Hindawi

Shock and Vibration

Volume 2021, Article ID 9469318, 25 pages
https://doi.org/10.1155/2021/9469318

Review Article
Vibration Analysis for Machine Monitoring and Diagnosis: A
Systematic Review

1 1,2
Mohamad Hazwan Mohd Ghazali and Wan Rahiman
1
School of Electrical and Electronic Engineering, Universiti Sains Malaysia Engineering Campus, Nibong Tebal 14300,
Penang, Malaysia
2
Cluster of Smart Port and Logistic Technology (COSPALT), Universiti Sains Malaysia Engineering Campus, Nibong Tebal 14300,
Penang, Malaysia

Correspondence should be addressed to Wan Rahiman; [email protected]

Received 4 July 2021; Revised 3 August 2021; Accepted 25 August 2021; Published 11 September 2021

Academic Editor: Gang Tang

Copyright © 2021 Mohamad Hazwan Mohd Ghazali and Wan Rahiman. This is an open access article distributed under the
Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided
the original work is properly cited.

Untimely machinery breakdown will incur significant losses, especially to the manufacturing company as it affects the production
rates. During operation, machines generate vibrations and there are unwanted vibrations that will disrupt the machine system,
which results in faults such as imbalance, wear, and misalignment. Thus, vibration analysis has become an effective method to
monitor the health and performance of the machine. The vibration signatures of the machines contain important information
regarding the machine condition such as the source of failure and its severity. Operators are also provided with an early warning
for scheduled maintenance. Numerous approaches for analyzing the vibration data of machinery have been proposed over the
years, and each approach has its characteristics, advantages, and disadvantages. This manuscript presents a systematic review of
up-to-date vibration analysis for machine monitoring and diagnosis. It involves data acquisition (instrument applied such as
analyzer and sensors), feature extraction, and fault recognition techniques using artificial intelligence (AI). Several research
questions (RQs) are aimed to be answered in this manuscript. A combination of time domain statistical features and deep learning
approaches is expected to be widely applied in the future, where fault features can be automatically extracted from the raw
vibration signals. The presence of various sensors and communication devices in the emerging smart machines will present a new
and huge challenge in vibration monitoring and diagnosing.

1. Introduction faults can be identified without stopping the machine or

tearing the machine down. The changes of these signals often
Machines are widely used in today’s industries and are indicate the presence of a fault. Acoustic analysis has the
paramount for factory operation. Careful monitoring of the advantages of short analysis time, high recognition effi-
machines must be done and when the machines do ciency, and nondestructive testing. However, it is very
breakdown unexpectedly, it will cause massive loss to the challenging to properly capture the acoustic signals due to
company. This can be prevented by diagnosing the machine several factors such as environmental conditions, different
to determine the fault or potential fault such as imbalance, parameter of recording software, and reflected acoustic
wear, misalignment, defective bearing, friction whirl, and signals [5]. Vibration signals analysis also has some ad-
cracking teeth in gearing [1, 2]. Several available diagnosis vantages and disadvantages. Real-time machine monitoring
methods have been applied over the years including oil can be achieved using vibration analysis and there are many
analysis, vibration signal analysis, particle analysis, corro- well-developed signal processing techniques that can be
sion monitoring, acoustic signal analysis, and wear debris applied. The limitations of vibration analysis are noise
analysis [3, 4]. Among these analyses, acoustic and vibration contamination and proper mounting position of the vi-
signal analysis emerge as popular choices because many bration sensors [6]. Another technique that can be used for
2 Shock and Vibration

machine monitoring and diagnosis is thermal imaging faulty system, data-driven methods are widely applied in
analysis. In this analysis, an infrared camera is usually used machine diagnosis and monitoring compared to model-
to detect many electrical faults in the machine based on the based methods. Thus, the main contribution of this article
thermal anomalies. The thermal images obtained are useful is the review of various data-driven vibration analysis
in detecting and locating the machine’s faults. However, this techniques and instruments used for monitoring and di-
technique is expensive and requires a longer time to process agnosing the machines. However, due to a wide range of
the thermal images compared to processing the acoustic and techniques, only the widely used ones are discussed in this
vibration signals. Vibration analysis is considered as the best article. There are several review articles regarding this area,
method in determining the machine condition [7]. and Table 1 discusses each of the review articles. Based on
According to Saucedo-Dorantes et al. [8], the percentage of Table 1, we aimed to fill the research gap in reviewing the
fault diagnosis techniques conducted with the means of vibration analysis for machine monitoring and diagnosis,
vibration analysis exceeds 82%. Machines are mostly made where the data acquisition system and comparison be-
up of moving parts that generate unwanted vibration and tween different vibration analysis techniques including the
with vibration analysis, and a decision on whether the latest deep learning approach are not discussed. We also
machine can continue to operate or needs to be shut down aimed to answer the following research questions de-
and repaired can be made [3]. scribed in Table 2.
The machine’s condition can be determined by the vi- This manuscript is organized as follows. In the next
bration amplitude and frequency, as both can reveal the section, the research approach for this review article is
severity and source of the machine problem, respectively [9]. discussed. Then, Sections 3–5 will follow the vibration
At first, without the help of vibration equipment, machine analysis steps, as shown in Figure 1. Firstly, in Section 3,
conditions can still be diagnosed with a human brain of the data acquisition stage is discussed. This stage involves
trained personnel coupled with the senses of touch and the types of vibration sensors used to obtain the vibration
hearing which acts as a vibration analyzer. However, human data and analyzers for analyzing the acquired data. The
perception is somewhat limited, and it is impossible to detect data acquisition stage is not discussed in most of the
problems that are beyond the capability of human senses of published review articles. Besides, different sensor
touch and hearing. Then, vibration analysis was based on a mounting techniques are also discussed in this section.
real-time spectral analyzer and now it can be categorized Section 4 is about the signal processing or feature ex-
into time, frequency, and time-frequency domain [10]. Time traction methods performed by researchers over the years,
and frequency domain analysis analyses the time series of based on the time, frequency, and time-frequency do-
data with respect to time and frequency, respectively. Time- mains. The final stage in vibration analysis, which is the
frequency domain analysis used both time and frequency fault recognition stage, is discussed in Section 5. In this
domains at the same time [3]. Vibration analysis for most section, different AI-based methods such as support
machine monitoring and diagnosis can be divided into low- vector machine (SVM), neural network (NN) including
speed and high-speed machines. Because currently there is deep learning, fuzzy logic, and genetic algorithm (GA) in
no universally accepted speed range to differentiate both the fault recognition step are discussed. The discussions
types of machines, machines with rotating speeds up to and findings of our review are explained in Section 6, and
600 rpm such as wind turbines and paper mills are con- we conclude our studies in Section 7.
sidered low-speed machines. According to Kim et al. [11],
low-speed machines can be very time-consuming and more
difficult to monitor compared to high-speed machines as the 2. Research Approach
rotating elements and the fault is typically low unless the
fault has reached above the background noise level. Vi- A comprehensive literature study was conducted to an-
bration monitoring of low-speed machines is highly asso- swer the five research questions. A Systematic Literature
ciated with low-speed bearing in ensuring the reliability of Review (SLR) approach was employed to collect the rel-
the machine. evant primary studies regarding the vibration analysis for
The vibration analysis for machine monitoring and machine monitoring and diagnosis [17]. Firstly, we
diagnosis typically consists of three main steps, which are classified the articles and the selected papers were then
data acquisition, signal processing, and fault recognition. analyzed and differentiated through the content analysis
To date, there are lots of techniques and instruments used method. The results can be classified into four main
in each of the aforementioned steps, and choosing the right categories:
ones might be quite challenging. This is because each
(1) Survey, where review articles made by other re-
method and instrument have its characteristics, advan-
searchers on vibration analysis for machine moni-
tages, and disadvantages. These methods can be divided
toring and diagnosis are discussed.
into two main groups which are model-based and data-
driven methods. Model-based methods require an ana- (2) Simulation, where the performance of the proposed
lytical model of the system whereas data-driven methods technique is evaluated through simulation methods.
do not need any assumption about the system’s model. In (3) Experiment/real-time deployments, where the
data-driven methods, advanced signal processing tech- methods developed are applied experimentally or in
niques are applied. Because it is very difficult to model a real environments.
Shock and Vibration 3

Table 1: List of review articles or survey regarding vibration analysis for machine monitoring and diagnosis.
Authors Scope Comments
Vishwakarma et al. Presented a review of some vibration feature extraction No discussion on the data acquisition system and AI-
[12] methods applied to different types of rotating machines based fault recognition techniques
Reviewed various techniques including the AI methods No discussion on the data acquisition system and the
Kumar et al. [3] used for fault diagnosis based on vibration analysis comparison between different feature extraction and
methods fault recognition methods are not presented
Presented the vibration analysis techniques in terms of
Only discussed four types of vibration analysis
Boudiaf et al. [13] their capabilities, advantages, and disadvantages in
techniques
monitoring rolling element bearings
Discussed a variety of vibration analysis approaches in No discussion on the data acquisition system and the
Aherwar [14] diagnosing the rotating machinery including the AI review was conducted in 2012; thus, it lacks the latest AI
methods methods such as deep learning
Sait and Sharaf- Presented a review of vibration-based analysis damage No discussion on the data acquisition system, frequency
Eldeen [15] detection techniques to monitor gearbox condition domain methods, and AI techniques
Discussed on the fault diagnosis in large electrical Most of the vibration analysis techniques are not
Sadeghi et al. [16]
machines using vibration signals discussed in this article

(4) Performance comparison, where the proposed analyzer can be divided into standalone and computer-based
technique is utilized and compared with other analyzer, whereas vibration sensor consists of accelerometer,
methods in terms of accuracy or robustness. velocity transducer, displacement sensor, and laser Doppler
vibrometer (LDV). The accelerometer can be further divided
The databases used for the electronic search were
into piezoelectric and microelectromechanical system
Multidisciplinary Digital Publishing Institute (MDPI), In-
(MEMS) accelerometer.
stitute of Electrical and Electronics Engineers (IEEE) Xplore,
Association for Computing Machinery (ACM) Digital Li- 4. Analyzer
brary, ScienceDirect, Web of Science, Wiley Online Library,
Researchgate, Springer, Scopus, and Google Scholar. To Analyzer is an instrument used to analyze the vibration data
refine the search results, a set of inclusion and exclusion produced by machinery. It is composed of a sensor (which is
criteria is used in determining the relevant articles, which presented in the later section of this paper), amplifier, filter,
can be observed in Table 3. and A/D converter. The signal from the vibration sensor
The article searching process starts by selecting the passes through the amplifier to increase the resolution and
main search term: “machine monitoring” or “machine signal-to-noise ratio. The amplified signal then passes
diagnosis.” The second search term was “vibration analy- through a filter so that aliasing would not be encountered in
sis.” So, the search sentence was as follows: “machine the digitization stage. The signal is digitized in A/D con-
monitoring” OR “machine diagnosis” AND “vibration verter, and then it goes through the processing unit where it
analysis.” The reason why we decided to use these simple can be portrayed as a time waveform or can be further
keywords is to get good coverage of potential studies. Next, processed to acquire frequency spectrum [10, 18]. The vi-
we refine the search terms by adding the “artificial intel- bration analyzer can be divided into conventional and
ligence” term such as SVM, NN, deep learning, fuzzy logic, computer-based vibration analyzer. A conventional vibra-
and GA as the third term. The first and second search terms tion analyzer is a standalone instrument that is specifically
are retained. The aim is to obtain the missing articles after built for vibration. It is a complex and expensive instrument,
searching with the “machine monitoring” or “machine usually used by vibration experts. This instrument can help
diagnosis” and “vibration analysis” terms. 100 related the user to determine the presence of a problem as well as its
studies were targeted to provide enough information for root cause and time for the machine to fail. There are single,
categorization and research trends. Next, we classified all dual, and four-channel analyzers available in the market. A
the selected articles into the aforementioned categories. single channel analyzer can only receive an input from one
The total number of suitable articles collected was 105, and accelerometer at a time, whereas a dual-channel analyzer can
some of the articles could be classified into more than one receive inputs from two differently placed accelerometers at
category. the same time [19]. A four-channel analyzer can accept input
from multiple sensors and capable of measuring horizontal,
vertical, axial, and early bearing detection simultaneously. It
3. Data Acquisition Methods (RQ 1) is usually used with a triaxial accelerometer. The key ad-
vantage of a four-channel analyzer is the ability to observe
According to Elango et al. [10], a trained worker without the operating deflection shape (ODS) of a machine. Nuawi
academic qualifications can carry out the data collection et al. [20] used a four-channel vibration analyzer in the
work but data processing work in determining the condition monitoring process of bearing condition and the application
of the machine requires an engineer. Referring to Figure 2, of a dual-channel analyzer for machine monitoring can be
there are two instruments that are crucial in the data col- seen in [21, 22]. Another cheaper alternative is a handheld
lecting stage which are analyzer and sensor. Vibration vibration meter. This battery-powered device is equipped
4 Shock and Vibration

Table 2: The research questions aimed to be answered in this study.

RQ Motivation
RQ 1: what are the current and future instruments used for the The answer to this question helps to determine the best instrument
machine’s vibration data acquisition process? to be applied for respective vibration analysis.
RQ 2: what are the most used vibration signal processing techniques Answering this question can help to determine the most applied
and the comparison between them including the advantages and techniques and identifying the benefits and drawbacks of each
limitations of each method? vibration’s signal processing methods.
RQ 3: what are the latest AI techniques used in machine vibration By answering this question, the latest AI method applied in machine
monitoring and diagnosis, and how does it perform compared to vibration monitoring and diagnosis can be identified and the
other widely used AI methods? performance can be compared with the widely used AI methods.
RQ 4: what are the future directions regarding vibration analysis for The answer to this question can help new researchers fill the research
machine monitoring and diagnosis? gap regarding this area.
Answering this question can help the readers identify the articles that
RQ 5: what are the most influential articles for each method?
can be used as references for their research.

