Article

Real-Time Detection of Faults in Rotating Blades Using

Frequency Response Function Analysis
Ravi Prakash Babu Kocharla 1 , Murahari Kolli 2 and Muralimohan Cheepu 3,4, *

1 Department of Mechanical Engineering, Prasad V. Potluri Siddhartha Institute of Technology,

Vijayawada 520007, India
2 Department of Mechanical Engineering, Lakireddy Bali Reddy College of Engineering,
Mylavaram 521230, India
3 Department of Materials System Engineering, Pukyong National University, Busan 48547, Republic of Korea
4 STARWELDS Inc., Busan 46722, Republic of Korea
* Correspondence: [email protected]; Tel.: +82-10-2714-3774

Abstract: Turbo machines develop faults in the rotating blades during operation in undesirable
conditions. Such faults in the rotating blades are fatigue cracks, mechanical looseness, imbalance,
misalignment, etc. Therefore, it is crucial that the blade faults should be detected and diagnosed in
order to minimize the severe damage of such machines. In this paper, vibration analysis of the rotating
blades is conducted using an experimental laboratory setup in order to develop a methodology to
detect faults in the rotating blades. The faults considered for the study include cracks and mechanical
looseness for which dynamic responses are recorded using a laser vibrometer. Analysis has been
carried out by comparing the frequency response function spectrums of the fault blade with those
of the healthy blade related to the resonance frequency. The Internet of Things and wireless sensor
networks are implemented to transmit the measured data to the cloud platform. A support vector
machine algorithm is used for preparing the learning model in order to extract and classify the faults
of the rotating blades. It can be clearly seen from the results that there is variation in the frequency
response function spectrums of healthy and faulty conditions of the rotating blades.

Keywords: fault detection; cracks; mechanical looseness; vibration spectrums

Citation: Kocharla, R.P.B.; Kolli, M.;

Cheepu, M. Real-Time Detection of 1. Introduction
Faults in Rotating Blades Using
Fault detection and diagnosis are concepts that have existed from the inception of
Frequency Response Function
machines. In the early days, the view of the manufacturers and the users towards fault
Analysis. Appl. Mech. 2023, 4,
diagnosis was that the safety and protection of a machine ensured its reliable operation.
356–370. https://doi.org/10.3390/
However, nowadays machines are expected to perform multiple tasks which make their
applmech4010020
operations more complex. Therefore, there comes the requirement to improve the fault
Received: 31 January 2023 diagnosis methods.
Revised: 20 February 2023 Rotating blades are important components in machines such as turbines, fans, pumps,
Accepted: 9 March 2023 etc., used in modern mechanical industries. These blades convert some form of energy into
Published: 15 March 2023 mechanical power. For example, wind turbines convert wind energy, when passed over the
blades, into mechanical power by developing aerodynamic force. Likewise, steam turbine
blades make the shaft to rotate when high pressure steam passes over them. Additionally,
Copyright: © 2023 by the authors.
equipment such as draught fans, blowers and compressors make use of the kinetic energy
Licensee MDPI, Basel, Switzerland. of blades in order to increase the fluid outlet pressure. Such blades can become damaged
This article is an open access article during critical operating conditions and have adverse effects on the safety and reliability
distributed under the terms and of the machines. Therefore, it becomes evident that developing a condition monitoring
conditions of the Creative Commons method that can detect the faults during blade operation is necessary. Every fault in the
Attribution (CC BY) license (https:// rotating blades may cause unwanted vibrations in the machine and carry the information
creativecommons.org/licenses/by/ of the fault. Therefore, vibration-based condition monitoring of the blades from the time of
4.0/). their inception can help us to detect blade faults at the earliest.

Appl. Mech. 2023, 4, 356–370. https://doi.org/10.3390/applmech4010020 https://www.mdpi.com/journal/applmech

Appl. Mech. 2023, 4 357

Sometimes, rotating blades also need to be monitored for defects developed during the
manufacturing process or made by less qualified personnel. Therefore, quality management
is essential during product development along with condition monitoring during the
performance stage of the blades. Any methodology developed for condition monitoring
can be used for quality assurance, along with the quality control of the blades. Emerging
technologies such as the Internet of Things and machine learning techniques, along with
cloud computing, can be implemented in developing real time fault detection methods
that can achieve efficient quality management of the blades. In doing so, sensors play
a vital role in measuring and acquiring data from the rotating blades. Pennacchi and
Vania [1] have conducted experiments on a flexible coupling connecting two shafts in order
to analyze the effect of angular misalignment. It is concluded that the vibration-based
condition monitoring of a rotor can be simulated to predict faults in rotating machines.
Djaidir et al. [2] have analyzed the dynamic behavior of gas turbine rotors with defects
using mathematical techniques of vibration response. In order to validate the usefulness
of the mathematical models, experiments are conducted and it has been found that the
developed approach is very useful in the fault diagnosis of gas turbines. Guan et al. [3] have
determined the rotational frequency from the derived vibration equations of misalignment
states (i.e., angular and offset) through shaft dynamic models and verified the results with
simulation and experimental results. Al Adawi and Ramesh Kumar [4] have investigated
the cause of high vibrations developed in the gas turbine, generator rotor and gear box of a
power station. This could help to detect the fault causing the vibration and prevent any
chance of catastrophic failure of the components.
Chaun Li et al. [5] investigated the influence of different domains of deep learning
algorithms in order to improve fault diagnosis capability. Andre et al. [6] developed a
condition monitoring system which is based upon cloud computing that includes data
mining and machine learning techniques. Failures in rotating machines are identified by
utilizing the data available in the real time Ethernet network, which will reduce the number
of sensors. This helps in reducing the maintenance cost of sensors, investment towards
signal processing hardware and developing a user-friendly system that can be understand
by the industry personnel. Marwin et al. [7] evaluated real time data using a learned model
for the predictive maintenance of a multipurpose machine. The clustering approach is
used for the segmentation of raw data, and then data are processed for predicting the
degradation state of the machine using a machine learning classifier. Wang et al. [8] have
proposed a new fault detection method based on a convolutional neural network (CNN)
using bottle neck optimization of multi-vibration signals of a wind power test rig. The
multi-vibration signals obtained from the sensors are first converted into 2D images and
then features are extracted using CNN technique, and then the faults are detected.
Min Xia et al. [9] proposed a framework in which IoT and cloud computing can be
assisted in the continuous development of a design in order to improve its efficiency and
cost effectiveness. Siliang Lu et al. [10] proposed a condition monitoring method in order to
diagnose the faults in motor bearings. This method works on the undersampling principle
as it reduces the sampling rate, data transmission time and power consumption while
measuring and collecting the different signatures of the vibration signals of the defective
bearings. Adrian Stetco et al. [11] presented a review on condition monitoring of wind
turbines for the detection of blade faults using machine learning models. It was proposed
that support vector machines, neural networks and decision trees are the most suited
machine learning algorithms in order to select and extract the features of the fault.
Andreas Theissler et al. [12] presented an overview of the most suitable machine learning
techniques used for the predictive maintenance of automotive applications. Yoonjae Lee et al. [13]
proposed a feature variable dimensional coordination method for fault diagnosis that
can result in high accuracy with less processing time. Mohanta et al. [14] reviewed the
detection of several sources of faults such as imbalance, bearing problems, wicket gate
problems and shear pin failure in hydro power stations by monitoring vibration signals of
turbine-generator sets. Chenyang Li et al. [15] proposed a condition monitoring system that
Appl. Mech. 2023, 4 358

