IFTCS

Expert Systems With Applications 238 (2024) 121956
Contents lists available at ScienceDirect
Expert Systems With Applications

journal homepage: www.elsevier.com/locate/eswa
Review
Development of Intelligent Fault-Tolerant Control Systems with Machine

Learning, Deep Learning, and Transfer Learning Algorithms: A Review
Arslan Ahmed Amin *, Muhammad Sajid Iqbal, Muhammad Hamza Shahbaz
Department of Electrical Engineering, FAST National University of Computer and Emerging Sciences, Chiniot Faisalabad Campus, Punjab, Pakistan
A R T I C L E I N F O A B S T R A C T
Keywords: Intelligent Fault-Tolerant Control (IFTC) refers to the applications of machine learning algorithms for fault
Fault diagnosis diagnosis and design of Fault-Tolerant Control (FTC). The overall goal of the FTC is to accommodate defects in
Intelligent fault-tolerant control the system components while they are in use and maintain stability with little to no performance reduction.
Algorithms
These systems are crucial for mission-critical and safety-related applications where the safety of people is at stake
Machine learning
and service continuity is crucial. In this review paper, a systematic study has been done for the development of
Deep learning
Transfer learning FTC with machine learning, deep learning, and transfer learning algorithms. The challenges and limitations faced
with their possible solutions through machine learning theories for the IFTC model are lined up. This paper
guides researchers on the different possible types of machine learning algorithms and their advanced forms like
deep learning and transfer learning. The differences among these are highlighted by the challenges and limi
tations of each. The paper is significant such that most of the important literature references from the Scopus
database particularly related to important electrical and mechanical industrial problems have been discussed to
guide the researchers who want to apply IFTC for specific industrial problems, being the research gap. Finally,
future research directions for the development of IFTC are highlighted.
1. Introduction control systems must be constructed with the capacity to tolerate any
errors. Fault-tolerant control (FTC) is a common term for this kind of
Faults in any real system are conceivable and impair the system’s control system. Specifically, FTC refers to control systems that can
stability and performance. For today’s technological systems to conform automatically withstand the loss of individual components. In the case of
to higher performance and safety criteria, sophisticated control systems such failures, they are able to keep the system stable and operating at a
are required. When actuators, sensors, or other system components fail, tolerable level. A fault-tolerant control system can recover from the loss
the performance of a complex system with a conventional feedback of a single component without compromising the system’s ability to
control design might degrade or become unstable. New methods for operate or remain stable overall (Amin & Mahmood-ul-Hasan, 2019b;
constructing control systems have been developed to address these Amin & Mahmood-Ul-Hasan, 2019).
problems and preserve the system’s ideal performance and stability In the first part, the standard FTC strategy and operational specifics
qualities while allowing for component failures (Amin & Hasan, 2019; are outlined briefly. FTC is divided into two primary categories: active
Amin & Mahmood-ul-Hasan, 2019a; X. Wang, 2020). Satellites, nuclear and passive. In an active fault-tolerant control system (AFTCS), the Fault
power facilities, chemical processing factories, and airplanes all use this Detection and Isolation (FDI) unit performs the primary role of detect
to ensure the safety of their passengers and crew. Any problem at all ing, localizing, and isolating problematic components on the device. In
with a single component of such a system might have catastrophic re the FDI unit, the algorithm employs the observer principle such that the
sults. Therefore, there is often a high need for robust fault tolerance, plant parameter is compared to a predetermined normal value to
safety, and dependability (W. J. Zhang & Wang, 2016). To increase generate a residual (Ahmed, Amin, Wajid, & Ahmad, 2020). No faults
reliability and availability while attaining the desired performance, are reported by the machine if the residual is within acceptable
Abbreviations: IFTC, Intelligent Fault Tolerant Control; FTC, Fault Tolerant Control; DNN, Deep Neural Network; KNN, K-Nearest Neighbor; AE, Auto Encoder;
RBM, Restricted Boltzmann Machine; ANN, Artificial Neural Network; DBM, Deep Belief Network; TCA, Transfer Component Analysis; IC Engine, Internal Com
bustion Engine; JDA, Joint Distribution Adaption; GAN, Generative Adversarial Network; TDA, Transfer Denoising Auto-encoder.
* Corresponding author.
E-mail address: [email protected] (A.A. Amin).
https://doi.org/10.1016/j.eswa.2023.121956
Received 21 February 2023; Received in revised form 30 September 2023; Accepted 1 October 2023
Available online 6 October 2023
0957-4174/© 2023 Elsevier Ltd. All rights reserved.
A.A. Amin et al. Expert Systems With Applications 238 (2024) 121956
parameters. If it is judged that the residual has exceeded the set limit, an number of system failures eventually surpassed the system’s intended
FDI unit will proclaim defective conditions. The controller is changed to failure tolerance (Bernier, Ortega, Ros, Rojas, & Prieto, 2000; H. Saltzer
accommodate new operating specifications. A decline in performance & Kaashoek, 2021).
due to faulty components is possible, but system dependability is assured The trustworthiness and accessibility of a system under specified
(S. Han et al., 2020). circumstances are what we mean when we talk about its reliability. The
The conventional AFTCS working can be demonstrated in state-space rate of component failure, the length of time the system is in operation,
as shown in the analysis by Wang et al. (Youqing Wang, Zhou, Qin, & and the conditions in which it is utilized all affect reliability. To be
Wang, 2008) to explain the architecture of the observer, reliable, a system or component must have a high chance of functioning
well for an established amount of time under specified operating con
ẋ = Ax + Bu (1)
ditions. The percentage of tests that fail is a major consideration in
reliability analyses. The rate of failure is the expected number of
y = Cx + Du (2)
breakdowns per given amount of time. It is often represented as. If a part
where, A, B, C, and D are system matrices of appropriate dimensions, breaks after 1000 h of use, its failure rate is 1x10-3 per hour, or λ = 1/
ẋ is state space input, and y is output. 1000.
The reliability function would be exponential, as shown in the graph
ẋ = Ax + Bu (3)
below, and the failure rate (λ) would be constant during the whole useful
lifetime, at the curve’s midpoint (Dubrova, 2013; Peebles, 2002).
y = Cx + Du (4)
x and y are estimated input and output produced by the observer. R(t) = e− λt
(13)
The exponential failure law is a useful tool for calculating dynamic
(ẋ − ẋ) = A(x − x) (5)
reliability functions.
(y − y) = C(x − x) (6) Mean time to failure (MTTF) is a crucial quantitative indicator of a
system’s dependability. The mean time between failure shows how long
ẋ = Ax + Bu + L(y − y) (7) an item typically lasts. It’s the average amount of time that a product is
expected to last before it stops working. Modeling assumptions for MTTF
In (7), L is the feedback gain. include the possibility that the faulty item cannot be repaired and an
interminable repair period. The MTTF may be calculated using the
ẋ − ẋ = A(x − x) + L(y − y) (8)
following formula:
(y − y) = C(x − x) (9) ∫∞
MTTF = R(t)dt (14)
0
ẋ − ẋ = (A + LC)(x − x) (10)
Alternatively, the reciprocal of the failure rate (λ) may be used to get
ėx = (A + LC)ex (11) MTTF.
