace metallic alloys using photo-thermo-mechanical radiometry. NDT&E Int.
2016, 84, 47–53.
101. Wang, Y.; Wright, N.T. A relationship between thermal diffusivity and finite deformation in polymers.
Int. J. Thermophys. 2005, 26, 1849–1859. [CrossRef]
102. Huan, H.; Mandelis, A.; Liu, L.; Melnikov, A. Evaluation of mechanical performance of NiCo nanocoated
aerospace aluminum alloy using quantitative photo-thermo-mechanical radiometry as a non-contact strain
gauge. NDT&E Int. 2017, 87, 44–49.
103. Tai, R.; Zhang, J.; Wang, C.; Mandelis, A. Thermal-Wave Fields in Solid Wedges Using the Green Function
Method: Theory and Experiment. J. Appl. Phys. 2013, 113, 133501. [CrossRef]
104. Liu, L.; Wang, C.; Yuan, X.; Mandelis, A. Curvature-insensitive methodology for thermal-wave
depth-profilometry in curvilinear solids. J. Phys. D-Appl. Phys. 2010, 43, 285403. [CrossRef]
105. Celorrio, R.; Mendioroz, A.; Apiñaniz, E.; Salazar, A.; Wang, C.; Mandelis, A. Reconstruction of radial thermal
conductivity depth profile in case hardened steel rods. J. Appl. Phys. 2009, 105, 083517. [CrossRef]
106. Zhang, J.; Xie, G.; Wang, C.; Mandelis, A. Laser induced thermal-wave fields in multi-layered spherical
solids based on Green function method. J. Appl. Phys. 2012, 112, 033521. [CrossRef]
107. Wang, M.; Mandelis, A.; Melnikov, A.; Wang, C. Quantitative lock-in thermography imaging of thermal-wave
spatial profiles and thermophysical property measurements in solids with inner corner geometries using
thermal-wave field theory. J. Appl. Phys. 2018, 124, 205106. [CrossRef]
120
applied
sciences
Article
Inducing Damage Diagnosis Capabilities in Carbon
Fiber Reinforced Polymer Composites by
Magnetoelastic Sensor Integration via 3D Printing
Dimitrios G. Dimogianopoulos 1 , Panagiotis J. Charitidis 2 and Dionysios E. Mouzakis 3, *
1 Department of Industrial Design and Production Engineering, University of West Attica, 12241 Athens,
Greece; [email protected]
2 Environmental Engineering School, Democritus University of Thrace, 67100 Xanthi, Greece;
[email protected]
3 Sector of Mathematics and Engineering Applications, Mechanics Laboratory, Hellenic Army Academy,
PO Vari P.O. 16673 Attica, Greece

Received: 15 January 2020; Accepted: 31 January 2020; Published: 4 February 2020
Abstract: This study investigates the possibility of inducing damage diagnosis capabilities in carbon
fiber reinforced polymer composite slabs using custom-built integrated sensors and conventional,
affordable equipment. The concept utilizes magnetoelastic strips integrated via 3D printing procedures
in composite slabs. Under external mechanical loading, the strip magnetization changes due to the
magnetoelastic phenomenon. Accordingly, electrical signals may be passively induced in conventional
reception coil circuits placed at a distance from the slab. Since these signals quantify the vibrating
slab’s response, which is affected by the slab’s structural integrity, damage may be detected when
specific signal characteristics change. Two main issues are examined, namely the ability of receiving
meaningful (with respect to noise) electrical signals from the built-in strips despite their contact-less
passive reception, and the potential of diagnosing damage using such signals. Hence, slabs of
various sizes and levels of structural damage (notches) have been vibrated at different frequencies
and amplitudes. Treating the experimental data from integrated strips by applying the proposed
processing framework allows for calculating eigenfrequencies sensitive to occurring damage (and its
severity), as verified by finite element models of the vibrating slabs. Accordingly, damage may be
detected and evaluated via the currently proposed experimental testing and analysis framework.
Keywords: damage detection; damage assessment; smart sensor; magnetoelastic strip; 3D printing;
additive manufacturing
1. Introduction
In the past decade considerable advances have been made in the technology of rapid manufacturing.
Not only have the techniques per se evolved, but the range of available materials for the so-called
3D printing technologies (or additive manufacturing) has also widened [1]. Numerous works in the
literature already refer to the next level of meta-structures which can be designed and manufactured
under the provocative term “4D printing”, referring to materials and structures with time-dependent
response or geometry [2]. Another interesting feature of the wide range of additive manufacturing
technologies is that sensors may be easily incorporated within the layered structure during its building
phase, resulting in smart systems, i.e., objects with built-in sensing capabilities [3]. Several review
articles (see [4,5] for typical examples) classify such smart systems according to their manufacturing
process, operating principle, and/or functionality. Different types of sensors, e.g., optical for interactive
devices [6], or embedded wires for capacitive sensing applications [7,8] have been suggested as
121
Appl. Sci. 2020, 10, 1029
potential smart systems. Mechanical strain sensing via various incorporated sensors in 3D printed
systems has also been proposed [9–11]. Magnetite/thermoplastic composite in 3D printing as direct
replacements for commercially available flow sensors [12], or custom-built devices (whiskers) with
built-in magnetostrictive materials for specific flow measuring applications [13] have been successfully
manufactured. Magnetic force sensors [14], or devices for sensing human tactile force and imperceptible
skin deformation due to heart pulses [15] for biomechanical applications are reported as feasible
options for 3D printed smart objects. Specifically, some applications related to stretchable sensors for
biomedical purposes are also reported in [16].
On the other hand, several custom-built sensors designed to monitor structures for damage
diagnosis purposes, have been proposed over the recent years, for instance [17–22]. Such sensors are
mostly designed to be patched onto the structure, so that the sensing element such as, fiber bragg
gratings [18,19], magnetoelastic strips [21], and PZT impedance [22] or MFC (macro-fiber composite)
transducers [23] to name but a few, could measure specific dynamic properties of the structure
under vibration.
Magnetoelastic sensors have been under intense investigation for many applications in the past
decade. Since they react to changes in imposed mechanical strain, by changing their magnetic constant,
they can be used as wireless and passive sensors at the same time. Magnetoelastic sensors are usually
made by rapid melt quenching, in the form of amorphous ferromagnetic ribbons. Most members of this
family are iron-rich alloys, such as the well-known Fe40 Ni38 Mo4 B18 marketed under the commercial
name Metglas® 2826 MB. The authors have successfully used this material as a damage sensor for
polymers and composite systems [21] and have gathered considerable experience in sensor application,
data gathering, and processing over the last years. A main advantage of such sensors against all other
types of relevant devices is that they require no wiring, electrical nor optical, and can be interrogated
wirelessly and in a passive state.
In general, the sensor is carefully configured in order to interfere as little as possible with the
monitored structure’s dynamics. Therefore, the sensor weight and volume are kept as low as possible,
meaning that in some cases even thermal-spraying techniques to achieve deposition of sensing element
(copper electrodes) onto the structure [24] may be used. A notable exception to this rule is found in [25],
where the sensor placed on the structure is deliberately chosen of considerable weight so as to change
at will the underlying dynamics.
In general, 3D printing procedures have been used for the sensor’s construction [17] or the
integration of the necessary circuitry/devices in order to facilitate operation of laser displacement
sensors [20] or MFC sensors/transducers [23]. In other words, the 3D printing process has very often
served as a means of facilitating the sensor’s construction process, or it’s mounting onto a given
structure and its effective operation, but it has rarely been used for manufacturing a structure with a
sensing element integrated within.
The aim of the current study is to use the additive manufacturing principle for integrating sensing
elements in a structure during its 3D printing process, and to diagnose structural damage based on
data from these built-in sensing elements and conventional, affordable equipment. The considered
structures are carbon fiber reinforced polymer (CFRP) composite slabs of different dimensions, which
incorporate magnetoelastic strips acting as sensing elements under external slab loading. Then, due to
the magnetoelastic phenomenon, such loading lead to changes in the strip’s magnetization, which
causes the induction of electrical signals in conventional reception coil circuits placed at a distance
from the slab. Since these signals quantify the vibrating slab’s response (itself depending on the slab’s
structural integrity) damage is detectable when changes of specific signal characteristics (selected
eigenfrequencies) occur. Even if, at times, this detection principle may prove not very practical to use,
it is a traditional approach for structural health monitoring purposes, as it can be sensitive to even
small damages [25].
This detection principle has also been used in a recent study [26], which features a similar
operational and testing concept to that proposed in the current work, but for the important detail of
122
Appl. Sci. 2020, 10, 1029
the magnetoelastic sensing material being attached onto (rather than integrated within) the structure.
Due to this critical difference, two main issues must be dealt with, namely the possibility of recording
meaningful (with respect to noise) electrical signals from the built-in strips despite their contact-less
reception, and the potential of diagnosing damage using such signals. Therefore, slabs of various
dimensions and damage levels have been vibrated at different frequencies and amplitudes. Via a
dedicated test data processing framework along with finite element method (FEM) analysis, it is
shown that magnetoelastic strips integrated within the material structure may successfully act as
passive embedded sensors, transmitting meaningful signals during mechanical loading and providing
qualitative and quantitative damage diagnosis results.
The rest of the paper is organized as follows: Section 2 presents the 3D printing process of CFRP
slabs with the integration of sensing elements, and the testing protocol used for the experiments.
In Section 3, the experimental data along with the relevant framework for data analysis and validation
via FEM are shown. Based on this framework, damage detection and assessment results are obtained
and commented upon. Finally, Section 4 presents some concluding remarks.
2.1. Slab Preparation by 3D Printing

A reprap project series clone model of a 3D printer (CTC i3, Zhuhai Electronic Ltd, Zhuhai City,
Guangdong, China) operated by an Arduino® controller, was used to prepare the slabs. The printer
operated in Fused Deposition Modeling (FDM) mode, had a single nozzle (diameter of 0.4 mm)
and used a 1.75 mm polymer filament. The filament was a short carbon fiber reinforced PET-G
polymer composite by 20 wt% (NEEMA3D™ Carbon: plus) of NEEMA3D™, Petroupolis, Greece.
The filament had a Young’s modulus of E = 3800 MPa and a yield strength of 52.5 MPa according to
the manufacturer’s specifications.
Rectangular slabs of 60 × 12 × 3 mm3 (hereafter referred to as type-A slabs) were prepared by
means of the FDM printer, which was paused once the desired layer height was reached, so as to allow
for attaching the 2826 MB Metglas® magnetoelastic (20 µm) μ strip of 25 × 5 mm2 (kindly donated by
Prof. D. Kouzoudis, University of Patras). The magnetoelastic strips were attached to the desired depth
by means of an epoxy adhesive. Subsequently, the magnetoelastic strip was covered by successive
material layers until obtaining the final slab thickness (Figure 1). In this way, one set of type-A slabs
with “hidden” built-in magnetoelastic strip sensors, hereafter referred to as M-sensors, was prepared.
Again, another series of type-A slabs without magnetoelastic strips (type-A blanc) was also prepared
for testing purposes.
Figure 1. Smart slab preparation showing the strip of magnetoelastic 2826 Metglas® ribbon material
and two views of the carbon fiber PET-G layers (a) during the process of ribbon (b) integration.
123
Appl. Sci. 2020, 10, 1029
Two sets of larger slabs of 150 × 25 × 3 mm3 were also prepared for the testing and may be seen in
Figure 2. As seen in Figure 2, the magnetoelastic strip was placed inside the specimens during the 3D
printing procedure, at heights of h = 1.15 mm for specimen’s type A and h = 1.75 mm for specimen
Types B and C, with respect to the specimen’s total thickness, measuring from the bottom of the slabs.
Figure 2. The three main slab types tested with built-in M-sensor. Type-A, -A blanc: slabs for DMA
testing, Type-B: slabs for high-amplitude vibration, Type-C1, -C2: notched slabs for damage diagnosis
with one (-C1), or two (-C2) notches at points a, b. Arrows on the left sides of slabs denote neutral axis
in bending levels.
The first set involved M-sensor slabs (hereafter referred to as type-B slabs) and the series contained
artificially damaged M-sensor slabs with one or two transverse notches (hereafter referred to as type-C1
or -C2 slabs, respectively), made with a diamond saw and featuring a depth of 1.5 mm.
Not that preparing two different slab sizes (type-A and type-B/-C1/-C2) was necessary because of
technical limitations imposed by the testing procedures, as explained in the next subsection. Details on
the objectives to be achieved and the testing protocol are given in the following subsection.
2.2. Experimental Procedures

Successful M-sensor operation depends on achieving two objectives:
• O1: That meaningful (with respect to noise) electrical signals passively emitted from M-sensors
could be generated/recorded, and
• O2: That the analysis of recorded signals could provide damage detection and assessment
(diagnosis) results, especially at high amplitude/low frequency vibrating operation.
The reason for focusing on high amplitude/low frequency operation in O2 is because both
objectives were achieved in a high frequency/low amplitude vibrating context [21], even though the
magnetoelastic strip was attached on the surface of the vibrating slab. Nonetheless, objective O1
was checked under all frequency—amplitude combinations, simply to ensure that the slabs with
built-in strips do operate as sensors under such conditions. Two testing procedures have been used for
checking whether O1 and O2 were achievable:
• P1: The procedure simulated high frequency and low amplitude operational conditions, typical
of systems with fast dynamics and involved type-A slabs both in M-sensor and blanc (type-A
μ Testing was performed on a Dynamic Mechanical Analyzer (DMA Q800) of
blanc) forms.
TA—Instruments (Figure 3), which led to choosing the specific dimensions of type-A slabs. Slabs
were subjected to three-point dynamic bending under a linear frequency scan of 1–100 Hz @ RT
and at v = 25 µm constant static deflection. Electrical signals created by induction to the remote
124
Appl. Sci. 2020, 10, 1029
interrogation coil (placed at 20 mm from the slab) were recorded by PioneerHill-Spectraplus©

Software for Data Acquisition at 4096 Hz via a PC-Soundcard 3.5 mm jack. Comparing signals
from type-A and -A blanc slabs allows for validating whether the hidden magnetoelastic material
could provide sensing capabilities to the (otherwise passive) slab.
• P2: The procedure involved series of type-B and -C1, -C2 slabs, fixed as cantilevers (Figure 4).
The free end suffered a load of quasi-sinusoidal form with frequency f = 4 Hz, mean value equal
to 12.5 mm and amplitude d = 12.5 mm. In other words, the free end featured deflections between
0 and 25 mm, applied via a custom-built mechanical exciter, which allowed for using a (larger
than type-A) size of type-B/-C1/-C2 slabs. Hence, low frequency and high amplitude operational
conditions typical of systems (structures) with slow dynamics were simulated. Electrical signals
created by induction to a low-cost interrogation coil (Vishay IWAS) placed at 20 mm from
the vibrating slab’s end, were recorded via a conventional digital oscilloscope at 200 KHz.
C comparing signals from pristine (type-B) or damaged (-C1, -C2) slabs should conclude on the
existence of frequency shifts and ensure that damage diagnosis results could be obtained using
the M-sensor concept.
(a) (b)
Figure 3. (a): DMA oscillation of smart type-A and -A blanc slabs. (b): Response of type-A, and A
blanc slabs (in V) versus time.
(a) (b)
Figure 4. (a): Single cantilever arrangement for testing type-B, -C1 and -C2 slabs. (b): Response of
type-B and -C2 (the most damaged) slabs (in V) versus time.
125
Appl. Sci. 2020, 10, 1029
Note that the coil distance of 20 mm from the slab was defined by a trial-and-error process,
and corresponds to the maximum distance for recording a meaningful signal with the given set up.
Obviously, placing the coil very close (or onto) the surface should provide better signal recording
capabilities, but it would undermine the practical aspects of the concept and could induce sensitivity
to the “lift-off” effects noted, for instance, in applications based on Eddy currents. All recorded signals
were analyzed with respect to their power spectra, estimated via the Welch method (usually coded in
various software packages such as MATLAB® ) with results presented in Section 3.
2.3. Finite Element Method Analysis

For achieving objective O2, the eigenfrequencies of type-B, -C1 and -C2 slabs involved in testing
should be experimentally computed, and compared to their corresponding theoretical values. The latter
were computed via FEM analysis, thus enabling to model and study complete type-B, -C1 and -C2
slabs, meaning that the magnetoelastic sensor (type-B), or even notches (-C1 and -C2 slabs) were
incorporated. FEM COMSOL Multiphysics 5.4 software was used to model the process and compute
the eigenfrequencies and mode shapes of the cantilever slabs with point loading (see Section 3.2).
Free tetrahedral and boundary triangular elements were selected for meshing. Mesh size was selected
as finer, because extremely finer meshes did not have a significant effect on the results. The required
number of elements for meshing was found equal to 166.301 in total, while minimum element quality
was set to 0.05. In all cases, the computations employed second-order polynomials.
3. Experimental Results and Analysis
3.1. Testing for Meaningful Signals with Respect to Noise (O1)

As described in Section 2.2, the DMA testing provided a conventional, yet rigorous means of
testing whether the objective O1 is achievable. The procedure P1 involved a relatively high vibration
frequency peak of 100 Hz and low vibration amplitude of 25 µm defined in the DMA operational
settings. Two type-A blanc and two -A slabs were initially tested in dynamic three-point bending see
Figure 3a. Typical responses of one type-A and one -A blanc in form of electrical signals (in V) versus
time are presented in Figure 3b. Obviously, based on the time plot alone, no quantitative distinction
(for instance, differences on signal variance or mean) between signals from type-A and -A blanc slabs
may be noted. This behavior is not unreasonable, because the external mechanical loading cannot
generate significant electrical power by means of the magnetoelastic phenomenon. Consequently,
when similar passive techniques (i.e., no excitation on the reception coil) are utilized, one should at best
expect qualitative rather than quantitative changes in the signals recorded. In other words, the spectral
characteristics of such signals should be analyzed.
Figure 5 presents the power spectra corresponding to response signals from two type-A blanc and
two A slabs showing clear differences. The insets show significant spikes at around 740 and 1220 Hz
for signals from the two type-A slabs (red and green lines, respectively). At these frequencies, at least,
there is a visible contribution from the M-sensor, superseding the effects of other phenomena (noise
and so on). The latter show up as an output response signal on the coil even when a type-A blanc slab
is tested.
Furthermore, in view of Figure 5, note that the slab excitation takes place at frequencies up to
100 Hz, yet the slab response shows mainly frequency content (harmonics) at higher frequencies (740
and 1220 Hz). The explanation may be two-fold. First, Figure 5 suggests that, at low frequencies (up to
200 Hz), a type-A slab actually produces a signal which has insignificant frequency content, barely
distinguishable from parasitic noise. At higher frequencies and especially at 740, 1220 Hz, or more,
the magnetoelastic strip seems to be generating a signal with more significant frequency content.
This lack of sensitivity might be alleviated if a more efficient (and, thus, expensive) coil was used,
but this would compromise one of the main goals of this work, namely performing damage diagnosis
via conventional, affordable solutions. Second, there is the possibility of the vibrating slab behaving in
126
Appl. Sci. 2020, 10, 1029
a nonlinear manner. External loading of the slab at a given frequency produces a response signal with a
shift of that frequency (and its harmonics), see Ch. 5 in [27]. Even though it is not trivial to state which
explanation is valid, the fact is that, at a specific frequency band, the M-sensor can provide a signal
qualitatively different than that recorded by vibrating type–A blanc slabs, thus achieving objective O1.
Figure 5. Power spectrum of response of two type-A (in green and red) and two -A blanc (in dark grey
and black) slabs tested following procedure P1. Evidence that the hidden strip acts like a sensor are
given in insets.
3.2. Testing for Damage Detection Capabilities (O2)

Section 2.2 described the second objective (O2) which should be achieved by following the testing
procedure P2, namely that of obtaining signals useful for detecting and assessing (diagnosing) damage
in the corresponding slabs, especially for vibrations of low frequency and high amplitude.
Procedure P2 was applied to three type-B slabs (pristine condition), two type-C1 slabs (thus, involving
one notch), and two more -C2 slabs (with two notches-see Figure 2). Testing involved vibrating the free
slab end at 4 Hz, with mean value of 12.5 mm and amplitude of 25 mm, with the response signals (as
generated by the reception coil) recorded at 200 KHz via a conventional oscilloscope. Even though the
sampling frequency is larger than those used in traditional structural monitoring systems, it is by no
means unobtainable, since no particular exotic equipment solutions are needed.
Figure 6 shows the power spectra of the signal responses from the type-B slabs, with insets
presenting frequency regions with the largest spikes from test data of at least two out of the three
type-B slabs tested. From tests with type-A (see Section 3.1), it was deduced that an M-sensor slab
provides meaningful signal (with respect to its frequency content) at high frequencies. At the same time,
a convenient fact is that small damage may be quite clearly indicated at higher modes of vibration [25].
Hence, one should check whether the spikes noted in Figure 6 actually correspond to eigenfrequencies,
so that they could be used for damage diagnosis purposes. Therefore, the FEM analysis described in
Section 2.3 was carried out.
Figure 7 presents indicative results of type-B, -C1 and -C2 slabs for two frequency ranges (around
47,000 and 90,000 Hz). These plots resulted from the finite elements modal vibration analysis and show
the eigenfrequencies and eigenmodes of the combined structure (slab and ribbon) at two significant
frequencies noted in signal power spectral analysis (Figure 6). As seen, type-B slabs show quite different
behavior to that of type-C1, -C2 slabs for both frequency bands. The results from eigenfrequencies
determination for a type-B (among others) slab are presented in Table 1 and compared with those
resulting from Figure 6, whereas their percent error is also shown. Interestingly, the six largest spikes
in Figure 6 are at frequency values extremely close to the eigenfrequencies of type-B slabs. The percent
error shown in Table 1 is almost insignificant, except for the lowest eigenfrequency at 4982.4 Hz, where
it reaches almost 5%. This is in agreement with findings of type-A slab testing (Section 3.1), where the
M-sensor’s output frequency content was significant mainly at higher frequencies. So, finite element
127
Appl. Sci. 2020, 10, 1029
modal analysis has confirmed the values of significant eigenfrequencies of the specimens as detected
in the experimentally obtained signals.
Figure 6. Power spectrum of response of type-B slab tested following procedure P2. Evidence that the
hidden strip acts like a sensor providing specific eigenfrequencies are given in the insets.
(a) (b)
(c) (d)
(e) (f)
(g) (h)
Figure 7. FEM setup and results: (a) Fixed constraints; (b) mesh., Displacements (mm) (c), (d) type-B at
47,402 and 89,994 Hz; (e,f) type-C1 at 47,253 and 90,038 Hz; (g,h) type-C2 at 47,017 and 90,064 Hz.
128
Appl. Sci. 2020, 10, 1029
Table 1. Eigenfrequencies of type-B, -C1 and -C2 slabs as determined by the solution of the finite
elements model compared with the ones picked up in signal power spectra.
FEM Eigenfrequencies in Hz for Power Spectra Eigenfrequencies Absolute Value of % Error:

Indicated Slab: (I) in Hz for Indicated Slab: (II) abs{(I)-(II)}/(I)
-B -C1 -C2 -B -C1 -C2 -B -C1 -C2
4982.4 4986.9 5158.5 5235 5230 5233 5 5 1.4
20,389 20,373 19,985 20,142 20,140 20,140 1.2 1.1 0.8
34,804 34,798 35,257 34,868 34,865 34,868 0.2 0.2 1.1
47,402 47,253 47,017 47,553 47,555 47,555 0.3 0.6 1.1
65,038 65,007 65,061 64,898 64,900 64,895 0.2 0.2 0.3
89,994 90,038 90,064 90,135 90,133 90,133 0.2 0.2 0.1
This result could be anticipated, even without the FEM analysis. In fact, every realistic
implementation of a quasi-sinusoidal excitation would trigger harmonics in the response signal,
with their magnitude in the signal power spectrum following a decreasing trend with growing
frequency values. On the rare occasion of spikes actually becoming larger with growing frequencies,
one may start thinking of such harmonics being close to eigenfrequencies of the tested structure.
Figure 8 shows the power spectra of the signal responses from two type-C1 and another two type-C2
slabs. Power spectra of signals from type-B slabs are also presented for comparison. Considering the
6 largest spikes (as for the case of test data from type-B slabs), it results that type-C1 and -C2 slabs
show spikes at approximately the same frequencies as type-B slabs (see Table 1) except for the spikes at
50,110 Hz and 40,090 Hz.
Figure 8. Power spectrum of response of type-C1 and -C2 slabs tested following P2 (type-B shown for
comparison): Hidden strip acts as a sensor and damage causes spikes at 40 and 50 KHz (see insets).
These values were also cross-validated via FEM analysis: Eigenfrequencies at the region of 50 KHz
were computed at 49,545 Hz and 49,946 Hz for -C1 and -C2 slabs, respectively. Again, as observed at
the region of 40 KHz, eigenfrequencies equal to 39,936 Hz and 40,088 Hz were computed for -C1 and
-C2 slabs, respectively. In all cases, the percent errors of the experimentally obtained (via power spectra)
eigenfrequency values with respect to those computed via FEM analysis are extremely small. In both
cases, the spike occurrences indicate alterations in structural condition, because at these frequencies no
significant content was present in signals from pristine slabs (see also Figure 6). Note that in many
other (mainly lower) frequency areas, one may find circumstantial evidence of spikes associated to
, from some, but not all,damaged slabs,
signals , ,
in some cases , undermined by noise. Hence,
somewhat
such frequency areas have been systematically disregarded, in favor of frequency bands where a
selected number of the most prominent spikes may be clearly denoted.
129
−
Appl. Sci. 2020, 10, 1029
There is also evidence that the damage level (one or two notches) may lead to spikes of different
amplitude, since at both frequency ranges of interest (around 40 and 50 kHz), -C1 slabs feature power
spectra with larger spikes than those from -C2 slabs. The analysis in [28] (specifically in Ch. 10),
proposes an estimation of the damping coefficient from such plots, by associating spikes of narrower
angle (i.e., more abrupt) to lower damping for a given frequency.
The fact that, at both frequencies, the spikes from -C1 slabs (in red and dark red) have a slightly
narrower angle than those from -C2 slabs (in green and dark green) may indicate that damping slightly
increases with notches, as should theoretically be the case [25].
Hence, damage is definitely detectable by considering a number of the most prominent spikes of
the tested slab (mainly at higher frequencies) and comparing with the corresponding group of spikes
from pristine slabs. Currently, frequencies around 40 and 50 KHz seem suitable for detecting damage,
with occurrence of new spikes pointing to structural alteration (notches) of the tested slabs.
𝑦𝑡 = 𝜃 ∙𝑝 𝑡 +𝑒 𝑡
3.3. Testing for Damage Assessment (O2)
𝑒 𝑡 ~𝑁𝐼𝐷 0, 𝜎
Even though detecting damage (notch) occurrence based on signals from testing with procedure P2
is definitely feasible, assessing the different damage levels is less trivial. Recall that damage detectionσ
μ σ
proved possible at two frequency regions (40 and 50 KHz- see Section 3.2). At these regions, spikes
from signals associated to -C1 and σ -C2 slabs differ only slightly. For an answer as to whether damage
level (one or two notches) may be − assessed via−P2, a final series of tests has been carried out. Four -C1
and another four -C2 slabs were tested following P2 and the associated power spectra wereθplotted in
Figure 9. Since four slabs of each damage level are available, ranges of spectrum−magnitudes−at 40 −
and 50−KHz for both damage levels may be formed. These ranges indicate the variation of spectral
θ
magnitude for signals θ
associated θ −with one or two notches.
to slabs
Figure 9. Power spectrum of response of type-C1 and -C2 slab tested following procedure P2. Evidence
that the damage level is only slightly short of being detectable at 40 and 50 KHz is given in insets.
At both frequencies, an obvious trend for signals from more damaged (-C2) slabs having spectra
of lower magnitude than their counterparts from less damaged (-C1) slabs is obvious. Spectra from
-C1 slabs range between −35.5 to −40.5θ dB at 40 KHz and −26 to −32.5 dB at 50 KHz. Spectra from -C2
slabs range between −40 to −43 dB at 40 KHz and −32 to −36.5 dB at 50 KHz. In both frequency areas,
there is one -C1
θ slab whose magnitude overlaps that corresponding to a -C2 slab. Hence, since most
of -C1 slabs have spectra of larger magnitude than those from -C2, procedure P2 provides indicative
(rather than definitive) answers on the damage level of the tested slab.
These conclusions are based on the (rather empirical) estimation of the magnitude of damping
values at each frequency found in [28]. Various approaches (for instance using dedicated equipment to
analyze mode shape) may help to characterize differences induced by the damage level to the slab
130
Appl. Sci. 2020, 10, 1029
vibration modes. In the current study, quantitative damping data have been obtained by the recorded
signals by investing effort in signal modeling procedures, as follows:
1. Response signals recorded by testing each -C1 or -C2 slab using procedure P2 were treated
as stochastic time-series sequences. This was quite realistic, since the recorded signals had
considerable noise due to electromagnetic parasitic phenomena related to the contact-less
recording principle. Furthermore, given that damage may be assessed by means of two frequency
regions (around 40,000 and 50,000 Hz), it was beneficial to perform high-pass filtering of the
raw signals, allowing for a lower number of frequency components to be modeled. One simple
means of doing so, was by differencing the raw signals (that is, forming the differences of each
signal value at instant t minus that at t − 1 for all t) as many times as necessary for effectively
undermining lower frequency (below 40 KHz) components. Naturally, a traditional high pass
filter may also be applied. Here, the raw signals recorded from testing with -C1 and -C2 slabs
(with their spectral characteristics shown in Figure 9) were differenced six times, for obtaining
signals with insignificant frequency content below 40 KHz.
2. Each signal was modeled by means of stochastic output–only AutoRegressive representations
with constant coefficients (CC-AR), identified via standard algorithms. These may be coded in
various programming languages, or even be found in software packages such as MATLAB® .
3. For each identified AR representation, poles at 40 and 50 KHz were monitored with respect to
their damping. Characteristic changes in damping values (at these frequencies) between different
data sets lead to damage assessment conclusions about the slabs generating these data.
The CC-AR representation is a standard yet effective means of modeling dependencies among
values of a stochastic sequence. The stochastic output-only CC-AR representation relating the current
(at time instant t) value of the modeled output signal to older ones has the following generic form:
P
L
y[t] = θi ·pi [t] + e[t]
i=1 (1)
e[t] ∼ NID 0, σ2e
with t designating the normalized discrete time and y[t], e[t] the output and one-step-ahead prediction
error [or residual, assumed to be a zero-mean uncorrelated sequence with variance σ2e signals,
respectively. The term NID(µ,σ2e ) stands for Normally Independently Distributed (with the indicated
mean µ and variance σ2e ). The terms pi [t] are referred to as regressors and involve lagged output values
ranging from y[t − 1] up to y[t − ny], with ny designating the maximum value of lags admitted for
the signal y[t] in (1) (model order). The i-th regressor coefficient is noted θi and has constant value.
For instance, a CC-AR representation designated as y[t] = 0.3·y[t − 1] + 0.7·y[t − 2] − 0.1·y[t − 3] + e[t]
has an order ny equal to 3 and is usually referred to as a CC-AR(3) model. Its regressor coefficients are
θ1 = 0.3, θ2 = 0.7 and θ3 = −0.1.
The objective is to identify the CC-AR representation, that is, to select the regressors pi [t] and
to estimate the associated coefficients θi in order to most effectively represent the dynamics of the
considered signal. The identification of CC-AR representation along with the estimation of coefficients
θi are trivial procedures extensively presented in, among others, [21]. The interested reader may refer
to this work for details on determining the model order and estimating the coefficients. Briefly, a signal
is effectively modeled by a CC-AR(ny) representation if one determines the lowest ny value for which
the associated CC-AR(ny) model accurately describes the frequency content of the signal (by assigning
poles at those frequencies) and the residual sequence e[t] is uncorrelated. The latter means that the
values of e[t] for various t feature insignificant mutual dependencies, as checked by specific hypothesis
tests [21].
At this point, note that identifying models for fault diagnosis applications is a somewhat different
task than performing identification for purposes such as control, signal pre-whitening and so on.
For fault diagnosis, the identified model should include poles with frequencies as close as possible to
131
Appl. Sci. 2020, 10, 1029
the peak locations (frequencies) of interest in the power spectrum plot. At the same time, the identified
CC-AR(ny) model should have as few regressors as possible: Extra regressors (over-modeling) may
lead to fewer correlated residuals e[t] but tend to assign poles even at frequency spikes corresponding
to noise or other parasitic phenomena. Over-modeling should, thus, be avoided, as far as possible.
In the current case, these peaks of interest were located at 40 and 50 KHz. Ideally, the user should
determine the value range for ny, for which all associated identified CC-AR(ny) models include poles
at the desired frequency peaks. In other words, one should plot CC-AR(ny) poles at each frequency as
a function of the corresponding value for ny (stabilization plot), and select a value range for ny leading
to obtaining poles at the desired frequency regions. At the same time, these CC-AR(ny) models should
produce residuals as uncorrelated as possible. Ideally, the model structure selected should have the
smallest ny of those in the range, thus avoiding any over-modeling issues.
In Figure 10a typical stabilization plot from a type-C1 test data and another from a type-C2 slabs
are shown. The considered values for ny ranged from 4 up to 180. With data from either -C1 or -C2
slabs, for ny ≥ 80 the associated CC-AR(ny) models consistently assign poles at frequencies around 40
and 50 KHz. This means that for a given ny satisfying the condition ny ≥ 80, there is always a set of
poles assigned to frequencies close to 40 KHz and another set for those close to 50 KHz. Hence, for the
selected value of ny, the relative difference of pole damping values at each frequency may provide
quantitative information on the damage level, exactly as when the power spectra of test signals are
(qualitatively) examined.
(a) (b)
Figure 10. Stabilization plots with ‘*’ indicating poles at each frequency: (a) type-C1 and (b) type-C2
slabs. For values ny ≥ ≥80, poles are consistently assigned at the monitored frequencies of interest
(around 40 and 50 KHz).
Four such models have been identified on data from type -C1 slabs. Another four CC-AR(80)
models have been identified on data from -C2 slabs. As shown in Figure 11a, all eight models feature
poles located at the frequency band of interest (around 40 and 50 KHz). The autocorrelation coefficients
of CC-AR(80) residuals computed with test data from one -C1 and one -C2 slabs (with an indicative
case shown in Figure 11b) appear uncorrelated, since they are located in-between the dashed lines
designating statistical limits at the 5% risk level (see [21]). Then, clearly a CC-AR(80) representation is
an adequate compromise for modeling test data from -C1 and -C2 slabs.
The pole-zero maps of these eight CC-AR(80) models in Figure 11a show that at 40 and 50 KHz
there is one group of four poles associated to models identified on data from -C1 slabs (red crosses)
which slightly overlaps another group of four poles associated to -C2 slabs (green crosses). It is,
thus, obvious that it is not possible to completely distinguish between the two groups, due to the
small overlap.
132
≥
Appl. Sci. 2020, 10, 1029
(a)
(b)
Figure 11. (a): Pole-zero maps in the complex Im-Re plane of CC-AR(80) models with red crosses
associated to poles corresponding to -C1 slabs and green crosses associated to poles corresponding
to -C2 slabs. (b): Autocorrelation plots indicating uncorrelated CC-AR(80) residuals (values mostly
in-between the dashed limits) for one signal from -C1 and another from -C2 slabs.
However, it is also obvious that the poles associated to -C2 slabs (green crosses) have larger
damping than that from poles associated to -C1 slabs (red crosses). This stems from the fact that
the green crosses are closer to the dashed damping line (designating region of poles with damping
coefficient equal to 0.1) than the red ones. In fact, damping of poles at 40 KHz ranges from 0.0389 to
0.0458 for -C1 slabs and from 0.0436 to 0.0508 for -C2 slabs. Damping of poles at 50 KHz ranges from
0.012 to 0.0272 for -C1 slabs and from 0.024 to 0.037 for -C2 slabs. These conclusions are identical to
those resulting from the analysis based on the power spectral magnitude plots, presented above.
Moreover, tests with ny values up to 130 were also attempted only to find that, even for such
high values of ny, the values in the damping of poles corresponding to -C1 and -C2 data respected the
same relative order. Poles from -C1 consistently featured higher damping than those from -C2 slabs.
Hence, if the relative order in damping values of poles resulting from slabs with different damage is
monitored, then procedure P2 is able to provide quantitative indications about the level of damage in
the tested slabs. Accordingly, the objective O2 was deemed as achievable, if only partly with respect
to damage assessment. It is expected that fine-tuning P2 (or dropping the affordability condition by
investing in high-tech equipment) should lead to smaller amplitude ranges for the power spectra of
signals, thus achieving easier distinction of damage levels.
4. Conclusions
This study proposed a methodology for manufacturing smart slabs with embedded magnetoelastic
sensors aiming at providing damage diagnosis capabilities with conventional, affordable equipment.
Successful application of the proposed damage diagnosis concept requires that the smart slab must
133
Appl. Sci. 2020, 10, 1029
be under continuous monitoring, with the latter starting for a slab in perfect operational condition.
Comparisons of experimental data over time will then detect and evaluate any occurring damage.
The concept considered magnetoelastic strips integrated via 3D printing procedures in CFRP slabs.
During the two testing procedures proposed, the smart slabs emitted (in a passive manner) electrical
signals under vibration loading. These signals were received in a contact-less manner by conventional
low-cost reception coils. The analysis of the power spectra of these signals validated the sensing
concept, since they allow for computing the slab eigenfrequencies at high frequency regions. Such
frequencies may thus be monitorable by combining conventional oscilloscopes, low-cost coils along
with extensive analysis of the signal involved. Furthermore, it was shown that damage occurrence
in slabs caused spikes in specific (high) frequency areas of the signals, and that following a specific
signal modeling procedure, quantitative indications for assessing damage severity could also be found.
Future work is underway to extend the concept from the current basic state (of a cantilever setup
suffering from notches) to damage diagnosis cases of actual structural components. These include
specimens of larger dimensions, implemented in representative to real-life structures, suffering from
various damage profiles such as cracks, delaminations, and so on.
Author Contributions: Conceptualization D.E.M.; methodology D.E.M. and D.G.D.; software D.G.D. and P.J.C.;
validation D.G.D. and P.J.C.; formal analysis D.G.D.; investigation D.E.M. and D.G.D.; resources D.E.M. and
D.G.D.; data curation D.E.M. and D.G.D.; writing—original draft preparation, D.G.D.; writing—review and
editing D.G.D. and D.E.M.; visualization D.G.D. and D.E.M.; supervision, project administration, D.E.M. and
D.G.D. All authors have read and agreed to the published version of the manuscript.
Acknowledgments: The authors would like to thank Professors Evangelia Kontou for kindly providing access
to experimental facilities and Evangelos Hristoforou for the useful discussions and suggestions on this work.
Both Professors are from the National Technical University of Athens (NTUA).
Abbreviations
The following abbreviations are used in this manuscript:
AR autoregressive representation (model)

CC-AR autoregressive representation with constant coefficients (model)
CFRP carbon fiber reinforced polymer
DMA dynamic mechanical analyzer
FBG fiber bragg gratings
FEM finite element method (model)
FDM fused deposition modeling
Hz Hertz
M-sensor specimen with integrated magnetoelastic strip acting as integrated sensor
MFC macro-fiber composite
MPa mega Pascal
NID Normally Independently Distributed
O1 objective 1 (Section 2.2)
O2 objective 2 (Section 2.2)
P1 experimental procedure 1(Section 2.2)
P2 experimental procedure 2(Section 2.2)
PC personal computer
PET-G polyethylene terephthalate glycol
PZT piezoelectric transducer
type-A slab of 60 × 12 × 3 mm3 with integrated Metglas® ribbon
type-B slab of 150 × 25 × 3 mm3 with integrated Metglas® ribbon
type-C1 slab of 150 × 25 × 3 mm3 with integrated Metglas® ribbon and one notch
type-C2 slab of 150 × 25 × 3 mm3 with integrated Metglas® ribbon and two notches
134
Appl. Sci. 2020, 10, 1029
References
1. Mouzakis, D.E. Advanced technologies in manufacturing 3D-layered structures for defense and aerospace.
In Lamination: Theory and Application; Intechopen: London, UK, 2018; Volume 5, pp. 571–596.
2. Khoo, Z.X.; Teoh, J.E.M.; Liu, Y.; Chua, C.K.; Yang, S.; An, J.; Leong, K.F.; Yeong, W.Y. 3D printing of smart
materials: A review on recent progresses in 4D printing. Virt. Phys. Prototyp. 2015, 10, 103–122. [CrossRef]
3. Ota, H.; Emaminejad, S.; Gao, Y.; Zhao, A.; Wu, E.; Challa, S.; Chen, K.; Fahad, H.M.; Jha, A.K.; Kiriya, D.; et al.
Application of 3D printing for smart objects with embedded electronic sensors and systems. Adv. Mater.
Technol. 2016, 1, 1600013. [CrossRef]
4. Ni, Y.; Ji, R.; Long, K.; Bu, T.; Chen, K.; Zhuang, S. A review of 3D-printed sensors. Appl. Spectroc. Rev. 2017,
52, 623–652. [CrossRef]
5. Xu, Y.; Wu, X.; Guo, X.; Kong, B.; Zhang, M.; Qian, X.; Mi, S.; Sun, W. The Boom in 3D-Printed Sensor
Technology. Sensors 2017, 17, 1166. [CrossRef]
6. Willis, K.; Brockmeyer, E.; Hudson, S.; Poupyrevet, I. Printed optics: 3D printing of embedded optical
elements for interactive devices. In Proceedings of the 25th annual ACM symposium on User interface
software and technology (UIST ‘12), Cambridge, MA, USA, 7–10 October 2012.
7. Shemelya, C.; Cedillos, F.; Aguilera, E.; Maestas, E.; Ramos, J.; Espalin, D.; Muse, D.; Wicker, R.; MacDonald, E.
3D printed capacitive sensors. In Proceedings of the Sensors, 2013 IEEE, Baltimore, MD, USA, 3–6 October
2013; pp. 1–4.
8. Shemelya, C.; Cedillos, F.; Aguilera, E.; Espalin, D.; Muse, D.; Wicker, R.; MacDonald, E. Encapsulated copper
wire and copper mesh capacitive sensing for 3-D printing applications. IEEE Sens. J. 2015, 15, 1280–1286.
[CrossRef]
9. Muth, J.T.; Vogt, D.M.; Truby, R.L. Embedded 3D Printing of Strain Sensors within Highly Stretchable
Elastomers. Adv. Mater. 2014, 26, 6307–6312. [CrossRef]
10. Agarwala, S.; Goh, G.L.; Yap, Y.L.; Goh, G.D.; Yu, H.; Yeong, W.Y.; Tran, T. Development of bendable
strain sensor with embedded microchannels using 3D printing. Sens. Actuators A Phys. 2017, 263, 593–599.
[CrossRef]
11. Christ, J.F.; Aliheidari, N.; Ameli, A.; Pötschke, P. 3D printed highly elastic strain sensors of multiwalled
carbon nanotube/thermoplastic polyurethane nanocomposites. Mater. Des. 2017, 131, 394–401. [CrossRef]
12. Huber, C.; Abert, C.; Bruckner, F.; Groenefeld, M.; Muthsam, O.; Schuschnigg, S.; Sirak, K.; Thanhoffer, R.;
Teliban, I.; Vogler, C.; et al. 3D print of polymer bonded rare-earth magnets, and 3D magnetic field scanning
with an end-user 3D printer. J. Appl. Phys. Lett. 2016, 109, 162401. [CrossRef]
13. Na, S.M.; Park, J.J.; Jones, N.J.; Werely, N.; Flatau, A.B. Magnetostrictive whisker sensor application of carbon
fiber-alfenol composites. Smart Mater. Struct. 2018, 27, 105010. [CrossRef]
14. Chatzipirpiridis, G.; Erne, P.; Ergeneman, O.; Pane, S.; Nelson, B.J. A magnetic force sensor on a catheter tip
for minimally invasive surgery. In Proceedings of the 37th Annual International Conference of the IEEE
Engineering in Medicine and Biology Society (EMBC), Milan, Italy, 25–29 August 2015.
15. Lee, H.B.; Kim, Y.W.; Yoon, J.; Lee, N.K.; Park, S.-H. 3D customized and flexible tactile sensor using a
piezoelectric nanofiber mat and sandwich-molded elastomer sheets. Smart Mater. Struct. 2017, 26, 045032.
[CrossRef]
16. Amjadi, M.; Kyung, K.U.; Park, I.; Sitti, M. Stretchable, skin-mountable and wearable strain sensors and their
potential applications: A review. Adv. Funct. Mater. 2016, 26, 1678–1698. [CrossRef]
17. Kong, Q.; Fan, S.; Bai, X.; Mo, Y.L.; Song, G. A novel embeddable spherical smart aggregate for structural
health monitoring: Part I. Fabrication and electrical characterization. Smart Mater. Struct. 2017, 26, 095050.
[CrossRef]
18. Bocherens, E.; Bourasseau, S.; Dewynter-Marty, V.; Py, S.; Dupont, M.; Ferdinand, P.; Berenger, H. Damage
detection in a radome sandwich material with embedded fiber optic sensors. Smart Mater. Struct. 2000, 9,
310. [CrossRef]
19. Ding, G.; Cao, H.; Xie, C. Multipoint cure monitoring of temperature and strain of carbon fibre-reinforced
plastic shafts using fibre Bragg grating sensors. Nondestruct. Test. Eval. 2019, 34, 117–134. [CrossRef]
20. Chang, S.W.; Lin, T.K.; Kuo, S.Y.; Huang, T.-H. Integration of high-resolution laser displacement sensors and
3D printing for structural health monitoring. Sensors (Basel) 2018, 18, 19. [CrossRef]
135
Appl. Sci. 2020, 10, 1029
21. Mouzakis, D.E.; Dimogianopoulos, D.G. Magnetoelastic metglas® sensors: Application of wireless detection
principle and stochastic nonlinear modelling for damage diagnosis in smart systems. In Glass Materials
Research Progress; Wolf, J.C., Lange, L., Eds.; Nova Science Publishers: New York, NY, USA, 2008; pp. 225–257.
22. Yang, Y.; Liu, H.; Annamdas, V.G.M.; Soh, C.K. Monitoring damage propagation using PZT impedance
transducers. Smart Mater. Struct. 2009, 18, 045003. [CrossRef]
23. De Medeiros, R.; Sartorato, M.; Vandepitte, D.; Tita, V. A comparative assessment of different frequency
based damage detection in unidirectional composite plates using MFC sensors. J. Sound Vib. 2016, 383,
171–190. [CrossRef]
24. Ogawa, M.; Huang, C.; Nakamura, T. Damage detection of CFRP laminates via self-sensing fibres and
thermal-sprayed electrodes. Nondestruct. Test. Eval. 2013, 28, 1–16. [CrossRef]
25. Khammassi, M.; Wali, R.; Al-Mutory, A.; Yousaf, A.; Sassi, S.; Gharib, M. Simplified modal-based method to
quantify delamination in carbon fibre-reinforced plastic beam. Nondestruct. Test. Eval. 2019, 34, 283–298.
[CrossRef]
26. Samourgkanidis, G.; Kouzoudis, D.A. Pattern matching identification method of notchs on cantilever beams
through their bending modes measured by magnetoelastic sensors. Theor. Appl. Fract. Mech. 2019, 103,
102266. [CrossRef]
27. Landau, L.D.; Lifshitz, E.M. Mechanics, 3rd ed.; Elsevier, Butterworth-Heinenann: Oxford, UK, 1981.
28. Rao, S.S. Mechanical Vibrations, 5th ed.; Prentice Hall: Upper Saddle River, NJ, USA, 2011.
136
applied
sciences
Article
Stress Estimation Using the Acoustoelastic Effect of
Surface Waves in Weak Anisotropic Materials
Jihyun Jun 1 , Young-Dae Shim 1 and Kyung-Young Jhang 2, *
1 Department of Mechanical Convergence Engineering, Graduate School, Hanyang University,
Seoul 04763, Korea; [email protected] (J.J.); [email protected] (Y.-D.S.)
2 School of Mechanical Engineering, Hanyang University, Seoul 04763, Korea