Start

Analyzer
Data Acquisition

Sensor

Time Domain

Feature Extraction/
Signal Processing Frequency Domain

Time-Frequency
Domain

Fault Recognition AI-Based

End

Figure 1: The framework of the article where data acquisition stage is discussed first, followed by feature extraction techniques, and fault
recognition approaches based on AI.

with an accelerometer and provides a display of vibration conventional vibration analyzer is faster and more accurate.
levels when in contact with machinery [19]. It requires very To overcome these limitations, dedicated hardware such as
little skill to use but its measurement capability is somewhat Digital Signal Processor (DSP) or Field Programmable Gate
limited and lacking in data storage performance. Array (FPGA) was used alongside a personal computer and
A computer-based vibration analyzer is an emerging sensor for user control and result display [18]. The computer
instrument where vibration data can be processed virtually processor usually has to handle the whole operating system
with the help of specific software and a personal computer. in addition to the virtual analyzer whereas the DSP only
This method has gained popularity because it is simple, performs one task. This makes the vibration analysis faster in
inexpensive, and easy to repair, and can perform most of the a computer equipped with DSP. In [26], vibration analysis
functions available in the conventional vibration analyzer on the rotating machine has been conducted by employing
such as oscilloscope, multimeter, and waveform generator. two DSPs. Rangel-Magdaleno et al. [27] has conducted a
LabVIEW is a widely used programming language in this vibration analysis on the CNC machine using an FPGA
method due to numerous arrays of data acquisition cards device. FPGA played a role in processing the vibration data
and measurement systems supported by it [23–25]. Ansari and the computer screen portrayed the results obtained for
and Baig [23] used the computer-based vibration analyzer to further analysis. Rodriguez-Donate et al. [28] developed an
monitor the condition of the machine, and they found that a online monitoring system for the induction motor with the
Shock and Vibration 5

Table 3: The inclusion and exclusion criteria used for articles searching in this study.
Exclusion
Inclusion criteria
criteria
Articles that are
Articles should fall into one of the four categories mentioned in this study not written in
English
Articles that are
not related to
the vibration
Articles should meet both the search terms analysis for
machine
monitoring and
diagnosis
Duplicated
Articles are published or accepted during the period between 1996 and 2021 (covering 25 years period)
articles
Articles should be listed at least in one of the research databases
Articles should be published or accepted at a conference, journal, magazines, or theses

Data Acquisition

Analyzer Sensor

Conventional/ Computer-Based Accelerometer Velocity Displacement LDV

Standalone Analyzer Analyzer Transducer Sensor

Piezoelectric MEMS

Figure 2: The data acquisition stage of the proposed review article.

implementation of FPGA, and they found that the FPGA- 5.1. Sensor Mounting Method. Choosing a mounting method
based system has better processing speed compared to DSP as well as implementing it correctly is an important factor in
and all the peripheral digital structures and processing unit vibration data collection. For continuous or online monitoring
can be included in a single chip. Compared to DSP, a of machine condition, vibration sensors are usually mounted
computer-based vibration analyzer using FPGA is better permanently at a specific location in the machine. Mounting
because it can achieve true parallelism. Both devices actually can be divided into four main methods, namely, stud-mounted,
provide better performances than using only a computer- adhesive-mounted, magnet-mounted, and nonmounted. Stud
based vibration analyzer [18]. More information regarding mounting is usually preferable for permanent mounting ap-
DSP and FPGA analyzers can be found in [29–31]. plications. The sensor is screwed in a stud and secured to the
machine. Apart from highly reliable and secure, this mounting
technique has the widest frequency response compared to
5. Sensor other methods. Make sure that the location where the sensor is
A sensor or transducer is a device that converts mechanical going to be mounted is clean and paint free because any ir-
signals to electrical signals [32]. The type of sensors used is regularities in the mounting surface will produce improper
usually based on the frequency range, sensitivity, design, and measurements or worst, damage to the sensor itself [19]. For
operational limitations. No matter what type of sensors is adhesive mounting, no extensive machining is required as
used, the stiffer the mounting of the sensor, the higher the epoxy, glue, or wax will be applied. If the machine cannot be
frequency range and its reading accuracy [33]. In vibration drilled for stud mounting, adhesive mounting is generally the
analysis, there are three widely used sensors for acquiring the best alternative. Although this mounting technique is easy to
vibration signal. These sensors are accelerometer, velocity, apply, the accuracy of the measurement is reduced because of
and displacement sensor. The noncontact LDV sensor is also the presence of damping in the adhesive [19]. In addition to
discussed in this section. The advantages and disadvantages that, it is also more difficult to remove the sensor compared to
of each vibration sensor can be seen in Table 4. other mounting methods. The magnetic mounting method is
6 Shock and Vibration

usually limited to temporary applications with a portable et al. [41] used the MEMS accelerometer in different rotating
analyzer and not preferable for permanent monitoring because machines for vibration monitoring. A performance com-
the high-frequency signals might be disrupted. The non- parison between conventional piezoelectric accelerometer
mounting method is typically applied by a probe tip, where and MEMS accelerometer can be seen in [43, 44]. It was
there is no external mechanism between the transducer and the found that MEMS accelerometer’s sensitivity is more stable
target surface. It is usually used in areas that are difficult to compared with the piezoelectric accelerometers and this
reach. The length of the probe tip, however, will affect the low-cost MEMS accelerometer can be a good alternative to
measurement, with longer probes lead to more inaccuracies. the high-cost piezoelectric accelerometer. This sensor has
also been applied in [26, 39].

5.2. Accelerometer. An accelerometer is a device used to

5.3. Velocity Transducer. A velocity transducer measures the
measure the vibration or acceleration of a structure in the SI
voltage produced by the relative movement of the object,
unit of g (m/). The working mechanism is that when the
usually in the m/s or cm/s unit. It works based on the
piezoelectric material in the accelerometer is subjected to a
concept of electromagnetic induction, and it can operate
force, it produces a charge corresponding to the force ap-
without any external device [45]. As the surface where the
plied. Because force is directly proportional to the accel-
sensor is mounted vibrates, the movement of the magnet in
eration, any change to this factor will produce a change in
the coil will produce a voltage proportional to the velocity of
the charge produced, which is then amplified [33]. Uniaxial
the vibration [46]. This voltage signal represents the vi-
accelerometer can only detect movement in one plane,
bration produced and is then feeds a meter or analyzer [33].
whereas triaxial accelerometer covers all the three dimen-
Velocity sensors are not recommended when diagnosing the
sions. Compared to the uniaxial accelerometer, triaxial ac-
high-speed machinery because the operational frequency
celerometer has a higher memory capacity but much more
range is limited from 10 Hz to 2 kHz [10]. Generally, velocity
expensive [34]. Accelerometer is a widely used sensor due to
transducer costs less than other sensors and coupled with its
its reliability, simplicity, and robustness. It can be further
easy installation feature, it is favorable in monitoring the
divided into a piezoelectric and MEMS accelerometer. Pi-
vibration of rotating machinery. However, it is big, heavy,
ezoelectric accelerometer relies on the piezoelectric effect of
and most velocity transducers are prone to reliability
quartz or ceramic crystals, which are usually preloaded, to
problems at operational temperatures that exceed 121°C
generate an electrical output that is proportional to the
[37, 47]. A velocity transducer was applied by Rossi [48] to
applied acceleration. Changes in the charge produced de-
measure the frame vibration of compressor, which usually
pend on this acceleration [35, 36]. Piezoelectric acceler-
comprises of frequencies below 10 Hz. Figure 3(b) shows the
ometer possesses several advantages such as better frequency
vibration measurement using velocity transducer.
and dynamic range, lightweight, and high sensitivity.
However, it is vulnerable to interference from the external
environment [37]. It also requires electronic integration in 5.4. Displacement Sensor. A displacement sensor, which is
order to obtain velocity and displacement data because it is sometimes called eddy current or proximity sensor measures
AC coupled [37]. Salami et al. [38] demostrated the appli- both relative vibration and position of the shaft. The dis-
cation of LabVIEW in monitoring and analyzing the vi- placement unit can be in m, cm, or mm. It is usually used in
bration signals, where piezoelectric accelerometer was used measuring low-frequency vibration of less than 10 Hz, but it
in their study. Igba et al. [39] installed the piezoelectric can also measure vibration up to 300 Hz [45]. However, they
accelerometer sensor on the operational turbines to obtain do not excel in measuring a shaft bending away from the
the vibration data for time domain analysis. Khadersab and probe location [47]. Unbalance and misalignment problems
Shivakumar [40] used the piezoelectric accelerometer to are the types of problems that can be detected by the dis-
obtain vibration data from rotating machinery to analyze the placement probe. For the measured vibration frequencies
bearing faults. Figure 3(a) shows the vibration measurement above 1 kHz, the amplitude is usually lost in the noise level
piezoelectric accelerometer. [47]. It has the advantages of a good dynamic range within a
MEMS accelerometer usually consists of movable proof specific frequency range, reasonable sensitivity, and a simple
mass with plates, supported by a mechanical suspension postprocessing circuit with negligible maintenance. How-
system to the frame [41]. When it is subjected to acceler- ever, it is difficult to install, susceptible to shocks, and some
ation, the proof mass tends to resist motion due to its own traditional displacement sensors are not calibrated for un-
inertia and therefore the spring is stretched or compressed. known metal materials [37]. Sarhan et al. [49] used the
As a result, force corresponds to the applied acceleration is displacement sensor in monitoring the cutting forces of the
created. MEMS accelerometer is DC coupled and very machining center under different cutting conditions. Sai-
suitable for measuring low-frequency vibration and accel- mon et al. [50] developed a low-cost fiber optic displacement
eration. It requires low processing power and provides sensor (FODS) for industrial applications that is immune to
superior sensitivity [41]. Modern MEMS accelerometer electromagnetic interference. The capability of a fiber optic
provides quite good data quality up to several tens of kHz. displacement sensor in capturing the amplitude and fre-
The drawback is that it suffers from a poor signal-to-noise quency of vibration was studied by Binu et al. [51], and based
ratio. Contreras-Medina et al. [42] used a low-cost MEMS on the results, this sensor can solve many sensing problems
accelerometer in detecting machinery failures. Chaudhury in aircraft.
Shock and Vibration 7

Table 4: The advantages and disadvantages of various sensors used in vibration analysis for machine monitoring and diagnosis.
Sensors Advantages Disadvantages
Piezoelectric Lightweight, high sensitivity, good Needs electronic integration to acquire velocity and displacement data,
accelerometer frequency, dynamic range vulnerable to interference from the external environment
Cheaper than piezoelectric sensor,
MEMS
requires low processing power, high Suffers from poor signal-to-noise ratio
accelerometer
sensitivity
Velocity Can operate without any external device, Limited operational frequency range, most velocity transducers are prone
transducer generally costs less than other sensors to reliability problems at operational frequency of more than 121°C
Displacement Good sensitivity, simple postprocessing
Succeptible to shock, difficult to install
sensor circuit with negligible maintenance
Ease of changing the measurement
points, ability to provide long-range
LDV Extremely high cost, limited portability
measurements without compromising
the signal quality

5.5. LDV. LDV is a noncontact optical measurement in- Although other sophisticated time domain approaches have
strument that can be applied to determine the vibration been used, the approach of visually looking at the time
velocities of any points on the surface of a particular ma- waveform should not be underestimated because numerous
chine [52, 53]. The working mechanism of LDV is based on information can be obtained in this manner. This infor-
the laser Doppler concept where a frequency-modulated mation includes the presence of amplitude modulation, shaft
coherent laser beam is reflected from a vibrating surface, and unbalance, transient, and higher-frequency components
Doppler shift of the reflected beam is compared with the [55]. However, simply looking into these vibration signals
reference beam,. Currently, a higher power infrared (in- cannot segregate the variations in vibration signals for
visible) fiber laser is more popular in LDV compared to the different machine failures due to the noisy data, especially at
He-Ne laser. The introduction of this technology has fulfilled the early stage of failure. Thus, a signal processing method is
the goal of achieving long-range measurements without required to obtain the important information from the time
compromising the signal quality [54]. Continuous-scan laser domain signals by converting the raw signals into appro-
Doppler vibrometry (CSLDV) has accelerated the mea- priate statistical parameters such as peak, RMS, crest factor,
surement at many points. The laser beam will scan con- and kurtosis. Several statistical parameters are usually
tinuously along a defined path across a structure according extracted from the time domain signal so that the most
to the desired scan frequencies. One major advantage of significant parameter, which can effectively differentiate
LDV is the ease of changing the measurement point, which between healthy and defective machine vibration signals,
can be done by just deflecting the laser beam. Despite that, can be chosen [56]. In this article, the statistical parameters
the application of LDV in machine monitoring and diag- of peak, RMS, crest factor, and kurtosis are discussed and the
nosis is limited because of the price and portability factors. advantages and disadvantages of each parameter can be seen
in Table 5.
6. Feature Extraction/Signal Processing Method
(RQ 2) 7.1. Peak. The peak is the maximum value of the signal, v(t),
over measured time and can be defined as [55]
Figure 4 shows the feature extraction/signal processing stage
involved in this study. It can be divided into time, frequency, peak � |v(t)|max. (1)
and time domain frequency analysis. The time domain
analysis involves the statistical features of peak, root-mean- If there is a presence of impacts, the peak values of the
square (RMS), crest factor, and kurtosis. The frequency vibration signal will vary. Under a fault condition, the peak
domain analysis consists of fast Fourier transform (FFT), value increases. The fault’s severity and type can be assessed
cepstrum analysis, envelope analysis, and spectrum analysis, based on the amplitudes of the corresponding peaks. Peak
whereas time-frequency domain analysis can be divided into value feature was studied by Lahdelma and Juuso [57] to
WT, Hilbert–Huang transform (HHT), Wigner–Ville dis- diagnose bearing and gear faults in the machine. The pro-
tribution (WVD), short-time Fourier transform (STFT), and posed approach is suitable for online analysis as the re-
Power Spectral Density (PSD). quirements for frequency range are small. Shrivastava and
Wadhwani [58] used statistical parameters such as peak,
7. Time Domain Analysis RMS, crest factor, and kurtosis to diagnose the rotating
electrical machine. Although all parameters can differentiate
The simplest vibration analysis for machine diagnosis is used between healthy and faulty conditions, they concluded that
to analyze the measured vibration signal in the time domain. determining the type of faults in this manner is not very
Vibration signals obtained are a series of values representing efficient. Igba et al. [39] used the peak values approach in
proximity, velocity, and acceleration, and in time domain monitoring the condition of wind turbine gearboxes because
analysis, the amplitude of the signal is plotted against time. faults can be detected based on the changes in their values.
8 Shock and Vibration

This approach can also counter the limitations of the RMS faults. It was found that the crest factor is the most sensitive
feature, where the RMS is not significantly affected by low- feature for gear failure and by applying this feature, the
intensity vibrations. achieved diagnostic accuracy is 93.33%. Shrivastava and
Wadhwani [58] applied the crest factor values along with
other time domain features for the fault detection and di-
7.2. RMS. RMS value presents the power content in vi-
agnosis of rotating electrical machines. They found that the
bration and useful in detecting an imbalance in rotating
crest factor feature is unable to classify between healthy
machinery. According to Vishwakarma et al. [59], this is the
bearing, bearing with defective ball, and bearing with de-
simplest and effective technique to detect faults especially
fective outer race. Aiswarya et al. [66] used the crest factor
imbalance in rotating machines. However, detecting the
feature along with other time domain features to diagnose
faults at the early stage is still a problem in this method and
faults in the turbo pump of a liquid rocket engine. Combined
this technique is only appropriate for the analysis of a single
with the SVM method in the fault classification stage, the
sinusoid waveform. The RMS value is more suitable for
proposed method can diagnose the fault effectively with
steady-state applications and analysis of single sinusoid
100% accuracy.
waveform [1]. RMS is preferred over peak value due to the
peak value’s sensitivity to noise. RMS value of a pure si-
nusoid is equal to the area under the half-wave, which is 7.4. Kurtosis. Kurtosis is a nondimensional statistical
0.707. RMS value can be represented by measurement of the number of outliers in distribution
􏽳�����������
and in vibration analysis, it corresponds to the number of
1 T2
RM � . 􏽚 v(t)dt, (2) transient peaks. A high number of transient peaks and a
T T1 high kurtosis value may be indicative of wear. Kurtosis is
not sensitive to running speed or load, and its effective-
where T represents time duration and v(t)is the signal.
ness is dependent on the presence of significant impul-
Referring to Igba et al. [39], the RMS method has two
siveness in the signal [67]. Kurtosis feature can provide
disadvantages. The first one is the RMS values of a vibration
the information regarding the non-Gaussianity or im-
signal are not affected by the isolated peaks in the signal,
pulsiveness of the vibration signals [68, 69]. In machine
reducing its sensitivity towards incipient gear tooth failure.
condition monitoring applications, kurtosis is usually
Next, it is also not significantly affected by short bursts of
preferable to crest factor but the latter is more widely
low-intensity vibrations. This will produce some compli-
used. This is because the meters that can record the crest
cations in detecting the early stages of bearing failure.
factor value are easily available and more affordable
Bartelmus et al. [60] applied the RMS values as a diagnostic
compared to the kurtosis meter. Kurtosis was among the
feature to diagnose gearbox fault, where the models of the
five parameters applied by Fu et al. [70] to be incorporated
behavior of gearboxes that correlate the transmission error
with the unsupervised AI method in diagnosing rolling
function and load variation are presented. Sheldon et al. [61]
bearing. Based on the results, the proposed method was
used the RMS feature in diagnosing wind turbine gearbox
found to have a sensitive reflection on fault identifications,
and stated that applying the RMS feature is not recom-
including a slight fault. Runesson [67] employed the
mended in detecting early stages of bearing failure. RMS was
kurtosis along with RMS value in monitoring the con-
among the statistical parameters applied by Krishnakumari
dition of a mechanical press. The results demonstrated
et al. [62] in fault diagnostics of the spur gear. The pa-
that kurtosis generally is not reliable but contains some
rameters are then combined with fuzzy logic and diagnostics
useful information in the monitoring of the gearbox of the
accuracy was found to be 95%, where DT reduces the de-
mechanical press machine. Other research studies that
mand for human expertise. Other applications of RMS
applied the kurtosis approach in vibration analysis for
values in vibration analysis for machine monitoring can be
machine monitoring can be seen in Table 6 [63, 71].
seen in Table 6 [63, 64].