utilizes the vibration data of machinery equipment measured and transmitted by a terminal
device based upon the IoT. Using these data, a diagnosis model was trained that can detect
the imbalance caused by a blade in a fan. Alexandre et al. [16] evaluated the performance
of various machine learning algorithms by acquiring data during predictive maintenance
in order to diagnose the rotating machines. In recent years, the condition monitoring
of industrial equipment has been carried out by making use of the concepts available
in emerging technologies such as the Industrial Internet of Things (IIoT) and the Fourth
Industrial Revolution. The sensors capable of acquiring the data in the form of images,
graphics and digitalization are processed to identify the condition of the equipment [17].
Omar et al. [18] presented a review of the challenges facing the condition monitor-
ing and fault diagnosis of induction motors using acoustic emissions. This review was
focused mainly on the faults in bearings, rotor and stator. Shiqing et al. [19] investigated
the capability of trumpet-shaped acoustic metamaterials with a high refractive index in
order to magnify the specific frequency range of the acoustic signals generated during
the operation of machines. Experimental studies of such metamaterials have proved to
have more efficient sensing with high performance when diagnosing faults in machinery.
Divya et al. [20] proposed an intelligent fault detection system to detect faults in an air
compressor. The system is based on the real-time empirical acoustic sensor data acquired
from the key sensitive points that are identified by performing sensitive point or position
analysis. The faulty features are determined using a wavelet-based approach, and a model-
based supervised classifier is employed to predict the fault class. Lu et al. [21] evaluated the
condition monitoring performance of a two-stage helical gearbox using airborne acoustics
in conjunction with modulation signal bispectrum analysis. Qurthobi et al. [22] conducted
an evidence-based systematic review on the different approaches and techniques used for
mechanical failure analysis based on acoustic signals. Caleb et al. [23] proposed a fault
diagnosis method for drills based on principal component analysis and artificial neural
networks. The sound data are collected for the undamaged and defective drills, and further
processed to identify the faults. Martin-del-Campo and Sandin [24] presented the measures
and heuristics for the automatic scoring of the acoustic emissions developed due to faulty
condition in the bearings of a rotating machine.
Saeed et al. [25] proposed a fault diagnosis approach in rotating equipment based on
permutation entropy, signal processing and artificial intelligence. Permutation entropy
is implemented to identify the faulty state from the extracted frequency components
using wavelet packet transform and envelope analysis of the vibration signals. The multi-
output adaptive neuro-fuzzy interface system is implemented to classify the fault type.
Scislo [26] introduced a vibration measurement system that works on a completely non-
contact approach, which includes sound pressure change as the excitation source and
the 3D Laser Doppler Vibrometer as the measurement system. The Industry 4.0 data
acquisition and analysis systems enabled the development of quality assurance and control
of the production process. On the other hand, the fabrication method of the blades and
wheels also might have an effect on the performance and quality aspects [27–32]. During
fabrication or manufacturing of the objects the induced stresses, heat and powder forms
of additive methods has an effect even at micron levels, which are sensitive to centrifugal
forces and torque [33–37]. Brito et al. [38] proposed a new approach for fault detection and
diagnosis in rotating machines in which an unsupervised anomaly detection algorithm was
used to identify and classify the faults through vibration analysis. Root cause analysis was
performed when unsupervised classification was not possible to classify the faults.
It can be clearly understood from the literature that the power generating machinery
uses blades that may fail during operation under critical loading conditions. Therefore,
there exists a need to develop a condition monitoring technique in order to diagnose blade
faults. The aim of this paper is to detect the faults of the rotating blades using vibration-
based condition monitoring. In order to achieve this, a rotating blade test rig was designed
and manufactured for healthy and faulty conditions. A laser beam vibrometer was used
to measure the vibration signals of the rotating blades and then processed through a Fast
 there exists a need to develop a condition monitoring technique in order to diagnose blade
faults. The aim of this paper is to detect the faults of the rotating blades using vibration-
based condition monitoring. In order to achieve this, a rotating blade test rig was designed
Appl. Mech. 2023, 4 and manufactured for healthy and faulty conditions. A laser beam vibrometer was used 359
to measure the vibration signals of the rotating blades and then processed through a Fast
Fourier Transform (FFT) analyzer in order to extract the frequency response function
(FRF)
Fourierspectrums.
TransformThe (FFT)support
analyzervector machine
in order learning
to extract techniqueresponse
the frequency was used in order
function (FRF)to
classify
spectrums. the The
faults based vector
support on the machine
extractedlearning
featurestechnique
of the rotating blades.
was used The work
in order pre-
to classify
sented in this
the faults basedpaper is believed
on the extractedtofeatures
be useful in finding
of the rotatingthe bladeThe
blades. faults of machinery
work presented in oper-
this
paperin
ating is remote
believed to be with
places useful in finding
minimum the blade
human faults of machinery operating in remote
intervention.
places with minimum human intervention.
2. Materials and Methods
2. Materials and Methods
In certain applications of rotating blades, such as thermal power plants and aerospace
In certain
jet engines, applications
turbine of subjected
blades are rotating blades,
to highsuch as thermalalong
temperatures power plants
with and aerospace
centrifugal loads.
jet engines, turbine blades are subjected to high temperatures along with centrifugal
However, as these blades are coated with thermal resistant coatings, they are able to with- loads.
stand thermal loads. Additionally, the effect of the elevated temperatures on the blades to
However, as these blades are coated with thermal resistant coatings, they are able is
withstand thermal
appreciable loads.the
only when Additionally, the effect
blade material has of the elevated
crystalline temperatures
defects such as on the blades
dislocations.
is appreciable
Therefore, only when
temperature the blade material
environment has crystalline
is not considered in this defects
work and such
it isasassumed
dislocations.
that
Therefore, temperature environment is not
faults are developed purely due to mechanical loads. considered in this work and it is assumed that
faultsAnareexperimental
developed purely
rig is due to mechanical
designed loads.
in the resemblance of rotating blades in order to
An experimental rig is designed in the resemblance of rotating blades in order to
develop a health monitoring system to detect the blade faults. A laser beam vibrometer is
develop a health monitoring system to detect the blade faults. A laser beam vibrometer
used to measure the vibration response so that investigations can be carried out by com-
is used to measure the vibration response so that investigations can be carried out by
paring measured FRF spectrums of the healthy and faulty blades. The IoT helps the meas-
comparing measured FRF spectrums of the healthy and faulty blades. The IoT helps the
ured data to be transmitted to the cloud platform, wherein the classification of the fault is
measured data to be transmitted to the cloud platform, wherein the classification of the
achieved by applying the suitable machine learning technique.
fault is achieved by applying the suitable machine learning technique.
2.1.
2.1. Manufacturing
Manufacturing of of the
the Experimental
Experimental Rig Rig along
along with
with Instrumentation
Instrumentation
For
For this research study, a test rig was manufactured in
this research study, a test rig was manufactured in accordance
accordance withwith modelled
modelled
dimensions
dimensions as shown in Figure 1. Finite element based modal analysis was carried
as shown in Figure 1. Finite element based modal analysis was carried out
out
using
using ANSYS
ANSYSR15.0 R15.0software
softwaretotodetermine
determine the natural
the naturalfrequency
frequencyof the rotating
of the blades.
rotating To
blades.
validate the the
To validate theoretical
theoreticalresults obtained
results obtainedfromfrom
the ANSYS
the ANSYS software, experiments
software, werewere
experiments con-
ducted
conductedon the testtest
on the rig.rig.
Knowing
Knowing both
boththethetheoretical
theoreticaland
andexperimental
experimentalresults
resultswith
with less
less
deviation, it was decided to carry out further analysis using the test rig as it can
deviation, it was decided to carry out further analysis using the test rig as it can be justified be justified
with
with real
realtime
timeoperating
operatingconditions.
conditions.TheThe blade faults,
blade suchsuch
faults, as cracks and mechanical
as cracks loose-
and mechanical
ness, were developed within the test rig for which vibration signals are
looseness, were developed within the test rig for which vibration signals are measured measured in order
in
to develop
order the blade
to develop the health monitoring
blade health system.
monitoring system.

Figure 1. (a) Geometric and (b) finite element models of the shaft.
Figure 1. (a) Geometric and (b) finite element models of the shaft.

