Condition Monitoring in Composite Materials: An Integrated Systems Approach

189
Condition monitoring in composite materials:

an integrated systems approach
B Culshaw1, S G Pierce1 and W J Staszekski2

1Department of Electronic and Electrical Engineering, University of Strathclyde, Glasgow, Scotland
2Department of Mechanical Engineering, University of Sheffield
Abstract: This paper presents the results of a collaborative project, the objective of which was to
investigate the potential offered by combining ultrasonic Lamb wave interrogation of composite
material structures, insonifying the whole structure, with a single wavefront integrating optical fibre
detector. The system, which was designed primarily for condition monitoring rather than defect
imaging, proved to be capable of detecting small (centimetre square) delaminations, millimetre-sized
holes and impact damages of a few joules, all with insonification wavelengths of the order of 2 cm.
Data extraction proved to be extremely important since the damage-detection process relies essentially
on analysing scattering signatures rather than attempting to image defects and damage. The system—
dubbed SISCO (Structurally Integrated Systems for comprehensive evaluation of COmposites)—
indicates that self-monitoring structures can be efficiently designed using controlled insonification
and a relatively sparse sensor array, presenting a contrast to the more conventional imaging approach
involving far more complex sensing arrays and detection and signal processing algorithms.
Keywords: condition monitoring, insonification, composite materials, integrated systems,

delamination, algorithm
1 INTRODUCTION
This paper presents an integrated systems concept con-

ceived to facilitate condition monitoring in composite
material systems and investigated as a European
Commission BRITE EURAM funded project—SISCO
(Structurally Integrated Systems for comprehensive
evaluation of COmposites). The paper will summarize
the objectives and outcomes of the SISCO project which
was completed in March 1997 and will also assess the
implications of the results of the project in terms of gen-
eric requirements for future condition monitoring sys-
tems. These needs particularly concern the requirements
for a careful system definition and for a broad interdis-
ciplinary approach to realizing the optimum system Fig. 1 The final goal—a plug in structural test assessing any
configuration. structure to determine its integrity and certify it for
continued service
2 SISCO PROJECT OBJECTIVES as a regular inspection procedure. This signature is