Received: 20 November 2019; Accepted: 18 December 2019; Published: 24 December 2019
Abstract: This paper proposes a novel stress measurement method using the acoustoelastic effect
of surface wave to estimate the stress of a homogeneous material plate with orthogonal anisotropy,
in which the surface wave velocities are measured in three different directions before and after
loading stress. The effectiveness of the proposed method was verified by numerical simulations and
experiments. For the simulations, the surface wave velocities in three directions were obtained from
a conventional perturbation model for weak anisotropic materials. The simulation results showed that
the stress estimation error was less than 3% for an anisotropic rate up to 2% under stress conditions
up to 90 MPa. Two specimens were prepared for the experiments, one was almost isotropic and
another that had a relatively larger anisotropy rate of 2.6%. Then, the stresses loaded by a tensile
test machine were estimated. The results showed good agreement with the given stresses for both
specimens. These results confirm that the proposed method can be applied to estimate the surface
stress state in anisotropic material plates. The proposed method is simple, practical, and is expected
to be useful for monitoring changes of surface stress before and after machining such as the punching
or bending of plate.
Keywords: acoustoelastic effect; stress estimation; surface waves; weak anisotropy material
1. Introduction
Monitoring of applied or residual stress is important for quality control of industrial parts
manufacturing and health management of structures. The conventional methods of measuring
stress include mechanical methods such as hole-drilling and nondestructive techniques such as X-ray
diffraction (XRD) and acoustoelastic method. However, mechanical methods are complicated and not
suitable for online monitoring. X-ray diffraction has the disadvantage of being harmful to the human
body and requires separate pretreatment of the specimen. Unlike these approaches, the acoustoelastic
method is harmless and has advantages that it can be applied to real products without pre-treatment
and it is suitable for online monitoring.
The acoustoelastic method utilizes the propagation velocity of an elastic wave, which changes
with the stress state of a medium [1]. The acoustoelastic effect appears in all kinds of ultrasonic
waves including longitudinal, transverse and surface waves. For many applications, however, surface
stresses are needed and in this case, surface waves are suitable since they only penetrate to a depth
of approximately one wavelength. In addition, the transmitter and receiver of the surface waves are
located on the same plane so that it is possible to perform inspection using only one side. Furthermore,
it is not necessary to know the exact thickness of the applied object to measure the wave velocity.
These features enhance the application of surface waves in the field. Accordingly, various studies on
stress estimation using acoustoelastic effect of surface wave have been conducted. In particular, many
137
Appl. Sci. 2020, 10, 169
studies have been carried out to analyze stresses in weld [2] or rail specimen [3,4]. In those studies,
however, the direction of the dominant stress is already known, so it is only possible to determine the
presence of the stress or to estimate the stress in the known direction of the dominant stress.
Meanwhile, there is also a method using longitudinal critical refraction (Lcr) waves, which have
advantage of high sensitivity to stress [2,5]. However, the Lcr wave is only applicable in contact
manner because it uses the critical refraction at the contact interface between the wedge and the test
object. On the other hand, surface waves are easy to extend in a non-contact manner such as laser
ultrasonic technology.
Early studies regarding the acoustoelastic effect of surface wave were conducted to identify the
linear relationships between stress and surface wave velocity in isotropic material [6,7]. However,
even if a material is initially isotropic, changes of the micro-structure occur during material production
and processing, which may result in anisotropic properties. In this case, no matter how weak the
anisotropy is, the use of the isotropic theory causes a large error in stress estimation and therefore
cannot be applied.
Only a very limited number of studies have investigated on the acoustoelastic effect in anisotropic
material. Delsanto et al. [8] and Mase and Delsanto [9] proposed a model that separates the contribution
of the anisotropic effect and the contributions of the acoustoelastic effect to evaluate how the surface
wave velocity changes from the isotropic state when a stress is applied in a weakly anisotropic material.
Although, this method has not been experimentally verified, it is worthwhile to analyze the effect of
anisotropy. However, it is very inconvenient for actual application, because it is necessary to know
in advance how the actual elastic modulus of the material differs from the isotropic elastic modulus
before the stress is applied. Another study by Thompson et al. [10] proposed a general solution for
plane wave propagation in a symmetry plane of an orthorhombic, biaxial stressed anisotropic material.
However, it does not provide a direct solution to the case of using surface waves. Additionally, it is
necessary to know the direction of anisotropy in advance, which leads to an inconvenient step of
measuring the shear wave velocity in many different polarization directions.
In this study, we propose a simpler and more practical method for estimating surface stress using
the acoustoelastic effect of surface waves in a plate that has orthogonal anisotropy on its surface. In the
proposed method, we modified the isotropic theory, in which the rate of change of surface wave
velocity according to the stress is expressed using two acoustoelastic coefficients, two principal stresses
and the angle between the principal stress and the wave propagation direction. The difference between
the proposed method and isotropic theory is that in isotropic theory, the surface wave velocity at the
unstressed state is constant regardless of the propagation direction, while in the proposed method it is
dependent on the propagation direction due to anisotropy.
Proposed method can estimate not only magnitude but also direction of principal stress by
measuring the rate of change of the surface wave velocity in any three directions before and after stress
is applied, which does not require any information of stress direction in advance. The requirement to
measure the initial surface wave velocities in three directions before stress is applied also will not be
a problem at all in the comparison of the stress states before and after the machining such as punching
or bending. Furthermore, this method is very simple and practical compared to the previous methods
of Delsanto et al. [8] and Thompson et al. [10]. Since there is no need to pre-measure the directional
elasticity of the material, or pre-check the direction of anisotropy. In addition, this approach requires
measuring only the surface wave velocities in any three directions before and after the stress is applied.
For verification of the proposed method, numerical simulations were performed. For the
simulation, we generated the surface wave velocity data with respect to the propagation direction using
Delsanto’s model for several typical cases when both anisotropy and stress exist. We then analyzed the
error that occurs when the stress is estimated by the proposed method using this data. In the simulation,
three different combinations of the anisotropy direction, principal stress direction and measurement
direction, and two different stress states, uniaxial stress and biaxial stress, were evaluated.
138
Appl. Sci. 2020, 10, 169
To verify the proposed method experimentally, stresses applied by a tensile testing machine
were estimated by measuring the surface wave velocities in three directions. The experiments were
carried out for two types of aluminum plates, one nearly isotropic and one with relatively greater
anisotropy. Two specimens were prepared for each type where one was used for the measurement of
the acoustoelastic coefficients and the other was used for stress estimation. Stress up to 90 MPa was
applied and the performance of stress estimation was verified by comparing the estimated stress with
the applied stress.
2. Theory
2.1. Stress Estimation Method in Isotropic Material

The acoustoelastic effect of surface wave in isotropic materials is expressed as follows,
Vθ − VR K + K2 K − K2
= 1 (σ1 + σ2 ) + 1 (σ1 − σ2 ) cos 2θ (1)
VR 2 2
which represents the relationship between the surface wave velocity change and the principal
stresses [9]. Here, VR is the reference velocity of the surface wave in the unstressed state, Vθ is the
surface wave velocity in the stressed state in the θ direction, and θ is the angle between the surface
wave propagation direction and the principal stress σ1 . σ1 and σ2 are principal stresses, and K1 and K2
are the acoustoelastic coefficients.
The acoustoelastic coefficients K1 and K2 represent the linear proportional coefficients between
the velocity change rate and stress when the surface wave propagates in the stress direction and in the
direction perpendicular to the stress direction under the uniaxial stress condition, respectively. That is,
when σ1 = σ and σ2 = 0, the coefficient can be expressed as follows:
V0 − VR 1
K1 = (2)
VR σ
V90 − VR 1
K2 = (3)
VR σ
where, V 0 is the surface wave velocity in the direction of stress and V 90 is the velocity in the direction
perpendicular to the stress.
To determine the three terms (σ1 , σ2 , θ) in Equation (1), three-directional surface wave velocities
can be used. As an example, using the velocities in the three-directions with a 45◦ difference, as shown
in Figure 1, the following three equations can be obtained.
Vθ − VR K + K2 K − K2
= 1 (σ1 + σ2 ) + 1 (σ1 − σ2 ) cos 2θ (4)
VR 2 2
Vθ−45 − VR K + K2 K − K2
= 1 (σ1 + σ2 ) + 1 (σ1 − σ2 ) cos 2(θ − 45) (5)
VR 2 2
Vθ−90 − VR K + K2 K − K2
= 1 (σ1 + σ2 ) + 1 (σ1 − σ2 ) cos 2(θ − 90) (6)
VR 2 2
In the above equations, Vθ , Vθ−45 , and Vθ−90 are surface wave velocities propagating in the θ,
θ − 45◦ , and θ − 90◦ directions, respectively, in the stressed material, and VR is the reference velocity in
the unstressed state. From Equations (4)–(6), the principal stresses, σ1 and σ2 , and the angle θ can be
determined as follows:
" #
1 1 Vθ V 1 Vθ Vθ−90 1
σ1 = + θ−90 − 2 + − · (7)
2 K1 + K2 VR VR K1 − K2 VR VR cos 2θ
139
σ σ σ
Appl. Sci. 2020, 10, 169
0 1
1
" #
1 1 Vθ V 1 Vθ Vθ−90 1
σ2 = + θ−90 − 2 − 1 − · (8)
2 K1 + K2 VR VR 90 K1 − K2 VR VR cos 2θ
2
 V Vθ−90 Vθ−45
 
1  −1  VR + VR − 2 VR 
θ
θ = tan   (9)

2  Vθ V
− θ−90 
VR VR
σ σ
Using these three equations, when theθacoustoelastic coefficients (K1 , K2 ) and the reference surface
wave velocity (VR ) in the unstressed condition are known, the principal stresses (σ1 , σ2 ) and the angle
(θ) can be estimated using the three-directional measurements of the surface wave velocity.
Figure 1. Schematic of the relationships of the angles between the principal stress direction and
three-measuring directions in 45◦ increments, in which θ θ is the angle from a measuring direction to the
principal stress (σ1 ) direction, and Vθ , Vθ−45 , Vθ−90 are surface wave velocities propagating in the θ,
θ − 45◦ , and θ − 90◦ directions, respectively.
2.2. Stress Estimation Method in Weakly Anisotropic Material

The method mentioned above is based on the theory for an ideal isotropic material, in which the
surface wave velocity in the unstressed condition is constant regardless of the direction of propagation.
However, the actual material is not always produced in a perfectly isotropic condition and has
anisotropic properties. In this case the surface wave velocity is directionally dependent, and the
isotropic method cannot be directly applied. As will be shown later in the numerical simulations,
a very large error occurs even in the case of a very small anisotropy.
To compensate for this anisotropic effect, in this study, Equation (1) for the isotropic case was
modified as follows,
Vθ − Vθ,R K + K2 K − K2
= 1 (σ1 + σ2 ) + 1 (σ1 − σ2 ) cos 2θ (10)
Vθ,R 2 2
where VR in Equation (1) is replaced with Vθ,R which is the surface wave velocity in the θ direction in
the unstressed state. This equation represents the rate of change of the surface wave velocity before
and after applying stress in the θ direction.
Then, the acoustoelastic coefficients K1 and K2 can be rederived as follows under the uniaxial
stress condition (σ1 = σ and σ2 = 0):
V0 − V0,R 1
K1 = (11)
V0,R σ
V90 − V90,R 1
K2 = (12)
V90,R σ
140
Appl. Sci. 2020, 10, 169
where V 0,R and V 90,R are the reference surface wave velocities parallel and perpendicular to the stress
direction in the unstressed state, respectively. As a result, Equations (11) and (12) replace VR in
Equations (2) and (3) with the initial directional velocities, V 0,R and V 90,R , respectively.
Next, for stress estimation using three-directional measurements of the surface wave velocity in
the same way as isotropic theory, we derive the following three equations from Equation (10),
" ! ! #
1 1 Vθ Vθ−90 1 Vθ Vθ−90 1
σ1 = + −2 + − · (13)
2 K1 + K2 Vθ,R Vθ−90,R K1 − K2 Vθ,R Vθ−90,R cos 2θ
" ! ! #
1 1 Vθ V 1 Vθ V 1
σ2 = + θ−90 − 2 − − θ−90 · (14)
2 K1 + K2 Vθ,R Vθ−90,R K1 − K2 Vθ,R Vθ−90,R cos 2θ
  Vθ Vθ−90 Vθ−45 
1  −1  Vθ,R + Vθ+90,R − 2 Vθ−45,R 
θ = tan   (15)
2  Vθ V
− θ−90 
Vθ,R Vθ−90,R
where, Vθ,R , Vθ−45,R , and Vθ−90,R are the reference surface wave velocities in the θ, θ − 45◦ and θ − 90◦
directions in the unstressed state, respectively, and Vθ , Vθ−45 , and Vθ−90 are the surface wave velocities
in the stressed state. As a results, Equations (13)–(15) replace the reference velocity VR in the unstressed
condition in Equations (7)–(9) with the initial velocity in each measurement direction.
3. Numerical Simulations
Numerical simulations were performed to verify the stress estimation performance in typical
stress states when the proposed method was applied to anisotropic materials. For this, the anisotropy
of a material is assumed to be orthogonal, and the acoustic anisotropy rate (η) is defined as the ratio
of the difference between surface wave velocity in the direction in which the wave velocity becomes
maximum (Vmax ) and the surface wave velocity in the direction in which the wave velocity becomes
minimum (Vmin ) to the minimum surface wave velocity (Vmin ) as follows:
Vmax − Vmin
η= (16)
Vmin
The numerical simulation requires three directional surface wave velocities in unstressed and
stressed states. For this, the model suggested by Delsanto et al. [8] and Mase and Delsanto [9] was
used. For the simple analysis applied in the study of Mase, only one elastic constant, C11 , was changed
by anisotropy, and thus Equation (17) can be applied
3
P
Vθ −VR 1 4 ′ aσk vk σ1 +σ2
VR = 2µ A2222 cos φC11 + (aσ0 + µ ) µ
k =1 (17)
3
P a∆k vk σ1 −σ2
+(a∆0 + µ ) cos 2θ µ
k =1
Here, VR is the surface wave velocity in the unstressed isotropic medium, and Vθ is the wave
velocity affected by anisotropy and the acoustoelastic effect in the θ direction. µ, A2222 , ασk , α∆k (k: 0, 1,
2, 3), and νk (k: 1, 2, 3) are material constants of the isotropic material, and C’11 is the anisotropic
variation of C11 . The left side of Equation (17) corresponds to the rate of change of the surface wave
velocity, the first term on the right side reflects the effect of anisotropy on the change of velocity, and
the other terms refer to the contribution of the acoustoelastic effect on the change of velocity In this
model, φ is the angle between the acoustic anisotropy direction and the surface wave propagation
direction, and θ is the angle between the principal stress direction and the surface wave propagation
direction, as shown in Figure 2. The acoustic anisotropy direction is defined as the direction in which
the velocity of the surface wave is maximum.
141
𝜙𝜙
𝜙 θθ
θ
Appl. Sci. 2020, 10, 169
σ θ θθθ
Figure 2. Principal stress (σ1σσ) direction, and three-measuring directions in 45◦ increments, in which
σ ϕ
σ 1σ) direction andϕφϕis the angle to the
is the angle from a measuring direction to the principal stress (σ
θθ−θ−
θ θ,
θ−θ − 45◦ ,
anisotropy direction, and Vθ , Vθ−45 , Vθ−90 are surface wave velocities propagating inθthe
θθθ θ−
θ− θ−
θ−
θ−
θ−
θ−θ− ◦
and θ − 90 directions, respectively.
The three-directional measurement technique was applied using the velocities of the surface waves
propagating towards the θ direction θ and −45◦ and −90
θθ −−− ◦ to the θθθSimulations were performed
−−−θ direction.
for the six cases of the combinations of the different angle conditions shown in Table 1 and different
stress conditions shown in Table 2. Table 1 shows the three cases for setting the anisotropy direction,
principal stress direction and propagation direction. Case 1 is the case where all of the anisotropy,
principal stress (σ1 ), and surface wave σσ
σ propagation directions are the same. In Case 2, the anisotropy
direction and the principal stress direction are coincident, but the surface wave propagation direction
is different. Case 3 is a general case where all directions are inconsistent.
Tableσ 2 shows the given σstresses for the two groups. Group A is the case where only uniaxial
stress (σσ2σ = 0) is applied and σσσ1σis varied among 30, 60 andσ 90 MPa. Group B is σ the case where biaxial
stress is applied while fixing σ1 at 90 MPa and varying σ2 among three stresses (σσσ2 = 30, 60, 90 MPa).
σ σ σσ
Table 1. Directions of the anisotropy, principal stress, and the surface wave propagation for
numerical simulation.
∅ Case 1 Case 2 Case 3

∅∅ θ
∅ 0◦ 30◦ 60◦
θθθ 0◦ 30◦ 30◦
Table 2. Two groups of applied stress setting for numerical simulation.
A B
σ1 30, 60, 90 MPa 90 MPa
σ2 0 MPa 30, 60, 90 MPa
142
Appl. Sci. 2020, 10, 169
The isotropic elastic constants of aluminum in the COMSOL database, which are listed in Table 3,
were used to calculate the surface wave velocity, where λ and µ are Lame constants and l, m and n are
Murnaghan constants.
Table 3. Material constant used to numerical verification.
λ [GPa] µ [GPa] C11 [GPa] l [GPa] m [GPa] n [GPa]

51
σ26 103 −250 −330 −350
σ
The anisotropy rate η changed from 0 to 0.02 in 15 steps by varying the value of C11 . That is,
the anisotropy was varied up to 2%. Figure 3 shows an example λ of the calculation results for η = 0.02, and
principal stress direction is coincidence with the anisotropy direction and principal stress magnitudes
are σ1 = 90 MPa and σ2 = 0. The velocity distribution of the unstressed state has maximum values in
the directions of 0◦ and 180◦ , and a minimum value in the directions perpendicular to them. Also, it is
possible to verify that the velocity distribution is slightly changed by the acoustoelastic effect when
stress is applied. λ μ
To demonstrate the difficulty of applying the isotropic− − to the −anisotropic material first,
theory
we applied the surface wave velocity data set obtained for the Case 1A to the isotropic theory of
Equations (7)–(9), prior toηapplying the proposed method. Figure 4 shows the stress estimation error
of σ1 , in which surface wave velocities in three directions of 0◦ , 45◦ , and 90◦ were used. η As can be
seen from the results, the stress estimation error is ridiculously large even for very weak anisotropy.
σ
Therefore, σ
it is very difficult to estimate the stress of anisotropic materials using the isotropic theory.
Next, to verify the performance of the proposed method, the surface wave velocity data set for
all six conditions shown in Tables 1 and 2 was generated. Stress estimation was performed by using
surface wave velocities in three directions of 0◦ , 45◦ , and 90◦ . The errors between the estimated stress
and given stress of σ1 were then calculated, where the results are shown in Figure 5. In all cases tested,
the errors were less than 3%. In particular, in Case 1A, where all directions of anisotropy, principal
stress (σ1 ) and surface wave propagation are identical, no errors are generated in the stress estimation
σ
regardless of the stress. This is because Delsanto’s model in Equation (17) used in this numerical
simulation is consistent with our modified Equation (10). Actually, by rearranging Equation (17) for the
uniaxial stress and angle conditions of Case 1A, it becomes equal to Equation (10) (see the Appendix A).
Figure 3. Surface wave velocity according to the propagation direction for unstressed and stressed
states in an anisotropic material.
143
estimation1
Appl. Sci. 2020, 10, 169
estimation error [%]

σ
σ
1
Figure 4. Stress estimation errors of the isotropic theory with respect to the anisotropy rate (Case 1A).
(a) (b)
(c) (d)
(e) (f)
Figure 5. Stress estimation errors of the proposed method with respect to the anisotropy rate (a) Case
1A; (b) Case 1B; (c) Case 2A; (d) Case 2B; (e) Case 3A; (f) Case 3B.
Meanwhile, in the case of A with uniaxial stress, the stress estimation errors are independent
of the given stress, and only increase with the anisotropy rate. In the case of B with biaxial stress,
144
Appl. Sci. 2020, 10, 169
the stress estimation errors increase with both anisotropy and stress. However, it must be noted again
that the estimation error in all cases are very small. Based on these results, it was confirmed that the
proposed method for stress estimation can properly estimate stress in a weakly anisotropic material.
4. Specimens and Experiments
4.1. Specimens and Measurement of the Anisotropy Rate

To verify the proposed method experimentally, stress estimation experiments were performed.
Two kinds of Al6061 material with different acoustic anisotropy rates were used as the test specimens.
One (SA) was produced by a hot rolling process and the other (SB) was produced by an extrusion
process. Two test specimens were prepared for each material as shown in Figure 6. One specimen was
used to check the anisotropy and acoustoelastic coefficient, and the other was used to verify the stress
estimation method. The size of the hot rolled specimen is 350 mm × 62 mm × 4 mm (length × width ×
thickness) and the size of the extruded specimen is 350 mm × 60 mm × 5 mm.
Figure 6. Aluminum plate specimens used in experiments.
To verify the surface wave velocity distribution and anisotropy rate of each specimen, the velocity
was measured in 24 directions with an interval of 15◦ . The experimental setup is shown in Figure 7.
PZT transducers with a center frequency of 2.25 MHz and wedges (ABWML-7T-90, Olympus) were
used for surface wave excitation and reception. A pulser-receiver (Olympus 5077PR) was used for
the excitation with a single pulse and a high-resolution digital oscilloscope (Lecroy HDO4034A) with
a 10 GS/s sampling rate was used for precise measurement of small velocity changes. The surface wave
propagation distance was fixed at 10 mm using a jig. The time of flight (TOF) was used to measure the
change rate of the surface wave velocities or surface wave velocity ratios used in Equations (11)–(15).
The TOF was measured based on the arrival time of the peak of the pulse signal. In addition, the delay
time irrespective of surface wave propagation on the specimens, such as the delay in the wedge, was
compensated by calibration in advance.
Figure 7. Surface wave velocity measurement set-up, in which the distance fixing zig holds the wedges
to keep the surface wave propagation distance constant.
145
Appl. Sci. 2020, 10, 169
The measurement results of the surface wave velocity distribution are shown in Figure 8, in which
the velocity was normalized with respect to the maximum velocity. Maximum velocity (Vmax ) and
minimum velocity (Vmin ) in the hot-rolled specimen are 2967.1 m/s and 2957.5 m/s, respectively, and
they are 2951.8 m/s and 2876.2 m/s, respectively, in extruded specimen. Figure 8a display the results of
the hot-rolled aluminum specimen (SA.1) and Figure 8b shows the results of the extruded aluminum
specimen (SB.1). The acoustic anisotropy rate of the hot-rolled aluminum specimen (SA.1) was 0.005
(0.5%) and that of the extruded aluminum specimen was 0.026 (2.6%). This demonstrates that specimen
SA is nearly isotropic while specimen SB has a weak anisotropic feature.
(a) (b)
Figure 8. Surface wave velocity distributions of the two specimens: (a) Hot rolled specimen (SA.1);
(b) Extruded specimen (SB.1).
4.2. Measurement of Acoustoelastic Coefficients

The acoustoelastic coefficients were obtained prior to the stress estimation experiments.
The measurement set-up is shown in Figure 9. Based on the definition of the acoustoelastic coefficient,
the change rates of the surface wave velocity were measured with stress in the stress direction (K1 ) and
in the perpendicular direction (K2 ) in the uniaxial stress state. A tensile tester (Zwick-Roell) was used
to apply the stress. The loading force was varied between 2500 N–22,500 N with an interval of 2500 N.
The loading interval of 2500 N corresponds to a stress of about 10 MPa. – The velocity measurement
set-up is identical to the set-up described in Figure 7.
Figure 10 shows the change of the received pulsed surface wave signal according to the applied
stress. Figure 10a corresponds to the K1 measurement and Figure 10b reflects K2 . As shown in the
zoomed-in box, the peak points of the pulse signal shift depend on the stress increment. That is,
the surface wave velocity in the stress direction decreases with the stress increment, while it increases
in the direction perpendicular to the stress.
Figures 11 and 12 show the change rates of the time of flight (TOF) with respect to the applied
stress in each direction for the SA.1 and SB.1 specimens, respectively. The results were linearly fitted
and the slopes of the fitted lines are the acoustoelastic coefficients defined in Equations (11) and (12).
The obtained acoustoelastic coefficients are listed in Table 4.
146
Appl. Sci. 2020, 10, 169
Figure 9. Set-up for acoustoelastic coefficient measurement. The change in surface wave velocity under
tensile stress loading is measured. For the measurement of K1, transducers (blue colored) are placed in
the loading direction, while for the measurement of K2, transducers (orange colored) are placed in the
direction perpendicular to the loading direction.
(a) (b)
Figure 10. Received signals obtained in the acoustoelastic coefficient measurements: (a) Measurement
of K1 —The peak moves to the right because the velocity of surface wave in the direction of tensile stress
decreases with increasing stress; (b) Measurement of K2 —The peak moves to the left because the velocity
of surface wave in the direction perpendicular to the tensile stress increases with increasing stress.
(a) (b)
Figure 11. Change rates of the TOF with respect to the applied stress in specimen SA.1 for measurements
of (a) K1 and (b) K2 .
147
Appl. Sci. 2020, 10, 169
(a) (b)
Figure 12. Change rates of the TOF with respect to the applied stress in specimen SB.1 for measurements
of (a) K1 and (b) K2 .
Table 4. Values of the acoustoelastic coefficients for specimens SA.1 and SB.1.
Specimen K1 [1/MPa] K2 [1/MPa]

A.1 −132.6 × 10−6 30.5 × 10−6
B.1 − × 10−6 −
−124.0 39.7 × 10−−6
− − −
4.3. Experiments for Stress Estimation

The stress estimation set-up is shown in Figure 13. The experimental set-up is similar to the
acoustoelastic coefficient measurement except for the velocity measurement direction. The surface
wave velocities were measured in three directions with difference of 45◦ , where the angle θ was set to
θ
30◦ . This measurement condition corresponds to the Case 2B in the numerical simulation. Stress was
applied by tensile loading from 7500 N to 22,500 N at an increment of 2500 N. Specimens SA.2 and
SB.2 were used for the stress estimation.
Figure 13. Experimental set-up for the proposed stress estimation.
The stress estimation results of the hot-rolled (SA.2 specimen) and extruded (SB.2 specimen)
aluminum plates are plotted in Figure 14 and the detailed values are given in Tables 5 and 6, respectively.
For the hot-rolled aluminum specimen, the estimated results match the given stresses and angle well.
Similarly, the stress and angle estimations for the extruded specimens also exhibit a good agreement
with the applied values even though this specimen has a larger anisotropy rate than the hot-rolled
148
Appl. Sci. 2020, 10, 169
specimen. The maximum error was 5.14 MPa at a given stress of 90.92 MPa in the hot-rolled specimen,
and the other errors were less than 5.00 MPa for both specimens. As a result, it was confirmed that
stresses can be estimated in weakly anisotropic materials using the proposed method. Note that the
experiments were carried out for the case where the stress in the thickness direction was uniform only
to verify the proposed method. If the stress varies with depth, surface waves of different wavelengths
can be used to probe the stress field at different depths.
(a) (b)
Figure 14. Stress estimation results (a) Hot-rolled specimen (SA.2) and (b) Extruded specimen (SB.2).
Table 5. Estimated results and errors for the hot-rolled specimen (SA.2).
Given stress [MPa] 30.40 40.48 50.57 60.66 70.75 80.83 90.92
Estimated stress [MPa] 28.90 36.42 47.01 61.46 66.54 77.82 85.78
Stress estimated error [MPa] 1.50 4.06 3.56 −0.80 4.21 3.01 5.14
Estimated angle θ [◦ ] 23.39 25.41 26.18 −25.35 20.17 26.23 27.73
Angle Estimated Error𝜽[◦ ] 6.61 4.59 3.82 4.65 9.83 3.77 2.27
Table 6. Estimated results and errors for the hot-rolled specimen (SB.2).
Given stress [MPa] 24.95 33.24 41.50 49.78 58.06 66.34 74.61
Estimated stress [MPa] 21.56 33.43 42.20 48.58 53.87 61.57 69.89
Stress estimated error [MPa] 3.39 −0.19 −0.70 1.20 4.19 4.77 4.72
Estimated angle θ [◦ ] 32.96 34.02 34.94 33.03 34.98 34.58 34.30
Angle Estimated Error [◦ ] −2.96 −4.02 −4.94 −3.03 −4.98 −4.58 −4.30
5. Conclusions
This study proposed a method to estimate stress in anisotropic materials, which involved
modification of the equation for the isotropic theory. In this method, the ratios of the surface wave
velocities in the stressed state to those in the stress-free state in three directions were utilized, enabling
simultaneous estimation of the principal stresses and their directions.
Numerical simulations and experiments were conducted for verification. The results of the
numerical simulation showed that the stress estimation error occurred at a very small order when the
acousitc anisotropy rate was less than 2%. Experiments were performed on hot-rolled and extruded
aluminum plates that had different anisotropy rates of, 0.05% and 2.6%, respectively. The stresses
applied by a tensile tester were estimated by measuring the ratios of the surface wave velocities in
the stressed state to those in the unstressed state in three-directions. The estimated results showed
good agreement with the given stresses in the range from 30 to 90 MPa. These results demonstrate
that the stress estimation method proposed in this study can effectively estimate the stress in weakly
anisotropic materials. The proposed method is simple, practical and is expected to be useful for
monitoring changes in surface stress before and after machining such as punching or bending of plates.
149
Appl. Sci. 2020, 10, 169
Author Contributions: Conceptualization, J.J. and Y.-D.S.; Methodology, J.J. and Y.-D.S.; Software, Y.-D.S.;
Validation, J.J., Y.-D.S. and K.-Y.J.; Formal Analysis, J.J.; Investigation, J.J. and Y.-D.S.; Resources, K.-Y.J.; Data
Curation, Y.-D.S.; Writing-Original Draft Preparation, J.J.; Writing-Review and Editing, K.-Y.J.; Visualization,
Y.-D.S.; Supervision, K.-Y.J.; Project Administration, K.-Y.J.; Funding Acquisition, K.-Y.J. All authors have read and
agreed to the published version of the manuscript.
Funding: This research was supported by the Nuclear Power Research and Development Program through
the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning
(NRF-2013M2A2A9043241).
Nomenclature
A2222 Material constants of the isotropic material
C11 Elastic constant of isotropic material C’11: Variation of C11 according to anisotropy
K1 , K2 Acoustoelastic coefficients
VR Velocity of the surface wave in the unstressed state for isotropic theory
Vθ Surface wave velocities in the θ direction in the stressed state
Vθ,R Surface wave velocity in the θ direction in the unstressed state
Surface wave velocity in the direction in which the wave velocity becomes
Vmax
maximum
Surface wave velocity in the direction in which the wave velocity becomes
Vmin
minimum
l, m, n Murnaghan constants
ασk , α∆k (k: 0, 1, 2, 3) Material constants of isotropic material
η acoustic anisotropy rate
Angle between the surface wave propagation direction and the principal stress (σ1 )
θ
direction
λ, µ Lame constants
φ Angle between the anisotropy direction and the surface wave propagation direction
νk (k: 1, 2, 3) Material constants of isotropic material
σ1 , σ2 Principal stresses
Appendix A
Considering the uniaxial stress (σ1 = σ, σ2 = 0) and angle condition of Φ = θ, Equation (17) can be rearranged
as follows,
X3 X3
Vθ − VR 1 a0k vk σ a∆k vk σ
= A2222 cos4 θC′11 + (aσ0 + ) − (a∆0 + ) cos 2θ (A1)
VR 2µ µ µ µ µ
k =1 k =1
By substituting σ = 0, the velocity in the unstressed state, in the θ direction can be obtained as follows:
1
Vθ,R = ( A2222 cos4 θC′11 + 1)VR (A2)
2µ
On the other hand, the velocity in the stressed state in the θ direction is represented as follows,
 3 3

 1 X a0k vk σ X a∆k vk σ 
 4 ′
Vθ =  A2222 cos θC11 + (aσ0 + ) + (a∆0 + ) cos 2θ + 1VR (A3)
2µ µ µ µ µ
k =1 k =1
Considering the condition of Case 1A shown in Tables 1 and 2, for θ = 0◦ , 45◦ and 90◦ , three-directional
surface wave velocities in the unstressed state can be expressed by Equations (A4)–(A6)
1 1
V0,R = ( A2222 cos4 0C′11 + 1)VR = ( A2222 C′11 + 1)VR (A4)
2µ 2µ
1
V45,R = ( A2222 cos4 45C′11 + 1)VR (A5)
2µ
150
Appl. Sci. 2020, 10, 169
1
V90,R = ( A2222 cos4 90C′11 + 1)VR = VR (A6)
2µ
The three-directional velocities in the stressed state are represented by Equations (A7)–(A9).
 3 3

 1 X a0k vk σ X a∆k vk σ 
 ′
V0 =  A2222 C11 + (aσ0 + ) + (a∆0 + ) + 1VR (A7)
2µ µ µ µ µ
k =1 k =1
 3

 1 1 X a0k vk σ 
V45 = 
 ′
A2222 C11 + (aσ0 + ) + 1VR (A8)
2µ 4 µ µ
k =1
 3

 X a0k vk σ 
V90 
= (aσ0 + ) + 1VR (A9)
µ µ
k =1
The acoustoelastic coefficients defined in Equations (11) and (12) then become the following:
3
P 3
P
a0k vk 1 a∆k vk 1
(aσ0 + µ )µ + (a∆0 + µ )µ
k =1 k =1
K1 = 1
(A10)
( 2µ A2222 C′11 + 1)
3
X a0k vk 1
K2 = (aσ0 + ) (A11)
µ µ
k =1
Finally, Equations (A4)–(A11) are substituted into the stress estimation equations shown in Equations (13)–(15)
to obtain σ1 = σ, σ2 = 0 and θ = 0, which are identical to the given conditions.
References
1. Hughes, D.S.; Kelly, J.L. Second-order elastic deformation of solids. Phys. Rev. 1953, 92, 1145. [CrossRef]
2. Javadi, Y.; Plevris, V.; Najafabadi, M.A. Using LCR ultrasonic method to evaluate residual stress in dissimilar
welded pipes. Int. J. Innov. Manag. Technol. 2013, 4, 170–174.
3. Gokhale, S.J.T.A. Determination of Applied Stresses in Rails Using the Acoustoelastic Effect of Ultrasonic
Waves. Master’s Thesis, Texas A&M University, College Station, TX, USA, 2007.
4. Hughes, J.; Vidler, J.; Khanna, A.; Mohabuth, M.; Kotooussov, A.; Ng, C. Measurement of residual stresses
in rails using Rayleigh waves. In Proceedings of the International Conference on Structural Integrity and
Failure, Perth, Australia, 3–6 December 2018.
5. Bray, D.E.; Tang, W.J.N.E. Subsurface stress evaluation in steel plates and bars using the LCR ultrasonic wave.
Nucl. Eng. Des. 2001, 207, 231–240. [CrossRef]
6. Jassby, K.; Saltoun, D. Use of ultrasonic Rayleigh waves for the measurement of applied biaxial surface
stresses in aluminum 2024-T 351 alloy. Mater. Eval. 1982, 40, 198–205.
7. Lu, W.; Peng, L.; Holland, S. Measurement of acoustoelastic effect of Rayleigh surface waves using laser
ultrasonics. In Review of Progress in Quantitative Nondestructive Evaluation; Springer: Boston, MA, USA, 1998;
pp. 1643–1648.
8. Delsanto, P.; Clark, A.V., Jr. Rayleigh wave propagation in deformed orthotropic materials. J. Acoust. Soc. Am.
1987, 81, 952–960. [CrossRef]
9. Mase, G.; Delsanto, P. Acoustoelasticity using surface waves in slightly anisotropic materials. In Review of
Progress in Quantitative Nondestructive Evaluation; Springer: Boston, MA, USA, 1988; pp. 1349–1356.
10. Thompson, R.B.; Lee, S.; Smith, J.F. Angular dependence of ultrasonic wave propagation in a stressed,
orthorhombic continuum: Theory and application to the measurement of stress and texture. J. Acoust.
Soc. Am. 1986, 80, 921–931. [CrossRef]
151
applied
sciences
Article
Using Acoustic Emission Measurements for
Ice-Melting Detection
Michael Stamm 1,2, * , Helge Pfeiffer 2 , Johan Reynaert 1 and Martine Wevers 2
1 Brussels Airlines, M&E, Brussels Airport, Building 32, 1930 Zaventem, Belgium;
[email protected]
2 KU Leuven, Department of Materials Engineering, Kasteelpark Arenberg 44, 3001 Leuven, Belgium;
[email protected] (H.P.); [email protected] (M.W.)