8. Frequency Domain Analysis

7.3. Crest Factor. A crest factor is the ratio of the peak value
of the input signal to the RMS value and is represented as Most real-world signals can be broken down into a com-
follows [55]: bination of unique sine waves. Each sine wave will appear as
peak a vertical line in the frequency domain, where the height and
Crest Factor � . (3) position of the line represent the amplitude and frequency,
RMS
√� respectively. In frequency domain analysis, the amplitude is
For a pure sine wave, the crest factor will be 2 � 1.414, plotted against frequency and compared to the time domain,
and for normally distributed random noise, the value will be and the detection of the resonant frequency component is
approximately 3. Compared to peak and RMS values, the easier. This is one of the reasons why frequency domain
crest factor is usually used when measurements are con- methods are favorable in detecting faults in the machine
ducted at different rotational speeds because it is inde- [59]. Several characteristics of the signal that are not visible
pendent of speed. Crest factors are also reliable only in the in the perspective of time domain can be observed using
presence of significant impulsiveness [1]. Jiang et al. [65] frequency domain analysis. However, frequency analysis is
employed the crest factor features and SVM to diagnose gear not suitable for signals whose frequencies vary over time.
Shock and Vibration 9

Piezoelectric
Velocity Mounting stud
accelerometer
transducer
Adhesive
Measured surface Drilled hole
Measured surface

(a) (b)

Figure 3: Vibration measurement using (a) piezoelectric accelerometer by means of adhesive mounting and (b) velocity transducer, stud-
mounted on the machine’s surface.

Feature Extraction/
Signal Processing

Time Frequency Time-Frequency

Domain Domain Domain

Peak RMS Crest Kurtosis WT WVD HHT PSD STFT

Factor

FFT Cepstrum Envelope Spectrum

Figure 4: The feature extraction/signal processing stage of the proposed review article.

The advantages and disadvantages of each frequency domain of time information. FFT is also unable to investigate the
method can be seen in Table 7. transient features efficiently in time and can predict the fault
but cannot determine the severity of fault [3]. However, it is
the quickest way to separate the frequencies of the signal for
8.1. FFT. Fourier transform (FT) converts a signal f(t) in the diagnosis process. A combination of time domain signal
the time domain to the frequency domain, generating the analysis and FFT is normally associated with diagnosing the
spectrum F(ω). FT is given by low-speed machine to produce more accurate results but the
∞
major concern is its reliance on the magnitude of the fault
F(ω) � Ff(t) � 􏽚 f(t)e− iωt dt. (4)
having an effect on the carrier frequency [11]. Saucedo-
−∞
Dorantes et al. [8] used the FFTand PSD method to diagnose
where ω is the frequency and t is the time. It can be con- the fault in a gearbox and detect the bearing defect in the
verted back to time domain from the frequency domain by induction motor. They found that the proposed method can
inverse Fourier transform (IFT). This can be obtained as perfectly detect the presence of wear at low operating fre-
1 ∞ quencies but is not suitable at high operating frequencies.
f(t) � F− 1 (F(ω)) � 􏽚 F(ω)eiωt dω. (5) Patel et al. [72] proposed the FFT method as an analysis tool
2π −∞
in monitoring a rotating machine, and major faults such as
FFT is an efficient and widely used algorithm to obtain misalignment and bearing can be monitored in this manner.
the FT of discretized time signals. The FFT plot of fault-free Other applications of FFTcan also be seen in Table 6 [23, 73].
industrial machines consists of only one peak, which rep-
resents the natural frequency of the operating machine.
Thus, defect in the machine can be identified when there is a 8.2. Cepstrum Analysis. Cepstrum analysis was developed in
presence of other peaks aside from the natural frequency the 1960s and can be defined as the power spectrum of the
peak in the plot. However, Goyal and Pabla [37] claimed that logarithm of the power spectrum [74]. Cepstrum analysis
during the conversion between domains, there is a little loss can be used to detect any periodic structure in the spectrum
10 Shock and Vibration

Table 5: The advantages and disadvantages of time domain methods.

Time domain
Advantages Disadvantages
methods
Peak Simple and easy technique Sensitive to noise
Simple and easy technique, directly
RMS related to the energy content of the Changes in RMS vibrations are only sensitive to high-amplitude components
vibration profile
Easily available and affordable crest
Crest factor Only reliable in the presence of significant impulsiveness
factor meter
High performance in detecting
periodic impulse force, highly
Kurtosis sensitive to shock, can be assimilated Costly kurtosis meter can be erroneous
with a shape factor, independent of
the signal amplitude

Table 6: Various reported vibration analysis techniques in machine monitoring and diagnosis.
Authors Methodologies Findings
The proposed method can effectively diagnose the condition
Incorporated the RMS and kurtosis values in NN to diagnose
[63] using only a few features of vibration data but the fault types
the rolling element bearings fault
and severity level cannot be determined
Applied the kurtosis and SVM method to diagnose roller The accuracy of the proposed method is 93.75% and can be
[71]
bearing fault applied even with a limited number of samples
Used the FFT and PSD methods to monitor the condition of the A PC-based vibration analyzer, incorporating the proposed
[23]
machine method was developed
The simulation results obtained from MATLAB/Simulink are
[73] Applied the FFT method to diagnose induction motor fault
in agreement with the experimental results
Combined the cepstrum analysis and NN method to detect and NN can diagnose gear faults with high accuracy, provided that
[78]
diagnose gear fault proper measured data are used
CPW approach is more suitable for applications that do not
Used two cepstrum analysis approaches, namely, automated
require bandpass filtering but applying both approaches
[79] cepstrum editing procedure (ACEP) and cepstrum
without prior knowledge can lead to false result in detecting
prewhitening (CPW) to detect bearing fault
bearing faults
Squared envelope method is an optimal approach in fault
diagnosis in terms of computational cost and simplicity
Applied the envelope analysis to diagnose faults in rotating
[81] compared to the improved synchronous average (ISA), the
machines under variable speed conditions
cepstrum prewhitening (CPW), and the generalized
synchronous average (GSA)
The squared envelope method is more suitable to analyze the
[86] Used the envelope analysis to diagnose bearing faults
cyclostationary signals compared to the envelope method
Combined the higher-order spectrum analysis and SVM to
[90] The proposed method achieved the accuracy of up to 99%
diagnose faults in power electronic circuit
Applied the power spectrum analysis (PSA) and SVM to Using the PSA with SVM classifier gives better result compared
[91]
diagnose rolling bearing fault to NN classifier
Using the proposed method as input to a NN classifier
Applied the WPT method to monitor the condition of the produced nearly 100% classification accuracy. Also, the
[100]
machine proposed method produced a better result compared to FFT
when the data are corrupted by noise
Combined the Hilbert transform and WPT to detect gearbox
[102] The proposed method is capable of detecting early gar fault
fault
The proposed method is more effective and has more
Applied a denoising method based on WT to diagnose rolling
[101] advantages compared to Donoho’s soft-thresholding denoising
bearing and gearbox
method
Proposed an orthonomal DWT (ODWT) method to monitor The proposed method outperforms the EEMD and Hilbert
[103]
and diagnose bearing faults at an early stage envelope spectrum analysis method
HHT outperforms FT, where FT can only differentiate
[115] Applied the HHT and FT methods to diagnose machine fault
characteristic frequency in low-frequency band
Proposed a new local mean parameter to improve the HHT Introducing the new parameter improves the HHT process and
[116]
method to detect gearbox fault makes the fault detection process simpler
Applied the STFT method to diagnose faults in a hydroelectric The basis for the effective fault diagnosis of hydroelectric
[120]
machine machines was proposed
Shock and Vibration 11

Table 6: Continued.
Authors Methodologies Findings
Combined STFT and SVM methods to diagnose faults in The proposed method has a huge potential to diagnose fault
[121]
induction motor intelligently in other real-time system applications
Combined the STFT and nonnegative matrix factorization The proposed method is able to determine the fault types and
[122]
(NMF) methods to detect rolling bearing element fault severity and yields 99.3% accuracy, superior to NN
The proposed method can reliably and quickly diagnose
[125] Applied the PSD method to diagnose Massey Ferguson gearbox
gearbox fault
The proposed method achieved an accuracy of 98.82% and
[126] Used the PSD and SVM methods to diagnose gearbox faults
97.16% for spur and helical gearbox dataset, respectively
SVM trained with 11 time domain statistical features and three
Used the SVM method to diagnose rolling bearing element fractal dimensions provides better results compared to the
[133]
fault based on the fractal dimensions SVM trained with only fractal dimensions or with time domain
statistical features
Applied the SVM, K-nearest neighbor (KNN), and Fischer The performance of the SVM method is much better compared
[134]
linear discriminant (FLD) methods to diagnose engine fault to the KNN and FLD method
Presented a real-time online monitoring approach for the disc Findings show that different types of disc slitting machines’
[135]
slitting machine based on WPT and SVM methods faults can be successfully detected with an accuracy of 95.6%
RMS values were among the statistical features employed as It was found that the deep learning methods are superior
[64] input parameters for the DNN to monitor the condition of compared to the NN method, which has low robustness in
gearbox diagnosing the condition of gearbox
Compared the combination of GA with three types of NN,
namely, multilayer perceptron (MLP), radial basis function The combination of GA with MLP and PNN gave 100% success
[145]
(RBF), and probabilistic neural network (PNN) to detect rate whereas RBF gave 99.31% rate
bearing fault
Applied the combination of EMD and NN methods to diagnose The proposed method can successfully diagnose roller bearing
[146]
the roller bearing fault fault but has an end effect complication
Combined the WPT, GA, NN, and SVM methods to diagnose
[147] The proposed method produced 100% classification accuracy
fault in diesel engine
Proposed a CNN approach that makes use of cyclic spectrum Based on the validation with benchmark vibration data
[148] maps (CSMs) of raw vibration signal to diagnose the motor collected from bearing tests, the proposed technique is superior
bearing in the rotating machine to its referenced methods in terms of the classification accuracy
Proposed a distribution-invariant deep belief network The proposed method is able to achieve a high diagnosis
[149]
(DIDBN) as a basis for intelligent fault diagnosis of machines accuracy even with new working conditions
Used the CNN method with 1D image of raw three-axis It was found that CNN trained with a higher number of kernels
[150]
accelerometer signal as the input in the first layer produced slightly better performance
Proposed a hybrid deep signal processing method to diagnose
The proposed method is superior to the manual extraction
bearing faults in the machine, where the signal processing,
[151] methods and commonly used deep learning structures in terms
feature extraction, and bearing fault diagnosis were
of accuracy, and it is not affected by the operation conditions
automatically conducted
Presented the augmented deep sparse autoencoder (ADSAE) Compared with other deep learning architectures, the proposed
[152] method in diagnosing gear faults, where data shifting technique method provides a higher accuracy (99%) and only requires a
was incorporated to enhance the SAE model few raw vibration signal data
Combined the decision tree and fuzzy logic methods to
The performance of the proposed method in diagnosing fault
[62] diagnose spur gear fault based on the statistical features such as
was found to be 95%
RMS, crest factor, and kurtosis
Proposed the fuzzy logic method to diagnose the operation of The proposed method can easily diagnose the operational status
[160]
rotating machines of the rotating system
Applied the fuzzy logic method to monitor and diagnose the The proposed method can successfully identify and classify the
[161]
condition of the pump faults of the five-plunger pump
Proposed the combination of DWT and fuzzy logic to predict The proposed approach has an error of less than 1% in
[162]
the presence of misalignment in rotating machinery predicting the degree of misalignment
Experts’ knowledge regarding the maintenance of the gas
Developed a gas turbine vibration monitoring approach based
[163] turbine in accordance to the vibration level detected can be
on Takagi–Sugeno fuzzy logic
successfully expressed by the proposed method
Proposed a combination of wavelet support vector machine The proposed method yielded a better diagnostic accuracy
[168] (WSVM) and immune genetic algorithm (IGA) to diagnose compared to the SVM and NN method, in addition to strong
gearbox fault generalization capability
Incorporating the GA to select the parameter of SVM can
Combined the GA, SVM, and EEMD methods to diagnose gear
[169] improve the generalization ability and classification accuracy of
faults
the diagnostic system
Applying the cross-validation method to optimize SVM
Applied the combination of GA and SVM in bearing fault
[170] outperforms the SVM method optimized by GA in bearing
diagnosis
fault diagnosis
12 Shock and Vibration

