Deep Learning-Based Intelligent Fault Diagnosis Methods Toward Rotating Machinery

Abstract

Fault diagnosis of rotating machinery plays a significant role in the industrial production and engineering field. Owing to the drawbacks of traditional fault diagnosis methods, such as heavily dependence on human knowledge and professional experience, intelligent fault diagnosis based on deep learning (DL) has aroused the interest of researchers. DL achieves the desirable automatic feature learning and fault classification. Therefore, in this review, DL and DL-based intelligent fault diagnosis techniques are overviewed. DL-based fault diagnosis approaches for rotating machinery are summarized and discussed, primarily including bearing, gear/gearbox and pumps. Finally, with respect to modern intelligent fault diagnosis, the existing challenges and possible future research orientations are prospected and analyzed.

Full text

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 1
Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.Doi Number
Deep Learning-based Intelligent Fault
Diagnosis Methods towards Rotating
Machinery
SHENGNAN TANG1, SHOUQI YUAN1, YONG ZHU1, 2
1. National Research Center of Pumps, Jiangsu University, Zhenjiang 212013, Jiangsu, China
2. State Key Laboratory of Fluid Power and Mechatronic Systems, Zhejiang University, Hangzhou 310027, Zhejiang, China
Corresponding authors: Shouqi Yuan (shouqiy@ujs.edu.cn) and Yong Zhu (zhuyong@ujs.edu.cn).
This work is supported by National Natural Science Foundation of China under grant 51779107 and grant 51805214, and in part by China Postdoctoral
Science Foundation under grant 2019M651722 and Open Foundation of the State Key Laboratory of Fluid Power and Mechatronic Systems (No. GZKF-
201905), Natural Science Foundation of Jiangsu Province under grant BK20170548 and the Youth Talent Development Program of Jiangsu University.
ABSTRACT Fault diagnosis of rotating machinery plays a significant role in the industrial production and
engineering field. Owing to the drawbacks of traditional fault diagnosis methods, such as heavily
dependence on human knowledge and professional experience, intelligent fault diagnosis based on deep
learning (DL) has aroused the interest of researchers. DL achieves the desirable automatic feature learning
and fault classification. Therefore, in this review, DL and DL-based intelligent fault diagnosis techniques
are overviewed. DL-based fault diagnosis approaches for rotating machinery are summarized and discussed,
primarily including bearing, gear/gearbox and pumps. Finally, with respect to modern intelligent fault
diagnosis, the existing challenges and possible future research orientations are prospected and analyzed.
INDEX TERMS Deep learning, deep neural network, intelligent fault diagnosis, rotating machinery.
I. INTRODUCTION
As an essential part and one of the most representative of
mechanical equipment, the rotating machinery relies on
rotation for purpose of a specific function. It has been widely
used in the field of mechanical transmission, including
aircraft engines, pump, wind turbine generator systems, gas
turbine engine, and power plants [
1
,
2
]. Owing to unavoidable
malfunction and downtime of the mechanical equipment in
the process of operation, fault diagnosis is of great
significance for rotating machinery in order to ensure the
reliability and safety [
3
-
6
].
In general, fault diagnosis methods are divided into the
followings, model-based methods, signal-based methods,
knowledge-based methods and composite methods [
7
]. In
view of traditional fault diagnosis methods, they are
primarily based on mechanism, feature frequency or fault
feature extraction [
8
]. On account of dependence on the
practical experience and professional knowledge, it is
difficult to detect the fault of rotating machinery with
complex structure by the use of traditional subjective fault
diagnosis methods [
9
,
10
]. Some improvement and
achievement have been made on fault diagnosis with respect
to the model-based methods and signal-based methods. A
kalman filter was improved and used to evaluate the state of
hydraulic actuator and leakage of hydraulic system by
Sepehri et al [
11
,
12
]. Du and Goharrizi et al. analyzed and
estimated vibration signal of hydraulic pump, pressure signal
of hydraulic cylinder and actuator via wavelet transform
[
13
,
14
]. The doubly iterative empirical mode decomposition
(EMD) and adaptive multifractal detrended fluctuation
analysis were employed to analyze fault diagnosis of the
bearing, the gear and the piston pump [
15
,
16
]. Although the
shortages of artificial data statistics have been compensated
by the methods discussed above to some extent, there are still
some limitations in fault diagnosis of rotating machinery
owing to the difficulty in feature extraction and complicate
mathematical model.
With the implement of “Industry 4.0” and “Internet +”,
artificial intelligence (AI) has been quickly integrated into
the various traditional industries [
17
]. Intelligent fault
diagnosis, which is combined with other feature extraction
methods, AI as the main body, has attracted more and more
attention. It is considered to be a powerful tool for big data
processing and fault diagnosis of mechanical equipment,
which provides a new exploration path for fault diagnosis
and health management of rotating machinery [
18
,
19
].

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
Great success has been achieved in fault diagnosis of
rotating machinery with traditional machine learning
methods, such as support vector machine (SVM) and
artificial neural network (ANN) [
20
-
23
]. Wavelet packet
decomposition and EMD were combined and used to feature
extraction, moreover, ANN was utilized to preliminary fault
diagnosis by Bin et al [
24
]. In order to achieve fault diagnosis
of hydraulic pipe, an integrated method including principal
component analysis (PCA), ANN and multiple adaptive
neural fuzzy inference system was proposed by Saeed et al
[
25
]. On account of on-line intelligent diagnosis based on
neural network, Schlechtingen et al. used it to fault diagnosis
of wind turbine generator [
26
]. In order to realize fault
identification of bearing, many various efforts have been
made on the exploration of novel methods. Amar et al.
proposed a neural network based on vibrational spectra [
27
],
Jiang et al. combined improved singular value decomposition
(SVD) and hidden markov model [
28
], and Zeng et al. used a
maximum interval classification method based on flexible
convex hull [
29
]. A novel diagnosis method for bearing was
proposed by Li et al., thereinto, the geometry of input data
was taken into account [
30
]. Zhu et al. combined multi-scale
fuzzy measure entropy, infinite feature selection and SVM to
explore the effectiveness of fault diagnosis for the bearing
[
31
]. In addition, intelligent diagnosis method was proposed
based on firefly neural network by Li et al [
32
]. However,
there are still some limitations and deficiencies in traditional
intelligent diagnosis methods. On the one hand, in
consideration of feature extraction, a large number of signal
processing technologies requires to be grasped and rich
experience in engineering practice needs to be possessed;
additionally, feature extraction and intelligent diagnosis are
treated separately, the relationship between them could not
be taken into account. On the other hand, with regard to
model training, the shallow model is used to characterize the
complex mapping relationship between signals and health
status, which leads to the obvious deficiency in diagnostic
ability and generalization performance of the model in the
face of mechanical big data [
33
,
34
].
Modern intelligent fault diagnosis technology is based on
the new theory and method of AI. In 2006, Hinton et al. first
proposed the deep learning (DL) theory in Science [
35
],
which triggered a wave of research on many different fields.
DL was ranked as the top 10 breakthrough technologies of
2013 by MIT Technology Review. In 2015, Hinton et al.
indicated that DL was thought to be one of machine leaning,
and breakthrough was analyzed and discussed in the respects
of image, video, audio and text processing [
36
]. It has been
adequately demonstrated that DL presents the broad
prospects on research and application. Through multi-layer
nonlinear network training, potential features of samples
have been learned and classification or prediction ability
have been improved with DL. DL methods that are widely
studied usually include deep belief network (DBN) [
37
,
38
],
stacked self-encoders (SAE) [
39
], convolutional neural
network (CNN) [
40
,
41
] and recurrent neural network (RNN)
[
42
]. Based on multivariate encoder information, a CNN was
designed to intelligently identify the failure of planetary gear
box by Lin et al. Not only did the deficiency of traditional
vibration analysis overcome, but also a potential intelligent
tool was provided to obtain the expected diagnosis towards
rotating machinery [
43
]. In accordance with multi-domain
features, an integrated kernel extreme learning machine was
proposed and used to gear box, rotor and motor bearing,
effective diagnosis was achieved by visualization with the
method [
44
].
Presently, on account of the wide use of DL in many
pattern recognition fields, intelligent fault diagnosis based on
DL has attracted much more attention of professional
researchers in machinery field. Therefore, this review will
focus the efforts on fault diagnosis of rotating machinery. It
will place an emphasis on fault diagnosis integrated with
deep neural network technology. Furthermore, a summary of
the applications will be given towards commonly used
rotating machinery such as bearing, gear and pump. Finally,
the above discussions are concluded and the possible
research directions are provided to inspire more researches in
this field.
II. DL BASED FAULT DIAGNOSIS
A. ARTIFICIAL INTELLIGENCE
As a new and interdisciplinary science, AI is aiming at
simulating some of human thinking processes and intelligent
behavior by the use of computer. It can be achieved in
computer by the following two different ways, one is
engineering approach which adopts traditional programming
technique; the other is modeling approach, such as generic
algorithm and ANN.
From SIRI to AlphaGo, rapid development of AI has been
supposed to be interesting, surprising and outstanding
[
45
,
46
]. AI approaches have been integrated into many
various fields, great achievements have been obtained in
man-machine game, pattern recognition, automatic
engineering and knowledge engineering [
47
-
49
]. Because of
the increase of machinery data and complication of fault
which result in high uncertainty during diagnosis process, AI
based methods will outperform traditional methods on
diagnosis efficiency. AI-based approaches can be divided
into the following two categories, knowledge-driven methods
and data-driven approaches [
50
].