Received: 29 October 2019; Accepted: 6 December 2019; Published: 9 December 2019
Featured Application: This work contributes to the development of a monitoring system for the
remote detection, localization and quantification of melting ice. Acoustic emission measurements
enable the remote monitoring in an environment where the region of interest is neither optically nor
physically accessible. The kerosene tank of civil aircraft is one example where the aforementioned
accessibility is only achieved with high costs and extensive labor. This is why inter alia aircraft
operators that perform the daily maintenance of civil aircraft are interested in such an inspection
method. Here the detection of ice inside the fuel tanks increases the operational efficiency as well
as the safety for the ground personnel.
Abstract: Aircraft operators being faced with water accumulation in fuel tanks on a daily basis and
are looking for reliable detection systems to determine the remaining amount of accumulated ice
during maintenance after flight. Using such a technology, an increase in the safety and efficiency
of the aircraft operation would be possible in this highly competitive market. This article presents
the use of the Acoustic Emission Technique (AE) for the reliable and non-invasive monitoring of the
melting of ice in fuel tanks. This technology is in principle based on the fact that a phase transition
comes frequently along with stress relaxation that can be used for monitoring the process. Therefore,
the melting of water can, in essence, be monitored with AE without accessing the ice directly.
The analysis of the AE signals has been carried out in the time domain since it was the melting of ice
needed to be monitored as a function of temperature rise time. The insights presented in this paper
can possibly lead to new technologies for ice detection, especially in remote areas that are not easily
accessible with other techniques.
Keywords: acoustic emission; melting ice; aviation; fuel system
1. Introduction
Since the beginning of aviation, innovation in technologies and operational procedures have
steadily increased comfort, performance and safety of air transport. Especially the latter aspect leads to
a continuously decreasing number of fatal accidents even though the number of passengers is steadily
increasing. In this context, it is important that also aging aircraft types are retrofit equipped with
modern technologies whenever feasible.
Besides new technological developments, the competition on the European as well as the global
aviation market increases continuously. This pushes aircraft operators towards more efficient processes.
In order to enhance efficiency, many key innovations focus on the reduction of scheduled and
unscheduled grounding time (downtime) of the aircraft which is the time an aircraft is not available
153
Appl. Sci. 2019, 9, 5387
for flying. However, the procedure to remove liquid water from the fuel systems, called the draining
procedure, requires time for the frozen water inside the tank to melt. The draining procedure makes use
of the fact that liquid water, as well as ice, have a higher density than kerosene. Therefore, liquid water
can be removed from the fuel tank by opening so-called drain valves at the lowest point of the tank
and let the water run out of the tank. As this only removes liquid water from the tank, one must wait
until all frozen water is molten. The time needed for the ice to melt depends on the inner temperature
of the tank and on the outer surrounding temperature. Especially at cold winter times, scheduled
downtimes are insufficient to melt all frozen water. A detailed description of the draining procedure
and the problem occurring during the daily aircraft operation can be found in Stamm et al. [1].
The real-time detection of remaining ice inside the fuel tank will thus contribute to the optimization
of the draining procedure that is performed to remove liquid water from the fuel tank enabling in this
way a considerable reduction of costly downtimes. For the operation in the hangar by aircraft technicians,
a mobile, small and easy-to-use AE system that can monitor the melting ice from outside the tank is
required. To match these requirements already in laboratory experiments, a low-priced system with
off-the-shelf components is used to perform presented experiments.
1.1. Water in Fuel Tanks

During the daily operation of aircraft, water accumulates inside the aircraft fuel tanks and freezes
if the outside air reaches sub-zero temperatures. Different possibilities and root causes of water
contamination in jet fuel are discussed in Baena-Zambrana et al. [2]. Besides the contaminated fuel
that is used to fill up the aircraft tanks, water vapor condensation inside the aircraft fuel tank also
leads to water accumulation inside the tank. Liquid water finally disturbs fuel quantity meters and
enables microbiological corrosion whereas frozen water can harm the internal systems and structures
by e.g., blocking valves or bursting pipes by freezing inside the pipes.
1.2. AE Monitoring of Melting Ice

AE is a well-established monitoring technique in the field of non-destructive testing (NDT)
and many reviews mention applications in the fields of condition and process monitoring and
defect detection. Li [3] gives a suitable review of using AE for monitoring metal cutting processes
and discusses related signal processing methodologies including time series analysis, FFT, wavelet
transform, etc.. Examples for AE-based condition monitoring and defect detection in the civil sector
are described in Behnia et al. [4], Zaki et al. [5] as well as Vestrynge et al. [6] who present their work
on monitoring concrete structures. Schiavi et al. [7] present their work which shows that AE can also
applied to monitor the brittle materials during compression tests.
Additionally, Boyd and Varley [8] list several chemical processes that were monitored with AE
such as the addition of sodium carbonate to calcium chloride [9], metal-acid reaction [10] or the
solid-phase transitions of C2 Cl6 [11].
Sawada et al. [12] measured the acoustic signals during the freezing and melting of super cooled
distilled water. Melting and freezing are accompanied by acoustic activity that is more intense during
the melting. This is the general behavior of phase transitions with a volume decrease being acoustically
more active than transitions with volume expansion [12]. Wentzell et al. [13] also measured the initial
melting phase of distilled water. Since this is believed to be predominated by crystal fracture due
to the high temperature gradient between the sample and the surrounding temperature, no or only
a negligible number of events that occur due to the phase transition itself are measured. Finally,
the mentioned literature provides only the time histogram of acoustic events during the melting of ice
and no additional analysis of the acquired data.
Unlike melting processes, AE measurements on crystallization processes are discussed in some
publications, e.g., by Lube and Zlatkin [14] and Ersen et al. [15] who use AE to estimate the crystal
purity during the growing of ruby and leuco-sapphire crystal and effect of malic and citric acid
on the crystallization of gypsum respectively. Furthermore, Wang and Huang [16] performed AE
154
Appl. Sci. 2019, 9, 5387
measurement on crystallization processes of salicylic acid and concluded that AE can be used as
a reliable technique to monitor the development of the crystallization process.
The possibility of using the AE technique to monitor the melting of ice, first using tab water,
is further explored and documented in this paper. The experimental setup that allows a controlled
environmental conditions will be explained in the following.
2. Experimental Setup
For the AE measurement during melting and freezing, an experimental setup was built that allows
the controlled freezing and melting of a sample while keeping the environmental noise at a minimal
level. In the following, the AE measurement setup that meets the industry requirements described
above and the thermostatic setup are explained.
2.1. Data Acquisition and Processing

A good review of AE measurements of chemical processes is given in Grosse and Ohtsu [17]
where the general approach, usage and advantages of acoustic emission measurements are discussed.
In addition, a general AE setup is described which usually contains an acoustic sensor, followed by
a filtering and amplification of the signals and dedicated signal analysis by the CPU. In the following
all three parts will be explained separately.
The most common AE sensors are piezoelectric (PZT) sensors that use the piezoelectric effect which
transforms elastic motions into voltages [17]. In this work, the commercial broadband piezoelectric sensor
“VS30-V” from the Vallen Systeme GmbH with a flat frequency response between 25 to 80 kHz [18].
The PZT sensor is connected to an AC coupled preamplifier “Model 5670” by Olympus NDT Inc.
This broadband preamplifier provides a very-low-noise 40 dB amplification of ultrasonic signals with
a flat (±1.5 dB) response in a frequency range of 50 kHz–10 MHz [19]. The advantage of the bandwidth
that does not match with the full operating frequency of the PZT is that frequencies below 50 kHz are not
amplified which reduces the picked up ambient noise. Thereafter, a “PicoScope 2205A” USB oscilloscope
digitizes and saves the waveforms with a sampling frequency of f s = 1.565 MHz. The oscilloscope
operates in a simple edge threshold trigger mode with a trigger level of VTHD = +10 mV which was
approximately two time the environmental noise at the beginning of the experiments. With every trigger,
a time windows with a length of t = 2 ms (trigger point at ttrigger = 20% = 0.4 ms) is saved in a file that
contains the time of occurrence.
The settings of the oscilloscope define some additional parameters that are commonly used in AE:
• hit definition time (HDT)

• hit lockout time (HLT)
• peak definition time (PDT)
Classical AE systems allow the explicit setting of these parameters for a optimized operation in
certain conditions. With the setup used in this study, these parameters are given by the operating
mode of the oscilloscope and no additional (software implemented) realization of these parameters is
implemented. The PDT = 1.6 ms is given by the recorded time window for each AE event. The HDT
and HLT are not explicitly set but implicitly given by the oscilloscope settings. The recorded time
window and the duration of the AE previous event determine the HDT such as D + HDT = 1.6 ms.
Due to the short duration of the expected AE events (D ≈ 15% of the recording time window) and the
low quantity of signals, no HLT is required and it can be assumed as zero.
The post-processing of the recorded transients is done in real time in an additional module of the
data acquisition software, which is coded in Python 3.7. Figure 1 shows a typical acoustic burst and
some parameters that are also listed in Grosse and Ohtsu [17] and typically used for parameter-based
analysis. The following parameters are calculated for each transient:
• Amplitude (AMPL) in [mV]
155
Appl. Sci. 2019, 9, 5387
• Duration (D) in [ms]

• Rise time (RT) in [ms]
• Energy (E) in [ms×mV]
Figure 1. A sketch of an acoustic burst with some key indicators that are used for parameter-based
analysis: peak-to-peak amplitude (AMPL), duration (D), rise time (RT), acoustic energy (E) and the
trigger threshold (THD).
2.2. Temperature Controlling System

To measure the acoustic activity of melting ice, not only an AE measurement system but also
a stable thermostatic setup is essential and this was achieved by an in-house-built device that monitors
and controls the temperature in a closed volume. To reach a stable temperature in the range of
−15 to +45 ◦C, two Peltier cooler modules (Adaptive APH-199-14-15-E) are driven by a controller of
Meerstetter Engineering GmbH controller (Meerstetter TEC-1090-HV) and the accompanying software
(TEC Service V 3.00).
A 10 kΩ NTC temperature sensor measures the air temperature inside the box and provides the
feedback for the software to adjust the cooling power of the Peltier elements. To guarantee an equally
distributed temperature inside the box with an inner volume of ca. 4 × 4 × 10 cm3 , a ventilator is
constantly circulating the air.
The explained setup provides, depending on the experiment, a constant temperature or a linear
temperature change until reaching a target temperature. This exactly enables the freezing and melting
of a water under controlled environmental conditions.
2.3. Sensor Setting

Figures 2 and 3 show a sketch and a photograph of the mechanical setup inside the
temperature-controlled box respectively. This concentrates on a good acoustic coupling between
the PZT sensor and the probe and an equally distributed air flow throughout the experiment.
156
Appl. Sci. 2019, 9, 5387
Figure 2. A sketch of the experimental setup inside the temperature-controlled box. An 1 mm thick aluminum
plate is holding the sample of water (a) as well as the PZT sensor (b) and therefore couples the two acoustically.
The internal van (c) ensures sufficient air movement and a uniform temperature inside the box.
Figure 3. A photo of the sensor setup. The specimen and the PZT sensor are positioned on an aluminum
plate which acts as a wave-guide. The plastic cylinder holds the water (for demonstration purposes colored
in blue) and has no bottom so that the water is directly touching the aluminum plate.
The sample holder for the water/ice and the PZT sensor were both positioned on a 1 mm thick
aluminum plate that serves as a wave-guide for the AE signals. To guarantee a good coupling between
the PZT and the plate, some vacuum grease (Dow Corning, High vacuum grease) is used as couplant.
As the sample holder has no bottom, the water/ice is directly touching the aluminum plate and no
additional couplant is needed. The hollow sample holder, thus sealed with vacuum grease, allows
the initially liquid water directly touching the plate. The PZT sensor is placed close to the sample (ca.
2 cm) and acoustically coupled to the plate with some vacuum grease.
Small foam pieces are placed under the aluminum plate to damp mechanical and acoustic
vibrations from the outer housing. The aluminum plate is providing a good acoustic coupling of the
ice and the PZT sensor and reproduces realistic conditions as in a real application (e.g., in an aircraft
fuel tank), where signals will also be transmitted via sheet-structures comparable to the aluminum
plate. Although later applications monitor melting ice systems that that are a factor of 10–100 larger
in size, a down-scaled experimental setup was designed to ensure an equal and thorough controlled
temperature during the measurements with in-house equipment.
As described in Section 2.2, for an optimal performance of the temperature control system, a good
air circulation inside the box is required. The drawback of the internal ventilator is its vibration that
interferes with the AE measurement and is therefore reliably damped with a piece of foam between
the ventilator and the outer housing of the box.
157
Appl. Sci. 2019, 9, 5387
2.4. Measurement
The explained measurement setup allows AE measurements on freezing water and melting ice
under constant environmental conditions in which only the temperature changes and the AE setup
stays untouched.
For the experiments explored in this paper, the water experiences temperature profile as discribed in
Table 1.
Table 1. Temperature profile during the measurement in the temperature–controlled box.
Phase Time Temperature

constant 1h T = 10 ◦C
freezing 3h ∆T↓ = −8 ◦C h−1
constant 2h T = −14 ◦C
melting 3h ∆T↑ = 8 ◦C h−1
constant 2h T = 10 ◦C
3. Results
The gray histograms in Figure 4 show the number of acoustic signals during the melting of a water
sample with different weights between m = 1.0 ± 0.1 g and m = 4.0 ± 0.1 g. The black lines show
the air temperature in the isolated box.
(A) (B)
(C) (D)
Figure 4. Time-dependent histogram of acoustic signals, “#hits” per minute, during the melting of
ice. Four histograms (A–D) for four different amounts of ice (mice = (1 g, 2 g, 3 g, 4 g), upper left to
lower right) plotted against the time (grey). On the second x-axis (right), the air temperature inside the
experimental setup is plotted (black line).
All four histograms show a high acoustic activity when the temperature reaches the melting
temperature. Two peaks of acoustic activity can be identified. The first peak appears at the beginning
of the melting process where the Celsius temperature just reaches a positive value. After an “activity
valley”, another, lower peak appears at the end of the melting phase.
Prior to the increase of temperature in Figure 4 at t < 6 h, the temperature is constant for
approximately two hours. This time included, none of the performed measurements show more than
five detected events until the temperature reaches T ≈ −2.5 ◦C.
158
Appl. Sci. 2019, 9, 5387
The temperature curve confirms the reliable performance of the temperature controlling facility.
The experimental setup reaches the target temperature within its limits and stabilizes the temperature
at the lower plateau at T = −14.00 ± 0.05 ◦C. Taking a possible systematic offset from the NTC
sensor into account, an air temperature uncertainty of less than ∆T = 0.1 K can be assumed .
During the constant raise of target temperature, the temperature shows a small variation of about
∆T ∼30 mK around the target temperature at t∼7 h. The registered time-dependent power that drives
the Peltier-cooling element confirms this fluctuation around the constantly rising target temperature
(not shown in Figure 4). Both can be explained with the latent heat of the melting process which results
an additional “energy consumption” in the system. The temperature control unit reacts with a certain
delay to this, which results in a slower temperature increase. This is followed by an “over-steering” of
the temperature control unit and an increase of the temperature above the target temperature. In the
upper right corner of each plot, the mass of water and total sum of acoustic events is indicated.
As described in Section 2.1, common AE parameters such as AMPL, D, RT and E are calculated
for each acoustic event. This allows a later filtering and detailed parameter-based analysis. However,
in this experiment, only the RT was used to distinguish between AE signal from the ice-melting and
background noise. For this, a RT-filter with a threshold of RT ≥ 0.1 ms was implemented. A different
representation of the data can be seen in Figure 5 where all detected AE events are plotted in a scatter
plot showing their AMPL and the time of occurrence (gray dots). The histograms in Figure 5 show the
accumulated energy per time unit during the melting (black line). The single events, as well as the
histograms which are very similar to the histograms in Figure 5 indicate that there is no significant
change in the AMPL or acoustic energy of the events along the melting process. Also, the other
parameters (D, RT) that are implicitly influencing the acoustic energy do not show any significant
change during the melting process.
(A) (B)
(C) (D)
Figure 5. To show the temporal distribution of the acoustic energy, four plots (A–D) of different
measurements (mice = (1 g, 2 g, 3 g, 4 g), upper left to lower right) show every AE event plotted with
its amplitude AMPL verses the time of occurrence (grey dots). Taking the energy of each event into
account, the temporal distribution of accumulated energy per time unit (black line) shows, similar to
Figure 4, two peaks at the start and end of the melting phase.
Figure 6 shows the acoustic activity during the melting of different samples with varying water
quantities. The solid markers indicate the number of events measured during the complete melting
159
Appl. Sci. 2019, 9, 5387
process. As seen in Figure 4, the AE events in the initial melting phase dominate the total number of
AE events. For the further analysis, the time between the two peaks divide the histograms in two parts
(“first peak”, “second peak”). Besides all events (“all” = “first peak”+“second peak”), the number of
events in the first peak (“peak”) are separately plotted in Figure 6 as circles. Only the missing “second
peak” the measurement with the smallest quantity makes this distinction impossible and it is assumed
that all AE events are in the “first peak”.
Figure 6. The total amount of acoustic signals during the melting of ice plotted against the quantity of
ice. The black and white data points show the complete sum of events and the number of events in the
first peak respectively.
The usage of a weight scale reduces the uncertainty of the quantity of water per measurement
that is assumed to be below ∆mice < 0.1 g.
4. Discussion
The measurements presented above show that the AE technique can monitor melting processes
of ice quantitatively as well as qualitatively. The acoustic activity of ice in an environment with
a constantly increasing temperature starts when the temperature reaches the melting temperature of
water. Independent of the amount of water, the acoustic activity is the highest at the beginning of the
melting process where rates of up to 125 AE events per minute where measured with the described
setup. This number is however strongly dependent on the physical measurement setup, used hardware
and signal filters.
Wentzell et al. [13] already reported a peak in the acoustic activity at the beginning of the melting
process of ice. They identified the temperature gradient between the surrounding and the frozen
sample as a root cause for crystal fractures that cause this peak. In the presented work, the very slow
temperature increase (∆T↑ = 8 K h−1 ) prevents high temperature gradients within the sample and
crystal fractures should not be a relevant source of acoustic events.
After this peak, the rate of the acoustic events decreases and a “valley” with significantly lower
AE activity appears. For the measurements with amounts of water mice > 2.0 g, the acoustic activity
again increases towards the end of the melting process.
This behavior can have multiple reasons that include phenomena related to the intrinsic acoustic
activity during the melting process or the acoustic coupling between source and sensor.
In Figure 7 a schematic, simplified of the temporal histogram of AE events during the melting
process is shown and divided into three parts (A, B, C) that might correspond to different situations
inside the sample holder.
160
Appl. Sci. 2019, 9, 5387
During the first part of the melting (A), the sample is dominated by ice. At a certain point in time,
the constantly increasing temperature results in the melting of ice and stress in the crystal due to the
sudden change in volume during the melting. Therefore, a significant amount of acoustic energy is
released in this initial stage. As seen in Figure 6, the number of acoustic events at the beginning of the
melting process is dominated by the first peak and linear dependent on the amount of ice.
After the first peak, the acoustic activity decreases and the second phase (B) with less acoustic
activity sets in. Therefore, the coupling of the AE source (melting ice) and the sensor changes.
An additional film of water between the sensor and the ice results in an additional surface that reflects
a part of the acoustic energy.
Finally, as described above and seen in all additional measurements, a remarkable increase of observed
acoustic activity is measured at the end of the melting process. In this phase (C), only a little ice amount
is left swimming at the surface of the water reservoir. This ice is therefore in contact with the air which is
warmer than the water in the sample holder resulting in a higher temperature gradient. This could be one
reason for a higher melting rate of the last bits of ice resulting in an increase of the acoustic activity.
To investigate this effect, measurements with a bucket that was closed with a cover and some vacuum
grease were done. In this case, the air inside the bucket was trapped and not interchangeable with the
external air. Therefore, it can be assumed that the effect of the ice touching “warm” air was reduced.
However, also these measurements show the three phases of melting including the second peak at the end
of the melting phase which suggests that the ice touching the “warm” air has no or only a minor effect.
Figure 7. A sketch of a temporal histogram of the acoustic activity during a melting process (above) and
the water/ice composition inside the sample holder (below). The physical composition of the ice in the
sample holder might explain the three phases in the histogram (A–C). For comparison of the sketch with
the measured data, measurement of the mice = 4.0 g sample is shown in the upper right corner.
5. Conclusions
The presented work shows that AE can monitor the melting process of ice which is accompanied
by acoustic signals observed in the ultrasonic frequency range above 30 kHz.
To conduct measurements in a controlled environment, an actively temperature-controlled box
which can be heated and cooled with Peltier elements was build. With this, measurements on melting
processes of different amounts of tap water were conducted. Basic signal analysis based on the time of
occurrence, amplitude and length of acoustic events enables the monitoring of the melting processes.
A systematic pattern of acoustic activity during the melting process with two peaks was measured.
Due to these two peaks, which occur at the beginning and the end of the melting phase, the start,
as well as the end of the melting, can be clearly estimated. In addition to this, the number of acoustic
161
Appl. Sci. 2019, 9, 5387
events in the first peak seems to be correlated with the amount of ice. These results show that AE
can monitor melting processes in systems with a good acoustic coupling between the AE sensor and
the melting ice. This is especially interesting when melting ice must be identified in areas which are
difficult to access, as for instance a fuel tank within an aircraft during the daily maintenance. Here,
information about ice in the fuel tank would improve the safety and efficiency of draining processes
which remove liquid water from the tanks in the hangar.
6. Outlook
The presented work shows that melting of ice can be monitored with AE. However, in this work
only very little information (time of occurrence, AMPL, E, D, RT) were deduced from the acoustic
signals and no spectral analysis was performed yet. Thus, additional information that can be deduced
from a spectral analysis with well-known methods such as modal analysis or principal component
analysis as done by Surgeon and Wevers [20], Johnson [21] respectively. One question is if the two peaks
in the histogram originate from different physical phenomena during the melting and if a detailed
spectral analysis can differentiate the signals in the first and second peak. However, for this, it must
be ensured that the spectral composition of the acoustic transients is not heavily influenced by the
experimental setup and the changing physical properties of the sample during each measurement.
In addition, the influence of physical parameters such as the cooling rate during the freezing,
the purity of the sample and related freezing temperature and chemical composition of the sample to
the acoustic activity have to be investigated. For this, a temperature measurement inside the ice/water
mixture during the experiment would provide further insights into the freezing/melting behavior
of the sample. Combined with repetitive measurements of the same setting and resulting statistics,
fundamental properties of the sample could be derived only by recording and analyzing the acoustic
signals during the melting.
Author Contributions: M.S. recorded and analyzed the presented data and was the major contributor in writing
this paper. H.P. contributed in the design and the interpretation of the experiment. J.R. was mainly responsible
for the technical clarifications and formulations regarding the daily aircraft maintenance. M.W. supervised and
reviewed the analysis of the data and contributed to the interpretation of the results. All authors read and
approved the final manuscript.
Funding: The Research leading to these results has received funding from the “NDTonAIR” project (Training
Network in Non-Destructive Testing and Structural Health Monitoring of Aircraft structures) under the action:
H2020–MSCA–ITN–2016–GRANT 722134.
Acknowledgments: Special thanks to Johan Vanhulst (KU Leuven) for the technical support and to the M&E
department of Brussels Airlines for daily cooperation.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the
study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to
publish the results.
Abbreviations
The following abbreviations are used in this manuscript:
AE acoustic emission technique

NDT non-destructive testing
PZT piezoelectric
AC alternating current
AMPL amplitude
D duration
RT rise time
E energy
THD threshold
HDT hit definition time
HLT hit lockout time
PDT peak definition time
162
Appl. Sci. 2019, 9, 5387
References
1. Stamm, M.; Pfeiffer, H.; Reynaert, J.; Wevers, M. Development of a technique for the detection and
quantification of water and ice in the fuel tanks of an aircraft. ECNDT 2018, 2018, 1–9.
2. Baena-Zambrana, S.; Repetto, S.; Lawson, C.; Lam, J.W. Behaviour of water in jet fuel—A literature review.
Prog. Aerosp. Sci. 2013, 60, 35–44. [CrossRef]
3. Li, X. A brief review: Acoustic emission method for tool wear monitoring during turning. Int. J. Mach.
Tools Manuf. 2002, 42, 157–165. [CrossRef]
4. Behnia, A.; Chai, H.K.; Shiotani, T. Advanced structural health monitoring of concrete structures with the
aid of acoustic emission. Constr. Build. Mater. 2014, 65, 282–302. [CrossRef]
5. Zaki, A.; Chai, H.K.; Aggelis, D.G.; Alver, N. Non-destructive evaluation for corrosion monitoring in
concrete: A review and capability of acoustic emission technique. Sensors 2015, 15, 19069–19101. [CrossRef]
[PubMed]
6. Verstrynge, E.; Pfeiffer, H.; Wevers, M. A novel technique for acoustic emission monitoring in civil structures
with global fiber optic sensors. Smart Mater. Struct. 2014, 23, 065022. [CrossRef]
7. Schiavi, A.; Niccolini, G.; Tarizzo, P.; Carpinteri, A.; Lacidogna, G.; Manuello, A. Acoustic emissions at high
and low frequencies during compression tests in brittle materials. Strain 2011, 47, 105–110. [CrossRef]
8. Boyd, J.W.; Varley, J. The uses of passive measurement of acoustic emissions from chemical engineering
processes. Chem. Eng. Sci. 2001, 56, 1749–1767. [CrossRef]
9. Sawada, T.; Gohshi, Y.; Abe, C.; Furuya, K. Acoustic emissions arising from the gelation of sodium carbonate
and calcium chloride. Anal. Chem. 1985, 57, 366–367. [CrossRef]
10. Cao, Z.; Wang, B.F.; Wang, K.M.; Lin, H.G.; Yu, R.Q. Chemical acoustic emissions from gas evolution
processes recorded by a piezoelectric transducer. Sens. Actuators B Chem. 1998, 50, 27–37. [CrossRef]
11. Wentzell, P.D.; Wade, A.P. Chemical acoustic emission analysis in the frequency domain. Anal. Chem. 1989,
61, 2638–2642. [CrossRef]
12. Sawada, T.; Gohshi, Y.; Abe, C.; Furuya, K. Acoustic emission from phase transition of some chemicals.
Anal. Chem. 1985, 57, 1743–1745. [CrossRef]
13. Wentzell, P.D.; Vanslyke, S.J.; Bateman, K.P. Evaluation of acoustic emission as a means of quantitative
chemical analysis. Anal. Chim. Acta 1991, 246, 43–53. [CrossRef]
14. Lube, E.; Zlatkin, A. In-process monitoring of crystal perfection during melt growth. J. Cryst. Growth 1989,
98, 817 – 826. [CrossRef]
15. Ersen, A.; Smith, A.; Chotard, T. Effect of malic and citric acid on the crystallisation of gypsum investigated
by coupled acoustic emission and electrical conductivity techniques. J. Mater. Sci. 2006, 41, 7210–7217.
[CrossRef]
16. Wang, X.; Huang, Y. An investigation of the acoustic emission generated during crystallization process of
salicylic acid. Powder Technol. 2017, 311, 350 – 355. [CrossRef]
17. Grosse, C.U.; Ohtsu, M. Acoustic Emission Testing; Springer Science & Business Media: Berlin/ Heidelberg,
Germany, 2008.
18. Vallen. AE-Sensor Data Sheet VS375-M. Vallen Systeme GmbH. 2018. Available online: www.vallen.de
(accessed on 9 December 2019).
19. Olympus. Olympus Ultrasonic Preamplifiers, Data sheet: Model 5670; Olympus NDT Inc.: Tokyo, Japan, 2018.
20. Surgeon, M.; Wevers, M. Modal analysis of acoustic emission signals from CFRP laminates. NDT&E Int.
1999, 32, 311–322.
21. Johnson, M. Waveform based clustering and classification of AE transients in composite laminates using
principal component analysis. NDT&E Int. 2002, 35, 367–376.
Sample Availability: The datasets used and/or analyzed during the current study are available from the
corresponding author on reasonable request.
163
applied
sciences
Article
The Teager-Kaiser Energy Cepstral Coefficients as
an Effective Structural Health Monitoring Tool
Marco Civera 1, * , Matteo Ferraris 2 , Rosario Ceravolo 2 , Cecilia Surace 2 and Raimondo Betti 3
1 Department of Mechanical and Aerospace Engineering-DIMEAS, Politecnico di Torino,
Corso Duca degli Abruzzi 24, 10129 Turin, Italy
2 Department of Structural, Geotechnical and Building Engineering-DISEG, Politecnico di Torino,
Corso Duca degli Abruzzi 24, 10129 Turin, Italy; [email protected] (M.F.);
[email protected] (R.C.); [email protected] (C.S.)
3 Department of Civil Engineering and Engineering Mechanics, Columbia University, New York, NY 10027,
USA; [email protected]

Received: 22 October 2019; Accepted: 15 November 2019; Published: 23 November 2019
Featured Application: This paper presents a damage sensitive feature, the Teager-Kaiser Energy
Cepstral Coefficients (TECCs), which can be used to train a Machine Learning algorithm to
perform damage detection and Structural Health Monitoring (SHM) on complex buildings and/or
mechanical systems.
Abstract: Recently, features and techniques from speech processing have started to gain increasing
attention in the Structural Health Monitoring (SHM) community, in the context of vibration analysis.
In particular, the Cepstral Coefficients (CCs) proved to be apt in discerning the response of a damaged
structure with respect to a given undamaged baseline. Previous works relied on the Mel-Frequency
Cepstral Coefficients (MFCCs). This approach, while efficient and still very common in applications,
such as speech and speaker recognition, has been followed by other more advanced and competitive
techniques for the same aims. The Teager-Kaiser Energy Cepstral Coefficients (TECCs) is one of these
alternatives. These features are very closely related to MFCCs, but provide interesting and useful
additional values, such as e.g., improved robustness with respect to noise. The goal of this paper is to
introduce the use of TECCs for damage detection purposes, by highlighting their competitiveness with
closely related features. Promising results from both numerical and experimental data were obtained.
Keywords: damage detection; cepstral analysis; Structural Health Monitoring; Machine Learning;
Teager-Kaiser Energy; gammatone filter
1. Introduction
Damage detection and Structural Health Monitoring (SHM) indicate the field of engineering
that is involved in the assessment of the structural integrity for civil, mechanical, and aerospace
applications, among others [1]. More specifically, vibration-based SHM [2] deals with the structural
response, as recorded by means of a sensor network, to extract damage-related features from its free or
forced oscillations. In this context, recent years have seen a continuous surge of interest in Machine
Learning (ML) algorithms, as they are perfectly suited for damage detection [3], if this is seen from
a Pattern Recognition/Novelty Detection standpoint [4]. The concept is simple: by knowing the current
state of a system (which is usually undamaged, but can be indistinctly previously damaged, as long
as it is stable and not dramatically changing under “normal” conditions), a baseline can be defined.
This is performed by training a “normality” model via ML techniques over the features that were
165
Appl. Sci. 2019, 9, 5064
extracted from the known set of data. Subsequently, the structural response from other scenarios can
be labelled as divergent from this model or not by regarding the value of a distance metric between
its current response and the baseline model. One can refer to the previously cited book of Farrar &
Worden for a detailed description of the basics of ML for SHM [1].
Some damage-related features (also known as damage sensitive features, DSFs) are needed for
training and testing a Machine Learning algorithm. Eventually, DSFs need to be as low in dimensionality
as possible, since a very large number of cases is required for the training phase, preserving the relevant
information while discarding unrelated effects. However, feature selection is a concern much broader
than its SHM applications. One of the fields where it has seen a major development is in speech,
speaker, and language recognition.
Virtual assistants, human-like speech synthesis, and audio indexing are all examples of ML
applied to Speech Processing for Artificial Intelligence (AI) purposes. Other applications are countless:
automatic transcription to text and conversational human-machine interfaces are the most obvious uses
of speech recognition. Speaker recognition, i.e., the process of automatically predicting the identity of
the speaker by a given utterance, is used for voice biometric authentication, telephone-based services,
speaker diarisation, and forensic activities. Finally, language recognition finds its main applications in
emergency call routing, spoken language translation, and audio surveillance, to cite a few. Speaker and
language recognition rely on similar techniques, as they mostly share the same problems: preprocessing
the signal, extracting its relevant features, and representing them by means of statistical models.
There are no conceptual impediments for the application of the related techniques for
vibration-based SHM if acceleration time histories (THs) from the structure of interest are considered
instead of audio samples. As for speaker recognition, distinctive traits, related to specific phenomena
of interest, have to be spotted and discerned from uninteresting effects. On the one hand, mechanical
vibrations and biomedical signals show the same issues, such as changing content along time
(both frequency- and amplitude-wise) during dynamic excitation, which cannot be properly depicted
by means of Discrete Fourier Transform (DFT) [5], and thus require joint time-frequency analysis
techniques for a two-dimensional, time-vs-frequency representation [6,7], such as the one obtained via
wavelet analysis [8–11]. On the other hand, speech signals present some difficulties—such as different
duration between samples, unknown input, and high level of non-stationarity—which are generally
much less impelling for civil and mechanical structures rather than in biological systems [12]. Indeed,
it is possible to figure the human speech production system out as a control system, where the plant
dynamics are related to the vocal tract characteristics, while the constriction and airflow mechanisms
are the biological counterparts of the controller. In this sense, the analogy with structural vibrations is
straightforward. Some other examples of contamination between the speech processing techniques and
structural health monitoring include some recent works on Wavelet Levels (WLs), on Hilbert-Huang
Transform (HHT), and on Empirical Mode Decomposition (EMD) [13,14]. Other proposals include
well-known techniques, such as the Continuous Wavelet Transform (CWT), the Unscented Kalman
Filter (UKF), and the Blind Source Separation (BSS) [5].
By far, the cepstrum [15] has been the main contribution to the field, being often coupled with
other techniques, such as autoregressive moving-average (ARMA) models [16] or directly applied
for the cepstral editing of signals [17,18]; an extensive review of the uses of cepstral features for the
monitoring of mechanical (non-biological) vibrations can be found in [19].
Mel-Frequency Cepstral Coefficients (MFCCs) are an example of low-dimensional, fixed dimension
feature vectors defined into a cepstral framework, which are very effective in speech processing and
already successfully tested as DSFs for mechanical systems [20,21] and subsequently refined [22].
MFCCs proved to be apt in detecting damage in experimental and numerically simulated data for
simpler structural elements and in more complex cases, such as historic masonry buildings. However,
this feature is not free of issues; one of the major problems of MFCCs is how they can be swayed
by additive noise with relative ease. Indeed, log-mel-amplitudes are influenced by the low-energy
component of the signal and this is more strongly evident in the first MF cepstral coefficient.
166
Appl. Sci. 2019, 9, 5064
Similarly to the MFCCs, the Teager-Kaiser Energy Cepstral Coefficients (TECCs) were first
introduced in the ambit of Speech Recognition in 2005 by Dimitriadis et al. [23], relying on the Kaiser
definition of the Teager Energy Operator [24]. The process for their extraction is two-stepped. Firstly,
the signal is filtered through a dense non-constant-Q Gammatone filterbank (details will follow shortly),
which replaces the triangular filters utilised in the standard MFCCs approach. This initial step is shared
with other closely related alternatives, such as, specifically, the Gammatone Cepstral Coefficients
(GTCCs) [25], which have also been tested here in this work. Secondly, the short-time average of the
output of the previously mentioned Teager-Kaiser Energy (TKE) Operator is computed. A deeper
discussion for both MFCCs and TECCs is postposed to the respective subsections; it is important to
remark here how the proposed TECC-based method relies on the TKE operator, which, in turn, is the
basis of several time-frequency analysis approaches for the estimation of instantaneous amplitude
and frequency [26]. These techniques already proved to be efficient in other fields of signal processing
apart speech analysis and very recently they have been successfully applied to Structural Health
Monitoring. For instance, the TKE has been combined with deep belief networks for the fault diagnosis
of reciprocating compressor valves [27] and with least-squares support vector machine (LS-SVM)
classifiers to detect bearing fault [28]. Nevertheless, its use is still very limited and few applications are
reported in the literature.
The proposed TECC-based approach has been tested on two numerical and one experimental
case studies. The first numerical case is a Finite Element (FE) model that represents the Santa
Maria and San Giovenale Cathedral bell tower in Fossano, Italy, which is a structure of relevant
importance for cultural heritage and it has been deeply investigated in the literature (for instance,
see [29]). This is representative of historical masonry buildings, which noteworthy often present
several structural problematics and require proper monitoring [30], especially in the case of high-rise
medieval structures [31,32]. The second case study regards the spar of the high-aspect-ratio (HAR)
prototype wing XB-1 studied in [33]. This case also presents its own difficulties, arising from the high
flexibility of the cantilevered structure. The experimental data that are described in [21] have also been
used for comparison. In this last case study, the damage is modelled as a breathing crack mechanism,
introducing a source of nonlinearities often encountered in damaged structures, where the presence
of crack often results in nonlinear behaviour [34,35]. As a benchmark, the original five alternative
methods that were proposed in [21] and the six new closely-related variants described in [22] have also
been applied to the numerical and experimental data; the results show interesting improvements.
It is important to stress how the choice of different structural systems, materials, and damage
mechanisms for this methodology’s validation is intended to show its reliability for novelty detection
in a wide range of applications and conditions, as “damage” (in the broader sense possible) might
manifest itself in many different ways, also depending on the specific constructional material [36],
and these different occurrences may not be all similarly easy to detect—especially in larger structures
with complicated geometries.
The outline of this paper is as following: in Section 2, the needed background for a full
comprehension of the context is recalled. The algorithm that is utilised for damage detection is
described in Section 3. In Section 4, the numerical case studies are presented. Section 5 briefly describes
the experimental data used for further validation. A comparison with previous works and similar
alternatives is also proposed in the Results Section, and Conclusions follow.
2. Theoretical Background
The novel method here proposed for SHM purposes, based on the Teager-Kaiser Energy Cepstral
Coefficients (TECCs), requires some previous knowledge regarding the definitions of cepstrum,
quefrency [37], and the melody-frequency scale (also abbreviated as the mel-scale) [38]. This Section
is intended to also provide a basic common ground for non-experts in signal and speech processing,
not familiar with the jargon, and to establish some definitions before continuing. For the sake of better
comprehensibility, the discussion is sequenced here into the following subsections: cepstral analysis,
167
Appl. Sci. 2019, 9, 5064
a brief recall about MFCCs, some hints about their (many) variants, and finally the description of the
TECCs themselves.
2.1. Cepstral Analysis

A somehow basic but effective description of the cepstrum (that can be applied to both its power
and complex definitions) is the “spectrum of the spectrum on logarithmic amplitude axis” [39], as it
is computed as the inverse transform of the logarithm of its estimated spectrum. First described
in [37], its original and main use is for pitch detection and determination [40], but in the years it has
found plenty of applications in speech processing, such as for the estimation of harmonics-to-noise
ratios [41]. Resorting to the cepstrum rather than the standard frequency spectrum gives some
interesting advantages, mainly the possibility to easily discern and eventually edit off uninteresting
harmonics. Some slightly different definitions exist, such as e.g., the discrete complex cepstrum
with unwrapped phase, which are useful for time-frequency analysis purposes [42]. Preserving the
phase information, the complex extension of the cepstrum allows for reconstructing the original
time-dependent signal; however, for the purpose of this research, the definition of the power cepstrum
is more convenient. Analytically, this can be defined by resorting to the direct and inverse z-trasform
definitions, as ([43], Chapter 24)
2 " I #2
ˆ −1 2 1 2 n−1
f , Z log H (z) = log H (z) z dz (1)
2πj Γc
where ‘ˆ’ denotes the power cepstrum of function f , Z−1 (·) is the inverse z-transform operator,
H (z) = Z( f ) represents the z-transform of an ordered, discretised time sequence f = x[n], and n
indicates the specific time instant; the contour of integration, Γc , lies within an annular region where
H (z) is defined as analytic and single-valued. Evaluating Equation (1) on the unit circle, the problem
is reconducted to the Fourier domain, and it can be restated as
" Z π #2
1 2
fˆ = log H (ω) e jωt
dt (2)
2π −π
For an angular frequency ω = 2π f and the Discrete Fourier Transform H (ω). The basic idea
exploited is the concept of homomorphic deconvolution, i.e., separating different components of one
given signal, if f itself is made up by a convolution of some distinct underlying signal components.
With respect to the simple, more common Fourier Transform in the frequency domain, this process
allows for better discerning between the content of interest and unnecessary/unrelated influences,
disturbances or echoes, as convolution in time domain becomes summation in cepstral domain.
Thus, thanks to the frequency-warping, the cepstral response simultaneously incorporates properties
from both the time and the frequency domains. Please note that, due to direct and inverse transform,
the measurement unit is the same as the departing one i.e., time expressed in seconds; nevertheless,
fˆ cannot be considered to be an ordered time series, because of the warping. Further details can be
found in [15]. However, the cepstrum of a signal maintains its same dimensionality, without any gain
in conciseness with respect to raw THs. Thus, specific features, i.e., the Cepstral Coefficients (CCs),
have to be extracted from it.
2.2. Cepstral Coefficients (CCs) and Mel-Frequency CCs

The Cepstral Coefficients (CCs) can be extracted from the cepstrum of a signal. In the field of
speech and sound processing, the mel-scale is intensively used for this purpose, by rearranging the
cepstrum in a nonlinear log-like scale of frequency. The idea is generally credited to the works of
Mermelstein [44] and Bridle and Brown [45], and it has since then been developed to become a standard
in speech processing.
168
Appl. Sci. 2019, 9, 5064
The mel-scale is a purely empirically definition, being intended to mimic the auditory system of
the human hear, which is a nonlinear sensing device with almost logarithmically spaced intervals of
equivalent perception. The definition of its base unit, the Mel, is one-thousandth of the pitch (℘) of
a simple tone with a 40 dB amplitude above the auditory threshold and a frequency that is equal to
1000 Hz [38,46]. The reason behind the thresholding lies in several experimental investigations (such
as [47] to name one), which proved that speech processing linear-like for small spectral values and
logarithmic-like for large ones yield to more robust identifications with data affected by additive or
convolutional artificial noise. Obviously, this definition is very specific, perceptual, and suited for the
aim of emulating the human ear. Nevertheless, this was found from previous studies to be apt for
SHM purposes; this might be explained by its logarithmic-like spacing, which is recurrent in many
natural events not limited to speech production and perception [21]. Another point of relevance is
that the definition itself is perceptual, psychoacoustical, and based on perceived equidistance between
subsequent pitches; thus, it is not directly provided with a mathematical formulation. Of the many
concurrent analytical definitions for the Mel–Hertz conversion, the most commonly accepted one is
the variant proposed by Fant in [48]. Other alternative definitions exist; a comparative study was
performed by Umesh et al. in [49].
The procedure to extract MFCCs from a non-stationary recorded signal of interest are the following:
(i) framing of the signal into more stationary tracts; (ii) windowing of the frames; (iii) computation of
the Power Spectral Density (PSD) in the frequency domain; (iv) frequency warping to re-order the
discrete power spectrum in the melody-frequency scale (Mel-scale); and finally, (v) L-points inverse
Discrete Cosine Transform (iDCT). This last passage can be also performed by means of inverse
Discrete Fourier Transform (iDFT), but iDCT is more technically convenient as more𝐿 adherent −1 to the
×
𝑀
Karhunen–Loève Transform. When considering (arbitrarily) L − 1 coefficients 𝒄∈ℝ to be extracted and M
critical bands, the resulting 𝜋(2𝑙
L ×1 , can be + 1)𝑚as
𝑐[𝑙]array
= of MFCCs,𝑎 log(𝐻 c ∈ R[𝑚]) cos defined ,
2𝑀
XM−1
π(2l+1)m
c[l] = am log(Hmel [m]) cos1 2M ,
m=0 ( 1 𝑖𝑓 𝑚 = 0 (3)
𝑀
𝑙 = 0,1, … , 𝐿 − 1, 𝑎 = M i f m = 0
l = 0, 1, . . . , L − 1, am = 22
M i f 𝑖𝑓
m𝑚 ≥≥1 1
𝑀
𝐻 [m][𝑚]
where am𝑎 is the amplitude multiplying the m-th point of the mel-spectrum Hmel . The procedure is
illustrated for the example of frame #5 in Figure 1.
Figure 1. The Mel-Frequency Cepstral Coefficients (MFCCs) extraction procedure; frame #5 is used as
an example. From left to right: the acceleration time history and its corresponding power spectrum,
the power cepstrum of the same, and the resulting MFCCs in the quefrency domain.
2.3. Similar Scales and Features Investigated

While vastly predominant in the common use, the mel-scale is not the sole nonlinear scale
viable. Other options proposed in the literature include the Bark Scale [50] and the Equivalent
rectangular bandwidth (ERB)-rate scale [51], to cite the two main competitors. In both cases, several
Hertz–Bark/Hertz–ERB relationships were proposed in an analytical form, similarly to the Mel-Scale,
169
Appl. Sci. 2019, 9, 5064
as defined in [52] and in [53], respectively. Some of these alternatives, such as the Bark-Frequency
Cepstral Coefficients (BFCCs) [54], have also been recently tested by the Authors for SHM purposes.
In this context, three novel scales, namely, the Energy COntent-based MMFCCs (ECO-MMFCCs),
the MOdal Frequency-centred MMFCCs (MO-MMFCCs), and the LOw-Frequency Restricted MMFCCs
(LO- MMFCCs) have been recently proposed [22].
Apart from different psychoacoustical, Mel-like scales as the ones just mentioned, non-triangular
weighting functions can also be used, such as gammatone filters [55]. Additionally, the acceleration
THs can be pre-processed before applying the extraction procedure. An option is the mentioned
Teager-Kaiser Energy (TKE) Operator. This, combined with the Gammatone filters, is the core of the
method proposed in this work and will be better detailed in the two next subsections.
To sum up and for comparison with the previous studies, Figure 2 reports a pictorial representation
of the tested approaches.
Figure 2. Conceptual relationships between the proposed methodology and the other tested approaches.
2.4. Advantages Over the Auto-Regressive Model Coefficients

The MFCCs, TECCs, and similar cepstral-based features can be compared to other DSFs to better
describe their advantages. As done in [21], the discussion is here focused on the Auto-Regressive
Model (AR) Coefficients, which are a classic feature widely used for SHM purposes [3,4]. An AR model
of order p can be defined (for a time-discrete signal x[n]), as
Xp
x[k ] = a j x [ k − 1 ] + ǫk (4)
j=1
With x[k] being the signal value at the k-th integer multiple of the sampling interval ∆t, a j the j-th
AR coefficient, and ǫk the modelling error at k∆t. The Akaike Information Criterion (AIC) can be used
to select a proper value of p to both prevent over- and under-fitting (i.e., to have a model complete
enough to describe properly the observed data, but not unnecessarily overcomplicated), as

AIC(p) = n ln σ2p + 1 + 2p (5)
170
Appl. Sci. 2019, 9, 5064

where n is the total amount of estimated data point, p is the specific order under investigation, and ln σ2p
is the natural logarithm of the mean of the sum of squared residual errors ǫk between the model and
the data at any data point.
However, with respect to AR coefficients, the cepstral features have the following useful qualities:
(1) they do not require a somehow arbitrary definition of the model order (even if some arbitrariness
does still exist in the definition of the L − 1 coefficients to be extracted from the M critical bands); (2) their
results are much less affected by the choice of L and M with respect to AR’s results accordingly to the
order p; and, (3) the effects of noise, echoes, and nonlinearities can be easier discerned and eventually
removed in the quefrency domain, as they affect separate and distant regions in the logarithmic scale.
A proper comparison between cepstral features and AR coefficients over some experimental data
is postponed to Section 6.2.1.
2.5. Teager-Kaiser Energy Cepstral Coefficients (TECCs)

The TECCs are known in speech recognition and similar applications to perform better than
MFCCs when noise is artificially added, and comparably to them for convolutional (non-additive)
noise and in clean conditions [23]; the extraction procedure of these coefficients is made up by two main
components that distinguish them from the MFCCs. These steps are the definition of the Teager-Kaiser
Energy (TKE) and the convolution with Gammatone filters, which are described here below. The whole
extraction procedure is graphically displayed in the flowchart of Figure 3.
Figure 3. Flowchart of the Teager-Kaiser Energy Cepstral Coefficients (TECCs) extraction procedure.
2.5.1. Teager-Kaiser Energy (TKE)

The Teager-Kaiser Energy (TKE) Operator, also simply known as the Teager Energy Operator,
was firstly mentioned by Teager & Teager [56] in 1990. However, this first occurrence did not report
an analytical definition of it, but only a vaguely defined as the “energy creating the sound”. Thus, it
was Kaiser [24], from which the complete name, to propose a proper definition of it as it is currently
used. An extensive description of it can be found in [57]; the basic concepts are recalled here to provide
a minimal context.
Rather than directly relying on the discretised signal x[n], extracting the CCs from the nonlinear
TKE operator allows for more correctly approximating the actual 𝑥[𝑛]
signal energy. The rationale is that
power spectrum/cepstrum approximate the total energy of the system as the square of its amplitude,
x2 , while the total energy of a vibrating, single degree-of-freedom undamped mechanical system is
given by the sum of 𝑥 kinetic and potential energies, that is to say,
1 1 .2
E = 1kx2 + 1mx (6)
2 2
𝐸 = 𝑘𝑥 + 𝑚𝑥
2 2
Unfortunately, this also means that the TKE is limited by its implicit assumption that the source
of the analysed signal can be approximated to a single undamped SDoF oscillator, which is a clearly
incomplete model. In its discrete form, the TKE of a given signal can be defined as
ψ(x[n]) = x[n]2 − x[n − 1]x[n + 1] (7)
From this definition, the magnitude of the frequency transform is used for the extraction of the
CCs. The process is the same as described in Section 2.2. and represented in Figure 1, except for the use
of TKE operator and of the Gammatone filter in lieu of acceleration time histories and triangular filters.
Noteworthy, the TKE operator can be approximately seen as a high-pass Volterra system, in both
continuous- and discrete-time. In the latter case, as used in this work, the operator might be further
171
Appl. Sci. 2019, 9, 5064
approximated as the product of an appropriate finite impulse response (FIR) filter and of the signal
local (along time) mean. This approximation is only legitimate if the local mean is much larger than
the variance of the signal system; however, this error is much more evident in speech signals, which
are commonly strongly non-stationary, rather than in the mechanical vibration of structures.
2.5.2. Gammatone Filter