such as harmonics, sidebands, or echoes [66]. This allows induction motor, and the proposed method can efficiently
faults such as bearing and localized tooth faults, which detect fault without any information regarding the model.
produce low-level harmonically related frequencies to be Application of envelope analysis in machine monitoring can
detected. There are four types of cepstrum which are real also be seen in Table 6 [81, 86].
cepstrum, complex cepstrum, power spectrum, and phase
spectrum, but power cepstrum is the most widely used
8.4. Spectrum Analysis/Comparison. Spectrum analysis is
cepstrum in machine diagnosis and monitoring. According
related to FFT in a way that FFT is often used in spectrum
to Goyal and Pabla [45], cepstrum analysis is important in
analysis to transform the signal from time to frequency
gearbox diagnosis. Dalpiaz et al. [75] compared the cepstrum
domain [87]. Spectrum comparison should be conducted on
analysis with other methods in monitoring the condition of a
a logarithmic amplitude scale (dB) as the changes on a
gearbox and found that cepstrum analysis is insensitive to
logarithmic axis can determine the state of the vibration.
gear cracks. To detect the rubbing phenomenon of the
However, ones have to deal with the small fluctuations of the
sliding bearing in the machine, the cepstrum analysis
rotating speed of the machine [55]. A fault that is able to
method was applied by Sako et al. [76]. It was found that the
change the vibration signature significantly over a short
proposed approach can even detect mild rubbing, which is
period of time can be determined by this method [88].
difficult to be achieved by using conventional abnormality
Spectrum analysis is a complex analysis that even with the
diagnostic methods. Experimental work was conducted by
amount of literature available, expert skills are still required
Aralikatti et al. [77] to diagnose the universal lathe machine,
to exploit the diagnostic capabilities of spectrum analysis.
where cepstrum analysis was employed to the time domain
Compared to the cepstrum analysis, spectrum analysis does
signal. The vibration signal was obtained by a triaxial ac-
not provide any information regarding the time localization
celerometer. It was concluded that analyzing the vibration
of frequency component [77]. Salami et al. [38] have
signal in the frequency domain does not guarantee the
employed the spectrum analysis method for machine con-
presence of a fault. Other applications of cepstrum analysis
dition monitoring, and it is observed that this approach can
can be discovered in Table 6 [78, 79].
produce smoothed and high-resolution spectral estimates of
the vibration signals compared to the FFT approach. This
8.3. Envelope Analysis. Envelope analysis, also known as spectral is useful for monitoring the state of machines.
amplitude demodulation or demodulated resonance anal- Ciabattoni et al. [89] proposed a novel statistical spectrum
ysis, was introduced by Mechanical Technology Inc. [80]. analysis (SSA) where the spectral content of vibration signals
This technique separates the low-frequency signal from was calculated using FFT and then transformed into sta-
background noise [59]. Envelope analysis is made up of a tistical spectral images in diagnosing faults of rotating
bandpass filtering and demodulation step that extracts the machines. Other applications of spectrum analysis can be
signal envelope, and its spectrum possibly contains the observed in Table 6 [90, 91].
desired diagnostic information [81]. It is widely used in
rolling element bearing and low-speed machine diagnosis 9. Time-Frequency Domain Analysis
and has the advantage of early detection of bearing problems
Time and frequency domains are integrated into the time-
[55, 81]. The challenge of this approach is determining the
frequency domain analysis. This means that the signal fre-
best frequency band to envelope. Envelope analysis needs a
quency component and their time-variant features can be
sharp filter and precise specification of the frequency band
determined simultaneously in this analysis. Vibration
for filtering in order to work smoothly [55]. Regarding
analysis approaches mentioned before (time domain and
bearing failure, the noise components make the envelope
frequency domain methods) mostly rely on the stationary
analysis difficult to determine the fault. The introduction of
assumption that is unable to detect the local features in time
the squared envelope analysis method has solved this
and frequency domain simultaneously [92]. Thus, such
problem, where a squared envelope can be computed as
methods are inappropriate for nonstationary signal analysis.
shown in [82]. It is highly preferable in analyzing the
As mentioned before, the time-frequency domain analysis
cyclostationary signals. Rubini and Meneghetti [83] com-
methods discussed in this study include WT, HHT, WVD,
pared the envelope analysis with the wavelet transform (WT)
STFT, and PSD. The advantages and disadvantages of each
method in diagnosing the incipient faults in ball bearings.
time-frequency domain method can be seen in Table 8.
The results demonstrated that after 30 min (48000 cycles),
the envelope analysis is no longer able to diagnose the
presence of the fault whereas the WT method is still relevant. 9.1. WT. WT technique was first proposed by Morlet back in
An envelope analysis approach has also been applied by 1974 [93]. It is a linear transformation decomposing a time
Widodo et al. [84] to preprocess the vibration signals of low- signal into wavelets, which are local functions of time,
speed bearing, and thus, determining the bearing charac- equipped with predetermined frequency content. Instead of
teristic frequencies. This method is then compared with the sinusoidal functions, wavelets are used as the basis [92, 94].
acoustic emission (AE) signal analysis, and during the fault A suitable wavelet basis has to be chosen according to the
recognition stage with the SVM technique, it produces worse signal structure to avoid misleading diagnosis results. WT
performance than the AE approach. Envelope analysis also provides a superior time localization at high frequencies
was applied by Leite et al. [85] to detect the bearing fault in compared to STFT. WT is preferable when dealing with
Shock and Vibration 13

Table 7: The advantages and disadvantages of frequency domain methods.

Frequency domain
Advantages Disadvantages
methods
FFT Easy to implement, fast technique Cannot efficiently analyze transient features in time
Easy to implement, useful in side band Can only be applied to the well separated harmonics, the fluctuations of
Cepstrum Analysis
analysis the curve of the spectrum are averaged-out due to the filtering
Excellent application in bearing system, works
Envelope Can lead to a gross diagnosis error, not suitable to be applied in the gear
well even in the presence of a small random
analysis system
fluctuation
Useful in detecting signal that changes
Spectrum significantly over a short period of time, higher
Require experts’ skills due to its complexity
analysis spectral estimation performance compared to
the FFT

nonstationary signals and in analyzing the transient signal frequency translation of itself [104]. The WVD of a signal
from the measured vibration signal [95]. According to Zou x(t)is represented aswhere x∗ is the conjugate of x and τ is
and Chen [96], the WT technique is highly sensitive to the delay variable. WVD has several advantages such as
stiffness variation compared to WVD. The WT method can better resolution than STFT, excellent accuracy, and window
be distinguished into discrete and continuous wavelet function is not needed for its analysis [45]. WVD is not
transform (DWT and CWT). In DWT, the power of two acts directly used by researchers to determine the time-frequency
as a scaling factor, and it is typically applied through a pair of structures of signals because of the cross-term interference
lowpass and highpass wavelet filters. The scaling factor is problem [93].
selected arbitrarily or via convolution for the CWT [45]. 1 +∞ τ τ
Both DWT and CWT are referred to as standard WT, which WX (t, ω) � 􏽚 x􏼒t + 􏼓 · x∗ 􏼒t − 􏼓 · e− jτω dτ, (6)
2π −∞ 2 2
cannot efficiently perform feature extraction of certain types
of signals due to its incapability to produce a sparse rep- Staszewski et al. [105] performed the fault detection
resentation. It has a low-frequency resolution for high- analysis on the gearbox using the original WVD method and
frequency components and poor time localization for low- its weighted form, and they claimed that compared to the
frequency components. This gives birth to the wavelet packet original WVD, its weighted form can reduce interference in
transform (WPT) technique. It is a more advanced form of the time-frequency domain, with a cost of a reduction in the
CWT, where it further decomposes the detailed information frequency resolution. Directional Wigner distribution
of the signal in the high-frequency region and improves the (dWD) was specifically developed for transient complex-
frequency resolution, making it applicable for the analysis of valued signal analysis and applied in rotating or recipro-
various nonstationary signals. Dalpiaz and Rivola [94] ap- cating machines by [106]. Baydar and Ball [107] used the
plied the WT method in monitoring the condition of the smooth pseudo WVD technique on acoustic and vibration
automatic packaging machine and found that WT is able to signals to diagnose the gearbox condition, and the results
determine the variation of vibration frequency content showed that acoustic signals were more effective in early
within the machine cycle. The advantage of CWT is that it fault detection, compared to vibration signals. An applica-
has a finer scale parameter compared to the DWT method, tion of the WVD method in diagnosing induction machines
but in terms of computational, DWT is more efficient [97]. was demonstrated by Climente-Alarcon et al. [104]. By using
Al-Badour et al. [98] employed the CWT and WPT in this proposed technique, more reliable diagnosis results
detecting faults in rotating machinery. WPT is actually an might be obtained in a situation where harmonics tracing is
extension of the DWT method but with finer frequency difficult.
resolution. They found that in terms of speed and spectral
characterization of the vibration signal, the WPT method is
better than the CWT method. Rangel-Magdaleno et al. [99] 9.3. HHT. David Hilbert first introduced the Hilbert
applied the DWT method to detect the incipient broken bar transform in 1905. Then, Huang et al. [108] introduced the
in the induction motor, and the detection accuracies HHT in 1998 to determine the characteristics of stationary,
achieved are 96.55% for the unload conditions, 80.5% for the nonstationary, and transient signals. HHT consists of em-
half-load, and 87.6% for the full-load condition. Other pirical mode decomposition (EMD) of signals and Hilbert
applications of the WT technique can be found in Table 6 transform, and by combining these two methods, a Hilbert
[100–103]. spectrum can be obtained, where the faults in a running
machine can be diagnosed [92]. Thus, in this manuscript,
any works regarding the EMD method fall into the HHT
9.2. WVD. Wigner introduced the WVD method and Ville section. A complicated multicomponent signal can be
applied it to process the signal, and thus, it was named the broken down into a series of intrinsic mode functions
Wigner–Ville distribution. It is a specific case of the Cohen (IMFs) using this method. By using the EMD technique, a
class distributions which yields a time-frequency energy complicated signal x(t) can be reconstructed with the help
density computed by correlating the signal with a time and of IMFs, expressed as
14 Shock and Vibration

Table 8: The advantages and disadvantages of time-frequency domain methods.

Time-frequency
Advantages Disadvantages
domain methods
Provide a superior time localization at high frequencies Suffers from the convolution of a priori basis functions
WT compared to STFT, more flexible compared to STFT, with the original signal, difficult to choose the mother
availability of various wavelet functions wavelet type
Has a good time and frequency resolution, does not require
WVD Vulnerable to interference, slower than STFT
window function for its implementation
Has a good time and frequency resolution, does not require Result misinterpretation due to IMFs generated at the
HHT
priori basis functions low-frequency region
Straightforward method and recommended for beginners Constant frequency resolution for the whole signal,
STFT in the time-frequency analysis, low computational difficult to find a fast and effective algorithm to calculate
complexity STFT
Can be directly computed by FFT, requires very little
PSD Frequency resolution is affected by the window size
processing power

∞ 9.4. STFT. STFT was pioneered by Gabor back in 1946 in the

x(t) � 􏽚 ci (t) + rn (t)dt, (7) communication field [117]. It has the ability to counter the
−∞
limitations of FFT and is mostly applied to extract the
where ci (t) is the th IMF and xn (t)is the residual signal that narrowband frequency content in nonstationary or noisy
represents the slowly varying or constant trend of the signals [37]. In the STFT method, the initial vibration signal
signal [92]. HHT has several advantages such as low is broken down into time segments by windowing, and then
computational time, and it does not associate with any FT is applied to each time segment [45]. The mathematical
convolution [45]. However, EMD, which is the main part equation for STFT is given by
of HHT, has certain drawbacks. People might misinter- ∞

pret the result due to unenviable IMFs generated at the STFT(f, τ) � 􏽚 x(t)ω(t − τ)e− j2πft dt, (8)
−∞
low-frequency region, and in addition to that, the low-
frequency components’ signals cannot be separated. Thus, where x(t) is the interpreted signal and ω(t) is the window
the ensemble EMD (EEMD) method was introduced to function centered at time Τ. STFT depends on the width of
overcome the limitation of EMD by introducing Gaussian the window. AA large window width is chosen to obtain
white noise to the EMD [92, 109]. However, in the low- greater accuracy in frequency, whereas to improve the ac-
frequency region, the energy leakage and modal aliasing curacy in time, a small window width is desired. The main
problems still exist. This is what motivates Torres et al. drawback of this approach is that it cannot achieve high
[110] to propose the complete EEMD with adaptive noise resolution in the time and frequency domain
(CEEMDAN) method, which can produce better modal simultaneously.
frequency spectrum separation outputs. Safizadeh et al. [118] proposed the STFT application in
Peng et al. [111] introduced a better version of HHT machinery diagnosis and proved that although STFT pro-
that incorporated the WPT technique to decompose the vides the time-frequency information with limited precision,
vibration signal into a set of narrowband signals. The and it is better than the conventional methods of machine
proposed method was shown to have a better resolution in diagnosis. To avoid cross-term effects, Burriel-Valencia et al.
the time and frequency domain compared to the wavelet- [119] implemented the STFT method for fault diagnosis of
based scalogram. Wu et al. [112] employed the HHT induction machines, where the spectrums in the frequency
approach in diagnosing the looseness faults of rotating domain in the relevant frequency band are filtered. The
machinery, and the proposed technique is successful in proposed method greatly reduced the computing time and
determining the faults at different components of the memory resources. The STFT method has also been applied
machine. Osman and Wang [113] proposed a normalized in fault diagnosis of hydroelectric machine [120], induction
HHT (NHHT) technique to tackle the problem of motor [121], and rolling element bearing [122] (refer to
selecting the appropriate distinctive IMF components, Table 6).
especially for bearing health condition monitoring.
However, the EMD of the proposed method only pro-
cesses signals over narrowbands. Chen et al. [114] pro- 9.5. PSD. PSD can be applied to measure the amplitude of
posed a combination of CEEMDAN and particle swarm oscillatory signals in the time series data and determine the
optimization least squares support vector machine (PSO- energy strength of frequencies, which may be useful for
LSSVM) to improve the diagnosis accuracy of rolling further analysis. From the complex spectrum, the one-sided
bearings. The diagnosis accuracy of several rolling bear- PSD can be computed in (m/s2 )2 /Hz as
ings fault types was improved by this method to 100%.
Other applications of HHT in machine monitoring and 2|X(f)|2
PSD(f) � , (9)
diagnosis can be observed in Table 6 [115, 116]. t2 − t1 􏼁
Shock and Vibration 15

where t2 − t1 is the time range and X(f) is the complex The learning and test data for the SVM are obtained from
spectrum of the vibration x(t) in a time range, which can be the feature extraction process and after training the SVM
expressed in units of (m/s2 /Hz). PSD can also be directly algorithm, the SVM matrix was obtained [128]. Optimiza-
calculated in the frequency domain if FFT of vibration signal tion methods such as GA and particle swarm optimization
is used, by applying the following formula [123]: (PSO) are usually incorporated with SVM in order to achieve
2 better results. One of the reasons why SVM is widely applied
Grms 􏼁 in vibration analysis for machine diagnosis is due to its
PSD � , (10)
f compatibility with large and complex datasets such as data
collected in the manufacturing industry [129]. SVM is very
where Grms is the root-mean-square of acceleration in a
useful as the number of features of classified entities will not
certain frequency f. PSD can analyze the faulty frequency
affect the performance of SVM [130]. This means that for the
bands without facing a slip variation issue and does not
base of the diagnosis system, there is no limited number of
necessarily focus on one specific harmonic [124]. It requires
attributes that can be selected. There is no requirement for
very little processing power and can be directly computed by
experts’ knowledge in SVM, as is the case with fuzzy logic,
FFT or by converting the autocorrelation function [45].
and no layers are involved in SVM structure, compared to
PSD technique has been used by Cusido et al. [124]
NN.
alongside WT to diagnose faults in induction machines, and
Poyhonen et al. [130] implemented the SVM technique
the proposed technique can successfully diagnose faults for
to diagnose faults in an electrical machine, and the results
every operating point of the induction motor. However, to
showed that the classification accuracy was high, except for
improve the diagnosis accuracy, good knowledge to deter-
the detection of eccentric rotors. Tabrizi et al. [131] com-
mine the suitable mother wavelet and sampling frequencies
bined the SVM with the WPT (for signal preprocessing) and
are still required. Mollazade et al. [123] also used the PSD
EEMD (for feature extraction) methods to detect small
values in the feature extraction stage and fuzzy logic in the
defects on roller bearings under different operating condi-
fault recognition stage of fault diagnosis of hydraulic pumps.
tions. A classification tree kernel-based SVM has also been
The classification accuracy of the proposed technique for
employed together with CWT to identify bearing faults, and
1000, 1500, and 2000 rpm conditions is 96.42%, 100%, and
this combination proves to be a promising method and
96.42%, respectively. Other applications of PSD in machine
superior to other SVM methods with common kernels in
monitoring and diagnosis can be seen in Table 6 [125, 126].
diagnosing the rolling element bearing fault [132]. Pinheiro
et al. [128] used the SVM method for fault diagnosis of a
10. Fault Recognition/AI-Based Technique rotary machine and successfully detected several unbalance
(RQ 3) faults. However, its performance is still questionable as a
small number of samples are used. Further implementation
The application of AI in vibration analysis for machine
of SVM in vibration analysis for machine diagnosis can be
monitoring and diagnosis has become increasingly popular,
found in Table 6 [90, 133–135].
and based on this review, AI-based techniques contribute
about 57% of the overall vibration analysis method in
machine diagnosis and monitoring, as shown in Figure 5.
10.2. NN. NN is made up of a large number of richly
This is because most of the techniques mentioned before
interconnected artificial processing neurons called nodes,
require huge expertise for successful implementation, which
connected to each other in layers forming a network [136].
makes them not suitable for common users [80]. Further-
NN has the ability to model processes and systems from raw
more, the expert is not immediately available. This is where
vibration data extracted from frequency and time-frequency
AI-based methods come in because a nonexpert user can
domain techniques mentioned before [137]. In the training
make reliable decisions without the presence of a machine
stage of NN, superior input variables might suppress the
diagnosis expert. AI can be defined as any task performed by
influence of the weak variables. Thus, the data must be
a program or a machine that is difficult enough that it re-
properly processed and scaled before being fed into the NN.
quires intelligence to accomplish it [127]. Several AI-based
Normalizing the raw vibration data to the values between 0
methods of vibration analysis for machine monitoring and
and 1 can help to reduce the effect of the input variable [137].
diagnosis are SVM, NN, fuzzy logic, and GA.
Training time increases according to the complexity of the
network, and this directly affects the accuracy of the results.
10.1. SVM. SVM was initially introduced by Vapnik and is Due to the robustness and efficiency in handling noisy data,
the most widely used classification algorithm. This method the backpropagation neural network (BPNN) is widely used
transforms the data set or sample space to a high-dimen- in machine diagnosis [138]. BPNN, pioneered by Rumelhart
sional, kernel-induced feature space by nonlinear trans- and McClelland in 1986, is made up of three layers, which
formation and then determines the best hyperplane [1]. The are input, hidden, and output layer [93]. The presence of
best hyperplane means the one with the largest margin hidden layers gives the NN an ability to explain nonlinear
between the two classes A and B, as shown in Figure 6. The systems, and the higher the number of hidden layers, the
data points from both classes that are closer to the hyper- deeper the NN [139]. Similar to the SVM method, NN does
plane and influence the position and orientation of the not need a knowledge base to detect the location of the faults,
hyperplane are called support vectors. unlike the fuzzy logic method. Ertunc et al. [140] compared
16 Shock and Vibration