effectively the transfer function (Fig. 2) between an
The underlying concept which SISCO sought to investi- ultrasonic source array designed to generate acoustic
gate was that of a single plug-in access terminal ( Fig. 1) waves within the structure and an optical fibre detection
through which a structural signature could be obtained array designed to integrate the wavefronts from the
acoustic waves. This transfer function represents the
The MS was received on 30 April 1997 and was accepted for publication combined delay and attenuation between the source and
on 26 August 1997. the detector. It is a very sensitive function of the material
I02797 © IMechE 1998 Proc Instn Mech Engrs Vol 212 Part I
Downloaded from pii.sagepub.com at PURDUE UNIV LIBRARY TSS on April 12, 2015
190 B CULSHAW, S G PIERCE AND W J STASZEKSKI
Fig. 2 Basic system configuration for integrated structural monitoring system
structure and composition. Due to echoes, it is also a particularly effective probing system. Lamb waves
influenced by the material outside the direct source– comprise a combination of longitudinal and shear waves,
detector path. It can be measured either as an impulse each with their own particular velocity components [1].
response or in the frequency domain. This paper concen- The requirement for constructive interference between
trates on the former. These integrated wavefronts are these components gives rise to a propagating mode spec-
interpreted using appropriate signal processing software trum (shown in Fig. 3a for aluminium) which depends
in terms of damage or ageing characteristics that alter upon the bulk and shear modulus and the Poisson ratio
the signature, which itself is indicative of the overall of the material. In Fig. 3, the vertical axes present the
structural condition. phase velocity of a particular propagating mode in km/s
The principal technical features of this concept and the horizontal axis is a normalized frequency thick-
include: ness product where the thickness of the plate through
which the mode is propagating in mm is multiplied by
1. The single access/interrogation port, which is
the temporal frequency of the propagating wave in MHz.
designed to operate with electrical and optical fibre
Each individual propagating mode can be characterized
interfaces. This enables two important system param-
by the number of nodes within the thickness of the plate
eters. The mechanical interfaces between the exci-
and by whether the opposite surfaces of the plates are
tation source and the structure and between the
displaced symmetrically (i.e. both going outwards from
structure and the detector remain constant and there-
the plane of the plate at the same time) or antisymmetri-
fore uncertainties in these interfaces, a common cause
cally. In the figure, S is the lowest order symmetric mode
of frustration in ultrasonic testing, are eliminated. 0
and similarly for all the other modes. For test purposes
Secondly, the detection system is immune to the
it is often convenient to arrange for only one propagat-
effects of electromagnetic pick-up from the excitation
ing mode to be allowed within the plate. The S and A
process. With conventional electrical source–detector 0 0
modes can all propagate down to zero frequency whereas,
configurations, electromagnetic radiation between the
for example, the A mode cuts off at around 2 MHz mm.
source and detection stages is frequently troublesome. 1
In the region around 1 MHz mm the S and A modes can
2. The system is designed for wide-area interrogation in 0 0
both propagate but their phase velocities are very different;
contrast to the more usual point-to-point ultrasonic
therefore it is relatively easy to ensure that one or other of
inspection systems. The wavefront integration fea-
the modes is launched through appropriate spatial phase
tures of the detection process and the polar radiation
matching, e.g. by using the wedge transducers shown later
profiles of the source then become important
in Fig. 6. From an inspection perspective, the principal
parameters.
features of Lamb waves are that they propagate through
3. The system is initially designed for condition monitor-
the entire thickness of the plate and they are also sensitive
ing only. It does not attempt to characterize ageing
to the principal material properties through the disper-
processes or defects, to locate them or even to define
sions curve relationships, especially when these are rapidly
the type of defect that is present. However, it tran-
changing [3]. The dispersion curves are also somewhat
spires that some information on these features can be
dependent on structural loading, particularly tensile
obtained from appropriate signature analysis.
stresses. Therefore the capacity for using these dispersion
The applications focus was in composite materials struc- curves as a probe is very significant.
tures, particularly those predominantly in plate or skin The test programme focused upon composite plates
form, wherein the special properties of Lamb waves form and in order to design the ultrasonic inspection tools, a
Proc Instn Mech Engrs Vol 212 Part I I02797 © IMechE 1998
CONDITION MONITORING IN COMPOSITE MATERIALS: AN INTEGRATED SYSTEMS APPROACH 191
(a)
(b)
Fig. 3 (a) Theoretical dispersion curves for phase velocities of Lamb waves in aluminium. (b) Dispersion
curve for a typical composite plate sample with experimental data points superimposed (see reference
[2] for further details)
detailed knowledge of the dispersion curves for Lamb gram is shown in Fig. 3b. (This is a computer program
waves in carbon fibre composite plates is essential. The used to calculate the dispersion curves for Lamb waves
discussion of the situation for the isotropic material— in anisotropic materials.) The simple A and S modes
aluminium—in the previous paragraph extracted the only exist for the zero mode numbers: higher order
principal features of these dispersion curves. For com- modes are hybrids of A and S components. Similar dis-
posite materials, the situation is considerably more com- persion curves are required to design the ultrasonic
plex since the dispersion curves depend upon factors interfaces for each of the composite plate samples
such as the volume fraction of reinforcing fibres and the inspected during the research programme. A quick
orientation of adjacent layers and the number of layers examination of Fig. 3b rapidly confirms the more com-