1
(y − y) = Cex (12) MTTF = (15)
λ
No fault is declared by FDI when the residual “ex ” goes to zero. If the For systems in which individual components are beyond repair, the
residual is removed, an error will be declared and the estimated FDI unit Mean Time to Failure (MTTF) is determined.
will be replaced. AFTCS’s complex structure and sluggish reaction due to The Mean Time to Repair (MTTR) is a simple metric used to rank the
excessive computations reflect major disadvantages. repair ability of products. It’s a measure of how long it typically takes to
In a conventional passive fault-tolerant control system (PFTCS) no fix a broken gadget or part. A thorough explanation of these factors,
FDIs are inserted and all the failures in the design process are pre- including the current level of knowledge along with its limits, and a
configured (Du & Li, 2023). The PFTCS operates with offline param specific nuclear power plant application, is provided.
eter adjustment design architecture and due to fewer computations, has Mean time to repair (MTTR) is an important metric for gauging the
a quicker response than AFTCS (Amin & Mahmood-ul-Hasan, 2021). reliability of repairable goods. It’s the average amount of time needed to
However, faults considered in the design process can only be tolerated. fix anything that’s broken. Detailed discussions of these characteristics,
Sliding mode control (SMC) is one of the most popular PFTCS techniques as well as a look at the current state of the art’s limits and an example
due to its advantages of robustness to external disturbances, system nuclear power plant application scenario, may be found (Kumar, Singh,
parameter variations, and model uncertainties (Shahbaz & Amin, 2023). & Tripathi, 2018). In (H. Li & Zhao, 2005), a semi-Markov process model
The hybrid features of these two techniques make each system and reliability function computation for FTCs is presented. In (Abdel
outstandingly suited to a hybrid fault-tolerant control system (HFTCS) salam, Diab, Tennakoon, & Griffiths, 2016), it is proposed that the
(Alsuwian, Riaz, et al., 2022; Alsuwian, Tayyeb, et al., 2022; Amin & reliability of multilayer modular converters might be improved by using
Mahmood-ul-Hasan, 2022). In the HFTCS, the system can react quickly FTC techniques in a bulk power transmission situation. In both good and
to defects by using the PFTCS feature in safety applications and then dysfunctional settings, there is a trend towards increased dependence.
optimize using the AFTCS property later. Differential equations and the study of dependability have both
The quantitative measure of fault tolerance will determine how benefited from the use of Markov chains. In (HONGYAN DUI, 2015), a
much failure a system can tolerate. One such metric is the number of reliability analysis is conducted by breaking the system’s lifecycle into
component failures a system can withstand before it fails itself, often stages and identifying potential failure points. To build automated
known as failure tolerance. The system’s fault tolerance decreases after a control systems and optimize these schemes for generally higher system
failure. While in operation, the system’s remaining failure tolerance lifespan and dependability, it is necessary to conduct important analyses
should be tracked to determine how near it is to crashing altogether. of these basic components and subcomponents.
When fault tolerance is implemented without a system to track how Resilience is the ability to lessen the severity and frequency of fail
much of it has been utilized, the concept of a safety margin is ignored. ures rather than eliminate them. Resilient systems are robust and can
This is a classic case of a system design flaw. When fault-tolerant systems recover from and adapt to a variety of different types of failures,
fail, the ensuing investigation always reveals several problems that were including malicious attacks, network outages, hardware problems, and
effectively hidden by the system but were never reported or fixed. The software flaws (Cassottana, Roomi, Mashima, & Sansavini, n.d.).
2
According to Jicheng Chen and Shi (2021), the Cyber-Physical System shown in Fig. 1.
(CPS) is seen as the central feature of next-generation control systems of Machine learning theories got the attention of many researchers and
the Industrial 4.0 era because of its thorough integration of physical engineers from industry (Khan & Yairi, 2018; R. Zhao et al., 2019).
systems, online computing, networking, and control. A reliable, stable, Algorithms like artificial neural networks (ANN) (Schmidhuber, 2015),
and resilient CPS is essential for fortification and security. Developing support vector machines (SVM) (Rezvani, Wang, & Pourpanah, 2019), k-
CPSs is complicated by security concerns because of the system’s Nearest Neighbor (kNN) (Kim, Choi, Park, Leung, & Nasridinov, 2021),
vulnerability to external factors such as physical limitations, unpre and probabilistic graphical model (PGM) (Savchynskyy, 2019) evolved
dictability, and malicious attacks. One technique to offer resilience with the research in machine learning theory (Yadav, Shukla, & Singh,
against cyberattacks and deliver robustness in the face of stochastic 2021). With a focus on FTC, researchers used an expert system-based
uncertainty is to integrate the resilient mechanism with the stochastic approach, genetic algorithm-based approach (Iqbal & Amin, 2022),
model predictive control (MPC) framework. The study (F. Wang, Qian, SVM-based approach (Araste, Sadighi, & Jamimoghaddam, 2023),
Yan, Yuan, & Zhang, 2020) describes a resilient robot prototype that can particle swarm optimization-based approach (Alsuwian, Iqbal, et al.,
continue operating after being damaged and was termed an R-Robot II. 2022), ANN-based approach (Heo & Lee, 2018), and intelligent
In this study, a closed-loop, passive-joint under-actuated robust robot approach (LeCun, Bengio, & Hinton, 2015) to model FTC. These ap
was examined and described. proaches learned the faults of the machine by using artificially extracted
The objective of fault tolerance is to mitigate system failures or features. The artificially extracted features are based on digitally
sustain function in the presence of defects by using redundancy and collected data. In the next step, the FTC model is trained, which results
error correction mechanisms. The primary emphasis is placed on in the automatic detection of the health level of the machine.
continuous operation. On the other side, resilience is defined as the Many deep learning algorithms have emerged and refined the FTC
capacity to recover quickly from failures and resume functioning nor modeling. The research in neural networks gives the concept of deep
mally. Resilience is the ability of a system to adapt to new circumstances learning in the 1980s (Goodfellow, Bengio, & Courville, 2016). Some
and recover quickly from failures. Resilience guarantees quick recovery relatable examples include autoencoders (AE), the restricted Boltzmann
and continuous service after unexpected occurrences, whereas fault machine (RBM), and deep belief network (DBN) (Dogan & Birant, 2021)
tolerance limits disruptions. Resilience and fault tolerance both improve algorithms. Furthermore, convolutional neural networks (CNN) make
system reliability by lessening the effect of failures and preventing them their way into the field through AlexNet (Krizhevsky, Sutskever, &
from occurring (W. J. Zhang & Lin, 2010). Hinton, 2017) and ResNet (K. He, Zhang, Ren, & Sun, 2016). These al
gorithms used the idea of self-learning the fault instead of artificial
2. Concept of intelligent fault-tolerant control feature selection. This further narrows down the involvement of human
beings. Moreover, models based on these algorithms aim for an end-to-
Machine learning is a branch of Artificial Intelligence (AI), which is end diagnostic system.
often characterized as a machine’s capacity to replicate intelligent Transfer learning theories are focused on designing the FTC model
human behavior. AI systems are used to do complex tasks in a way for engineering applications. The undoubtful success of deep learning-
similar to how humans solve problems. The purpose of artificial intel based FTC models has a drawback within it. There is an assumption
ligence is to create computer models with “intelligent behaviors” akin to that the data is always properly labeled. Yet this assumption fails when it
those of humans. This refers to machines that can interpret a natural comes to real-life engineering problems (K. He et al., 2016). The first
language text, recognize a visual scene, or perform a physical activity. factor to start with is that machines are working under normal condi
Data, whether it is numbered, photos, or text, is the basis of machine tions and faults happen minimally. Hence most of the collected data
learning. Bank transactions, photographs of persons or even particular belongs to a healthy state of the machine, which is of less use. The
bakery items, repair records, time series data from sensors, and sales second factor is the unavailability of machine health data at a low price.