using the bearing datasets from Case Western Reverse

University, but hyperparameters for the CNN model can still
43%
be improved. A combination of transfer learning and DNN
has been employed by Qian et al. [143] in diagnosing the
57%
rotating machines under different working conditions.
Transfer learning focuses on how to store the knowledge or
solution obtained when solving a problem and apply it to
different but related problems so that the amount of data
collection and training costs can be reduced [144]. The
proposed method is robust, and there is no requirement for
AI-Method
further training when supplied with new datasets from
Non AI-Method
different working conditions. Similar applications can also
Figure 5: The percentage difference between AI and non-AI be found in Table 6 [78, 145–152].
methods in vibration analysis for machine diagnosis and
monitoring.
10.3. Fuzzy Logic. Compared to conventional logic, fuzzy
logic aims at modeling the imprecise modes of reasoning to
make rational decisions in an environment of uncertainty
and imprecision [153, 154]. There are four main stages of the
Class A Marging
fuzzy logic system, which are fuzzification, an inference
mechanism, rule-base, and defuzzification component. The
Separating fuzzification stage converts the input data into fuzzy sets
Hyperplane before the fuzzy inference stage makes a reliable conclusion
based on the rules created in the rule-base stage. Finally, the
Class B defuzzification stage produces quantifiable results. Fuzzy
Support Vector logic is associated with membership functions, which role is
Figure 6: The structure of SVM technique. to map the nonfuzzy input values to fuzzy linguistic terms
and vice versa. To obtain a diagnostic system with excellent
sensitivity, the rules and membership functions can be tuned
the performance of NN and the combination of NN and [155]. However, properly determine the fuzzy rules and
fuzzy logic, which is known as adaptive neurofuzzy inference optimize the membership functions are the biggest challenge
systems (ANFIS). Envelope analysis was applied for the in fuzzy logic. Fuzzy logic is easier to implement compared
signal processing step. They concluded that the ANFIS to SVM and NN. Apart from that, unlike other AI methods
method was superior to the NN, especially in diagnosing such as SVM and NN, it does not rely on the datasets as there
fault severity. Castelino et al. [137] used the application of is no training or testing stage in fuzzy logic. In particular
NN in the vibration monitoring carried out on industrial cases, this method can only provide a general diagnosis, as
rotary machines that were running in real operating con- the specific fault symptom of a machine cannot be regularly
ditions. The results showed that NN performed better for determined. However, this is the only available alternative
nonstationary signals in the time-frequency domain com- when collecting the fault data is not possible [156]. Lasurt
pared to the frequency domain. Recently, the deep learning et al. [157] compared the performance of fuzzy logic with
or deep neural network (DNN) has been applied widely in NN in fault diagnosis of electrical machines, and they
machine monitoring and diagnosis. It is a type of NN that claimed that fuzzy logic performs better in detecting dif-
contains more than one hidden layer. Compared to the NN ferent faults for a wide range of operating conditions,
and SVM method, the DNN approach can adaptively learn compared to NN. Wu and Hsu [158] combined the method
the hierarchical representation from raw data through of DWT and fuzzy logic to detect gear fault and the results
multiple nonlinear transformations and approximate obtained showed that the recognition rate of the proposed
complex nonlinear functions instead of extracting the fault method is over 96% under various experimental conditions.
feature manually [141]. However, due to its deep architec- Mukane et al. [159] applied the FFT method for signal
ture, a high number of parameters are involved, which leads processing and fuzzy logic for fault recognition in identi-
to the risk of overfitting. Widely applied DNN architectures fying machinery faults. Multiple faults can be identified in
include autoencoders (AE), convolutional neural network this manner including the severity of the faults. Other ap-
(CNN), restricted Boltzmann machines, and deep belief plications of fuzzy logic in vibration analysis for machine
networks. Table 9 shows the comparison between NN and diagnosis can be found in Table 6 [160–163].
deep learning/DNN in machine monitoring and diagnosis.
Hoang and Kang [142] applied the CNN technique, 10.4. GA. GA, derived from a study of a biological system,
which is a class of DNN that is most commonly applied for can solve both constrained and unconstrained optimiza-
image analysis, to diagnose the rolling element bearing in tion problems based on a natural selection process
rotary machines. Raw vibration signals in time domain are [164, 165]. At each step, GA randomly selects the best
converted into a 2D form for the CNN to perform vibration individuals based on their quality from the current pop-
image classification. This method achieved 100% accuracy ulation to become parents for the children of the next
Shock and Vibration 17

Table 9: NN vs. deep learning/DNN. in turn, produces a more accurate measurement. However,
Characteristics NN Deep learning/DNN
due to its costly application, it is not widely used. LDV with
multichannel measurements is expected to be widely applied
Prior knowledge Required Not required
in the future, where the cost of implementing the LDV is
Computational burden Lower Higher
Feature extraction Manual Automated reduced.
Accuracy Lower Higher Regarding the signal processing techniques (RQ 2), the
Robustness Lower Higher works on improving the detection and diagnosis of faults in
the time-frequency domain have attracted numerous at-
generation, in order to reach an optimal solution. This step tention from researchers. This is because it can be imple-
will continue until a terminating condition is reached. mented to investigate the nonstationary signals as failure
There are three main processes that occur at each GA, signals are not repetitive at the earliest stage. Usually, these
namely, selection, crossover, and mutation. GA is usually nonstationary signals contain abundant information on
used to optimize the monitoring system parameters and machine faults. Time and frequency domain techniques for
boosting the speed and accuracy of fault diagnosis. Samanta machine monitoring are based on the assumption of sta-
et al. [145] presented a combination of ANN and SVM with tionary signals, and this is not suitable for detecting short-
GA for bearing fault detection. Based on the result, SVM duration dynamic phenomena, especially in rotating ma-
performs better than NN, and GA helps to reduce the chinery. However, traditional methods should not be
training time of both methods. Han et al. [166] used the GA omitted as it is preferable in certain applications. For low-
in the process of diagnosing the induction motor and found speed machines, envelope analysis is widely applied as it can
that GA helps the diagnosis system to perform better by detect low energy signals and the envelope spectrum is
choosing critical features and optimizing the network further analyzed using time-frequency domain methods
structure. Hajnayeb et al. [167] claimed that removing such as WT and HHT, where the noise level in the vibration
some features from the input features will result in quicker signals is reduced. To make use of the advantages of a certain
and more accurate diagnosis systems. GA is usually method and to make up for the limitations, some researchers
combined with other AI methods such as SVM, fuzzy logic, applied the fusion of certain techniques. Based on
and NN. In NN, the GA can be used as an alternative to Figures 7(a)–7(c), the most applied time, frequency, and
learning the weight values and to optimize the topology of a time-frequency domain methods in machine monitoring
NN. For the fuzzy control, the GA can be used to tune the and diagnosis are RMS, FFT, and WT techniques,
associated membership function parameters as well as respectively.
generating the fuzzy rules. The applications of GA can also To answer the RQ 3, researchers also move towards
be observed in Table 6 [168–170]. The advantages and implementing the intelligence system in the vibration
disadvantages of each fault recognition method can be seen analysis for automated decision making, and from the
in Table 10. review and referring to Figure 7(d), SVM is the most widely
used method mainly due to its high classification accuracy
11. Discussion and low computational time. Besides, it was found that the
application of the time domain parameters is directly
More than 100 articles were discussed in this study, covering proportional to the applications of AI methods. This is
a topic associated with the vibration analysis in machine because time domain features can improve the perfor-
monitoring and diagnosis in terms of instruments used in mance of AI methods and have a low computational cost,
the data acquisition stage, feature extraction methods, and which would not put much computational burden on AI
fault recognition by AI techniques. Because there is a large methods. The works on refining the algorithms for lower
amount of literature on this field, a review of all the literature computational cost and easy implementation of the vi-
is impossible, and some papers might be omitted. For the bration analysis are still in progress for the AI-based
data acquisition process and to answer the RQ 1, most of the techniques. Based on the previous studies regarding vi-
studies applied the simpler and cheaper alternative of the bration analysis for machine monitoring and diagnosis,
computer-based analyzer, and with the help of DSP and most of the reported studies applied the vibration analysis
FPGA, this analyzer is nearly as good as the conventional method on one test rig or machine, where the results
analyzer. Accelerometer is still the best sensor option for produced are excellent but the same performance cannot be
vibration analysis, and this has been proved by most of the guaranteed when applied to other machines. This also
articles reviewed. However, due to the high cost of the pi- applies to the environment, where most of the reported
ezoelectric accelerometer, researchers are continuously works are conducted in a controlled environment and the
working towards the application of the MEMS accelerom- performance might differ if employed in an industrial
eter, which can provide the same or better performance. setting. Similar cases applied to the fault recognition using
Velocity transducer is preferable to be applied in diagnosing AI, especially in SVM and NN methods. These data-driven
low-speed machinery compared to accelerometers as the methods are based on the training of historically obtained
absolute accelerations measured are much smaller in value datasets fed to the algorithm, and when entirely new
for similar vibration displacements. Noncontact sensors also datasets are used, generalization issues might occur. Thus,
have a huge potential in machine monitoring as mounting it is important to train the AI algorithms with appropriate,
the sensor on the machine is not a concern anymore which diverse, and optimized datasets. It is also recommended to
18 Shock and Vibration

Table 10: The advantages and disadvantages several AI-based methods.

AI-based
Advantages Disadvantages
method
Compatible with large and complex dataset, high Difficult to determine the appropriate kernel to be applied,
SVM
accuracy perform poorly when the data contain noise
Fault tolerance process, have a self-learning ability Have Complicated design process, takes a long time to process a large
NN
a strong capability to process data network, black box solution
Difficult to obtain the knowledge rules, difficult to determine the
Fuzzy logic Strong robustness, simple design and easy to understand
right membership function
Can process an extremely large range of data, can be
GA High computing cost, time-consuming procedure
applied for the optimization purpose

18% 21%
31%
39%
19%
36%
15%
21%

Peak Crest Factor FFT Envelope

RMS Kurtosis Cepstrum Spectrum
(a) (b)
10%
13%
14% 40%
36%
24%
25%
8%
30%

WT
WVD STFT SVM Fuzzy Logic
HHT PSD NN GA
(c) (d)

Figure 7: Percentage of methods used for (a) time domain, (b) frequency domain, (c) time-frequency domain, and (d) AI-based approaches,
based on the reported articles.

test the AI techniques at different operating or environ- In terms of noise, most of the reported existing
mental conditions than the training datasets to counter the methods can effectively distinguish noise from vibration
generalization problem. It was discovered that almost 80% signals. However, this is based on the assumption of
of the previous studies applied the feature extraction or Gaussian distribution vibration signals. In an industrial
signal processing techniques such as envelope analysis, environment, vibration signals are usually corrupted with
STFT, WT, and HHT, whether with the combination of AI non-Gaussian noise, due to the abnormal operation of
method or not. Although these techniques achieved good gears or bearings and random disturbances that occur in
performance in monitoring and diagnosing the machine the machine, which is not widely considered in the re-
condition, expert knowledge regarding signal processing is ported techniques. This is because industrial machines are
still required. Furthermore, the feature extractor has to be complex systems made up of various components such as
reconstructed for every specific fault diagnosis task. This is shafts, bearings, and gearboxes, which run simulta-
one of the reasons researchers are migrating to the deep neously. Thus, some fault signatures are often covered by
learning approach where fault features are automatically the machine natural frequencies and submerged by high
extracted from the raw vibration signals. non-Gaussian noise, and as a result, the faulty frequencies
Shock and Vibration 19

Table 11: The most influencing articles for each methods.