The Gammatone is a linear filter that is obtained from the impulse response of a sinusoid
modulated by a gamma distribution function, that is to say,
g(t) = atn f −1 e−2πbERB( fc )t cos(2π fc t + φ) (8)
for a given centre frequency fc . In (8), a is the impulse response function amplitude, n f is the desired
filter order, b the filter bandwidth, and ERB( fc ) is the Equivalent Rectangular Bandwidth defined
by [51]; φ indicates the phase of the carrier. Therefore, the filter frequency response becomes
A 6 A 6
G(ω) = + ) (9)
2 (2πbERB( fc ) + j(ω − ωc ))4 2 (2πbERB( fc ) + j(ω + ωc ))4
with filter gain A and non-dimensionalised ωc = 2π fc . The main advantages of Gammatone filters,
with respect to the symmetric triangular ones originally employed for MFCCs extraction, are that
(i) they are broader and smoother and (ii) they are asymmetrical and non-constant-Q, i.e., the filter
frequency response overlap is not constantly 50%. This second point allows for them to emphasise the
lower frequencies content, which is of greater relevance in both speech processing and Structural Health
Monitoring, while larger bandwidths and less sharp transitions improve the noise robustness [58];
moreover, a Gammatone filterbank is denser in the frequency domain than the mel-frequency-based
triangular filterbank. These advantages with respect to the MFCCs and similar alternatives are
documented for making the TECCs more effective in the ambit of speech and sound processing.
3. Damage Detection Algorithm

The chosen feature, the TEC coefficients, has been used to train a machine learning (ML) algorithm,
obtaining a model of the structural response of interest on unaltered conditions. The algorithm applied
is essentially the one that is described in [21], being the proposed DSF, and not the ML procedure
utilised, the novelty of this paper; the procedure is only briefly restated for completeness.
As for any ML approach, two phases exist—training and testing—where the former can be
executed offline, while the latter may be performed online. During the training phase, the pattern
of the DSF coming from the baseline model is ‘learnt’ by the algorithm, which builds a statistical
model out of it. This model is then used as a comparison for the same feature when extracted from the
structural response under unknown conditions. A metric of distance between any test case and the
reference model is then utilised to discern whether these conditions under inspection correspond to
a structural change that is substantial enough to be linked with damage. This threshold value, as will
be better explained later, is derived via statistical means. This classic statistical framework is well
described in [59].
Signal pre-processing on the acceleration THs was executed, as indicated in [22]. For s acquisition
channels, ntr realisation and thus ntr ·s THs for training the ML algorithm, it is then possible to extract
ntr feature vectors, i.e., c(i) ∈ Rs·(L−1)×1 , where i = 1, . . . , ntr represents the i-th vectors of L CCs, where
L has been arbitrarily chosen and set equal for all of the alternatives tested here (both numerically
and experimentally) to enforce comparability; note that the size L − 1 derives from the first CC, c(i) [0],
being always discarded to mitigate the DC component and other input-specific effects [20].
The following hypotheses have been made. First, the cepstral features are assumed to follow
a Gaussian multi-variate distribution (which is reasonable for a large enough training set and very
commonly used for ML); therefore, all c(i) are considered as being practically uncorrelated and thus
172
Appl. Sci. 2019, 9, 5064
independent identically distributed (i.i.d.) vectors. This is necessary to completely define the ‘normality’
model (i.e., the baseline model of the unaltered reference situation) by means of its sample covariance
matrix Str and the sample mean vector mtr alone. These can be computed, respectively, as
1 Xntr (i) T
Str = c − mtr c(i) − mtr (10)
ntr i=1
and
1 Xntr (i)
mtr = c (11)
ntr i=1
The Squared Mahalanobis Distance (SMD) of a d-dimensional point x from the baseline model is
used as a damage metric; x corresponds to one element of the test set, i.e., to a single test feature vector
c. The physical meaning behind the use of SMD is that features coming from more different structural
e
conditions will be more distant in the feature space; the advantages of this approach, which is the
favourite (and often the default) option for outlier detection, are well-known and properly described
in [21]. Basically, it relies on the sample covariance matrix Str ; this allows for accounting for feature
variability under confounding factors, such as temperature variation, wind speed, or operational
conditions, as long as the given samples are measured during these different, non-damage-related
external conditions. The SMD can be analytically defined as
D2 (x) = (x − mtr )T Str −1 (x − mtr ) (12)
and the procedure can be easily extended and generalized to multiple elements in the test set. As
a Damage Index (DI), D2 (x) needs a threshold (hereinafter, Γ) to discern actual outliers—hopefully,
linked to damaged conditions—to statistical fluctuations of the normality model; it is, essentially,
a lower limit of discordancy to call for novelty. Γ is defined here by exploiting a known property
of the inspected distribution. The distribution of the SMD of a d-variate point x follows a scaled
F-distribution [60], defined by two degrees of freedom, the dimension of the scalar x and the number
of observations in the statistical population used to define the model. In this specific case, they take the
form of d = s·(L − 1) and n = ntr , which leads to
ntr (ntr − s·(L − 1)) 2

D (x) ∼ Fs·(L−1),ntr −s·(L−1) (13)
n2tr − 1 ·s·(L − 1)
where D2 (x) can be directly compared to the F-distribution after being properly scaled. Noteworthy,
this is valid as long as x, which belongs to the test set, has not been used to define the sample statistics’
estimators. Γ is then simply defined correspondingly to the (1 − α)− quantile of F, with α arbitrarily
set to 1%. Moreover, it is mandatory that the total number of training data is larger than the degrees of
freedom of the system analysed (i.e., of the product between the number of cepstral coefficients per
channel and the number of channels) to produce the needed statistics.
4. The Numerical Case Studies

Two numerical case studies are introduced here. These are representative of several cases that
are frequently encountered in Aerospace, Mechanical, and Civil Engineering. In both cases, a 10%
root mean square (RMS) white Gaussian noise (WGN) has been added to the simulated acquisitions;
response THs have been normalised and standardised (by subtracting the mean and dividing by the
standard deviation) at any frame before the feature extraction.
173
Appl. Sci. 2019, 9, 5064
4.1. Santa Maria and San Giovenale Cathedral Bell Tower

The Finite Element Model (FEM) of the Santa Maria and San Giovenale Cathedral bell tower
(as visible in Figure 4) has been used as a numerical case study for the proposed damage-sensitive
feature, as previously proposed for similar alternatives in [22]. The baseline FE Model of the structure
‘as is’ has been calibrated accordingly to the results of a campaign of ambient vibration tests [29], aimed
to establish the mechanical properties of the deteriorated (or cracked) masonry walls. Table 1 provides
material properties, with Young’s moduli defined accordingly to the story level, as also pictorially
described in Figure 2. The story levels are labelled as follow. Level 0: from ground floor to 9.9 m; level
1: from there to 28.2 m; level 2: from the previous level top m to 35 m, and level 3 corresponds to the
belfry (35 m to 46 m high). The octagonal belfry at the top of the bell tower has a wooden rooftop and
0.5 m thick masonry walls, while the thickness of the external perimeter is 1.5 m for the lower levels.
By assuming 𝛼a Rayleigh’s
𝛽 viscous model, the damping parameters α and β were defined from the
experimentally determined
𝑓 = 1.297first two flexural eigenfrequencies, f1 = 1.297 Hz and f2 = 4.253 Hz.
𝑓 = 4.253
Figure 4. Structural scheme of the bell tower (to the right in the image) and picture of its front view (to
the left); autofrettage interventions are clearly visible on the external façades.
Table 1. Fine-tuned Bell Tower baseline Finite Element Model’s (FEM’s) mechanical properties.
Structure-Wise Properties Story-Wise Properties

Rayleigh’s Damping Parameter [-] Young’s Modulus E [MPa]
[𝐌𝐏𝐚]
α 0.0993 Level 0 2.690 103
𝛼 β 0.0180 Level 1 2.690
1.320 10
103
𝛽 [kg/m3 ]
Density Level 2 103
1.250 10
1.320
ρ m 2000 Level 3 103
2.470 10
1.250
Poisson’s
𝜌 ratio [-] 2.470 10
ν 0.3
8-noded, 6 DoF-per-node rectangular shell elements were used everywhere in the FEM. The mesh
size was 0.5 × 0.5 m wherever possible. The distributed connections with the Cathedral were modelled
by means of three-dimensional (3D) spring elements at the contact points [29]. For all simulations
(both on the baseline model and on the damaged scenarios), the acceleration THs were recorded at
the nodes closest to the actual sensor positions visible in Figure 5, along with the 20 directions of
the corresponding acquisition channels. The sampling frequency fs,num 1 was set to 100 Hz. Seismic
174
Appl. Sci. 2019, 9, 5064
strong ground motions were applied as input; 300 spectrum-compatible earthquakes were artificially
generated for this aim accordingly to the Italian normative requirements for earthquake engineering at
the time of the research (Norme Tecniche per le Costruzioni - NTC, D.M. 14th January 2008, Chapter 7)
and in compliance with the Eurocodes. The limit State of safeguard of life was considered. The seismic
intensity parameters are reported in Table 2. Some of the values were fixed, while the remaining
parameters were left free to vary in between realistic ranges, generating 10 combinations. Additive
white Gaussian noise (AWGN) was then inserted to obtain 10 slightly different dynamical inputs from
any of these 30 cases. Elastic spectra were employed, with the smallest period of the response being
equal to 0.02 s, the largest period 4.00 s, the beginning of the stationary part of the accelerogram at 2 s,
and the duration of the stationary part TLVL (NTC §3.2.3.6) free to float in a given range. The baseline
FE model was modified to obtain several damage scenarios. These are, again for comparison’ sake,
the same as utilised in [22]. In all cases, the damage was simulated as a reduction of the Young’s
modulus E at given locations to emulate damage in the masonry external walls. Table 3 lists all the
scenarios; the corresponding first − natural frequency is also reported to highlight the frequency shift
induced by the presence of damage. −
Figure 5. Left: scheme of the sensor placement and direction of the acquisition channels (red, green,
and yellow arrows). Right: the updated Finite Element (FE) model.
Table 2. Seismic intensity parameters.
Fixed Parameters 𝐸 Constrained Parameters

return period TR 475 years TLVL 10–25 s
rated life VN 50 years Damping coefficient 0.05–0.07
Probability of exceedance PVR 0.9
Undrained shear strength CU 1.0
topographic category T1
stratigraphic category C
peak𝐸ground acceleration a g 0.109 g
seismic input duration 35 s
175
Appl. Sci. 2019, 9, 5064
Table 3. Baseline model and other scenarios for numerical validation (Fossano bell tower).
Case Name Description

00 Undamaged As it is, no Young’s modulus reductions
01 1st Damaged 5.00% reduction of the E modulus at Level 0 (all four façades)
02 2nd Damaged 5.00% reduction of the E modulus at Level 1 (all four façades)
03 3rd Damaged 𝐸 at Level 2 (all four façades)
5.00% reduction of the E modulus
04 4th Damaged 5.00% reduction of the E modulus at Level 3 (all four façades)
05–10 5th–10th Damaged 15.00% local reduction of the E modulus at specific, realistic points (as reported in Figure 6)
11 11th Damaged 10.00% global reduction of the E modulus of all levels, all façades
12 12th Damaged 50.0% reduction of spring stiffness at the linking with the Cathedral, x- and y- directions
13 13th Damaged 50.0% reduction of spring stiffness at the linking with the Cathedral, x-direction only
14 14th Damaged 50.0% reduction of spring stiffness at the linking with the Cathedral, y-direction only
15 1st Unaltered𝐸 +0.25% global increase of the E modulus of all levels, all façades
16 2nd Unaltered +1.00% global increase of the E modulus of all levels, all façades
17 3rd Unaltered −0.25% global reduction of the E modulus of all levels, all façades
18 4th Unaltered −1.00% global reduction of the E modulus of all levels, all façades
(a) (b)
(c) (d)
(e) (f)
Figure 6. (a) to (f): Damage Scenarios 05–10, in the same order. The scenarios have been derived by
observation of similar structures damaged by earthquakes; areas with reduced Young’s Modulus are
portrayed in grey.
176
𝑙 = 258
Appl. Sci. 2019, 9, 5064
Cases 01 to 04 only present a 5% reduction of E along one of the four stories of the structure,
on all façades. Cases 05–10 are realistic post-earthquake scenarios, which are based on the Authors’
previous experience and intended to model typical damage pattern encountered in similar cases;
they are portrayed in Figure 6. Case 11 is a global stiffness reduction of 10% throughout the whole
structure. Cases 12–14 emulate damage to the linking with the adjacent cathedral, with 50% of the
spring stiffness. Finally, four ‘unaltered’ scenarios were also accounted for: as for case 11, a global
variation of E was applied; specifically, a reduction/increase equal to 1% and 0.25%. This was intended
to simulate statistical fluctuations of the material properties, which cannot be unequivocally related to
the presence of occurring damage.
4.2. High Aspect Ratio Flexible Wing

The second case study concerns the spar of a prototype wing, the linear and nonlinear dynamics
of which have been investigated in recent studies [61,62]. Importantly, the prototype skin is supposed
to completely transfer the aerodynamic loads to the spar, making the structural behaviour of the latter
a good approximation of the dynamical response of the whole spar-skin ensemble. The HAR, highly
flexible wing is cantilevered, its geometry narrowing toward the tip with taper ratio changing at
l = 258 mm (hereinafter, the mid-length section). Table 4 reports other geometrical and mechanical
properties. The wing planform presents a tapered swept configuration, with trapezoidal planform,
swept leading edge, and inward-kinked trailing edge. In this case, the fatigue damage that is induced
by subsequent load cycles has been modelled as a reduction of the Young’s modulus at the two most
probable locations, i.e., at the clamped end and at the mid-length section, as portrayed in Figure 7 and
enlisted in Table 5. The input, a perpendicular acceleration applied on the wingtip, was defined as
a white Gaussian noise of peak value 1.5 g, sampled at fs,num 2 = 256 Hz for 5 s. 50 realisations were
numerically simulated on the undamaged case for training; 10 others per case were used for testing the
trained model. Five output channels were considered, mimicking the actual sensors array (plus the
point of application of the laser Doppler velocimeter) actually used in the previous experimental
studies and depicted in Figure 7.
ECO- and MO-MMFCCs, firstly defined in [22] for the specific case of the Santa Maria and San
Giovenale Cathedral bell tower, were adapted to this second numerical case. For the ECO-MMFCCs,
90% of the energy content was found at 43.76 Hz, by taking the mean over the five output channels.
Therefore, this value has been set as the new cut-off frequency. The centre frequencies for MO-MMFCCs
were defined from the modal analysis of the wing spar and benchmarked and validated against
experimental data obtained in [61,62]; they correspond, in order of increasing natural frequency, to the
1st, 2nd, 3rd, and 4th flapwise bending mode; the 1st torsional mode; the 5th flapwise flexural; the 1st
chordwise bending mode; the 2nd torsional; and, the 6th flexing modes. The tenth mode (the 7th
flapwise) fell over the maximum investigable frequency at 333.88 Hz and was therefore replaced by
adding ‘1′ to the begin of the array, thus resulting in
fc = [1 5.49 23.16 55.80 103.99 125.11 172.09 189.15 219.90 255.62]
To conclude, the low frequency range of LO-MMFCCs was not changed and it remains defined
between 0 and fs /2, which here becomes 128 Hz.
For this second case, the major challenge comes from even large reductions in E resulting in
minimal differences in terms of frequency shift, as can be seen in the last column of Table 5.
177
Appl. Sci. 2019, 9, 5064
Table 4. Geometrical and mechanical properties of the wing spar.

333.88
′
Density ρ 2850 kg/m3
Young’s Modulus E 7.31 × 1010 Pa
𝑓 = [1 5.49 23.16 55.80 103.99 125.11 172.09 189.15 219.90 255.62
Poisson Ratio v 0.33
] -
Free length (clamp to tip) ltip 706 mm
Thickness t 2 mm
𝑓 /2 Max width at clamped section bmax 180.00 mm
Mid-length width at the section of changing tampering (l = 258 mm) bl=258 56.10 mm
Min width at the tip section bmin 17.04 mm
(a)
(b)
Figure 7. (a) Areas subject to decreased Young’s modulus (b) Damage scenarios.
Table 5. Baseline model and other scenarios for numerical validation (HAR wing).
ρ 1st Natural
Case Name Description
Frequency [Hz]
00 Undamaged As it is, no Young’s modulus reductions ∙ 10 5.490
01 1st Unaltered 1.00% reduction of the E modulus in areas 2A, 3A, and 4A 5.488
02 2nd Unaltered 1.00% reduction of the E modulus in area 5A 5.489
03 1st Damaged 50.00% reduction oftipE in areas 2A, 3A, and 4A 5.270
04 2nd Damaged 50.00% reduction of E in area 3A 5.330
05 3rd Damaged 50.00% reduction of E in areas 2A, 3A, 4A and 5A 5.168
06 4th Damaged 50.00% reduction of E in areas 3A and 5A 5.224
07 5th Damaged 50.00% reduction of E in areas 2A and 4A 5.460
max
5. The Experimental Case Study (Frame Behaving Nonlinearly)

258 258
For the experimental validation, a laboratory three-story frame, which behaves nonlinearly due to
damage occurrence, has been investigated; thisminis the same case studied by [21], again to ensure the
comparison of the results, and proposed by [3]. Please note that this is quite different from the two
numerical cases, where the structural damage was not a source of nonlinearity. This is due to the goal
178
Appl. Sci. 2019, 9, 5064
of this experimental setup being to mimic breathing crack behaviour [63]. A bumper–column bilinear
mechanism achieves this: the column, hanging from the third floor, meets the bumper at the second
story. This mechanism generates a directional nonlinear response, with stiffness increasing when the
two are touching and pushing against each other. A scheme of the test structure is reported in Figure 8;
the shaker and the four sensors’ locations are highlighted. A detailed description of the experimental
setup and procedure is provided in [59]. The severity of the damage is modelled by moving closer or
farther the column’s tip—when they are closer, the two elements come in touch for lower amplitudes
of the driving force, thus causing major nonlinearities in the recorded response. The experimental THs
have a total duration of 25.6 s, with fs,exp = 320 Hz. The dynamic input is a band-limited excitation
(ranging 20–150 Hz), as defined in [3]. The 17 states (nine undamaged and eight damaged, with the
first nine configurations being intended to emulate different operational and environmental conditions)
were used for this investigation; they are summarized in Table 6. All the output channels were used
for the sensor setup considered. The test set was made up of 10 realisations for each scenario, totalling
180 simulated THs.
Figure 8. The experimental test setup schematics.
Table 6. Baseline, Undamaged and Damaged Scenarios for experimental validation.
Case Description
1 Undamaged Baseline
2 Undamaged Added mass of 1.2 kg at the base
3 Undamaged Added mass of 1.2 kg at the 1st floor
4 Undamaged 87.5% stiffness reduction in one column of the 1st inter-story
5 Undamaged 87.5% stiffness reduction in two columns of the 1st inter-story
6 Undamaged 87.5% stiffness reduction in one column of the 2nd inter-story
7 Undamaged 87.5% stiffness reduction in one column of the 2nd inter-story
8 Undamaged 87.5% stiffness reduction in one column of the 3rd inter-story
9 Undamaged 87.5% stiffness reduction in two columns of the 3rd inter-story
10 Damaged Distance between bumper and column tip 0.20 mm
15 Damaged Bumper 0.20 mm from column tip, 1.2 kg added at the base
16 Damaged Bumper 0.20 mm from column tip, 1.2 kg added on the 1st floor
17 Damaged Bumper 0.10 mm from column tip, 1.2 kg added on the 1st floor
As for the second numerical case study, the definitions of ECO- and MO-MMFCCs have been
changed to fit the structure of interest. For ECO-MMFCCs, the 90% bound of fco is approximately
78 Hz (averaging over the channels 3, 4, and 5 for the 50 realisations of Case 1); the fs,exp /2 limit of
LO-MMFCCs becomes 160 Hz. The array of the centre frequencies for MO-MMFCCs underwent some
179
𝑓 =
𝑓 =
Appl. Sci. 2019, 9, 5064
major rethinking. Being the structure basically assimilable to a three degrees-of-freedom oscillator
(floor level does not count as it is subject to the input-induced rigid motion), only the first three
frequencies have been retained. The remaining seven filters have been linearly spaced up to fs /2,
i.e., 160 Hz, thus resulting in
fc = [31 54 71 83 96 109 122 135 148 160]
6. Results
The results from previous works are reported for direct comparison. The methods applied are,
by following the same nomenclature (see also Figure 2), (1) the log-scale CCs; (2) the linear CCs; (3) the
fs
classic definition of MFCCs; (4) the MFCCs with a cut-off frequency fco = 4; and, (5) the MFCCs
fs
with a cut-off frequency fco = from [21]; the three proposed Mel-Modified Cepstral Coefficients
8
(MMFCCs), i.e., the so-called (i) ECO-MMFCCs; (ii) MO-MMFCCs; and, (iii) LO-MMFCCs, from [22],
plus the Bark-scale based (iv) BFCCs; finally, the novel features here presented, (I) the GTCCs and (II)
the TECCs.
For better readability, this Section is split between the numerical and the experimental data. All
the results are expressed in terms of type 1 (false positives, i.e., false alarms) and type 2 (false negatives
or false acceptance) errors. In the latter case, the system is declared healthy while not being so;
the opposite happens for false alarms. Being life-safety the main aim of any SHM procedure, type
1 errors are generally overlooked respect to their type 2 counterparts. Nevertheless, economical,
practical, and psychological concerns make these as valuable as the other ones. A simple yet effective
reasoning is that an alarm system constantly affected by false alarm will most probably be ignored
when actual damage is effectively spotted. This nullifies any possible gain from the deployment of the
SHM apparatus.
6.1. Results from the Numerical Simulations

Regarding the first numerical case study (the Fossano bell tower), cases 00 to 04—i.e., the baseline
and the story-uniform damage cases—are visually compared in Figure 9 to prove the algorithms
capability to discern simpler configurations and widely extended damage from the normality model.
Figure 10 shows the results for cases 05 to 10 (realistic damage scenarios). To estimate the robustness of
each MFCC-based variant to false alarms, cases 15 to 18 (i.e., the scenarios with small fluctuations of
the Young’s modulus respect to the baseline) are presented in Figure 11. Apart from the Teager-Kaiser
Energy Cepstral Coefficients, the standard definition of MFCCs is also shown for comparison. Table 7
reports all results. In all figures, the black dashed line represents the defined threshold; bars exceeding
it are coloured in red and they stand for the realisations labelled as damaged, while blue bars
indicate values of DI under the prescribed threshold. From a statistical pattern recognition standpoint,
this means that the cepstral features that correspond to that specific realisation have not departed
considerably enough from their respective values for the reference baseline and are therefore labelled
as undamaged.
As can be inferred from the first two columns of Table 7, the GTCCs, the Logarithmic Scale CCs,
and the three Mel-Modified Scales (ECO-, MO-, and LO-MMFCCs) have all relatively low type 1 errors.
The Damage Index values inside any given damage case are quite similar; that is a valuable proof of the
robustness of the method. Bark-scale based CCs wrongly labelled as damaged 60% of the undamaged
inputs coming from the same scenario used for the training. The TECCs were the most performing
feature; in Figure 9.a one can clearly see that two out of the three mislabeled realisations for TECCs
fall very short of the threshold. The third column in Table 7 (cases 05–14) shows that all algorithms
work effectively in terms of damage detection. The classic definition of the Mel-Scale, the Bark-Scale,
and the linear Scale nevertheless present some relatively larger fluctuations along the ten realisations
of the same damage scenario with different inputs. The Log-Scale and the Mel-modified Scales provide
more stable results, with a flat trend inside each damage scenario. This seems to indicate that a kind
180
Appl. Sci. 2019, 9, 5064
of nonlinear scale performs better than the constantly spaced filters. The filter shape might be even
more incisive, as GTCCs and TECCs are all spaced according to the Bark Critical Bands yet perform
better than triangular filters-based options both in terms of type I and II errors. Most of the algorithms
recognise substantial gaps (Figures 8 and 9) between different scenarios, which reflect the difference of
damage extension; this seems to indicate a possibility to also exploit them for instances of damage
severity assessment. Finally, the last column of Table 7 (cases 15 to 18) evidently portrays how only the
TECCs correctly label all the proposed cases as unmodified respect to the baseline; the other features
proved to be quite unsatisfactory, especially the Mel-Scale and Bark-Scale. That seems to once more
validate the assumption that the straight application of the original definition of the mel-scale, tuned
for speech processing aims, is not well-performing and proper corrections are needed. In particular,
the scenarios with a variation in the material properties of ±0.25% are most often correctly labelled
as healthy; otherwise, the ±1.00% deviation of the Young’s modulus provides DI values above the
threshold. This phenomenon could be justified by the fact that even a relatively small difference
between the pristine and modified mechanical properties of the model’s material cause a greater
variation in the power spectrum of the structure, which is also reflected in the power and is wrongly
seen as damage-induced. This aspect is not helpful in the damage detection process; the TKE operator
seems to be more resilient to these slight changes. By summing up all of the damage scenarios of this
first case, it turns out that of all features, the TECCs are the only one always correctly labelling the
cases with realistic damage patterns (05–14) and the ones with global, small fluctuations of E, unrelated
to damage (15–18). Indeed, the TECCs fall short of a perfect score for only three mislabeled realisations
(out of 10) in case 01, which was also the most demanding scenario due to the proximity of the damage
to the fixed end.
(a)
(b)
Figure 9. Damage indexes for TECCs (a) and MFCCs (b); first numerical case study (Fossano bell
tower), scenarios 00–04 (story-level damage).
181
Appl. Sci. 2019, 9, 5064
(a)
(b)
tower), scenarios 05–14 (realistic damage).
(a)
Figure 11. Cont.
182
Appl. Sci. 2019, 9, 5064
(b)
tower), scenarios 15–18 (no damage).
Table 7. Results for the first numerical case study (bell tower).
Case 00 Cases 01–04 Cases 05–14 Cases 15–18

Type 1 Errors [%] Type 2 Errors [%] Type 2 Errors [%] Type 1 Errors [%]
MFCCs 60.00 0.00 0.00 45.00
fcut−off = 14 fs 30.00 0.00 0.00 45.00
fcut−off = 18 fs 30.00 0.00 0.00 45.00
Log-Scale 20.00 0.00 0.00 65.00
Linear 60.00 0.00 0.00 45.00
ECO-MMFCCs 20.00 0.00 0.00 62.50
MO-MMFCCs 10.00 0.00 0.00 62.50
LO-MMFCCs 20.00 0.00 0.00 65.00
BFCCs 60.00 0.00 0.00 47.50
GTCCs 20.00 0.00 0.00 60.00
TECCs 0.00 7.50 0.00 0.00
Table 8 reports the results for the second case study (the highly flexible prototype wing); Figure 12
graphically illustrates them. It can be inferred that, in this case, the exact detection of damage was
more difficult. MFCCs and similar 1variants proved to be able to spot everywhere the damage but at
𝑓 = 𝑓
the cost of not actually discerning it4from the two cases with small perturbations. The TECCs, on the
1
other hand, did not perform𝑓 perfectly,
= 𝑓yet they avoided all but one false alarm and were able to discern
8
the damaged conditions decently in the inspected scenarios. If one does not take into account Case 07,
which presents an imperceptible decrease of 0.03 Hz and went completely unnoticed by the trained
algorithm, the type 2 error percentage drops to 7.50%, with two errors in Case 04, no one in Case 05,
and a single mislabeled element in Case 06.
183
Appl. Sci. 2019, 9, 5064
Table 8. Results for the second numerical case study (HAR wing).
Cases 00–02 Cases 03–07

Type 1 Errors [%] Type 2 Errors [%]
MFCCs 66.67 0.00
fcut−off = 41 fs 66.67 0.00
fcut−off = 81 fs 66.67 0.00
Log-Scale 70.00 0.00
Linear 66.67 0.00
ECO-MMFCCs 66.67 0.00
MO-MMFCCs 66.67 0.00
LO-MMFCCs 66.67 0.00
BFCCs 66.67 0.00
GTCCs 66.67 0.00
TECCs 3.33 26.00 (*)
(*) 7.50% on Cases from 03 to 06.
(a)
(b)
Figure 12. Damage indexes for TECCs (a) and MFCCs (b); second numerical case study (highly flexible
prototype wing).
As in the previous case of the Fossano bell tower, this seems to point out the larger robustness of
the TECCs feature to small changes, assumed here as statistical variations of the structural parameters
and not directly linked to damage occurrence. A possible explanation comes from the TKE Operator
being notably sensitive to frequency and amplitude changes, while much more robust than MFCCs to
noise and minimal variations in the field of speech recognition accuracy.
184
1
Appl. Sci. 2019, 9, 5064
6.2. Results from the Experimental Data

The results from experimental data are displayed here in Figure 13. As before, TECCs and
classic MFCCs are only depicted due to space constraints. Table 9 summarises the results, in terms of
percentage of type 1 and 2 errors.
(a)
(b)
Figure 13. Damage indexes for TECCs (a) and MFCCs (b); and, experimental data.
Table 9. Results for the experimental case study.
Cases 1–9 Cases 10–17

Type 1 Errors [%] Type 2 Errors [%]
MFCCs 3.33 0.00
fcut−off = 41 fs 5.56 0.00
fcut−off = 81 fs 7.78 1.25
Log-Scale 4.44 1.25
Linear 5.56 0.00
ECO-MMFCCs 4.44 0.00
MO-MMFCCs 2.22 22.50
LO-MMFCCs 12.22 25.00
BFCCs 3.33 0.00
GTCCs 4.44 0.00
TECCs 0.00 13.75
185
Appl. Sci. 2019, 9, 5064
In interpreting the data, one must consider the different definition of ‘damage’ between the two
numerical and the experimental case studies. The Fossano bell tower and the HAR prototype wing
behave linearly in both damaged and pristine conditions; again, the difference between the two is only
a matter of shifting natural frequencies. The experimental data have a completely different ‘meaning’ of
damage, as here, differences in mass and stiffness are accounted as different operational conditions and
the distinction between damaged and undamaged response lies solely on the presence of nonlinearities.
In this latter case, the TECCs proved to be the only feature able to reach a 0% type 1 error in the test set,
but only in spite of a relatively high type 2 error percentage, only surpassed by the recently proposed
MO- and LO-MMFCCs. Nevertheless, a closer inspection of the several damaged and undamaged
configurations allows for highlighting that this error is clustered in only three scenarios, respectively,
cases 10, 15, and 16. Case 10 is the one with the slightest nonlinearity and, thus, is the most challenging
to detect. The algorithm also struggled for cases 15 and 16, where changes in the frame’s mass are
involved. On the other hand, all of the undamaged scenarios are recognised as such by means of the
TECCs, while classic MFCCs generate some misclassifications for case 5 (which is the one with the
largest stiffness reduction close to the base of the frame, where such structural changes have the most
impact in the vibrational response).
Therefore, experimental results seem to validate what is expected from numerical simulations,
that is to say: (i) a greater generalisation capability of the TECCs feature over closely-related alternatives;
(ii) a strong robustness to false alarms but at the cost of some false negatives; and, (iii) a sensibility
to shifts in the frequency content higher or at least comparable to MFCCs in its original form and
proposed variants. On the other hand, the TECCs proved to be less sensitive to the presence of limited
nonlinearities during changing operational conditions than to mass changes respect to the MFCCs and
similar features; but this outcome might be specific of the experimental data utilised here.
6.2.1. Further Analysis of the Experimental Data

Some further tests were performed to better define the advantages of the proposed TECCs feature.
For brevity sake, only the results for experimental data are reported; the same findings were observed
for numerical data.
Firstly, the number and distribution of output channels were investigated. Four setups have been
considered: setup #0, including all channels (numbered 2 to 5, see Figure 8), for which the results are
already reported in Figure 13 and Table 9; setup #1 (channels 3, 4, and 5); setup #2 (channels 3 and
4); and finally, setup #3 (constituted by channels 4 and 5). Table 10 enlists the results for these last
three setups. Remarkably, this test evidenced what was already observed in [21], that is to say, MFCCs
and similar features produce less accurate results when all of the acquisition channels are considered.
This point was confirmed here for type 1 errors in two out of the three new setups considered. The same
effects were encountered here for TECCs as well, with the same conditions: MFCCs systematically fail
in labelling case 5 as undamaged, due to its relatively large shift in frequency, while TECCs struggle in
classifying cases 10, 15, and 16 as damaged due to their lower level of nonlinear distortion.
Secondly, a comparison with a non-cepstral feature, the aforementioned AR coefficients, was also
performed and reported in Table 10. The AIC approach was applied similarly to what done in [21].
An order of 12 was used for sensor setup #1 and #2; order 10 was applied instead for the sensor setup
#3. It can be seen that the percentage of type 1 error is generally higher than most of the cepstral
features that are proposed here.
186
Appl. Sci. 2019, 9, 5064
Table 10. Results according to the sensor setup considered.
Sensor Setup #1 Sensor Setup #2 Sensor Setup #3

Cases 1–9 Cases 10–17 Cases 1–9 Cases 10–17 Cases 1–9 Cases 10–17
Type 1 Type 2 Type 1 Type 2 Type 1 Type 2
Errors [%] Errors [%] Errors [%] Errors [%] Errors [%] Errors [%]
MFCCs 4.44 0.00 2.22 1.25 2.22 0.00
fcut−off = 41 fs 3.33 0.00 1.11 0.00 1.11 0.00
fcut−off = 81 fs 5.56 0.00 3.33 1.25 3.33 0.00
Log-Scale 5.56 1.25 3.33 2.50 1.11 1.25
Linear 3.33 0.00 1.11 1.25 1.11 0.00
ECO-MMFCCs 3.33 0.00 1.11 0.00 1.11 0.00
MO-MMFCCs 2.22 25.00 3.33 30.00 1.11 27.50
LO-MMFCCs 10.00 25.00 0.00 23.75 2.22 25.00
BFCCs 1.11 0.00 0.00 2.50 1.11 0.00
GTCCs 2.22 0.00 2.22 1.25 2.22 1.25
TECCs 0.00 7.50 0.00 11.25 0.00 6.25
AR coefficients 13.33 0.00 12.22 0.00 6.67 0.00
The Central Processing Unit (CPU) time that was required for the training phase was then
evaluated for all types of DSFs on non-optimised (yet comparable) versions of the code. This was
tested by means of the Matlab® stopwatch timer tic–toc; Table 11 reports the results. One can clearly
see that the computational time is comparable over the several sensor setups considered. Note that
the timing accounts for training operations (feature extraction and population statistics estimation)
and for threshold definition; the time required for data preprocessing, differently from what done
in [21], was not considered, as it is identical for all features. This returns some faster results with
respect to what is reported there. Moreover, the codes were run 10 times to avoid any computational
variability; the average result is reported. As expected, the difference between MFCCs, similar variants,
GTCCs and TECCs are minimal. It can be observed that GTCCs takes slightly longer than MFCCs
(in the order of some fractions of a second), since there it is needed to build up the more complexly
shaped gammatone filterbank. The TECCs require that additional time plus a bit more to compute the
TKE operator out of the provided time arrays. In the case with all acquisition channels considered,
these two operations make the code run in almost 30% more time (0.43 s on average). However, this
delay is basically irrelevant if compared to the one that is needed to extract other features, such as the
Auto-Regressive (AR) model coefficients [21].
Lastly, the standard deviation is proposed as a metric of dispersion for the results of the several
realisations that belong to the same damage case, to quantitatively express the robustness of the
classification. Table 12 enlists the results for all 17 cases. Only the values for the sensor setup #0
and for MFCCs, log-scale CCs, linear CCs, BFCCs, GTCCs, and TECCs are reported here due to
space constraints. The same behaviour was observed for the other setups as well. By considering ten
realisations per investigated case, it can be observed that the use of TECCs as a DSF produce a much
smaller scattering of the results, with the standard deviation never exceeding σ = 6, while the same
measure goes well over 50 or even 70 for MFCCs in some cases.
187
Appl. Sci. 2019, 9, 5064
Table 11. CPU time accordingly to the sensor setup considered (training phase).
Sensor Setup #0 Sensor Setup #1 Sensor Setup #2 Sensor Setup #3