of composite material within the plate. Composites can plex mode spectrum, especially at low frequencies. The
be highly anisotropic, the most extreme being a uniaxial curves in Fig. 3 have been derived theoretically (Fig. 3a
material, though it could be argued that as the orien- analytically and Fig. 3b using finite element modelling).
tation becomes more random, an isotropic approxi- These dispersion curves have also been confirmed
mation could be reasonable. Even in this case the experimentally [2] using laser-generated ultrasound as a
influence of the boundary condition between adjacent broad temporal and spatial frequency probe.
layers can be extremely important. A typical Lamb wave The SISCO consortium comprised Bertin et Cie
dispersion curve calculated using the POMMARD pro- as the principal contractor who also contributed very
substantially by their skills in modelling the dispersion plate of the order of 600 mm square and later in the
characteristics of composite materials via POMMARD. form of a representative structure including, for example,
Trinity College Dublin, through the Department of bracing members. From each sample a reference signal
Mathematics, modelled the basic scattering character- was extracted and then various forms of defects were
istics of damage and defect sites when insonified with introduced, including holes and impacts. In addition,
the Lamb wave spectrum. Several user partners made controlled delaminations of the order of 1 cm2 were
available different sample structures and different appli- introduced into some samples. The dimensions of typical
cation contexts within which to test the system concept. defects ranged from 1 to 10 mm, with the exception of
These included CASA of Spain, a major aerospace impact damage whose physical extent is far more difficult
manufacturer with interest in carbon fibre reinforced to quantify but is likely to be of the same order. In order
plastics, Vetrotex from France, who manufacture glass to operate within the correct frequency range on the
fibre composites, predominantly for use in automobiles, dispersion curves (i.e. typically around 1 MHz mm) the
GEC Alsthom from the United Kingdom, who fabricate phase velocity for the S mode (which was chosen
0
mixed glass and carbon composites, often including because of its much lower dispersion in the 1 MHz mm
foam inserts, as components for railway vehicles, and region) produces a wavelength in the region of 20 mm.
British Gas, whose interests lie in energy storage tanks. Consequently, any manifestations of defects must be
Strathclyde University contributed both with ultrasonic interpreted through the scattering properties of the
source design and optical fibre detection systems and defect whose dimensions, which are always less than the
performed the system trials. Towards the end of the pro- illuminating wavelength, are too small to enable the
ject, an informal collaboration with Sheffield University reconstruction of images from the modified wavefront.
focusing on specific features within the data analysis A typical experimental arrangement is illustrated in
problem also contributed to the progress of the project. Fig. 4. The samples were fabricated with one or two
embedded optical fibres (see, for example, Fig. 5) which
were equipped with optical connectors to facilitate inter-
3 WHAT SISCO ACHIEVED connection of the optical interrogation and detection
system [4–7]. Insonification was usually based upon
The project entailed testing a significant number (over standard wedge transducers whose impedance was opti-
two dozen) of representative samples with optical fibres mized using a 1 : 3 piezoelectric composite source trans-
embedded within them. These samples included carbon ducer configuration; some tests were conducted with low
fibre sheets of thicknesses up to 3 mm, glass fibre sheets profile surface-mounted transducers ( Fig. 6). These
and tubes of thicknesses up to 10 mm and laminated ultrasonic sources, typically operating at around
carbon–glass structures, both as individual sheets up to 250 kHz except for the very thick samples, had a typical
a few millimetres in thickness and as a part of composite total beam divergence of about 30° full angle and a 3 dB
sandwiches with a hard foam as the filling. Each of these bandwidth of about 10 per cent. While this does not
material systems was initially tested as a straightforward simulate the whole-field insonification, it provides
Fig. 4 Defect detection using optical fibres and long wavelength ultrasound—typical overall source: detec-
tor geometry
Fig. 5 Photograph of a composite plate sample showing optical fibre connectors enabling access to optical
fibre wavefront sensors within the plate
Fig. 6 Conventional wedge ultrasonic sources based on 1 : 3 connectivity composite actuators and new low
profile geometries for future system integration
a suitably straightforward initial probe to evaluate ence [4], incorporated the optical fibre which was
the principle. The source was then stepped along the embedded within the composite as one arm of an all-
sample to eventually produce the desired whole-field fibre actively stabilized homodyne Mach Zehnder inter-
insonification. ferometer. Local changes in the ultrasonic pressure field
The optical fibre detector, described in detail in refer- alter the phase delay of the light propagating through
the fibre and this is detected interferometrically. The technique. Here a five-cycle pulse launched from the
design of the interferometer was optimized by using both transducer (arranged as in Fig. 4) traverses the optical
a low noise laser source and a carefully designed elec- fibre from right to left, which then, if there are no defects
tronic receiver system. The optical fibre detector has very between the fibre and the rear edge, will be reflected from
different properties from those of the more commonly the rear edge to produce what can be termed the ‘back
used piezoelectric devices. First of all, it is not mechan- wall echo’. This is the situation for the upper trace in
ically resonant so its detection bandwidth is extremely Fig. 7. The large pulse at the centre left of the trace is
high with uniform sensitivity to pressure fields over a the first encounter of the Lamb wave with the fibre. The
range for at least d.c. up to 25 MHz, which is well above later, smaller due to attenuation, pulse is the back wall
the frequencies used here. The detection fibre is typically echo. There is evidence of some scattering between the