Number
Methods Articles’s name
of citation
Fault diagnostics of rolling element bearing using the NN method and the time domain features of RMS
Time domain and kurtosis conducted by [63]. From the results, the proposed method can still effectively diagnose the 762
machine even with the reduced number of inputs.
A research conducted by [86] to study the relationship between the classical envelope analysis and
Frequency domain spectral correlation analysis in the diagnostics of bearing faults. It was shown that the envelope analysis 613
provided the same results as the complex spectral correlation function.
A study on applying the CWT method in the feature extraction of mechanical vibration signals
Time-frequency
conducted by [101]. The results proved that the proposed method is more effective than the Donoho’s 1050
domain
“soft-thresholding denoising” method.
AI Similar to [63]. 762

become nondominant in the spectrum, which makes the technology and development in the data acquisition process
machine diagnosis and monitoring process more difficult and signal processing techniques including the instrument
[171, 172]. applied. Nowadays, even inexperienced users can conduct
For RQ 4, recent applications of smart machines present a effective vibration monitoring without the presence of an
new and huge challenge in monitoring and diagnosing their expert. In this study, we have conduct a systematic review of
condition. The presence of various sensors and communication vibration analysis for machine monitoring and diagnosis,
devices in the smart machine will produce very noisy data. which can be divided into data acquisition, feature extrac-
Thus, research studies on a vibration technique that is precise tion, and fault recognition stages. Several RQs have been
and robust and can handle a huge amount of noisy data ef- answered in this study which might provide useful infor-
ficiently are desired. Sometimes the collected vibration data mation on this area. From the study, several key factors are
from the sensors are insufficient for machine diagnosis and determined:
monitoring. This is why the need for the Digital Twin modeling
(i) With the advancement of powerful software and the
approach arises. According to [173, 174], the Digital Twin
Internet, a computer-based analyzer is preferable in
model can map various characteristics of the physical machine
the future due to its low cost and performance,
into the virtual world to produce a digital replica of the ma-
which is as good as the standalone analyzer.
chine that is transferable, detachable, modifiable, reproducible,
repeatable, and erasable. Thus, extrapolating the vibration data (ii) Noncontact sensor is the future of vibration analysis
acquired from the sensors is possible based on the mathe- for machine monitoring and diagnosis due to its
matical representation of the machine. By combining the AI- flexibility and independence of any mass-loading
based data-driven approach and physics-based simulation effects, without compromising the signal quality.
model, Digital Twin can obtain additional information re- (iii) Time and frequency domain methods are suitable
garding the prediction of machine failures. This model can also for stationary signals and time-frequency domain
be applied to run several simulations in different operational techniques are preferable for nonstationary signals
and environmental conditions, increasing the robustness of the and early fault detection.
diagnosis technique. However, producing an effective and (iv) Deep learning, especially the deep transfer learning
proper Digital Twin model remains a challenge due to the method, is starting to be applied in vibration
nonlinear dynamics, and uncertainty that occurs in the op- analysis for machine monitoring and diagnosis as it
erating smart machines [175]. Thus, works on constructing a helps in minimizing the requirement for expert
Digital Twin model that can properly represent the actual knowledge in the complicated feature extraction
conditions of the machine still can be explored. In addition, a step. Traditional AI methods such as SVM, NN, and
technique to monitor and diagnose the machine’s condition fuzzy logic still require expert knowledge in the
from a remote location without the need to visit the machine is feature extraction stage of newly fed datasets.
worth exploring. For AI-based techniques, the performance
(v) Traditional time domain features such as RMS and
can be further evaluated by studying the effects of simultaneous
crest factor are still relevant in the future and its
fault occurrence on the machine. In terms of AI algorithms,
application with AI will continue to increase.
incorporating the transfer learning approach into the algorithm
can be further explored since the deep transfer learning method
is still in its early stages. Table 11 shows the most cited articles Data Availability
reviewed in this manuscript for each method to answer the RQ
The data used to support the findings of this study are
5.
available from the corresponding author upon request.
12. Conclusion
Conflicts of Interest
Vibration analysis for machine monitoring and diagnosis
has become cheaper and cheaper thanks to the emerging The authors declare no conflicts of interest.
20 Shock and Vibration

Acknowledgments and Control (ICMIC), pp. 37–42, Algiers, Algeria, November

2016.
The authors would like to thank Synvue Sdn Bhd (1236749- [14] A. Aherwar, “An investigation on gearbox fault detection
D) for insights and feedback of the project development. using vibration analysis techniques: a review,” Australian
This work was supported by the Collaborative Research in Journal of Mechanical Engineering, vol. 10, no. 2, pp. 169–
Engineering, Science, and Technology (CREST) under grant 183, 2012.
304/PELECT/6050424/C121. [15] A. S. Sait and Y. I. Sharaf-Eldeen, “A review of gearbox
condition monitoring based on vibration analysis techniques
diagnostics and prognostics,” Rotating Machinery, Structural
References Health Monitoring, Shock and Vibration, Volume 5, vol. 5,
[1] A. Aherwar and M. S. Khalid, “Vibration analysis techniques pp. 307–324, 2011.
for gearbox diagnostic: a review,” International Journal of [16] I. Sadeghi, H. Ehya, J. Faiz, and H. Ostovar, ““Online fault
Advances in Engineering & Technology, vol. 3, no. 2, pp. 4–12, diagnosis of large electrical machines using vibration signal-a
2012. review,” in,” in Proceedings of the 2017 International Con-
[2] D. H. C. de Sá Só, D. P. Viana, A. A. de Lima et al., “Di- ference on Optimization of Electrical and Electronic Equip-
agnostic and severity analysis of combined failures composed ment (OPTIM) & 2017 Intl Aegean Conference on Electrical
by imbalance and misalignment in rotating machines,” In- Machines and Power Electronics (ACEMP)., pp. 470–475,
ternational Journal of Advanced Manufacturing Technology, Brasov, Romania, May 2017.
vol. 114, no. 9, pp. 3077–3092, 2021. [17] B. Kitchenham, S. Charters, D. Budgen et al., Guidelines for
[3] S. Kumar, M. Lokesha, K. Kumar, and K. Srinivas, “Vibration Performing Systematic Literature Reviews in Software Engi-
based fault diagnosis techniques for rotating mechanical neering, UK, Tech. Rep., Keele University and University of
components: review paper,” in Proceedings of the Interna- Durham, Keele, England, 2007.
tional Conference on Advances in Manufacturing, Materials [18] K. Shahzad, P. Cheng, and B. Oelmann, “Architecture ex-
and Energy Engineering, pp. 1–6, Karnataka, India, June 2018. ploration for a high-performance and low-power wireless
[4] R. Ranjan, S. K. Ghosh, and M. Kumar, “Fault diagnosis of vibration analyzer,” IEEE Sensors Journal, vol. 13, no. 2,
journal bearing in a hydropower plant using wear debris, pp. 670–682, 2012.
vibration and temperature analysis: a case study,” Proceed- [19] C. Scheffer and P. Girdhar, Practical Machinery Vibration
ings of the Institution of Mechanical Engineers - Part E: Analysis and Predictive Maintenance, Elsevier, Netherlands,
Journal of Process Mechanical Engineering, vol. 234, no. 3, Europe, 2004.
pp. 235–242, 2020. [20] M. Nuawi, S. Abdullah, A. Ismail, and N. Kamaruddin, “A
[5] A. Glowacz, “Ventilation diagnosis of angle grinder using novel analysis (i-kaz 3d) for three axial vibration signal in
thermal imaging,” Sensors, vol. 21, no. 8, p. 2853, 2021. bearing condition monitoring,” Statistics, vol. 6, no. 7,
[6] A. Glowacz, R. Tadeusiewicz, S. Legutko et al., “Fault di- pp. 318–322, 2008.
agnosis of angle grinders and electric impact drills using [21] S. Kumar, R. Sehgal, R. Kumar, and S. Bhandari, “Vibrations
acoustic signals,” Applied Acoustics, vol. 179, p. 108070, 2021. analysis of 4 jaw flexible coupling considering unbalancing in
[7] D. N. Brown and T. Jensen, “Machine-condition monitoring two planes,” International Journal of Science and Technology,
using vibration analysis,” 2020, https://www.bksv.com/ vol. 1, no. 11, pp. 602–608, 2012.
media/doc/bo0253.pdfAvailable:. [22] Q. He and D. Du, “A new dual-channel analyzer for vibration
[8] J. J. Saucedo-Dorantes, M. Delgado-Prieto, J. A. Ortega- of rotary machine,” in Proceedings oif the ASME Power
Redondo, R. A. Osornio-Rios, and R. D. J. Romero-Tron- Conference, pp. 253–256, Orlando, FL, USA, May 2008.
coso, “Multiple-fault detection methodology based on vi- [23] S. A. Ansari and R. Baig, “A pc-based vibration analyzer for
bration and current analysis applied to bearings in induction
condition monitoring of process machinery,” IEEE Trans-
motors and gearboxes on the kinematic chain,” Shock and
actions on Instrumentation and Measurement, vol. 47, no. 2,
Vibration, vol. 2016, Article ID 5467643, 13 pages, 2016.
pp. 378–383, 1998.
[9] W.-B. Zoungrana, A. Chehri, and A. Zimmermann, “Au-
[24] A. Gani and M.-J. E. Salami, “A labview based data acqui-
tomatic classification of rotating machinery defects using
sition system for vibration monitoring and analysis,” in
machine learning (ml) algorithms,” Human Centred Intel-
Proceedings of the Student Conference on Research and De-
ligent Systems, vol. 189, pp. 193–203, 2020.
[10] S. Elango, J. G. Aravind, and S. Boopathi, “Vibration analysis velopment, pp. 62–65, Shah Alam, Malaysia, July 2002.
of bearing by using mechanical stethoscope,” International [25] P. Bilski and W. Winiecki, “A low-cost real-time virtual
Journal of Advanced Science and Research, vol. 3, no. 1, spectrum analyzer,” in Proceedings of the IEEE Instrumentation
pp. 1137–1149, 2018. and Measurement Technology Conference, pp. 2216–2221,
[11] Y. H. Kim, A. C. C. Tan, and V. Kosse, “Condition moni- Ontario, Canada, May 2005.
toring of low-speed bearings - a review,” Australian Journal [26] G. Betta, C. Liguori, A. Paolillo, and A. Pietrosanto, “A dsp-
of Mechanical Engineering, vol. 6, no. 1, pp. 61–68, 2008. based fft-analyzer for the fault diagnosis of rotating machine
[12] M. Vishwakarma, R. Purohit, V. Harshlata, and P. Rajput, based on vibration analysis,” IEEE Transactions on Instru-
“Vibration analysis & condition monitoring for rotating mentation and Measurement, vol. 51, no. 6, pp. 1316–1322,
machines: a review,” in Proceedings of the 5th International 2002.
Conference of Materials Processing and Characterization, [27] J. D. J. Rangel-Magdaleno, R. D. J. Romero-Troncoso,
pp. 2659–2664, Elsevier, Hyderabad, India, March 2016. R. A. Osornio-Rios, E. Cabal-Yepez, and A. Dominguez-
[13] A. Boudiaf, A. Djebala, H. Bendjma, A. Balaska, and Gonzalez, “Fpga-based vibration analyzer for continuous cnc
A. Dahane, “A summary of vibration analysis techniques for machinery monitoring with fused fft-dwt signal processing,”
fault detection and diagnosis in bearing,” in Proceedings of IEEE Transactions on Instrumentation and Measurement,
the 8th International Conference on Modelling, Identification vol. 59, no. 12, pp. 3184–3194, 2010.
Shock and Vibration 21

[28] C. Rodriguez-Donate, R. Romero-Troncoso, A. Garcia- [43] J. Doscher, “Adxl105: a lower-noise, wider-bandwidth ac-
Perez, and D. A. Razo-Montes, “Fpga based embedded celerometer rivals performance of more expensive sensors,”
system for induction motor failure monitoring at the start-up Analog Dialogue, vol. 33, no. 6, pp. 27–29, 1999.
transient vibrations with wavelets,” in Proceedings of the [44] S. Thanagasundram and F. S. Schlindwein, “Comparison of
International Symposium on Industrial Embedded Systems, integrated micro-electrical-mechanical system and piezo-
pp. 208–214, Le Grande Motte, France, June 2008. electric accelerometers for machine condition monitoring,”
[29] G. Betta, C. Liguori, and A. Pietrosanto, “A multi-application Proceedings of the Institution of Mechanical Engineers - Part
fft analyzer based on a dsp architecture,” IEEE Transactions C: Journal of Mechanical Engineering Science, vol. 220, no. 8,
on Instrumentation and Measurement, vol. 50, no. 3, pp. 1135–1146, 2006.
pp. 825–832, 2001. [45] D. Goyal and B. S. Pabla, “The vibration monitoring methods
[30] L. M. Contreras-Medina, R. de Jesus Romero-Troncoso, and signal processing techniques for structural health
E. Cabal-Yepez, J. de Jesus Rangel-Magdaleno, and monitoring: a review,” Archives of Computational Methods in
J. R. Millan-Almaraz, “Fpga-based multiple-channel vibra- Engineering, vol. 23, no. 4, pp. 585–594, 2016.
tion analyzer for industrial applications in induction motor [46] A. Fernandez, Seismic Velocity Transducers, https://power-
failure detection,” IEEE Transactions on Instrumentation and mi.com/content/seismic-velocity-transducers[Online].
Measurement, vol. 59, no. 1, pp. 63–72, 2009. Available:, 2020.
[31] M. Kashiwagi, C. d. Costa, and M. H. Mathias, “Digital [47] M. P. Boyce, Gas Turbine Engineering Handbook, Elsevier,
systems design based on dsp algorithms in fpga for fault Oxford, England, 2011.
identification in rotary machines,” Journal of Mechanics & [48] G. Rossi, “Vibration analysis for reciprocating compressors,”
Industry Research, vol. 2, no. 1, pp. 1–5, 2014. ORBIT Magazine, vol. 32, no. 2, pp. 10–15, 2012.
[32] B. Deb Majumder, J. K. Roy, and S. Padhee, “Recent advances [49] A. A. Sarhan, A. Matsubara, S. Ibaraki, and Y. Kakino,
in multifunctional sensing technology on a perspective of “Monitoring of cutting force using spindle displacement
multi-sensor system: a review,” IEEE Sensors Journal, vol. 19, sensor,” in Proceedings of the Japan – USA Symposium on
no. 4, pp. 1204–1214, 2019. Flexible Automation, pp. 1–4, Denver, CO, USA, July 2004.
[33] C. Sanders, “A guide to vibration analysis and associated [50] S. M. Saimon, N. H. Ngajikin, M. S. Omar et al., “A low-cost
techniques in condition monitoring,” 2020, https://www. fiber based displacement sensor for industrial applications,”
scribd.com/document/320879762/Vibration-Analysis- TELKOMNIKA (Telecommunication Computing Electronics
Guideaccessed: 20 October 2020. and Control), vol. 17, no. 2, pp. 555–560, 2019.
[34] K. W. Cheung, M. J. Starling, and P. D. McGreevy, “A [51] S. Binu, V. P. Mahadevan Pillai, and N. Chandrasekaran,
comparison of uniaxial and triaxial accelerometers for the “Fibre optic displacement sensor for the measurement of
assessment of physical activity in dogs,” Journal of Veterinary amplitude and frequency of vibration,” Optics & Laser
Behavior, vol. 9, no. 2, pp. 66–71, 2014. Technology, vol. 39, no. 8, pp. 1537–1543, 2007.
[35] H. N. Norton, Handbook of Transducers, Prentice-Hall, [52] N. Hasheminejad, C. Vuye, W. Van den bergh, J. Dirckx, and
Hoboken, NJ, USA, 1989. S. Vanlanduit, “A comparative study of laser Doppler
[36] S. Xianzhong, J. Zhuangde, L. Peng, G. Lin, and J. Xingdong, vibrometers for vibration measurements on pavement ma-
“A novel pvdf based high-gn shock accelerometer,” in terials,” Infrastructure, vol. 3, no. 4, p. 47, 2018.
Proceedings of the 7th International Symposium on Mea- [53] X. Ma, M. Li, L. Ye et al., “Research on self calibration
surement Technology and Intelligent Instruments, pp. 107– technology for precision workpiece measurement based on
110, Huddersfield,England, April 2005. laser Doppler vibrometer,” Integrated Ferroelectrics, vol. 199,
[37] D. Goyal and B. S. Pabla, “Condition based maintenance of no. 1, pp. 160–171, 2019.
machine tools-A review,” CIRP Journal of Manufacturing [54] S. J. Rothberg, M. S. Allen, P. Castellini et al., “An inter-
Science and Technology, vol. 10, pp. 24–35, 2015. national review of laser Doppler vibrometry: making light
[38] M.-J. E. Salami, A. Gani, and T. Pervez, “Machine condition work of vibration measurement,” Optics and Lasers in En-
monitoring and fault diagnosis using spectral analysis gineering, vol. 99, pp. 11–22, 2017.
techniques,” in Proceedings of the 1st International Confer- [55] I. Howard, “A review of rolling element bearing vibra-
ence on Mechatronics, pp. 690–700, Kuala Lumpur, Malaysia, tion’detection, diagnosis and prognosis,” Report for the
June 2001. Defence Science and Technology Organization Canberra
[39] J. Igba, K. Alemzadeh, C. Durugbo, and E. T. Eiriksson, (Australia). Report no. DSTO-RR-0013, October. Victoria,”
“Analysing rms and peak values of vibration signals for DSTO Aeronautical and Maritime Research Laboratory.,
condition monitoring of wind turbine gearboxes,” Renew- Tech. Rep., Melbourne, Victoria, Australia, 1994.
able Energy, vol. 91, pp. 90–106, 2016. [56] P. Gangsar and R. Tiwari, “Multiclass fault taxonomy in
[40] A. Khadersab and S. Shivakumar, “Vibration analysis rolling bearings at interpolated and extrapolated speeds
techniques for rotating machinery and its effect on bearing based on time domain vibration data by svm algorithms,”
faults,” Procedia Manufacturing, vol. 20, pp. 247–252, 2018. Journal of Failure Analysis and Prevention, vol. 14, no. 6,
[41] S. B. Chaudhury, M. Sengupta, and K. Mukherjee, “Vibration pp. 826–837, 2014.
monitoring of rotating machines using mems accelerome- [57] S. Lahdelma and E. Juuso, “Signal processing in vibration
ter,” International Journal of Scientific Engineering and Re- analysis,” in Proceedings of the 5th International Conference
search, vol. 2, no. 9, pp. 1–11, 2014. on Condition Monitoring and Machinery Failure Prevention
[42] L. M. Contreras-Medina, R. Romero-Troncoso, J. R. Millan- Technologies, pp. 867–878, Edinburgh, Scotland, July 2008.
Almaraz, and C. Rodriguez-Donate, “Fpga based multiple- [58] A. Shrivastava and S. Wadhwani, “An approach for fault
channel vibration analyzer embedded system for industrial detection and diagnosis of rotating electrical machine using
applications in automatic failure detection,” in Proceedings of vibration signal analysis,” in Proceedings of the International
the International Symposium on Industrial Embedded Sys- Conference on Recent Advances and Innovations in Engi-
tems, pp. 229–232, Le Grande Motte, France, June 2008. neering, pp. 1–6, Jaipur, India, May 2014.
22 Shock and Vibration