reports are examples of data. The data that has been gathered and pre Hence, the labeling required by the deep learning models results in
pared for use as training data will be used to train a machine-learning higher costs. Moreover, this unavailability of the labeled data within the
model. More data enhances the application’s performance. engineering problem scope makes the deep learning models unfit. Also,
In order to extract and transform features, deep learning algorithms the data collected is majorly from the healthy status of the machine
employ a cascade of numerous layers of nonlinear processing units. Each hence it is not possible to train the FTC model with real unhealthy data.
new layer learns numerous levels of interpretations that correspond to In this regard, transfer learning theory surfaces with the possibility of
various levels of abstraction; the layers constitute a hierarchy of con solving practical engineering problems. Within transfer learning algo
cepts. Each new layer utilizes the output from the preceding layer as rithms, the knowledge learned from one task or multiple tasks can be
input. In a more precise way, deep learning can be defined as: “Deep transferred to related ones having a novel issue (Bin Yang, Lei, Jia, &
learning is a technique that provides us not just the relationship between Xing, 2019). The concept of transfer learning goes back to 1995 when it
two or more variables, but also the rules that govern that relationship as was introduced by the name of the learning-to-learn technique (S. J. Pan
well as the logic behind it” (W. J. Zhang, Yang, Lin, Ji, & Gupta, 2018). & Yang, 2010) and has had sufficient improvements since the 2010s.
Autonomous vehicles employ deep learning as a crucial technology to Some known algorithms are transferred component analysis (TCA)
detect stop signs and differentiate between a person and a lamppost. At (Matasci, Volpi, Kanevski, Bruzzone, & Tuia, 2015), joint distribution
its most fundamental level, deep learning may be understood as an adaption (JDA) (Sinno Jialin Pan, Tsang, Kwok, & Yang, 2011), and
automated kind of predictive analytics. Deep learning algorithms are TrAdaboost (Long, Wang, Ding, Sun, & Yu, 2013). After 2015, transfer
arranged hierarchically with ever greater degrees of complexity and learning is combined with deep learning to further apply it in the field of
abstraction than linear machine learning algorithms. Transfer learning computer vision (W. Dai, Yang, Xue, & Yu, 2007) through transfer
is a strategy for machine learning in which a model developed for one denoising auto-encoder (TDA) (Venkateswara, Chakraborty, & Pan
task serves as the foundation for another. Pre-trained models are chanathan, 2017) and joint distribution adaption (JDA) (Long, Wang,
commonly used as the basis for deep learning tasks in computer vision Cao, Sun, & Yu, 2016). The literature wisdom further indicates the
and natural language processing since they save time and money when application of feature-based approaches (Bin Yang et al., 2019), gener
compared to constructing neural network models from scratch and ative adversarial network (GAN) based approaches (Long, Zhu, Wang, &
perform considerably better on related tasks (T. Ahmad, Madonski, Jordan, 2017), instance-based approaches (L. Guo, Lei, Xing, Yan, & Li,
Zhang, Huang, & Mujeeb, 2022; “Machine Learning, Explained | MIT 2019), and parameter-based approaches (Fei Shen, Chao Chen, Yan, &
Sloan,” 2022; Oroojlooy & Hajinezhad, 2022; Z. Yu, Zhang, Jiang, Fu, & Gao, 2015) being used by the researchers working on developing the
Jin, 2022). The overall hierarchy and classification of algorithms are FTC models.
3
Fig. 1. Hierarchy and classification of intelligent fault diagnosis algorithms.
Our contribution to this review paper is a methodical investigation of

TABLE 1
the creation of FTC using transfer learning, deep learning, and machine
Search Strings and Research Articles found in Scopus.
learning techniques. The IFTC model’s difficulties and restrictions are
listed, along with any potential solutions offered by machine learning Sr. Search String Number of Papers Selected
No. Conference Journal Total for
theories. The main types of machine learning algorithms and their
Papers Papers Review
advanced versions, such as deep learning and transfer learning, are
1 Fault Tolerant 21 24 45 10
explained in this paper to researchers. The difficulties and restrictions of
Control AND
each bring out the distinctions between them. In order to help re Support Vector
searchers who want to use IFTC for particular industrial problems, the Machine
study is notable in that the majority of relevant literature references, in 2 Fault Tolerant 107 143 250 16
particular those connected to significant electrical and mechanical in Control AND KNN
dustrial difficulties, have been presented. Last but not least, prospective Control AND
research focuses for IFTC development are mentioned. Probabilistic
In order to conduct this review, we have used the Scopus database to Graphical Modelling
find relevant research papers. Different search strings were used to find 4 Fault Tolerant 10 8 18 10
Control AND
the relevant research articles from the last ten years (2013 – 2023). The
Decision Tree
details of the search strings applied, and the results found on the Scopus 5 Fault Tolerant 44 101 145 19
database for the above-mentioned years are shown in Table 1 and their Control AND
distribution is shown in Fig. 2. Artificial Neural
The articles were divided into two categories against each search Network
string conference paper and journal paper. We have found a total of
Control AND
1020 articles related to our review topic. Most relevant articles were Convolutional
then further reduced according to the following criteria. Neural Network
Control AND ResNet
• The redundant papers were removed from the obtained results.
• More preference was given to journal articles based on the publica Control AND
tion date, journal type, and relevance to the topic. Moreover, the Transfer Learning
recent article is selected from an author if there were more than 1 9 Fault Tolerant 55 153 208 12
article from the same author. Control AND Fuzzy
Logic
• An article with a greater number of citations is preferred.
• Some conference papers were also included in the review. The papers Control AND Auto
from the conferences with a long sequence were preferred. Encoder
4
Fig. 2. Distribution of Conference and Journal Papers in Scopus Database Against Search Strings.
• Apart from Machine learning, deep learning, and transfer learning, (2016a), digital signal processing techniques are engaged by researchers
some papers related to feature extraction, feature selection, data to precisely locate the fault or to categorize the fault. Within this section,
collection, and health state estimation are also included in this the implementation, and the process of FTC development using classical
review. machine learning algorithms are focused on. The detailed three-step
• Some well-known research papers and books related to mathemat process for the intelligent fault diagnosis is shown in Fig. 3. These
ical modeling and standard definitions are also included in the steps are discussed in detail in the following subsections.
references.