B.
DL
As a distinguished development of AI, DL can be
understood as feature learning or representation-learning,
which possesses multiple and high levels representations of
data, concretely, through DL, low-level features from
simple and nonlinear modules were composed to form more
abstract high-level representations in terms of categories or
features, complex functions and distributed feature
representations of data can be obtained [
51
]. Very deep

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
neural networks can be considered to be typical DL model.
DNN plays an essential role in deep models, mainly
including DBN, SAE, CNN, RNN and GAN. As one of
unsupervised learning ways, DBN is a DNN which
possesses the stacked structure and consists of multiple
Restricted Boltzmann Machines. Similarly, SAE is used to
deal with high-dimensional data by means of unsupervised
learning. It is composed of multiple auto-encoders, which
presents a three-layer neural network including both
encoding and decoding processes [
52
]. CNN is a supervised
learning network, whose structure seems to be more
complex with convolution layers, sample layers and full
connected layers. With especial ring structure, RNN
represents a unique advantage in settling learning problems
with sequential data via unsupervised learning. It is
interesting that GAN is comprised by generally both non-
linear function models, that is, a generative model and a
discriminative model respectively [
53
]. In order to
overcome the deficiency of insufficient training data, deep
transfer learning (TL) has achieved that the learning from
the source domain is employed to the target domain [
54
,
55
].
Deep reinforcement learning is the learning of the
intelligent system from the environment to the behavior
mapping, with the purpose of maximizing the long-term
cumulative reward, which really realizes the machine's
capability of self-learning and self-thinking [
56
] Compared
with traditional machine learning, it should be noted that
the key advantage of DL is layers of features are
automatically learned from raw data through a general-
purpose learning procedure, not dependent on engineering
skills and domain expertise [36].
In view of the advantages of DL, it has been applied to
many different fields such as language processing,
automatic speech recognition, and audio recognition
[
57
,
58
]. Meanwhile, this has aroused the interest of
researchers in the field of mechanical engineering, making
it play an essential role in intelligent fault diagnosis
combined with other methods and technologies [
59
,
60
].
C.
Overview of DL based Fault Diagnosis
Intelligent fault diagnosis is the combination of AI and fault
diagnosis, which expresses comprehensive use of domain
expertise and AI technology and strong capability of
processing considerable mechanical data [
61
,
62
].
Three different steps are included in traditional intelligent
fault diagnosis, namely, signal collection, feature extraction
and fault classification. Since some exhausted and
handcrafted signal feature extraction technologies could be
required in those methods, diagnosis results will finally be
affected. Moreover, the ability to learn the complex non-
linear relationships between features and patterns will be
hindered with the shallow structures such as SVM [
63
]. With
respect to new intelligent fault diagnosis, in place of feature
extraction and selection, the features can be automatically
learned from raw signals, which presents more intelligent
than conventional approaches [
64
,
65
].
Some good results have been achieved in applications to
gear, gearbox, bearing, rolling, pump, wind turbine and
nuclear power plant with modern intelligent fault diagnosis
[
66
-
69
]. Deep CNN was employed to fault diagnosis of wind
turbine, bearing and gearbox by Liu et al., and it is worth
noting that the spatiotemporal pattern network was integrated
[
70
,
71
]. Xu and his colleagues proposed a new intelligent
diagnosis method based on elaborately designed deep neural
network for failure detection of wind turbine, which solved
the problem of unbalanced distribution with regard to
SCADA data [
72
]. Combined a CNN with a Naïve Bayes
data fusion proposal, Chen et al. applied DL theory to nuclear
power plant inspection [69]. Zhang et al. constructed a new
unsupervised learning method called general normalized
sparse filtering, which was used for fault diagnosis of rolling
bearing and planetary gearbox [
73
].
III. APPLICATIONS OF DL TOWARD FAULT DIAGNOSIS
IN ROTATING MACHINERY
Combined with the above analysis, it can be proved that it
has acquired some improvements and achievements for
machinery fault diagnosis illuminated from the applications
of DL technique in other fields. As shown in Table 1, the
applications of DL-based methods in machinery fault
diagnosis have been summarized. In order to evaluate the
diagnosis performance of methods, the following evaluation
indicators are employed, including the diagnosis accuracy,
the training accuracy, the average testing accuracy, the
prediction accuracy, the clustering effect from visualization.
This review will play an emphasis on intelligent fault
diagnosis of typical rotating machinery, including bearing,
gear and pump. Furthermore, DL-based approaches for
improving diagnosis accuracy will be analyzed and discussed
in the following. TABLE I
SUMMARY OF DL-BASED METHODS FOR MACHINERY FAULT DIAGNOSIS
Technique
Applications
Diagnosis
Effect
(Ref.)
CNN with first-layer
kernel dropout
Bearing fault
diagnosis
99.77%
accuracy
[74]
CNN and second
generation wavelet
transform
Motor bearing
fault diagnosis
99.63%
accuracy
[75]
CNN and wavelet
packet energy
Spindle bearing
fault diagnosis
99.8%
accuracy
[76]
CNN with hierarchical
structure
Bearing fault
diagnosis
desirable
performance
[40]
CNN
Bearing fault
diagnosis
99.75%
[77]
CNN
Bearing fault
diagnosis
100%
accuracy
[78]
CNN
Rotor and
bearing fault
diagnosis
100.00%
accuracy for
rotor; 93.33%
for bearing
[79]
CNN
Bearing fault
diagnosis
99.89%
accuracy
[80]
DBN and sparse
autoencoder
Bearing fault
diagnosis
97.82%
average
accuracy
[81]
RNN-based
Bearing fault
99.85%
[82]

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
autoencoders
diagnosis
accuracy
Autoencoder and
particle swarm
optimization-SVM
Bearing fault
diagnosis
94.34%
[83]
SAE and softmax
regression
Bearing fault
diagnosis
95.26%
[84]
Deep convolutional
auto-encoding NN
Bearing fault
diagnosis
99.7%
[85]
Kernel auto-encoder
based on firefly
optimization
Bearing fault
diagnosis
95.95%
average
accuracy
[86]
Deep convolution
variational autoencoder
network
Self-priming
centrifugal
pump and
bearing fault
diagnosis
98.845 and
97.62%
average
accuracy
respectively
[87]
DBN
Bearing fault
diagnosis
93.17%
average
accuracy
[88]
DBN and singular
value decomposition
Bearing fault
diagnosis
100%
[89]
DBN and Hilbert
envelope spectrum
Bearing fault
diagnosis
99.55%
accuracy
[90]
DBN
Bearings fault
pattern
recognition
99.93%
accuracy
[91]
DBN based on multi-
layer neural networks
Gearbox fault
diagnosis
98.1 %
[92]
CNN
Gearbox fault
diagnosis
96.8% mean
accuracy
[93]
SAE, dropout
technique and ReLU
activation function
Gearbox fault
diagnosis
99.34%
[94]
GAN and stacked
denoising autoencoders
Gearbox fault
diagnosis
98.4%
accuracy
[95]
TL
Gearbox and
bearing fault
diagnosis
100%
[96]
Multimodal deep
support vector
Gearbox fault
diagnosis
97.08%
[97]
Support tensor machine
Gear fault
diagnosis
99.50%
[98]
DNN
Bearings and
gears fault
diagnosis
100%
[39]
CNN
Bearing, self-
priming
centrifugal
pump, and
axial piston
hydraulic pump
fault diagnosis
99.79%,
99.481%,
100%
prediction
accuracy
[99]
SAE
Spacecraft fault
diagnosis
98.35%
[100]
Refrigerant charge
fault detection-based
CNN
Heat pump
system
99.9%
[101]
A. INTELLIGENT FAULT DIAGNOSIS OF BEARING
As one of well-known and widely-used rotary machinery,
bearing is of great significance but its brokendown occupies
nearly 45-55% of equipment fault, which will lead to
accidents, downtime, even severe damage and economic loss
[
102
,
103
]. Therefore, it is of vital importance to investigate
intelligent fault diagnosis methods of bearing, especially the
DL technique.
In order to overcome the imbalanced distribution of
machinery health conditions, a new learning method called
deep normalized CNN (DNCNN) was investigated to
classify the faults of bearing by Lei et al [
104
]. Three bearing
datasets are employed to validate the diagnosis accuracy of
the proposed methods, in which single faults and compound
faults with various imbalanced degrees are taken into account.
In Figure 1(a-c), it can be seen that DNCNN presents the
superiority than S-CNN and R-CNN in terms of learning
features from the vibration signals, in which the features
cluster well. By the use of the confusion matrices, the
imbalanced classification results were successfully obtained,
that is, 95.4% of the samples were correctly classified by the
proposed method, and only 4% of the samples were
misclassified.
FIGURE 1. The visualization of learned features for Dataset A: (a) S-CNN,
(b) R-CNN, (c) DNCNN. S-CNN represents the convolutional neural
networks (CNN) using sigmoid function, R-CNN represents the CNN
using ReLU, DNCNN represents deep normalized convolutional neural
network, respectively [104].
As one method of machine learning, TL makes it possible
that one pretrained model is employed again to another task
with the purpose of reducing the distribution discrepancy and
enhancing the predictive performance [
105
,
106
]. Generally,
both developing model and pretraining model are included,
moreover, the latter is widely used in machine learning.