The boundary condition for the rotating blades is shown in Figure 2. For shell 63
(4 node shell element), every node has 6 degrees of freedom, 3 translations, Ux, Uy and Uz,
and 3 rotations: Rotx, Roty and Rotz. Ux, and Uy degrees of freedom of the nodes, which
are at supports, are constrained to simulate the actual condition of rotating blades.
 The boundary condition for the rotating blades is shown in Figu
Appl. Mech. 2023, 4, FOR PEER REVIEW node shell element), every node has 6 degrees of freedom, 3 translatio 5

Appl. Mech. 2023, 4 and 3 rotations: Rotx, Roty and Rotz. Ux, and Uy degrees of freedom 360 o
are at supports, are constrained to simulate the actual condition of rot
The boundary condition for the rotating blades is shown in Figure 2. For shell 63 (4
node shell element), every node has 6 degrees of freedom, 3 translations, Ux, Uy and Uz,
and 3 rotations: Rotx, Roty and Rotz. Ux, and Uy degrees of freedom of the nodes, which
are at supports, are constrained to simulate the actual condition of rotating blades.

Figure 2. Boundary conditions of the rotating blades.

Figure
Figure 2. Boundary
2. Boundary conditions
conditions of the rotatingof the rotating
blades. blades.
The experiments were conducted using a set-up as shown in Figure 3a, comprising
The
of shaft, experiments
coupling, hub,were conducted
blades, motor, using a set-up
etc. FRF signalsas of
shown in Figure
vibration are 3a, comprising
measured by laser
of shaft,The experiments
coupling, hub, blades, were
motor, conducted
etc. FRF signals using
of a
vibration set-up
are as
measured
vibrometer, as shown in Figure 3b. In order to fetch the acceleration signal of vibration, shown
by laser inaFig
vibrometer, as shown in Figure 3b. In order to fetch the acceleration signal of vibration, a
ofwas
laser shaft, coupling,
focused hub, shaft
on the rotating blades,
drivenmotor, etc. These
by a motor. FRF signals
signals were of vibration
obtained forare
laser was focused on the rotating shaft driven by a motor. These signals were obtained for
each case, i.e., healthy,
vibrometer, crack and
as shown mechanical
inmechanical
Figure 3b. looseness.
In order Thetoblade
fetch faults, such as cracks si
each case, i.e., healthy, crack and looseness. The blade faults,the
suchacceleration
as cracks
and
andmechanical
mechanical looseness,
looseness,onwere
were developed
developed within
within the test
the driven rig for
test rig forby which
which vibration
vibrationThese signals
signals
were
laser was focused the rotating shaft a motor. signal
weremeasured
measuredin inorder
order to developthe
to develop theblade
bladehealth
health monitoring
monitoring system.
system. FRFFRF spectrums
spectrums
wereeach
were case,and
extracted,
extracted, i.e.,then
and healthy,
then crack
comparisons
comparisons were
were and
made mechanical
made between
between looseness.
healthy
healthy andandfaultfaultThe blade fa
conditions.
conditions.
and mechanical looseness, were developed within the test rig for whic
were measured in order to develop the blade health monitoring syste
were extracted, and then comparisons were made between healthy an

Figure 3. 3.(a)(a)Experimental
Figure Experimental rig
rig and (b) laser
and (b) laservibrometer.
vibrometer.

The detailed discussion of the experiment set-up is as follows. A shaft of 20 mm di-

ameter and 600 mm length, made of steel, was supported by two ball bearings at its ends.
The blades were attached to a hub, which was centrally located on the shaft. This entire
assembly was connected to the motor with a flexible coupling designed to transmit torque
Appl. Mech. 2023, 4 361

The detailed discussion of the experiment set-up is as follows. A shaft of 20 mm

diameter and 600 mm length, made of steel, was supported by two ball bearings at its ends.
The blades were attached to a hub, which was centrally located on the shaft. This entire
assembly was connected to the motor with a flexible coupling designed to transmit torque
without any axial, radial or angular misalignment. The rotational motion of the shaft was
attained by a power-driven motor of Kirloskar make of 1 HP and 3000 rpm speed. The
shaft-motor assembly was mounted on a supporting frame in alignment with the shaft
center using bolts and nuts.
Frequency response function spectrums are the measurement of vibration signals of the
rotating blades. A laser vibrometer is a non-contact type vibration measuring instrument
and is used to measure the frequency spectrum of the rotating blades. A Fast Fourier
Transform analyzer is connected to the vibrometer in order to convert the vibration signal
into a convenient form which can be later assessed for detecting the faults of the rotating
blades. The experimental work of this research was conducted on the test rig, which is
located in the Machine Dynamics laboratory at PVPSIT, India.
Non-contact type sensors can reduce the use of a greater number of contact sensors
to measure the detailed vibration response of a rotating blade. Non-contact type sensors
such as laser vibrometers are available in 1D or 3D based upon the detailed vibration
measurement. The 3D laser vibrometers provide more detailed vibration information
compared to other vibrometers. Therefore, the selection of the vibrometer depends upon
the amount of vibration details. For the fault detection of rotating blades, 1D vibrometer
information is sufficient to perform frequency response function analysis.
The laser vibrometer is of Polytec make, consisting of an OFV511 compact sensor
head along with a laser unit, as shown in Figure 3b. This laser unit delivers a beam
of helium neon laser light of 316 nm through an optical fiber and is received by a high
precision interferometer available in the head. The laser beam is split by a beam splitter
into two beams, namely a reference beam and a measurement beam. The measurement
beam is then focused onto the rotating blades by making it pass through a second beam
splitter. The reflected measurement beam is then merged with the reference beam by the
third beam splitter and focused on the photo detector. A Doppler frequency shift of the
measurement beam is manifested with the change in the optical path length per unit of
time. An acousto-optic modulator placed in the reference beam creates the frequency shift,
and the modulation frequency is directly proportional to the velocity of the rotating blades.

2.2. Modal Testing

The process of determining the modal characteristics is referred to as modal anal-
ysis. The modal characteristics include natural frequencies, mode shapes, stiffness and
damping of the vibrating component. The knowledge of these modal characteristics of
any component helps to predict its dynamic behavior. The modal characteristics contain
the most vital information in the design of rotating machinery. They are required for
carrying out the dynamic analysis of components in the finite element environment by
knowing vibration response. The vibration responses of the components are then extracted
from modal testing in the form of FRFs to obtain natural frequencies. The known natural
frequencies are helpful to avoid resonance during operation, and in turn help to elevate the
operational life of the components. Additionally, the information of the vibration response
is utilized to detect the faults in the components with reference to the vibration response of
the healthy component.
The natural frequencies of the rotating blades are determined by conducting the modal
testing by using the FRF test. The experimental rig for the modal testing of the rotating
blades is shown in Figure 4. The laser vibrometer measures the vibration responses, which
are analyzed using FFT and recorded on a computer. From the FRFs of the accelerated
blade, natural frequencies are identified.
Appl.
Appl.Mech.
Appl. Mech. 2023,44,4,FOR
Mech.2023,
2023, FORPEER
PEERREVIEW
REVIEW 36277

Figure
Figure 4.4.Cracked
Figure4. Crackedblade.
Cracked blade.
blade.

