=Paper=
{{Paper
|id=Vol-3360/p11
|storemode=property
|title=Rotating Machinery Fault Diagnosis based on Artificial
Intelligence and Vibration Analysis
|pdfUrl=https://ceur-ws.org/Vol-3360/p11.pdf
|volume=Vol-3360
|authors=Salah Mahmood Al-Khafaji,Alaa Abdulhady Jaber
|dblpUrl=https://dblp.org/rec/conf/system/Al-KhafajiJ22
}}
==Rotating Machinery Fault Diagnosis based on Artificial
Intelligence and Vibration Analysis==
https://ceur-ws.org/Vol-3360/p11.pdf
Rotating Machinery Fault Diagnosis based on Artificial
Intelligence and Vibration Analysis
Salah Mahmood Al-Khafaji1 , Alaa Abdulhady Jaber1
1
University of Technology, Baghdad, Iraq
Abstract
As science and technology advance, rotary machines in modern industries become more advanced and complex, making
maintenance more difficult. Multiple failures can occur in machine components, requiring monitoring and periodic mainte-
nance. Multiple types of failure may occur, causing severe damage. As a result, modern technologies for early fault detection
have developed. Vibration analysis of rotating machinery is one of the most often used condition monitoring techniques.
Artificial intelligence (AI) approaches are widely used for selecting the features affected by faults. This paper discusses three
different types of artificial intelligence methods, which are Artificial Neural Networks, K-nearest neighbor, and Support Vec-
tor Machine, and presents a comprehensive review of recent studies on fault diagnosis for various rotary machine elements,
comprising the type of failure, feature extraction method, and classification technique performance.
Keywords
Artificial intelligence, Fault diagnosis, Artificial neural network, Support vector machine, K-nearest neighbour, Rotating
machinery
1. Introduction fault. Elaborating that the authenticity and uniqueness
of the extracted features are of major concern [6, 7, 8, 9].
The use of rotating machinery in modern industries is This paper discusses the applications of three types of
growing more sophisticated and complicated as science artificial intelligence methods in machinery faults diag-
and technology development, and maintenance becomes nosis, which are the Artificial Neural Networks (ANN),
more difficult as the sophistication of these machines in- K-nearest neighbor (KNN), and Support Vector Machine
creases. Due to the extensive working hours with severe (SVM), and presents a comprehensive review of previous
loads of this type of equipment, its components are sub- studies on fault diagnosis techniques for several com-
jected to multiple faults, requiring monitoring and peri- ponents of rotary machines, including the type of fail-
odic maintenance. Multiple types of failure may occur in ure, feature extraction method, and the performance of
various parts, such as bearings, gears, pulleys, shafts, etc. classification technique. The remaining of this paper is
These faults affect the machine’s performance and might organized as follows. Section 2 introduces a theoretical
result in major issues and financial losses. As a result, basis of AI techniques. Section 3 reviews the applications
modern technologies for early detection of faults, includ- of AI techniques in fault diagnosis of rotating machinery.
ing condition monitoring and fault diagnostics are being Finally, the conclusions are drawn in Section 5.
investigated and developed to ensure that the machines
operate effectively. The main goals of fault diagnostic
investigation are to determine the machine’s normal op- 2. Theoretical basis of Artificial
erating condition, identify the type of fault, and predict Intelligence techniques
the fault’s progression. One of the most common types
of condition monitoring is the vibration analysis of rotat- 2.1. Artificial neural networks (ANN)
ing machines. However, vibration signals are typically
non-stationary, non-linear, and intermixed with noise. The ability of biological neurons in the human brain has
consequently, recent research has forecasted the domi- inspired researchers to invent a computational structure,
nance of AI over technological innovations [1, 2, 3, 4, 5]. namely the Artificial Neural Network. Frank Rosenblatt,
As a result, the features from these signals are extracted, a psychologist, came up with the first artificial neural
and artificial intelligence (AI) techniques are employed to network in 1958. ANN is made up of a group of linked
identify the exact sensitivity of features for each type of neurons that are arranged in layers to form a network.
Usually, ANN consists of an input layer, an output layer,
SYSYEM 2022: 8th Scholar’s Yearly Symposium of Technology, Engi-
neering and Mathematics, Brunek, July 23, 2022
and at least one hidden layer. The number of neurons in
" me.20.12@grad.uotechnology.edu.iq (S. M. Al-Khafaji); the input and output layers is defined by the number of
20039@uotechnology.edu.iq (A. A. Jaber) input and output variables required to define the problem,
0000-0001-5709-195X (A. A. Jaber) as well as the nature of the problem, while the process
© 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License
Attribution 4.0 International (CC BY 4.0). of trial and error dictates the number of hidden layers
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073
CEUR Workshop Proceedings (CEUR-WS.org)
85
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
and neurons within each layer. Each neuron in a layer limited amount of training material, SVMs have lately
(excluding the input layer) sums the input value with garnered much attention. An SVM, for instance, will
the related weight to determine a single value threshold. classify a two-class dataset by locating a splitting plane
Then, the single value threshold is added with a bias to that separates the area containing the data. Each side
form a net value. Finally, a non-linear activation function of the hyper-plane will have its own class of points. A
is applied to the net value, creating an output value, as linear boundary in the input feature space can be the best
shown in Figure 1. For a supervised learning method, separation plane, while in other instances, a non-linear
input values are compared with the output values, and boundary might be utilized to separate the target classes
then a back propagation algorithm is utilized for training when a linear boundary would not be able to successfully
the ANN model by adjusting the weights between each separate them [21, 22], as shown in Figure 3. SVMs are
neuron in the multiple layers [10]. Many applications being used in various research fields, including biological
have employed ANN due to its great performance [11, 12], sequence analysis, facial identification, and mechanical
such as pattern recognition, fault prediction and classi- problem diagnostics [23]. Even though its performance
fication, speech recognition, handwritten and printed varies depending on the application, SVM is a robust,
text recognition, image processing, and cancer diagnosis effective, and simple tool for various applications such as
[13, 14, 15, 16]. speech recognition, texture categorization, face detection,
heart disease,and fault diagnosis [24, 25, 26].