Through the integration of auto-balanced high-order
Kullback-Leibler divergence, smooth conditional distribution
alignment and weighted joint distribution alignment, a novel
TL framework was designed for fault diagnosis of rotator
bearing and gearbox under varied conditions [
107
].By
introducing grey wolf optimization algorithm, a new TL-
based method was constructed for diagnosis of the bearing. It
was worth mentioning that long-short term memory RNN
was employed to gain some auxiliary datasets [
108
]. Inspired
by the idea of TL, a deep CNN was proposed to be used for
fault diagnosis of unlabeled data by Lei et al., which made it
possible that labeled data from one machine after being

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
trained could effectively classify the unlabeled data from
other machines [
109
]. Furthermore, as one of CNN, a
transfer neural network based on feature was explored for
state identification of bearings. In comparison to the other
methods such as CNN and multi-layer adaptation CNN, the
average classification accuracy of the proposed method was
the highest one which achieved 84.32%. It has been
demonstrated that more desirable transfer results and transfer
performance were obtained with FTNN. Seen from Figure
2(f), in consideration of the learned transferable features, the
distribution was adapted efficaciously, furthermore, the
among-class distance was expanded [
110
]. In order to
overcome the limitations in training and the performance
degradation, a new deeper 1D CNN based on the residual
learning was developed for fault diagnosis of wheelset
bearings, and the effectiveness was approved by visualization
Figure 3 [80].
FIGURE 2. The visualization of the learned features on the dataset B
(source domain) and the dataset D (target domain): (a) CNN, (b) TCA, (c)
DAFD, (d) DDC, (e) MACNN, and (f) FTNN. CNN represents convolutional
neural networks, TCA represents transfer component analysis, DAFD
represents, DDC represents, MACNN represents multi-layer adaptation
CNN, and FTNN represents feature-based transfer neural network [110].
A new deep TL with three-layer sparse encoder was
investigated by Wen et al, which was validated by the use of
motor bearing dataset. Compared with other traditional
methods, such as DBN, ANN, sparse filter, and SVM, this
proposed method presents good performance and the
prediction accuracy achieved 99.82% [
111
].
He et al. presented a composite deep signal processing
approach, which integrated vibration analysis and deep
learning [
112
]. Vibration analysis was embedded into the
discrete Fourier transform - inverse discrete Fourier
transform autoencoder, which achieved that time-frequency
characteristics were learned adaptively and effective
convergence was obtained in view of learning procedure.
Real bearing data was employed to validate the performance
of the proposed method, which presented obviously higher
diagnosis accuracy compared with those of popular deep
neural network (DNN), CNN and SVM. Specifically, the
testing accuracy reached 100.00% while below 95.50% in
other methods when shaft speeds were set as 45 and 60.
FIGURE 3. Visualization of these five methods in noise environment
(SNR=−16 dB). ADCNN represents adaptive deep CNN, Wen-CNN
represents CNN proposed by Wen et al, MSCNN represents multiscale
CNN, WDCNN represents deep convolutional neural networks with wide
first-layer kernels, Der-1DCNN represents deeper 1D CNN [80].
Motivated by the thought of enhancing the generalization
ability and robustness of diagnosis model through utilizing
the structural domain information among multiple bearing
fault types, a new deep output kernel learning was presented
in order to overcome the randomness of some deep learning
methods [
113
]. In comparison to one the-state-of-the-art
signal analysis method, four shallow models and four deep
models, it showed higher accuracy of 100.00% and shorter
training time of less than 7 s.
Combined compressed sensing with a convolutional DBN,
a new improved deep model with powerful feature learning
ability was constructed to analyze the single fault and
compound faults of rolling bearing by Shao et al [
114
]. It
should be pointed that the analysis efficiency was enhanced
by compressed sensing and the generalization performance
was enhanced via exponential moving average technique.
The average testing accuracy of the proposed method
achieved 94.80%, which was be superior to other traditional
methods of no more than 90.00%, including the standard
DBN, CNN, deep auto-encoder (DAE), BP neural network
and SVM. From the visualization of PCA (Figure 4), it can
be proved that the better clustering result was obtained from
the proposed method, which expressed the superiority in
capturing potential features.
FIGURE 4. Three dimensional visualization of different features using
PCA. (a) Compressed data features, (b) extracted 22 features, and (c)
deep features [114].

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
In regard to the diversity of the fault data distribution and
the data reconstruction ability, a sparse stacked denoising
autoencoder is developed for the fault diagnosis of rolling
bearings [
115
]. With the introduction of optimized transfer
learning algorithm, the problem of the domain adaption was
solved, and the accuracy of the target domain achieved
96.70% in one of motor loads. It was demonstrated that the
quality of the target domain was influenced by the
performance of the source domain algorithm, however, it is
limited to only depend on the enhancement of the source
domain.
In consideration of unlabled data in practical engineering,
combined with Gath-Geva clustering algorithm, a stacked
denoising autoencoder was investigated for roller bearing
fault diagnosis without principal component analysis and
data mark [
116
]. The proposed method presented the
superior clustering effect. Moreover, its classification
accuracy was higher compared to those of the other
combination models, and the highest one reaches up to
100.00%.
B. INTELLIGENT FAULT DIAGNOSIS OF GEAR AND
GEARBOX
It was indicated that the gearbox failure was the primary
contributor to equipment fault, which took up nearly 40% in
mechanical transmission field according to the investigation
performed under the assistance of the Institute of Electrical
and Electronic Engineers (IEEE) [
117
]. Hence, in view of the
fault diagnosis for gear and gearbox, the methods based on
DL will be highlighted in the following.
Combined CNN and extreme learning machine, a new
model without any extra training and fine tuning was
established by Chen et al., gearbox dataset and motor bearing
dataset were selected to verify the effectiveness of the
proposed method, as depicted in Figure 5 [
118
]. It was
demonstrated that the feature learning capability was
improved by the CNN employed as an automatic feature
extractor, and the classification performance and the learning
speed were promoted through the extreme learning machine.
In view of gearbox, the results indicated that the training
accuracy and average test accuracy reached 100.00%±0.00
and 99.83% ±0.24 respectively, which achieved the
superiority in contrast to the other methods such as standard
CNN. With regard to motor bearing, the training accuracy
and average test accuracy gained 100.00%±0.00 and 99.92%
±1.24 respectively, which exhibits the better classification
performance.
With respect to signal processing-based methods, a
wavelet packet transform, a distance evaluation technique
and a support vector regression (SVR)-based generic multi-
class solver were combined for fault diagnosis of bearing and
gearbox [
119
]. The proposed method presented the superior
representative capability and the higher diagnosis accuracy,
which was mainly attributed to the influences of wavelet
basis functions on the proposed whole framework.
FIGURE 5. The structure of the proposed CNN-ELM model for fault
diagnosis [118].
Motivated by the idea of TL, a new intelligent fault
diagnosis scheme named deep transfer network with joint
distribution adaptation was exploited to overcome the
applicability limitations for the traditional diagnosis methods
[
120
]. Three datasets including wind turbine, bearing and
gearbox fault dataset, were employed to verify the
performance of the proposed framework, which displayed
some good results in accordance to various working
conditions, the types and severities of fault. In order to
demonstrate the performance of the proposed method, the
average diagnosis accuracy, missing alarm rate, and false
alarm rate were chosen as evaluation indicators, meanwhile,
eight state-of-the-art intelligent diagnosis approaches were
used as comparisons. With regard to gearbox, the average
diagnosis accuracy of the proposed scheme outbalanced
those of other methods, which reached up to more than 96%.
Similarly, a working condition-robust fault diagnosis method
based on an improved joint distribution adaptation was
exploited to achieve the acquisition of more useful samples
and reduction of the input dimension [
121
]. The vibration
signal datasets of roller bearings and a gearbox were used to
validate the fault diagnosis performance of the proposed
method, which obviously demonstrated its effectiveness
although its computational time was completely
unsatisfactory.
In consideration of unexpected diagnostic results via
utilizing the spectrum signal, modern spectrum signals
through preprocessing current signals was incorporated into
DNN by Li et al [
122
]. Compared with SVM and BPNN, the
proposed method represented superior diagnostic results for
faults detection in planetary gears, the testing accuracy rate
of which achieved 96.69% with standard deviation of 1.05%.
Furthermore, the diagnosis advantage of the proposed
method was proved by the visualization of fault
characteristics from PCA, as shown in Figure 6(d), which
presented better clustering effect and little overlapping than
those of others.
A new DL method was developed for fault diagnosis of
planetary gear through combining power spectral entropy of
variational mode decomposition and DNN, which was
trained through unsupervised training and supervised fine
tuning [
123
]. It is beneficial to fault classification via the
reduction of raw signals by the use of BP. Compared with
other methods such as SVM and BP, the proposed method
exhibited the higher overall recognition rate of 100%.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
FIGURE 6. Scatterplot of the main characteristic components: (a)
signals in the time domain, (b) direct spectrum signals, (c) spectrum
signals without the power frequency, and (d) modified spectrum signals.
Reprinted with permission from ref. [122].
Based on time-frequency analysis and DNN, a deep
residual learning was constructed for fault diagnosis in
planetary gearbox [
124
]. Its performance was demonstrated
under nonstationary running conditions, which implied
admirable results for the incipient fault detection, especially
when rotating speed was variable. The proposed scheme
presented the higher diagnosis accuracy, which reached up to
95.4% under faulty condition.
A deep CNN was constructed for gearbox fault diagnosis
under different operating conditions, which was compared
with different SVM classifiers optimized by the use of a grid
search technique [
125
]. With regard to vibration signals of
different directions, the proposed method showed the
superiority to other traditional methods. The identification
accuracy achieved 93.6% and the computational cost was
reduced.