2.3.
2.3.Online
OnlineCondition
ConditionMonitoring
Monitoring
The
Themain
mainaim aimofofthe
thecondition
conditionmonitoring
monitoringisisto toprevent
preventmachines
machinesfrom fromexperiencing
experiencing
catastrophic
catastrophicfailure
failureby bythe
theearly
earlydetection
detectionof ofthe
thefaults.
faults.This
Thishelps
helpsto todiagnose
diagnosethe thefaults
faults
and eliminate further damage
and eliminate further damage to the to the machine.
the machine. Therefore, an effective structural
machine. Therefore, an effective structural health health
monitoring
monitoring system
systemisis
monitoringsystem neededfor
isneeded
needed forthe
for thedetection
the detectionand
detection and
and diagnosis
diagnosis
diagnosis ofof
of thethe
the faults
faults
faults in in
inrealreal
realtime
timetime
in-
in-
industrial
dustrial applications.
dustrialapplications.
applications.In In
Inthis this
thissection,section,
section,aabrief a brief description
briefdescription
descriptionof ofthe of the
theproposed proposed
proposedcondition condition
conditionmoni-
moni-
monitoring
toring
toringsystem
systemsystem
based
based based
on
onrealontime
real real online
time time
onlineonline technology
technology
technology is given
isisgiven
given for for detection
forfault
fault fault detection
detection and and
anddiag-
diag-
diagnosis.
nosis. Wireless
nosis.Wireless
Wireless sensor
sensorsensor networks
networks
networks (WSNs)
(WSNs) (WSNs)
and
andthe
theand
IoTthe
IoT areIoT
are are utilized
utilized
utilized in
inthe in the proposed
theproposed
proposed method-
method-
methodology,
ology,
ology,as as is shown
asisisshown
shown in
inFigure in Figure
Figure 5.5.The 5. Thesystem
Theentire
entire entire
systemsystem
designdesign
design comprisescomprises
comprises the the disposition
thedisposition
disposition of
ofthe
the
of the
WSNs, WSNs,
WSNs,monitoringmonitoring
monitoringplaceplaceandplace
andcloud and cloud
cloudplatform. platform.
platform.

Figure
Figure 5.5.Schematic
Figure5. Schematicdiagram
Schematic diagramof
diagram the
ofthe
of online
theonline condition
onlinecondition monitoring
conditionmonitoring system.
monitoringsystem.
system.

2.3.1.
2.3.1.Disposition
Dispositionof ofthe WSNSystem
theWSN System
In applications
Inapplications evolving
applicationsevolving for
evolvingfor forautomation,
automation,
automation, attention
attention
attention isis
being
isbeing
beinggrabbed
grabbed
grabbedglobally
globally by by
globally wireless
bywire-
wire-
communication
less
lesscommunication
communication technologies.
technologies.One One
technologies. such
Onesuch standard
suchstandard adopted
standard adoptedin thein
adopted proposed
in the
theproposedsystem
proposed is IEEE
system
system isis
803.15.4.
IEEE
IEEE803.15.4.Figure
803.15.4. 6 shows
Figure
Figure the functional
66shows
shows the
thefunctional diagram
functional diagramof the
diagram ofWSNs
of the
theWSNs system
WSNs proposed
system
system proposedin this
proposed in
in
work.
this The
work. measurements
The measurements of the
of condition
the conditionmonitoring
monitoring parameters
this work. The measurements of the condition monitoring parameters such as vibration parameters such as
such vibration
as and
vibration
natural
and
andnaturalfrequencies
natural frequenciesare obtained
frequencies are by using
areobtained
obtained by ausing
byusingvibration sensorsensor
aavibration
vibration focused
sensor on the on
focused
focused rotating
onthe blades
therotating
rotating
of the
blades machine.
blades of ofthe When
themachine. several
machine.When sensor
Whenseveral nodes
severalsensor are
sensornodes used
nodesare for fault
areused
usedfordetection
forfault in more
faultdetection
detectionin than one
inmore
more
machine,
than
than one it is required
one machine, that one
machine, itit isis required of the
required that sensor
that oneone of nodes
of the becomes
the sensor
sensor nodes the basic
nodes becomes node,
becomes the which
the basic helps
basic node,to
node,
handle
which
whichhelpsthe allocation
helpsto tohandle of
handlethe the resources,
theallocation
allocationof blending
ofthe of
theresources, the data
resources,blending and
blendingof scheduling
ofthe
thedata
dataandof the network
andscheduling
scheduling
seeing
of
ofthe the
thenetworkprior condition
networkseeingseeingthe of data
theprior transmission.
priorcondition
conditionof ofdataThis functionality
datatransmission.
transmission.This implementation
Thisfunctionality
functionality prevents
imple-
imple-
unnecessary
mentation
mentationprevents data and decreases
preventsunnecessary
unnecessarydata the flow
dataand of data
anddecreases in
decreasesthe the wireless
theflow
flowof communication
ofdata
dataininthe
thewirelessnetwork,
wirelesscom-com-
thus increasing
munication
munication the life
network,
network, of the
thus
thus sensors.the
increasing
increasing Sensors
the life ofsend
life of the the vibration
the sensors.
sensors. Sensors
Sensorsinformation
send
send the of all the
the vibration
vibration
machines
information
information to the
of basic
ofall
all thenode
the and then
machines
machines to
tothe
thethe datanode
basic
basic are transmitted
node and
andthen
thenthe indata
the the form
data are of packets in
aretransmitted
transmitted to the
inthe
the
monitoring
form
formof place.
ofpackets
packetsto Then,
tothe the
themonitoringinterface
monitoringplace. of the
place.Then, coordinator
Then,the theinterface node
interfaceof with
ofthe the expert
thecoordinator
coordinatornode system
nodewith is
with
carried out through serial port communication.
the
theexpert
expertsystem
systemisiscarried
carriedout outthrough
throughserialserialportportcommunication.
communication.
Appl. Mech.
Appl. Mech. 2023,
2023, 44, FOR PEER REVIEW 3638

Figure 6. Data transmission through IoT.

IoT.