4. Data collection and feature extraction
Further contents of the paper have been arranged in the following
sections. Section 3 covers machine learning algorithms; section 4 de The very first step for the development of FTC is data collection. Data
scribes the deep learning algorithms and section 5 depicts the transfer is collected through sensors. These sensors collect data in the form of
learning algorithms used for the IFTC development. The discussion and acoustics, vibrations, current readings, and temperature values. Nor
future research directions are described in section 6. Finally, the mally vibration data and acoustic data are connected to bearings (Qiao,
conclusion of the study is presented. Lei, & Li, 2019; Rai & Upadhyay, 2016; Yanxue Wang, Xiang, Markert, &
Liang, 2016b), gears, and gearbox faults (Li et al., 2016; Wong, Zhong,
3. Machine learning algorithms in FTC Yang, & Vong, 2017). To detect the fault in engines, instantaneous speed
data is used (Lin, Tan, Ma, & Mathew, 2015; R. Wang, Chen, & Guan,
The process of machine fault diagnosis includes a human effort to 2021; Yoon & He, 2015). For the machines that require electric power to
manually record the machine’s health status in pursuit of developing an function, their fault detection can be done using the current data (Naha,
FTC model. This reduces the overall accuracy. In Chen et al. (2016), Samanta, Routray, & Deb, 2017; Safizadeh & Yari, 2017; T. Yang, Pen,
Hasan and Kim (2018), Yanxue Wang, Xiang, Markert, and Liang Wang, & Chang, 2016). However, this current detection can also be
5
Fig. 3. Intelligent fault diagnosis (Krizhevsky et al., 2017).
performed with a simple current transformer. The intuition that can be to pre-rank the candidate characteristics among the models. Data vari
picked here is that multiple sensors detect interrelated values from the ance, Pearson correlation coefficient, Relief algorithm, Fisher score,
machines. These readings can be used in a novel manner to generate a class separability, chi-squared, information gain, and gain ratio are a
highly accurate diagnosis procedure (Duan et al., 2018). few of them. These variable ranking models evaluated characteristics
After the data collection step, next comes the feature extractions. from a variety of angles, resulting in a variety of ranking outcomes. A
This is done by using time-domain analysis, frequency-domain analysis, weighted voting system was implemented to re-rank features based on
and time–time-frequency-domain analysis. As a next step, these features the influence of the ranking outcomes on the Radial Basis Function
are categorized through filters, wrappers, or embedded methods. For (RBF) classification. The method was applied for high-dimensional
feature extraction, three types of analysis can be done. machinery fault diagnosis. In (Reyes, Morell, & Ventura, 2015; Urban
Time-domain analysis: Through this analysis, features can be owicz, Meeker, La Cava, Olson, & Moore, 2018) the authors imple
extracted based on mean, standard deviation, root mean square values, mented Relief and Relief-F filters to categorize the features according to
etc. These terms vary according to the speed and load conditions of the their sensitivity. In (X. Cao, Tian, Ji, & Qiu, 2021) they used the concept
machine. As they are time-dependent, one major category of time- of information gain to develop an FTC model for better results. The
domain analysis is dimensional analysis. The second category is concept was borrowed from the information gap theory. In (Sun, Yin,
dimensionless parameters which are a function of machine operation. Ding, Qian, & Xu, 2022), the minimum redundancy maximum relevance
They can be extracted in the form of skewness, impulse, shape indicator, (RMR) filter is used to identify the features with minimum similarity.
etc. Fisher score is a sort of filter that maximizes the feature distance con
Frequency-domain analysis: This analysis comes up when it was cerning another class of features and also minimizes this distance for the
observed that certain features cannot be extracted from time-domain features that are available within the same class (Sun, Wang, Ding, Xu, &
analysis. This analysis extracts features using the frequency spectrum. Lin, 2021). (Hallajian, Motameni, & Akbari, 2022; Oh, Lee, Lee, Joo, &
Time-frequency-analysis: The randomness of the machine health is Lee, 2022) used the previous distance characteristic of the feature and
extracted using time–frequency analysis (Duan et al., 2018; Rostaghi, filtered the sensitive features by using distance evaluation (DE).
Khatibi, Ashory, & Azami, 2021). This is done by using short-time Wrapper-based methods: This technique is focused on the interaction
Fourier analysis, wavelet transform (WT), or empirical model decom of the feature selection and the selected machine learning algorithm.
position (EMD). This kind of analysis is constraint-free when it comes to With the help of two experiments on the rolling bearing fault and rotor
non-stationary processes. system fault, the authors (Van & Kang, 2016) showed that the hybrid
filter and wrapper method is more accurate. In essence, the wrapper
5. Feature selection method compares the output accuracy of the algorithm and on that
basis, a set of features is selected. In case, where the accuracy falls below
After feature extraction, feature selection comes up as an important the expected output of the features, the set is altered in the next itera
topic to be considered. The computational cost, dimensionality curse, tion.(Xiong, Su, Jiang, & Jiang, 2019) shows the Las Vagas wrapper
and sacked information are major parameters where feature selection is (LVW), frequently used in industry to distinguish between the sensitive
concerned. There are several ways of feature selection. Some of the well- features set.
known methods available in open literature include the following. Embedded methods: As the name suggests this method embeds the
Filter-based methods: This method is independent of the classifying feature selection process within the classifier. The optimization objec
algorithm being used within the system. It filters the collected data and tive of the algorithm is parameterized using the feature set. The classifier
preprocesses it before handing it over to the algorithm. In P. Shi, Liang, algorithm training is done with the help of the features already selected
Han, and Zhang (2017) to pick the most significant input characteristics by the classifier. G. Li, Gu, and Ding (2020), Saud and Shakya (2021)
from all possibly relevant features, a hybrid model that incorporates identified two ways L1 and L2 to parameterize the classifier objective
various feature selection models is used. They utilized eight filter models function. The over-fitting of data was not seen using both methods.
6
Moreover, L1 can give better accuracy as it focuses on sparse parame feature parameters in the time domain, specifically intended for gear
ters. These parameters overcome the issue of redundancy of damage identification, and in the frequency domain. To select sensitive
information. features and remove fault-unrelated features, a two-stage feature se
Apart from these categories, there are some feature selection lection and weighting technique (TFSWT) using the Euclidean distance
methods worth mentioning. In (F. Jiang et al., 2018), a technique evaluation technique (EDET) was presented and used. The gear crack
combining ensemble empirical mode decomposition (EEMD), correla levels were identified with high accuracy using a weighted K nearest
tion coefficient analysis, and support vector machine (SVM) was sug neighbor (WKNN) classification algorithm. Moreover, a comprehensive
gested to fuse multi-sensor information for bearing fault diagnosis. To study on feature selection methods can be found in (Pudjihartono,
rebuild new vibration signals, EEMD is used to break down the vibration Fadason, Kempa-Liehr, & O’Sullivan, 2022).
signal into a collection of intrinsic mode functions (IMFs), and a corre
lation coefficient ratio factor (CCRF) is used to choose sensitive IMFs for 6. Health state recognition
further feature fusion analysis. Second, the reconstructed signal is used
to create an original feature space. Following that, correlation co The next inline procedure after the extraction and selection of fea
efficients among the vibration signals of the examined multi-sensors are tures is to evaluate the diagnosis model. This model helps to evaluate the
utilized to give weights, and the so-called fused features are retrieved relationship between the chosen features and the classifier. The model is
using the weights and original feature space. Finally, for bearing defect trained with samples that are labeled. After the training completion of
diagnostics, a trained SVM is used as the classifier. The original vibration the model, it is tested using unlabeled data. There are several machine
signals, the initial IMF, the suggested reconstruction signal, and the learning theories available and with the help of these many FTC ap
proposed technique all have diagnostic values of 73.33%, 74.17%, proaches exist in the literature. However, only four are discussed here
95.83%, and 100%, respectively. It is shown in Gerdes, Galar, and due to their research popularity.
Scholz (2016) that the decision trees can also be used to classify the a) Expert system-based approaches:
failures and the effect of feature extraction parameters can also be This approach is focused to provide the user with precise knowledge
analyzed by using decision trees in a robust sensor network. (Y. Li, Yang, about machine health diagnosis to significantly reduce human input.