By the use of the maximum correntropy and artificial fish
swarm algorithm, a new deep autoencoder feature learning
method was designed and optimized by Shao et al. for the
fault diagnosis of gearbox and electrical locomotive roller
bearing [
126
]. Compared with other approaches such as
standard deep autoencoder, BP and SVM, the proposed
method possessed the admirable diagnosis effectiveness
including robustness, and the average testing accuracy
reached 94.05% with a smaller standard deviation of 1.34.
In order to overcome the dependence on numerable
labelled data and time consuming of handcrafted feature
extraction in traditional supervised diagnosis, a new deep
semi-supervised method of multiple association layers
networks was investigated by Zhang et al [
127
]. As shown in
Figure 7, the wavelet packet transform was employed to
preprocess raw signals, moreover, the labeled and unlabeled
data was together used to train the model and method. It can
be concluded that the recognition accuracy of the proposed
method presented the advantage in comparison to SAE and
DBN with less labeled data. The recognition rate increased
from 78.58% to 93.26% with the increase of the labeled
samples from 2% to 100%. Additionally, it is worth to note
that the optimization of the hyper-parameter may be a key
challenge and have a great influence on the performance of
the neural networks.
FIGURE 7. The framework of the proposed method for fault diagnosis of
planetary gearbox [127].
C. INTELLIGENT FAULT DIAGNOSIS OF PUMPS
With the function diversity and structural complexity of
hydraulic system, it seems to be more challenging for fault
identification and classification [
128
-
130
]. As power source
of hydraulic system, hydraulic pump plays an indispensable
role in reflecting the working state of the system [
131
,
132
].
Meanwhile, with respect to the wide use of centrifugal pump,
whose operating state directly affects production and safety.
According to the statistics on the mechanical and electrical
equipment defects, more than 50% are connected with pump
failures [
133
,
134
]. Therefore, it is of great significance to
diagnose pump faults accurately and effectively in order to
ensure the safety and reliability of the system. Although
some researches have achieved admirable results on
machinery intelligent fault diagnosis, there are still little
investigations on pumps.
As an essential and famous DL, DNN has aroused great
attention in intelligent fault diagnosis, which has also
stimulated interest in research for pumps. A new data-driven
method based on CNN with LeNet-5 was developed by Wen
et al [
135
]. In regard to axial piston hydraulic pump, two
fault conditions were taken into account, and the piston shoes
and swashplate wearing and valve plate wearing were
included. The prediction accuracy achieved 100%. As for
self-priming centrifugal pump, four faults conditions were
analyzed, including bearing roller wearing, inner race
wearing, outer race wearing, and impeller wearing fault
condition. From the results of confusion matrix, it can be
observed that the prediction accuracy of 99.481% was

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
obtained, moreover, the most misclassification was 0.4%.
Similarly, a simple improved CNN was proposed for fault
diagnosis of hydraulic pump by Yan et al [
136
]. Two
operating conditions including stable and variable pump
speeds were investigated, the accuracy rate exceeded 95%
and 90% in view of the worst results.
Based on image-processing technique, a probabilistic
neural network was introduced by Lu et al, and it was
achieved that the feature was automatically extracted in a
two-dimensional space [
137
]. The speeded-up robust features
and t-Distributed Stochastic Neighbor Embedding (SNE)
were employed to automatic feature extraction and
dimensionality reduction respectively. By the use of t-SNE,
the feature information was more clustered and presented the
potent capability of separability (Figure 8(A)). It can be
concluded from the cross-validation results that the proposed
method presented the high diagnosis accuracy. The
classification accuracy was more than 96% for the self-
priming centrifugal pump. For the axial piston hydraulic
pump, the average classification accuracy achieved as high as
98.71%.
FIGURE 8. The first three features extracted using t-SNE (A) and without
using t-SNE (B) [137].
Through introducing data indicator containing time and
frequency, Wang et al. investigated a DBN for multiple faults
diagnosis of the axial piston hydraulic pump, which achieved
the advantageous classification accuracy of 97.40% in
comparison to SVM and ANN [
138
]. It deserved to be
mentioned that the restricted Boltzmann machine was used to
realize the automatic learning of fault features.
In consideration of the complex dynamic behavior for
rotary machinery, symbolic analysis plays an essential role
[
139
]. In combination with hierarchical symbolic analysis
(HSA), a CNN was used for fault diagnosis of centrifugal
pump [
140
]. The diagnosis accuracy improved as the number
of hierarchical layers increased, moreover, the computation
time admirably reduced after using HSA. It achieved the
maximum of 98.50% when hierarchical layer was 3. By
means of data fusion which achieved the transformation of
multi-sensor-signals to images, another improved CNN was
proposed by Wang et al., and the prediction accuracy reached
up to 99.47%. It presented the obvious better diagnosis
effectiveness in comparison with other intelligent methods
[
141
].
Owing to the long operating time and computing
complication, a novel intelligent fault diagnosis scheme was
developed combined deep structure with SVM, which
realized the learning of the hidden features [
142
-
144
]. The
similar conclusions were obtained that the accuracy rate
increased with the number of network layers. In contrast to
other methods, the proposed method exhibited the superior
diagnosis performance. In consideration of the accuracy rate
and computing time, the optimum result achieved up to
97.75% with standard deviation of 0.20.
IV. CONCLUSIONS AND PERSPECTIVES
Relevant studies on fault recognition methods have been
performed by our research group [
145
,
146
]. Additionally,
PCA and XGBoost were integrated to diagnose hydraulic
valves. It is worth noting that we have conducted many
investigations on fault diagnosis and signal processing for
hydraulic pumps and centrifugal pumps, which mainly
concentrate on conventional intelligent methods [
147
-
150
].
Furthermore, we gradually begin to study intelligent fault
diagnosis methods such as SVM for hydraulic pumps [
151
],
which provides a theoretical foundation for the following
researches on DL-based fault diagnosis approaches. In the
present and future, we will put emphasis on the DNN-based
methods and explore multi-information fusion technique with
well generalization capability, moreover, remote diagnosis
system will be exploited and constructed.
In accordance with the analysis and discussions above, the
methods based on DL can not only adaptively extract the
hidden complex and changeable fault information, but also
overcoming the reliance on diagnostic knowledge and
engineering experience of traditional methods. Although
these methods have achieved some expected results in rotary
machinery, there are still some challenges in the current
researches and the corresponding future research directions
are as follows:
(a) A large number of studies only used experiments or
existing datasets to validate the effectiveness of the proposed
methods, and the underlying mechanism of improved
diagnostic accuracy has not been analyzed in details.
(b) Many researches primarily focus on the single physical
source information, diagnosis accuracy requires to be
improved owing to small data size. It is significant to pay
more attention to multi-source information, which can
comprehensively reflect the state of equipment. But multi-
source signal has diversity and complexity problems, which
need to be further studied.
(c) The commonly used single marker system has
interpreted fault information out of context, and the
introduction of multi-marker system could be promising to
explore the identification of multiple faults.
(d) On account of many present methods, only the
diagnosis accuracy is improved. However, in the face of the

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
fault with more coupled and concurrency characteristics, it is
urgent for further exploring the identification of complex
faults and the generalization performance of the method.
Based on the thinking of DL, intelligent fault diagnosis
strategies are overviewed in this review. The applications of
DL-based techniques in fault diagnosis of rotating machinery
are thoroughly analyzed and discussed, mainly bearing, gears
and pumps. The diagnosis performance of these emerged
methods is highlighted, which provides ideas and guidance
for the exploration and applications of novel intelligent fault
diagnosis in rotary machinery extending to other machineries.
REFERENCES
[
1
] C. Shen, D. Wang, F. Kong, and P. W.Tse, “Fault diagnosis of rotating
machinery based on the statistical parameters of wavelet packet
paving and a generic support vector regressive classifier,”
Measurement. Vol. 46, no.4, pp. 1551-1564, May 2013.
[
2
] B.M. Ebrahimi, J. Faiz, and M.J. Roshtkhari, “Static-, dynamic-, and
mixed-eccentricity fault diagnoses in permanent-magnet synchronous
motors,” IEEE T. Ind. Electron., Vol. 56, no.11, pp. 4727-4739, Nov.
2009.
[
3
] Y. Lei, J. Lin, Z. He, and M. Zuo, “A review on empirical mode
decomposition in fault diagnosis of rotating machinery,” Mech. Syst.
Signal Process. Vol. 35, no. 1, pp. 108-126, Feb. 2013.
[
4
] S. Yin, and O. Kaynak, “Big data for modern industry: challenges and
trends,” Proc. IEEE. Vol. 103, no.2, pp. 143-146, 2015.
[
5
] W. Qiao, and D. Lu, “A survey on wind turbine condition monitoring
and fault diagnosis,” IEEE T. Ind. Electron., Vol. 62, pp. 6536-6545,
2015.
[
6
] Y. Lei, J. Lin, M. Zuo, and Z. He, “Condition monitoring and fault
diagnosis of planetary gearboxes: a review,” Measurement. Vol. 48,
pp. 292-305, Feb. 2014.
[
7
] Z. Gao, C. Cecati, and S. X. Ding, “A survey of fault diagnosis and
fault-tolerant techniques-Part I: Fault diagnosis with model-based and
signal-based approaches,” IEEE T. Ind. Electron., Vol. 62, no.6, pp.
3757-3767, Mar. 2015.
[
8
] Y. Tian, C. Lu, and Z. Wang, “Approach for hydraulic pump fault
diagnosis based on wpt-svd and svm,” Appl. Mech. Mater., Vol. 764-
765, pp. 191-197, 2015.