CPU Time [s] CPU Time [s] CPU Time [s] CPU Time [s]
MFCCs 1.60 1.31 1.00 1.03
fcut−off = 14 fs 1.63 1.31 1.01 1.03
fcut−off = 18 fs 1.62 1.35 1.04 1.04
Log-Scale 1.62 1.37 1.01 1.02
Linear 1.56 1.30 0.99 0.99
ECO-MMFCCs 1.56 1.31 1.00 1.02
MO-MMFCCs 1.58 1.29 1.00 0.99
LO-MMFCCs 1.59 1.29 1.01 0.99
BFCCs 1.58 1.32 1.02 1.00
GTCCs 1.78 1.38 1.06 1.07
TECCs 2.03 1.62 1.15 1.16
Table 12. Standard deviation of Damage Index (DI) (sensor setup #0).
σ(DI(n))-Cases 1 to 17
MFCCs 0.32, 0.20, 0.28, 0.19, 0.28, 0.32, 0.26, 0.21, 0.22, 6.21, 41.74, 62.64, 52.52, 72.21, 6.16, 8.63, 81.06.
Log-Scale 0.24, 0.29, 0.15, 0.22, 0.28, 0.22, 0.27, 0.22, 0.23, 3.29, 22.23, 33.42, 31.63, 32.26, 3.45, 5.39, 35.00.
Linear 0.29, 0.30, 0.22, 0.19, 0.27, 0.22, 0.24, 0.31, 0.24, 6.55, 39.27, 67.89, 46.22, 52.90, 5.50, 7.85, 72.59.
BFCCs 0.31, 0.20, 0.26, 0.19, 0.26, 0.35, 0.22, 0.21, 0.19, 6.73, 44.16, 68.87, 47.46, 64.38, 6.72, 9.36, 90.09.
GTCCs 0.20, 0.28, 0.15, 0.16, 0.20, 0.31, 0.32, 0.19, 0.24, 4.19, 25.88, 45.77, 51.07, 73.55, 3.80, 6.54, 41.64.
TECCs 0.18, 0.14, 0.17, 0.23, 0.14, 0.17, 0.13, 0.26, 0.22, 0.75, 4.78, 5.57, 4.25, 4.64, 0.84, 1.70, 4.35.
7. Conclusions
Any Structural Health Monitoring system relies on data, almost always pre-processed, and on
features extrapolated from them. Mel-Frequency Cepstral Coefficients (MFCCs) have recently been
proven to be effective in damage detection, relying on the cepstrum of the recorded structural response,
even if margins for improvements were evident. The Teager-Kaiser Energy (TKE) operator has been
proposed here as the basis for a similar feature, less subject to false positive mislabeling when used for
Machine Learning. That is adherent to what is encountered for speech signals and well-known in the
field of speaker and speech recognition. The investigation reported here spans over different structures,
with different input and very different setups—varying acquisition parameters, such as the number
of the output channels, the sampling frequency, and so on. Moreover, in the first and in the second
numerical study, the damage was modelled as a linear reduction of stiffness inserted in a linear system,
while the experimental case emulated the damage as a pointwise source of nonlinearity in an otherwise
linear system and stiffness local reduction was intended as a change in the operational conditions that
were unrelated to damage. This shows the excellent capability of generalisation of this approach.
The proposed damage sensitive feature, the Teager-Kaiser Energy Cepstral Coefficients (TECCs),
has been benchmarked against the MFCCs and some similar variants. Interesting numerical and
experimental results were achieved for both the linear and nonlinear models of damage.
The main conclusions are the following:
1. the TECCs, MFCCs, and similar features perform efficiently both with damage emulated by
a reduction of the Young’s modulus E or by the presence of nonlinearities, even if the signal
is noisy;
2. the Mel-Scale performed similarly to other nonlinear scales such as the Bark Scale or logarithmic
spacing, with no clear best option among all the investigated case studies;
3. the algorithms that resort to a Gammatone filterbank generally produced better results than the
ones with triangular filters; even if the TECCs still outperforms the GTCCs;
4. the TECCs present a strong reduction of the order of magnitude of the Damage Index (DI) with
respect to MFCCs and similar options, in absolute terms; this does not affect the damage detection
188
Appl. Sci. 2019, 9, 5064
algorithm, as the relative distance between the ‘normality’ model and the damaged cases remains
and seems even to increase;
5. the Teager-Kaiser Energy Cepstral Coefficients outperforms all the competitors in terms of
little or no type 1 errors, but it is slightly more prone to type 2 errors respect to MFCCs and
derived features;
6. with respect to non-cepstral features, such as the AR coefficients, the main benefits lie in the
less computational cost and greater robustness to noise and to confounding influences, such as
environmental and operational effects, unrelated to damage.
The outcomes of this research leave plenty of room for improvements. Indeed, apart from
the TECCs and other cepstral-based alternatives, wavelet-based alternatives are of great interest.
The Mel-Frequency Discrete Wavelet Coefficients may be a relevant alternative. Thanks to the Discrete
Wavelet Transform, the filter spacing issue is by-passed; filter shaping is reduced to the selection of the
mother wavelet. The Authors are committed to pursuing further studies in this direction.
Author Contributions: Conceptualization, M.C., M.F., R.C., C.S. and R.B.; Data curation, M.C. and M.F.; Formal
analysis, M.C. and M.F.; Funding acquisition, R.C. and C.S.; Investigation, M.C. and M.F.; Methodology, M.C.,
R.C., C.S. and R.B.; Project administration, R.C., C.S. and R.B.; Resources, R.C., C.S. and R.B.; Software, M.C., M.F.
and R.B.; Supervision, R.C., C.S. and R.B.; Validation, M.C. and M.F.; Visualization, M.C.; Writing—original draft,
M.C. and M.F.; Writing—review & editing, M.C., R.C., C.S. and R.B.
Funding: This research was funded by the INTE project by Compagnia di San Paolo (2017-18) “System Identification,
model updating and damage assessment of heritage buildings and structures”, as a joint partnership of Politecnico
di Torino (IT) and Columbia University (USA).
Acknowledgments: The Authors wish to thank Luciana Balsamo and Marica Leonarda Pecorelli for their precious
help and advice. The Authors also wish to acknowledge the Engineering Institute at Los Alamos National
Laboratory for making available to the public domain the experimental data used in this work.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the
study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to
publish the results.
References
1. Farrar, C.R.; Worden, K. Structural Health Monitoring: A Machine Learning Perspective; John Wiley & Sons:
Hoboken, NJ, USA, 2013.
2. Doebling, S.W.; Farrar, C.R.; Prime, M.B. A Summary Review of Vibration-Based Damage Identification
Methods. Shock Vib. Dig. 1998, 30, 91–105. [CrossRef]
3. Figueiredo, E.; Park, G.; Farrar, C.R.; Worden, K.; Figueiras, J. Machine learning algorithms for damage
detection under operational and environmental variability. Struct. Health Monit. 2011, 10, 559–572. [CrossRef]
4. Sohn, H.; Farrar, C.R.; Hunter, N.F.; Worden, K. Structural Health Monitoring Using Statistical Pattern
Recognition Techniques. J. Dyn. Syst. Meas. Control 2001, 123, 706–711. [CrossRef]
5. Amezquita-Sanchez, J.P.; Adeli, H. Signal Processing Techniques for Vibration-Based Health Monitoring of
Smart Structures. Arch. Comput. Methods Eng. 2016, 23, 1–15. [CrossRef]
6. Bonato, P.; Ceravolo, R.; Stefano, A. De Time-Frequency and Ambiguity Function Approaches in Structural
Identification. J. Eng. Mech. 1997, 123, 1260–1267. [CrossRef]
7. Bonato, P.; Ceravolo, R.; De Stefano, A.; Molinari, F. Cross-Time Frequency Techniques for the Identification
of Masonry Buildings. Mech. Syst. Signal Process. 2000, 14, 91–109. [CrossRef]
8. Law, S.S.; Zhu, X.Q.; Tian, Y.J.; Li, X.Y.; Wu, S.Q. Statistical damage classification method based on wavelet
packet analysis. Struct. Eng. Mech. 2013, 46, 459–486. [CrossRef]
9. Liu, J.; Wang, Z.; Liu, J.-L.; Wang, Z.-C.; Ren, W.-X.; Li, X.-X. Structural time-varying damage detection using
synchrosqueezing wavelet transform. Smart Struct. Syst. 2015, 15, 1738–1991. [CrossRef]
10. Wang, C.; Ren, W.X.; Wang, Z.C.; Zhu, H.P. Time-varying physical parameter identification of shear type
structures based on discrete wavelet transform. Smart Struct. Syst. 2014, 14, 831–845. [CrossRef]
11. Pnevmatikos, N.; Blachowski, B.D.; Hatzigeorgiou, G.; Pnevmatikos, N.G.; Blachowski, B.;
Hatzigeorgiou, G.D.; Swiercz, A. Wavelet analysis based damage localization in steel frames with bolted
connections. Smart Struct. Syst. 2016, 18, 1738–1991. [CrossRef]
189
Appl. Sci. 2019, 9, 5064
12. Civera, M.; Zanotti Fragonara, L.; Surace, C. Nonlinear Dynamics of Cracked, Cantilevered Beam-like
Structures Undergoing Large Deflections. In Proceedings of the 2019 IEEE 5th International Workshop on
Metrology for AeroSpace (MetroAeroSpace), Torino, Italy, 19–21 June 2019.
13. Pines, D.; Salvino, L. Structural health monitoring using empirical mode decomposition and the Hilbert
phase. J. Sound Vib. 2006, 294, 97–124. [CrossRef]
14. Civera, M.; Filosi, C.M.; Pugno, N.M.; Silvestrini, M.; Surace, C.; Worden, K. Assessment of vocal cord
nodules: A case study in speech processing by using Hilbert-Huang Transform. J. Phys. Conf. Ser. 2017, 842,
012025. [CrossRef]
15. Childers, D.G.; Skinner, D.P.; Kemerait, R.C. The cepstrum: A guide to processing. Proc. IEEE 1977, 65,
1428–1443. [CrossRef]
16. Zheng, H.; Mita, A. Damage indicator defined as the distance between ARMA models for structural health
monitoring. Struct. Control Health Monit. 2008, 15, 992–1005. [CrossRef]
17. Kim, J.T.; Lyon, R.H. Cepstral analysis as a tool for robust processing, deverberation and detection of
transients. Mech. Syst. Signal Process. 1992, 6, 1–15. [CrossRef]
18. Peeters, C.; Guillaume, P.; Helsen, J. A comparison of cepstral editing methods as signal pre-processing
techniques for vibration-based bearing fault detection. Mech. Syst. Signal Process. 2017, 91, 354–381.
[CrossRef]
19. Randall, R.B. A History of Cepstrum Analysis and its Application to Mechanical Problems. Mech. Syst.
Signal Process. 2017, 97, 3–19. [CrossRef]
20. Balsamo, L.; Betti, R.; Beigi, H. Structural Damage Detection Using Speaker Recognition Techniques.
In Proceedings of the 11th International Conference on Structureal Safety and Reliability (ICOSSAR),
New York, NY, USA, 16–20 June 2013.
21. Balsamo, L.; Betti, R.; Beigi, H. A structural health monitoring strategy using cepstral features. J. Sound Vib.
2014, 333, 4526–4542. [CrossRef]
22. Ferraris, M.; Civera, M.; Ceravolo, R.; Surace, C.; Betti, R. Using Enhanced Cepstral Analysis for Structural
Health Monitoring; Springer: Singapore, 2020; pp. 150–165.
23. Dimitriadis, D.; Maragos, P.; Potamianos, A. Auditory Teager Energy Cepstrum Coefficients for Robust
Speech Recognition. In Proceedings of the 9th European Conference on Speech Communication and
Technology, Lisbon, Portugal, 4–8 September 2005.
24. Kaiser, J.F. On a simple algorithm to calculate the “energy” of a signal. In Proceedings of the International
Conference on Acoustics, Speech, and Signal Processing, Albuquerque, NM, USA, 3–6 April 1990; pp. 381–384.
25. Schluter, R.; Bezrukov, I.; Wagner, H.; Ney, H. Gammatone Features and Feature Combination for Large
Vocabulary Speech Recognition. In Proceedings of the 2007 IEEE International Conference on Acoustics,
Speech and Signal Processing—ICASSP ’07, Honolulu, HI, USA, 15–20 April 2007; pp. 649–652.
26. Vakman, D. On the analytic signal, the Teager-Kaiser energy algorithm, and other methods for defining
amplitude and frequency. IEEE Trans. Signal Process. 1996, 44, 791–797. [CrossRef]
27. Tran, V.T.; AlThobiani, F.; Ball, A. An approach to fault diagnosis of reciprocating compressor valves using
Teager—Kaiser energy operator and deep belief networks. Expert Syst. Appl. 2014, 41, 4113–4122. [CrossRef]
28. Henríquez Rodríguez, P.; Alonso, J.B.; Ferrer, M.A.; Travieso, C.M. Application of the Teager-Kaiser energy
operator in bearing fault diagnosis. ISA Trans. 2013, 52, 278–284. [CrossRef]
29. Ceravolo, R.; Pistone, G.; Fragonara, L.Z.; Massetto, S.; Abbiati, G. Vibration-Based Monitoring and Diagnosis
of Cultural Heritage: A Methodological Discussion in Three Examples. Int. J. Archit. Herit. 2016, 10, 375–395.
[CrossRef]
30. Carpinteri, A.; Lacidogna, G. Damage monitoring of an historical masonry building by the acoustic emission
technique. Mater. Struct. 2006, 39, 161–167. [CrossRef]
31. Carpinteri, A.; Lacidogna, G. Structural Monitoring and Integrity Assessment of Medieval Towers.
J. Struct. Eng. 2006, 132, 1681–1690. [CrossRef]
32. Carpinteri, A.; Invernizzi, S.; Lacidogna, G. In situ damage assessment and nonlinear modelling of a historical
masonry tower. Eng. Struct. 2005, 27, 387–395. [CrossRef]
33. Pontillo, A.; Hayes, D.; Dussart, G.X.; Lopez Matos, G.E.; Carrizales, M.A.; Yusuf, S.Y.; Lone, M.M. Flexible
High Aspect Ratio Wing: Low Cost Experimental Model and Computational Framework. In Proceedings of
the 2018 AIAA Atmospheric Flight Mechanics Conference, Reston, VA, USA, 8–12 January 2018.
190
Appl. Sci. 2019, 9, 5064
34. Lanzoni, L.; Tarantino, A.M. Equilibrium configurations and stability of a damaged body under uniaxial
tractions. Z. Angew. Math. Phys. 2014, 66, 171–190. [CrossRef]
35. Tarantino, A.M. Equilibrium paths of a hyperelastic body under progressive damage. J. Elast. 2014, 114,
36. Bedon, C. Diagnostic analysis and dynamic identification of a glass suspension footbridge via on-site
vibration experiments and FE numerical modelling. Compos. Struct. 2019, 216, 366–378. [CrossRef]
37. Bogert, B.P. The quefrency alanysis of time series for echoes; Cepstrum, pseudo-autocovariance,
cross-cepstrum and saphe cracking. Time Ser. Anal. 1963, 09–243.
38. Stevens, S.S.; Volkmann, J.; Newman, E.B. A Scale for the Measurement of the Psychological Magnitude
Pitch. J. Acoust. Soc. Am. 1937, 8, 185–190. [CrossRef]
39. Satyam, M.; Sudhakara, R.; Devy, C.G. Cepstrum Analysis: An Advanced Technique in Vibration Analysis of
Defects in Rotating Machinery. Def. Sci. J. 1994, 44, 53–60. [CrossRef]
40. Noll, A.M. Cepstrum Pitch Determination. J. Acoust. Soc. Am. 1967, 41, 293–309. [CrossRef] [PubMed]
41. De Krom, G. A Cepstrum-Based Technique for Determining a Harmonics-to-Noise Ratio in Speech Signals.
J. Speech Lang. Hear. Res. 1993, 36, 254–266. [CrossRef] [PubMed]
42. Oppenheim, A.V.; Schafer, R.W.; Buck, J.R. Discrete-Time Signal Processing; Prentice Hall: Upper Saddle River,
NJ, USA, 1999; ISBN 0130834432.
43. Beigi, H. Fundamentals of Speaker Recognition; Springer: Berlin/Heidelberg, Germany, 2016; ISBN 9781489979223.
44. Mermelstein, P. Distance measures for speech recognition, psychological and instrumental. Pattern Recognit.
Artif. Intell. 1976, 116, 374–388.
45. Bridle, J.S.; Brown, M.D.; Chamberlain, R.M. An experimental automatic word recognition system; Technical
Report JSRU No. 1003; Joint Speech Research Unit: Ruislip, England, 1974.
46. Stevens, S.S.; Volkmann, J. The Relation of Pitch to Frequency: A Revised Scale. Am. J. Psychol. 1940, 53,
47. Hermansky, H.; Morgan, N.; Hirsch, H.-G. Recognition of speech in additive and convolutional noise based
on RASTA spectral processing. In Proceedings of the IEEE International Conference on Acoustics Speech
and Signal Processing, Minneapolis, MN, USA, 27–30 April 1993; Volume 2, pp. 83–86.
48. Fant, G. Acoustic Theory of Speech Production: With Calculations Based on X-ray Studies of Russian Articulations;
Mouton: The Hague, Netherlands, 1970; ISBN 9027916004.
49. Umesh, S.; Cohen, L.; Nelson, D. Fitting the Mel Scale. In Proceedings of the 1999 IEEE International
Conference on Acoustics, Speech, and Signal Processing, Phoenix, AZ, USA, 15–19 March 1999.
50. Zwicker, E. Subdivision of the Audible Frequency Range into Critical Bands (Frequenzgruppen). J. Acoust.
Soc. Am. 1961, 33, 248. [CrossRef]
51. Irino, T.; Patterson, R.D. A time-domain, level-dependent auditory filter: The gammachirp. J. Acoust. Soc. Am.
1997, 101, 412–419. [CrossRef]
52. Traunmüller, H. Analytical expressions for the tonotopic sensory scale. J. Acoust. Soc. Am. 1990, 88, 97–100.
[CrossRef]
53. Moore, B.C.J.; Glasberg, B.R. Suggested formulae for calculating auditory-filter bandwidths and excitation
patterns. J. Acoust. Soc. Am. 1983, 74, 750–753. [CrossRef]
54. Herrera, A.; Del Rio, F. Frequency bark cepstral coefficients extraction for speech analysis by synthesis.
J. Acoust. Soc. Am. 2010, 128, 2290. [CrossRef]
55. Slaney, M. An Efficient Implementation of the Patterson-Holdsworth Auditory Filter Bank; Technical Report
Apple Computer No. 35, Perception Group—Advanced Technology Group, Cupertino, California (USA).
1993. Available online: https://engineering.purdue.edu/~{}malcolm/apple/tr35/PattersonsEar.pdf (accessed
on 21 November 2019).
56. Teager, H.M.; Teager, S.M. Evidence for Nonlinear Sound Production Mechanisms in the Vocal Tract. In Speech
Production and Speech Modelling; Springer: Dordrecht, The Netherlands, 1990; pp. 241–261.
57. Maragos, P.; Kaiser, J.F.; Quatieri, T.F. On amplitude and frequency demodulation using energy operators.
IEEE Trans. Signal Process. 1993, 41, 1532–1550. [CrossRef]
58. Skowronski, M.D.; Harris, J.G. Increased mfcc filter bandwidth for noise-robust phoneme recognition.
In Proceedings of the IEEE International Conference on Acoustics Speech and Signal Processing,
Orlando, FL, USA, 13–17 May 2002; pp. 801–804.
191
Appl. Sci. 2019, 9, 5064
59. Figueiredo, E.; Park, G.; Figueiras, J. Structural Health Monitoring Algorithm Comparisons Using Standard Data
Sets; Los Alamos National Lab.(LANL): Los Alamos, NM, USA, 2009.
60. Ververidis, D.; Kotropoulos, C. Information Loss of the Mahalanobis Distance in High Dimensions:
Application to Feature Selection. IEEE Trans. Pattern Anal. Mach. Intell. 2009, 31, 2275–2281. [CrossRef]
[PubMed]
61. Civera, M.; Zanotti Fragonara, L.; Surace, C. Video Processing Techniques for the Contactless Investigation
of Large Oscillations. J. Phys. Conf. Ser. 2019, 1249, 012004. [CrossRef]
62. Civera, M.; Zanotti Fragonara, L.; Surace, C.; Civera, M.; Zanotti Fragonara, L.; Surace, C. Using Video
Processing for the Full-Field Identification of Backbone Curves in Case of Large Vibrations. Sensors 2019, 19,
2345. [CrossRef] [PubMed]
63. Bovsunovsky, A.; Surace, C. Non-linearities in the vibrations of elastic structures with a closing crack: A state
of the art review. Mech. Syst. Signal Process. 2015, 62–63, 129–148. [CrossRef]
192
applied
sciences
Article
The Rapid Detection Technology of Lamb Wave for
Microcracks in Thin-Walled Tubes
Shunmin Yang , Mingquan Wang * and Lu Yang
Science and Technology on Electronic Test and Measurement Laboratory, North University of China,
Taiyuan 030051, China
* Correspondence: [email protected]; Tel.: +86-139-3454-8995 or +86-0351-355-7406

Received: 25 July 2019; Accepted: 27 August 2019; Published: 1 September 2019
Abstract: Thin-walled tubes are a kind of pressure vessel formed by a stamping and drawing process,
which must withstand a great deal of sudden pressure during use. When microcrack defects of a
certain depth are present on its inner and outer surfaces, severe safety accidents may occur, such as
cracking and crushing. Therefore, it is necessary to carry out nondestructive testing of thin-walled
tubes in the production process to eliminate the potential safety hazards. To realize the rapid detection
of microcracks in thin-walled tubes, this study could be summarized as follows: (i) Because the
diameters of the thin-walled tubes were much larger than their thicknesses, Lamb wave characteristics
of plates with equal thicknesses were used to approximate the dispersion characteristics of thin-walled
tubes. (ii) To study the dispersion characteristics of Lamb waves in thin plates, the detection method
of the A0 mode was determined using the particle displacement–amplitude curve. (iii) Using a
multi-channel parallel detection method, rapid detection equipment for Lamb wave microcracks
in thin-walled tubes was developed. (iv) The filtering peak values for defect signal detection
with different depths showed that the defect detection peak values could reflect the defect depth
information. (v) According to the minimum defect standard of a 0.045-mm depth, 100,000 thin-walled
tubes were tested. The results showed that the missed detection rate was 0%, the reject rate was
0.3%, and the detection speed was 5.8 s/piece, which fully meets the actual detection requirements of
production lines. Therefore, this study not only solved the practical issues for the rapid detection
of microcracks in thin-walled tubes but also provided a reference for the application of ultrasonic
technology for the detection of other components.
Keywords: Lamb wave; microcracks testing; multi-channel detection; signal-to-noise ratio;

thin-walled tube
1. Introduction
1.1. Purpose and Significance

Thin-walled tubes are pressure vessels formed by stamping and drawing. In the production
process, due to errors in the processing parameters and faults in the production equipment, microcrack
defects form in the inner and outer walls of thin-walled tubes. However, thin-walled tubes must
withstand large sudden changes in pressure during use. When the depths of the microcracks on the
inner and outer surfaces are greater than a certain threshold, severe safety accidents can occur, such as
cracking and crushing. Therefore, it is necessary to carry out nondestructive testing of thin-walled
tubes in the production process to eliminate the potential safety hazards.
For the nondestructive testing of thin-walled tubes, the commonly used testing methods include
eddy current [1,2], magnetic powder [3], X-ray [4,5], liquid penetrant [6–8], and ultrasonic [9] methods.
With eddy current detection, it is difficult to distinguish irregular surface defects and detect them
193
Appl. Sci. 2019, 9, 3576
quantitatively, which can easily lead to misjudgment [10]. Magnetic particle testing is more suitable for
surface defect detection, and it is difficult to quantitatively detect the depth of a flaw [10]. The X-ray
method is mainly used to detect the volume of the defect inside the object. For an area defect, there
will be an overlap of the front and back, and the angle between the ray beam and the orientation of the
crack cannot exceed 10◦ ; otherwise, it cannot be detected [10]. Penetration detection is limited to the
defects of the surface opening. The penetrating liquid pollutes the parts and the environment and is
not suitable for on-line detection [10]. Ultrasonic testing is sensitive to the area of the defect inside the
object and can locate the relative size and location of the defect. The detection depth is much greater
than other detection methods, and thus, it has been widely used [10]. For thin-walled tubes, ultrasonic
Lamb waves are usually used for testing [11].
The ultrasonic detection principle of thin-walled tubes is the same as that of pipe, and there are
many studies on the principle, method, and application of Lamb wave detection of pipe for reference.
Therefore, based on the analysis of Lamb wave detection of pipe, the ultrasonic detection method
of thin-walled tubes was explored. Defect types in pipe include lamination [12], corrosion [13], and
cracking [14–16], which can occur on both inner and outer surfaces. The modes of Lamb waves in the
pipe are mainly divided into longitudinal [17,18], torsional [19–23], and flexural [24,25].
According to the relevant references, the following work must be completed to realize the Lamb
wave detection of pipe defects. (i) Based on analyzing and studying the dispersion characteristics
of Lamb waves, the transducer frequency and Lamb wave mode with the highest sensitivity to the
detected defect type should be selected. (ii) To study the interaction between defect and Lamb waves in
the pipe, simulation should be carried out under a three-dimensional (3D) model. (iii) A test platform
should be built to compare the difference between simulation and test results by scanning artificial
defects in the pipe with transducers. (iv) Signal processing is used to determine the location and size
of defects in the pipe. The difficulties in the detection process mainly focus on the two aspects: mode
selection and 3D model simulation. On the one hand, the dispersion characteristics of Lamb waves in
pipe have three modes to be analyzed and studied. Compared with only symmetric and antisymmetric
modes in a thin plate, the difficulty of modal analysis and selection is increased. On the other hand, a
3D model is needed to analyze the interaction between defects and Lamb waves in the pipe. Compared
with the 2D model of a thin plate, there are great challenges in terms of computational complexity.
When the diameters of hollow cylindrical components are much larger than their thicknesses,
the dispersion characteristics of Lamb waves in hollow cylindrical components can be approximated
by studying the characteristics of Lamb waves in plates with equal thicknesses [26–28]. He et al. [29]
found that when the diameter to thickness ratio of hollow cylindrical components is less than 0.05,
the dispersion curves of the hollow cylindrical components were similar to those of components with
the corresponding plate thicknesses. In this study, the diameters and thicknesses of the thin-walled
tubes were 35 mm and 0.8–1.2 mm, respectively. Selecting the Lamb-wave dispersion characteristics
of plates with equal thicknesses to approximate the propagation characteristics of Lamb waves in
thin-walled tubes could not only simplify the model but also facilitate the application of practical
engineering practice.
1.2. The State of the Art
1.2.1. Transducers for Lamb Waves

Qiu et al. [30] proposed a multi-physical simulation method for the Lamb wave propagation
of piezoelectric transducers under load conditions. The results showed that the phase velocity and
amplitude changes of Lamb waves obtained by simulation were in good agreement with experiment.
He et al. [31] designed two electromagnetic acoustic transducers to generate and receive S0 Lamb
waves in ferromagnetic steel plates, which were kept within 5 and 19 mm from the sample surface,
respectively. This allowed the electromagnetic acoustic transducer to interact with the sample with
less force and facilitated easier scanning. Zhang et al. [32] established the finite element model and
194
Appl. Sci. 2019, 9, 3576
detection system of laser ultrasound and realized the simulation and test detection of multilayer
bonded composite materials. The results showed that this method not only facilitated the visualization
of debonding but also provided accurate bonding quality information.
1.2.2. Lamb Mode

Gupta et al. [33] studied the interaction of a basic symmetric Lamb wave mode S0 with laminations
in eight layers of glass-fiber-reinforced polymer (GFRP) laminated composite by numerical simulation
and experimental verification. The results showed that the sensitivity of S0 mode reflection depended
on the thickness direction and layering direction of the composite laminate. Among all laminates
tested, the S0 mode was highly sensitive to DL23 and DL34 (Delamination is considered between the
layers 2&3 and 3&4 locations, which are abbreviated as DL12 and DL34, respectively. The thickness is
0.42–1.26 mm). Mori et al. [34] experimentally studied the transmission characteristics of Lamb wave S0
and A0 modes at the contact edges of two plates. The results showed that the transmission coefficient of
the S0 mode increased monotonically with increasing contact pressure, but the transmission coefficient
of the A0 mode had a non-monotonic relationship with contact pressure and local minimum frequency.
Cho et al. [35] calculated the reflection and propagation coefficients of the selected guided wave
modes using simulation and compared the results with experimental data. The results showed
that defect characterization was possible if a suitable criterion for modal selection was established.
Ochoa et al. [36] used two zeroth-order Lamb wave modes to detect various types of invisible impact
damage in composite materials. Because the digital shearography and ultrasonic C-scan technology
could identify 5- and 10-J impact damage and the impact damage of 3-J could not be detected, the
results of the Lamb wave detection were confirmed by comparison.
1.2.3. Characteristics of Lamb Waves

Pant et al. [37] analyzed the influence of material properties, such as E11 (axial elastic modulus),
E22 (transversal elastic modulus), G12 (plane shear modulus), and ρ (density), on Lamb wave
propagation/dispersion of three kinds of composite laminates. These composite laminates are
unidirectional, cross-layered, and quasi-isotropic, which are all composed of 16 layers of unidirectional
carbon fiber epoxy prepreg. The results showed that when Lamb waves are used for damage detection,
the A0 mode should be selected when f d (frequency times thickness) is small, and the S0 mode should
be selected when f d is large. Shen et al. [38] proposed an effective numerical method with which
to study the nonlinear scattering and mode conversion of Lamb waves interacting with breathing
cracks. The results showed that the scattering phenomenon was related to the incident wave mode.
Symmetric and antisymmetric scattering modes exhibit alternate properties, which, in turn, carry most
of the energy and dominate the harmonic scattering. Ai et al. [39] used the finite element method
to study the propagation characteristics of Lamb waves in thin plates, and when compared with the
frequency dispersion curve, the error of the simulation was 1.3%. Senyurek et al. [40] studied the
notch and impact damage of slats, and the results showed that Lamb waves could provide time and
cost efficiency in aircraft structure detection compared with other technologies. The finite element
method was used by Alkassar et al. [41] to simulate the propagation of Lamb waves in a thin metal
plate. The results showed that when there was damage in the plate, a new packet-wave formed by a
mode conversion phenomenon could be observed.
1.2.4. Defects Identification Method

Yelve et al. [42] studied the interaction between the Lamb wave of the S0 mode and the damage in
an aluminum plate and used an artificial neural network (ANN) to locate the damage. The results
showed that the ANN was an effective tool for damage location in a plate structure. The local
wavenumber method of Lamb waves was used by Fan et al. [43] to characterize the flat bottom defects
in isotropic thin plates. The simulation and experimental results showed that the average wavenumber
195
Appl. Sci. 2019, 9, 3576
could be used to not only identify the shape, position, and size of the damage but also quantify the
depth of the damage.
1.3. The Contribution

In the development process of on-line Lamb wave detection equipment for microcracks in
thin-walled tubes, the following work items have been accomplished.
(i) The dispersion characteristics of Lamb waves in thin plates with the same thickness were used to
approximate the ultrasonic characteristics of thin-walled tubes, which reduced the dimensions of
the model and laid a foundation for the engineering application of Lamb waves.
(ii) Based on the calculation and simulation of phase velocity, incident angle, and particle amplitude
displacement, the Lamb wave mode with the highest sensitivity to the inner wall and outer wall
cracks of a thin-walled cylinder was determined.
(iii) Multi-channel technology was adopted to solve the problem of rapid detection and improve the
stability and reliability of the equipment.
(iv) The defect signal extraction circuit was designed, which greatly reduced the amount of data
acquired and improved the data processing speed.
(v) Rapid on-line detection equipment for microcracks in thin-walled tubes was developed, and the
engineering application of rapid detection of microcracks by Lamb wave was realized.
(vi) Qualitative and quantitative non-destructive testing of the minimum depth crack of 10 × 0.15 ×
10 0.15 0.045
0.045 (length × width × depth) was realized.
2. Theoretical Principle
2.1. Testing Object and Artificial Defect

The structural sketch of a thin-walled tube is shown in Figure 1. Each tube was a steel pressure
vessel with a variable wall thickness formed by cold stamping and stretching. The wall thickness of
the mouth was 0.8 mm, the wall thickness near the bottom was 1.2 mm, and the inner diameter of the
mouth was 37 mm. The detection range was a thin-walled tube 255 mm away from the mouth.
0.8 mm
37 mm
1.2 mm
Figure 1. Structural sketch of a thin-walled tube.
196
Appl. Sci. 2019, 9, 3576
Because thin-walled tubes are stamped and stretched in the production process, the inner
and outer surfaces are prone to scratches with axial and circumferential distributions. In the actual
production process, it was determined by manual inspection that the main defects are axially distributed.
Therefore, to study the ultrasonic signal characteristics of these defects, groove defects with axial and
circumferential distributions were fabricated on the inner and outer surfaces of thin-walled tube samples.
Artificial defects were manufactured by the national measurement unit using electro-processing
technology and measured by an optical image method. The lengths and widths of the defects
were 10 ± 0.1 mm and 0.15 ± 0.01 mm, respectively. Three depths were used: 0.045 ± 0.005 mm,
0.073 ± 0.005 mm, and 0.10 ± 0.005 mm, and the detailed information is shown in Table 1.
Table 1. Artificial defects of different depths.
Defect Types Defect Position (Away from the Mouth)/mm Defect Depth/mm
Inner circumferential distribution 40 0.045 ± 0.005
Inner axial distribution 60 0.045 ± 0.005
Outer axial distribution 50 0.045 ± 0.005
2.2. Dispersion Characteristics of Lamb Waves

In the process of Lamb wave flaw detection, different modes of the Lamb wave will be excited
when different incident angles are used, and thus, different sensitivities can be obtained for defects
at different locations. Therefore, theoretical analysis is needed to determine which mode is most
sensitive to the axial and circumferential distribution defects before testing. The Rayleigh–Lamb
frequency equation is used to describe the wave characteristics of Lamb waves [11]. They are expressed
mathematically as follows.
Symmetric mode:
πfd 2 πfd
4pq tan q + (p2 − 1) tan p = 0, (1)
cp cp
Antisymmetric mode:
2 πfd πfd
(p2 − 1) tan q + 4pq tan p = 0, (2)
cp cp
q q
cp 2 cp 2
where p = ( cT ) − 1, q = ( cL ) − 1, cp is the phase velocity of Lamb wave, cT is the S-wave velocity,
cL is the P-wave velocity, f is the frequency of Lamb wave, and d is the thickness of the plate.
In calculating the dispersion characteristics of the Lamb waves, the P-wave velocity cL and S-wave
velocity cT of the steel plate were 5970 and 3200 m/s, respectively. The Rayleigh–Lamb frequency
equation was solved numerically using Matlab, and the phase velocity cp vs. frequency-times-thickness
f d dispersion curves of the Lamb wave was obtained, as shown in Figure 2.
Because the wall thickness of the thin-walled tubes varied from 0.8 to 1.2 mm, when a 5 MHz
transducer was used for detection, the frequency-times-thickness was 4–6 MHz·mm, as shown by the
red dotted line in Figure 2. As shown in the figure, the possible modes of the Lamb wave included the
A0 , S0 , A1 , S1 , A2 , and S2 modes. Because the phase velocities of the A1 , S1 , A2 , and S2 modes vary
greatly, the position of the defect echo signal is not fixed on the time axis, and it is difficult to identify
the defect signal. Therefore, only the A0 and S0 modes were considered in the detection process.
197
Appl. Sci. 2019, 9, 3576
Figure 2. Relationship between the phase velocity cp and frequency-times-thickness f d for Lamb waves
p
in a steel plate (the S-wave velocity was 5970 m/s, the P-wave velocity was 3200 m/s, and Poisson’s
ratio was 0.28).
2.3. Mode Selection

Through the dispersion curves of the Lamb waves, we determined that the optional modes were
A0 or S0 . However, how to distinguish between them required further study of the direction of particle
vibrations
A0 S0 and the distributions of particle displacements for the A0 and S0 modes in plates. The
particle displacement-amplitude curve of the Lamb waves reflects the variation A0 and distribution
S0 of
Lamb wave energy in the plate, which is one of the main bases for mode selection. The in-plane and
out-of-plane displacement amplitudes of the particle vibrations are denoted as U and V, respectively.
If the vertical distance between the particle and waveguide axis is x, their mathematical expressions
are as follows [44]:
Symmetric mode:
(kz2 − s2 ) sin(0.5sd)
Us = cos(q′ x)B + s cos(sx)B, (3)
2q(k′ 2sin (0.5q′ d)
s2 )sin(0.5sd) '
'
2 sin(0.5 '
)
cos( ) cos( )
(k2 − s2 ) sin(0.5sd)
Vs = z ( 2 2 )sin(0.5 ′) sin(q′ x)B + kz sin(sx)B, (4)
2kz sin(0.5q d sin( ' )
) sin( )
2 sin(0.5 ' )
Antisymmetric mode:
(kz2 −s2 ) cos(0.5sd) (kz2 −s2 ) cos(0.5sd)

Ua = (k′2 cos
s2 )cos(0.5sd) sin(q' ′ x)D − s sin(sx)D, Va = ( 22kz 2cos
)cos(0.5
(0.5q′ d) )
cos'(q′ x)D + kz cos(sx)D, (5)
2q (0.5q′ d)
2 ' cos(0.5 '
)
sin( ) sin( ) 2 cos(0.5 '
)
cos( ) cos( )
q q
2π f cp 2 cp 2
where kz = 2cπp , s = kz ( cT2) − 1, q' ′ = kz ( c2L ) − 1, and cp is the phase velocity of Lamb wave.
( ) 1 ( ) 1
In the symmetric mode, the vibration of a particle inp a plate is symmetrical in the center of the
plate, and its in-plane displacement is symmetric along the thickness of the plate, while the out-of-plane
displacement is antisymmetric. Under the antisymmetric mode, the vibration phase of the particle in
the plate is the same on the upper and lower surfaces, and the out-of-plane displacement is symmetric
along the thickness of the plate, while the in-plane displacement is antisymmetric along the thickness
of the plate. For the steel plate with 1-mm thickness, the particle displacement-amplitude curves of the
A0 and S0 modes are shown in Figure 3.
A0 S0
198
Appl. Sci. 2019, 9, 3576
In-plane displacement
Out-of-plane displacement In-plane displacement
Out-of-plane displacement
0.5 0.5
Plate thichness /mm

Plate thichness /mm
0 0
-0.5 -0.5
-1.0 -0.5 0 0.5 1.0 -1.0 -0.5 0 0.5 1.0
Normalized displacement Normalized displacement
(a) (b)
Figure 3. Particle displacement-amplitude curves of A0 0and S0 S

modes
0 in steel plate ((a) A0 mode;
0
(b) S0 mode). S 0
In Figure 3, the dashed line is the in-plane displacement along the thickness direction; that is, the
vibration direction of the particle is parallel to the plate surface, while the real line is the out-of-plane
displacement, that is, the vibration direction of the particle is perpendicular to the plate surface. The
figure shows that the in-plane displacement of mode S0 had a larger component in the plate, which is
S0
suitable for detecting the defects perpendicular to the plate surface, and the out-of-plane displacement
of mode A0 had a larger component in the plate, which is suitable for detecting defects on the inner
and outer surfaces of theAplate. Therefore, the A0 mode, which is sensitive to the surface defects of the
0
inner and outer walls of thin-walled tubes, was selected as the detection mode. This is consistent with
A0
the conclusion of Ref. [37].
2.4. Calculation of Incident Angle

The method of water immersion was adopted in the detection process, so it was necessary to
calculate the incident angle of the A0 mode Lamb wave produced by the ultrasonic transducer when
the Lamb waves propagate from water into the steel. The mathematical formula is as follows:
A0
cL
α = arcsin , (6)
cp
α arcsin
where α is the incident angle, cL is the P-wave velocity in water, and cp is the phase velocity of Lamb
wave in steel. The relationship between the incident angle and frequency-times-thickness is shown in
α L p
Figure 4.
As shown in Figure 4, when the frequency-times-thickness was 4–6 MHz mm, the incident angle
of the mode A0 corresponded to 30.5–29.99◦ . In practical applications, 30◦ was chosen as the incident
angle to obtain the A0 mode of the Lamb waves.
199
α arcsin
α
Appl. Sci. 2019, 9, 3576
Figure 4. Relationship between the incident angle and frequency-times-thickness in steel (the P-wave
velocity in water was 1480 m/s).
0
3. Testing Equipment 0
3.1. Appearance and Core Components

As shown in Figure 5, the testing equipment was mainly composed of three parts: a Programmable
Logic Controller (PLC) control unit, a Personal Computer (PC) control unit, and a mechanical device.
The control unit of the PLC was mainly responsible for the motion control of the mechanical device
and communication with the PC through the RS232 serial module. The PC control unit mainly
included a multi-channel pulse transmitter/receiver, the extraction device of the detection signal
feature, the multi-channel data acquisition card, and the PC. The function of the multi-channel pulse
transmitter/receiver was to detect the object using an excitation pulse to trigger the ultrasonic transducer
and receive the detection signal from the detected object. The feature extraction device for the testing
signal mainly ensured that only the defect echo signal was extracted through the hardware gate to
reduce the amount of data acquisition and improve the speed of data processing. The mechanical
device mainly performed the mechanical actions of grasping, rotating, and moving the thin-walled
tube into place for testing.
Figure 5. Appearance of testing equipment (1. PLC control unit; 2. PC control unit; 3. mechanical device).
200
Appl. Sci. 2019, 9, 3576
The core components of the mechanical equipment are shown in Figure 6, which mainly include
the material-preparation container, the testing container, the ultrasonic transducer, and the built-in
manipulator. The material-preparation container was mainly used to place the thin-walled tube to be
inspected. By moving up and down, the thin-walled tube was conveyed to a position that the built-in
manipulator could grasp. There was an overflow outlet at the mouth of the testing container, which
was used to discharge the overflow amount of water as the thin-walled tube slowly sank into the testing
container. There was a water inlet at the bottom of the testing container, which was used to replace
water when the thin-walled tube was removed from the testing container after the detection to ensure
the coupling between the ultrasonic transducer and the thin-walled tube. A pair of ultrasonic probes
were installed on the side of the testing container to scan the axial and circumferential distribution
defects of the thin-walled tube. The movement of the testing container and the rotation of the built-in
manipulator carrying the thin-walled tube could cooperate, which allowed the spiral scanning of the
thin-walled tube by the ultrasonic probes. The built-in manipulator held the thin-walled tube from
the inside. Meanwhile, an airbag at the top of the built-in manipulator sealed the opening of the
thin-walled tube to prevent water from penetrating the thin-walled tube during the detection process.
The built-in manipulator moved back and forth along the guide rail to grasp, rotate, and move the
thin-walled tube into place.
Figure 6. Core components of testing equipment (1. material-preparation container; 2. testing container;
3. ultrasonic transducer; 4. built-in manipulator).
The detection process was as follows. Firstly, when all five material-preparation containers were
placed into the thin-walled tubes, the built-in manipulator moved backward to the central position of
the material-preparation container. When the material-preparation container rose to a certain position,
the built-in manipulator held the thin-walled tube from the inside and allowed the top airbag to
seal the opening of the thin-walled tube. The material-preparation container quickly dropped to the
original position. Secondly, the built-in manipulator carrying the thin-walled tube moved forward
to the center of the testing container and began to rotate at 320 rpm. The testing container quickly
rose to a certain position, and the PLC control unit notified the PC through the RS232 serial port
to begin the detection. The PC started the data acquisition card and the pulse transmitter/receiver,
and the testing container began to rise slowly and uniformly so that the ultrasonic transducer could
achieve full coverage spiral scanning of the thin-walled tube to be detected. Thirdly, after the detection,
the built-in manipulator stopped rotating, and the testing container quickly dropped to the original
position. The PC stopped the data acquisition and transferred the data processing results to the PLC
control unit through an RS232 serial port. Finally, the built-in manipulator continued to move forward
201
Appl. Sci. 2019, 9, 3576
and based on the data processing results, the thin-walled tube identified as the unqualified product
was placed on the conveyor belt of rejected product. Then, the built-in manipulator moved on, and the
remaining thin-walled tubes were placed on the conveyor belt of qualified product to complete the
detection process.
3.2. Multichannel Detection Method

Because the detection speed required 6 s/piece, there were three high-speed requirements for
single-piece detection: (i) the assistant time was zero; (ii) the rotational speed of thin-walled tube was
over 3000 rpm; (iii) the sampling frequency was over 100 MHz. These three requirements directly
affected the stability, reliability, and service life of the system, and the design risk was very high. If five
groups of 10-channel parallel structures were adopted, the detection time could be extended to 30 s
(18 s for a detection time, 12 s for auxiliary time; the rotational speed could be reduced to 300 rpm, and
the sampling frequency could be reduced to 100 kHz). Thus, not only was the impact of mechanical
transmission alleviated but also sufficient processing time was provided for data acquisition and defect
signal processing. The problem of detection time, which is difficult to solve, was transferred to a
multi-channel problem that was relatively easy to solve. Furthermore, this laid a foundation for the
stable and reliable design of the system.
4. Results and Analysis

In the detection process, the ultrasonic transducer was a point focus probe with a frequency of
2.5 MHz, a diameter of 12 mm, a focal length of 20 mm, and a focal spot of 2 mm. The diameter of the
thin-walled tube was 37 mm, the length of the detection area was 255 mm, and the rotation speed was
320 rpm. Thus, 0.1875 s was required for the tube to rotate once. Based on the single-channel sampling
frequency of 0.75 kHz, the number of points sampled during the rotation of the tube was calculated to
be 140. The total sampling length was 20,800 points. For the longitudinal defect with a 10-mm length
(i.e., the defect distributed along the axis), the theoretical detection period was 10/1.72 = 5.81 (where 10
is the length of longitudinal defect, in-unit of mm; 1.72 is the pitch; that is, the axial interval of the
helix formed by ultrasonic transducer scanning along the circumferential direction of the thin-walled
tube, in-unit of mm; 5.81 is the number of defect waves obtained by the ultrasonic transducer scanning
the longitudinal artificial defect with 10 mm length, in-unit of cycle). For the transverse defect with a
10-mm length (i.e., the defect distributed along the circumference), the theoretical detection period
was 12/1.72 = 6.98 (where 12 is the diameter of the transducer, in-unit of mm; the transverse defect is
distributed along the circumference of the thin-walled tube, and the transducer is also spiral-scanned
along the circumference of the thin-walled tube, so the number of cycles scanned by the transducer
for the transverse defect depends on the ratio of the diameter of the transducer to the pitch; 1.72 is
the pitch, in-unit of mm; 6.98 is the number of defect waves obtained by the ultrasonic transducer
scanning the transverse artificial defect with a 10-mm length, in-unit of cycle). The actual number of
defect detection cycles is shown in Figure 7a–h. The number of transverse defect detection cycles was
seven, and the number of longitudinal defect detection cycles was 8–10. The actual detection period of
the longitudinal defect was larger than the theoretical value, while that of the transverse defect was
equal to the theoretical value. This was due to the different scanning and defect distribution directions.
3.75
Amplitude/V
Amplitude/V
3.75
2.50
2.50
1.25 1.25
0 0
0 500 1000 1500 0 500 1000 1500
Sampling points/points Sampling points/points
(a) (b)
Figure 7. Cont.
202
Appl. Sci. 2019, 9, 3576
Amplitude/V
3.75
Amplitude/V
3.75
2.50 2.50
1.25 1.25
00 0
500 1000 1500 0 500 1000 1500
Sampling points/points
(c) (d)
Amplitude/V
3.75
Amplitude/V
3.75
2.50 2.50
1.25 1.25
00 500 1000 1500 0
0 500 1000 1500
(e) (f)
Amplitude/V
Amplitude/V
3.75 3.75
2.50 2.50
1.25 1.25
00 500 1000 1500 00 500 1000 1500
(g) (h)
Amplitude/V
3.75
Amplitude/V
3.75
2.50 2.50
1.25 1.25
0 0
0 500 1000 1500 0 500 1000 1500
(i) (j)
Figure 7. Signal with and without flaw ((a) Depth of inner transverse flaw was 0.045 mm; (b) Depth of
inner longitudinal flaw was 0.045 mm; (c) Depth of outer transverse flaw was 0.045 mm; (d) Depth of
outer longitudinal flaw was 0.045 mm; (e) Depth of inner longitudinal flaw was 0.073 mm; (f) Depth of
outer longitudinal flaw was 0.073 mm; (g) Depth of inner longitudinal flaw was 0.10 mm; (h) Depth of
outer longitudinal flaw was 0.10 mm; (i) Signal without flaw and with strong noise; (j) Signal without
flaw and with weak noise).
4.1. Analysis of Qualitative Detection

The signals of the ultrasonic testing with and without flaws are shown in Figure 7. For (a), (c) and
(b), (d) in Figure 7, the defects with same depth, same distribution direction, and different location,
from the amplitude observation for defect wave, seemed to indicate that (a), (c) and (b), (d) in Figure 7
overlap; in fact, they were ultrasonic echo signals of defects at different locations, but the difference
was very small. For (a), (b) and (c), (d) in Figure 7, the defects with the same depth, same location, and
different distribution directions, the difference between them was obvious. Among them, Figure 7a–h
showed the echo signals of defects with different depths, and Figure 7i,j showed the echo signal
without defects. Therefore, regardless of the signal’s periodicity or amplitude, the differences in the
echo signals, when flaws were present or absent, were very significant, which showed that the method
adopted in this study exhibited a good detection performance for qualitative detection.
4.2. Analysis of Quantitative Detection

To better analyze the quantitative relationship between the defect depth and signal amplitude, all
the defect signals were smoothed first to eliminate the influence of single-peak extrema. Second, the
peak values of the echo signals at different depths were extracted and plotted. Finally, the first and last
peaks of the echo signals with different depths were eliminated, and the mean of residual peaks in the
signals was calculated to make the mean curve approach the peak value of the defect signals as stably
3.75 3.75
as possible. 0.045 0.045
Amplitude/V
Amplitude/V
0.073 2.86 0.073

2.76
As shown in Figure 8, for longitudinal 0.10internal2.39
defects, when the defect depth increased0.10 from
2.05
0.045
1.36
to 0.073 mm, the echo signal peak value increased 1.36
from 1.36 to 2.05 V. When the defect depth
increased from 0.073 to 0.10 mm, the echo signal peak value increased from 2.05 to 2.76 V. For the
longitudinal
0
0 defect,
500 when the 1000
defect depth 1500
increased from0
0 0.045 to 0.073
500 mm, the1000
peak value of
1500the
203
Appl. Sci. 2019, 9, 3576
echo signal increased from 1.36 to 2.39 V. When the defect depth increased from 0.073 to 0.10 mm, the
peak value of the echo signal increased from 2.39 to 2.86 V.
3.75 3.75
0.045 0.045
Amplitude/V
Amplitude/V
0.073 2.86 0.073
2.76
0.10 2.39 0.10
2.05
1.36 1.36
0 0
0 500 1000 1500 0 500 1000 1500
(a) (b)
Figure 8. The peak of flaw signal with different depths ((a) The peak of inner flaw signal with different
depths; (b) The peak of outer flaw signal with different depths).
The results showed that the quantitative relationship of the internal longitudinal defects was
slightly better than that of external longitudinal defects. There was a linear quantitative relationship
between the depth of the longitudinal defect and the peak value of the detection signal. The depth of
the detected defect could be represented by the peak value of the defect detection signal.
5. Issues and Conclusions
5.1. Issues
5.1.1. Misjudgment
Based on the minimum defect standard of 10 × 0.15 × 0.045 (length × width × depth) (taking a
thin-walled tube with a 0.045-mm depth artificial defect as a standard test block, when the signal
amplitude of the Lamb wave for the defect in the detected thin-walled tube is greater than or equal
to the signal amplitude of the Lamb wave for the artificial defect in the standard test block, and the
signal periods of the Lamb wave for the defect is greater than or equal to the signal periods of Lamb
wave for the artificial defect in the standard test block, the tested thin-walled tube will be judged
as unqualified product. Otherwise, the tested product will be judged as a qualified product. The
missed detection rate is the ratio of unqualified products that are identified as qualified products to the
total detected quantity, while the reject rate is the ratio of unqualified products to the total detected
quantity), 100,000 thin-walled tubes were tested. The test results showed that the leakage rate was
0%, the rejection rate was 0.3%, and the average detection time of each product was 5.8 s, which fully
meets the actual application requirements of a production line. Owing to the airbag located at the top
of the built-in manipulator being cut, a small amount of water penetrated the thin-walled tubes during
detection, resulting in misjudgment. This problem was solved by modifying the sawing technology of
the mouth of the thin-walled tube.
5.1.2. Signal-to-Noise Ratio