125 mm in diameter and indeed does exhibit radial initial pulse and the back wall echo, indicative of small
mechanical resonances which, depending upon the fibre inhomogeneities within the material. In the presence of
composition, typically start somewhere in the region of a defect between the fibre and the back wall, two things
25–35 MHz and recur at intervals thereafter. The other happen. A reflected pulse is observed at a characteristic
major contrast between optical fibre detection and piezo- time well before the time at which the back wall echo
ceramic detection systems is in the spatial properties and arrives. In addition, since some energy has been
beam-forming capability of the optical fibre element. In extracted from the pulse on the way to the back wall,
effect the optical fibre integrates the amplitude and phase the back wall echo pulse has been attenuated. This corre-
distribution of the acoustic field along its length. In the sponds to the lower trace in Fig. 7. The traces in Fig. 7
experiments described here, the fibre extends over the show a particularly clear example of this illustrating
full width of the plate sample so that this integration effect of introducing a hole several millimetres in diam-
can extend over either the beam width itself or when the eter a short distance from the optical fibre. The presence
beam is sufficiently diverged, usually after several echoes of the hole is visible as the pulse appearing immediately
from the edges, the fibre will integrate the ultrasonic after the first pulse in the trace. The decrease in the back
wavefront over the full width of the sample. Further, the wall echo is also apparent. In many cases, more subtle
fact that the fibre can be placed within the sample implies signal processing is needed to extract features that are
some mode selectivity. Piezoceramic detectors have not immediately visible on the traces.
almost always been placed at the surface of the sample. This overall approach has several benefits. The initial
Placing the fibre within the sample implies that it will pulse travelling from right to left can act as a continually
not respond to Lamb waves, which have a pressure null updating system calibration and provides a reference for
in the transverse direction at the position at which the the effective transfer function between the optical fibre
fibre detector is placed. and the return echo from the back wall. This reference
Figure 7 shows some typical traces obtained using this can be used in a wide variety of different ways. The
Fig. 7 S Lamb wave reflection from holes: upper trace shows defect-free sample, lower trace shows
0
10 mm hole
simplest is to take the peak amplitude of this reference hole is only 60 mm and the total beamwidth from the
and refer this peak amplitude to the amplitudes of scatter source (30°) produces constructive interference from the
signatures in the gap between the reference pulse and hole and the impact at the ‘shadow’ position, which could
the returning back wall echo. Figure 8 shows the use of account for this apparently anomalous result.
this approach and the relationship between the ratio of Cracks are particularly difficult to detect simply
the reference pulse to the reflected (from the hole) pulse because introducing a controlled crack is inherently a
amplitude and the hole diameter. The insonification difficult process—indeed, within the confines of the pro-
wavelength is approximately 2 cm so that the holes are ject the partners concluded that the best approach was
all a small fraction of the wavelength in dimensions. The to model the effect of cracks using the same analytical
broader bandwidth, especially the higher frequency com- technique as that used to model scatter from holes. This
ponents, accounts for the higher sensitivity for the single- was completed and demonstrated that a through-crack
cycle pulse. of dimensions comparable to an equivalent hole would
Impact damage ( Fig. 9) can be detected using broadly produce broadly similar scattering amplitudes. The mod-
the same technique. The slight difference is that in this elling had proved to be a useful guideline for the case
case it is easier to detect the drop in the back wall echo of scatter from holes and so gives some confidence that
rather than the scatter and this serves as an example of cracks will also be detectable. Experimental confir-
the need, even at a very elementary level, for an adaptive mation, however, remains a task for later.
approach to signal processing depending upon the type Delaminations are viewed by many as the most elusive
of damage to be detected. It also gives an initial hint of of defects within CFRP (carbon fibre reinforced plastic)
the fact that these systems can, in principle, give a good and GFRP (glass fibre reinforced plastic) structures.
indication of not only the presence of the defect but also These reasons are obvious—the delaminations are not
its type. The impacts, the locations of which are indi- visible from the surface and for in-plane propagation of
cated in Fig. 9, were produced using an 8 mm steel ball ultrasonic probe waves usually produce negligible scat-
with an impact energy of 5 J. The trace in the lower half ter. The only possible detection mechanism relies on the
of Fig. 9 is the amplitude of the back wall echo nor- observation that instead of propagating a single Lamb
malized to the first incident pulse on the fibre. A 5 mm wave through the entire sheet, through the delamination,
hole (at 105 mm) provides a behavioural reference. The the Lamb wave is propagated through two sheets, the
dip in the back wall echo after impacts 1 and 3 is large total thickness of which is the same as that of the initial
and evident from the trace. The deviation between the single sheet. A quick examination of the dispersion curve
position of the dip and that of the impact is simply a shows that, depending on the appropriate selection of
consequence of the sample interval (25 mm) by which propagating Lamb wave mode, this will modify the
the source was stepped. Paradoxically, impact 2 gives an phase velocity and therefore result in some wavefront
apparent increase in the back reflection amplitude. curvature changing the relative phase of the launch and
However, the spacing between impact 2 and the reference detected wavefronts. The ultrasonic detection system
Fig. 8 Reflection coefficients from holes in composite sample