2.2. K-nearest neighbour
K-NN method is one of the most basic and straightfor-
ward classification approaches. It is an instance-based
learning technique based on grouping components with
similar features; it determines the class category of a
test case based on its k closest neighbors. Evelyn Fix
and Joseph Hodges developed it in the early 1950s, and
Thomas Cover extended it subsequently[17]. Every train-
ing sample in KNN classification algorithms is repre-
sented as a two-dimensional space based on the value of
each of its characteristics. The testing sample is therefore
displayed in the same space as its K closest neighbors. Figure 1: ANN
The classes of each closest neighbor of K are counted,
and the class with the most votes is selected as the testing
sample’s classification. Typically, the distance between
the testing sample and each training sample is used to 3. Applications of AI techniques
estimate the K-nearest neighbors. The distance between
the testing sample and each training sample is typically in rotating machinery
used to estimate the K-nearest neighbors. As shown in
Figure 2, three factors influence KNN performance: K 3.1. ANN applications
value, Euclidean distance, and parameters’ normaliza- Mohammed et al. [27] developed a method for crack de-
tion. KNN is particularly effective for huge training data tection using Power Spectrum Density (PSD) in a motor-
sets, although it takes more time to compute than other shaft-generator system. The vibration signals were col-
methods. Along with its simplicity, it is generally used inlected from three piezoelectric accelerometers placed in
the fault diagnosis of rotating machinery[18, 19]. Like- different places; one attached to the motor bearing hous-
wise, it is used in medical prediction, data mining, face ing, the other attached to the generator bearing hous-
recognition, and financial modeling [20]. ing, and the last placed in the center of the shaft cover-
ing guard. The vibration signals were fed to the charge
2.3. Support Vector Machine amplifiers, which are connected to an analog to digital
converter involving a dSPACE-DS1102 DSP controller.
Support vector machines are a model of artificial intelli- Data acquisition instruments were included in the DSP
gence technology commonly used for data classification program for acquiring data from the model and are sup-
and regression. SVMs, which Vapnik introduced in the ported by Matlab software. The peak position component
middle of the 90s, are supervised learning techniques method (PPCM) was used to identify the highest peaks
based on statistical learning theory. Because of their and their positions from the PSD analysis and form a ma-
greater capacity to construct an accurate representation trix for the input data of ANN. The Figure 4 shows the vi-
of the connection between the input and output from a bration spectrum of the first accelerometer. A multilayer
86
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
Table 1
ANN designing parameters[29]
Number of input layer neurons 24
Number of hidden layer neurons 17
Number of output layer neurons 11
Number of hidden layers 1
Hidden layer activation function Sigmoid
Output layer activation function Linear
Training algorithm To be identified
Learning rate 0.05
MSE stopping criteria 10e–4
Minimum performance gradient 10e–5
Maximum number of epochs 50,000
chine (EDM) was used to create the faults in the bearing’s
elements. A 14-bit USB-6009 National Instrument data
acquisition is used to collect signals from three calibrated
Figure 2: K-NN
single-axis ADXL001 accelerometers. Labview, Matlab,
and C language software were used for data acquisition
to extract and analyze vibration data. After the DWT,
perceptron ANN for each accelerometer was employed to the statistical features were extracted and used for de-
classify four health cases: normal shaft and a shaft with signing ANN. Only the standard deviation (STD) was
40%, 50%, and 60% pre-crack. The Levenerg-Marquart chosen as an input to ANN as it was the most sensitive
learning algorithm was used as a training method. ANN feature for bearing faults. A multilayer perceptron neural
for the data of the first and the third accelerometer distin- network was utilized for fault detection. Table 1 shows
guished four conditions with 100% accuracy, and for the the parameters used for designing the neural network.
second accelerometer data, it achieved 99.7% accuracy in ANN was trained to differentiate between various types
fault classification. of robot-bearing faults and achieved a remarkable level
Ali et al. [28] proposed a feature extraction method of fault diagnosis with an accuracy of 100%.
based on Empirical Mode Decomposition (EMD) energy Luwei et al. [30]used frequency domain data fusion to
entropy and ANN to classify seven bearing conditions: a classify rotating machine faults with the help of ANN.