Li, Xu, & Huang, 2017) describes a unique defect diagnosis technique The detailed implementation of this approach is shown in Fig. 4. The
based on modified multi-scale symbolic dynamic entropy (MMSDE) and raw data after going through a diagnosis task is collected from the dy
minimum redundancy maximum relevance (mRMR) to identify various namic database and shared with the inference engine. The node
planetary gearbox health states. In (Y. Li et al., 2017) MMSDE was knowledge base stores the unhealthy features related to machine diag
suggested to analyze the dynamical properties of time series over a range nosis and provides it to the inference engine when a comparison is
of scales by quantifying their regularity. The mRMR method was pro required for the sake of accuracy. The involvement of humans in un
posed to enhance the fault characteristics, and MMSDE offers clear ad derstanding the diagnosis is provided through the user interface.
vantages in the detection of dynamical changes and computing Furthermore, to respond to user queries an explanation system is
performance. Finally, to complete the fault pattern detection, the ob installed. Another expert system-based approach to detecting the
tained additional features were input into the least square support vector rotating machine fault can be found in (Kafeel et al., 2021). In this work,
machine (LSSVM). (Singh & Shaik, 2019) describes a support vector the authors have used the vibration sensor and signal processing tech
machine (SVM) and Stockwell transform-based technique for detecting, niques to identify and classify the faults. The classification results ob
classifying, and locating defective bearings in a three-phase induction tained in this work have shown that hybridization of spectral and
motor. A new approach was suggested in (Xiao, Chen, Chen, Hu, & Tang, temporal characteristics, classified using SVM along with Gaussian
2023) to classify the different degrees of gear fractures automatically kernel provided the best results with 98.2% accuracy, 100% specificity,
and accurately. To describe the gear conditions, this technique retrieved 96.6% sensitivity, and 1.8% error rate.
Fig. 4. The architecture of expert-based IFTC.
7
b) Rule-based Engine: have a clear understanding of the FTC based on SVM. The theory of SVM
(GELGELE & WANG, 1998) used the If-then rule-based system tar is different from the neural networks in the way that a hyperplane is
geting automotive engines. (J. Dai, Tang, Huang, & Wang, 2019) used to distinguish between dataset elements. Given the dataset
{ }m
implemented signal-based FTC for hydraulic systems, (Diaz et al., 2017) xi , yi i=1 with m samples and yi ∊{ − 1, 1}, a hyperplane f(x) = 0 is
and (S. Ahmad, Ahmad, & Kim, 2022) have proposed the rule-based separated the given dataset into two classes, and the hyperplane is
fault diagnosis for bearings, and centrifugal pumps respectively. Be shown as
sides many applications, the rule-based system lacks compatibility with
∑
the increased number of variables.
m
f (x) = wT x + b = wT xi + b = 0, (16)
Fuzzy logic-based engine: To deal with the non-numeric and i=1
imprecise data, fuzzy logic-based fault diagnosis is introduced.
Where w and b are the parameters to find the hyperplane. The
(Changsong Li et al., 2022) used the fuzzy fault-tolerant model for the
created hyperplane is subject to
power systems. The fuzzy model was implemented with Weighted
( )
Corrective Fuzzy Reasoning Spiking Neural P Systems with real numbers yi f (xi ) = yi wT xi + b ≥ 1, i = 1, 2, • • • , m (17)
(rWCFRSNPSs) by (T. Wang et al., 2020). In Alsuwian, Amin, et al.
(2022) the authors applied this concept to motor engines. The FTC based Therefore, the cost function of SVM is
on fuzzy attributes was used for bearing and efficient results were seen 1
in (Berredjem & Benidir, 2018). The problem with the fuzzy-based en min ‖w‖2 (18)
w, b 2
gine is that it is difficult for the model to gain knowledge as the dataset
based on fuzzy information is hard to capture. Henceforth, reducing the ∴s.t.yi (wT xi +b) ≥ 1, i = 1, 2, • • •, m AFR in a certain ratio in the
FTC accuracy. In (Riaz, Amin, & Tayyeb, 2022), an AFTCS based on a compression stroke. For improved engine performance, it’s very critical.
fuzzy logic controller was suggested to improve the reliability of the The literature studies were done on the use of SVM for making the
air–fuel ratio control system of an IC engine. machine health diagnosis model (Zhongsheng Wang & Fan, 2015), these
c) Neural network-based engine: studies include the characteristic Frequency Band Energy Entropy and
To enhance the accuracy of the FTC, neural networks (NN) are Support Vector Machine (Pang, Tang, Zhou, & Tian, 2018), fault diag
considered as the next option (Z. Chen, Chen, Wu, Cheng, & Lin, 2019). nosis of turbo-pump rotors (Belosludtsev, Borovik, Danilchenko, &
The built-in characteristics of the neural network-based diagnosis model Sekisov, 2021), and automotive systems (Q. Shi & Zhang, 2021). The
are that it has inherently received the memory, learning, and feature highlighted issues were further analyzed and resolved through different
association (Y. Zhao, Li, Zhang, & Zhang, 2019). Internal combustion SVM approaches used for rolling bearing fault diagnosis(Y. Li, Xu, Zhao,
engine fault diagnosis is done by (Cervantes-Bobadilla et al., 2023) and & Huang, 2016; Y. Li, Yang, Wang, Liu, & Liang, 2018), to identify the
(Namigtle-Jiménez, Escobar-Jiménez, Gómez-Aguilar, García-Beltrán, gear faults (D. Yang, Liu, Li, Li, & Ma, 2015), wind turbine fault diag
& Téllez-Anguiano, 2020) using neural networks based on probability nosis (B. Tang, Song, Li, & Deng, 2014), diesel engine fault detection
concepts. (S. Zhang et al., 2021) modeled the FTC using the concept of a (Cai, Weng, & Zhang, 2017), and for induction motor fault detection and
multi-layer perceptron. The model was then used in the case of bearings. condition monitoring (M. Z. Ali, Shabbir, Liang, Zhang, & Hu, 2019;
For the same problem, (Choudhary, Mian, & Fatima, 2021) used the Choudhary, Goyal, & Letha, 2021). However, SVM has its drawbacks
neural network model along with thermal images for decision-making to which include instability when faced with large data, kernel parametric
achieve the FTC model. The combined effect of the above two schemes is restrictions, and inability to handle multi-layer tasks.
unable to still bridge one gap between two main issues. First, it is Aside from ANN and SVM, other approaches also exist. These include
difficult to collect huge data from the machine for diagnosis in the en kNN, PGM, and decision tree methods.
gineering field. Second, the reasoning block in Fig. 4 cannot get the KNN.
required amount of knowledge to provide the explanation block for KNN is one of the supervised learning models and is useful in clas
reasoning. Fuzzy-based decision-making solves this problem of sification problems. The process includes finding the closest entry near
reasoning. However, ANN requires large data for training which is not an unlabeled sample. In the open literature, the kNN approach is used
available in real-world engineering problems. for detecting the faults in plastic bearing faults (Peña et al., 2018),
d) Case-based engine: rolling bearing fault detection and estimation (L. Jiang, Shi, & Xuan,
From the literature, case-based diagnosis models are focused to solve 2014; L. Jiang, Xuan, & Shi, 2013; J. Ma, Xu, Huang, & Huang, 2017;
privileged engineering problems for the existing problem samples. The Van & Kang, 2016; Yao, Zhen, Wu, & Guan, 2017), accelerometer and
fault diagnosis model was built by using the existing knowledge avail load cell rolling bearing faults (Safizadeh & Latifi, 2014), a wind tur
able at railway companies and train manufacturers (Q. Xu et al., 2018). bine’s bearing fault diagnosis (X. An & Tang, 2017), automobile gearbox
In the automotive industry, (James, Gandhi, & Deshmukh, 2018) makes fault detection (Gharavian, Almas Ganj, Ohadi, & Heidari Bafroui, 2013;
use of the fault diagnosis data available from the company to build the Z. Li et al., 2013; Park, Kim, & Choi, 2018), fixed axis gearbox faults
FTC. Within the aircraft industry, (Weiser & Ossmann, 2022) prepared detection (Vanraj, Dhami, & Pabla, 2018), and to diagnose the stator
an FTC model for modern commercial aircraft. Computer-aided (CAD) faults of the single-phase induction motor (Glowacz & Glowacz, 2017).