Fig. 9 Impact damage detector results: upper is location and lower is back wall echo amplitude
was not phase locked to the source and so such phase Appendix) that were designed to extract the so-called
modifications were not obvious. However, perhaps these novelty index (also defined in the Appendix), the detec-
resultant wavefront curvatures could manifest them- tion of the elusive delamination became very much sim-
selves through the integrating properties of the optical pler (Fig. 11) and allowed quantitative thresholds to be
fibre detector. defined. Figure 11 refers to an artificially introduced
The effect of quite a large delamination (Fig. 10) was delamination 20 mm square in an otherwise undamaged
found to be quite small using the generic classes of detec- sample. Here the discrimination between the presence
tion technique that have been used thus far. Enhancing and absence of the delamination is very good indeed.
the detection sensitivity using correlation processing sys- The processing algorithms have been tuned to extract
tems gave, in this case, little improvement. However, on the delamination characteristics.
turning to more complex and more advanced signal pro- This has been a very brief summary of the experimen-
cessing systems using wavelet transforms (effectively tal programme. The references cited within this summary
a generalization of the Fourier transform—see the present the experimental procedures and the signal
Fig. 10 Detection of large delamination: upper is location and lower is detected signal versus time after
correlation with the input pulse detected by an optical fibre
processing approaches in much more detail. This experi- unlike the more conventional view, in SISCO a limited
mental and analytical data are used to provide a back- number of sensors was used and predetermined prob-
ground for the discussion that follows. ing mechanisms were incorporated. The sensors were
designed specifically to respond to this probe. The
inherent logic in this approach is that outputs from an
4 IMPLICATIONS FOR RESEARCH AND array of direct sensors (for strain, temperature, etc.)
DEVELOPMENT IN SMART STRUCTURES within a structure are meaningless without a knowledge
of the combined strain and temperature driving forces.
The conventional view of the smart structure ( Fig. 12) Additionally, the outputs from such sensor arrays must
is that it comprises a complete structural interrogation also be fed into a comprehensive and viable finite
and evaluation system—sometimes dubbed the ‘sensory element model of the structure under examination. In
structure’—together with a control and response net- SISCO the use of the ultrasonic probe is simply to pro-
work. SISCO is then one of the plethora of projects vide a known consistent and repeatable ‘load’, designed
aimed at establishing the first of these [8]. However, to be sensitive to faults and ageing processes of interest
Fig. 11 Novelty index processing for artificially induced 20 mm×20 mm delamination detection
Fig. 12 Concept diagram for a ‘generic’ smart structure based on condition monitoring using unknown
stimuli
and to extract inferentially from the response to this array is a complex scientific problem which has been
known driving function the appropriate structural data. identified only relatively recently.
From a system design perspective this results in a For the final system user the question over the incor-
number of benefits: poration of smart structures always reduces to econ-
omics. Will the additional cost of incorporating
1. The complexity of the sensor array is significantly
structural integrity monitoring with either integrated or
reduced.
externally triggered corrective action be outweighed by
2. The response of the system is at temporal frequencies
the savings in running costs and/or initial mechanical
which are well outside the range of any thermal or
construction costs, the latter through the ability to
mechanical interfering phenomena. Therefore the
improve design procedures and operate closer to the
measurement process is relatively clean and direct and
structural limit? This issue remains complex, com-
can be tuned to the responses anticipated.
pounded by the complex ownership structure of many
3. The signal processing required when identifying an
major structural artefacts, continually changing legis-
anomalous state is substantially simplified, though at
lation, constraints imposed by the insurers and the social
this stage only for the good/bad decision concerning
and environmental expectations of society at large.
structural integrity. How this could be approached as
While these are critical factors in determining the future
an input to a control system has yet to be examined.
application of smart structures technology, at the techni-
The disadvantages include the need to incorporate cal research and development stages it is far more fruitful
within the structure both the sensor and excitation to focus on initially the system performance and, to some
source arrays and to design the latter for a suitable extent, on the trade-off between performance and system
whole-field examination process. Optimizing this design cost/complexity in implementation. From this perspec-
and coupling this to the strategic location of the sensing tive SISCO has given input into a debate on the relative
merits of monitoring the response of a structure to a and development is an interdisciplinary activity. This
predetermined probe and the alternative approach, project, and indeed other experiences in similar projects,
namely assessing the structural state through monitoring has emphasized this feature. The need for contributors
its response to unknown stimuli which can often be to both convey the impact of the integrated system on
inferred through measurements of temperature, strain their specific interests to others and to appreciate the
fields, etc. impact of their input on the compromises to be intro-
The latter approach requires a significantly more duced by the incorporation of the ‘intelligent’ elements
complex sensor array plus a detection network which is obvious. However, it most definitely requires time and
responds to a wider range of sensor outputs. The former effort on the part of the project team to ensure that the