healthy condition, degraded roller, degraded outer race, From four sensors placed on each bearing, higher-order
degraded inner race, failure roller, degraded outer race, spectra components were extracted at different speeds
and failure inner race. The EMD approach is based on and various fault conditions. ANN was performed in
the simple assumption that each signal is made up of two stages. In the first stage, five types of one-against-
various simple intrinsic modes of oscillations, named all (OAA) ANN were trained using the Resilient Back-
intrinsic mode functions (IMFs). This adaptive decom- propagation learning algorithm to specify the presence
position approach is especially useful for non-linear and of five faults: bent shaft, loose bearings, shaft misalign-
non-stationary signal analysis. The data was collected ment, cracked shaft, and rubbing in the shaft. For the first
from high-sensitivity accelerometers attached to four and second ANN, the accuracy was 100% and very good
bearings. Ten time-domain features were extracted in classification accuracy for the remaining ANN. In the
addition to the EMD energy entropy to create a feature second stage, all-against-all (AAA) ANN was performed
vector for ANN input layer. The three zones of run-to- to determine the particular fault type using the same
failure vibration signals are shown in Figure 5 for the learning techniques as in the first stage. The receiver
outer race failure. A statistical criterion was applied to operating characteristics (ROC) curve was utilized for es-
decide the most effective IMFs for bearing diagnosis. A timating the accuracy of AAA network classification (see
back-propagation algorithm was employed for training Figure 7). The classifier’s performance is measured using
the ANN. The proposed method was able to classify bear- the area under the curve (AUC).The larger AUC value
ing states with an average accuracy of 93%. means its performance is better (perfect classification).
Jaber and Bicker [29]developed an intelligent bearing The approximate AUC values for the five fault conditions
fault diagnosis system using the discrete wavelet trans- mentioned were 1, 0.992, 0.999, 0.986, and 0.992, respec-
form (DWT). Inner and outer races bearing faults were tively, indicating excellent classification performance.
simulated on the elbow joint of the PUMA 560 robot, Jaber and Ali [31]developed a fault detection system
which is shown in Figure 6. An Electrical Discharge Ma- for a pulley-belt rotating test rig. Two ADXL335 vibra-
87
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
Figure 3: SVM A.Linear B.Non-linear
Figure 4: Vibration spectra for different cut depths at fre-
quency range of 800 to 950 Hz[27]
Figure 6: PUMA 560 robot[29]
microcontroller. The Arduino microcontroller was used
as a low-cost data acquisition device. For ANN train-
ing data, five time-domain features from each sensor
were extracted as input data: the mean value, RMS, skew-
ness, kurtosis, and standard deviation. Also, five different
faults were studied: unbalance, driving pulley fault, a side
cut-out in the belt, belt slippage, and misalignment in
pulleys. Labview software was employed for collecting
Figure 5: Bearing run-to-failure vibration signals[28] vibration signals and extracting their data features. The
extracted features are then uploaded to Matlab to design
a multilayer ANN. As a result, the designed ANN was
able to identify each fault perfectly. The performance
tion sensors placed on each bearing were employed to
plot is shown in Figure 8.
extract vibration signal features. These sensors are wired
Sharma et al. extracted [32]frequency-domain features
to an Arduino MEGA 2650 based on the ATmega 2560
from the vibration signals of a three-phase induction mo-
88
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
lized for fault classification. To train ANN, two different
techniques, scaled conjugate gradient (SCG) and Leven-
berg–Marquardt (LM), are used, and their performance is
compared. For greater network sizes, the SCG algorithm
outperforms the LM algorithm, and the proposed ANN
could identify each fault condition with great accuracy.
Rao and Reddy [33]used wavelet transforms to iden-
tify a method for detecting irregularities such as open
cracks or grooves on a rotating stepped shaft with sev-
eral discs. The vibration signals were extracted from the
displacement sensor and transformed into discrete and
continuous wavelet transforms (DWT and CWT) at a
specific rotor speed. The detailed process and techniques
of analysis are shown in the Figure 9. The difference in
wavelet coefficients of rotors with and without grooves
is analyzed to identify the damage or groove locations.
A reduction in the shaft’s diameter was used to model
the cracks and grooves classified as radial cracks. A nu-
merical analysis of various cases was used to find each
case’s first five natural frequencies. In this research, a
Figure 7: ROC plot[30] three-layer feed-forward ANN was utilized. The DWT
coefficient of various crack locations and depths was
used to train ANN. The Levenberg-Marquardt algorithm
is employed to train ANN. Consequently, the overall pre-
diction accuracy of the developed ANN is 99.53%. It was
able to detect cracks as small as 1% of their diameter.
Espinoza and Sinha [34]developed a smart vibration-
based machine learning model that uses ANN. The model
is then tested blindly to detect the rig’s healthy and faulty
conditions when running at different speeds. The test rig
has four accelerometers placed on four bearings. Four
scalar features were extracted from vibration data sam-
ples and arranged in three scenarios. In the first scenario,
each sample is obtained randomly from an accelerometer
from only one bearing. The second scenario considers the
measurement from only one bearing with a fixed location.
The third scenario considers a simultaneous vibration
Figure 8: ANN perfomance plot[31] signal collection from all four bearings. The scalar fea-
tures are: root mean square (RMS),kurtosis, variance, and
skewness. Spectral analysis is also used to examine the
tor. The vibration signal was collected from a single-axis dynamic behavior of the rig under various conditions.
accelerometer. The extracted features are used to classify ANN of multilayer perceptron (MLA) based on the back
three bearing conditions: healthy, inner race defect, and propagation was utilized to classify acquired data. De-
outer race defect. The electric discharge machine (EDM) pending on the results obtained in the particular scenario,
was used to create 2 mm holes in the inner and outer either Bayesian regularization functions or scaled conju-
races. For each condition, five features are extracted gate gradient are used to train the network. The results
from the frequency domain: mean frequency, median found that the first scenario observed has an almost 25%
frequency, lower band power, upper band power, and chance that the health conditions will be misdiagnosed
band power ratio. The acquired features were normal- as faulty. The second scenario performs relatively bet-
ized to increase classification accuracy and distinguish ter than the first because consistent data from a specific
any bias. As a result, ten sets of five normalized charac- bearing point exposes some features of machine behavior.