[
9
] Y. Lei, F. Jia, X. Zhou, and J. Lin, “A deep learning-based method for
machinery health monitoring with big data,” J. Mech. Eng., Vol. 51,
no. 21, pp. 49-56, Nov. 2015.
[
10
] K. Worden, W. J. Staszewski, and J. J. Hensman, “Natural computing
for mechanical systems research: A tutorial overview,” Mech. Syst.
Signal Process. Vol. 25, no. 1, pp. 4-111, Jan. 2011.
[
11
] L An, and N Sepehri. “Hydraulic actuator circuit fault detection using
extended Kalman filter,” Am. Control Conf., IEEE, Vol. 5, no. 1, pp.
4261-4266, 2003.
[
12
] L An, and N Sepehri. “Leakage fault identifcation in a hydraulic
positioning system using extended Kalman filter,” Am. Control Conf.,
Proc. IEEE, Vol. 4, pp. 3088-3093, 2004.
[
13
] W. Du, Z. Zhou, and X. Huang, “The application of wavelet analysis
to hydraulic cylinder leakage detection,” Machine Tools and
Hydraulic Pressure. Vol. 6, pp. 318-319, 2003.
[
14
] A Y Goharrizi, and N Sepehri, “A wavelet-based approach to internal
seal damage diagnosis in hydraulic actuators,” IEEE T. Ind. Electron.,
Vol. 57, pp. 1755-1763, Sep. 2010.
[
15
] S. Xia, J. Zhang, S. Ye, B. Xu, J. Xiang, and H. Tang, “A mechanical
fault detection strategy based on the doubly iterative empirical mode
decomposition,” Appl. Acoust., Vol. 155, pp. 346-357, Dec. 2019.
[
16
] W. Du, M. Kang, and M. Pecht, “Fault diagnosis using adaptive
multifractal detrended fluctuation analysis,” IEEE Trans. Ind.
Electron., Vol. 67, no. 3, pp. 2272-2282, Mar. 2020.
[
17
] S. Zhou, S. Wang, et al., “Innovative thinking and methods,” China
Railway Publishing House, pp. 42, Sept. 2016.
[
18
] S. Chan, Y. Lu, and Y. Wang, “Data-driven cost estimation for
additive manufacturing in cyber manufacturing,” J. Manuf. Syst., Vol.
46, pp. 115-126, Jan. 2018.
[
19
] J. Wang, Y. Ma, L. Zhang, R. X. Gao, and D. Wu, “Deep learning for
smart manufacturing: methods and applications,” J. Manuf. Syst., Vol.
48, Part C, pp. 144-156, Jul. 2018.
[
20
] B. Samanta. “Gear fault detection using artificial neural networks and
support vector machines with genetic algorithms,” Mech. Syst. Signal
Process, Vol. 18, no. 3, pp. 625-644, May 2004.
[
21
] A. Hajnayeb, A. Ghasemloonia, S.E. Khadem, and M.H. Moradic.
“Application and comparison of an ANN-based feature selection
method and the genetic algorithm in gearbox fault diagnosis,” Expert
Syst. Appl., Vol. 38, no. 8, pp. 10205-10209, Aug. 2011.
[
22
] X. Zhang, B. Wang, and X. Chen, “Intelligent fault diagnosis of roller
bearings with multivariable ensemble-based incremental support
vector machine,” Knowl.-Based Syst., Vol. 89, pp. 56-85, Nov. 2015.
[
23
] R. Jegadeeshwaran, and V. Sugumaran, “Fault diagnosis of
automobile hydraulic brake system using statistical features and
support vector machines,” Mech. Syst. Signal Process, Vol. 52-53, pp.
436-446, Feb. 2015.
[
24
] G.F. Bin, J.J. Gao, X.J. Li, and B.S. Dhillonc, “Early fault diagnosis
of rotating machinery based on wavelet packets-empirical mode
decomposition feature extraction and neural network,” Mech. Syst.
Signal Process, Vol. 27, pp. 696-711, Feb. 2012.
[
25
] R A Saeed, A N Galybin, and V Popov, “3D fluid-structure modelling
and vibration analysis for fault diagnosis of Francis turbine using
multiple ANN and multiple ANFIS,” Mech. Syst. Signal Process, Vol.
34, no. 1-2, pp. 259-276, Jan. 2013.
[
26
] M. Schlechtingen, and I F. SANTOS, “Comparative analysis of neural
network and regression based condition monitoring approaches for
wind turbine fault detection,” Mech. Syst. Signal Process, Vol. 25, no.
5, pp. 1849-1875, Jul. 2011.
[
27
] M. Amar, I. Gondal, and C. Wilson, “Vibration spectrum imaging: A
novel bearing fault classification approach,” IEEE T. Ind. Electron.,
Vol. 62, no. 1, pp. 494-502, Jan. 2015.
[
28
] H. Jiang, J. Chen, G. Dong, T. Liu, and G. Chen, “Study on Hankel
matrix-based SVD and its application in rolling element bearing fault
diagnosis,” Mech. Syst. Signal Process, Vol. 52-53, pp. 338-359, Feb.
2015.
[
29
] M. Zeng, Y. Yang, J. Zheng, and J. Cheng, “Maximum margin
classification based on flexible convex hulls for fault diagnosis of
roller bearings,” Mech. Syst. Signal Process. Vol. 66-67, pp. 533-545,
Jan. 2016.
[
30
] Y. Li, C. K. L. Lekamalage, T. Liu, P. Chen and G. Huang, “Learning
representations with local and global geometries preserved for
machine fault diagnosis,” IEEE Trans. Ind. Electron., Vol. 67, no. 3,
pp. 2360-2370, Mar. 2020.
[
31
] K. Zhu, L. Chen, and X. Hu, “A multi-scale fuzzy measure entropy
and infinite feature selection based approach for rolling bearing fault
diagnosis,” J. Nondestruct. Eval., Vol. 38, no. 4, Dec. 2019.
[
32
] W. Li, S. Weng, and S. Zhang, “A firefly neural network and its
application in bearing fault diagnosis,” Chin. J. Mech. Eng., Vol. 51,
no. 7, pp. 99-106, Apr. 2015.
[
33
] Y. Lei, F. Jia, J. Lin, S. Xing, and S. X. Ding, “An intelligent fault
diagnosis method using unsupervised feature learning towards
mechanical big data,” IEEE T. Ind. Electron., Vol. 63, pp. 3137-3147,
Jan. 2016.
[
34
] H. Shao, H. Jiang, H. Zhang, and T. Liang, “Electric locomotive
bearing fault diagnosis using novel convolutional deep belief
network,” IEEE T. Ind. Electron., Vol. 65, pp. 2727-2736, Aug. 2018.
[
35
] G. E. Hinton, and R. R. Salakhutdinov, “Reducing the dimensionality
of data with neural networks,” Science, Vol. 313, no. 5786, pp. 504-
507, Jul. 2006.
[
36
] Y. L. Cun, Y. Bengio, and G. E. Hinton, “Deep learning,” Nature, Vol.
521, no. 28, pp. 436-444, May 2015.
[
37
] P. Tamilselvan, and P. Wang, “Failure diagnosis using deep belief
learning based health state classification,” Reliab. Eng. Syst. Safe.,
Vol. 115, pp. 124-135, Jul. 2013.
[
38
] H. Shao, H. Jiang, X. Zhang, and M. Niu, “Rolling bearing fault
diagnosis using an optimization deep belief network,” Meas. Sci.
Technol., Vol. 26, no. 11, pp. 115002, Nov. 2015.
[
39
] F. Jia, Y. Lei, J. Lin, X. Zhou, and N. Lu, “Deep neural networks: a

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
promising tool for fault characteristic mining and intelligent diagnosis
of rotating machinery with massive data,” Mech. Syst. Signal Process,
Vol. 72-73, pp. 303-315, May 2016.
[
40
] X. Guo, L. Chen, and C. Shen, “Hierarchical adaptive deep
convolution neural network and its application to bearing fault
diagnosis,” Measurement, Vol. 93, pp. 490-502, Nov. 2016.
[
41
] L. Li, “Sparsity-promoted blind deconvolution of ground-penetrating
radar (GPR) data,” IEEE Geosci. Remote Sens. Lett., Vol. 11, no. 8,
pp. 1330-1334, Jan. 2014.
[
42
] H R Li, and S S Gu, “A fast parallel algorithm for a recurrent neural
network,” Acta Automatica Sinica, Vol. 30, no. 4, pp. 516-522, Jul.
2004.
[
43
] J. Jiao, M. Zhao, J. Lin, and J. Zhao, “A multivariate encoder
information based convolutional neural network for intelligent fault
diagnosis of planetary gearboxes,” Knowl-Based Syst., Vol. 160, pp.
237-250, Jul. 2018.
[
44
] S. Pang, X. Yang, X. Zhang, and X. Lin, “Fault diagnosis of rotating
machinery with ensemble kernel extreme learning machine based on
fused multi-domain features,” ISA T., to be published. DOI:
10.1016/j.isatra.2019.08.053.
[
45
] Siri, https://en.wikipedia.org/wiki/Siri (Accessed on 2017/4/20).
[
46
] AlphaGo, https://deepmind.com/research/alphago/ (Accessed on
2017/4/20).
[
47
] A. Graves, A. Mohamed, and G. Hinton, “Speech recognition with
deep recurrent neural networks,” in ICASSP2013, pp.1-5, 2013.