approach can be based on a much sparser sensor array system is fully understood and effectively conveyed.
tuned to a known generic output characteristic (the Formalization of the project goals—and this applies uni-
characteristics of the driving interrogation force), but versally in the smart structures area—needs this under-
requires the presence of the interrogating function, pref- standing to be coupled to a thorough knowledge and
erably integrated in the structure, in order to ensure that background in the specialist areas concerned, which
its application is consistent over long periods of time. embrace materials science, structural engineering, opto-
From a pure engineering perspective the use of the electronics, ultrasonics, signal processing and appli-
known interrogation function is probably preferable cations engineering.
provided that this function can be designed to highlight
the conditions of interest, primarily because it simplifies
the detection and signal processing. SISCO also high-
lighted the significant contributions that signal pro- 5 CONCLUSIONS: WHERE TO NEXT?
cessing can make to system efficacy, exemplified by the
results in Fig. 11. The SISCO project demonstrated that a combination of
The project highlighted other generic issues. In par- ultrasonic interrogation and optical fibre detection offers
ticular, the need to fully and reliably integrate the source the unique prospect of comprehensive defect detection
and sensor excitation functions into the material was within composite sheet samples and highlights the flexi-
very apparent. The effects of non-uniformity in the bility of the wide-area insonification/integrating detector
interface between the sensor and the structure and of combination. The project also highlighted the necessity
variability in the interface between the source and the for optimizing the signal processing and data-acquisition
structure were evident in the earlier phases of the SISCO media and interpretation procedures. Most certainly,
programme. The interface issue also highlighted a par- considerably more research is required in that area.
ticular benefit of the use of ultrasonic probing signals In the context of ‘smart structures’, the project exem-
compared to more conventional, low-frequency, quasi- plified an alternative approach which incorporates a con-
static strain monitoring. Conventional acrylate-coated trolled probing function with a relatively sparse sensor
optical fibres could be mechanically embedded and were array for structural integrity monitoring, in contrast to
a reliable and efficient interface between the structure the more conventional approach, which uses a very dense
and the sensor for the relatively fast dynamic interro- sensing array but an uncontrolled and unspecified (i.e.
gating signal. In contrast, there is considerable docu- naturally occurring temperature fields or in-use mechan-
mented evidence [9, 10] that quasi-static strain ical loading) probe.
monitoring requires thin coated polyimide interfaces to The project benefited very substantially from the
ensure intimate connection and reliable strain transfer extremely diverse interests and expertise that were avail-
between the sensor element and the structure under test. able within the consortium and also at the later stage
Polyimide-coated optical fibres are a special item and with the informal collaboration in the signal processing
therefore significantly less readily available than acry- area. The success of novel signal processing techniques
late-coated fibres. Additionally, the more intimate con- in extracting elusive signatures was particularly useful in
tact between the polyimide-coated optical fibre and the delamination detection. These most encouraging results
reinforcing fibre within the polymer matrix of the CFRP indicated the potential benefits that could be accrued
or GFRP can result in embedding-induced microbend from pursuing a more comprehensive programme
losses, which are sometimes also a function of applied focused upon integrating the materials and structural
mechanical strain. The ability to introduce connectors engineering with whole-field insonification and wide-
between the structure and the interrogation area integrating detection together with flexible signal
system/excitation system is also very important. The con- processing tools and, in parallel, a comprehensive struc-
nection system shown in Fig. 5, while simply a prototype tural model. The experience gained in integrating these
based upon a modified communications connector, does techniques and advancing their application in a coordi-
demonstrate process compatibility. This is generic to any nated manner as a design process for self-monitoring
structurally integrated sensor system. structures offers great promise, not only in the devel-
It is a well-worn cliché that smart structures research opment of this basic concept but also in the generic
evolution of tools for smart structures, research develop- 11 Chui, Ch. K. An introduction to wavelets. In Wavelet
ment and application. Analysis and Its Applications, Vol. 1, 1992 (Academic Press
Inc., Boston, Massachusetts).
12 Daubechies, I. Ten Lectures on Wavelets, 1992 (SIAM,
Philadelphia, Pennsylvania).
ACKNOWLEDGEMENTS
13 Staszekski, W. J. Wavelet based compression for vibration
analysis. J. Sound Vibr., 1997 (submitted ).
The SISCO project was supported in part by the BRITE
EURAM programme as BRE2.CT94-0990 and by UK
EPSRC GR/J/122603. The authors wish to acknowledge APPENDIX (summarized from reference [6 ])
the significant contributions from the participants in
SISCO (Bertin, INCA, CASA, British Gas, Vetrotex, Wavelet analysis
GEC Alsthom) and from Geof Tomlinson and Keith
Worden of the Department of Mechanical and Classical Fourier analysis provides a spectral represen-
Processing Engineering, Sheffield University. tation that is independent of time. However, many
vibration processes exhibit non-stationary behaviour
which cannot be effectively described using this analysis.
REFERENCES There exist a number of different time-variant methods
for non-stationary processes. These include: adap-
1 Viktorov, I. A. Rayleigh and Lamb Waves—Physical Theory tive techniques (short-time Fourier transform), time-