teristics were collected for each bearing condition. Six The third scenario indicates the best performance, with
sets were used for training ANN, while the remaining 100% accuracy. The model was trained at a rotational
four were used for testing. A single-layer ANN was uti- speed of 1800 rpm and was tested blindly using test rig
data at 2400 rpm without training. The results showed
89
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
Figure 9: Analysis block diagram[33]
Figure 10: Overall performances, training and testing at 1800 RPM[34]
100% accuracy in machine condition diagnosis, as shown and used for training the KNN to detect the remaining
in Figure 10. faults. The results showed an accuracy of 100% for the
first step, and the overall accuracy exceeded 90% for the
3.2. K-NN applications second step.
Pandya et al.[36] presented a fault classification
He et al. [35] proposed a two-step plastic-bearing fault di- method based on acoustic emission for bearing health
agnostics method. This approach investigated cage fault, monitoring. Five conditions were examined: healthy
rolling element fault, and surface contact faults in the bearing, outer race, inner race, ball, and combined defect.
inner and outer race. Two accelerometers attached to the The data was collected using an acoustic emission sen-
surface of the bearing house were used to collect vibra- sor installed on the housing of the test bearing and an
tion data with NI PCI-4472B data acquisition. Frequency OROS 3 SERIES acoustic analyzer, and it was processed
domain features were extracted from vibration signals by using NVGATE software. The Hilbert–Huang Transform
the envelope analysis technique and fed to the statistical (HHT) was employed in this method. Band-pass filtering
classification method. The first step only used the statis- with empirical mode decomposition was performed on
tical classification method to classify the outer race fault, the collected signals, and IMFs were optimized. IMFs
as shown in Figure 11. As a second step, entropy and were used to extract nine time-frequency domain fea-
distance (time-domain features) were extracted by EMD tures: peak, mean, RMS, kurtosis, crest factor, impulse
90
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
Figure 11: Frequency domain features[35]
Figure 12: Theoretical and predict classification[37]
factor, shape factor, ring down count, and energy. Multi-
ple supervised machine learning techniques were used features (RMS, standard deviation, skewness, crest factor,
for the fault classification using WEKA software, and the kurtosis, peak level, K factor, mean value, variance, and
weighted KNN was the most efficient technique. Then, a median). PCA was used to reduce features. The waterfall
modified KNN method based on an asymmetric proximity fusion model was used as the fusion approach to merge
function (APF) was developed to improve classification sensor data effectively. The results of the experiments
accuracy even further. The results demonstrate that the indicate that the proposed method improves fault identi-
proposed APF-KNN algorithm with optimized features fication and diagnostic accuracy, as shown in the Table 2.
surpasses the KNN method by 96.6667%.
Wang et al.[37] developed a bearing fault diagnosis Tain et al.[39] proposed a method for detecting and
method based on combining the Kernel Principal Com- monitoring bearing faults in an electric motor. Four faults
ponent Analysis (KPCA) with the KNN algorithm. KPCA were investigated: the outer race, inner race, rolling el-
is a method for applying the kernel method to generalize ement, and cage fault. Vibration signals were extracted
linear Principal Component Analysis (PCA) to non-linear from two accelerometers installed on the bearing hous-
cases. An electro-discharge machine was used to artifi- ing and decomposed in the frequency domain to obtain a
cially seed a single point fault in the bearing inner race, set of sub-signals. Different fault features were extracted
outer race, and rolling element. The accelerometers were based on cross-correlation and spectral kurtosis (SK). The
used to sample vibration signals positioned at the drive principal component analysis (PCA) technique was uti-
end. The extracted features are the clearance factor, the lized to reduce the redundancy of fault features. KNN is
greatest peak, impulse factor, kurtosis, mean absolute used to combine the features with a health index, which is
difference, peak factor, peak-to-peak value, rad ampli- further analyzed for defect detection. Following various
tude, RMS value, skewness, variance, waveform index, experiments, a gearbox was mounted to create signals
and kurtosis value. KPCA was applied to the extracted that mask bearing signals and cause false-negative de-
signal and used to define KNN. As a result, this method tection. As a result, this method was effective in fault
could classify bearing faults with an accuracy of 96.67%, classification and isolating bearing signals from gear sig-
as shown in Figure 12. nals to detect previously unknown faults.