2.3.2. Fault Diagnosis Method in Rotating Blades

2.3.2. Fault Diagnosis Method in Rotating Blades
The sensor used in this work is a laser vibrometer. Its function is to measure the
The sensor used in this work is a laser vibrometer. Its function is to measure the vi-
vibration signals of the healthy and faulty conditions in the blades. These signals will
bration signals of the healthy and faulty conditions in the blades. These signals will help
help in understanding the frequency response function spectrum for each condition of the
in understanding the frequency response function spectrum for each condition of the
blades, so as to detect faults in real time applications. This system is used to process the
blades, so as to detect faults in real time applications. This system is used to process the
data acquired from the sensors and helps to detect and diagnose the faults in the machines.
data acquired from the sensors and helps to detect and diagnose the faults in the ma-
When the spectrums extracted by this system increase, it becomes difficult to classify the
chines. When the spectrums extracted by this system increase, it becomes difficult to clas-
detected faults and it is necessary to apply the dimensionality set of the spectrum. Therefore,
sify the detected faults and it is necessary to apply the dimensionality set of the spectrum.
it is proposed to classify the faults by implementing machine learning techniques such
Therefore, it is proposed to classify the faults by implementing machine learning tech-
as support vector machines, neural networks, decision trees, etc. This process of fault
niques such as support vector machines, neural networks, decision trees, etc. This process
detection is divided into three phases, namely preprocessor, feature extraction of spectrums
of fault
and detection
finally is divided
classification of the into three
fault and phases, namely preprocessor, feature extraction of
its diagnosis.
spectrums and finally classification
The vibration response data of the rotating of the fault and its diagnosis.
machine are collected and analyzed in
The vibration response data of the
preprocessor stage. These analyzed data are purely mechanically rotating machine are collected
based and andhelp
analyzed
to detectin
preprocessor stage. These analyzed data are purely mechanically
the faults when processed in real time. The data processing includes the removal of the based and help to detect
the faults
weak data when
by theprocessed
host system in that
realare time. The data
entered duringprocessing includesbythe
data acquisition removal
WSNs. The of the
faults
weak
in the data by the
rotating host system
machine can bethat areindividual
either entered during data acquisition
or combinational. by WSNs.
In order The faults
to differentiate
in the rotating machine can be either individual or combinational.
the type of the fault, the spectrums are extracted in the feature extraction stage. The support In order to differentiate
the type
vector of the fault,
machine the spectrums
algorithm is proposed are extracted
in order in to the feature
analyze theextraction
spectrums stage. The sup-
obtained for
port vector machine algorithm is proposed in order to analyze
various cases of faults in the machine. From the results obtained from the analysis, the the spectrums obtained for
various cases of faults in the machine. From the results obtained
type of the fault is listed and the severity of each fault is identified. In the last phase, the from the analysis, the
type of thefeatures
extracted fault is arelisted and the severity
compared in order of to each fault
classify is type
the identified.
of theIn the in
fault last phase,
the the
rotating
extracted
blades of afeatures
machine. are compared in order to classify the type of the fault in the rotating
blades of a machine.
2.3.3. Cloud Platform of IoT
2.3.3.TheCloud Platform
method used of for
IoTthe extraction of features from FRF spectrums used in the
detection of faultsused
The method is found
for the to be most suited
extraction methodfrom
of features and FRF
accurate enough
spectrums in classifying
used in the de-
the fault.
tection The support
of faults is foundvectorto be most machinesuited algorithm
method and is more relevant
accurate enough for infault prediction
classifying the
in real time applications using the cloud platform of the
fault. The support vector machine algorithm is more relevant for fault prediction in real IoT. For performing such IoT
operations,
time applications it is required
using the to choose a cloud service
cloud platform of the IoT.from the
For vendors. Services
performing such IoTprovided
operations, by
the cloud vendors include remote control, data analysis, machine
it is required to choose a cloud service from the vendors. Services provided by the cloud learning, deep learning,
etc.,
vendorsdepending
includeuponremote thecontrol,
requirements of the application.
data analysis, machine learning, deep learning, etc., de-
pending It is necessary to store the data
upon the requirements andapplication.
of the analyze it in real time. Therefore, a cloud platform
is selected, namely to
It is necessary ThingSpeak.
store the data In order to implement
and analyze it in realThingSpeak
time. Therefore,in this research
a cloud work,
platform
the foundation is the instance creation of the parameter that
is selected, namely ThingSpeak. In order to implement ThingSpeak in this research work, serves as the interconnection
between
the foundation the IoTisand data storage
the instance services.
creation of theAdditionally,
parameter that a programming tool that links
serves as the interconnection
the real time application environment is also provided
between the IoT and data storage services. Additionally, a programming tool that by ThingSpeak. Python code
linksis
developed for the extraction of the features and for classifying
the real time application environment is also provided by ThingSpeak. Python code is them for detecting the faults.
This code receives
developed for the data from the
extraction of thesensors
featuresthrough
and for the classifying
main functional them block as they are
for detecting the
linked together with the IoT. The data sent by the sensors
faults. This code receives data from the sensors through the main functional block will be in the form of a message,
as they
and need totogether
are linked be extractedwith before
the IoT. being
The processed
data sent by by the
the Python
sensors code.
will be Due to the
in the service
form of a
being
message, and need to be extracted before being processed by the Python code. Due toand
provided by the IoT, communication happens among the real time application the
sensors via cloud platform. All the devices in the IoT are interlinked and communicate
service being provided by the IoT, communication happens among the real time applica-
among themselves with the aid of a lightweight protocol. Therefore, in the work of fault
tion and sensors via cloud platform. All the devices in the IoT are interlinked and com-
detection and diagnosis using the IoT, the transmission of data is carried out in the form of
municate among themselves with the aid of a lightweight protocol. Therefore, in the work
lightweight protocol or messages.
of fault detection and diagnosis using the IoT, the transmission of data is carried out in
In the proposed condition monitoring system of fault detection, a support vector
the form of lightweight protocol or messages.
machine algorithm helps to process the vibration data transmitted by the IoT network,
Appl. Mech. 2023, 4, FOR PEER REVIEW 9

Appl. Mech. 2023, 4 364

In the proposed condition monitoring system of fault detection, a support vector ma-
chine algorithm helps to process the vibration data transmitted by the IoT network, as
shown in Figure 7. The efficiency of the fault classification can be improved by comparing
as shown in Figure 7. The efficiency of the fault classification can be improved by com-
low and high ranges in feature extraction, taking into account accuracy, precision and
paring low and high ranges in feature extraction, taking into account accuracy, precision
specificity.
and specificity.

Figure 7. Actual
Actual set-up
set-up of
of nodes
nodes in
in IoT.

3. Results
3. Results and
and Discussions
Discussions
The first
The first three
threeresonance
resonancefrequencies
frequencies of the rotating
of the blades
rotating for both
blades healthy
for both and faulty
healthy and
faulty conditions were evaluated using finite element analysis software and areinlisted
conditions were evaluated using finite element analysis software and are listed Tablein1,
and the corresponding mode shapes are shown in Figure 8.
Table 1, and the corresponding mode shapes are shown in Figure 8.

1. Resonance
Table 1.
Table Resonance frequency
frequency values
values of
of rotating
rotating blades.
blades.
Condition 1st Mode
1st Frequency (Hz)
Mode Frequency 2nd2nd
Mode Frequency
Mode (Hz)
Frequency 3rd
3rdMode
ModeFrequency (Hz)
Frequency
Condition
Healthy 77.01 (Hz) 599.04 893.71
(Hz) (Hz)
Single crack 74.78 596.8 886.2
TwoHealthy
cracks 74.92 77.01 599.04
594.6 893.71
878.1
Single crack
Looseness 73.24 74.78 596.8
593.2 886.2
873.5
Two cracks 74.92 594.6 878.1
Looseness 73.24 593.2
The natural frequencies corresponding to the first mode shape of the rotating 873.5blades as
shown in Figure 8a are 77.01 Hz, 74.78 Hz, 74.92 Hz and 73.24 Hz for healthy, single crack,
The natural
two cracks frequencieslooseness,
and mechanical corresponding to the first
respectively. mode
It can shape of
be clearly the that
seen rotating blades
the natural
as shown in of
frequencies Figure 8a are
the first mode77.01 Hz,are
shape 74.78 Hz, 74.92
different Hz and
for the healthy73.24andHzfaulty
for healthy,
conditions single
of
crack, two cracks
the rotating blades,andandmechanical looseness,
also a significant respectively.
decrease It can bevalue
in the frequency clearly forseen that
all the the
faulty
conditions
natural compared
frequencies withfirst
of the that of the
mode healthy
shape conditionfor
are different ofthe
thehealthy
blades and is observed.
faulty condi- The
same trend can be seen for the natural frequencies given in Table
tions of the rotating blades, and also a significant decrease in the frequency value for all1 of the second and third
mode
the shapes
faulty of the rotating
conditions compared blades,
withasthat
shown of thein Figure
healthy 8b,c. Therefore,
condition of theit isblades
evident is that
ob-
the decrease
served. in the trend
The same natural frequency
can be seen of forthetherotating
naturalblades indicates
frequencies the presence
given in Table of 1 aoffault
the
in one form
second or other.
and third mode shapes of the rotating blades, as shown in Figure 8b,c. Therefore,
The abnormal
it is evident that thecondition
decrease in of the
the natural
blade isfrequency
found by using of the the Campbell
rotating bladesdiagram,
indicatesfrom the
which the critical speed of the
presence of a fault in one form or other.blades was identified, comparing the excitation and natural
frequencies of the rotating
The abnormal condition blades.
of theWhenever
blade is found the rotating
by using blades operate atdiagram,
the Campbell the resonancefrom
which the critical speed of the blades was identified, comparing the excitationconcentrated
condition, crack will initiate due to the higher stresses developed at the stress and natural
region of theofblade.
frequencies These blades.
the rotating stressesWhenever
keep on increasing
the rotating andblades
the crack
operatepropagates until the
at the resonance
point where
condition, the stress
crack intensity
will initiate duefactor
to theofhigher
the crack exceeds
stresses the fracture
developed at thetoughness of the
stress concen-
blade material before catastrophic failure occurs [39,40] That
trated region of the blade. These stresses keep on increasing and the crack propagatesis why stress intensity factors
are tothe
until be point
estimated
where forthethestress
propagating
intensitycrack
factor lengths basedexceeds
of the crack upon the thelinear elastic
fracture fracture
toughness
mechanics. For the above condition of the failure, it is seen in the
of the blade material before catastrophic failure occurs [39,40] That is why stress intensity literature that the final
crack length is estimated to be 70% of the width of the blade, which
factors are to be estimated for the propagating crack lengths based upon the linear elastic means that the blade
does notmechanics.
fracture fail at the For
crack thelength
abovebelow
conditionfinalofcrack length.itTherefore,
the failure, is seen in the cracks shouldthat
literature be
detected before they reach final length.
the final crack length is estimated to be 70% of the width of the blade, which means that
Vibrations
the blade are produced
does not fail at the in the blades
crack length due below to the
finalfaults,
cracknamely
length. crack,
Therefore,mechanical
cracks
looseness, imbalance, misalignment, etc. That means that vibration data can be used to
should be detected before they reach final length.
detect such faults. Generally, faults such as imbalance and misalignment are monitored
regularly during preventive maintenance works, but the cracks and mechanical looseness
of the blades becomes a challenging task to detect. Therefore, it becomes necessary to
Appl. Mech. 2023, 4 365