and Applications, 1967 (Plenum Press, New York). frequency techniques ( Wigner–Ville distribution) and
2 Pierce, S. G., Culshaw, B., Philp, W. R., Lecuyer, F. and time-scale procedures (wavelet transform). A funda-
Farlow, R. Broadband Lamb wave measurements in alu- mental difference between wavelet analysis and other
minium and carbon/glass fibre reinforced composite mate- methods is that instead of seeking to decompose a signal
rials using non-contacting laser generation and detection. into its harmonics, which are global functions, the signal
Ultrasonics, 1997, 35, 105–114. is broken down into local functions called wavelets. The
3 Rogers, W. P. Elastic property measurement using concept of wavelet analysis has many different origins,
Rayleigh–Lamb waves. Res. Nondes. Eval., 1995, 6, from mathematics to signal processing. A brief introduc-
185–208.
tion to the relevant wavelet theory is given in this appen-
4 Pierce, S. G., Philp, W. R., Gachagan, A., McNab, A.,
dix. More detailed analysis can be found in reference
Hayward, G. and Culshaw, B. Surface bonded and embed-
ded optical fibres as ultrasonic sensors. Appl. Optics, 1996, [11].
35(25), 5191–5197.
5 Pierce, S. G., Philp, W. R., Culshaw, B., Gachagan, A.,
(a) Continuous time-scale wavelet transform
McNab, A., Hayward, G. and Lecuyer, F. Surface bonded The wavelet transform is a linear transformation that
optical fibre sensors for the inspection of CFRP plates using
decomposes a given function x(t) into a superposition
ultrasonic Lamb waves. Smart Mater. Structs, 1996, 5,
of the elementary functions g (t) derived from an ana-
776–787. a,b
6 Staszekski, W. J., Pierce, S. G., Worden, K., Philp, W. R., lysing wavelet g(t) by scaling and translation, i.e.
A B
Tomlinson, G. R. and Culshaw, B. Wavelet signal processing t−b
for enhanced Lamb wave defect detection in composite g (t)=g* (1)
a,b a
plates using optical fibre detection. Opt. Engng, 1997, 36,
1877–1888. where * denotes complex conjugation; b is a translation
7 Pierce, S. G., Staszekski, W. J., Gachagan, A., James, I. R., parameter indicating the locality and a is a dilation or
Philp, W. R., Worden, K., Culshaw, B., McNab, A.,
scale parameter. The continuous time-scale wavelet
Tomlinson, G. R. and Hayward, G. Ultrasonic condition
monitoring of composite structures using a low profile
transform is defined as
P A B
source and an embedded optical fibre sensor. Presented at 1 +2 t−b
SPIE 1997, Symposium on Smart Structures and Materials, W x (a, b)= x(t)g* dt
(2)
g √a a
San Diego, California, 2–6 March 1997. −2
8 Culshaw, B. Smart Stuctures and Materials, 1996 (Artech The normalization by factor 1/√a ensures that the inte-
House, Norwood and London). grated energy given by each wavelet g (t) is indepen-
9 Michie, W. C., et al. Optical fibre sensors for monitoring a,b
dent of the dilation a.
of structures (OSMOS ). In SPIE Symposium on Smart
There are many similarities between the classical
Structures and Materials, San Diego, California,
February 1996.
Fourier approach and wavelet analysis. However, the
10 Culshaw, B. and Michie, W. C. Fibre optic strain and tem- major difference is manifested by the space (time) localiz-
perature measurement in composite materials—a review ation of the wavelet transform. The locality of the analy-
of the OSTIC programme. In Recent Advances in Adaptive sis can be seen if the translation in equation (2) is rela-
and Sensing Materials and Their Applications, 1992, ted to moving the window g(t) over the time signal and
pp. 791–810 ( Technomic Press, Lancaster, Pennsylvania). the wavelet scaling (dilation) to the filter frequency
bandwidth scaling. Thus any local features in the signal (Fig. 13), which is defined by
can be identified from the position and scale of the wave-
lets into which it is decomposed. In the frequency y(t)=ei2pf0|t| e−|t|2/2 (6)
domain the wavelet transform also allows the analysis
to focus on a given bandwidth using a different sized The continuous time-scale wavelet transform has a
window depending on the frequency of interest. In con- number of useful properties which are widely studied
trast to Fourier analysis, where the sine and cosine waves and can be found elsewhere [10, 11]. In general, this
are used as a set of basis functions, the wavelet transform transform offers varying time and frequency resolutions
involves an infinite set of possible basis functions. of the analysis. This is not a feature of classical Fourier
For the function g(t) to qualify as an analysing wave- analysis and other time-frequency representations (e.g.
let, it must satisfy the appropriate admissibility con- the Wigner–Ville distribution).
dition, in fact a bandpass filter: The wavelet transform is a two-dimensional complex
function which can be presented graphically in different
P
+2 |g( f )|2
0<C = df <2 (3) ways. The application in this paper involves the represen-
g |f|
−2 tation of the transform in terms of its modulus and
The condition given by equation (3) is required for phase. However, only the modulus is used in the analysis
obtaining the synthesis formula for the wavelet trans- since it better conveys the physical energy of the ana-
form, given as lysed signal.
P P
1 +2 +2 1
x(t)= W (a, b) g*
C g √a (b) Orthogonal wavelet transform
g −2 −2
A B
t−b da db The two-dimensional time-scale framework of the con-
× (4)
a a2 tinuous wavelet transform shows a lot of redundancy
through many wavelet coefficients of equal magnitude.
The locality of the transform also requires that g(t) Thus the extension to this analysis is a process of discret-
decays at infinity: ization of a and b in such a way that a true orthonormal
P
+2 basis of functions g(t) can be obtained. Within a discrete
|g(t)| dt<2 (5) time-scale framework the function g(t) is called an
−2 orthogonal wavelet if the family of wavelets given by
Many practical applications impose, additionally, equation (1) forms an orthonormal basis. A number of
regularity and vanishing moments on the wavelets. The different bases have been proposed to construct ortho-
analysis performed in this paper uses the Morlet wavelet gonal wavelets. The simplest basis can be given by the
Fig. 13 Wavelet analysis functions: (a) Morlet wavelet and (b) fourth-order Daubechies wavelet
well-known Haar function: the analysed signal can be represented as a sum of m