Safizadeh and Latifi[38] developed a unique bearing Gohari and Eydi [40] used KNN to identify shaft un-
fault diagnostic approach based on KNN and a combina- balance in multi-disc rotors and compared the results
tion of an accelerometer and a load cell. This method was with the Decision Tree (DT) Algorithm. Various masses
used to classify three bearing conditions: a healthy, outer were mounted in three radiuses to create an unbalanced
race defect, and ball defect. Spark erosion techniques condition. Two ADXL335 accelerometers attached to
were employed to create the defects in bearing elements. the shaft were employed to collect vibration data from
A Piezoelectric IMI 608A111 accelerometer was used to the test setup. A data acquisition device (ADVANTECH
collect vibration data, and a SEWHACNM SM601 load 4711A) was used for data collection. A program was de-
cell was used with a DACELL DNAM100 amplifier to mea- veloped in Labview Software for signal processing and
sure the load. The output of each sensor was connected recording, and filtering of unwanted noises. Eight sta-
to the NI-USB-9233 DAQ. Two frequency domain features tistical features were extracted from time and frequency
were extracted (amplitude in ball pass frequency of the domain signals: the peak value, skewness, average, RMS
outer race and ball spin frequency) with ten time-domain error, absolute average, the peak of average ratio, crest
91
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
Table 2
Effectiveness of each technique[38]
Bearing condition Accelerometer detection Load cell detection Data fusion technique
Healthy Outer race fault Healthy Healthy
Outer race fault Outer race fault Outer race fault Outer race fault
Outer race fault Outer race fault Outer race fault Outer race fault
Outer race fault Outer race fault Outer race fault Outer race fault
Ball fault Ball fault Outer race fault Ball fault
Ball faul Ball fault Outer race fault Ball fault
Figure 13: KNN perfomance in unbalance locating [40] Figure 14: Classification accuracy of EKNN[41]
factor, standard deviation, kurtosis factor, and peak in shows the flowchart of (OWT-KNN). EDM was used to
the frequency domain. The extracted features were fed implant faults in the inner race and the ball. The vibration
to the KNN and the DT algorithms. The study revealed signals were collected from three accelerometers placed
that the KNN surpasses the Decision Tree in estimating in different positions and subjected to a multilayer or-
unbalance parameters and can classify the faults with an thogonal wavelet transform. Peak-to-peak values were
average accuracy of 86.6%. The Figure 13 shows the KNN found on each scale to create the feature vectors. Lately,
performance of unbalanced locating. the created feature vectors are utilized to generate the
Lu et al. [41] proposed a fault diagnosis method for classification model and train the KNN classifier. The
rotating machinery called Enhanced K-Nearest Neighbor classification results show that this approach can achieve
(EKNN), based on KNN and spare coding. This method a 100% fault classification.
investigated four bearing conditions: normal conditions,
inner race faults, outer race faults, and roller faults. An 3.3. SVM Applications
accelerometer attached to the bearing was utilized to col-
lect vibration signals. Fast Fourier Transform (FFT) was Jiang et al. [43] used SVM and multi-sensor informa-
applied to gather frequency domain features for creating tion fusion to develop a fault diagnosis approach for
the training dataset. Then, discriminative features were rotating machines. This approach investigated differ-
extracted by applying Spare Filtering (SP) to the training ent cases for machine elements: three conditions for the
dataset. Finally, the discriminative features were opti- gears (normal, missing, and chipped tooth), four con-
mized by the L-BFGS algorithm and transferred into the ditions for the bearing (normal, defect, inner race de-
feature vector. The feature vector was fed to the EKNN fect, and outer race defect), and three conditions for the
and the traditional KNN. As shown in Figure 14, the pro- shaft (normal, 3mm crack depth, and 5mm crack depth).
posed method could classify faults with 99% accuracy, IMI 608A11 accelerometers collected vibration signals
surpassing the traditional KNN. with the Dewetron data acquisition system. Twelve time-
Li et al. [42] presented a bearing fault detection method domain features were extracted for each condition, in-
based on the Orthogonal Wavelet Transform K-Nearest cluding mean, peak, amplitude square, RMS, root ampli-
Neighbor Algorithm (OWT-KNN). The OWT can decom- tude, standard deviation, skewness, kurtosis, waveform
pose the signals into the equivalent local detail signals at factor, pulse factor, and margin factor. The multi-sensor
each scale in terms of time and frequency. The Figure 15 information fusion model was utilized to establish a mul-
92
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
Figure 16: Accuracy results[46]
tion of bearing faults. This method studied the following
cases: healthy, inner and outer race degradation, and in-
ner and outer race failure. A run-to-failure test provided
the data for this investigation. Vibration data was col-
lected by four accelerometers attached to four bearings.
The RMS was calculated from the collected time-domain
signals; the Hilbert transformation was used to detect the
envelope of a time-domain signal. Then, the envelope sig-
nal is converted into the frequency domain using the fast
Fourier transformation. Fault-specific frequencies are
found in the frequency spectrum, and energy associated
with each frequency band is collected. Four frequency do-
Figure 15: OWTKNN Flowchart[42]
main features were extracted in addition to RMS, and all
these features were used to train the SVM classifier. The
obtained classification accuracy for healthy, inner and
outer race degradation and inner and outer race failure
tidimensional vector by extracting the same character is 99.3%, 86.2%, 97.7%, 87.8%, and 84.2%, respectively.
from different sensors. The SVM was applied to each Huo et al. [46] proposed a fault diagnosis method for
case for fault classification. According to the results, the rotating shafts based on Multi-Scale Entropy (MSE). This
highest accuracies achieved by the SVM classifier were method investigated two shaft conditions: healthy and 4
93.33%, 100%, and 99.67% of gears, rolling bearing, and mm cracked shaft. The vibration data for each condition
cracked shaft cases, respectively. were collected from the PT 500 machinery diagnostic
Tabrizi et al. [44] proposed a combined automatic system. WPD and EMD were employed to decompose
method for detecting tiny defects on roller bearings that the signals to obtain reconstruction vectors and IMFs
uses Wavelet Packet Decomposition (WPD) with Ensem- data sets. Afterward, Shannon entropy criteria were uti-
ble Empirical Mode Decomposition (EEMD). Tri-axial ac- lized to select the largest entropy in the decomposed
celerometers were used to collect vibration data at three vectors. The MSE method was then used to define fault
different shaft speeds and external loads. Afterward, the symptoms and create feature vectors. Finally, the feature
original signals were extracted from the noisy signals vectors were fed to the SVM for fault classifying. Experi-
using WPD, and EEMD was applied to decompose the mental results showed that WPD combination with MSE
vibration signals into IMFs. Then, a feature vector was achieved a classification accuracy of 97.3%, whereas EMD
created from the normalized IMFs energy and fed to the combination with MSE had a better classification rate of
SVM. With Daubechies DB10 Denoising, the proposed 98.5%. The Figure 16 shows the accuracy results using
method was able to detect the defects with 100% accuracy. EMD, MSE, and SVM.