[
48
] K. Yan and D. Zhang, “Correcting instrumental variation and time-
varying drift: a transfer learning approach with autoencoders,” IEEE
T. Instrum. Meas., Vol. 65, no. 9, pp. 2012-2022, Jun. 2016.
[
49
] M. Oquab, L. Bottou, I. Laptev, and J. Sivic, “Learning and
transferring mid-level image representations using convolutional
neural networks,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit.,
2014, pp. 1717-1724.
[
50
] J. Lee, F. Wu, W. Zhao, M. Ghaffari, L. Liao, and D. Siegel,
“Prognostics and health management design for rotary machinery
systems-Reviews, methodology and applications,” Mech. Syst. Signal
Process, Vol. 42, no. 1-2, pp. 314-334, Jan. 2014.
[
51
] Z. Zhou, “Machine learning,” Beijing: Tsinghua University Press.
2016, pp. 113.
[
52
] G.E. Hinton, S. Osindero and Y.-W. Teh, “A fast learning algorithm
for deep belief nets,” Neural Comput., Vol. 18, pp. 1527-1554, July
2006.
[
53
] J. Schmidhuber. “Deep learning in neural networks: an overview,”
Neural Networks., Vol. 61, pp. 85-117, Jan. 2015.
[
54
] C. Tan, F. Sun, T. Kong, W. Zhang, C. Yang, and C. Liu, “A survey on
deep transfer learning,” The 27th ICANN, LNCS 11141, pp. 270-279,
2018.
[
55
] S.J. Pan, and Q. Yang, “A survey on transfer learning, “IEEE Trans.
Knowl. Data Eng., Vol. 22, pp. 1345-1359, 2010.
[
56
] K. Arulkumaran, M. P. Deisenroth, M. Brundage, and A. A. Bharath,
“Deep reinforcement learning: a brief survey,” IEEE Signal Proc.
Mag., Vol. 34, No. 6, pp. 26-38, Nov. 2017.
[
57
] H. Venkateswara, S. Chakraborty, and S. Panchanathan, “Deep-
learning systems for domain adaptation in computer vision,” IEEE
Signal Process. Mag., Vol. 34, no. 6, pp. 117-129, Nov. 2017.
[
58
] G. Hinton, L. Deng, D. Yu, D. George E., M. Abdel-rahman, J.
Navdeep, S. Andrew, V. Vincent, N. Patrick, S. Tara N., and K. Brian,
“Deep neural networks for acoustic modeling in speech recognition:
The shared views of four research groups,” IEEE Signal Process.
Mag., Vol. 29, no. 6, pp. 82-97, Oct. 2012.
[
59
] H. V. Nguyen, H. T. Ho, V. M. Patel and R. Chellappa, “DASH-N:
joint hierarchical domain adaptation and feature learning,” IEEE
Transactions on Image Processing, vol. 24, no. 12, pp. 5479-5491,
Dec. 2015.
[
60
] L. Zhang, W. Zuo and D. Zhang, “LSDT: latent sparse domain
transfer learning for visual adaptation,” IEEE Transactions on Image
Processing, vol. 25, no. 3, pp. 1177-1191, Mar. 2016.
[
61
] Y. Lei, “Intelligent fault diagnosis and remaining useful life prediction
of rotating machinery,” Butterworth-Heinemann. 2016.
[
62
] F. Jia, Y. Lei, L. Guo, J. Lin, S. Xing, “A neural network constructed
by deep learning technique and its application to intelligent fault
diagnosis of machines,” Neurocomputing, Vol. 107, pp. 241-265, Jan.
2018.
[
63
] G. Jiang, H. He, P. Xie and Y. Tang, “Stacked multilevel-denoising
autoencoders: a new representation learning approach for wind
turbine gearbox fault diagnosis,” IEEE T. Instrum. Meas., vol. 66, no.
9, pp. 2391-2402, Sept. 2017.
[
64
] R. Zhang, H. Tao, L. Wu and Y. Guan, “Transfer learning with neural
networks for bearing fault diagnosis in changing working conditions,”
IEEE Access, vol. 5, pp. 14347-14357, Jun. 2017.
[
65
] J. Dai, J. Tang, S. Huang, Y. Wang. “Signal-based intelligent hydraulic
fault diagnosis methods: review and prospects,” Chin. J. Mech. Eng.,
Vol. 32, pp. 75, Dec. 2019.
[
66
] W. Lu, B. Liang, Y. Cheng, D. Meng, J. Yang and T. Zhang, “Deep
Model Based Domain Adaptation for Fault Diagnosis,” IEEE T. Ind.
Electron., vol. 64, no. 3, pp. 2296-2305, Mar. 2017.
[
67
] J. Xie, L. Zhang, L. Duan and J. Wang, “On cross-domain feature
fusion in gearbox fault diagnosis under various operating conditions
based on Transfer Component Analysis,” 2016 IEEE International
Conference on Prognostics and Health Management (ICPHM),
Ottawa, ON, 2016, pp. 1-6.
[
68
] G. Jiang, H. He, J. Yan and P. Xie, “Multiscale Convolutional Neural
Networks for Fault Diagnosis of Wind Turbine Gearbox,” IEEE T. Ind.
Electron., vol. 66, no. 4, pp. 3196-3207, Apr. 2019.
[
69
] F. Chen and M. R. Jahanshahi, “NB-CNN: deep learning-based crack
detection using convolutional neural network and Naïve Bayes data
fusion,” IEEE T. Ind. Electron., vol. 65, no. 5, pp. 4392-4400, May
2018.
[
70
] T. Han, C. Liu, L. Wu, S. Sarkar, and D. Jiang, “An adaptive
spatiotemporal feature learning approach for fault diagnosis in
complex systems,” Mech. Syst. Signal Process, Vol. 117, no. 2, pp.
170-187, Feb. 2019.
[
71
] T. Han, C. Liu, W. Yang, and D. Jiang, “A novel adversarial learning
framework in deep convolutional neural network for intelligent
diagnosis of mechanical faults,” Knowl.-Based Syst., Vol. 165, pp.
474-487, Feb. 2019.
[
72
] Longting Chen, Guanghua Xu, Qing Zhang, and X. Zhang, “Learning
deep representation of imbalanced SCADA data for fault detection of
wind turbines,” Measurement, Vol. 139, pp. 370-379, Jun. 2019.
[
73
] Z. Zhang, S. Li, J. Wang, Y. Xin, and Z. An, “General normalized
sparse filtering: a novel unsupervised learning method for rotating
machinery fault diagnosis,” Mech. Syst. Signal Process, Vol. 124, pp.
596-612, Jun. 2019.
[
74
] Zhang W, Li C, Peng G, and Y. Chen, “A deep convolutional neural
network with new training methods for bearing fault diagnosis under
noisy environment and different working load,” Mech. Syst. Signal
Process., Vol. 100, pp. 439-53, Feb. 2018.
[
75
] J. Pan, Y. Zi, J. Chen, Z. Zhou and B. Wang, “LiftingNet: a novel deep
learning network with layerwise feature learning from noisy
mechanical data for fault classification,” IEEE Trans. Ind. Electron.,
vol. 65, no. 6, pp. 4973-4982, Jun. 2018.
[
76
] X. Ding and Q. He, “Energy-fluctuated multiscale feature learning
with deep ConvNet for intelligent spindle bearing fault diagnosis,”
IEEE Trans. Instrum. Meas., vol. 66, no. 8, pp. 1926-1935, Aug. 2017.
[
77
] D. Lee, V. Siu, R. Cruz, and C. Yetman, “Convolutional neural net
and bearing fault analysis,” in Pro. Int. Conf. Data Mining (p. 194).
The Steering Committee of The World Congress in Computer Science,
Computer Engineering and Applied Computing (WorldComp) (2016,
January).
[
78
] D T Hoang, and H J Kang, “Rolling element bearing fault diagnosis
using convolutional neural network and vibration image,” Cogn. Syst.
Res., Vol. 53, no. 1, pp. 42-50, Jan. 2019.
[
79
] S. Guo, T. Yang, and G. Wei, “A novel fault diagnosis method for
rotating machinery based on a convolutional neural network,”
Sensors, Vol. 18, no. 5, pp. 1429, May 2018.
[
80
] D. Peng, Z. Liu, H. Wang, Y. Qin and L. Jia, “A Novel Deeper One-
Dimensional CNN With Residual Learning for Fault Diagnosis of
Wheelset Bearings in High-Speed Trains,” IEEE Access, vol. 7, pp.
10278-10293, Dec. 2019.
[
81
] Z. Chen and W. Li, “Multisensor Feature Fusion for Bearing Fault
Diagnosis Using Sparse Autoencoder and Deep Belief Network,”

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
IEEE Trans. Instrum. Meas., vol. 66, no. 7, pp. 1693-1702, Jul. 2017.
[
82
] H Liu, J Z Zhou, Y Zheng, W. Jiang, and Y. Zhang, “Fault diagnosis
of rolling bearings with recurrent neural network-based
autoencoders,” ISA T., Vol. 77, pp. 167-178, Jun. 2018.
[
83
] D T Hoang, and H J Kang, “A bearing fault diagnosis method based
on autoencoder and particle swarm optimization-support vector
machine,” Intelligent Computing Theories and Application. Springer,
Cham, 2018.
[
84
] S. Tao, T. Zhang, J. Yang, X. Wang and W. Lu, “Bearing fault
diagnosis method based on stacked autoencoder and softmax
regression,” 2015 34th Chinese Control Conference (CCC),
Hangzhou, 2015, pp. 6331-6335.