FTC modeling was done by (Diagnosis and Fault-tolerant Control 2, The problem that originates while using the kNN is the non-optimal
2021), using fuzzy and kernel k-means hybrid diagnosis model for the selection of the parameters. This was resolved by using modified kNN
ignition system of automotive. To summarize, the fault diagnosis model for various mechanical applications (S. Dong, Luo, Zhong, Chen, & Xu,
based on an expert system approach acts as an algorithm for the infer 2017; S. Dong, Xu, & Chen, 2016; X. Zhao & Jia, 2018). To summarize,
ence engine. The model will be able to accommodate the required kNN is worthwhile to use. However, they offer greater computational
automation of the health state of the machine. However, the collected cost and the difficulty of finding the exact value for k.
data limitation still hinders the development of an expert system-based PGM.
inference engine. Moreover, this system lacks self-learning ability. This approach is focused on the use of probabilistic theory concepts
for finding the optimal fault diagnosis model. PGM is classified into two
parts, namely the Bayesian and Markov models. (J. Yu, Huang, & Zhao,
7. SVM based approaches
2016) and (J. Yu, Bai, Wang, & Shi, 2018) used the naive Bayesian
classifier to identify the positive and negative sides of classification for
Like NN, SVM also falls under supervised machine learning tech
the gearbox fault diagnosis. Another study was conducted for the
niques. It is very useful in classification problems. Various applications
detection and estimation of simultaneous faults by using Bayesian
of this machine learning theory are seen in the following sub-section to
8
classifiers (Y.-L. He, Wang, Kwong, & Wang, 2014). Two more studies the data. As a result, we must make sure that the data pretreatment
related to fault diagnosis of rolling bearings (J. Yu, Ding, & He, 2018) procedure, which involves eliminating outliers, filtering out missing
and rotary machinery (Asr, Ettefagh, Hassannejad, & Razavi, 2017) used values, and eliminating undesired characteristics, is carried out with the
the non-naïve Bayesian classifiers. Both these studies have reported a highest degree of accuracy. For some problems, it is even hard to find the
very high degree of accuracy. Furthermore, L. Dong, Li, Wang, and Lin training data. For example, if we want to train some model for 6G ap
(2020), Y. Jia, Xu, and Wang (2018) used the hidden Markov model plications, then we don’t have enough training data. For these sorts of
within the classifier. PGM has the advantage of working with huge data, problems, we need to mimic the behavior of the application first and
but it still has over-fitting issues, which will ultimately make it difficult then we can train the ML model which makes the whole process slow.
to model the classifier with high accuracy. Another problem is the mismatch of the data when it becomes hard to
Decision tree. sort out the relationship with an increased number of variables using
Fig. 5 shows a decision tree model. This model was tested by classical machine learning theories.
(Amarnath, Sugumaran, & Kumar, 2013; Y. Zhang, Li, Gao, Wang, & c) Robustness of Algorithm.
Wen, 2018) for applications like gears, automotive, and rotors. (B.-S. Another major problem in ML techniques is that, as the amount of
Yang, Di, & Han, 2008; Ziwei Wang et al., 2017) used the random forest data increases, the model can eventually be useless. The most accurate
for bearings application. (G. Tang, Pang, Tian, & Zhou, 2018) made use model at this time might prove inaccurate in the future and need to be
of particle swarm optimization to optimize the random forest results. revised. Therefore, in order to preserve the algorithm’s functioning,
continuous monitoring and upkeep are required. Arguably this is the
8. Challenges most taxing problem in the ML field.
The above section discussed the application of classical machine 9. Deep learning algorithms in FTC
learning theories for the development of FTC. Two main shortcomings
are identified as follows: The advancement in technology resulted in the gathering of more
a) High cost. information from machines for the FTC model. It has been made possible
Although AI has been introduced into this field. Still, the data was due to the field of the Internet of Things (IoT). It gives rise to the concept
collected through human input. When it comes to the collection of large of big data. The issue with the classical machine learning theories is that
data for meaningful knowledge, many humans will be needed to identify they don’t match big data collection. Hence, the need for more sophis
the sensitive features. Which will ultimately enhance the cost of labor. ticated algorithms was overcome by deep learning theories. As shown in
However, with the development of machine learning (ML), a number of Fig. 6 two steps are involved in an FTC procedure.
methods have been made available that can drastically cut the
requirement for human intervention in these procedures. For instance,
patterns and correlations within the data can be found without explicit 9.1. Big data collection
labeling by using unsupervised learning techniques like clustering and
dimensionality reduction. Reduced labor costs are another benefit of The term big data is understood by its characteristics. (Mayer-
semi-supervised learning, which integrates a small amount of manually Schönberger & Cukier, 2013) states four characteristics that explain the
labeled data with a big amount of unlabeled data. Even though there meaning of the well-known industrial term big data. Volume, veracity,
may still be situations when human input is required, AI and machine variety, velocity.
learning have developed strategies that can lower the labor expenses Big volume: The volume of data enhances the duration of the ma
related to data collecting and labeling. Although these techniques chine’s operation. It becomes huge when it is collected from a group of
reduce the human cost and the overall computational cost increases machines. For example, wind turbines when working in a smart grid.
which is still a challenge to tackle. Low density: The collected big data has a low density considering the
b) Data Quality and Availability. complete information about the health of the machine. Henceforth, to
The machine learning method heavily relies on data. Lack of high- fill up this gap some bad quality data is mixed with big data (X. Xu, Lei,
quality data is one of the major problems that machine learning ex & Li, 2020).
perts encounter. It can be exceedingly taxing to process noisy and erratic Multi-source data structures: The data is collected from various
data. We don’t want our system to produce predictions that are unreli nodes within the machine through sensors. The received data is het
able or flawed. Therefore, improving the result depends on the quality of erogeneous due to different storage structures.
Monitoring data: Data monitoring is a fast-track job performed by
Fig. 5. The decision tree (J. Yu et al., 2016).
9
Fig. 6. Diagnosis procedure of IFTC (Mayer-Schönberger & Cukier, 2013).
high-speed transmission channels. First are the production units of The various modified forms of AE used in the literature include
modern industries where machines work in the group and generate huge sparse AE, auto-denoising AE, and stacked AE. These models showed
data. Second, the normal data collection for diagnosis is based on mostly higher accuracy in comparison to SVM and ANN when applied to di
healthy data. The machine’s faults occur occasionally; hence the agnose faults across bearing, automotive, and rotor systems. (F. Jia, Lei,
received data is of less use with this density. Third, every sensor on a Guo, Lin, & Xing, 2018) implemented an end-to-end diagnosis system
machine reads and records the data in a different pattern. Hence the using normalized AE with the optimization objective to learn higher
received data is from multiple storage patterns. Finally, the development density and heterogeneous nature features. This was achieved by
of electronics has resulted in more speed of computation through GPU, describing networks that are connected through normalized sparse AE.