Appl. Mech. 2023, 4, FOR PEER REVIEW 10

monitor the condition of the blades in order to detect cracks and mechanical looseness
during the operation of the machine.

Figure 8.
Figure 8. (a) First mode, (b) second mode and (c) third mode of the rotating
rotating blades.
blades.

Experiments
Vibrations areare conducted
produced on ablades
in the rotating
dueblades
to the rig for two
faults, different
namely crack,blade faults,
mechanical
i.e., crack and
looseness, mechanical
imbalance, looseness. These
misalignment, faults
etc. That are selected
means based up
that vibration on can
data the actual
be used facts
to
observed
detect such in real time
faults. applications
Generally, faultsof rotating
such as blades.
imbalance Theandacceleration data are
misalignment areextracted
monitored for
the rotating
regularly bladespreventive
during by conducting modal tests
maintenance which
works, butaretheused to assess
cracks the health condition.
and mechanical looseness
This
of the blades becomes a challenging task to detect. Therefore, it becomesonnecessary
gives an understanding of how to investigate the effect of the faults the blades, to
such
monitor as crack and mechanical
the condition looseness
of the blades whichtoare
in order observed
detect cracksquite
and frequently
mechanicalinlooseness
rotating
blades.
during theIn order to implement
operation the present work, one sensor node is utilized to measure the
of the machine.
vibration of the rotating
Experiments blades. The
are conducted on avibration data are
rotating blades rigtransmitted to theblade
for two different monitor station
faults, i.e.,
using an IEEE 802.15.4 wireless communication device. The received
crack and mechanical looseness. These faults are selected based up on the actual facts ob- vibration signal is
the waveform in time domain that consists of statistical features. As
served in real time applications of rotating blades. The acceleration data are extracted for these time domain
features
the rotatingare difficult
blades by to conducting
use when classifying
modal tests thewhich
faults, are
theyused
are transformed
to assess theinto frequency
health condi-
signals and the frequency band is identified.
tion. This gives an understanding of how to investigate the effect of the faults on the
blades,Measured
such assignals
crack andof vibration
mechanicalare indicated
looseness in the form
which of harmonics
are observed forfrequently
quite healthy and in
faulty conditions of the rotating blades. The FRF of a healthy blade
rotating blades. In order to implement the present work, one sensor node is utilized to is shown in Figure 9, and
it is clearly
measure seen
the that there
vibration of is a higher
the rotating harmonic
blades. at 77 Hz,
The whichdata
vibration indicates that the resonance
are transmitted to the
of the blades exist within that range and that blades may
monitor station using an IEEE 802.15.4 wireless communication device. The become excited during turbine
received vi-
start
bration upsignal
or shut down
is the at this fundamental
waveform in time domain natural frequency.
that consists The first
of statistical fault condition
features. As these
considered is a crack on one of the rotating blades. Figure 10 shows the FRF spectrum of
time domain features are difficult to use when classifying the faults, they are transformed
the rotating blades, and it is observed that the resonance frequency shows more peaks with
into frequency signals and the frequency band is identified.
Measured signals of vibration are indicated in the form of harmonics for healthy and
faulty conditions of the rotating blades. The FRF of a healthy blade is shown in Figure 9,
and it is clearly seen that there is a higher harmonic at 77 Hz, which indicates that the
 blade. The second fault condition considered for the rotating blades was cracks on tw
blades which are adjacent to each other. The FRF spectrum indicates a greater number
peaks with increased resonance frequency range, as shown in Figure 11. The third fa
condition for the rotating blades is the mechanical looseness of the blade for which t
Appl. Mech. 2023, 4 FRF spectrum has multiple peaks, especially on lower side of resonance frequency 366 ran
of the blade, as shown in Figure 12.
For FRF spectrums, it was also observed that the amplitude or peak of the spectru
different amplitudes instead of the single peak seen for healthy blades. Additionally, the
is different for different types of faults. The amplitude of vibration for the healthy blad
natural frequency of the cracked blade is reduced compared with the healthy blade. The
was found
second to becondition
fault 0.00018 m/s 2, as shown in Figure 9. For the faulty conditions such as sin
considered for the rotating blades was cracks on two blades which
crackareblade, two
adjacent to cracks in adjacent
each other. blades and
The FRF spectrum mechanical
indicates looseness,
a greater number the with
of peaks amplitude
increased resonance frequency range, as shown in Figure 11. The third
vibrations were 0.00022 m/s , 0.00045 m/s and 0.00047 m/s , as seen in Figures
2 2 2 fault condition for10, 11 a
the rotating blades is the mechanical looseness of the blade for which the FRF spectrum
12, respectively. Therefore, it can be concluded that the intensity of the crack increas
has multiple peaks, especially on lower side of resonance frequency range of the blade, as
the stiffness
shown inof the blade
Figure 12. reduces and the amplitude of vibrations increases.

Figure 9. FRF
Figure of Healthy
9. FRF of Healthyblade.
blade.

Figure 10. FRF

Figure of single
10. FRF crack
of single crackblade.
blade.

For FRF spectrums, it was also observed that the amplitude or peak of the spectrum
is different for different types of faults. The amplitude of vibration for the healthy blades
was found to be 0.00018 m/s2 , as shown in Figure 9. For the faulty conditions such as
single crack blade, two cracks in adjacent blades and mechanical looseness, the amplitude
of vibrations were 0.00022 m/s2 , 0.00045 m/s2 and 0.00047 m/s2 , as seen in Figures 10–12,
respectively. Therefore, it can be concluded that the intensity of the crack increases, the
stiffness of the blade reduces and the amplitude of vibrations increases.
ppl. Mech.
Appl. Mech.2023,
2023,
Appl. 4,4,FOR
Mech. 2023,PEER
FOR 4PEERREVIEW
REVIEW 367 1
12

Figure
Figure 11.FRF
11.
Figure FRFFRF
11. of two
of two
of cracks
cracksin
cracks
two in adjacent blades.
adjacent blades.

Figure 12. FRF of blade looseness.

Figure 12. FRF
Figure of blade
12. FRF of bladelooseness.
looseness.