To solve the problem of the rapid detection of thin-walled tubes, a multi-channel parallel detection
method was adopted. Owing to the differences in the components between channels, some channels
had high signal-to-noise ratios and no noise, as shown in Figure 7a–d,g,h. Other channels had echo
noise with small amplitudes, as shown in Figure 7e,f,i,j. In the follow-up equipment development
process, it is necessary to further optimize the design of related hardware circuits to further improve
the consistency of multi-channel detection signals.
5.2. Conclusions
In the development process of on-line Lamb wave detection equipment for microcracks in
thin-walled tubes, the following conclusions were drawn.
204
Appl. Sci. 2019, 9, 3576
(i) Based on the calculation and simulation of phase velocity, incident angle, and particle amplitude
displacement, the Lamb wave A0 mode with the highest sensitivity to the inner wall and outer
wall cracks of thin-walled tubes was determined. This is consistent with the results of Ref. [37].
(ii) A defect signal extraction circuit was designed, which greatly reduced the amount of data acquired
and improved the data processing speed.
(iii) Rapid on-line detection equipment for microcracks in thin-walled tubes was developed, and the
engineering application of rapid detection of microcracks by Lamb waves was realized.
(iv) For the 0.045-mm depth of the artificial defect, whether the inner wall, or outer wall of the
longitudinal defect, or the transverse defect, there was almost no difference in the shape and
amplitude of the defect waveform. However, for the inner wall of the longitudinal defect and the
transverse defect, or the outer wall of the longitudinal defect and the transverse defect, waveform
amplitude and contour had certain differences.
(v) For the artificial defects with depths of 0.073 and 0.10 mm, the longitudinal defects of the inner
and outer walls were quite different in terms of waveform profile and amplitude.
(vi) This study not only solved the practical issue of the rapid detection of microcracks in thin-walled
tubes but also provided a reference for the application of ultrasonic technology to detect
other components.
Author Contributions: Conceptualization, S.Y. and M.W.; Methodology, S.Y.; Software, S.Y. and L.Y.; Validation,
L.Y., and S.Y.; Formal Analysis, M.W.; Investigation, L.Y.; Resources, M.W.; Data Curation, L.Y.; Writing—Original
Draft Preparation, S.Y.; Writing—Review and Editing, S.Y. and M.W.; Visualization, L.Y.; Supervision, M.W.;
Project Administration, S.Y. and M.W.; Funding Acquisition, M.W.
Funding: This work was supported by the National Key Scientific Instrument and Equipment Development
Projects of China (Grant No. 2013YQ240803) and the International Science and Technology Cooperation Program
of Shanxi, China (Grant No. 201803D421032).
References
1. Shim, H.; Choi, M.S.; Lee, D.H.; Hur, D.H. A prediction method for the general corrosion behavior of Alloy
690 steam generator tube using eddy current testing. Nucl. Eng. Des. 2016, 297, 26–31. [CrossRef]
2. Buck, J.A.; Underhill, P.R.; Morelli, J.; Krause, T.W. Intranets: Evaluation of machine learning tools for
inspection of steam generator tube structures using pulsed eddy. In Proceedings of the 43th Annual Review
of Progress in Quantitative Nondestructive Evaluation, Atlanta, GA, USA, 17–22 July 2016; AIP: Atlanta, GA,
USA, 2016; Volume 36. [CrossRef]
3. Liu, Y.; Lian, Z.; Lin, T.; Shen, Y.; Zhang, Q. A study on axial cracking failure of drill pipe body. Eng. Fail. Anal.
2015, 59, 434–443. [CrossRef]
4. Dong, B.; Fang, G.; Liu, Y.; Dong, P.; Zhang, J.; Xing, F.; Hong, S. Monitoring reinforcement corrosion and
corrosion-induced cracking by X-ray microcomputed tomography method. Cem. Concr. Res. 2017, 100,
5. Lauridsen, T.; Willner, M.; Bech, M.; Pfeiffer, F.; Feidenhans, R. Detection of sub-pixel fractures in X-ray
dark-field tomography. Appl. Phys. A 2015, 121, 1243–1250. [CrossRef]
6. Irek, P.; Slania, J. Material factors in relation to development time in liquid-penetrant inspection. Part 1.
Material factors. Arch. Metall. Mater. 2016, 61, 509–514. [CrossRef]
InvestIgatIon programme and preliminary tests. Arch. Metall. Mater. 2016, 61, 1697–1708. [CrossRef]
Testing of model plates. Arch. Metall. Mater. 2017, 62, 41–49. [CrossRef]
9. Rose, J.L. Ultrasonic Waves in Solid Media; Cambridge University Press: New York, NY, USA, 2004; pp. 73–143.
10. Hellier, C. Handbook of Nondestructive Evaluation; McGraw-Hill: New York, NY, USA, 2003; pp. 1–7.
11. Rose, J.L. Ultrasonic Guided Waves in Solid Media; Cambridge University Press: New York, NY, USA, 2014;
pp. 48–137.
205
Appl. Sci. 2019, 9, 3576
12. Li, Z.; He, C.; Liu, Z.; Wu, B. Quantitative detection of lamination defect in thin-walled metallic pipe by
using circumferential Lamb waves based on wavenumber analysis method. NDT E Int. 2019, 102, 56–67.
[CrossRef]
13. Howard, R.; Cegla, F. Detectability of corrosion damage with circumferential guided waves in reflection and
transmission. NDT E Int. 2017, 91, 108–119. [CrossRef]
14. Liu, K.; Wu, Z.; Jiang, Y.; Wang, Y.; Zhou, K.; Chen, Y. Guided waves based diagnostic imaging of
circumferential cracks in small-diameter pipe. Ultrasonics 2016, 65, 34–42. [CrossRef]
15. Yeung, C.; Ng, C.T. Time-domain spectral finite element method for analysis of torsional guided waves
scattering and mode conversion by cracks in pipes. Mech. Syst. Signal Process. 2019, 128, 305–317. [CrossRef]
16. Zhanga, X.; Tangb, Z.; Lva, F.; Yang, K. Scattering of torsional flexural guided waves from circular holes and
crack-like defects in hollow cylinders. NDT E Int. 2017, 89, 56–66. [CrossRef]
17. Lowea, P.S.; Sandersonb, R.; Pedrama, S.K.; Boulgouris, N.V.; Mudge, P. Inspection of pipelines using the
first longitudinal guided wave mode. Phys. Procedia 2015, 70, 338–342. [CrossRef]
18. Liu, Z.; Hu, Y.; Fan, J.; Yin, W.; Liu, X.; He, C.; Wu, B. Longitudinal mode magnetostrictive patch transducer
array employing a multi-splitting meander coil for pipe inspection. NDT E Int. 2016, 79, 30–37. [CrossRef]
19. Heinlein, S.; Cawley, P.; Vogt, T.K. Reflection of torsional T (0,1) guided waves from defects in pipe bends.
NDT E Int. 2018, 93, 57–63. [CrossRef]
20. Nakamura, N.; Ogi, H.; Hirao, M. EMAT pipe inspection technique using higher mode torsional guided
wave T (0,2). NDT E Int. 2017, 87, 78–84.
21. Niu, X.; Duan, W.; Chen, H.P.; Marques, H.R. Excitation and propagation of torsional T (0,1) mode for guided
wave testing of pipeline integrity. Measurement 2019, 131, 341–348. [CrossRef]
22. Kim, Y.; Park, K. Characterization of axial and oblique defects in pipes using fundamental torsional guided
modes. NDT E Int. 2017, 92, 149–158. [CrossRef]
23. Muggleton, M.; Kalkowski, M.; Gao, Y.; Rustighi, E. A theoretical study of the fundamental torsional wave in
buried pipes for pipeline condition assessment and monitoring. J. Sound Vib. 2016, 374, 155–171. [CrossRef]
24. Zhang, X.; Tang, Z.; Lv, F.; Pan, X. Helical comb magnetostrictive patch transducers for inspecting spiral
welded pipes using flexural guided waves. Ultrasonics 2017, 74, 1–10. [CrossRef]
25. Zhang, X.; Tang, Z.; Lü, F.; Pan, X. Excitation of dominant flexural guided waves in elastic hollow cylinders
using time delay circular array transducers. Wave Motion 2016, 62, 41–54. [CrossRef]
26. Li, J.; Rose, J.L. Natural beam focusing of non-axisymmetric guided waves in large-diameter pipes. Ultrasonics
2005, 44, 35–45. [CrossRef] [PubMed]
27. Nishino, H.; Takashina, S.; Uchida, F.; Takemoto, M.; Ono, K. Modal analysis of hollow cylindrical guided
waves and applications. Jpn. J. Appl. Phys. 2001, 40, 364–370. [CrossRef]
28. Velichko, A.; Wilcox, P.D. Excitation and scattering of guided waves: Relationship between solutions for
plates and pPipes. J. Acoust. Soc. Am. 2009, 125, 3623–3631. [CrossRef] [PubMed]
29. He, C.; Li, L.; Wu, B. Guided circumferential waves in hollow cylinders. Chin. J. Mech. Eng. 2004, 40, 7–12.
[CrossRef]
30. Qi, L.; Ya, X.X.; Li, X.D.; Yu, S.F. Multiphysics simulation method of lamb wave propagation with piezoelectric
transducers under load condition. Chin. J. Aeronaut. 2019, 32, 1071–1086. [CrossRef]
31. He, J.; Dixon, S.; Hill, S.; Xu, K. A new electromagnetic acoustic transducer design for generating and
receiving S0 lamb waves in ferromagnetic steel plate. Sensors 2017, 17, 1023. [CrossRef]
32. Zhang, K.; Zhou, Z. Quantitative characterization of disbonds in multilayered bonded composites using
laser ultrasonic guided waves. NDT E Int. 2018, 97, 42–50. [CrossRef]
33. Gupta, S.; Rajagopal, P. Effect of ply orientation and through-thickness position of delamination on the
reflection of fundamental symmetric S0 Lamb mode in GFRP. Ultrasonics 2018, 90, 109–119. [CrossRef]
34. Mori, N.; Biwa, S. Transmission characteristics of the S0 and A0 Lamb waves at contacting edges of plates.
35. Cho, Y.; Hongerholt, D.D.; Rose, J.L. Lamb wave scattering analysis for reflector characterization. IEEE Trans.
Ultrason. Ferroelectr. Freq. Control 1997, 44, 44.
36. Ochôa, P.; Infante, V.; Silva, J.M.; Groves, R.M. Detection of multiple low-energy impact damage in composite
plates using Lamb wave techniques. Compos. Part B Eng. 2015, 80, 291–298. [CrossRef]
37. Pant, S.; Laliberte, J.; Martinez, M.; Rocha, B.; Ancrum, D. Effects of composite lamina properties on
fundamental Lamb wave mode dispersion characteristics. Compos. Struct. 2015, 124, 236–252. [CrossRef]
206
Appl. Sci. 2019, 9, 3576
38. Shen, Y.; Cesnik, C.E.S. Nonlinear scattering and mode conversion of Lamb waves at breathing cracks: An
efficient numerical approach. Ultrasonics 2019, 94, 202–217. [CrossRef] [PubMed]
39. Ai, C.; Cai, X.; Li, J.; Liu, K. Intranets: Finite element simulation of Lamb wave detection in thin plate. In
Proceedings of the International Conference on Mechatronics Engineering & Information Technology, Xi’an,
China, 27–28 August 2016; Atlantis Press: Xi’an, China, 2016. [CrossRef]
40. Senyurek, V.Y. Detection of cuts and impact damage at the aircraft wing slat by using Lamb wave method.
Measurement 2015, 67, 10–23. [CrossRef]
41. Alkassar, Y.; Agarwal, V.K.; Alshrihi, E. Simulation of Lamb wave modes conversions in a thin plate for
damage detection. Procedia Eng. 2017, 173, 948–955. [CrossRef]
42. Yelve, N.P.; Mulla, I.B. Locating damage in thin metallic plates using Lamb waves and Artificial Neural
Network. In Proceedings of the International Conference on Nascent Technologies in Engineering,
Navi Mumbai, India, 27–28 January 2017.
43. Fan, G.; Zhang, H.; Zhang, H.; Zhu, W.; Chai, X. Lamb wave local wavenumber approach for characterizing
flat bottom defects in an isotropic thin plate. Appl. Sci. 2018, 8, 1600. [CrossRef]
44. Viktorov, I.A. Rayleigh and Lamb Waves: Physical Theory and Applications; Plenum Press: New York, NY, USA,
1967; pp. 75–82.
207
applied
sciences
Article
A Deep Learning Based Method for the
Non-Destructive Measuring of Rock Strength through
Hammering Sound
Shuai Han 1 , Heng Li 2 , Mingchao Li 1, * and Timothy Rose 3, *
1 State Key Laboratory of Hydraulic Engineering Simulation and Safety, Tianjin University,
Tianjin 300354, China
2 Department of Building and Real Estate, The Hong Kong Polytechnic University, Hong Kong, China
3 Science and Engineering Faculty, Queensland University of Technology, Queensland, QLD 4000, Australia
* Correspondence: [email protected] (M.L.); [email protected] (T.R.)

Received: 14 July 2019; Accepted: 20 August 2019; Published: 23 August 2019
Abstract: Hammering rocks of different strengths can make different sounds. Geological engineers
often use this method to approximate the strengths of rocks in geology surveys. This method is quick
and convenient but subjective. Inspired by this problem, we present a new, non-destructive method
for measuring the surface strengths of rocks based on deep neural network (DNN) and spectrogram
analysis. All the hammering sounds are transformed into spectrograms firstly, and a clustering
algorithm is presented to filter out the outliers of the spectrograms automatically. One of the most
advanced image classification DNN, the Inception-ResNet-v2, is then re-trained with the spectrograms.
The results show that the training accurate is up to 94.5%. Following this, three regression algorithms,
including Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Random Forest (RF) are
adopted to fit the relationship between the outputs of the DNN and the strength values. The tests
show that KNN has the highest fitting accuracy, and SVM has the strongest generalization ability.
The strengths (represented by rebound values) of almost all the samples can be predicted within an
error of [−5, 5]. Overall, the proposed method has great potential in supporting the implementation
of efficient rock strength measurement methods in the field.
Keywords: transfer learning; spectrogram analysis; selecting samples; non-destructive testing;

hammering sound; regression algorithm
1. Introduction
Rock strength is an important parameter for analyzing geological problems. In the field, geological
engineers are always concerned with how to estimate the strength of rocks rapidly, accurately, and
conveniently. There are many methods for estimating rock strength. The most widely used is the
unconfined compressive strength (UCS) test [1]. However, this method damages rock mass, and is
time-consuming and laborious in many cases. Moreover, it is sometimes impossible to damage the
rocks to determine their strength, especially in the field. For these reasons, many non-destructive
methods have been presented to predict the UCS indirectly, such as the rebound method and sonic
technique. The rebound method establishes the relationship between rock strength and rebound
value produced by rebound hammers. However, variations in operational procedures may lead to
large deviations, and there are too many equations used in this method, making it difficult to select a
suitable one. The sonic technique method establishes the relationship between rock strength and sonic
information. However, there are operational challenges in using sonic instruments, which are often
costly and impractical for use in the field.
209
Appl. Sci. 2019, 9, 3484
Due to the lack of facile methods for determining rock strength during geology surveys, experienced
geologists usually prefer to determine rock strength from hearing the sounds produced by hammering
a rock with a geological hammer. The rationale is that the hammering sound can reflect the surface
strength (the outer surface at a depth of 30–50 mm) of the rock to some extent [2], which is directly
related to its UCS. This method is convenient but subjective. The estimation heavily depends on the
experience of geological engineers and is easily affected by many factors, such as noise, dust, and even
the health condition of the geological engineer.
In recent years, deep learning, as one of the emerging fields of machine learning, has gained
increasing interest in various fields. More and more researchers find that many complex problems can
be solved well based on deep neural networks (DNNs). Generally, the two most active application
fields of deep learning are speech recognition and image processing [3,4]. Transfer learning is also an
active field of machine learning and has gained more attention because it can help build effective deep
learning networks without vast amounts of data [5,6].
In this research, we aimed to develop a rapid, accurate, and convenient method for measuring
the surface strength of rocks in the field. Inspired by the success of deep learning and the practice
of estimating rock strength from hammering sounds, a retrained DNN was developed as the core of
the proposed method. In the following of this paper, a literature review is firstly presented, and the
development of non-destructive methods of rock strength and the usage of DNN in geology are
introduced. Then the methodology of the research is described in detail, including the overall process,
the DNN technique, the short-time Fourier transformation, the data filtering algorithm, and the
regression algorithms. Finally, an experiment is carried out to test the proposed method.
2. Literature Review
2.1. Non-Destructive Methods for Measuring Rock Surface Strengths

Non-destructive testing is always the research hotspot of engineering [7]. For the non-destructive
measuring of rock strength, the most commonly used non-destructive methods are the rebound
method [8] and the sonic technique [9]. Many scholars have completed successful research with
the rebound method [10]. For example, Yaşar and Erdoğan [11] carried out a series of experiments
to investigate the relationship between Schmidt hammering rebound value and physicomechanical
properties, including UCS, porosity, and unit volume weight. Aoki and Matsukura [12] developed a
portable and simple equipment to test rock strength in the field, and validated the effectiveness of
the equipment using the rebound method. Lai et al. [13] estimated the mean quantification of rock
mass of five distinct locations through Schmidt hammer rebound tests. Despite these positive results,
it remains difficult to find out a general formula to describe the relationship between rebound values
and the surface strengths of rocks. Moreover, the measurements are of low accuracy, especially when
the operation is not standardized.
In contrast, sonic technology is a relatively accurate method. For example, Sharma and Singh [14]
established an empirical equation to predict rock strength based on P-wave velocity. Tziallas et al. [15]
fitted the relationship between Young’s modulus, UCS, and the velocity of an ultrasonic wave.
Liu et al. [16] proposed a machine learning based method to determine the UCS of rocks with P-wave
and some other indexes such as mineral composition and specific density. Son and Kim [17] used the
sound signal obtained by hammering a rock to calculate the total energy of the sound, and then used
the total energy calculation to calculate the strength of the rock. Azimian [18] developed a model for
predicting UCS with P-wave and Schmidt hammer rebound. However, the equipment used in this
experiment was laboratory-based and had limited application in the field.
2.2. Deep Learning in Geological Engineering

Deep learning has been used in many fields, and geological engineering is no exception.
For example, Palafox et al. [19] adopted a deep convolutional neural network (CNN) to automatically
210
Appl. Sci. 2019, 9, 3484
recognize volcanic rootless cones and impact craters from images of Martian terrain. Xu et al. [20]
established a DNN to classify different kinds of land covers automatically. Sidahmed et al. [21] trained
a DNN to recognize reservoir rock types to help in identifying hydrocarbon resources. Furthermore,
Yu et al. [22] combined the deep CNN and region growing algorithm to recognize landslides.
Training an effective DNN needs a large number of training data; however, in many situations,
there are not enough samples for researchers, such as in geology tasks. Fortunately, transfer learning
offers a good solution to such problems [23,24]. For instance, Li et al. [25] proposed a classification
method for recognizing the features of microscopic sandstone pictures based on transfer learning.
Zhang et al. [26] used transfer learning to identify different geological structures from images.
Generally, there are two methods for transfer learning: (1) continue to train the pre-trained network
model to adjust the structure or weights of the network [27], or (2) remove the last layer of the pre-trained
network model, then use a new dataset to train a new output layer [25]. Considering that the former is
not effective when the sample types and sizes are different, the latter was adopted in this research.
3. Methodology
3.1. Overall Process

The overall process can be presented as Figure 1. Firstly, several rocks were selected as subjects.
For each rock, the strength data was measured with a rebounder and the hammering sound was collected
using a geological hammer and a voice recorder. Next, the hammering sounds were transformed into
spectrograms, and then the spectrograms were filtered with a clustering algorithm. On the other hand,
the measured strength data was pre-processed. After that, some spectrograms and their corresponding
strength data (Batch A) were used to retrain the Inception-ResNet-v2 model, and the other spectrograms
and their corresponding strength data (Batch B) were used for the final test of the method.
In the next sub-sections, the core technologies of this research are described in details, including
the Inception-ResNet-v2 model, the short-time Fourier transformation (STFT), the clustering algorithm,
and the regression algorithms.
Select rocks Dataset

Geological hammer Rebounder
& voice recorder
Batch A Batch B
Collect hammering Collect strength
sound data For re-training For testing For testing
[Section 3.3] STFT
Pre-process
Spectrograms (three sigma Retrain the Inception-ResNet-v2 network
rule) [Section 3.2]
Clustering
[Section 3.4]
algorithm
Make the predictions Continuous
Valid spectrograms Mean strengths （Regression algrithms）
[Section 3.5]
Feature Labels
Dataset Final test of the research
Figure 1. Flow chart of this research. The left scheme shows how to collect and process the data, the
Figure 1. Flow chart of this research. The left scheme shows how to collect and process the data, the
right scheme shows how to train the deep neural network (DNN) model and how to test our method.
3.2. Inception-ResNet-v2 Model

Inception-ResNet-v2 was developed from the Inception Net model by Google [28]. Different from
Inception v1 to v3, the Inception-ResNet-v2 model takes advantage of the residual networks, successfully
211
Appl. Sci. 2019, 9, 3484
improving the accuracy and convergence speed of the original model. In a residual network, as shown
in Figure 2, the output of the previous layer is inputted into the middle layer and the next layer
together with the output of the middle layer [29]. Therefore, when adjusting the weights using the back
propagation method, the gradient from the upper layer can skip over the middle layer up to the lowest
layer, ensuring that all the weights can be adjusted effectively. More details of the Inception-ResNet-v2
model can be found in the publications [28,29].
middle layer
Figure 2. Diagram of a residual network unit.
Generally, the fine-tuning process includes four key steps: (1) train a deep learning network
based on the source data. In this research, the Inception-ResNet-v2 model has been well-trained.
(2) remove the output layer of the network, and reform the size of the output layer according to the
target data, (3) initialize the weights of output layer and keep the weights of other layers unchanged,
and (4)‘re-train the network, as illustrated in Figure 3.
Source Pre-trained
Source data
Label Model
Copy Copy
Target Fine-tuned
Target data Random
Label Model
Figure 3. Process of fine-tuning a deep learning network.
3.3. Short-Time Fourier Transformation (STFT) and Spectrogram

In this research, hammering sounds were firstly transformed into spectrograms because of the
Inception-ResNet-v2′′ s′ strength in image identification. Hence, the STFT was utilized to process the
sounds, as follows:
∞
X
Sx (n, ω) = x(m)w(n − m)e−jωm (1)

−∞
where w(n) represents the window function, and ω is the frequency in radians [30]. In an STFT process,
ωω
it is important to determine the size of the window. A large window leads to poor time resolution,
while a small window leads to a poor frequency resolution. Either window size will degrade the
quality of a spectrogram. Moreover, during programming, the sampling frequency fs, the time lapse
L, and the frequency discretization N also affect the resolution of the spectrograms. However, it is
difficult to determine the exact parameters in different cases. In this research, we found out a set of
parameters through repeated experiments to make the spectrograms look clear.
Spectrograms generated by the STFT are presented in Figure 4. In a spectrogram, there are three
dimensions that correspond to frequency (vertical axis), time (horizontal axis), and sound pressure
(color gradient) or power spectral density (PSD). In the spectrograms example presented in Figure 4,
yellow represents high sound pressure, and blue represents low sound pressure. It is easy to distinguish
how sound pressures are distributed along with the frequencies in a certain period, how the sound
pressures change over time in a certain frequency range, and how the principal frequency changes
over time. Specifically, Figure 4a corresponds to the hammering sound of a rock with low strength,
Figure 4b corresponds to a rock with moderate strength, and Figure 4c corresponds to a rock with high
212
Appl. Sci. 2019, 9, 3484
strength. It is obvious that in Figure 4c the proportion of high frequency is the largest, the attenuating
speed is the slowest, and the differentiation of different frequency bands is the clearest.
Figure 4. Examples of spectrograms. The rebound values of (a), (b), and (c) are 20, 54, and 73
respectively. The horizontal axes represent time (/ms), and the vertical axes are frequencies (/kHz).
Additionally, we found that there were no differences in parts of the spectrograms that had
frequencies greater than 5 kHz, no matter for high-strength rocks or for low-strength rocks; thus,
the parts with high frequency do not contribute to the analysis. Moreover, research has shown that
human’s most sensitive sound frequency is 2 kHz–5 kHz [31]. Therefore, only the 0–5 kHz part of the
spectrograms was used for identifying strengths.
3.4. Selection of Training Samples Based on Clustering
3.4.1. Script Program for Cutting Hammering Sound

In this study, thousands of hammering sounds were transformed into spectrograms. Therefore, a
simple script was programmed to split the hammering sounds automatically. First, for a sound file
that contained hundreds of hammering sounds, the time series was iterated to determine a series of
mutational points, [t1 , t2 , . . . , tn ], according to the amplitude, as shown in Figure 5. Next, it was assumed
that the hammerings occurred at 10 milliseconds before the mutational points and each hammering
sound lasted 150 ms, as shown in the right part of Figure 6. The two durations, 10 milliseconds and
150 ms, were determined by the statistical data in this research. The individual hammering sounds
were then extracted, and the STFT was used to transform all the sound fragments into spectrograms.
However, these spectrograms could not be directly used for training because some sound fragments
may not contain the right hammering sound. For example, some sound fragments may contain loud
talking voices, or some may be produced by hammering at a wrong point by mistake. Therefore,
invalid spectrograms should be eliminated to maximize the effectiveness of training.
Figure 5. Process of producing spectrograms.
213
Appl. Sci. 2019, 9, 3484
3.4.2. Binarization and Feature Extraction

For a set of spectrograms that obtained from the same rock, the biggest difference between the valid
and invalid spectrograms is the distribution of frequencies. According to this principle, we suggest
binarizing the spectrograms firstly to emphasize their features, as shown in Figure 6. The binarization
threshold of each spectrogram can be determined using Otsu’s method [32]. Otsu’s method entails
dividing an image into two sections (background and object) according to the gray-scale gradient.
The formula of Otsu’s method is as follows:
g = w1 × w2 × (µ1 − µ2 )2 (2)
where w1 is the proportion of the pixels of the object, w2 is the proportion of the pixels of the background,
µμ1 is the mean gray value of the object, and µμ2 is the mean gray value of the background. The optimal
gray threshold is determined by finding out the maximum of g.
Figure 6. Progress of binarization and feature extraction.
Following this, each binarization was divided into 10 equal parts along the vertical axis.
The number of white pixels in every part was calculated, then normalized by dividing them by
the total amount of the white pixels, as shown in Figure 6. In this way, each spectrogram could be
represented by 10 features. For example, the spectrogram shown in Figure 6 can be represented by
[0.1323, 0.07887, 0.0657, 0.0665, 0.0682, 0.0815, 0.1029, 0.1698, 0.1687].
For a set of spectrograms that obtained from the same rock, the features of the valid spectrograms
are similar to each other, while the invalid spectrograms are various. In this study, we assume that the
spectrograms can be divided into three categories with a clustering method, and the categories that
contain the most spectrograms are regarded as valid spectrograms.
3.4.3. Clustering Based on the Modified K-means Algorithm

The K-means algorithm, as a classical clustering algorithm, is famous for its simplicity and strong
clustering ability [33]. As mentioned in Section 3.4.2, we set the value of K as three to divide the
binarization spectrograms into three clusters, meaning that there is a major cluster that contained the
most binarization spectrograms, and the other two clusters represented two extremes that differed
from the major cluster.
However, the K-means algorithm is highly random because the initial seeds are selected randomly,
making it easily plunge into local optima. To solve this problem, the initial seed selection is modified.
The modified K-means algorithm is as follows:
214
Appl. Sci. 2019, 9, 3484
Algorithm Modified K-means Algorithm

Input: Dataset X = {x1 , x2 , . . . , xn }, numbers of clusters N = 3
Output: Clustering result LabelX
1: For i = 1 to n−1
2: For j = i to n
3: Distance(i, j) = the Euclidean distance between xi and xj
4: End
5: End
6: MaxDistance = the maximum of Distance
7: If number(MaxDistance) > 1
8: Select one of the MaxDistances randomly and find out its corresponding xmax_i , xmax_j
9: End
10: Seeds(1) = xmax_i , Seeds(2) = xmax_j , Seeds(3) = Mean(xmax_i , xmax_j )
11: LabelX = Kmeans(X, Seeds)
The modified seed selection method identifies three seeds that have the longest distance from
each other, and can ensure that the differences between the three categories are maximized.
The next step is to determine which categories are valid. The category that contains the most
spectrograms is regarded as a valid category firstly. By experience, the amount of valid spectrograms
occupies approximately more than 85% of the total. Therefore, for the other two categories, if the sum
of them is less than 15%, then the spectrograms of the two categories are determined to be invalid;
if the sum of them is larger than 15%, then only the smaller category is determined to be invalid.
3.5. Prediction Using Machine Learning

Using the re-trained Inception-ResNet-v2 model to determine the surface strength of rocks is
virtually an image classification process. In training the DNN, the inputs are the spectrograms, and the
labels are the values of surface strengths. After training, for a new spectrogram, the DNN can give the
probabilities that the spectrogram belongs to each of the strengths, and classify it into a class according
to the maximum probability, as shown in Table 1.
Table 1. Prediction result of a spectrogram.
Strength 10 20 30 40 50 60 70 80 90 100
Probability 0.0002 0.0000 0.0098 0.0005 0.0000 0.8806 0.1017 0.0004 0.0006 0.0051
It can be seen that the prediction results are discrete. However, surface strength is a continuous
variable. To resolve this problem, the probabilities are regarded as the spectrograms’ features extracted
by the DNN model. Therefore, every spectrogram has 10 features and one label (strength value).
Then, the relationship between the features and the strength values can be fitted with regression
algorithms. The regression algorithms used in this research included the K-Nearest Neighbor [34],
the Support Vector Machine [35,36], and the Random Forest [37].
4. Experiment and Analysis
4.1. Data Collection

Data for re-training the DNN model comprised the surface strengths and hammering sounds.
In this experiment, two batches of data, Batch A and Batch B, were collected, and the rocks in
Batch B were different from the rocks in Batch A. Batch A was used for re-training and conducting the
preliminary tests. Batch B was used for the final test of the method.
The surface strengths of rocks were measured by an N-type rebound device. An N-type rebound
device can measure the rebound values by hitting a rock. As mentioned in Section 2.1, current research
215
Appl. Sci. 2019, 9, 3484
has demonstrated that there is a strong link between the rebound value and the UCS. Therefore, in this
study, the surface strengths of the rocks were represented by their rebound values.
The objects used in the experiment were rocks that existed in the natural environment. Every rock
was intact and at least 0.05 m3 in volume. Rocks were not limited to a particular type. It was not
possible to measure the strength of the whole rock mass due to the rock anisotropy, and what was
measured was just the strength of one point on the rock. To avoid damage to the surface of the rock
caused by measuring one point too many times (especially when the rock was weak), three measuring
points close to each other were set on one rock. Each point was hit five times by the rebound hammer.
In total, there were 15 rebound values for each rock. Then, the surface strength of a rock could be
calculated by filtering the values with the three sigma rule and calculating the mean of the remaining
rebound values. Moreover, before measuring, the weathered layers of the rocks were removed. Table 2
shows the measurements of 15 rock samples A1–A10 (strengths of Batch A) and B1–B5 (strengths of
Batch B). In addition, the rocks included granites, basalts, killas, and andesites. However, the types
and mineral compositions of rocks were not regarded as the influential factors.
The reason why we divided the whole dataset into two batches was that an algorithm (even DNN)
may be sensitive to the data that similar to the training data. The rocks in Batch B were different from
the rocks in Batch A, and the strengths of Batch-B rocks were also different from the Batch-A rocks.
A part of Batch A was for the primary test of our method, and the whole data of Batch B was for the
further verification of the generalization of the method.
Table 2. Rebound values of A and B.
No. Mean Variance No. Mean Variance No. Mean Variance

A1 20 1.12 A6 54 1.56 B1 61 1.88
A2 73 1.13 A7 67 1.18 B2 48 2.37
A3 48 1.50 A8 58 1.72 B3 64 2.39
A4 39 1.73 A9 26 1.00 B4 72 2.43
A5 62 1.15 A10 44 1.88 B5 66 1.92
After measuring the strengths, each rock was hammered 200–260 times with a geological hammer
around the three measuring points. A voice recorder was used to record the hammering sounds.
The hammering rate was approximately one-two times per second, and the hammering force was
slightly varied every 20–30 times to ensure the variability of the hammering force. The voice recorder
had two channels, and its sampling frequency was 24 kHz.
It should be noted that in fact the hammering force mainly affects the amplitude of the hammering
sound (sound level). Therefore, we did not hammer the rocks too hard, but put the voice recorder very
near the hammering points to obtain clear sounds, and in this way prevented (or reduced) damage of
the rock surface. Despite that, some rocks were still damaged in our experiment, and in these cases,
new hammering points near the old ones were selected to continue the experiment. However, for those
rocks of which nearby areas of the measuring points were all damaged, the measuring processes were
terminated immediately.
4.2. Producing Spectrograms

With the script described in Section 3.4.1, 2410 hammering sounds were extracted from the sound
files. The parameters used in STFT were as follows: the size of the window function R = 64, the time
lapse L = 32, the sampling frequency fs = 24 kHz, and the frequency discretization N = 8192.
As shown in Table 2, there are 10 different rebound values in total, each of which corresponded
to 200–260 spectrograms. According to the rebound values, all the spectrograms of Batch A were
assigned to 10 different file folders, and the folders were named by their rebound values. After that,
the modified K-means clustering algorithm described in Section 3.4.3 was used to remove the invalid
spectrograms in every folder. Table 3 shows the filtering result of Batch A.
216
Appl. Sci. 2019, 9, 3484
Table 3. Clustering results of spectrograms.
No. Total Cluster 1 Cluster 2 Cluster 3 Removed Remained

1 201 24 52 125 Cluster 1 177
2 217 85 103 29 Cluster 3 188
3 258 73 16 169 Cluster 2 242
4 239 136 67 36 - 239
5 260 96 18 146 Cluster 2 242
6 232 105 99 28 Cluster 3 204
7 252 10 81 161 Cluster 1 242
8 253 139 20 94 Cluster 2 233
9 259 154 90 15 Cluster 3 250
10 239 2 74 163 Cluster 1 237
Total 2410 - - - - 2254
4.3. Re-Training the DNN

In Batch A, there were 2254 spectrograms. Figure 7 shows some of the spectrograms with
different strengths.
These spectrograms were then used to re-train the Inception-ResNet-v2 model. The parameters
set for training comprised an initial learning rate of 0.002, a learning rate decay of 0.7, an epoch
number of 100, and a batch size of 10. About 80% of the data were used to train the network, and the
remaining 20% were used for validation. The evaluation indicators included the accuracy and the loss.
After training, the structure and weights of the network were determined and were not changed any
further. The training process is illustrated in Figure 8.
Figure 7. Spectrograms with different strengths.
As presented in Figure 8, in the first 1500 iterations, the accuracy grew rapidly. After about 5000
iterations, the accuracy reached 0.9, and then grew slowly. At the end of the training, the accuracy
was 0.945. In the first 3000 iterations, the loss dropped quickly, then held steady between 0.50 and
1.00. We stopped the training at the 18000th step, because the loss showed that it would not be further
decreased, and more training would lead to an over-fitting.
217
Appl. Sci. 2019, 9, 3484
1.00 3.50
0.90
3.00
0.80
Accuracy
Accuracy Loss
Loss
0.70 2.50
Accuracy
0.60
2.00
Loss
0.50
1.50
0.40
0.30 1.00
0.20
0.50
0.10
0.00 0.00
1 1500 3000 4500 6000 7500 9000 10500 12000 13500 15000 16500 18000
Iteration
Figure 8. Re-training of the Inception-ResNet-v2 model.
4.4. Predicting Rock Surface Strength

The training results demonstrate that the fine-tuned DNN can classify the spectrograms with
high accuracy. In the next step, the outputs of the network were regarded as the features of the
spectrograms. Three regression algorithms, including the KNN, the SVM, and the RF, were then tried
to fit the relationship between the features and the strengths.
4.4.1. Predictions of Batch A

First, Batch A was used as the input to test the method. The regression algorithms were used as
follow: (1) in the SVM, the Gaussian kernel function was adopted; (2) in the KNN, the number of N
was set to 20; (3) in the RF, the number of trees was 800, and the maximum depth was nine. Figure 9
displays the regression results.
218
Appl. Sci. 2019, 9, 3484
Figure 9. Predictions of Batch A.
The R2 is used to measure the goodness of fit (GOF). Figure 9a,c, and e show that all the three
algorithms can reach a high R2 of more than 0.95. The R2 of the KNN algorithm is the largest and is
more than 0.98. The norm of residual is another measure of GOF, and a lower norm signifies a better fit.
Among the three algorithms, KNN has the smallest norm of residual. Figure 9b,d,f are the distributions
of the errors of the three algorithms. It also indicates that KNN is more accurate than the other two
algorithms: the range of the errors is within [−20, 20], the errors
− that larger than −2.5 and smaller than
−
2.5 occupy 96.88%, and the errors in the range of [−5, 5] occupy 97.86%. Comprehensively,
− KNN gets
the best performance, and the followings are the SVM and the RF.
4.4.2. Predictions of Batch B

The samples in Batch B had neither participated in the deep network re-training nor the regression
process. Moreover, four out of five of the labels (rebound values) of Batch B were out of the range of
219
Appl. Sci. 2019, 9, 3484
Batch A. The configurations of the regression algorithms were the same as that in Section 4.4.1. Table 4
shows the mean errors and variances between the predictions and the real strength values.
As presented in Table 4, the predictions of the SVM have the minimum mean errors except
for the fourth and the fifth sample sets, and have the minimum variance except for the second set.
The minimum mean error of the 5th samples sets is predicted by KNN. The minimum variance of the
second set is also predicted by KNN. Overall, SVM predictions are the best, the following is KNN,
then RF.
Table 4. Regression results of Batch B.
Mean Errors Variance

Strength
KNN SVM RF KNN SVM RF
48.00 −1.75 −1.11 −1.96 2.19 1.23 2.33
61.00 1.68 1.17 1.49 2.14 2.31 2.41
64.00 −1.75 −0.58 −1.79 6.30 4.66 5.60
66.00 2.44 1.62 1.18 3.15 1.81 3.29
72.00 5.19 6.95 5.73 5.65 4.34 6.47
4.5. Discussion
Based on the results presented in Section 4.4, it can be seen that the predictions of Batch A are
significantly better than the predictions of Batch B. For one thing, the data of Batch A and Batch B
were collected by two groups of researchers, and there might be some non-standard operations when
collecting Batch B—for example, the hitting directions and the hitting speeds when using the rebound
device. Further, the re-trained deep learning network was used for image classification, and there were
only 10 clusters in the training sample. By enlarging the size of training data, the prediction of Batch B
can be improved. For Batch A, the KNN algorithm made the best predictions, meaning that the fitting
accuracy of KNN was the highest. For Batch B, the SVM achieved the best results, meaning that it had
the strongest generalization ability in this regression.
5. Conclusions
In this paper, a new non-destructive measuring method for rock surface strength is presented
based on the DNN technique and spectrogram analysis. The process comprises four steps: (1) collect
hammering sounds and strength data, (2) produce spectrograms of the hammering sounds and
remove invalid spectrograms using a modified K-means algorithm, (3) re-train the Inception-ResNet-v2
model by taking the spectrograms and strength data as the inputs and labels respectively, and (4) use
regression algorithms to make the prediction results continuous. The validation shows that the strengths
(represented by rebound values) of almost all the samples can be predicted within an error of [−5, 5].
Moreover, the combination of the re-trained DNN and KNN has the highest fitting accuracy, and
the combination of the re-trained DNN and SVM has the strongest generalization ability. Therefore,
if the size and the number of clusters of training samples are large, we recommend KNN for regression.
Otherwise, SVM is recommended.
The proposed method was accomplished using Python and Tensorflow programming; however,
it should be noted that the contribution of this research is not about a new deep neural network, and both
the Inception-ResNet-v2 model and transfer learning are well-established techniques. This research is
an application of them, and is aimed at presenting an effective and simple method for field survey.
Overall, the proposed method offers great potential in supporting the implementation for efficient rock
strength measurement methods in the field.
A noted limitation of this research is related to the use of a rebound device to capture the raw
rock strength data, under experimental conditions. The precision of the rebound method is low, and
the measurement is easily affected by hitting directions and hitting speed. In the subsequent work,
220
Appl. Sci. 2019, 9, 3484
higher precision techniques, such as acoustic emission techniques, will be considered to optimize the
performance of the method further.
Author Contributions: S.H. collected and analyzed the data, wrote the code and the paper; H.L. gave professional
geological guidance; M.L. provided the idea and edited the manuscript; T.R. gave professional geological guidance.
Funding: This research was funded by the National Natural Science Foundation for Excellent Young Scientists of
China (Grant no. 51622904), the Tianjin Science Foundation for Distinguished Young Scientists of China (Grant no.
17JCJQJC44000), and the Research Grants Council of Hong Kong (PolyU 152093/14E).
References
1. Quinones, J.; Arzúa, J.; Alejano, L.R.; García-Bastante, F.; Ivars, D.M.; Walton, G. Analysis of size effects on
the geomechanical parameters of intact granite samples under unconfined conditions. Acta Geotech. 2017, 12,
1229–1242. [CrossRef]
2. Selçuk, L.; Yabalak, E. Evaluation of the ratio between uniaxial compressive strength and Schmidt hammer
rebound number and its effectiveness in predicting rock strength. Nondestruct. Test. Eval. 2015, 30, 1–12.
[CrossRef]
3. Hannun, A.; Case, C.; Casper, J.; Catanzaro, B.; Diamos, G.; Elsen, E.; Prenger, R.; Satheesh, S.; Sengupta, S.;
Coates, A. Deep speech: Scaling up end-to-end speech recognition. arXiv 2014, arXiv:1412.5567.
4. Zhou, B.; Lapedriza, A.; Xiao, J.; Torralba, A.; Oliva, A. Learning deep features for scene recognition using
places database. In Proceedings of the Advances in Neural Information Processing Systems (NIPS), Montreal,
Canada, 8–13 December 2014; MIT Press: Cambridge, MA, USA; pp. 487–495.
5. Cireşan, D.C.; Meier, U.; Schmidhuber, J. Transfer learning for Latin and Chinese characters with deep neural
networks. In Proceedings of the 2012 International Joint Conference on Neural Networks (IJCNN), Brisbane,
Australia, 10–15 June 2012; pp. 1–6.
6. Zhou, J.T.; Pan, S.J.; Tsang, I.W.; Yan, Y. Hybrid Heterogeneous Transfer Learning through Deep Learning.
In Proceedings of the 28th AAAI Conference on Artificial Intelligence, Québec City, Québec, Canada,
27–31 July 2014; pp. 2213–2220.
7. Noori Hoshyar, A.; Rashidi, M.; Liyanapathirana, R.; Samali, B. Algorithm Development for the
non-destructive testing of structural damage. Appl. Sci. 2019, 9, 2810. [CrossRef]
8. Haramy, K.; DeMarco, M. Use of the Schmidt hammer for rock and coal testing. In Proceedings of the 26th
US Symposium on Rock Mechanics (USRMS), American Rock Mechanics Association, Rapid City, SD, USA,
26–28 June 1985; pp. 549–555.
9. McNally, G. Estimation of coal measures rock strength using sonic and neutron logs. Geoexploration 1987, 24,
10. Goudie, A.S. The Schmidt Hammer in geomorphological research. Prog. Phys. Geogr. 2006, 30, 703–718.
[CrossRef]
11. Yaşar, E.; Erdoğan, Y. Estimation of rock physicomechanical properties using hardness methods. Eng. Geol.
2004, 71, 281–288. [CrossRef]
12. Aoki, H.; Matsukura, Y. A new technique for non-destructive field measurement of rock-surface strength:
An application of the Equotip hardness tester to weathering studies. Earth Surf. Process. Landf. 2007, 32,
1759–1769. [CrossRef]
13. Lai, G.T.; Mazlan, N.A.; Nadzir, M.S.M.; Rafek, A.G.; Serasa, A.S.; Hussin, A.; Ern, L.K.; Yeok, F.S. Uniaxial
Compressive Strength of Antarctic Peninsula Rocks: Schmidt Hammer Rebound Test. Sains Malays. 2017, 46,
677–684.
14. Sharma, P.; Singh, T. A correlation between P-wave velocity, impact strength index, slake durability index
and uniaxial compressive strength. Bull. Eng. Geol. Environ. 2008, 67, 17–22. [CrossRef]
15. Tziallas, G.P.; Tsiambaos, G.; Saroglou, H. Determination of rock strength and deformability of intact rocks.
Electron. J. Geotech. Eng. 2009, 14, e12.
16. Liu, Z.; Shao, J.; Xu, W.; Wu, Q. Indirect estimation of unconfined compressive strength of carbonate rocks
using extreme learning machine. Acta Geotech. 2015, 10, 651–663. [CrossRef]
221
Appl. Sci. 2019, 9, 3484
17. Son, M.; Kim, M. Estimation of the Compressive Strength of Intact Rock Using Non-Destructive Testing
Method Based on Total Sound-Signal Energy. Geotech. Test. J. 2017, 40, 643–657. [CrossRef]
18. Azimian, A. Application of statistical methods for predicting uniaxial compressive strength of limestone
rocks using nondestructive tests. Acta Geotech. 2017, 12, 321–333. [CrossRef]
19. Palafox, L.F.; Hamilton, C.W.; Scheidt, S.P.; Alvarez, A.M. Automated detection of geological landforms on
Mars using Convolutional Neural Networks. Comput Geosci. 2017, 101, 48–56. [CrossRef]
20. Xu, G.; Zhu, X.; Fu, D.; Dong, J.; Xiao, X. Automatic land cover classification of geo-tagged field photos by
deep learning. Environ. Model. Softw. 2017, 91, 127–134. [CrossRef]
21. Sidahmed, M.; Roy, A.; Sayed, A. Streamline Rock Facies Classification with Deep Learning Cognitive Process.
In Proceedings of the SPE Annual Technical Conference and Exhibition, Society of Petroleum Engineers,
San Antonio, TX, USA, 9–11 October 2017; pp. 1–11.
22. Yu, H.; Ma, Y.; Wang, L.; Zhai, Y.; Wang, X. A landslide intelligent detection method based on CNN and
RSG_R. In Proceedings of the Mechatronics and Automation (ICMA), Takamatsu, Japan, 6–9 August 2017;
pp. 40–44.
23. Carneiro, G.; Nascimento, J.; Bradley, A.P. Unregistered multiview mammogram analysis with pre-trained
deep learning models. In Proceedings of the Medical Image Computing and Computer-Assisted Intervention,
Munich, Germany, 5–9 October 2015; pp. 652–660.
24. Schwarz, M.; Schulz, H.; Behnke, S. RGB-D object recognition and pose estimation based on pre-trained
convolutional neural network features. In Proceedings of the Robotics and Automation, Seattle, WA, USA,
26–30 May 2015; pp. 1329–1335.
25. Li, N.; Hao, H.; Gu, Q.; Wang, D.; Hu, X. A transfer learning method for automatic identification of sandstone
microscopic images. Comput. Geosci. 2017, 103, 111–121. [CrossRef]
26. Zhang, Y.; Li, M.; Han, S. Automatic identification and classification in lithology based on deep learning in
rock images. Acta Petrol. Sin. 2018, 34, 333–342.
27. Chi, J.; Walia, E.; Babyn, P.; Wang, J.; Groot, G.; Eramian, M. Thyroid nodule classification in ultrasound
images by fine-tuning deep convolutional neural network. J. Digit. Imagin. 2017, 30, 477–486. [CrossRef]
28. Szegedy, C.; Vanhoucke, V.; Ioffe, S.; Shlens, J.; Wojna, Z. Rethinking the inception architecture for computer
vision. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Seattle, WA,
USA, 27–30 November 2017; pp. 2818–2826.
29. He, K.; Zhang, X.; Ren, S.; Sun, J. Deep residual learning for image recognition. In Proceedings of the
IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, 26 June–1 July 2016;
pp. 770–778.
30. Allen, J. Short term spectral analysis, synthesis, and modification by discrete Fourier transform. IEEE Trans.
Acoust. Speech Signal Process. 1977, 25, 235–238. [CrossRef]
31. Gelfand, S.A. Essentials of Audiology, 2nd ed.; Thieme: New York, NY, USA, 2001.
32. Otsu, N. A threshold selection method from gray-level histograms. IEEE Trans. Syst. Man. Cybern. 1979, 9,
62–66. [CrossRef]
33. Li, M.; Han, S.; Shi, J. An enhanced ISODATA algorithm for recognizing multiple electric appliances from
the aggregated power consumption dataset. Energ. Build. 2017, 140, 305–316. [CrossRef]
34. Li, M.; Miao, L.; Shi, J. Analyzing heating equipment’s operations based on measured data. Energ. Build
2014, 82, 47–56. [CrossRef]
35. Hearst, M.A.; Dumais, S.T.; Osuna, E.; Platt, J.; Scholkopf, B. Support vector machines. IEEE Intell. Syst.
Their Applic. 1998, 13, 18–28. [CrossRef]
36. Ballabio, C.; Sterlacchini, S. Support vector machines for landslide susceptibility mapping: The Staffora River
Basin case study, Italy. Math. Geosci. 2012, 44, 47–70. [CrossRef]
37. Breiman, L. Random forests. Mach. Learn. 2001, 45, 5–32. [CrossRef]
222
applied
sciences
Article
Detecting the Void behind the Tunnel Lining by
Impact-Echo Methods with Different Signal
Analysis Approaches
Rongning Cao 1,2 , Meng Ma 1,2, * , Ruihua Liang 1,2 and Chao Niu 3
1 Key Laboratory of Urban Underground Engineering of Ministry of Education, Beijing Jiaotong University,
Beijing 100044, China
2 School of Civil Engineering, Beijing Jiaotong University, Beijing 100044, China
3 CCCC Highway Consultants Co., Ltd., Beijing 100088, China