[
85
] X. Zhang, X. Zhou, and C. Tang, “A deep convolutional auto-
encoding neural network and its application in bearing fault
diagnosis,” Journal of Xi’an Jiaotong University, 2018.
[
86
] F. Wang, X. Liu, B. Guo, G. Deng, Q. Han, and H. Li, “Application of
kernel auto-encoder based on firefly optimization in intershaft bearing
fault diagnosis,” J. Mech. Eng., Vol. 55, no.7, pp. 58-64, Apr. 2019.
[
87
] B. She, F. Tian, and W. Liang, “Fault diagnosis based on a deep
convolution variational autoencoder network,” Chinese Journal of
Scientific Instrument, Vol. 39, no. 10, pp. 27-35, Oct. 2018.
[
88
] J. Tao, Y. Liu, and D. Yang, “Bearing fault diagnosis based on deep
belief network and multisensor information fusion,” Shock Vib., Vol.
7, pp. 1-9, Jan. 2016.
[
89
] Y. Li, X. Wang, M. Zhang, and H. Zhu, “An approach to fault
diagnosis of rolling bearing using SVD and multiple DBN
classifiers,” Journal of Shanghai Jiaotong University, Vol. 49, no. 5,
pp. 681-686, 694, May 2015.
[
90
] X. Wang, Y. Li, T. Rui, H. Zhu, and J. Fei. “Bearing fault diagnosis
method based on Hilbert envelope spectrum and deep belief
network,” Journal of Vibroeng., Vol. 17, no. 3, 1295-1308, May 2015.
[
91
] M. Gan, C. Wang, and C. Zhu, “Construction of hierarchical diagnosis
network based on deep learning and its application in the fault pattern
recognition of rolling element bearings,” Mech. Syst. Signal Process,
Vol. 72-73, no. 2, pp. 92-104, May 2016.
[
92
] Z. Chen, C. Li, and R. Sanchez, “Multi-layer neural network with
deep belief network for gearbox fault diagnosis,” J. Vibroeng., Vol. 17,
no. 5, pp. 2379-2392, Aug. 2015.
[
93
] Z. Chen, C. Li, and R.V. Sanchez, “Gearbox fault identification and
classification with convolutional neural networks,” Shock and Vib.
Vol. 2015, pp. 1-10, Oct. 2015.
[
94
] G. Liu, H. Bao, and B. Han, “A stacked autoencoder-based deep
neural network for achieving gearbox fault diagnosis,” Mathematical
Problems in Engineering., Vol. 5, pp. 1-10, Jul. 2018.
[
95
] Z. Wang, J. Wang, and Y. Wang, “An intelligent diagnosis scheme
based on generative adversarial learning deep neural networks and its
application to planetary gearbox fault pattern recognition,”
Neurocomputing, Vol. 310, pp. 213-222, Oct. 2018.
[
96
] H. Zheng, R. Wang, and J. Yin, “A new intelligent fault identification
method based on transfer locality preserving projection for actual
diagnosis scenario of rotating machinery,” Mech. Syst. Signal Process,
Vol. 135, pp. 106344, Jan. 2020.
[
97
] C. Li, R.-V. Sanchez, G. Zurita, M. Cerrada, D. Cabrera, and R. E.
Vásquezc, “Multimodal deep support vector classification with
homologous features and its application to gearbox fault diagnosis,”
Neurocomputing, Vol. 168, pp. 119-127, Nov. 2015.
[
98
] Z. He, H. Shao, J. Cheng, X. Zhao, and Y. Yang, “Support tensor
machine with dynamic penalty factors and its application to the fault
diagnosis of rotating machinery with unbalanced,” Mech. Syst. Signal
Process, to be published. DOI: 10.1016/j.ymssp.2019.106441.
[
99
] L Wen, X Y Li, L Gao, and Y. Zhang, “A new convolutional neural
network based data-driven fault diagnosis method,” IEEE Trans. Ind.
Electron., Vol. 65, no. 11, pp. 5990-5998, Nov. 2018.
[
100
] K. Li, and Q. Wang, “Study on signal recognition and diagnosis for
spacecraft based on deep learning method,” 2015 Prognostics and
System Health Management Conference (PHM). IEEE, 2015.
[
101
] Y. H. Eom, J. W. Yoo, and S. B. Hong, “Refrigerant charge fault
detection method of air source heat pump system using convolutional
neural network for energy saving,” Energy, Vol. 187, no. 15, pp.
115877, Nov. 2019.
[
102
] C. Lanham, “Understanding the tests that are recommended for
electric motor predictive maintenance,” Baker Instrument Company.
2002.
[
103
] S. Nandi, H.A. Toliyat, and X. Li, “Condition monitoring and fault
diagnosis of electrical motors-a review,” IEEE Trans. Energy Convers.
Vol. 20, no. 4, pp. 719-729, Dec. 2005.
[
104
] F. Jia, Y. Lei, N. Lu, and S. Xing, “Deep normalized convolutional
neural network for imbalanced fault classification of machinery and
its understanding via visualization,” Mech. Syst. Signal Process, Vol.
110, pp. 349-367, Sept. 2018.
[
105
] K. Weiss, T.M. Khoshgoftaar, and D. Wang, “A survey of transfer
learning,” J. Big Data., Vol. 3, pp. 9-49, May 2016.
[
106
] H. Ren, W. Liu, and M. Shan, and X. Wang, “A new wind turbine
health condition monitoring method based on VMD-MPE and
feature-based transfer learning,” Measurement, Vol. 148, pp. 1-8, Dec.
2019.
[
107
] W. Qian, S. Li, and X. Jiang, “Deep transfer network for rotating
machine fault analysis,” Pattern Recognition, Vol. 96, pp. 1-14, Dec.
2019.
[
108
] Z. Wu, H. Jiang, K. Zhao, and X. Li, “An adaptive deep transfer
learning method for bearing fault diagnosis,” Measurement, Vol. 151,
pp. 1-14, Feb. 2020.
[
109
] L. Guo, Y. Lei, S. Xing, T. Yan and N. Li, “Deep convolutional
transfer learning network: a new method for intelligent fault diagnosis
of machines with unlabeled data,” IEEE Trans. Ind. Electron., vol. 66,
no. 9, pp. 7316-7325, Sept. 2019.
[
110
] B. Yang, Y. Lei, F. Jia, and S. Xing, “An intelligent fault diagnosis
approach based on transfer learning from laboratory bearings to
locomotive bearings,” Mech. Syst. Signal Process, Vol. 122, pp. 692-
706, May 2019.
[
111
] L. Wen, L. Gao and X. Li, “A new deep transfer learning based on
sparse auto-encoder for fault diagnosis,” IEEE T. Syst. Man Cy.-S.,
vol. 49, no. 1, pp. 136-144, Jan. 2019.
[
112
] M. He, and D. He, “A new hybrid deep signal processing approach
for bearing fault diagnosis using vibration signals,” Neurocomputing,
to be published, DOI: 10.1016/j.neucom.2018.12.088.
[
113
] W. Mao, W. Feng, and X. Liang, “A novel deep output kernel
learning method for bearing fault structural diagnosis,” Mech. Syst.
Signal Process, Vol. 117, pp. 293-318, Feb. 2019.
[
114
] H. Shao, H. Jiang, H. Zhang, W. Duan, T. Liang, and S. Wu,
“Rolling bearing fault feature learning using improved convolutional
deep belief network with compressed sensing,” Mech. Syst. Signal
Process, Vol. 100, pp. 743-765, Feb. 2018.
[
115
] M. Sun, H. Wang, P. Liu, S. Huang, and P. Fan, “A sparse stacked
denoising autoencoder with optimized transfer learning applied to the
fault diagnosis of rolling bearings,” Measurement, Vol. 146, pp. 305-
314, Nov. 2019.
[
116
] F. Xu, W. P. Tse, and Y. L. Tse, “Roller bearing fault diagnosis using
stacked denoising autoencoder in deep learning and Gath-Geva
clustering algorithm without principal component analysis and data
label,” Appl. Soft Comput., Vol. 73, pp. 898-913, Dec. 2018.
[
117
] Z. Gao, S.X. Ding, and C. Cecati, “Real-time fault diagnosis and
fault-tolerant control,” IEEE Trans. Ind. Electron., Vol. 62, no. 6, pp.
3752-3756, May 2015.
[
118
] Z. Chen, K. Gryllias, and W. Li, “Mechanical fault diagnosis using
convolutional neural networks and extreme learning machine,” Mech.
Syst. Signal Process, vol. 133, pp. 106272, Mar. 2019.
[
119
] C. Shen, D. Wang, F. Kong, and P. W. Tse, “Fault diagnosis of
rotating machinery based on the statistical parameters of wavelet
packet paving and a generic support vector regressive classifier,”
Measurement, Vol. 46, no. 4, 1551-1564, May 2013.
[
120
] T. Han, Liu C, Yang W, and D. Jiang, “Deep transfer network with
joint distribution adaptation: a new intelligent fault diagnosis
framework for industry application,” ISA T. 2018.
[
121
] W. Qian, S. Li, P. Yi, K. Zhang, “A novel transfer learning method
for robust fault diagnosis of rotating machines under variable working
conditions,” Measurement, vol. 138, pp. 514-525, May 2019.