It can be clearly
It can seen from the results that every
typetype of fault present in the rotating
It can be be clearlyseen
clearly seenfrom
from the
theresults
resultsthat every
that every of fault
type of present in the rotating
fault present in the rotatin
blades gives
blades different
gives differentFRF
FRF spectrums with
spectrums with multiple
multiple peaks
peaks within
within the resonance
the resonance frequencyfrequency
blades gives different FRF spectrums with multiple peaks within the resonance frequenc
range of the
range blade.
of the blade.These
TheseFRFFRF spectrums
spectrums areare then
then usedused to develop
to develop a learning
a learning model using
model using
range of the blade.
a support vector
These FRF
machinelearning
spectrums are then
learningtechnique.
technique. This
used toproved
This model
developbe a learning model usin
a support vector machine model provedtoto thethe
be best in in
best thethe clas-
a support vectorofmachine
classification the faultslearning
compared technique. Thislearning
model proved
models. to be the best in the clas
sification of the faults compared withwith
thethe other
other learning models. Hence,
Hence, condition
condition moni-
sification of theoffaults
monitoring compared
the rotating with
blades is the other
the most learning
important models.
activity Hence,
that results condition mon
in minimizing
toring
theof
of the rotating
costs
blades is the most important activity thatequipment.
results in minimizing the
toring theincurred
rotatingin blades
maintenance
is theand maximizing
most important the activity
safety of the
that results in minimizing th
costs incurred in maintenance and maximizing the safety of the equipment.
costs4.incurred
Conclusions
in maintenance and maximizing the safety of the equipment.
4. Conclusions
The most common failure cause of turbo machinery is blade faults. These faults may
4. Conclusions
occur in many ways, but the most important ones are cracks and mechanical looseness of
The most common failure cause of turbo machinery is blade faults. These faults may
the
Theblades.
most Therefore,
common itfailure
is necessary
causetoofdetect
turbothese faults well
machinery is in advance
blade by providing
faults. These faults ma
occura in many
condition ways, but
monitoring the most
system in important
order to ones
safeguard are
the cracks
machine.and In mechanical
this work, anlooseness
IoT of
occur in
the blades. many ways,
Therefore, but the most important ones are cracks and mechanical looseness ao
based fault diagnosis it is necessary
method to detect
was designed whichthese faultsmeasuring
included well in advance by providing
the vibration data
the blades.
condition Therefore,
monitoring it is necessary to detect these faults well in advance by providing
of the blades using a system in ordersensor,
laser vibrometer to safeguard
a wirelessthe machine.
sensor network Insystem
this work, an IoT based
to transmit
condition
faultthe monitoring
data frommethod
diagnosis system
the machine
wasto in order
the cloudto
designed safeguard
platform
which and the
included machine.
a support vector
measuring Inthethis
machinework, an
learning
vibration IoTof
data base
the
fault diagnosis
algorithm to method
predict thewas designed
condition of thewhich
rotatingincluded
blades. measuring
FRF spectrums
blades using a laser vibrometer sensor, a wireless sensor network system to transmit the the
were vibration
compared data
for of th
blades
data from using
theamachine
laser vibrometer
to the cloudsensor, a wireless
platform and a sensor
supportnetwork system to
vector machine transmit
learning th
algo-
data from the machine to the cloud platform and a support vector
rithm to predict the condition of the rotating blades. FRF spectrums were compared for machine learning algo
rithm to predict
the healthy the condition
and faulty conditions of the rotating
of the rotatingblades.
blades,FRF
andspectrums
their features werewerecompared
extractedfo
the healthy
to help and faulty
to classify the conditions
type of fault. of the
Therotating
following blades,
are theandkey their features
results were from
obtained extracte
the
Appl. Mech. 2023, 4 368

the healthy and faulty conditions of the rotating blades, and their features were extracted
to help to classify the type of fault. The following are the key results obtained from the
present work:
• The accuracy of the experimental setup was well obtained by verifying with simulation results;
• A remote fault detection method for rotating blades was developed by implementing
the Internet of Things;
• A support vector machine algorithm was trained with vibration responses for healthy
and faulty conditions of rotating blades with accuracy and precision;
• Fatigue cracks and mechanical looseness were detected using frequency response
function spectrums.
This proposed work aims to provide an online condition monitoring method for
Industry 4.0 applications.

Author Contributions: Conceptualization, R.P.B.K., M.K. and M.C.; methodology, R.P.B.K., M.K.
and M.C.; software, R.P.B.K., M.K. and M.C.; validation, R.P.B.K., M.K. and M.C.; formal analysis,
R.P.B.K., M.K. and M.C.; investigation, R.P.B.K., M.K. and M.C.; resources, R.P.B.K., M.K. and M.C.;
data curation, R.P.B.K., M.K. and M.C.; writing—original draft preparation, R.P.B.K., M.K. and M.C.;
writing—review and editing, M.C.; visualization, R.P.B.K., M.K. and M.C.; supervision, M.K.; project
administration, R.P.B.K., M.K. and M.C. All authors have read and agreed to the published version of
the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Pennacchi, P.; Vania, A. Diagnosis and model based identification of a coupling misalignment. Shock. Vib. 2005, 12, 293–308.
[CrossRef]
2. Djaidir, B.; Hafaifa, A.; Kouzou, A. Faults detection in gas turbine rotor with vibration analysis under varying conditions. J. Theor.
Appl. Mech. 2017, 55, 393–406. [CrossRef]
3. Guan, Z.; Chen, P.; Zhang, X.; Zhou, X.; Li, K. Vibration analysis of shaft misalignment and diagnosis method of structure faults
for rotating machinery. Int. J. Perform. Eng. 2017, 13, 337–347. [CrossRef]
4. Saif, S.K.; Adawi, A.L.; Kumar, G.R.R. Vibration diagnosis approach for industrial gas turbine and failure analysis. Br. J. Appl. Sci.
Technol. 2016, 14, 1–9.
5. Li, C.; Sanchez, R.V.; Zurita, G.; Cerrada, M.; Cabrera, D. Fault diagnosis for rotating machinery using vibration measurement
deep statistical feature learning. Sensors 2016, 16, 895. [CrossRef]
6. Andre, L.D.; Afonso, C.T.; Guilherme, S.S.; Dennis, B.; Rodrigo, N. A cloud-based condition monitoring system for fault detection
in rotating machines using PROFINET process data. Comput. Ind. 2021, 126, 103394.
7. Marwin, Z.; Felix, M.; Veronika, L.; Christian, K.; Samuel, K. A machine learning-based workflow for automatic detection of
anomalies in machine tools. ISA Trans. 2022, 125, 445–458.
8. Wang, H.; Shi, L.; Liuyang, S.; Lingli, C. A novel convolutional neural network based fault recognition method via image fusion
of multi-vibration-signals. Comput. Ind. 2019, 105, 182–190. [CrossRef]
9. Xia, M.; Teng, L.; Yunfei, Z.; Clarence, W. Closed-loop design evolution of engineering system using condition monitoring through
internet of things and cloud computing. Comput. Netw. 2016, 101, 5–18. [CrossRef]
10. Siliang, L.; Peng, Z.; Xiaoxian, W.; Yongbin, L.; Fang, L.; Jiwen, Z. Condition monitoring and fault diagnosis of motor bearings
using undersampled vibration signals from a wireless sensor network. J. Sound Vib. 2018, 414, 81–96.
11. Adrian, S.; Fateme, D.; Xingyu, Z.; Valentin, R.; David, F.; Mike, B.; John, K.; Goran, N. Machine learning methods for wind
turbine condition monitoring: A review. Renew. Energy 2019, 133, 620–635.
12. Andreas, T.; Judith, P.V.; Marcel, K.; Gordon, E. Predictive maintenance enabled by machine learning: Use cases and challenges in
the automotive industry. Reliab. Eng. Syst. Saf. 2021, 215, 107864.
13. Yoonjae, L.; Byeonghui, P.; Minho, J.; Jongsu, L.; Changwoo, L. A quantitative diagnostic method of feature coordination for
machine learning model with massive data from rotary machine. Expert Syst. Appl. 2023, 214, 119117.
14. Rati, K.M.; Thanga, R.C.; Srikanth, A.; Aparna, A.; Ripul, G. Sources of vibration and their treatment in hydro power stations-A
review. Int. J. Eng. Sci. Technol. 2017, 20, 637–648.
Appl. Mech. 2023, 4 369