[59] M. Vishwakarma, R. Purohit, V. Harshlata, and P. Rajput, [75] G. Dalpiaz, A. Rivola, and R. Rubini, “Gear fault monitoring:
“Vibration analysis & condition monitoring for rotating comparison of vibration analysis techniques,” in Proceedings
machines: a review,” Materials Today: Proceedings, vol. 4, of the 3rd International Conference Acoustical and Vibratory
no. 2, pp. 2659–2664, 2017. Surveillance Methods and Diagnostics Techniques, pp. 623–
[60] W. Bartelmus, F. Chaari, R. Zimroz, and M. Haddar, 632, Las Vegas, NV, USA, January 1998.
“Modelling of gearbox dynamics under time-varying non- [76] T. Sako, K. Tokumo, and O. Yoshie, “Rubbing detection of
stationary load for distributed fault detection and diagnosis,” sliding bearings in early stage by frequency modulation
European Journal of Mechanics - A: Solids, vol. 29, no. 4, analysis,” Electronics and Communications in Japan, vol. 97,
pp. 637–646, 2010. no. 12, pp. 11–23, 2014.
[61] J. Sheldon, G. Mott, H. Lee, and M. Watson, “Robust wind [77] S. S. Aralikatti, K. Ravikumar, and H. Kumar, “Fault diag-
turbine gearbox fault detection,” Wind Energy, vol. 17, no. 5, nosis of single point cutting tool using spectrum, cepstrum
pp. 745–755, 2014. and wavelet analysis,” in Proceedings of the 1st International
[62] A. Krishnakumari, A. Elayaperumal, M. Saravanan, and Conference on Manufacturing, Material Science and Engi-
C. Arvindan, “Fault diagnostics of spur gear using decision neering, pp. 1–9, Telangana, India, August 2019.
tree and fuzzy classifier,” International Journal of Advanced [78] H. Li, Y. Zhang, and H. Zheng, “Gear fault detection and
Manufacturing Technology, vol. 89, no. 9-12, pp. 3487–3494, diagnosis under speed-up condition based on order ceps-
2017. trum and radial basis function neural network,” Journal of
[63] B. Samanta and K. R. Al-Balushi, “Artificial neural network Mechanical Science and Technology, vol. 23, no. 10,
based fault diagnostics of rolling element bearings using pp. 2780–2789, 2009.
time-domain features,” Mechanical Systems and Signal [79] C. Peeters, P. Guillaume, and J. Helsen, “A comparison of
Processing, vol. 17, no. 2, pp. 317–328, 2003. cepstral editing methods as signal pre-processing techniques
[64] Z. Chen, X. Chen, C. Li, R.-V. Sanchez, and H. Qin, “Vi- for vibration-based bearing fault detection,” Mechanical
bration-based gearbox fault diagnosis using deep neural Systems and Signal Processing, vol. 91, pp. 354–381, 2017.
networks,” Journal of Vibroengineering, vol. 19, no. 4, [80] R. B. Randall, Frequency Analysis, Bruel and Kjaer,
pp. 2475–2496, 2017. Copenhagen, Denmark, 1987.
[65] L. Jiang, Y. Liu, X. Li, and A. Chen, “Gear fault diagnosis [81] D. Abboud, J. Antoni, S. Sieg-Zieba, and M. Eltabach,
“Envelope analysis of rotating machine vibrations in variable
based on svm and multi-sensor information fusion,” Journal
speed conditions: a comprehensive treatment,” Mechanical
of Central South University, vol. 41, no. 6, pp. 2184–2188,
Systems and Signal Processing, vol. 84, pp. 200–226, 2017.
2010.
[82] D. Ho and R. B. Randall, “Optimisation of bearing diagnostic
[66] N. Aiswarya, S. Suja Priyadharsini, and K. Moni, “An effi-
techniques using simulated and actual bearing fault signals,”
cient approach for the diagnosis of faults in turbo pump of
Mechanical Systems and Signal Processing, vol. 14, no. 5,
liquid rocket engine by employing fft and time-domain
pp. 763–788, 2000.
features,” Australian Journal of Mechanical Engineering,
[83] R. Rubini and U. Meneghetti, “Application of the envelope
vol. 16, no. 3, pp. 163–172, 2018.
and wavelet transform analyses for the diagnosis of incipient
[67] J. Runesson, “Vibration analysis for condition monitoring of
faults in ball bearings,” Mechanical Systems and Signal
mechanical presses,” Master’s Thesis, Lund University, Lund, Processing, vol. 15, no. 2, pp. 287–302, 2001.
Sweden, 2019. [84] A. Widodo, E. Y. Kim, J.-D. Son et al., “Fault diagnosis of low
[68] P. H. Westfall, “Kurtosis as p,” The American Statistician, speed bearing based on relevance vector machine and
vol. 68, no. 3, pp. 191–195, 2014. support vector machine,” Expert Systems with Applications,
[69] J. Hebda-Sobkowicz, R. Zimroz, and A. Wyłomańska, “Se- vol. 36, no. 3, pp. 7252–7261, 2009.
lection of the informative frequency band in a bearing fault [85] V. C. M. N. Leite, J. G. B. Da Silva, G. F. C. Veloso et al.,
diagnosis in the presence of non-Gaussian noise-comparison “Detection of localized bearing faults in induction machines
of recently developed methods,” Applied Sciences, vol. 10, by spectral kurtosis and envelope analysis of stator current,”
no. 8, p. 2657, 2020. IEEE Transactions on Industrial Electronics, vol. 62, no. 3,
[70] S. Fu, K. Liu, Y. Xu, and Y. Liu, “Rolling bearing diagnosing pp. 1855–1865, 2014.
method based on time domain analysis and adaptive fuzzy- [86] R. B. Randall, J. Antoni, and S. Chobsaard, “The relationship
means clustering,” Shock and Vibration, vol. 2016, 2016. between spectral correlation and envelope analysis in the
[71] K. H. Zhu, X. G. Song, and D. X. Xue, “Roller bearing fault diagnostics of bearing faults and other cyclostationary ma-
diagnosis based on imf kurtosis and svm,” Advanced Ma- chine signals,” Mechanical Systems and Signal Processing,
terials Research, vol. 694-697, pp. 1160–1166, 2013. vol. 15, no. 5, pp. 945–962, 2001.
[72] H. K. Patel, D. Shah, and A. Raghuwanshi, “Real time [87] J. Trout, “Vibration analysis explained,” 2020, https://www.
machine health monitoring and vibrational analysis using fft reliableplant.com/vibration-analysis-31569[Online]. Avail-
approach,” International Journal of Engineering and Ad- able:.
vanced Technology, vol. 8, no. 5, pp. 1833–1836, 2019. [88] P. J. Loughlin and G. D. Bernard, “Cohen-posch (positive)
[73] T. Ameid, A. Menacer, H. Talhaoui, and I. Harzelli, “Broken time-frequency distributions and their application to ma-
rotor bar fault diagnosis using fast fourier transform applied chine vibration analysis,” Mechanical Systems and Signal
to field-oriented control induction machine: simulation and Processing, vol. 11, no. 4, pp. 561–576, 1997.
experimental study,” International Journal of Advanced [89] L. Ciabattoni, F. Ferracuti, A. Freddi, and A. Monteriu,
Manufacturing Technology, vol. 92, no. 1–4, pp. 917–928, “Statistical spectral analysis for fault diagnosis of rotating
2017. machines,” IEEE Transactions on Industrial Electronics,
[74] Y. Liu, Z. Jiang, H. Haizhou, and J. Xiang, “Asymmetric vol. 65, no. 5, pp. 4301–4310, 2017.
penalty sparse model based cepstrum analysis for bearing [90] J. Cui, Y.-R. Wang, and Q. Liu, “The technique of power
fault detections,” Applied Acoustics, vol. 165, p. 107288, 2020. electronic circuit fault diagnosis based on higher-order
Shock and Vibration 23

spectrum analysis and support vector machines,” In Pro- Mechanical Systems and Signal Processing, vol. 11, no. 5,
ceedings of the Chinese Society of Electrical Engineering, pp. 673–692, 1997.
vol. 27, no. 10, pp. 62–66, 2007. [106] Y.-S. Han and C.-W. Lee, “Directional Wigner distribution
[91] T. Guishuang, S. Wang, and C. Zhang, “A method for rolling for order analysis in rotating/reciprocating machines,”
bearing fault diagnosis based on the power spectrum analysis Mechanical Systems and Signal Processing, vol. 13, no. 5,
and support vector machine,” in Proceedings of the IEEE 10th pp. 723–737, 1999.
International Conference on Industrial Informatics, [107] N. Baydar and A. Ball, “A comparative study of acoustic and
pp. 546–549, Beijing, China, July 2012. vibration signals in detection of gear failures using wigner-
[92] Z. Feng, M. Liang, and F. Chu, “Recent advances in time- ville distribution,” Mechanical Systems and Signal Processing,
frequency analysis methods for machinery fault diagnosis: a vol. 15, no. 6, pp. 1091–1107, 2001.
review with application examples,” Mechanical Systems and [108] N. E. Huang, Z. Shen, S. R. Long et al., “The empirical mode
Signal Processing, vol. 38, no. 1, pp. 165–205, 2013. decomposition and the hilbert spectrum for nonlinear and
[93] Y. Wei, Y. Li, M. Xu, and W. Huang, “A review of early fault non-stationary time series analysis,” Proceedings of the Royal
diagnosis approaches and their applications in rotating Society of London. Series A: Mathematical, Physical and
machinery,” Entropy, vol. 21, no. 4, p. 409, 2019. Engineering Sciences, vol. 454, no. 1971, pp. 903–995, 1998.
[94] G. Dalpiaz and A. Rivola, “Condition monitoring and di- [109] S. Huang, X. Wang, C. Li, and C. Kang, “Data decomposition
agnostics in automatic machines: comparison of vibration method combining permutation entropy and spectral sub-
analysis techniques,” Mechanical Systems and Signal Pro- stitution with ensemble empirical mode decomposition,”
cessing, vol. 11, no. 1, pp. 53–73, 1997. Measurement, vol. 139, pp. 438–453, 2019.
[95] B. Bao Liu, S.-F. Ling, and Q. Qingfeng Meng, “Machinery [110] M. E. Torres, M. A. Colominas, G. Schlotthauer, and
diagnosis based on wavelet packets,” Journal of Vibration P. Flandrin, “A complete ensemble empirical mode de-
and Control, vol. 3, no. 1, pp. 5–17, 1997. composition with adaptive noise,” in Proceedings of the IEEE
[96] J. Zou and J. Chen, ““A comparative study on time- International Conference on Acoustics, Speech and Signal
–frequency feature of cracked rotor by wigner–ville distri- Processing (ICASSP), pp. 4144–4147, Prague, Czech Republic,
bution and wavelet transform,” Journal of Sound and May 2011.
Vibration, vol. 276, no. 1-2, pp. 1–11, 2004. [111] Z. K. Peng, P. W. Tse, and F. L. Chu, “A comparison study of
[97] A. Giantomassi, F. Ferracuti, S. Iarlori, G. Ippoliti, and improved Hilbert-Huang transform and wavelet transform:
S. Longhi, “Electric motor fault detection and diagnosis by application to fault diagnosis for rolling bearing,” Me-
kernel density estimation and kullback-leibler divergence chanical Systems and Signal Processing, vol. 19, no. 5,
based on stator current measurements,” IEEE Transactions pp. 974–988, 2005.
on Industrial Electronics, vol. 62, no. 3, pp. 1770–1780, 2015. [112] T. Wu, Y. Chung, and C. Liu, ““Looseness diagnosis of
[98] F. Al-Badour, M. Sunar, and L. Cheded, “Vibration analysis rotating machinery via vibration analysis through hil-
of rotating machinery using time-frequency analysis and bert–huang transform approach,” Journal of Vibration and
wavelet techniques,” Mechanical Systems and Signal Pro- Acoustics, vol. 132, no. 3, pp. 1–9, 2010.
cessing, vol. 25, no. 6, pp. 2083–2101, 2011. [113] S. Osman and W. Wang, “A normalized hilbert-huang
[99] J. Rangel-Magdaleno, H. Peregrina-Barreto, J. Ramirez- transform technique for bearing fault detection,” Journal of
Cortes, R. Morales-Caporal, and I. Cruz-Vega, “Vibration Vibration and Control, vol. 22, no. 11, pp. 2771–2787, 2016.
analysis of partially damaged rotor bar in induction motor [114] W. Chen, J. Li, Q. Wang, and K. Han, “Fault feature ex-
under different load condition using dwt,” Shock and Vi- traction and diagnosis of rolling bearings based on wavelet
bration, vol. 2016, Article ID 3530464, 11 pages, 2016. thresholding denoising with ceemdan energy entropy and
[100] G. G. Yen and K.-C. Lin, “Wavelet packet feature extraction pso-lssvm,” Measurement, vol. 172, Article ID 108901, 2020.
for vibration monitoring,” IEEE Transactions on Industrial [115] Z. Y. Xuan and M. Ge, “Application of the hilbert - huang
Electronics, vol. 47, no. 3, pp. 650–667, 2000. transform for machine fault diagnostics,” Applied Mechanics
[101] J. Lin and L. Qu, “Feature extraction based on morlet wavelet and Materials, vol. 182-183, pp. 1484–1488, 2012.
and its application for mechanical fault diagnosis,” Journal of [116] M. Rezaee and A. Taraghi Osguei, “Improving empirical
Sound and Vibration, vol. 234, no. 1, pp. 135–148, 2000. mode decomposition for vibration signal analysis,” Pro-
[102] X. Fan and M. J. Zuo, “Gearbox fault detection using hilbert ceedings of the Institution of Mechanical Engineers - Part C:
and wavelet packet transform,” Mechanical Systems and Journal of Mechanical Engineering Science, vol. 231, no. 12,
Signal Processing, vol. 20, no. 4, pp. 966–982, 2006. pp. 2223–2234, 2017.
[103] W. Fan, Q. Zhou, J. Li, and Z. Zhu, “A wavelet-based sta- [117] D. Gabor, “Theory of communication. part 1: the analysis of
tistical approach for monitoring and diagnosis of compound information,” Journal of the Institution of Engineers: Elec-
faults with application to rolling bearings,” IEEE Transac- tronics, vol. 93, no. 26, pp. 429–441, 1946.
tions on Automation Science and Engineering, vol. 15, no. 4, [118] M. Safizadeh, A. Lakis, and M. Thomas, “Using short-time
pp. 1563–1572, 2017. fourier transform in machinery diagnosis,” in Proceedings of
[104] V. Climente-Alarcon, J. A. Antonino-Daviu, M. Riera- the 4th WSEAS International Conference on Electronic, Signal
Guasp, R. Puche-Panadero, and L. Escobar, “Application of Processing and Control, pp. 1–7, Rio de Janeiro, Brazil, April
the Wigner-Ville distribution for the detection of rotor 2005.
asymmetries and eccentricity through high-order har- [119] J. Burriel-Valencia, R. Puche-Panadero, J. Martinez-Roman,
monics,” Electric Power Systems Research, vol. 91, pp. 28–36, A. Sapena-Bano, and M. Pineda-Sanchez, “Short-frequency
2012. fourier transform for fault diagnosis of induction machines
[105] W. J. Staszewski, K. Worden, and G. R. Tomlinson, “Time- working in transient regime,” IEEE Transactions on In-
frequency analysis in gearbox fault detection using the strumentation and Measurement, vol. 66, no. 3, pp. 432–440,
wigner-ville distribution and pattern recognition,” 2017.
24 Shock and Vibration