Senanayaka et al. [45] presented a method based on Gu et al. [47] developed a fault diagnosis for rolling
the SVM algorithm for the early detection and classifica- bearings based on SVM and PCA. Four deep groove ball
93
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
Table 3
Classifier Results[49]
Method Best Model Accuracy Training time
10% Data Validation Cubic SVM 95.6% 15.947
15-Fold Cross Validation Quadratic SVM 94.7% 60.62
10-Fold Cross Validation Quadratic SVM 94.7% 19.401
5-Fold Cross Validation Quadratic SVM 93.8% 10.283
Figure 18: PCA using the pre-processing data[50]
Figure 17: Classification results[48]
can be derived by integrating the time-frequency infor-
bearings conditions were examined: normal condition, mation of all SSCs. Time-frequency entropy (TFE) was
roller flaking, inner ring flaking, and outer ring flaking. calculated from TFS to build CFBEE. Then a feature vec-
An IMI 601A1 accelerometer was used with a UUA300 tor was obtained from calculated CFBEEs and fed to the
data acquisition card for vibration signal monitoring. A SVM, and the SVM was able to classify rotor faults, as
wavelet packet was used to decompose the collected sig- shown in Figure 17.
nals and create the feature vectors. Then, the feature Parmar and Pandaya [49] developed a fault diagnosis
vectors were integrated by PCA and fed to the SVM clas- method for cylindrical bearings based on SVM. Four con-
sifier. Different kernel functions and SVM classification ditions were investigated: healthy bearing, defects in the
algorithms were used to compare and verify the influence inner race, outer race, and the rolling element (defect
of kernel functions and classification algorithms on the sizes are around 0.5 mm). A three-axis sensor was used
accuracy of the SVM classifier. Finally, the developed to collect vibration signals, and CoCo-80 Dynamic Signal
method achieved a performance of more than 97%. Pang Analyzer processed the collected signals. The collected
et al. [48] proposed a novel rotor fault diagnosis method signals were analyzed by wavelet packet decomposition
based on Characteristic Frequency Band Energy Entropy (with mother wavelet ’sym20), then mathematical param-
(CFBEE) and SVM to investigate three rotor faults: im- eters were extracted, such as mean value, mean square
balance, rubbing, and oil film instability failures. The value, skewness, kurtosis, standard deviation, crest factor,
shaft vibration signals were collected using two eddy- and energy. Then these parameters were normalized and
current sensors attached to the mounting blocks. The fed to two classification methods: ANN and SVM. The
time-frequency features were decomposed by improved results showed that ANN classification accuracy did not
singular spectrum decomposition (ISSD) into a series of exceed 90%, while the SVM (cubic model) was able to
singular spectrum components (SSCs). The Hilbert Trans- classify faults with 95.6% accuracy. Table 3 shows the
form (HT) demodulates the obtained SSCs to determine SVM classification results.
their instantaneous amplitude (IA) and instantaneous Lee et al. [50] developed a method to detect misalign-
frequency (IF). IF and IA reflect the time-frequency infor- ment in a rotating machine shaft based on the SVM algo-
mation of SSC, and the Time-Frequency Spectrum (TFS) rithm. A gyro vibration sensor was attached to the shaft’s
end between the rotor and the shaft to collect normal
94
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
and abnormal vibration data. The SVM was applied to [5] N. Brandizzi, V. Bianco, G. Castro, S. Russo, A. Wa-
the collected data without preprocessing, and its classi- jda, Automatic rgb inference based on facial emo-
fication accuracy was 49.71%. Then, the FFT technique tion recognition, in: CEUR Workshop Proceedings,
was used to extract the power spectrum from the time volume 3092, CEUR-WS, 2021, pp. 66–74.
domain data, and the PCA algorithm was used to reduce [6] F. Fallucchi, M. Gerardi, M. Petito, E. W. De Luca,
the dimensions (see Figure 18). The SVM algorithm was Blockchain framework in digital government for
applied to the processed data and could predict the nor- the certification of authenticity, timestamping and
mal and abnormal conditions with an average accuracy data property, in: Proceedings of the 54th Hawaii
of 98.8%. International Conference on System Sciences, 2021,
p. 2307.
[7] B. Nowak, R. Nowicki, M. Woźniak, C. Napoli,
4. Conclusion Multi-class nearest neighbour classifier for
incomplete data handling, in: Lecture Notes
Rotating machinery fault diagnosis plays an important
in Artificial Intelligence (Subseries of Lec-
role in saving maintenance costs, downtime, and safety
ture Notes in Computer Science), volume
risks. In this study, a variety of AI approaches for rotat-
9119, Springer Verlag, 2015, pp. 469–480.
ing machinery diagnostics are discussed. The theoretical
doi:10.1007/978-3-319-19324-3_42.
approach and fault diagnostic application of ANN, K-NN,
[8] S. Illari, S. Russo, R. Avanzato, C. Napoli, A cloud-
and SVM have been reviewed. This study advises per-
oriented architecture for the remote assessment
forming feature selection on the feature vector before em-
and follow-up of hospitalized patients, in: CEUR
ploying AI classification techniques to identify the most
Workshop Proceedings, volume 2694, CEUR-WS,
sensitive feature for each fault case. Further, this study
2020, pp. 29–35.
suggest using other AI classification techniques such as,
[9] A. Simonetta, A. Trenta, M. C. Paoletti, A. Vetrò,
Random Forest, Decision Tree, Naive Bayes, and Deep
Metrics for identifying bias in datasets, SYSTEM
learning. As AI techniques grow more sophisticated, it is
(2021).
expected that AI approaches will remain interesting and
[10] B. Paya, I. Esat, M. Badi, Artificial neural network
effective for detecting faults in rotating machinery.
based fault diagnostics of rotating machinery using
wavelet transforms as a preprocessor, Mechanical
References systems and signal processing 11 (1997) 751–765.