15. Chenyang, L.; Lingfei, M.; Hanru, T.; Ruqiang, Y. Lifelong Condition Monitoring Based on NB-IoT for Anomaly Detection of
Machinery Equipment. In Proceedings of the 8th International Conference on Through-Life Engineering Service—TESConf 2019,
Cleveland, OH, USA, 27–29 October 2019.
16. Alexandre, S.R.; Vinicius, W.K.; Iuri, C.F.; Nasser, J. An analysis of machine learning algorithms in rotating machines maintenance.
IFAC PapersOnLine 2022, 55, 252–257.
17. Roque, A.; Sobral, J. Motion Amplification Technology as a Tool to Support Maintenance Decisions. In Proceedings of the MPMM
2018—Maintenance Performance Measurement and Management Conference, Coimbra, Portugal, 21–23 June 2018; pp. 58–64.
18. Omar, A.; Fahad, A.; Mahmoud, M.; Muhammad, I.; Adam, G.; Faisal, A.; Jaroslaw, K.; Witold, G. Sounds and acoustic
emission-based early fault diagnosis of induction motor: A review study. Adv. Mech. Eng. 2021, 13, 1–19.
19. Shiqing, H.; Yubin, L.; Weijie, T.; Rongfeng, D.; Qingbo, H.; Fengshou, G.; Andrew, D.B. Sensing with sound enhanced acoustic
metamaterials for fault diagnosis. Front. Phys. 2022, 10, 1054. [CrossRef]
20. Divya, M.N.; Narayanappa, C.K.; Gangadharaiah, S.L. Modeling a Fault Detection Predictor in Compressor using Machine
Learning Approach based on Acoustic Sensor Data. Int. J. Adv. Comput. Sci. Appl. 2021, 12, 650–667.
21. Kaibo, L.; James, X.G.; Hongwei, F.; Xiuquan, S.; Bing, L.; Fengshou, G. Acoustics Based Monitoring and Diagnostics for the
Progressive Deterioration of Helical Gearboxes. Chin. J. Mech. Eng. 2021, 34, 82.
22. Qurthobi, A.; Maskeliunas, R.; Damaševicius, R. Detection of Mechanical Failures in Industrial Machines Using Overlapping
Acoustic Anomalies: A Systematic Literature Review. Sensors 2022, 22, 3888. [CrossRef] [PubMed]
23. Caleb, V.; Kwang-Seok, M.; Suk-Hwan, L.; Ki-Ryong, K. Sound Based Machine Fault Diagnosis System Using Pattern Recognition
Techniques. J. Korea Multimed. Soc. 2017, 20, 134–143.
24. Martin-del-Campo, S.; Sandin, F. Online feature learning for condition monitoring of rotating machinery. Eng. Appl. Artif. Intell.
2017, 64, 187–196. [CrossRef]
25. Saeed, R.; Mehdi, S.A.; Stefania, S.; Francesco, F. Fault diagnosis in industrial rotating equipment based on permutation entropy,
signal processing and multi-output neuro-fuzzy classifier. Expert Syst. Appl. 2022, 206, 117754.
26. Scislo, L. Single-Point and Surface Quality Assessment Algorithm in Continuous Production with the Use of 3D Laser Doppler
Scanning Vibrometry System. Sensors 2023, 23, 1263. [CrossRef]
27. Sarila, V.K.; Moinuddin, S.Q.; Cheepu, M.; Rajendran, H.; Kantumuchu, V.C. Characterization of Microstructural Anisotropy in
17–4 PH Stainless Steel Fabricated by DMLS Additive Manufacturing and Laser Shot Peening. Trans. Indian Inst. Met. 2023, 76,
403–410. [CrossRef]
28. Paladugu, S.R.M.; Sreekanth, P.R.; Sahu, S.K.; Naresh, K.; Karthick, S.A.; Venkateshwaran, N.; Ramoni, M.; Mensah, R.A.; Das, O.;
Shanmugam, R. A Comprehensive Review of Self-Healing Polymer, Metal, and Ceramic Matrix Composites and Their Modeling
Aspects for Aerospace Applications. Materials 2022, 15, 8521. [CrossRef] [PubMed]
29. Shiva, A.; Cheepu, M.; Kantumuchu, V.C.; Kumar, K.R.; Venkateswarlu, D.; Srinivas, B.; Jerome, S. Microstructure characterization
of Al-TiC surface composite fabricated by friction stir processing. IOP Conf. Ser. Mater. Sci. Eng. 2018, 330, 012060. [CrossRef]
30. Ross, N.S.; Gopinath, C.; Nagarajan, S.; Gupta, M.K.; Shanmugam, R.; Kumar, M.S.; Boy, M.; Korkmaz, M.E. Impact of hybrid
cooling approach on milling and surface morphological characteristics of Nimonic 80A alloy. J. Manuf. Process. 2022, 73, 428–439.
[CrossRef]
31. Kantumuchu, V.C.; Cheepu, M.M. The Influence of Friction Time on the Joint Interface and Mechanical Properties in Dissimilar
Friction Welds. J. Met. Mater. Res. 2022, 5, 1–7. [CrossRef]
32. Shanmugam, R.; Ramoni, M.; Thangamani, G.; Thangaraj, M. Influence of additive manufactured stainless steel tool electrode on
machinability of beta titanium alloy. Metals 2021, 11, 778. [CrossRef]
33. Sarila, V.; Koneru, H.P.; Cheepu, M.; Chigilipalli, B.K.; Kantumuchu, V.C.; Shanmugam, M. Microstructural and Mechanical
Properties of AZ31B to AA6061 Dissimilar Joints Fabricated by Refill Friction Stir Spot Welding. J. Manuf. Mater. Process. 2022,
6, 95. [CrossRef]
34. Ragavanantham, S.; Kumar, S.S.; Shyam, M.S. Optimisation of Shutter Speed in Machine Vision Technique for Monitoring
Grinding Wheel Loading. Appl. Mech. Mater. 2015, 766, 878–883. [CrossRef]
35. Cheepu, M.; Kantumuchu, V.C. Numerical Simulations of the Effect of Heat Input on Microstructural Growth for MIG-Based
Wire Arc Additive Manufacturing of Inconel 718. Trans. Indian Inst. Met. 2023, 76, 473–481. [CrossRef]
36. Nagarajan, N.; Solaiyappan, A.; Mahalingam, S.K.; Nagarajan, L.; Salunkhe, S.; Nasr, E.A.; Shanmugam, R.; Hussein, H.M.A.M.
Meta-Heuristic Technique-Based Parametric Optimization for Electrochemical Machining of Monel 400 Alloys to Investigate the
Material Removal Rate and the Sludge. Appl. Sci. 2022, 12, 2793. [CrossRef]
37. Sarila, V.K.; Koneru, H.P.; Pathapalli, V.R.; Cheepu, M.; Kantumuchu, V.C. Wear and Microstructural Characteristics of Colmonoy-
4 and Stellite-6 Additive Layer Deposits on En19 Steel by Laser Cladding. Trans. Indian Inst. Met. 2023, 76, 457–464. [CrossRef]
38. Brito, L.C.; Susto, G.A.; Brito, J.N.; Duarte, M.A.V. An explainable artificial intelligence approach for unsupervised fault detection
and diagnosis in rotating machinery. Mech. Syst. Signal Process. 2022, 163, 108105. [CrossRef]
Appl. Mech. 2023, 4 370

39. Ravi Prakash Babu, K.; Raghu Kumar, B.; Devaraju, A.; Murahari, K.; Satyanarayana, K.; Sai Kumar, G. Finite element modelling
aspects in the fracture assessment of a low-pressure steam turbine blade. Int. J. Interact. Des. Manuf. 2022. [CrossRef]
40. Adhikari, R.; Bolar, G.; Shanmugam, R.; Koklu, U. Machinability and surface integrity investigation during helical hole milling in
AZ31 magnesium alloy. Int. J. Lightweight Mater. Manuf. 2023, 6, 149–164. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.