[120] Z. Wu, Y. Li, H. Liu, and L. Liu, “Application of short-time dimension,” Mechanical Systems and Signal Processing,
fourier transform in the vibration analysis of the large hy- vol. 21, no. 5, pp. 2012–2024, 2007.
droelectric machine,” Journal of Hydroelectric Engineering, [134] A. Moosavian, H. Ahmadi, and A. Tabatabaeefar, “Fault
vol. 24, no. 6, pp. 115–120, 2005. diagnosis of main engine journal bearing based on vibration
[121] T. P. Banerjee, S. Das, J. Roychoudhury, and A. Abraham, analysis using Fisher linear discriminant, k-nearest neighbor
“Implementation of a new hybrid methodology for fault and support vector machine,” Journal of Vibroengineering,
signal classification using short -time fourier transform and vol. 14, no. 2, pp. 894–906, 2012.
support vector machines,” in Soft Computing Models in [135] Y. Zhu, Q. Yan, and J. Lu, “Fault diagnosis method for disc
Industrial and Environmental, E. Corchado, P. Novais, slitting machine based on wavelet packet transform and
C. Analide, and J. Sedano, Eds., Springer, Berlin, Germany, support vector machine,” International Journal of Computer
pp. 219–225, 2010. Integrated Manufacturing, vol. 33, no. 10-11, pp. 1118–1128,
[122] H. Gao, L. Liang, X. Chen, and G. Xu, “Feature extraction 2020.
and recognition for rolling element bearing fault utilizing [136] J. Runge and R. Zmeureanu, “Forecasting energy use in
short-time fourier transform and non-negative matrix fac- buildings using artificial neural networks: a review,” Energies,
torization,” Chinese Journal of Mechanical Engineering, vol. 12, no. 17, p. 3254, 2019.
vol. 28, no. 1, pp. 96–105, 2015. [137] M. R. Castelino, H. S. Kumar, P. P. Srinivasa, and G. S. Vijay,
[123] K. Mollazade, H. Ahmadi, M. Omid, and R. Alimardani, “An “Artificial neural network-based vibration signal analysis of
intelligent combined method based on power spectral rotary machines-case studies,” in Proceedings of the Inter-
density, decision trees and fuzzy logic for hydraulic pumps national Conference on Emerging Trends in Engineering,
fault diagnosis,” International Journal of Intelligent Systems pp. 211–218, Karnataka, India, July 2014.
and Technologies, vol. 3, no. 4, pp. 251–263, 2008. [138] G. Zurita, V. Sanchez, and D. Cabrera, “A review of vibration
[124] J. Cusid, L. Romeral, J. A. Ortega, J. A. Rosero, and A. Garcia machine diagnostics by using artificial intelligence methods,”
Espinosa, “Fault detection in induction machines using Revista Investigaci’on and Desarrollo, vol. 1, no. 16,
power spectral density in wavelet decomposition,” IEEE pp. 102–114, 2016.
Transactions on Industrial Electronics, vol. 55, no. 2, [139] M. A. Korany, H. Mahgoub, O. T. Fahmy, and H. M. Maher,
pp. 633–643, 2008. “Application of artificial neural networks for response sur-
[125] K. Heidarbeigi, H. Ahmadi, and M. Omid, “Fault diagnosis face modelling in hplc method development,” Journal of
of massey ferguson gearbox using power spectral density,” in Advanced Research, vol. 3, no. 1, pp. 53–63, 2012.
Proceedings of the 18th International Conference on Electrical [140] H. M. Ertunc, H. Ocak, and C. Aliustaoglu, “Ann-and anfis-
Machines, pp. 1–4, Vilamoura, Portugal, September 2008. based multi-staged decision algorithm for the detection and
[126] M. Heydarzadeh, N. Madani, and M. Nourani, “Gearbox diagnosis of bearing faults,” Neural Computing & Applica-
fault diagnosis using power spectral analysis,” in Proceedings tions, vol. 22, no. 1, pp. 435–446, 2013.
of the IEEE International Workshop on Signal Processing [141] Y. Chen, X. Qin, J. Wang, C. Yu, and W. Gao, “FedHealth: a
Systems, pp. 242–247, Dallas, TX, USA, October 2016. federated transfer learning framework for wearable health-
[127] D. Dou and S. Zhou, “Comparison of four direct classifi- care,” IEEE Intelligent Systems, vol. 35, no. 4, pp. 83–93, 2020.
cation methods for intelligent fault diagnosis of rotating [142] D.-T. Hoang and H.-J. Kang, “Rolling element bearing fault
machinery,” Applied Soft Computing, vol. 46, pp. 459–468, diagnosis using convolutional neural network and vibration
2016. image,” Cognitive Systems Research, vol. 53, pp. 42–50, 2019.
[128] A. A. Pinheiro, I. M. Brandao, and C. Da Costa, “Vibration [143] W. Qian, S. Li, J. Wang, Y. Xin, and H. Ma, “A new deep
analysis in turbomachines using machine learning tech- transfer learning network for fault diagnosis of rotating
niques,” European Journal of Engineering Research and machine under variable working conditions,” in Proceedings
Science, vol. 4, no. 2, pp. 12–16, 2019. of the 2018 Prognostics and System Health Management
[129] O. AlShorman, M. Irfan, N. Saad et al., “A review of artificial Conference, pp. 1010–1016, Chongqing, China, October 2018.
intelligence methods for condition monitoring and fault [144] F. Yang, W. Zhang, L. Tao, and J. Ma, “Transfer learning
diagnosis of rolling element bearings for induction motor,” strategies for deep learning-based phm algorithms,” Applied
Shock and Vibration, vol. 2020, Article ID 8843759, 20 pages, Sciences, vol. 10, no. 7, p. 2361, 2020.
2020. [145] B. Samanta, K. R. Al-Balushi, and S. A. Al-Araimi, “Artificial
[130] S. Poyhonen, M. Negrea, A. Arkkio, H. Hyotyniemi, and neural networks and genetic algorithm for bearing fault
H. Koivo, “Support vector classification for fault diagnostics detection,” Soft Computing, vol. 10, no. 3, pp. 264–271, 2006.
of an electrical machine,” in Proceedings of the 6th Inter- [146] Y. Yu and C. Junsheng, “A roller bearing fault diagnosis
national Conference on Signal Processing, pp. 1719–1722, method based on emd energy entropy and ann,” Journal of
Beijing, China, October 2002. Sound and Vibration, vol. 294, no. 1-2, pp. 269–277, 2006.
[131] A. Tabrizi, L. Garibaldi, A. Fasana, and S. Marchesiello, [147] F. Bi and Y. Liu, “Fault diagnosis of valve clearance in diesel
“Early damage detection of roller bearings using wavelet engine based on bp neural network and support vector
packet decomposition, ensemble empirical mode decom- machine,” Transactions of Tianjin University, vol. 22, no. 6,
position and support vector machine,” Meccanica, vol. 50, pp. 536–543, 2016.
no. 3, pp. 865–874, 2015. [148] M. M. M. Islam and J.-M. Kim, “Motor bearing fault di-
[132] C. Wu, T. Chen, R. Jiang, L. Ning, and Z. Jiang, “A novel agnosis using deep convolutional neural networks with 2d
approach to wavelet selection and tree kernel construction analysis of vibration signal,” Advances in Artificial Intelli-
for diagnosis of rolling element bearing fault,” Journal of gence, Springer, in Proceedings of the Canadian Conference
Intelligent Manufacturing, vol. 28, no. 8, pp. 1847–1858, 2017. on Artificial Intelligence, pp. 144–155, May 2018.
[133] J. Yang, Y. Zhang, and Y. Zhu, “Intelligent fault diagnosis of [149] S. Xing, Y. Lei, S. Wang, and F. Jia, “Distribution-invariant
rolling element bearing based on svms and fractal deep belief network for intelligent fault diagnosis of
Shock and Vibration 25

machines under new working conditions,” IEEE Transac- [166] T. Han, B. S. Yang, and Z. J. Yin, “Feature-based fault di-
tions on Industrial Electronics, vol. 68, no. 3, pp. 2617–2625, agnosis system of induction motors using vibration signal,”
2020. Journal of Quality in Maintenance Engineering, vol. 13, no. 2,
[150] D. Kolar, D. Lisjak, M. Paja˛k, and D. Pavković, “Fault di- pp. 163–175, 2007.
agnosis of rotary machines using deep convolutional neural [167] A. Hajnayeb, A. Ghasemloonia, S. E. Khadem, and
network with wide three axis vibration signal input,” Sensors, M. H. Moradi, “Application and comparison of an ann-based
vol. 20, no. 14, p. 4017, 2020. feature selection method and the genetic algorithm in
[151] M. He and D. He, “A new hybrid deep signal processing gearbox fault diagnosis,” Expert Systems with Applications,
approach for bearing fault diagnosis using vibration signals,” vol. 38, no. 8, pp. 10 205–10 209, 2011.
Neurocomputing, vol. 396, pp. 542–555, 2020. [168] F. Chen, B. Tang, and R. Chen, “A novel fault diagnosis
[152] L. Xueyi, L. Jialin, Q. Yongzhi, and H. David, “Semi-su- model for gearbox based on wavelet support vector machine
pervised gear fault diagnosis using raw vibration signal based with immune genetic algorithm,” Measurement, vol. 46,
on deep learning,” Chinese Journal of Aeronautics, vol. 33, no. 1, pp. 220–232, 2013.
no. 2, pp. 418–426, 2020. [169] C. Y. Xiao, B. Q. Shi, Z. J. Hao, and S. M. Zhu, “Gear incipient
[153] L. A. Zadeh, “Fuzzy logic,” Computer, vol. 21, no. 4, diagnosing based on eemd and genetic-support vector ma-
pp. 83–93, 1988. chine,” Applied Mechanics and Materials, vol. 397-400,
[154] J. Serrano-Guerrero, F. P. Romero, and J. A. Olivas, “Fuzzy pp. 2104–2110, 2013.
logic applied to opinion mining: a review,” knowledge-based [170] R. Semil and P. Jaiswal, “Bearing fault diagnosis using
systems,” Knowledge-Based Systems, vol. 222, Article ID support vector machine with genetic algorithms based op-
107018, 2021. timization and k fold cross-validation method,” Interna-
[155] P. Jayaswal, A. K. Wadhwani, and K. B. Mulchandani, tional Journal of Recent Technology and Engineering, vol. 8,
“Machine fault signature analysis,” International Journal of no. 2, pp. 3242–3250, 2019.
Rotating Machinery, vol. 2008, pp. 1–10, 2008. [171] A. Afia, C. Rahmoune, and D. Benazzouz, “Gear fault di-
[156] J. M. F. Salido and S. Murakami, “A comparison of two agnosis using autogram analysis,” Advances in Mechanical
learning mechanisms for the automatic design of fuzzy di- Engineering, vol. 10, no. 12, Article ID 1687814018812534,
agnosis systems for rotating machinery,” Applied Soft 2018.
Computing, vol. 4, no. 4, pp. 413–422, 2004. [172] J. Nowicki, J. Hebda-Sobkowicz, R. Zimroz, and
[157] I. Lasurt, A. F. Stronach, and J. Penman, “A fuzzy logic A. Wyłomańska, “Dependency measures for the diagnosis of
approach to the interpretation of higher order spectra ap- local faults in application to the heavy-tailed vibration sig-
plied to fault diagnosis in electrical machines,” in Proceedings nal,” Applied Acoustics, vol. 178, Article ID 107974, 2021.
of the 19th International Conference of the North American [173] T. Fei, C. Jiangfeng, Q. Qinglin, M. Zhang, H. Zhang, and
Fuzzy Information Processing Society, pp. 158–162, Atlanta, S. Fangyuan, “Digital twin-driven product design,
GA, , USA, July 2000. manufacturing and service with big data,” International
[158] J.-D. Wu and C.-C. Hsu, “Fault gear identification using Journal of Advanced Manufacturing Technology, vol. 94,
vibration signal with discrete wavelet transform technique no. 9-12, pp. 3563–3576, 2018.
and fuzzy-logic inference,” Expert Systems with Applications, [174] D. Shangguan, L. Chen, and J. Ding, “A digital twin-based
vol. 36, no. 2, pp. 3785–3794, 2009. approach for the fault diagnosis and health monitoring of a
[159] R. V. Mukane, N. M. Gurav, S. Y. Sondkar, and complex satellite system,” Symmetry, vol. 12, no. 8, p. 1307,
N. C. Fernandes, “Labview based implementation of fuzzy 2020.
logic for vibration analysis to identify machinery faults,” in [175] J. Wang, L. Ye, R. X. Gao, C. Li, and L. Zhang, “Digital twin
Proceedings of the International Conference on Computing, for rotating machinery fault diagnosis in smart
Communication, Control and Automation, pp. 1–5, Pune, manufacturing,” International Journal of Production Re-
India, August 2017. search, vol. 57, no. 12, pp. 3920–3934, 2019.
[160] R. F. M. Marcal, M. Negreiros, A. A. Susin, and
J. L. Kovaleski, “Detecting faults in rotating machines,” IEEE
Instrumentation and Measurement Magazine, vol. 3, no. 4,
pp. 24–26, 2000.
[161] J. Wang and H. Hu, “Vibration-based fault diagnosis of
pump using fuzzy technique,” Measurement, vol. 39, no. 2,
pp. 176–185, 2006.
[162] A. Umbrajkaar and A. Krishnamoorthy, “Vibration analysis
using wavelet transform and fuzzy logic for shaft mis-
alignment,” Journal of Vibroengineering, vol. 20, no. 8,
pp. 2855–2865, 2018.
[163] B. Alili, A. Hafaifa, and A. Iratni, “Faults detection based on
fuzzy concepts for vibrations monitoring in gas turbine,”
Diagnostyka, vol. 21, no. 4, pp. 67–77, 2020.
[164] J. Kusyk, C. S. Sahin, M. Umit Uyar, E. Urrea, and S. Gundry,
“Self-organization of nodes in mobile ad hoc networks using
evolutionary games and genetic algorithms,” Journal of
Advanced Research, vol. 2, no. 3, pp. 253–264, 2011.
[165] S. Katoch, S. S. Chauhan, and V. Kumar, “A review on
genetic algorithm: past, present, and future,” Multimedia
Tools and Applications, vol. 80, no. 5, pp. 8091–8126, 2021.