[
122
] F. Li, X. Pang, and Z. Yang, “Motor current signal analysis using
deep neural networks for planetary gear fault diagnosis,”
Measurement, Vol. 145, pp. 45-54, Oct. 2019.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
[
123
] Yong L, Gang C, Chang L, and X. Chen, “Study on planetary gear
fault diagnosis based on variational mode decomposition and deep
neural networks,” Measurement, Vol. 130, pp. 94-104, Dec. 2018.
[
124
] S. Ma, F. Chu, and Q. Han, “Deep residual learning with
demodulated time-frequency features for fault diagnosis of planetary
gearbox under nonstationary running conditions,” Mech. Syst. Signal
Process, Vol. 127, pp. 190-201, Jul. 2019.
[
125
] G. Qiu, Y. Gu, and Q. Cai, “A deep convolutional neural networks
model for intelligent fault diagnosis of a gearbox under different
operational conditions,” Measurement, Vol. 145, pp. 94-107, Oct.
2019.
[
126
] H. Shao, H. Jiang, H.i Zhao, and F. Wang, “A novel deep
autoencoder feature learning method for rotating machinery fault
diagnosis,” Mech. Syst. Signal Process, Vol. 95, pp. 187-204, Oct.
2017.
[
127
] K. Zhang, B. Tang, Y. Qin, and L. Deng, “Fault diagnosis of
planetary gearbox using a novel semi-supervised method of multiple
association layers networks,” Mech. Syst. Signal Process, Vol. 131,
pp. 243-260, Sept. 2019.
[
128
] Y. Zhu, P. Qian, S. Tang, and W. Jiang, “Amplitude-frequency
characteristics analysis for vertical vibration of hydraulic AGC
System under nonlinear action,” AIP Advances, Vol. 9, no. 3, pp.
035019, Mar. 2019.
[
129
] Y. Zhu, S. Tang, C. Wang, W. Jiang, J. Zhao, and G. Li, “Absolute
stability condition derivation for position closed-loop system in
hydraulic automatic gauge control,” Processes, Vol. 7, no. 10, pp. 766,
Oct. 2019.
[
130
] Y. Zhu, S. Tang, C. Wang, W. Jiang, X. Yuan, and Y. Lei,
“Bifurcation characteristic research on the load vertical vibration of a
hydraulic automatic gauge control system,” Processes, Vol. 7, no. 10,
pp. 718, Oct. 2019.
[
131
] J. Du, S Wang, and H Zhang, “Layered clustering multi-fault
diagnosis for hydraulic piston pump,” Mech. Syst. Signal Process, Vol.
36, no. 2, pp. 487-504, Apr. 2013.
[
132
] J. Ma, H. Li, and B. Xu. “Fault model identification of hydraulic
pump based on mlse and extension theory,” Machine tool &
hydraulics, Vol. 43, no. 11, pp. 182-187, Jun. 2015.
[
133
] S. Kumar, and J.M. Bergada, “The effect of piston grooves
performance in an axial piston pumps via CFD analysis,” Int. J. Mech.
Sci., vol. 66, pp. 168-179, Jan. 2013.
[
134
] S. Kumar, J.M. Bergada, and J. Watton, “Axial piston pump grooved
slipper analysis by CFD simulation of three dimensional NVS
equation in cylindrical coordinates,” Comput. Fluids, Vol. 38, no. 3,
pp. 648-663, Mar. 2009.
[
135
] L. Wen, X. Li, L. Gao and Y. Zhang, “A New Convolutional Neural
Network-Based Data-Driven Fault Diagnosis Method,” IEEE Trans.
Ind. Electron., vol. 65, no. 7, pp. 5990-5998, Jul. 2018.
[
136
] Yan J, Zhu H, Yang X, Y. Cao, “Research on fault diagnosis of
hydraulic pump using convolutional neural network,” J. Vibroeng.,
Vol. 18, no. 8, pp. 5141-5151, Dec. 2016.
[
137
] C. Lu, Y. Wang, M. Ragulskis, and Y. Cheng, “Fault diagnosis for
rotating machinery: a method based on image processing,” PLoS
ONE, 11(10): e0164111. Oct. 2016.
[
138
] S. Wang, J. Xiang, Y. Zhong, and H. Tang, “A data indicator-based
deep belief networks to detect multiple faults in axial piston pumps,”
Mech. Syst. Signal Process, Vol. 112, pp. 154-170, Nov. 2018.
[
139
] Y. Li, Y. Ya ng, G. Li, M. Xu, and W. Huang, “A fault diagnosis
scheme for planetary gearboxes using modified multi-scale symbolic
dynamic entropy and mRMR feature selection,” Mech. Syst. Signal
Process, Vol. 91, pp. 295-312, Jul. 2017.
[
140
] Y. Yang, H. Zheng, Y. Li, M. Xu, and Y. Chen, “A fault diagnosis
scheme for rotating machinery using hierarchical symbolic analysis
and convolutional neural network,” ISA T., Vol. 91, pp. 235-252, Aug.
2019.
[
141
] Wang H, Li S, Song L, and L. Cui, “A novel convolutional neural
network based fault recognition method via image fusion of multi-
vibration-signals,” Comput. Ind., Vol. 105, pp. 182-190, Feb. 2019.
[
142
] Z. Qi, B. Wang, Y. Tian, and Peng. Zhang, “When ensemble learning
meets deep learning: A new deep support vector machine for
classification,” Knowl.-Based Syst., Vol. 107, pp. 54-60, Sept. 2016.
[
143
] L. C. Jiao, J. Zhao, and S. Y. Yan, “Deep learning, optimization and
recognition,” Beijing, China: Tsinghua Univ. Press, 2017, pp. 116-117.
[
144
] K. Li, R. Zhang, F. Li, and L. Su, “A new rotation machinery fault
diagnosis method based on deep structure and sparse least squares
support vector machine,” IEEE Access, Advances in Prognostics and
System Health Management. Vo. 7, pp. 26571-26580, Feb. 2019.
[
145
] Y. Lei, W. Jiang, A. Jiang, and Y. Zhu, “Fault diagnosis method for
hydraulic directional valves integrating PCA and XGBoost,”
Processes, Vol. 7, no. 9, pp. 589, Sept. 2019.
[
146
] Z. Zheng, Z. Wang, Y. Zhu, S. Tang and B. Wang, “Feature
extraction method for hydraulic pump fault signal based on improved
empirical wavelet transform,” Processes, Vol. 7, no. 11, pp. 824, Nov.
2019.
[
147
] C. Wang, X. Chen, N. Qiu, Y. Zhu, and W. Shi, “Numerical and
experimental study on the pressure fluctuation, vibration, and noise of
multistage pump with radial diffuser,” J. Braz. Soc. Mech. SCI., Vol.
40, no. 10, pp. 481, 2018.
[
148
] S. Tang, Y. Zhu, W. Li, and J. Cai, “Status and prospect of research
in preprocessing methods for measured signals in mechanical
systems,” Journal of Drainage and Irrigation Machinery Engineering,
Vol. 37, no. 9, pp. 822-828, 2019.
[
149
] Y. Zhu, S. Tang, L. Quan, W. Jiang, and L. Zhou, “Extraction method
for signal effective component based on extreme-point symmetric
mode decomposition and Kullback-Leibler divergence,” J. Braz. Soc.
Mech. SCI., Vol. 41, no. 2, pp. 100, 2019.
[
150
] J. Pei, F. Zhang, D. Appiah, and B. Hu, “Performance prediction
based on effects of wrapping angle of a side channel pump,” Energies,
Vol. 12, no. 1, pp. 139, Jan. 2019.
[
151
] Z. Liu, W. Jiang, W. Tan, and Y. Zhu, “Fault identification method
for hydraulic pumps based on multi-feature fusion and multiple
kernel learning SVM,” Chin. Mech. Eng., Vol. 27, no. 24, pp. 3355-
3361, Dec. 2016.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2019.2963092, IEEE Access
VOLUME XX, 2017 9
SHENGNAN TANG received her master degree
in Yanshan University, Qinhuangdao, China, in
2013. Now she is a doctoral student in Jiangsu
University. She has participated in more than 10
research projects at provincial or ministerial
levels. She has authored or co-authored more than
10 publication papers. Her current research
interests include intelligent fault diagnosis based
on deep learning and their application in rotating
machinery.
SHOUQI YUAN received the master degree
and Ph.D. degree in Jiangsu institute of
technology (present Jiangsu university),
Zhenjiang, China, in 1990 and 1995,
respectively. He visited Cranfield University
from 1995 to 1996. He is currently a Professor
and academic leader of National Key Discipline
Fluid Machinery and Engineering. He has
presided and completed more than 30 projects
including National Science Fund for
Distinguished Young Scholars, National Natural
Science Foundation and National 863 project. He has authored or co-
authored more than 300 publication papers, and 13 books, and holds more
than 11 PCT patents granted, including 2 U.S. patents. His research area
mainly focused on fluid machinery and engineering. He is a member of the
IEEE ITS and RS and a Senior Member of IET. He received the 2 National
Prize for Progress in Science and Technology and 2 second prize of
national teaching achievement.
YONG ZHU received the master degree and
Ph.D. degree in Yanshan University,
Qinhuangdao, China, in 2013 and 2017,
respectively. Since 2017, he has been an Assistant
Professor and master's supervisor in Jiangsu
University. He has presided over and participated
in more than 20 research projects at provincial or
ministerial levels. He has authored or co-authored
more than 50 publication papers, and 2 books,
and holds more than 20 patents granted. He
received the 1 Science and Technology Progress
Award of the Ministry. His research area mainly focused on prognostics
and health management, intelligent information processing and fault
diagnosis of rotating machinery.