IoT, and edge computing. (H. Shao, Jiang, Zhao, & Wang, 2017) implemented an AE-based diag
nosis model with the objective function designed based on correntropy
and then used a fish-warm algorithm to optimize the parameters. For
9.2. Deep learning-based diagnosis
motor bearing fault diagnosis, (M. Ma, Sun, & Chen, 2018) used AE to
construct an FTC model based on RNN. In (H. Yu, Wang, & Li, 2018), a
Within the context of deep learning, the model learns the features
hybrid model for diagnosis is attempted by the researchers. The theme is
from the data collection. At the same time, the diagnosis is also
the jointly constructed multiple AE models. These combined AE models
happening using the knowledge of learned features. This is done by
are designed to receive multi-scale signals and learn the robust features
using two layers. One is feature extraction and the other is the classifi
of FTC. These jointly reconstructed AE models work with other opti
cation layer. The feature extraction layer is based on multiple archi
mization techniques. The selection of parameters for jointly recon
tectures of various techniques like stacked AE, CNN, Res Net, etc.
structed AE models is based on the optimization method rather than the
However, the classification model is based on some ANN techniques due
BP algorithm. The improved results achieved through jointly recon
to its efficiency when dealing with multiple architectural data. In the
structed AE models were used by (Z.-X. Yang, Wang, & Zhong, 2016) in
following subsections, four deep learning models and their applications
the case of bearings and wind turbines. RProp and differential evalua
for developing FTC are discussed.
tion algorithms are used for optimization (Saufi, Ahmad, Leong, & Lim,
a) Stacked AE-based approach.
2018).
In the literature, it can be seen that AE has been widely used in
AE is an unsupervised technique. It can automatically process the
developing FTC models (X. Guo, Shen, & Chen, 2017; C. Lu, Wang, Qin,
{ }m feature extraction and is independent of the knowledge block. Hence, AE
& Ma, 2017; Qu, He, Deutsch, & He, 2017). Given the dataset xi , yi i=1 models can’t be directly used for machine health diagnosis. It becomes
with m samples, the represented features hi are defined as essential to add a classifier on top of the model hierarchy. And thus, as a
(
hi = fθ (xi ) = σf wT xi + b
)
(19) drawback, the AE model still needs to be trained well enough.
b) ANN-based approaches.
Where σ f demonstrates the activation function of the encoder A better way to diagnose fault is through ANN. ANN works on the
network. The reconstructed sample ̂
x i is illustrated as same logic as a human brain. Following are some applications of ANN
( T ) for modeling FTC.
xi = go w hi + b′ (20) The ANN works with both forward and backpropagation. When it is a
Where go represents the activation function of the decoder network. multi-layer perceptron model it is known as a backpropagation neural
To minimize the error between input samples and reconstructed ones, network (BPNN). The forward propagation of the neural network is
we used shown in Fig. 7. The input is in the form of samples and is treated with
different multiple hidden layers before the required output is mapped
1 ∑ through this input. The dataset is mapped to real numbers i.e. (x, y)
min00L(xi , ̂
xi) = mi = 1xi − x12 (21)
2m (Shahbaz & Amin, 2021). Where × represents the selected feature and y
The same represented features of lth AE can be calculated as translates the health state to this feature, the output layer is expressed
( ) as,
hli = fθl hl−i 1 , l = 2, 3, • • • , L, (22)
Where hli are features of AE, and θl is the training parameter of lth AE.
10
Fig. 7. The architecture of the backpropagation neural network (Shahbaz & Amin, 2021).
( )
nh− 1
∑
( )
xhi = σh whj .xh−i 1 + bhj , j = 1, 2, • • • , nh , h = 1, 2, • • • , H, (23) TABLE 2
i
i=1 Applications of ANN in fault-tolerant.
( ) Objects References Strategies
xhi i
represents the output of the jth neuron, nh is the number of
neurons, σh demonstrate the activation function, whj is weigh, and bhj is a Engines (Zhai, Yu, Qian, Lee, & Theera-Umpon, 2017; Mihály, BPNN
Luca, & Cristea, 2023; Khazaee et al., 2016) RBFNWNN
bias function. The predicted output of BPNN is Gears (Y. H. Ali et al., 2022; B. An et al., 2022; Li et al., 2018) RBFN
( ) WNNBPNN
∑
nH
Motors (Li et al., 2014; Pasqualotto & Zigliotto, 2022; Shifat & BPNN
(̂y i )k = σ OUT wOUT • xH
+ bOUT
, k = 1, 2, • • • , l, (24)
j i j
Hur, 2021)
i=1
Bearings (Bangalore & Tjernberg, 2015; Chikkam & Singh, 2023; BPNN
y i )k is predicted output, σ OUT is activation function of output, bOUT
(̂ is Gunerkar, Jalan, & Belgamwar, 2019)
j
the bias of the output layer, and wOUT
j demonstrates the weight of the
output layer. The root mean square error is used to design a neural Ahmadi, Amirmostofian, & Bagheri, 2021) made use of the ANN-based
network that precisely predicts the actual output (yi )k , the root mean approach for fault diagnosis in the bearings, and rotor system respec
square error is demonstrated as tively. The probability-based NN was used by (X. Chen et al., 2014). The
NN with the sequential processing steps for adaptive resonance diag
1∑ l
[ ]2
nosis in the rotor system was applied and it produced efficient results
minEi = y i )k
(yi )k − (̂ (25)
w,b 2 k=1 (Jin, Hou, Chen, & Lu, 2022). Among all these, (Barakat, El Badaoui, &
Guillet, 2013; Barakat, Lefebvre, Khalil, Druaux, & Mustapha, 2013)
A gradient descent algorithm is used to estimate the states in this
made use of the ANN-based fault-tolerant model of bearings.
BPNN as follows
c) DBN-based approach.
∂Ei ∂Ei The parametric limitation and big data collection gave rise to the
w←w − η • , b←b − η • (26)
∂w ∂b concept of generative stochastic NN. Fig. 8 shows an RBM architecture.
∴η is the learning rate. RBM is a type of generative stochastic NN. The specification can be
The different applications of the ANN-based fault-tolerant model are understood in a way that RBM has only binary inputs. Moreover, RBM is
discussed here. Moreover, Table 2 is prepared to have a view of four an energy-based model. A DBN is constructed by jointly reconstructing a
major applications within the domain of mechanical engineering sce set of RBM models. In the RBM structure, there are visible units v, and
narios. In the (Shahbaz & Amin, 2021), an AFTCS based on ANN was hidden units h. The variables v and h are subject to the joint configu
proposed for the air–fuel control system of a spark ignition internal ration as follows.
combustion engine. ∑
m ∑
n ∑
m ∑
n
(Hendrickx et al., 2020; Nezamivand Chegini, Haghdoust Manjili, E(v, h, θ) = − wi.j vi hj − bi vi − aj hj , (27)
i=1 j=1 i=1 j=1
11
Fig. 8. RBM architecture (D. Han, Zhao, & Shi, 2017).