Received: 4 July 2019; Accepted: 8 August 2019; Published: 10 August 2019
Abstract: A void behind the lining in a tunnel is considered to be a critical condition as it can
significantly impair the tunnel service life. In this study, we adopted the impact-echo (IE) method to
detect the voids. We designed two test conditions (tunnel lining with and without a void) for our
experiments performed in a laboratory environment. The influences of void size and impact-void
position were analysed using numerical simulations. The vibration response signals were analysed
in the time, frequency, and time–frequency domains using various signal analysis approaches. The
results were comparatively analysed to determine the best approach for void detection. The study
helped establish that a tunnel void can be evaluated through the vibration energy (amplitude and
duration) in the time domain, the resonance frequency and dynamic stiffness in the frequency domain,
and the energy distribution in time–frequency domain. The wavelet transform analysis is the most
appropriate method to observe the energy flow during the state changing and the dynamic stiffness
method can determine the void position precisely.
Keywords: non-destructive test; impact-echo method; tunnel void; experimental model; numerical
simulation; signal analysis
1. Introduction
With the rapid development of railway and highway constructions, the number of tunnels increases,
nevertheless, because of the complicated geological conditions and inappropriate constructions, the
tunnel diseases such as the lining cracks, tunnel leakage, vault roof slab, voids behind the lining, etc.
are also increasing. Particularly, a void located behind the lining in a tunnel may cause leakage and
deformation of the surrounding rocks that can significantly impair the tunnel service life.
The non-destructive testing (NDT) methods commonly adopted for the investigation of a
tunnel void includes ground-penetrating radar (GPR) [1,2], ultrasonic resilience method [3], infrared
photographing technology of temperature field [4], etc. Similarly, impact-echo (IE) is one another
widely used NDT method to detect the piles [5] and other engineering structures [6], and it is also
employed to detect the voids behind a tunnel lining. This method is especially known for locating the
voids in the very early stage of evaluation whereby the stress waves get reflected between the interfaces;
accordingly, the IE method is based on the analysis of the vibration response signal. Ni et al. [7]
assessed the quality of long drilled piles, such as the pile length, by analysing the test results in time
domain and time–frequency domain. Cheng et al. [8] used the transfer function simulated by the IE
method to analyse the bond condition of the concrete and substrate layers. The effect of the acoustic
223
Appl. Sci. 2019, 9, 3280
impedance and thickness of the substrate layer on the simulated transfer function was analysed in
the frequency domain. Aggelis et al. [9] reported the effectiveness of grouting through a study that
combined the time domain characteristics, spectral content, and wavelet transform. Song and Cho [10]
analysed the response signals in the time–frequency domains through short-time Fourier transform
(STFT) to evaluate the bonding state of tunnel shotcrete. They also studied the effects of the ground
types, thickness of the shotcrete, undulating surfaces, and impact sources in both frequency and
time–frequency domains [11]. Davis et al. [12] calculated the dynamic stiffness, mobility and damping,
and peak/mean mobility ratio of the tested concrete elements to evaluate the tunnel lining grouting.
Ryden et al. [13] evaluated the quality of backfill material in the segmental lining of a tunnel based on
the IE amplification factor (Q-factor).
In a previous study, the IE method was used to study the grouting quality, the tunnel shotcrete
bonding state, the material quality, etc. The response signals derived from IE method were analysed
in the time, frequency, time–frequency domains using various methods such as Q-factor, dynamic
stiffness, STFT, etc. The methods of analysis and the aspects of the response signal were selected based
on the quality or bonding state of the material under study. Various analysis methods were used
to analyse the properties of the data obtained by the IE method. They are usually compared in the
situation of detecting the bonding state or the material quality. However, the tunnel structure that we
studied in this article is in a complex loaded situation compared with the structure mentioned above
such as the concrete slab. Accordingly, in order to study the specific aspects of the tunnel lining and
to select an appropriate method of analysis to detect the void behind tunnel lining, it is necessary to
apply various methods and analyse the test results. The properties and applicability of these methods
should be compared.
In this paper, to compare different types of signal analysis method for detection of tunnel voids,
an experimental study was first performed to compare two conditions, tunnel lining with and without
voids. The responses were then analysed in the time, frequency and time–frequency domains using
several methods that include fractal box-counting dimension for time history analysis, dynamic
stiffness, STFT, wavelet analysis, and Hilbert Huang transform (HHT). Then, the influences of void
size and impact-void position were calculated using numerical simulations. It was ensured that the
best methods were used for numerical analysis.
2. Experimental Study
2.1. Experimental Outline

Impact echo method is based on the stress wave that is excited by an impact force. During the
impact echo test, an impact force is applied on the surface of the structure by a hammer. The generated
stress waves which include the body waves (P wave and S wave) and the surface wave (Rayleigh
wave) propagate in the structure. Reflections and refractions happen at the interface of two media
with different wave impedance. The reflectional signals are then collected by the sensor adjacent to the
impact point. The P wave propagates faster than other types of waves, and if the transducer is placed
close to the impact point, the response is dominated by P-wave echoes [14]. The signals are recorded
and restored by the data acquisition instrument and are analysed in the computer. The response signals
are analysed in the time domain, the frequency domain and the time–frequency domain. Finally, the
property of the structure, such as the location of void and the thickness of the structure [6], can be
detected from the patterns of waveform versus time and spectra versus frequency. The test flew is
shown in Figure 1.
224
Appl. Sci. 2019, 9, 3280
Figure 1. Test flow chart.
When the IE method is used to detect the thickness of a slab structure, the excited P wave reflects
in the slab structure repetitively, which forms a periodic modal related to the thickness of the slab. In
turn, the resonance frequency corresponding to the thickness occurs. Therefore, the thickness of a slab
structure can be detected by the thickness frequency according to the IE method. When there is a
void in the slab, the excited frequency needs to be higher than the thickness frequency. For example,
if the velocity of the P wave is 4000 m/s and the depth of the void is 0.2 m, the resonance frequency
is calculated as about 10 kHz, which should be contained in the excited frequency range. Therefore,
when a void is inside the slab structure, another resonance frequency corresponding to the void can be
seen except for the resonance frequency related to the thickness. Then the void can be detected by the
IE method by observing the resonance frequencies.
In the situation of detecting voids behind the tunnel lining, the resonance frequencies of the
conditions with and without void will be the same, which are both related to the thickness of the tunnel
lining. Thus, the void cannot be detected by the difference in the resonance frequency. However, a
void can be recognised by IE method due to different responses caused by different reflective ratios.
The reflective ratios are caused by the distinct difference in the wave impedance among tunnel lining,
soil and air. The wave impedance Z is expressed as:
Z = ρVP , (1)
P
where ρ is density and VP is P wave velocity. For the P waves propagate from structure A to structure
ρ
B, the reflective ratio is expressed
P as:
Ar Z − ZA
R= = B , (2)
Ai r ZBB + ZAA
i B A
where Ar is the reflected amplitude and Ai is the incident amplitude. When ZA is much larger than ZB ,
the reflected
Ar ratio R is closed to −1, which means
Ai the amplitude of the reflected waveAremains the same,
while the phase of the reflected changes. This situation happens at the interface of tunnel lining and
ZB −
air, where almost all of the energy is reflected. When ZA is approximately equal to ZB , the reflected
ratio R is closed to 0, which means most of the energy propagates through the interface. This situation
Z A soil.
is similar to the condition that the tunnel lining is tightly surrounded by the
Z BThis section presents the design of a laboratory principle experiment (Figure 2), whereby different
signal analysis approaches were then employed to compare the vibration responses for the conditions
A and B.
225
Appl. Sci. 2019, 9, 3280
(a) Condition A (b) Condition B

Figure 2. The basic principle of detecting the void behind tunnel lining by impact-echo (IE) method.
The experiment was carried out in a brick box of size 3.6 × 0.94 × 1.44 m3 (Figure 3a). A rockwool
material with a thickness of 0.1 m was placed inside the bunker to weaken the wave reflection at the
boundaries. The soil was filled into the bunker. The horseshoe tunnel models were buried in the soil.
The cross-sectional size of the tunnel model was 1:16 of a real railway tunnel and the thickness of the
tunnel model was 0.025 m (Figure 3b). The tunnel models were made of polyvinyl chloride (PVC)
material instead of the concrete. The material coefficients of the soil, the concrete, the PVC, and the
air are shown in Table 1. The wave impedance of the PVC and the concrete is within one order of
magnitude, which is much larger than the wave impedance of the air. This illustrates that the concrete
tunnel can be instead by the PVC, a cheaper material, in this principle experiment. Two different
conditions (A and B) were designed. Condition A referred to a tight contact between the tunnel lining
−
and the surrounding soils. The reflected ratio RA at the interface is −0.31. For Condition B, a void
behind the tunnel wall was modelled using a pre-embedded empty box of 285 × 145 × 155 mm3 placed
between the lining and the soil. The reflected ratio RB at the interface is −−1.
(a) Layout of the experiment table (unit: mm)
(b) The tunnel model (c) Accelerometer arrangement
Figure 3. The test system.
13
400
226
12
300 11
/(N/Hz)
10
200
e/N
Appl. Sci. 2019, 9, 3280
Table 1. Material parameters used in the principle experiment.
Material Density ρ/(kg/m3 ) P Wave Velocity V p /(m/s) Wave Impedance Z/(kg/m2 s)

Soil 2000 576 1.1 × 106
Concrete (C30) 2400 3691 8.8 × 106
PVC 1380 1500 2.1 × 106
Air 1.2 343 411.6
The test system consisted of a hand hammer, two unidirectional accelerometers, a data acquisition
instrument with 16 channels, and a computer. The hand hammer, consisted of a hammer head, a
force sensor and a hammer body, was used to apply an impact force on the tunnel model. The type
of the hammer is Dytran Dytranpulse™5800B4T, developed by the Dytran Instruments company in
California, USA. The head of the hammer is made by aluminium with the weight of 100 g and Young’s
modulus of 70 GPa. There is a special acceleration compensated piezoelectric force sensor in the
hammer head at the striking face, ensuring a smooth frequency spectrum, which is used to measure
the value of the impact force. The hammer has a sensitivity of 10 mV/lbf and a maximum force of
1000 lbf. The upper limit of the frequency range is 75 kHz. The type of unidirectional accelerometers is
LC0104, developed by LANCE in Ohio, USA. They are used to collect the accelerations perpendicular
to the tunnel lining, which has a sample frequency range of 0.5 Hz-9000 Hz, a sensitivity of 100 mV/g,
a resolution of 0.0002 g, and an acceleration range of 50g. The type of data acquisition instrument is
INV3018C, developed by the China Orient Institute of Noise & Vibration (COINV).
The testing point B was arranged at the tunnel lining behind the empty box on the right tunnel
model, while testing point A was arranged at the same location on the left tunnel model, as shown
in Figure 3a. According to the thickness and the P wave velocity of the tunnel lining in the test, the
thickness resonance frequency is obtained approximately at 28,800 Hz. Since the thickness frequency
of the two conditions is the same, this experiment of detecting the void behind tunnel lining does not
intend to focus on the amplitude of resonance frequency, but to analyse the dynamic responses of these
two conditions in a lower frequency range. The sampling frequency of the accelerometers is set at
2560 Hz. The impact points are 2 cm nearer to the testing points. For each condition, the impact was
repeated 10 times. For each time, the peak of impact force is controlled to be close to 400 N, and the
sampling frequency is 2560 Hz, as shown in Figure 4. It should be noted that the hammer should be
controlled to be perpendicular to the surface of the tested object. To obtain this goal, the coherence
function of the response and the force was calculated in each test. If the coherence function was close
to 1, the impulse force would be qualified. Otherwise, the data would be removed.
13
400
12
300 11
Amplitude/(N/Hz)
10
200
Force/N
9
100
8
0 7
6
-100
5
0.00 0.01 0.02 0.03 200 400 600 800 1000 1200
Time/s Frequency/Hz
(a) (b)
Figure 4. (a) Time history and (b) spectrum of the impact force.
227
Appl. Sci. 2019, 9, 3280
2.2. Time Domain Analysis

Although the impact forces generated by the hand hammer are controlled within a range, the
signals still need to be normalized in order to compare the responses of different conditions. Each
impact force and its related response need to multiply by a coefficient Ci , which is obtained as:
400
400
Ci = ,
peak
peak
(3)
Fi
peak
peak
where Fi is the peak value of the i-th test. Figure 5 shows the time-domain waveforms of the two
conditions, with an intercepted time duration of 0–50 ms. For Condition A, the waveform attenuates
smoothly with the amplitude almost decaying to zero in 10 ms. For condition B, an obvious secondary
peak occurs at 7 ms, while the minor peaks continuing until 15 ms. This phenomenon of a time-domain
waveform with a stronger reflection energy and a prolonged response is suggestive of the existence of
a void. It indicates that the stress waves reflect at the interface between the tunnel and the void, and
the reflective ratio of concrete–air interface is higher than that of concrete–soil interface.
40
30 Condition A
Acceleration/g
20
10
0
-10
-20
-30
0 5 10 15 20 25 30 35 40 45 50
40 Time/ms
30 Condition B
Acceleration/g
20
10
0
-10
-20
-30
0 5 10 15 20 25 30 35 40 45 50
Time/ms
Figure 5. Time histories of the two conditions (A and B).
The vibration response can be analysed quantitatively using the fractal box-counting dimension
method. The concept of fractal dimension was proposed by Mandelbrot [15]. The vibration signal
is irregular but self-similar. It occupies a larger area than the one-dimensional line, smaller area
than the two-dimensional plane. Its fractal dimension is measured by creating multiscale covers
around the signal graph. Square boxes with side length (δ) are used to cover the signal waves on the
time-amplitude plane. Then, the number of covered δ boxes (N) can be obtained (Figure 6). The fractal
dimension is defined as:
lgN
D = −lim , (4)
δ→0 lgδ
lg
lim lg
0
δ δ
⋯ δ
⋯ δ
∆
− δ−
228
Appl. Sci. 2019, 9, 3280
Figure 6. Number of boxes with side length δ.

δ
For the measured response signals in this paper, square boxes of a series of side lengths (δ1 , δ2 , · · · ,
δn ) were used to cover the signal waves on the time-amplitude plane, and then a series of covered box
quantities14 (N 1 , N 2 , · · · , Nn ) would be obtained. The minimum
14 δ should be larger than the sampling
interval 12∆t, because the signal wave was a straight line on 12every sampling interval. After calculating
the fitting line of (−lgδ − lgN), the fractal dimension would10be obtained as the slope of the fitting line
10
δ
(Figure 7).8 8
lg
lg
6 6
4 14 4 14
2 12 2 12
0 10 0 10
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1
8 -lg 8 -lg
lg N
lg N
6 6
4 4
− δ−
2 2
0 0
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1
-lg δ -lg δ
− δ−−lgN).
Figure 7. The fitting lines of (−lgδ
The results show that the fractal dimensions of the conditions A and B are 1.221 and 1.288,
respectively. This illustrates that when the waveforms of the two conditions are in the coordinate
system, with the same time scale and the same amplitude scale, the waveform B is more complicated.
Therefore, when the void exists, the fractal dimension is larger, and the waveform in the time domain
is more complicated.
2.3. Frequency Domain Analysis

Figure 8 shows the frequency spectra of ten test runs conducted on each of the two conditions. In
condition A, all the spectra are similar, and a dominant frequency of approximately 690 Hz is observed.
The amplitudes increase steadily between 0 and 500 Hz. In condition B, larger differences among the
tested spectra are observed. The dominant frequency remains the same (~690 Hz) as that of condition
A. Besides, the amplitude fluctuates between 0 and 500 Hz with two additional peaks between 300 and
500 Hz. For each condition, the average frequency spectra of these test runs are calculated, except for
the incoherence test runs (test 1 of condition A and test 4 for condition B), as shown in Figure 8c. The
229
Appl. Sci. 2019, 9, 3280
amplitude of condition B is higher than that of condition A in the frequency range of 800 Hz–1100 Hz.
There are more dominant frequencies in condition B.
0.9 0.9
Test 1 Test 1
0.8 Test 2
0.8 Test 2
Test 3 Test 3
Amplitude/(g/Hz)
Amplitude/(g/Hz)
0.7 Test 4 0.7 Test 4

Test 5 Test 5
0.6 Test 6 0.6 690Hz
Test 6
0.5
Test 7 690Hz Test 7
Test 8 0.5
Test 8
0.4 Test 9 Test 9
Test 10
0.4
Test 10
0.3 Average 0.3 Average
0.2 0.2
415Hz
0.1 0.1
0.0 0.0 305Hz 500Hz
0 200 400 600 800 1000 1200 1400 0 200 400 600 800 1000 1200 1400
Frequency/Hz Frequency/Hz
(a) 10-time experiments data of condition A (b) 10-time experiments data of condition B
0.5 Condition A
Condition B
Amplitude/(g/Hz)
0.4
0.3
0.2
0.1
0.0
0 200 400 600 800 1000 1200 1400
Frequency/Hz
(c) Comparison of the average data for the two conditions
Figure 8. Fourier spectra for two conditions.
The dynamic stiffness was also used to study the influential differences between the two conditions.
It expresses the resistance to deformation under the impact force [16]. In this test, dynamic stiffness
reflects the supporting performance of the tunnel-soil structure. The impact force F(t) and the obtained
acceleration response a(t) can be transformed into the frequency domain, using Fourier transform, as
F(f ) and a(f ), respectively. The acceleration response is expressed as the superposition of a series of
simple harmonic vibrations, so the amplitude of velocity can be represented as:
( )
) a( f ) ,
V ( f() = (5)
2 ff
2π
The velocity admittance is defined as:
V ( (f ) )
Gv =
v , (6)
F( f( ) )
Then, the dynamic stiffness can be calculated by:

2 2 ( )
2π f 2π f · F( f )
Kd =
d = , (7)
Gv v V ((f ) )
Figure 9 compares the Kd of the two conditions. The Kd in condition B is lower than condition
A. The reason is that the wave propagation in condition B is approximated as guided by a free-free
boundary condition while the wave in condition A is guided by a free-constrained boundary condition.
This phenomenon shows that the void existence can significantly reduce the dynamic stiffness in most
frequencies, which means it reduces the ability of the tunnel structure to resist the load.
230
Appl. Sci. 2019, 9, 3280
8
1.0x10 Condition A
Condition B
Dynamic stiffness/(N/m)
7
8.0x10
7
6.0x10
7
4.0x10
7
2.0x10
0.0
0 200 400 600 800 1000
Frequency/Hz
Figure 9. Dynamic stiffness analysis.
2.4. Time–Frequency Analysis

The frequency-domain analysis is widely used to research the frequency distribution of the signals.
However, it is only suitable to the stationary process, not for the non-stationary process. In other
words, the frequency domain analysis illustrates the frequency components of the signals but fails to
figure out the moment that a certain frequency component appears.
The time–frequency analysis expresses the signal as a function of both time and frequency using
the local transform. It helps understand changes in energy with both frequency and time upon the
occurrence of a certain frequency. The methods usually adopted for time–frequency analysis are
short-time Fourier transform, wavelet transform, Hilbert Huang transform, and so on.
2.4.1. Short-Time Fourier Transform (STFT)

The STFT is used to describe the frequency spectra of local signal sections changing over time.
The original signal is first divided into several sections with equal time length by an appropriate
sliding time window function η(t). These sections are approximated as stationary processes. Then, by
applying the Fourier transform from ( )t to f to each section, the frequency spectra vary with time are
obtained. The STFT can be expressed as [17]:
Z ∞
′
STFTz (t, f ) = z(t′ )η∗ (t'′ −* t)e' −j2π f t 2dt′ , (8)
STFT ( ,−∞)  ( ) (
'
'
) d
where z(t′ ) is( 'the

) original signal, η(t) is the
( ) window function centered around zero.
The width of the time window should be properly selected. For example, if a wide time window
is selected, the frequency resolution will increase, but the time resolution will decrease, which means
these frequency components can only be determined to appear in this wide time band, not a certain
instantaneous moment.
Figure 10 shows 3-dimensional (3D) graphs of the two conditions calculated by the STFT. For
condition A, three energy peaks can be observed at approximately 700, 900, and 1200 Hz. For condition
B, the number of energy peaks increases, and these peaks are all higher than those of condition A.
Among them, the energy peak at approximately 700 Hz is the highest. On the low-frequency band
between 0 and 100 Hz, the energy distributes widely along the time axis.
231
Appl. Sci. 2019, 9, 3280

Figure 10. Time–frequency spectra by STFT.
In summary, when the void exits, the vibration energy becomes stronger, which is indicative of
increased reflections of the stress waves from the tunnel-air interface. In addition, more energy moves
to the frequency of approximately 700 Hz, which in turn becomes the dominant frequency. It can also
be seen that the low-frequency energy attenuates slowly when the void exists.
2.4.2. Wavelet Transform 1

WTz ( )
Wavelet transform uses the mother wavelet function
R 
z( ) ( )d
to express the original signal, which can be
expressed as: Z
1 t−b
() WTz (a, (b) = )√ z(t)ϕ( )dt, (9)
a R a
where z(t) is the original signal. The ϕ( t−b a ) is the mother wavelet function. The a and b are the
scaling and translating parameters, respectively. The scaling parameter represents the dilation or
compression of the mother wavelet, and the translating parameter determines the mother wavelet
moving along the time domain. The dilative mother wavelet corresponds to low-frequency resolution.
()
The compressive mother wavelet corresponds to high-frequency resolution [18]. By calculating the
cross-correlation of the original signal z(t) and the mother wavelet function under different scales, the
frequency components are obtained. The moment that a certain frequency appears is obtained by
moving the mother wavelet along the time domain.
For the STFT, when the width of the time window is determined, the resolution of time and
frequency on the whole time–frequency range is constant. However, in signal analysing, the
high-frequency resolution is needed on the low-frequency range, while the high-time resolution
is needed on the high-frequency range. On this aspect, the wavelet transform meets the requirements
more closely.
Figure 11 shows the contour maps of the two conditions analysed by Cmor wavelet transform.
Except for the characters that are shown in Section 2.4.1 by STFT, new phenomena are found in
Figure 11. When the void exists, the frequency components between 1000 Hz and 1200 Hz appears
from 0 s to 0.005 s, which is the period of the primary peak in Figure 5. In addition, the frequency
components above 1200 Hz occurs from 0.005 s to 0.01 s, which is the period of the secondary peak in
Figure 5.
232
Appl. Sci. 2019, 9, 3280
Figure 11. Time–frequency spectra by wavelet transform.
For the STFT, it could only be roughly determined when the frequency components appear
because of the low time resolution. However, for the wavelet transform, the moment that a certain
frequency component appears could be clearly figured out. This means that the wavelet transform can
tell more detail about the characters.
2.4.3. Hilbert Huang Transform (HHT)

The signal can also be presented in the time–frequency domain using the Hilbert Huang Transform
(HHT). The HHT decomposes a signal into intrinsic mode functions (IMFs) along with a trend, and then
obtains instantaneous frequency data. Compared to other transforms, HHT is more like an algorithm
(an empirical approach) that can be applied to a data set, rather than a theoretical tool. The calculation
of the HHT includes two steps: empirical mode decomposition (EMD) and Hilbert transform. By
means of EMD, the acceleration data can be decomposed into different simple non-sinusoidal signals,
called IMFs [19]. Each IMF is transformed into a time–frequency domain by Hilbert transform, and
then the time–frequency spectrum can be obtained after the superposition.
Figure 12 shows the time–frequency spectra of the signals analysed by HHT. The original signals
were separated into several IMFs, and the first step of IMF was in the range of 600–1200 Hz. According
to the previous analysis, the study would focus on the first-step IMF (600–1200 Hz). For condition A,
the vibration energy peak is at 700 Hz and 5 ms, and the energy attenuates significantly after 5 ms,
almost approaching zero at 10 ms. For condition B, there are two energy peaks, of which one appears
at 1100 Hz and 3.5 ms, while the other one is at 700 Hz and 6 ms. The energy remains high until 10 ms.
Figure 12. The time–frequency spectrum.
233
Appl. Sci. 2019, 9, 3280
In summary, when the void exists, another vibration energy peak of 1100 Hz appears at 3.5 ms
that is indicative of the reflection of the stress waves at the tunnel-air interface. This phenomenon also
explains that the stress waves take longer time to propagate to the soil, which causes the energy peak
of 700 Hz to appear 1 ms later in condition B. In addition, the vibration lasts longer when a void exists.
By means of HHT, a signal was separated into several sub-signals dominant in different frequency
ranges. As the key frequency band for the sub-signals could be studied individually, the stress wave
propagation and the frequency component distribution could be seen clearly, however, it is less holistic
because of the separation of the original signal.
The STFT and wavelet transform analysis to focus on the overall vibration energy in a
time–frequency plane that could be observed clearly. Similarly, the energy flow between different
conditions could also be studied clearly. In principle, unlike the STFT, the wavelet transform can
change the scale while analysing different frequencies of the signal. As a result, both high-time and
high-frequency resolution can be achieved using wavelet a transform. Based on these aspects, the
wavelet analysis method would be chosen for the following numerical study to analyse the signals in
time–frequency domains.
3. Numerical Study
The laboratory experiment compared the vibration responses of the two conditions (with and
without voids) using IE method. The approach proved to be more satisfactory in the experimental
study was employed in the signal analysis using the numerical model, whereby the size and the
position of the void were considered.
3.1. Numerical Model

A finite element (FE) model considering the soil, tunnel, and void on the soil-structure interface
was built in Midas GTS, shown as Figure 13. The material parameters are shown in Table 2. The
coordinate system based on the right-hand rule was shown, and the dimensions of the model were 84
× 84 × 40 m3 . The external diameter of the tunnel was 6 m, and the thickness of the tunnel lining was
0.3 m. In order to simulate the wave shape, the size of the mesh needs to be divided smaller than 1/8 of
the smallest wavelength, which is expressed as:
Vs
∆l ≤ , (10)
8 fmax
where V s is the shear wave velocity and fmax is the highest analysis frequency, which is 150 Hz in
this model. According to the equation, the element of soil should be smaller than 0.25 m and the
element of tunnel lining should be smaller than 1.88 m. In this model, the tunnel elements were set
at 0.3 m. The soil elements near to the top of the tunnel were set as 0.2 m, and the size was getting
bigger as the elements spread from the centre to the boundary of the model. The viscoelastic artificial
boundaries were applied to the model to simulate the infinite soil. The viscoelastic artificial boundary
is a continuous distributed parallel spring-damping system with calculated stiffness and damping
factors. Once these coefficients are input into the Midas GTS software, the equivalent spring-damping
elements will be applied on the boundary nodes.
234
Appl. Sci. 2019, 9, 3280
(a) Cross section (b) Longitudinal profile
Figure 13. Tunnel-soil FE model.
Table 2. Material parameters used in the numerical models.
Young’s Compressive Shear Wave

Poisson’s Density
Material Modulus Wave Velocity Velocity
Ratio v ρ/(kg/m3 )
E/(kPa) V p /(m/s) V s /(m/s)
Soil 475,000 0.32 2000 576 297
Tunnel lining 30,000,000 0.20 2400 3691 2260
In the experimental study, the effect of void size on the response was not considered. In contrast,
for the numerical study, four conditions with varying void sizes were designed as listed in Table 3. The
cross-sections of the voids were fan-shaped with different radius, and their longitudinal length was
kept the same as 4 m. An impact force signal was applied on the inner surface of the tunnel lining in
the model. The duration of the force was 0.0032 s, and the peak value was 88.26 kN. The frequency
range influencing the impact force was between 1 and 150 Hz (Figure 14). The vibration detecting
point was 0.1 m close to the impact point. To analyse the effect of the impact-void position, the impact
and response points were varied with reference to the centre of void along the outer surface of the
tunnel lining. The distance between the impact point and the centre of the void, represented by Dx
(Figure 13b), was varied from 1 to 6 m.
100 1.0
80 Table 3. Conditions with different0.8

void sizes.
Amplitude
60
Force/kN
Conditions Size/m3
Void 0.6
40 Condition S0 0 (without void)
0.4
Condition S1 0.39
20
Condition S2 3.14
0.2
0
Condition S3 7.07
0.0
0.0 0.1 0.2 0.3 0 20 40 60 80 100 120 140 160
Time/s Frequency/Hz
100 1.0
80 0.8
Amplitude
60
Force/kN
0.6
40
0.4
20
0.2
0
0.0
0.0 0.1 0.2 0.3 0 20 40 60 80 100 120 140 160
Time/s Frequency/Hz
(a) (b)
Figure 14. (a) Time history and (b) spectrum of the impact force.
235
Appl. Sci. 2019, 9, 3280
3.2. Effect of Void Size

Figure 15 illustrates the time histories of the four void size conditions. For condition S0, the
acceleration is relatively smaller with a peak value of 6.49 m/s2 . For condition S1, the acceleration is
larger with a peak value of 7.02 m/s2 . This is because, part of the vibration energy is absorbed by the
soil beside the void, and part of the energy reflects at the tunnel-void interface. For conditions S2 and
S3, with the increase of the void size, the peak value is higher, which means more energy reflects at the
interface rather than transmitting into the soil. In short, as the void is larger, the peak value of the
waveform is higher, which means more energy is reflected back.
12
Condition S0
6
012 Condition S0
6
-6
0
-12
0.0
-6 0.1 0.2 0.3 0.4
12
-12
0.0 0.1 0.2 Condition
0.3 S1
0.4
Acceleration/(m/s2)
6
12
06 Condition S1
Acceleration/(m/s2)
-6 0
-12-6
0.0 0.1 0.2 0.3 0.4
-12
12 0.0 0.1 0.2 0.3
Condition 0.4
S2
612
Condition S2
06
-6 0
-12-6
0.0
-12
0.1 0.2 0.3 0.4
12 0.0 0.1 0.2 0.3 0.4
612 Condition S3
Condition S3
06
-6 0
-6
-12
0.0
-12 0.1 0.2 0.3 0.4
0.0 0.1 0.2 0.3 0.4
Time/s
Time/s
Figure 15. Time histories of void conditions.
Figure 16 shows normalised Fourier spectra of the four void size conditions. For condition S0, the
vibration amplitude distributes in a wide frequency range, and the dominant frequency is not obvious.
The normalised peak amplitude is approximately 0.50. With the increase of void size, the amplitude
increases, and the dominant frequency can be clearly observed. In general, as the void is bigger, the
frequency distributes a narrower range, and the dominant frequency at 90 Hz is more obvious.
1.01.0 ConditionS0
S0
0.80.8 Condition
0.6
0.6
0.4
0.40.2
0.20.0
0.0 0 20 40 60 80 100 120 140 160
0
1.0 20 40 60 80 100 120 140 160
1.00.8 Condition S1
0.80.6 Condition S1
0.60.4
Amplitude
0.40.2
Amplitude
0.20.0
0.0 0 20 40 60 80 100 120 140 160
1.0
0 20 40 60 80 100 120 140 160
1.00.8 Condition S2
0.80.6 Condition S2
0.60.4
0.40.2
0.20.00 20 40 60 80 100 120 140 160
0.01.0
0 20 Condition
40 60S3 80 100 120 140 160
0.8
1.00.6 Condition S3
0.80.4
0.60.2
0.40.0
0.2 0 20 40 60 80 100 120 140 160
0.0
0 20 40 Frequency/Hz
60 80 100 120 140 160
Frequency/Hz
Figure 16. Normalised Fourier spectra of void conditions.
236
1.4x1010
Condition S0
Condition S1
1.2x1010 Condition S2
Condition S3
Appl. Sci. 2019, 9, 3280 1.0x1010
8.0x109
the 9dynamic stiffness varies with frequency for the four conditions. With
Figure 17 illustrates that6.0x10
the increase of void size, the4.0x10
dynamic
9 stiffness decreases in almost all the frequencies.
2.0x109 10
1.4x10
Condition S0
0.0 Condition S1
1.2x10010 20 40 60 S2 80 100
Condition 120 140 160
10
Condition S3
Frequency/Hz
1.0x10
8.0x109
6.0x109
4.0x109
2.0x109
0.0
0 20 40 60 80 100 120 140 160
Frequency/Hz
Figure 17. Dynamic stiffness analysis.
Figure 18 shows the colour contours of the wavelet spectra for the four void size conditions. With
the increase of the void size, the spectrum becomes larger, and an increasing energy concentration
can be obviously found. More energy tends to concentrate over 80–120 Hz, which in turn becomes
the dominant frequency band. The vibration duration is longer especially in the range of 60–120 Hz.
While the frequency is lower, the vibration energy lasts longer.
(a) Condition S0 (b) Condition S1
(c) Condition S2 (d) Condition S3
Figure 18. Wavelet spectra for different void size conditions.
237
Appl. Sci. 2019, 9, 3280
When the impulse force is applied to the tunnel, the P wave propagates into the structure as a
semi-spherical wave. Figure 19 shows a sectional draw of the reflection of the stress wave with different
void sizes. According to the wave impendence introduced in Section 2.1, there is a strong reflection
(red part) in the tunnel-void interface and a weak reflection (blue part) in the tunnel-soil interface.
While the void becomes larger, the angle of the red part is larger, which means more energy reflects
back and the attenuation lasts longer. This may cause the spectrum becomes larger in proportion to the
void size.
(a) (b)
Figure 19. The reflections of P wave in the conditions with (a) small void and (b) big void.
3.3. Effect of Distance between Impact and Void Position

When the impact location is far from the void, the void existence is not distinctly recognised.
Accordingly, in this subsection, different impact positions were analysed. The distance between the
impact location and the void centre was defined as Dx, which was varied from 0 to 6 m for the different
void size conditions. Figure 20 shows the dynamic stiffness Kd of the responses. The average values of
the dynamic stiffness were calculated in the range of 20–100 Hz due to the steady amplitudes.
2.0x109
1.8x109
1.6x109
1.4x109
Condition S1
9 Condition S2
2.0x101.2x10 9
Condition S3
Condition S0
9
1.8x1091.0x10 0 1 2 3 4 5 6
Dy/m
1.6x109
Figure 20. Dynamic stiffness of different void positions.
1.4x109
For condition S0, Kd is not affected by Dx. WhenCondition
Dx changes
S1
from 0 to 1 m, in the range of void
size, Kd for condition S1 decreases
1.2x109 slightly, while for conditions
Condition S2 S2 and S3 increase slightly.
Condition S3
For conditions S1, S2, and S3, when Dx changes from 1 S0
Condition to 3 m, Kd increases significantly. That is,
9
the dynamic stiffness changes 1.0x10
quickly when the impact point
0 1 2 3 4 5 moves
6 through the void edge. When Dx
changes from 3 to 6 m, Kd increases slowly. The dynamic stiffness for void conditions becomes almost
/m
the same as that of the non-void condition when Dx reaches to 6 m.
4. Conclusions
To investigate the effect of a void behind a tunnel lining using the IE method, both experimental
and numerical studies were performed. The different signal analysis approaches were compared, and
238
Appl. Sci. 2019, 9, 3280
the influence of the void size and impact-void position were analysed. The following conclusions can
be drawn:
• In the analysis of the vibration response in the time, frequency, and time–frequency domains,
the fractal box-counting dimensions and dynamic stiffness can be used for void recognition
quantitatively, and the wavelet analysis can provide a relatively better visual energy distribution
among the different time–frequency analysis approaches.
• When the void is larger, the dynamic stiffness decreases. When the impact location is beyond the
range of void size, the dynamic stiffness changes quickly.
By comparison of different signal analysis approaches, wavelet transform is believed to be the
most appropriate one to show the energy flow during the void state changing. Besides, the dynamic
stiffness could be used to detect the void position accurately.
Author Contributions: Data curation, R.C. and R.L.; formal analysis, R.C. and C.N.; funding acquisition, M.M.;
investigation, R.L. and C.N.; methodology, M.M.; project administration, M.M.; software, R.C.; supervision, M.M.;
writing—original draft, R.C.; writing—review & editing, R.C. and M.M.
Funding: This research was funded by fundamental research funds for the central universities, grant number
2018JBM036 and the research funding of China academy of railway sciences, grant number 2018YJ164.
Conflicts of Interest: The authors declare no conflict of interest. The funders played role in the design of the
study, the collection and the decision to publish the results.
References
1. Lalagüe, A.; Lebens, M.A.; Hoff, I.; Grøv, E. Detection of Rockfall on a Tunnel Concrete Lining with
Ground-Penetrating Radar (GPR). Rock Mech. Rock Eng. 2016, 49, 2811–2823. [CrossRef]
2. Cardarelli, E.; Marrone, C.; Orlando, L. Evaluation of tunnel stability using integrated geophysical methods.
J. Appl. Geophys. 2003, 52, 93–102. [CrossRef]
3. Huang, J.; Qi, H.; Diao, Z. Application of ultrasonic-rebound combined method in concrete strength test for
tunnel secondary lining. Water Resour. Hydropower Northeast 2011, 23. [CrossRef]
4. Wang, W.; Xiang, Y. Non-destructive testing and its application in tunnels. J. Railw. Eng. Soc. 2001, 3, 85–88.
5. Sansalone, M.; Carino, N. Detecting Voids in Metal Tendon Ducts using the Impact-Echo Method. Mater. J.
Am. Concr. Inst. 1992, 89, 8.
6. Sansalone, M.J.; Streett, W.B. Impact-Echo: Nondestructive Evaluation of Concrete and Masonry; Bullbrier Press:
Jersey Shore, PA, USA, 1997; ISBN 0961261064.
7. Ni, S.-H.; Lehmann, L.; Charng, J.-J.; Lo, K.-F. Low-strain integrity testing of drilled piles with high slenderness
ratio. Comput. Geotech. 2006, 33, 283–293. [CrossRef]
8. Cheng, C.-C.; Yu, C.-P.; Liou, T. Evaluation of interfacial bond condition between concrete plate-like structure
and substrate using the simulated transfer function derived by IE. NDT E Int. 2009, 42, 678–689. [CrossRef]
9. Aggelis, D.; Shiotani, T.; Kasai, K. Evaluation of grouting in tunnel lining using impact-echo. Tunn. Undergr.
Space Technol. 2008, 23, 629–637. [CrossRef]
10. Song, K.-I.; Cho, G.-C. Bonding state evaluation of tunnel shotcrete applied onto hard rocks using the
impact-echo method. NDT E Int. 2009, 42, 487–500. [CrossRef]
11. Song, K.-I.; Cho, G.-C. Numerical study on the evaluation of tunnel shotcrete using the Impact-Echo method
coupled with Fourier transform and short-time Fourier transform. Int. J. Rock Mech. Min. Sci. 2010, 47,
1274–1288. [CrossRef]
12. Davis, A.G.; Lim, M.K.; Petersen, C.G. Rapid and economical evaluation of concrete tunnel linings with
impulse response and impulse radar non-destructive methods. NDT E Int. 2005, 38, 181–186. [CrossRef]
13. Ryden, N.; Aurell, O.; Nilsson, P.; Hartlén, J. Impact echo Q-factor measurements towards non-destructive
quality control of the backfill in segmental lined tunnels. In Nondestructive Testing of Materials and Structures;
Springer: Berlin/Heidelberg, Germany, 2013; pp. 915–919. [CrossRef]
14. Sansalone, M.; Carino, N. Impact-Echo: A Method for Flaw Detection in Concrete Using Transient Stress Waves;
National Bureau of Standards: Gaithersburg, MD, USA, 1986. [CrossRef]
15. Mandelbrot, B.B. The Fractal Geometry of Nature; WH Freeman: New York, NY, USA, 1983; Volume 173,
ISBN 0716711869.
239
Appl. Sci. 2019, 9, 3280
16. Ma, M.; Liu, J.; Ke, Z.; Gao, Y. Bearing Capacity Estimation of Bridge Piles Using the Impulse Transient
Response Method. Shock Vib. 2016, 2016, 1–8. [CrossRef]
17. Ge, Z.; Chen, Z. Matlab Guide to Time-Frequency Analysis; Posts & Telecom Press: Beijing, China, 2006;
ISBN 9787115141132.
18. Lee, I.-M.; Han, S.-I.; Kim, H.-J.; Yu, J.-D.; Min, B.-K.; Lee, J.-S. Evaluation of rock bolt integrity using Fourier
and wavelet transforms. Tunn. Undergr. Space Technol. 2012, 28, 304–314. [CrossRef]
19. Shi, Z.; Liu, L.; Peng, M.; Liu, C.; Tao, F.; Liu, C. Non-destructive testing of full-length bonded rock bolts
based on HHT signal analysis. J. Appl. Geophys. 2018, 151, 47–65. [CrossRef]
240
applied
sciences
Article
A PEC Thrice Subtraction Method for Obtaining
Permeability Invariance Feature in Conductivity
Measurement of Ferromagnetic Samples
Dongdong Wen 1, * , Mengbao Fan 1, *, Binghua Cao 2 , Zhian Xue 1 and Ping Wang 3,4
1 School of Mechatronic Engineering, China University of Mining and Technology, Xuzhou 221116, China
2 School of Information and Control Engineering, China University of Mining and Technology, Xuzhou
221116, China
3 School of Automation Engineering, Nanjing University of Aeronautics and Astronautics,
Nanjing 210016, China
4 Nondestructive Detection and Monitoring Technology for High Speed Transportation Facilities, Key
Laboratory of Ministry of Industry and Information, Nanjing University of Aeronautics and Astronautics,
Nanjing 210016, China
* Correspondence: [email protected] (D.W.); [email protected] (M.F.); Tel.: +86-0516-83885829 (M.F.)