[11] G. Capizzi, G. Lo Sciuto, C. Napoli, M. Woźniak,
[1] R. Giuliano, The next generation network in 2030: G. Susi, A spiking neural network-based long-term
Applications, services, and enabling technologies, prediction system for biogas production, Neural
in: 2021 8th International Conference on Electrical Networks 129 (2020) 271 – 279.
Engineering, Computer Science and Informatics [12] G. Capizzi, G. Lo Sciuto, C. Napoli, E. Tramontana,
(EECSI), IEEE, 2021, pp. 294–298. An advanced neural network based solution to en-
[2] F. Bonanno, G. Capizzi, G. Sciuto, C. Napoli, G. Pap- force dispatch continuity in smart grids, Applied
palardo, E. Tramontana, A cascade neural net- Soft Computing Journal 62 (2018) 768 – 775.
work architecture investigating surface plasmon [13] O. I. Abiodun, A. Jantan, A. E. Omolara, K. V. Dada,
polaritons propagation for thin metals in openmp, N. A. Mohamed, H. Arshad, State-of-the-art in
Lecture Notes in Computer Science (including sub- artificial neural network applications: A survey,
series Lecture Notes in Artificial Intelligence and Heliyon 4 (2018) e00938.
Lecture Notes in Bioinformatics) 8467 LNAI (2014) [14] N. Ganesan, K. Venkatesh, M. Rama, A. M. Palani,
22–33. doi:10.1007/978-3-319-07173-2_3. Application of neural networks in diagnosing can-
[3] N. Dat, V. Ponzi, S. Russo, F. Vincelli, Supporting cer disease using demographic data, International
impaired people with a following robotic assistant Journal of Computer Applications 1 (2010) 76–85.
by means of end-to-end visual target navigation [15] G. C. Cardarilli, L. Di Nunzio, R. Fazzolari, D. Gi-
and reinforcement learning approaches, in: CEUR ardino, A. Nannarelli, M. Re, S. Spanò, A pseudo-
Workshop Proceedings, volume 3118, CEUR-WS, softmax function for hardware-based high speed
2021, pp. 51–63. image classification, Scientific reports 11 (2021)
[4] R. Brociek, G. Magistris, F. Cardia, F. Coppa, 1–10.
S. Russo, Contagion prevention of covid-19 by [16] D. S. Maitra, U. Bhattacharya, S. K. Parui, Cnn
means of touch detection for retail stores, in: CEUR based common approach to handwritten character
Workshop Proceedings, volume 3092, CEUR-WS, recognition of multiple scripts, in: 2015 13th In-
2021, pp. 89–94. ternational Conference on Document Analysis and
Recognition (ICDAR), IEEE, 2015, pp. 1021–1025.
95
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
[17] E. Fix, J. L. Hodges Jr, Discriminatory analysis- rotating machine faults using frequency domain
nonparametric discrimination: Small sample perfor- data fusion and artificial neural networks, Machines
mance, Technical Report, California Univ Berkeley, 6 (2018) 59.
1952. [31] A. A. Jaber, K. M. Ali, Artificial neural network
[18] J. S. Mohammed, J. A. Abdulhady, Rolling bearing based fault diagnosis of a pulley-belt rotating sys-
fault detection based on vibration signal analysis tem, International Journal on Advanced Science,
and cumulative sum control chart, FME Transac- Engineering and Information Technology 9 (2019)
tions 49 (2021) 684–695. 544–551.
[19] T. A. Dhomad, A. Jaber, Bearing fault diagnosis [32] A. Sharma, R. Jigyasu, L. Mathew, S. Chatterji, Bear-
using motor current signature analysis and the ar- ing fault diagnosis using frequency domain features
tificial neural network, International Journal on and artificial neural networks, in: Information
advanced scince Engineering Information Technol- and Communication Technology for Intelligent Sys-
ogy 10 (2020). tems, Springer, 2019, pp. 539–547.
[20] K. Taunk, S. De, S. Verma, A. Swetapadma, A brief [33] K. Babu Rao, D. Mallikarjuna Reddy, Fault detection
review of nearest neighbor algorithm for learning in rotor system by discrete wavelet neural network
and classification, in: 2019 International Confer- algorithm, Journal of Vibration and Control (2021)
ence on Intelligent Computing and Control Systems 10775463211030754.
(ICCS), IEEE, 2019, pp. 1255–1260. [34] N. F. Espinoza Sepúlveda, J. K. Sinha, Blind applica-
[21] T. Marwala, Condition monitoring using com- tion of developed smart vibration-based machine
putational intelligence methods: applications in learning (svml) model for machine faults diagnosis
mechanical and electrical systems, volume 200, to different machine conditions, Journal of Vibra-
Springer, 2012. tion Engineering & Technologies 9 (2021) 587–596.
[22] W. L. Lim, The application of artificial neural net- [35] D. He, R. Li, J. Zhu, Plastic bearing fault diagnosis
works for sensor validation in diesel engine condi- based on a two-step data mining approach, IEEE
tion monitoring and fault diagnosis, Ph.D. thesis, Transactions on Industrial Electronics 60 (2012)
University of Newcastle upon Tyne, 2009. 3429–3440.