levels:
G
1, for 0∏t∏ 1
2 x (t)= ∑ xm g (t) (12)
H(t)= −1, for 1 ∏t<1 (7) m k m,k
2 k
0, otherwise These levels represent the time behaviour of the signal
The Haar system has a compact support but suffers within different frequency bands and give the contri-
bution to the whole signal energy. Since each wavelet
from a lack of regularity and thus it does not offer effec-
tive signal decomposition. Here the support of the func- level has a constant percentage bandwidth, high levels
tion means the smallest closed set outside which the cover a wide frequency band and therefore the analysis
function vanishes identically. Regularity is a measure of uses high-frequency resolution. Each level is determined
the smoothness of a function. Smoothness of a function by a number of wavelets, which can be fitted into the
corresponds in turn to the decay of its Fourier transform. interval [0, 2r−1]. It can be shown that level m fits 2m
The natural extension and improvement of the analysis Daubechies wavelets within this interval. The orthogonal
can be obtained with the Daubechies wavelets [12]. For wavelet transform and orthonormal wavelet bases have
each integer r these wavelets construct an orthonormal attracted a great deal of interest. For more details the
basis and are defined in the form of the functions with reader is referred to reference [11].
binary dilation and dyadic translation:
g (t)=2m/2g (2mt−k) (8)
m,k r Wavelet novelty measure
Except for the trivial case r=1, for which the orthog-
onal basis reduces to the Haar wavelet analysis, none of Daubechies wavelets (12 provide an effective tool for
the Daubechies wavelets can be given in an explicit form. data compression, which can be used for feature selec-
All can be determined recursively from the so-called sca- tion procedures for fault detection [13]. This method is
ling functions corresponding to a finite impulse response related to the well-known novelty approach. The wavelet
( FIR) filter. The scaling function must be a solution of novelty measure establishes a description of normality
the functional equation: using features representing undamaged conditions and
then tests for abnormality or novelty.
P
2r−1 +2 In wavelet novelty detection the data are compressed
Q(t)=√2 ∑ c Q(2t−k), Q(t) dt=0 (9)
k in order to extract important features relevant to
k=0 −2
detecting faults. The algorithm of compression is based
An example of the fourth (r=2)-order Daubechies
on the orthogonal wavelet transform. The data com-
wavelet function is given in Fig. 13b. This fractal shape
pression is achieved by setting wavelet coefficients xm
function has the ability to represent any signal with much k
given in equation (11) below a threshold T to zero. The
better accuracy than the Haar function.
bigger the threshold the higher the compression ratio.
The orthogonal wavelet transform can now be defined
This procedure determines the most important features
as
of the data related to the fault. The reconstructed
P
+2 (uncompressed ) data are then used to form the template
xm = x(t)g (t) dt (10)
k m,k representing the normal (no fault) condition. The nov-
−2 elty index is defined as a weighted distance between the
and is seen as an extension to the continuous time-scale template a(t) and vector b(t):
wavelet transform, given by equation (2), with an ortho-
gonal basis of decomposition, binary dilation and dyadic d2 =da(t)−b(t)d =[a(t)−b(t)]TR[a(t)−b(t)]
a,b R
translation. By analogy to the continuous wavelet trans- (13)
form, the orthogonal wavelet synthesis formula is given where b(t) is the reconstructed vector of vibration data
by from which the working condition is to be evaluated, xT
x(t)= ∑ ∑ xm g (t) (11) denotes the transpose of vector x and R is a matrix that
k m,k
m k induces the norm d d . If R is a unit matrix I, the scalar
R
There exist many different ways of displaying graphi- measure given by equation (12) is the Euclidean dis-
cally the orthogonal wavelet transform. Summation of tance. The novelty index accepts or rejects the hypothesis
the magnitude of the wavelet coefficients leads to the that the vector b(t) belongs to the same class of vectors
concept of wavelet levels. After wavelet decomposition, as the template a(t).

Condition Monitoring in Composite Materials: An Integrated Systems Approach

Uploaded by

Copyright:

Available Formats

Condition Monitoring in Composite Materials: An Integrated Systems Approach

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

Condition Monitoring in Composite Materials: An Integrated Systems Approach

Uploaded by

Copyright:

Available Formats

189

Condition monitoring in composite materials:

B Culshaw1, S G Pierce1 and W J Staszekski2

Keywords: condition monitoring, insoniﬁcation, composite materials, integrated systems,

This paper presents an integrated systems concept con-

2 SISCO PROJECT OBJECTIVES as a regular inspection procedure. This signature is

Fig. 2 Basic system conﬁguration for integrated structural monitoring system

Fig. 8 Reﬂection coeﬃcients from holes in composite sample

well-known Haar function: the analysed signal can be represented as a sum of m

You might also like