Received: 17 May 2019; Accepted: 5 July 2019; Published: 7 July 2019
Featured Application: The work is potential to obtain a permeability invariance (PI) feature in
pulsed eddy current (PEC) signals for reducing the permeability and environment magnetic field
influence in the conductivity measurement of ferromagnetic samples.
Abstract: Conductivity, as an important index of structural health monitoring, can be used to evaluate
heat treatment condition, and sort different materials or measure the stress of mechanical parts.
However, the permeability of a measured sample has significant impact on the detected signal
in pulsed eddy current (PEC) testing, which is prone to measurement errors due to the effect of
permeability change. In this paper, a thrice subtraction method is investigated and utilized to obtain
a permeability invariance (PI) feature for reducing permeability effect in conductivity measurement
of ferromagnetic samples. The thrice subtraction method is based on the PEC signals of sample
and air, the difference signal between the difference PEC signal and its normalization signal, and
the difference signal between the difference normalization signal and its standard deviation. In the
thrice subtraction signals, the behavior of the obtained PI feature is analyzed by experiments and
simulations. The results demonstrate that the thrice subtraction method is a practicable program and
the PI feature is potential to measure the conductivity of ferromagnetic samples. The work reported
in this paper provides an effective approach to obtain a PI feature for estimating the conductivity of
ferromagnetic samples without a permeability effect.
Keywords: pulsed eddy current; ferromagnetic sample; conductivity; permeability
1. Introduction
Conductivity, as a parameter of ferromagnetic materials, can serve as an index to sort different
materials, and evaluate heat treatment condition or measure the stress of mechanical parts [1–3]. Due
to the advances in electromagnetic nondestructive evaluation, pulsed eddy current (PEC) testing has
become a feasible and preferred technique for measuring conductivity of materials in recent years [4,5].
However, for ferromagnetic samples, the measurement of conductivity is affected by the inhomogeneity
of tested samples and the confounding cross-sensitivity to conductivity and permeability, which
is undesirable in precision PEC testing [6,7]. Especially, the influence of permeability is complex
241
Appl. Sci. 2019, 9, 2745
and irregular. On the other hand, correspondingly, the permeability of ferromagnetic material will
change when the environmental magnetic field changes. Therefore, reducing the permeability effect
or the influence of environmental magnetic field in the measurement of electrical conductivity of
ferromagnetic samples is an important problem to be solved in PEC testing. It is important and
meaningful to improve the detection accuracy of conductivity of ferromagnetic samples.
In PEC testing, the measurements of conductivity and permeability have been studied by many
scholars in recent years. Adewale et al. [8] decoupled the influence of permeability and conductivity
in PEC measurements and the contributions of the electromagnetic properties (permeability and
conductivity) were analyzed with a view to isolate the influence of these two properties. It was
found that the conductivity effects are prominent in the rising edge of the transient response and the
permeability effects are prominent in the stable phase of the transient response. Chen et al. [4] studied
the electrical conductivity measurement of ferromagnetic metallic materials using the PEC technique.
In the measurement of conductivity, an inverse problem method based on a PEC analytical model was
proposed to determine the conductivity and permeability of ferromagnetic plates. Dziczkowski et
al. [9] described the method of determining the equivalent parameters of the coil, a scaling method
for a conductivity meter based on the elimination of liftoff on conductivity results, and a proposition
for an effective method to measure the conductivity of rough elements. Lu et al. [1,10] found a
conductivity lift-off invariance phenomenon and proposed an inverse problem method to determinate
the conductivity and permeability of ferrite metallic plates. Wang et al. [2] proposed an eddy current
measurement method to determinate electrical conductivity of samples, which indicated that the
logarithm of phase signature of impedance change basically varies linearly with electrical conductivity.
Yu et al. [11] observed a conductivity invariance phenomenon of eddy current non-destructive testing
(NDT), and used it to estimate the magnetic permeability of metals without the influence of its
conductivity. Ye et al. [12] proposed a novel approach based on decay time (DT) using PEC and a
highly sensitive magnetoresistive (MR) sensor in conductivity measurement of materials. The results
indicate that the DT parameter is linearly correlated with material conductivity allowing it to be used
for estimating the conductivity from the PEC measurement. According to the above studies, it is found
that the conductivity measurements of ferromagnetic samples with a permeability effect are rarely
studied by using PEC testing when permeability changes. These works also revealed that the change
of PEC signals caused by conductivity and permeability is indistinguishable in the measurement
of ferromagnetic samples. In other words, when the conductivity is measured, the influence of
permeability can not be reduced or eliminated in PEC signals. Thus, it is important and significant
to reduce the influence of permeability on PEC signals, and accurately measure the conductivity of
ferromagnetic samples when permeability changes.
It is known that liftoff point of intersection (LOI), as a desirable signal feature, is independent of
liftoff distance [13]. The LOI is a common intersection point of PEC signals when liftoff distance changes.
For the permeability effect on PEC signals, a similar method can be considered in the conductivity
measurement to reduce the influence of permeability. Thus, similarly, a common intersection invariance
point feature is needed in PEC signals of ferromagnetic samples for reducing the permeability effect
when sample permeability changes or the environmental magnetic field changes. Therefore, in the
measurement of conductivity, it is important and significant to propose a method to obtain a permeability
invariance (PI) feature for reducing permeability effect or environmental magnetic field influence.
The main contribution of this paper is that a thrice subtraction method is proposed to obtain
a PI point feature in PEC signals of ferromagnetic samples for reducing the permeability effect or
environmental magnetic field influence. Correspondingly, the availability of the thrice subtraction
method and the obtained PI feature is analyzed and discussed. The rest of this paper is organized as
follows: in Section 2 the PEC analytical model is formulated from a harmonic eddy current model, the
thrice subtraction method is introduced in Section 3, the simulation results are analyzed in Section 4,
in Section 5 the experiments are performed for verifying the simulation results and finally, Section 6
presents the concluding remarks.
242
Appl. Sci. 2019, 9, 2745
2. Formulation of the PEC Analytical Model

The PEC analytical model, as an effective tool, has been used to analyze the change of PEC signals
caused by thickness, conductivity and permeability [4,14]. According to previous studies, it can be
found that it is very difficult to directly construct a PEC analytical model. However, on the other hand,
it is simpler to construct a PEC analytical model based on the harmonic eddy current analytical model.
It is known that a periodic pulse signal can be expressed as a Fourier series of trigonometric functions.
Therefore, in this paper, the PEC analytical model is reconstructed through using the harmonic eddy
current analytical model. The main process is shown in Figure 1.
Pulse excitation Impedance
() change ∆Z(ω)
Pulse excitation Impedance
Fourier y(t) Harmonic eddy
change ∆Z(ω)
transform current response
Fourier Harmonic eddy
Spectral
transform
pulse Coil impedance
current response
signal (ω ) in air Z (ω)
Spectral pulse Coil impedance
signal Y(ω) in air Zair(ω)
[ (ω)/(∆Z(ω)+Z (ω)) ]×ΔZ(ω)

Inverse Fourier
[Y(ω)/(∆Z(ω)+Z air(ω)) ] ×ΔZ(ω)
transform
Inverse Fourier
Difference PEC
transform
signal ∆ ( )
Difference PEC
signal ∆u(t)
Figure 1. Flow chart of pulsed eddy current (PEC) model.

()
In the process of constructing the PEC analytical model, a pulse () voltage signal y (t) serves as
an excitation signal of the coil sensor. The difference PEC signal ∆u(t) is the voltage change of the
（） ( )
coil sensor when a plate is tested. The harmonic eddy current responses （ ∆Z） (ω) and Zair (ω) in the
frequency domain are the basis of the PEC model construction, which have been built in previous
studies [15,16]. The case for the coil sensor above a conductive plate was considered, as illustrated in
Figure 2. The PEC probe was an air-core cylindrical coil. The plate under research was assumed to
be linear, homogeneous and isotropic. The electrical conductivity, permeability and thickness of the
0 0
measured metallic plate are denoted as σ, µ and µ d, respectively, where µ = µ0 µr (µ0 is the permeability
of the vacuum, µr is the relative permeability
µ of the metallic plate). An artificial magnetic insulation
boundary was placed at h.
h
r1 r2
Z2
Z1
μμ σσ d
μμ0
Figure 2. Coil sensor above a plate.
Based on the harmonic eddy current response and spectral pulse signals, the difference PEC
response can be formulated by performing Kirchhoff’s Law in the frequency domain. It signifies that
the algebraic sum of the voltages of the coil sensor is always equal to zero in a closed loop, which can be
expressed as ∆U = Uplate − Uair = I × ((Zplate − Zair) ) . Correspondingly, the difference PEC response in
the frequency domain can be obtained. Where Uplate and Zplate are the voltage and impedance signal of
coil sensor when a plate is tested, Uair and Zair are the voltage and impedance signal of the coil sensor
when the coil sensor is in air, the voltage signal Uair and the impedance signal Zair serve as a reference
243
Appl. Sci. 2019, 9, 2745
signal, ∆U is the difference PEC response between the detection signal Uplate and reference signal Uair ,
I is the current of coil sensor, ∆Z = Zplate − Zair is the impedance change response when a plate is
tested. Correspondingly, the difference PEC response in the frequency domain can be expressed as
Y (ω)
∆U (ω) = × ∆Z(ω) (1)
∆Z(ω) + Zair (ω)
where Y(ω) denotes the Fourier transform of pulse voltage signal y(t), Zplate (ω) = ∆Z(ω) + Zair (ω),
Y(ω)/(∆Z(ω) + Zair (ω)) = I.
By using inverse Fourier transform (IFT), the difference PEC signal in the time domain can be
expressed as
∆u(t) = IFT[∆U (ω)]. (2)
Based on the results, a numerical calculation program can be written in MATLAB software
(MTLAB R2014a, Math Works, Inc, Natick, Massachusetts, United States, 2014). Correspondingly,
some scientific explanations can be verified by numerical method.
In Equation (1), when the air-core coil is excited by harmonic excitation [15], the impedance
change of air-core coil can be written as
∞
X χ2 (λ0i r1 , λ0i r2 ) 2
∆Z(ω) = j2πωµ0 n2c 2
(e−λ0i z1 − e−λ0i z2 ) R0i,1i . (3)
7 2
i=1 λ0i h [ J0 (λ0i h)]
On the other hand, for the case of the air-core coil in air [15], the impedance can be expressed as
∞
X χ2 (λ0i r1 , λ0i r2 )
Zair (ω) = j4πωµ0 n2c 2
[λ0i (z2 − z1 ) + e−λ0i (z2 −z1 ) − 1]. (4)
7 2
i=1 λ0i h [ J0 (λ0i h)]
In Equation (3), the R0i,1i can be expressed as
(λ0i µr )2 − λ21i + e−2λ1i d [λ21i − (λ0i µr )2 ]

R0i,1i = (5)
(λ1i + λ0i µr )2 − e−2λ1i d (λ1i − λ0i µr )2
where nc = Nc /(r2 − r1 )(z2 − z1 ), Nc denotes the number of the coil turns, r2 , r1 are the outer and inner
radius of the coil, (z2 − z1 ) denotes the height of the coil, λ0i is the i − th positive root of the Bessel
function J1 (λ0 h), Jm is the m − th order Bessel function of the first kind, R0i,1i is a constant related to the
q
conductivity, permeability and thickness of measured metallic plates, λ1i = λ20i + jωσµ.
In order to obtain the difference PEC response for simulating transient induced voltage, the periodic
pulse signal is applied. Combined with the superposition principle and inverse Fourier transform, the
PEC transient response can be reconstructed from the harmonic eddy current response [17,18]. For the
T-periodic pulse, the formula can be expressed as
t τ−t
y(t) = A[(1 − e− a )step(t) + (e a − 1)step(t − τ)]τ < T (6)
where, a is the time constant, step(t) is the step function, A is the amplitude value of excitation current,
τ is the excitation time, T is the excitation period.
3. Thrice Subtraction Method

By analyzing previous research ideas [13,19], a thrice subtraction method was proposed to obtain
a PI feature for reducing the effect of permeability change. The thrice subtraction method included
three subtraction steps and the operating process is shown in Figure 3. The first subtraction was
implemented to eliminate the effect of PEC signal from air. A difference PEC signal was obtained after
244
Appl. Sci. 2019, 9, 2745
the subtraction process. Then, the normalization signal of the difference PEC signal was obtained and
the second subtraction was performed between the difference PEC signal and its normalization signal,
which was used to amplify the variation of the difference PEC signal. Next, the standard deviation of
the difference normalization signal was calculated and the third subtraction was implemented, which
was used to reduce the effect of permeability. Finally, the PI feature could be extracted from thrice
subtraction signals when permeability changed. The detailed steps are described below.
First, the subtraction between PEC signals of sample and air is implemented to obtain the difference
PEC signal. The subtraction process can be shown as
() () ()
∆U (t) = U (t)plate − U (t)air (7)
() ()
where, U (t)plate is the PEC signal of sample, U (t)air is the PEC signal of air.
Second, the subtraction between the difference PEC signal and its normalization signal is performed
to obtain the difference normalization signal. The difference normalization signal can be formulated as
()
() ∆U (t)
Udnorm
= ∆U (t) − max( ( )) (8)
max(∆U (t))
()
∆U (t) max( ( ))
where, maxmax( denotes the normalization signal of the difference PEC signal, max(∆U (t)) denotes
(∆U (t)) ( ))
the maximum value of the difference PEC signal.
Then, the third subtraction between the difference normalization signal and its standard deviation
is calculated to obtain the thrice subtraction signal. The thrice subtraction signal can be expressed as
Ustd = Udnorm − std(Udnorm ) (9)

( )
where, std(Udnorm ) denotes the standard deviation of the difference normalization signal.
(
By changing )the permeability of samples, the different thrice subtraction signals can be obtained
and the PI feature can be extracted by mutual subtraction between them as
Ustd−1 − Ustd−2 = 0. (10)

1 2 0
Finally, by solving Equation (10), the amplitude and time of the PI feature can be obtained. It is
easy to think that the common point of both different signals, called the PI feature point, is unaffected
by the change in permeability. In other words, the PI feature is immune to the permeability effect.
Normalization Standard deviation

PEC signal
of sample Difference Change
Difference Three subtraction permeability Permeability
PEC signal normalization
signal invariance feature
PEC signal signal
of air
Figure 3. Flow diagram of thrice subtraction method.
4. Simulation Results
4.1. PI Feature Acquisition

According to the PEC analytical model, the numerical simulation code was written in MATLAB
software to simulate the PEC signal with different relative permeability and verify the thrice subtraction
method. In simulations, the height, inner and outer radius of the probe coil were 6 mm, 4 mm and
6 mm, respectively. There were 300 turns of coil. The boundary was set to h = 20r2 = 120 mm. The
pulse excitation had a 10 V amplitude, 200 Hz frequency and 50% duty cycle. The time constant of
pulse20excitation
2 120 mmwas set to 33 µs. The excitation signal of pulse voltage is shown in Figure 4. In Figure 5,
the difference PEC signals are shown when permeability μs changes. It can be observed that the difference
245
Appl. Sci. 2019, 9, 2745
PEC signals integrally decreased when relative permeability decreased. Especially, in the partially
enlarged drawing shown in Figure 6, the variation of difference PEC signal is obvious. The results
indicated that the difference PEC signals did not show intersection phenomenon and the permeability
effect was obvious when permeability changes.
10
10
10
8
88
signal/V
signal/V
signal/V
6
66
Voltage
Voltage
4
Voltage
44
2
22
0
00 00 1 2 3 4 5
0 11 22 Time/ms
33 44 55
Time/ms
Time/ms
Figure 4. Pulse excitation voltage signal.
1.4
1.4
1.4 μᵣ =
μᵣ = 30
30
1.2 μᵣ = 30
1.2 μᵣ = 50
signals/V
μᵣ == 50
50
signals/V
1.2 μᵣ
signals/V
1 μᵣ = 70
μᵣ == 70
70
11 μᵣ
μᵣ = 90
μᵣ == 90
μᵣ 90
0.8
0.8
PEC
0.8
PEC
PEC
0.6
0.6
0.6
Difference
Difference
Difference
0.4
0.4
0.4
0.2
0.2
0.2
0
00 0 0.5 1 1.5
00 0.5
0.5 Time/s 11 1.5
1.5
Time/s
Time/s
Figure 5. The difference in PEC signals.
1.4
1.4
1.4 μᵣ =
μᵣ = 30
30
μᵣ = 30
signals/V
1.2
signals/V
1.2 μᵣ = 50
signals/V
1.2 μᵣ == 50
μᵣ 50
1 μᵣ = 70
μᵣ == 70
70
11 μᵣ
μᵣ = 90
μᵣ == 90
90
μᵣ
PEC
0.8
PEC
0.8
PEC
0.8
Difference
0.6
Difference
0.6
Difference
0.6
0.4
0.4
0.4
0.2
0.2
0.2
0
00 0.1
0.1 0.2
0.3
0.2
0.3 0.4
0.4 0.5
0.5 0.6
0.6
0.1 0.2
0.3
Time/ms0.4 0.5 0.6
Time/ms
Time/ms
Figure 6. Partially enlarged drawing.
Under the condition that the rise time of pulse excitation is constant, when excitation pulse
frequency increased, the difference PEC signals changed as shown in Figure 7. It was observed that the
change of excitation pulse frequency led to the change of excitation pulse period. In experiments, the
difference PEC signals were recorded when it was triggered by the rising or falling of pulse excitation.
It can be seen that the difference PEC signal did not change in the first half part of the difference PEC
246
Appl. Sci. 2019, 9, 2745
signal, as shown in Figure 7a, when the excitation pulse frequency changes at lower excitation pulse
frequency. When the pulse frequency was higher, the recorded difference PEC signal in first half part of
periodic pulse was incomplete and there was no stationary part in difference PEC signal, as shown in
Figure 7b. When the rising time of pulse excitation changed, as shown in Figure 8, it can be observed
that the response time of the difference PEC signals to stationary section was shorter.
1.5 200 Hz 1.5
333 Hz
Difference PEC signals/V
11.5 250 HzHz

200 1.5
signals/V
1
333 Hz
500 Hz
250Hz
250
300 HzHz
signals/V
0.5 1 1
0.5 1000 Hz
500 Hz
250Hz
300 Hz 0.5
00.5 1000 Hz
PECPEC
0
0.5 0 0
Difference
0.5
Difference
1 1
−0.5 −0.5
1.5 1.5
−1 −1
0
−1.5 1 2 3 4 5 −1.5 0 1 2 3
0 1 Time/ms
2 3 4 5 0 1 Time/ms 2 3
Time/ms Time/ms
(a) (b)
Figure 7. The difference PEC signals with different frequencies. (a) lower frequency; (b) higher
frequency.
Rising time = 0.12 ms

Rising time
Rising time == 0.12
0.25 ms
ms
1.5 Rising time
Rising time == 0.25
0.36 ms
ms
1.5
Rising time = 0.36 ms

1
1
0.5
0.5
00
00 0.5
0.5 1 1.5
1.5
Time/ms
Time/ms
Figure 8. The difference in PEC signals with different rising times.
By using the thrice subtraction method based on the above difference PEC signals, the thrice
subtraction signals could be obtained from the difference PEC signals, as shown in Figure 9. It was
observed that the amplitude of thrice subtraction signals was smaller than that in the difference PEC
signals. An intersection point phenomenon was exhibited in the thrice subtraction signals, which is
called a PI signal feature. It was observed that at the intersection point of thrice subtraction signals,
the thrice subtraction signals were independent of permeability effect. The PI feature was immune
to permeability change. According to above analyses, it is known that the thrice subtraction signal
was affected by excitation pulse frequency and rising time of pulse excitation. Thus, some limits are
needed in the application of PI phenomenon. In order to keep the method practical, the excitation
pulse frequency should be kept lower when rising time of pulse excitation is longer.
247
Appl. Sci. 2019, 9, 2745
0.25
0.25 μᵣμᵣ==30
30
signals/V
subtraction signals/V
0.2 μᵣμᵣ==50
50
0.2
μᵣμᵣ==70
70
0.15
0.15 μᵣμᵣ==90
90
Thrice subtraction
0.1
0.1
0.05
0.05
PI
PI
00
Thrice
0.05
−0.05
0.1
−0.1
00 0.5
0.5 11 1.5
1.5
Time/ms
Time/ms
Figure 9. Thrice subtraction signals. Permeability invariance (PI) is shown by the red circle at the
intersection of the signals.
4.2. Behavior of PI Feature

When conductivity and thickness of samples changed, the PI time and amplitude were changed,
as shown in Figures 10 and 11. It was found that the PI amplitude and time changed when the thickness
changed from about 0 to 2 mm and the conductivity changed from about 0 to 10 MS/m, which signified
that the thickness and conductivity of sample could be measured by testing the PI amplitude and time.
When the thickness was larger than 2 mm and conductivity changed from 0 to 3 MS/m, the PI time and
amplitude changed with increasing conductivity. When the thickness was larger than 2 mm and the
conductivity changed from 3 MS/m to 10 MS/m, the PI time did not change. The results demonstrated
that the detection range of conductivity was from 0 to 10 MS/m when the thickness was smaller than
2 mm and the detection range of conductivity was from 0 to 3 MS/m when the thickness was larger
than 2 mm. It also indicated that the conductivity measurement was only suitable for a certain range
when sample thickness was determined.
−4
x 10
5.1 4
−4
x 10
x 10
5 5.1
5.5x 10 4.9
5
5.55 4.8
4.9
PIs time/s
4.7
4.8
54.5
PIs time/s
0 4.6
4.7
4.54 4.5
4 5 0 4.6
4 2 4.4
4.5
4 0 10 5
4.4 2
0 10
Figure 10. PIs time change with thickness and conductivity.
When the conductivity of sample was 3.2 MS/m, the PI time and amplitude curves are depicted
with thickness in Figure 12. It was observed that the PI time and amplitude were changed when
thickness was smaller than 0.043.6 mm. When sample thickness was larger0.02 than 4 mm, the PI time and
PIs amplitude/V
amplitude were almost unchanged,

0.02 which signified that the thickness of sample did not affect the PI
0.04
feature with increasing thickness. The PI time and amplitude curves are0.01 0.02
shown with conductivity in
PIs amplitude/V
0
Figure 13 for a thickness samples
0.02 of 0.8 mm. It was found that the PI time and amplitude increased
with increasing conductivity. The PIs 0 0.01conductivity. The results
changes along a curve with increasing
-0.02
0 4
indicated that the PI time and amplitude could serve as an index of conductivity when sample thickness
was constant, which signified that the conductivity of ferromagnetic 0.01
samples
0 could be determined by
-0.02 0 2
4 5
0 10 0.01
2 248 0
5
0 10
4
x 10
5.1
Appl. Sci. 2019, 9, 2745 x 10
5
5.5 4.9
testing PI time or amplitude. In other words, the thrice subtraction method was a feasible program to
5 4.8
PIs time/s
obtain a PI feature for reducing the effect of permeability in conductivity measurements of ferromagnetic
4.7
samples. The results indicated 4.5 that the behavior of PI feature was affected by the conductivity and
thickness of the samples. Thus, in practical measurement of sample 0 4.6
conductivity, the thickness of
4
sample should keep constant and the other variables should also remain 4.5
unchanged in testing. When
4 5
the thickness of sample was bigger than2 skin depth of eddy current in4.4 conductivity measurement,
the thickness did not affect the change of PI feature, 0 10as shown in Figures 10 and 11. However, the
measurement range of conductivity was changed. In order to measure the conductivity of the sample,
the thickness of sample should be known and the size of sample should also be the same.
0.04 0.02
PIs amplitude/V
0.02
0.01
0
0
-0.02
4
−0.01
2 0
5
0 10
4
Figure 3 and conductivity.
11. PIs amplitude change with thickness
x 10 x 10
5.1 x 10−4 −3
20 x 10
5.15
PIs amplitude/V
15
20
4.9
PIs time/s
5 10
PIs amplitude/V
15
4.8
4.9
PIs time/s
105
4.7
4.8
50
4.6
4.7
05
4.6
1 2 3 4 1 2 3 4
Thickness/mm Thickness/mm
−5
1 2 3 4 1 2 3 4
Thickness/mm Thickness/mm
(a) (b)
Figure 12. PI time and amplitude curves with thickness. (a) PI time curve; (b) PI amplitude curve.
−3
x 10
−4
x 10
4 3
5.1 x 10 −2 x 10
PIs amplitude/V
5.15 2
−4
PIs time/s
4
PIs amplitude/V
−6
5
4.9
PIs time/s
6
−8
4.9
4.8 8
−10
4.8
4.7 10
−12
2 4 6 8 10 12 2 4 6 8 10
4.7 Conductivity/MS/m Conductivity/MS/m
2 4 6 8 10 2 4 6 8 10
Conductivity/MS/m Conductivity/MS/m
(a) (b)
Figure 13. PI time and amplitude curves with conductivity. (a) PI time curve; (b) PI amplitude curve.
249
Appl. Sci. 2019, 9, 2745
5. Experimental Setup and Results
5.1. Experimental Setup

As shown in Figure 14, the experimental setup included a computer, signal generator (DG1022Z),
data acquisition card (NI USB-6356) and probe. The pulse excitation signal was obtained from the
signal generator and applied to the probe. The data acquisition card communicated with computer
by USB interface. The data acquisition card was set to trigger on the rising edge, which means that
the PEC signal was recorded when the signal rising edge began. The probe signal was processed
by low pass filter and was enlarged in LabVIEW software (LabVIEW 2015, NI, United States, 2015).
The ferromagnetic material can be seen as many small magnetic domain structures. When the
environmental magnetic field changes, the magnetic domain organization is ordered in the direction of
the magnetic field. Therefore, a magnet can be used to change the environmental magnetic field so that
the permeability of the ferromagnetic material is considered to have changed. In order to change the
permeability of ferromagnetic samples, a magnet piece was used to change the magnetic field caused
by material permeability. The magnet piece was only used to change the magnetic field in the testing
environment, which means that any magnet piece could be used in testing. The distance between
magnet and sample was set to change the size of the magnetic field, as shown in Figure 15.
Data
Data
acquisition
acquisition
card
card
Signal
Signal
generator
generator
Probe
Probe
Sample Magnet Computer
Sample Magnet Computer
Figure 14. Experimental setup.
Distance
Distance
Magnet
Magnet
Sample
Sample
Figure 15. Sample and magnet.
5.2. Results and Discussion

In experiments, the pulse excitation had a 10 V amplitude, a 200 Hz frequency and a 50% duty
cycle. The sampling frequency was 1 MHz and the sampling point was 2500. When the rising edge was
triggered, the data acquisition card began to record the probe signals. The samples included 45# steel,
65Mn steel, A3 steel and nodular cast iron plates. The sample had a length of 100 mm, width of 100 mm
and height of 10 mm. The conductivities of the samples are listed in Table 1, which were measured
using the four-wire Kelvin resistivity measurement method [20,21] in a laboratory environment on
samples with a polished top surface.
250
Appl. Sci. 2019, 9, 2745
Table 1. Conductivity of experiment samples.
65Mn Steel 45# Steel A3 Steel Nodular Cast Iron

Conductivity
4.47 3.76 3.41 1.56
(MS/m)
4.47 3.76 3.41 1.56
Figure 16 shows the difference PEC signals when the distance between the magnet and sample
changed. It was observed that the difference PEC signals integrally increased or decreased when the
magnet distance changed. In terms of 45# steel, 65Mn steel, A3 steel and nodular cast iron samples,
the difference PEC signals did not exhibit intersection phenomenon and the amplitude of difference
PEC signals was obviously affected. The results were consistent with the simulation results, which
indicated that the difference PEC signals were affected when permeability changes and the PI feature
did not appear.
3.5
4 Distance = 5 mm
Distance = 5 mm Distance = 10 mm
Distance = 10 mm 3
Distance = 15 mm
3 Distance = 15 mm 2.5 Distance = 20 mm

Distance = 20 mm
2
2
1.5
1
1
0.5
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time/s Time/s x 10
−3
x 10
−3
(a) (b)
3 Distance = 5 mm
Distance = 5 mm 4 Distance = 10 mm
Distance = 10 mm
2.5 Distance = 15 mm
2 Distance = 20 mm 3
1.5 2
1
1
0.5
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time/s x 10
−3
Time/s x 10
−3
(c) (d)
Figure 16. The difference PEC signals. (a) A3 steel; (b) 45# steel; (c) 65Mn steel; (d) nodular cast iron.
Based on the above differences PEC signals, the thrice subtraction method was used to obtain
thrice subtraction signals and the PI feature. Figure 17 shows that a PI feature appeared in thrice
subtraction signals in terms of 45# steel, 65Mn steel, A3 steel and nodular cast iron samples. At the PI
feature, the thrice subtraction signals were almost independent of the permeability effect. Furthermore,
the PI features were extracted from thrice subtraction signals, as shown in Figure 18. It was observed
that the PI time and amplitude increased along a curve with increasing conductivity. The change rule
of conductivity was consistent with the actual measurement. The result was in agreement with the
simulation result under a certain condition, which demonstrated that by testing PI amplitude and time,
the conductivity of ferromagnetic samples could be determined without the permeability effect. The
251
Appl. Sci. 2019, 9, 2745
results pointed that the thrice subtraction method was a feasible program to obtain a PI feature for
reducing the permeability effect in the conductivity measurement of ferromagnetic samples.
2.52.5 2 2 Distance
Distance
Distance = 5= mm
5 mm Distance = 5= mm
5 mm
Thrice subtraction signals/V

Distance = 10 mm

2 2 Distance = 10
Distance
mm
= 15 mm 1.51.5 Distance
Distance = 15
= 15 mmmm
1.51.5 Distance = 20 mm
1 1
1 1
0.50.5
0.50.5
PIPI PIPI
0 0 0 0
−0.50.5
−0.50.5 0
0 0 0.50.5 1 1 1.51.5 2 23 0 0.50.5 1 1 1.51.5 2 23
Time/s Time/s
Time/s x 10
Time/s x
x 1010
−3 x 10
−3
(a) (b)
3 3 Distance
Distance
= 5= mm
5 mm Distance = 5= mm
5 mm
Distance
1.51.5 Distance = 10 mm Thrice subtraction signals/V Distance = 10 mm
Distance = 10 mm 2.52.5 Distance = 10 mm

Distance = 15
Distance = 15 mm mm Distance = 15
Distance = 15 mm mm
Distance
Distance = 20
= 20 mmmm 2 2 Distance
Distance = 20
= 20 mmmm
1 1
1.51.5
0.50.5 1 1
0.50.5
PIPI PIPI
0 0 0 0
−0.50.5
0 0 0.50.5 1 1 1.51.5 2 23 0 0 0.50.5 1 1 1.51.5 2 23
Time/s x 10 Time/s
Time/s x 10
Time/s x 10
−3
x 10
−3
(c) (d)
Figure 17. Thrice subtraction signals. (a) A3 steel; (b) 45# steel; (c) 65Mn steel; (d) nodular cast iron.
4−4
x 10
x 10
65Mn
65Mn steel
steel
3.05
3.05 45#
45# steel
steel 0 0
45#45# steel
PI amplitude/V
3 3 steel
PI amplitude/V
2.95
PI time/s
2.95
PI time/s
A3 steel 65Mn steel 0.05

−0.05
nodular
2.92.9 A3 steel 65Mn steel nodular
cast iron
2.85 cast iron
2.85
nodular 0.1
−0.1
2.82.8 nodular
A3A3 steel
steel
cast
cast iron
iron
2.75
2.75
2 2 3 3 4 4
2 2 3 3 4 4 Conductivity/MS/m
Conductivity/MS/m
Conductivity/MS/m
Conductivity/MS/m
(a) (b)
Figure 18. PIs with samples of different materials. (a) PI time curve; (b) PI amplitude curve.
In PEC testing, the formulation of the PEC model was established in a simplified condition. Thus,
the simulation results fit one specific case in terms of a certain condition. The simulation results were
available to verify the change law of the PEC response with different thickness, conductivity and
permeability. The simulations may have led to the inconsistency with experimental results under the
same conditions. On the other hand, due to the measurement error, there may have been deviation in
the conductivity measured by the four-wire Kelvin resistivity measurement method. According to
experiment and simulation results, it was feasible to verify the variation and trend of the conductivity
252
Appl. Sci. 2019, 9, 2745
in terms of real value. Although there were some shortcomings in the simulations and experiments, the
change law of conductivity and variation of conductivity was consistent with the actual measurement,
which demonstrated that the measurement results were feasible and the proposed method was effective.
6. Conclusions
Permeability, as an influence factor, will lead to measurement error in conductivity measurement
of ferromagnetic samples when it changes. In this paper, the effect of permeability was considered in
the PEC measurement of conductivity of ferromagnetic samples. In order to reduce the permeability
effect, a thrice subtraction method was proposed to obtain a PI feature in conductivity measurement
of ferromagnetic samples. The PI phenomenon was verified and the behavior of the PI feature
was analyzed with changes in conductivity and thickness. Following that, the experiments were
implemented to verify the availability of the thrice subtraction method and the practicality of the
obtained PI feature in the conductivity measurement. The results demonstrated that the thrice
subtraction method was a feasible program to obtain a PI feature for reducing the permeability effect.
The PI feature has the potential to enable the measurement of conductivity of ferromagnetic samples
without the permeability effect. The next step for further research is to measure ferromagnetic materials
with unknown conductivity and evaluate the thickness of ferromagnetic samples.
Author Contributions: Conceptualization, D.W., M.F., B.C. and Z.X.; methodology, D.W.; software, D.W.;
validation, M.F., B.C., P.W. and Z.X.; formal analysis, Z.X.; investigation, D.W. and Z.X.; resources, M.F. and
B.C.; data curation, D.W.; writing—original draft preparation, D.W. and Z.X.; writing—review and editing, M.F.,
B.C. and P.W.; visualization, D.W.; supervision, M.F. and P.W.; project administration, B.C. and P.W.; funding
acquisition, M.F.
Funding: This work was supported in part by the National Natural Science Foundation of China under grant
51677187, and in part by the Priority Academic Program Development of Jiangsu Higher Education Institutions.
Acknowledgments: The authors would like to thank the editors and reviewers for giving significant and
insightful comments.
References
1. Lu, M.; Xu, H.; Zhu, W.; Yin, L.; Zhao, Q.; Peyton, A.; Yin, W. Conductivity Lift-off Invariance and
measurement of permeability for ferrite metallic plates. NDT E Int. 2018, 95, 36–44. [CrossRef]
2. Wang, C.; Fan, M.; Cao, B.; Ye, B.; Li, W. Novel Noncontact Eddy Current Measurement of Electrical
Conductivity. IEEE Sens. J. 2018, 18, 9352–9359. [CrossRef]
3. Rodrigues, N.M.; Rosado, L.S.; Ramos, P.M. A portable embedded contactless system for the measurement
of metallic material conductivity and lift-off. Measurement 2017, 111, 441–450. [CrossRef]
4. Chen, X.; Lei, Y. Electrical conductivity measurement of ferromagnetic metallic materials using pulsed eddy
current method. NDT E Int. 2015, 75, 33–38. [CrossRef]
5. Lefebvre, J.H.V.; Dubois, J.M.S. Lift-off point of intercept (LOI) behavior. Rev. Prog. Quant. Nondestruct. Eval.
Vols 24a 24b 2005, 760, 523–530.
6. Lu, M.Y.; Zhu, W.Q.; Yin, L.Y.; Peyton, A.J.; Yin, W.L.; Qu, Z.G. Reducing the Lift-Off Effect on Permeability
Measurement for Magnetic Plates from Multifrequency Induction Data. IEEE Trans. Instrum. Meas. 2018, 67,
7. Tian, G.Y.; He, Y.; Adewale, I.; Simm, A. Research on spectral response of pulsed eddy current and NDE
applications. Sens. Actuators A Phys. 2013, 189, 313–320. [CrossRef]
8. Adewale, I.D.; Tian, G.Y. Decoupling the Influence of Permeability and Conductivity in Pulsed Eddy-Current
Measurements. IEEE Trans. Magn. 2013, 49, 1119–1127. [CrossRef]
9. Dziczkowski, L. Elimination of Coil Liftoff from Eddy Current Measurements of Conductivity. IEEE Trans.
Instrum. Meas. 2013, 62, 3301–3307. [CrossRef]
10. Lu, M.; Xie, Y.; Zhu, W.; Peyton, A.J.; Yin, W. Determination of the magnetic permeability, electrical
conductivity, and thickness of ferrite metallic plates using a multi-frequency electromagnetic sensing system.
IEEE Trans. Ind. Inform. 2018. [CrossRef]
253
Appl. Sci. 2019, 9, 2745
11. Yu, Y.T.; Zou, Y.; Al Hosani, M.; Tian, G.Y. Conductivity Invariance Phenomenon of Eddy Current NDT:
Investigation, Verification, and Application. IEEE Trans. Magn. 2017, 53. [CrossRef]
12. Ye, C.; Su, Z.; Rosell, A.; Udpa, L.; Udpa, S.; Capobianco, T.; Tamburrino, A. A decay time approach for linear
measurement of electrical conductivity. NDT E Int. 2019, 102, 169–174. [CrossRef]
13. Tian, G.Y.; Li, Y.; Mandache, C. Study of Lift-Off Invariance for Pulsed Eddy-Current Signals. IEEE Trans.
Magn. 2009, 45, 184–191. [CrossRef]
14. Ulapane, N.; Alempijevic, A.; Miro, J.V.; Vidal-Calleja, T. Non-destructive evaluation of ferromagnetic
material thickness using Pulsed Eddy Current sensor detector coil voltage decay rate. NDT E Int. 2018, 100,
15. Fan, M.; Huang, P.; Ye, B.; Hou, D.; Zhang, G.; Zhou, Z. Analytical modeling for transient probe response in
pulsed eddy current testing. NDT E Int. 2009, 42, 376–383. [CrossRef]
16. Chen, X.; Lei, Y. Excitation current waveform for eddy current testing on the thickness of ferromagnetic
plates. NDT E Int. 2014, 66, 28–33. [CrossRef]
17. Fan, M.B.; Cao, B.H.; Yang, X.F. Analytical time-domain model of transient eddy current field in pulsed eddy
current testing. Acta Phys. Sin. 2010, 59, 7570–7574.
18. Mandache, C.; Lefebvre, J.H.V. Transient and harmonic eddy currents: Lift-off point of intersection. NDT E Int.
2006, 39, 57–60. [CrossRef]
19. Wood, J. Invariant pattern recognition: A review. Pattern Recognit. 1996, 29, 1–17. [CrossRef]
20. Bowler, N.; Huang, Y. Electrical conductivity measurement of metal plates using broadband eddy-current
and four-point methods. Meas. Sci. Technol. 2005, 16, 2193–2200. [CrossRef]
21. Miccoli, I.; Edler, F.; Pfnur, H.; Tegenkamp, C. The 100th anniversary of the four-point probe technique: The
role of probe geometries in isotropic and anisotropic systems. J. Phys. Condens. Matter 2015, 27, 223201.
[CrossRef] [PubMed]
254
MDPI
St. Alban-Anlage 66
4052 Basel
Switzerland
Tel. +41 61 683 77 34
Fax +41 61 302 89 18
www.mdpi.com
Applied Sciences Editorial Office

E-mail: [email protected]
www.mdpi.com/journal/applsci
MDPI
St. Alban-Anlage 66
4052 Basel
Switzerland
Tel: +41 61 683 77 34
Fax: +41 61 302 89 18
www.mdpi.com ISBN 978-3-0365-2291-3

Nondestructive Testing NDT PDF

Uploaded by

Copyright:

Available Formats

Nondestructive Testing NDT PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nondestructive Testing NDT PDF

Uploaded by

Copyright:

Available Formats

Nondestructive

ISBN 978-3-0365-2292-0 (Hbk)

About the Editor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

Preface to ”Nondestructive Testing (NDT)” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

Katsufumi Hashimoto, Tomoki Shiotani and Masayasu Ohtsu

Alessandro Grazzini, Sara Fasana, Marco Zerbinatti and Giuseppe Lacidogna

Elsa Garavaglia, Anna Anzani, Fabio Maroldi and Fabio Vanerio

Claudia Barile, Caterina Casavola, Vincenzo Moramarco, Carmine Pappalettere and

Dimitrios G. Dimogianopoulos, Panagiotis J. Charitidis and Dionysios E. Mouzakis

Jihyun Jun, Young-Dae Shim and Kyung-Young Jhang

Michael Stamm, Helge Pfeiffer, Johan Reynaert and Martine Wevers

Shunmin Yang, Mingquan Wang and Lu Yang

Shuai Han, Heng Li, Mingchao Li and Timothy Rose

Rongning Cao, Meng Ma, Ruihua Liang and Chao Niu

Featured Application: Damage assessment of a reinforced concrete (RC) by means of different

2. Experimental Setup and Monitoring Systems

Figure 3. Fiber optic sensor (FOS) installation (dimensions in cm).

7.1. Total AE Activity

Stages 1-3 Stages 4-6 Stages 7-8 Δt

7.2. The b-Value Analysis

7.3. AE Source Location

-1.6 -1.4 -1.2 -1

8. FOS Strain Measurements

8.1. AE vs. FOS Measurements

1-3 4-6 7-8 1-3 4-6 7-8

1-3 4-6 7-8 1-3 4-6 7-8

8.2. Cracks Pattern

Figure 14. Crack pattern at loading step #7 and numbered AE sensors.

9. DIC Strain Assessment

10. Numerical Analysis

Keywords: non-destructive testing; concrete; impact-echo; elastic wave; SIBIE; 3D visualization;

1.1. Non-Destructive Testing for Concrete

1.2. Impact-Echo Method

Figure 1. Schematic of impact-echo method applied to the testing interior concrete.

1.3. SIBIE Procedure

2.1. Sensor Array with Accelerometers and Laser Doppler Vibrometer

Figure 3. (a) Accelerometer array unit; (b) Overview of on-site implementation.

2.2. Computing Procedure for 3D SIBIE

Figure 5. Volume of interest and cross section for 3D SIBIE procedure.

Figure 7. FFT signal data stacking procedure.

3. Examination of 3D SIBIE Procedure

3.1. Application of Accelerometer Array for Detecting Signals

3.1.1. Reinforced Concrete Bridge Deck

Figure 9. Visual rack inspection procedure.

Figure 10. Cracks filled with resin (a,b).

Figure 11. 3D SIBIE result at intact part.

Figure 12. 3D SIBIE result at damaged part.

3.1.2. Concrete Specimen with Simulated Defect

Figure 14. Result of 3D SIBIE procedure for damage-simulated concrete specimen.

Figure 15. Result of 3D SIBIE procedure for intact concrete specimen.

3.2. Application of Laser Doppler Vibrometer for Detecting Signals

3.3. 3D SIBIE Procedure Improved with Signal Amplification

Figure 18. Principle of PIT (Pile Integrity Test).

Figure 19. Detected signals at accelerometers with a hammer strike.

Figure 20. Signal attenuation along wave propagation in sensor distance.

Figure 21. Improved 3D SIBIE results for damage-simulated concrete specimen.

Figure 22. Improved 3D SIBIE results for intact concrete specimen.

Keywords: non-destructive techniques (NDT); historical buildings; maintenance; stone columns;

Figure 4. Ultrasonic test carried out on columns of Via Crucis, Ghiffa.