[23] J. Fulcher, Advances in applied artificial intelligence, [36] D. Pandya, S. H. Upadhyay, S. P. Harsha, Fault
IGI Global, 2006. diagnosis of rolling element bearing with intrinsic
[24] D. A. Pisner, D. M. Schnyer, Support vector machine, mode function of acoustic emission data using apf-
in: Machine learning, Elsevier, 2020, pp. 101–121. knn, Expert Systems with Applications 40 (2013)
[25] G. Capizzi, G. Lo Sciuto, C. Napoli, E. Tramontana, 4137–4145.
M. Woźniak, A novel neural networks-based tex- [37] Q. Wang, Y. B. Liu, X. He, S. Y. Liu, J. H. Liu,
ture image processing algorithm for orange defects Fault diagnosis of bearing based on kpca and knn
classification, International Journal of Computer method, in: Advanced Materials Research, volume
Science and Applications 13 (2016) 45 – 60. 986, Trans Tech Publ, 2014, pp. 1491–1496.
[26] R. R. Sarra, A. M. Dinar, M. A. Mohammed, K. H. [38] M. Safizadeh, S. Latifi, Using multi-sensor data fu-
Abdulkareem, Enhanced heart disease prediction sion for vibration fault diagnosis of rolling element
based on machine learning and 𝜒2 statistical opti- bearings by accelerometer and load cell, Informa-
mal feature selection model, Designs 6 (2022) 87. tion fusion 18 (2014) 1–8.
[27] A. Mohammed, R. Neilson, W. Deans, P. Mac- [39] J. Tian, C. Morillo, M. H. Azarian, M. Pecht, Mo-
Connell, Crack detection in a rotating shaft using tor bearing fault detection using spectral kurtosis-
artificial neural networks and psd characterisation, based feature extraction coupled with k-nearest
Meccanica 49 (2014) 255–266. neighbor distance analysis, IEEE Transactions on
[28] J. B. Ali, N. Fnaiech, L. Saidi, B. Chebel-Morello, Industrial Electronics 63 (2015) 1793–1803.
F. Fnaiech, Application of empirical mode decom- [40] M. Gohari, A. M. Eydi, Modelling of shaft unbal-
position and artificial neural network for automatic ance: Modelling a multi discs rotor using k-nearest
bearing fault diagnosis based on vibration signals, neighbor and decision tree algorithms, Measure-
Applied Acoustics 89 (2015) 16–27. ment 151 (2020) 107253.
[29] A. A. Jaber, R. Bicker, Fault diagnosis of industrial [41] J. Lu, W. Qian, S. Li, R. Cui, Enhanced k-nearest
robot gears based on discrete wavelet transform and neighbor for intelligent fault diagnosis of rotating
artificial neural network, Insight-Non-Destructive machinery, Applied Sciences 11 (2021) 919.
Testing and Condition Monitoring 58 (2016) 179– [42] W. Li, Y. Cao, L. Li, S. Hou, An orthogonal wavelet
186. transform-based k-nearest neighbor algorithm to
[30] K. C. Luwei, A. Yunusa-Kaltungo, Y. A. Sha’aban, detect faults in bearings, Shock and Vibration 2022
Integrated fault detection framework for classifying (2022).
96
Salah Mahmood Al-Khafaji et al. CEUR Workshop Proceedings 85–97
[43] L.-l. Jiang, H.-k. Yin, X.-j. Li, S.-w. Tang, Fault diag-
nosis of rotating machinery based on multisensor
information fusion using svm and time-domain fea-
tures, Shock and Vibration 2014 (2014).
[44] A. Tabrizi, L. Garibaldi, A. Fasana, S. Marchesiello,
Early damage detection of roller bearings using
wavelet packet decomposition, ensemble empirical
mode decomposition and support vector machine,
Meccanica 50 (2015) 865–874.
[45] J. S. L. Senanayaka, S. T. Kandukuri, H. Van Khang,
K. G. Robbersmyr, Early detection and classification
of bearing faults using support vector machine al-
gorithm, in: 2017 IEEE Workshop on Electrical Ma-
chines Design, Control and Diagnosis (WEMDCD),
IEEE, 2017, pp. 250–255.
[46] Z. Huo, Y. Zhang, L. Shu, A comparative study of
wpd and emd for shaft fault diagnosis, in: IECON
2017-43rd Annual Conference of the IEEE Industrial
Electronics Society, IEEE, 2017, pp. 8441–8446.
[47] Y.-K. Gu, X.-Q. Zhou, D.-P. Yu, Y.-J. Shen, Fault
diagnosis method of rolling bearing using principal
component analysis and support vector machine,
Journal of Mechanical Science and Technology 32
(2018) 5079–5088.
[48] B. Pang, G. Tang, C. Zhou, T. Tian, Rotor fault
diagnosis based on characteristic frequency band
energy entropy and support vector machine, En-
tropy 20 (2018) 932.
[49] U. Parmar, D. Pandya, Experimental investigation
of cylindrical bearing fault diagnosis with svm, Ma-
terials Today: Proceedings 44 (2021) 1286–1290.
[50] Y. E. Lee, B.-K. Kim, J.-H. Bae, K. C. Kim, Misalign-
ment detection of a rotating machine shaft using
a support vector machine learning algorithm, In-
ternational Journal of Precision Engineering and
Manufacturing 22 (2021) 409–416.
97