Where θ represents the parameters of RBM. the marginal distribution Where xi represents the input dataset with k samples. By adding the
is calculated as softmax classification layer, the deep layer mapped properly as
[ ( ) ( ) ( )]
1 ∑ v
y i = P yi = 1|hli , θc , ⋯, P yi = q|hli , θc , ⋯, P yi = k|hli , θc
̂ (32)
P(v|θ) = exp[ − E(v, h, θ) ] (28)
Z(θ) h ( ) ( )/∑ ( )
k
Where P yi = 1|hli , θc = exp wcq • hLi +bcq q=1 exp wq • hi +bq ,
c L c
Where Z(θ) is the partition function. The activation condition of

hLi are the represented features in the Lth layer from the i th sample, ̂ yi
visible and hidden units is
represents the predicted one-hot label, and θc = {wc , bc } is the training
( )
∑
m parameter of the classification layer.
P(vi = 1|h) = σs bi + wi.j • hj (29) (D. Han et al., 2017) made use of variations of RBM, like single
Gaussian RBM, and stacked RBM for developing the diagnosis model of
j=1
( )
∑
And P(hi = 1|v) = σ s ai + nj=1 wi.j • vj (30). machine health. Optimization methods for DBN were stated in (H. Shao,
Jiang, Wang, & Wang, 2017; J. Xie et al., 2018). They include the
Where σ s is the activation function of the sigmoid. The estimations of
Nesterov method and particle swarm. The output accuracy of the
RBM are done by.
∑ modified models was found better.
θ = arg maxIn[P(θ|x1 , x2 , • • • , xk ) ] = 1 k In[P(xi |θ) ] (31).
̂
k i=1
θ
Fig. 9. Convolution and pooling process (Gu et al., 2018).
12
To achieve an accurate diagnosis model, DBN based health model an eye-catcher for many researchers. The use of transfer learning for
completes the process of feature extraction by externally training the developing the FTC model is discussed in this section. Furthermore, the
stacked RBM models. The problem with the BP algorithm, gradient section is detailed in subsections covering the motivation, use, and ap
vanishing, is overcome by training the RBM models with sufficiently proaches for developing an FTC based on transfer learning. The problem
labeled data. with the machine learning-based techniques was the unavailability of
d) CNN-based approach. sufficiently labeled data when engineering problems were considered.
CNN is a supervised learning technique. It has already proved itself in There is an extra and heavy cost involved in acquiring the right amount
the fields of image processing, computer vision, and target tracking (Gu of data which is also properly labeled. Hence, the idea of learning from
et al., 2018). CNN model consists of pooling layers, convolutional layers, the experience gained in diagnosis from one machine can be reused with
and fully connected layers. Fig. 9 shows the process of action of con other machines too. This idea leads to the concept of transfer learning
volutional and pooling layers. when dealing with real-world engineering problems.
The diagnosis model prepared using CNN can be of two types based In transfer learning, the diagnosis knowledge collected from one
on the dimensional approach. This architectural difference between 1D machine can be applied to gain knowledge from another machine or the
and 2D is expressed in Table 3 with the broad spectrum of CNN-based same machine. The diagnosis is however performed for a different task
fault-tolerant models in the literature. When it comes to fault diag than the one it is collected from. Hence, transfer learning works with
nosis models based on CNN, 2D CNN faces the curse of dimensionality source and target domains (S. J. Pan & Yang, 2010). Considering the
when working with 1D signals. This was overcome by (Ding & He, 2017; FTC, transfer learning can be divided into two streams depending on the
Y. Han, Tang, & Deng, 2018; Islam & Kim, 2019) using wavelet machine and its transferring knowledge. (W. Zhang, Li, Peng, Chen, &
transform. Zhang, 2018) establishes the basis for data transferring between iden
CNN-based diagnosis model can learn directly from the monitored tical or different machines, the two streams of applying transfer
input data. This raw data doesn’t require any pre-processing if sufficient learning. Within both streams, the source domain has labeled data,
data is available with proper labeling for feature extraction. Further however, the target domain may or may not be labeled. Transfer
more, CNN works with multi-scale data by sharing weights among the learning has been used by researchers for developing FTC models and
layers, and hence the problem of over-fitting vanishes from the model. Table 4 shows the details. Broadly speaking transfer learning can be
e) ResNet-based approach. achieved in four ways that are discussed in the coming up subsections.
The architecture of ResNet is shown in Fig. 10. A ResNet constructs a Feature-based approach.
deep learning architecture by piling up certain convolutional layers As the case may arise, the feature-based diagnosis approach is a valid
called residual blocks. The layers are connected in two ways. One is the choice for the scenarios, TDM, where cross-domain discrepancy can
forward channel connection and the second is the shortcut connection. occur (Long et al., 2016). Furthermore, the feature-based approach
To improve the efficiency of the input signals while it is passing through starts working by setting the common feature space. Next, the distrib
residual blocks, zero paddings are performed on the signals. The purpose uted discrepancy of the features is measured and optimized using the
of a shortcut connection is to keep the balance between the input and a minimization optimization algorithms. In the end, the cross-domain
combined output of forwarding channel connections. classifier segregates the features based on the knowledge learned after
The efficiency achieved by (S. Ma, Chu, & Han, 2019; W. Zhang, Li, & getting trained from source domain sample data. This is depicted in
Ding, 2019; M. Zhao, Kang, Tang, & Pecht, 2018, 2019) in implementing Fig. 11.
the ResNet-based FTC results in better diagnostics than other deep- GAN-based approach.
learning models. Moreover, it showed can be used with complex In GAN based approach, the generative model reads the available
models for efficient results. distribution of the target domain and reproduces the fake samples with
maximum similarity. Then comes the discriminator model. This works
10. Challenges on the parametric tuning of the generative model to help it achieve
maximum dissimilarity between original and fake samples (Warde-
To summarize, the present research consists of two factors. One is Farley & Xu, 2014). A deep adversarial CNN was implemented by (T.
data collection and the second is the deep learning algorithm. As the Han, Liu, Yang, & Jiang, 2019). The benefit of GAN is its equal oppor
literature cited has indicated aside from the benefits of high accuracy, tunity for use in TIM and TDM machines. The fake data generated is very
these methods still lack the optimized self-learning feature extraction useful in training the target domain samples. Hence, it is no issue for
procedure. The availability of properly labeled data is a must for the GAN-based FTC if the target data samples are not labeled properly.
performance of these FTC models. Hence, the need for a complete end- Instance-based approach.
to-end FTC model is still craved. In case the target domain data samples are not available in sufficient
amounts to train the model properly, an instance-based approach is
11. Transfer learning algorithms in FTC used. In this approach, the data available in the source domain at that
instance is used to improve the functionality of the model. In (W. Dai
The topic of transfer learning is an uprising concept and has become et al., 2007), the TrAdaboost approach is used, it is derived from the
Adaboost algorithm. To summarize, the instance-based approach is
favorable to work with TIM machines, but in the case of TDM machines,
TABLE 3 where the data sample discrepancy is huge, they may fail to perform as
Applications of CNN In fault-tolerant. expected.
Strategies Input data References Parameter-based approach.
2D CNN Time & frequency (Verstraete, Ferrada, Droguett, Meruane, & (Weiss, Khoshgoftaar, & Wang, 2016) relates the parameter
spectrum Modarres, 2017; X. Chen, Peng, Cheng, & approach in literature. The approach assumes that less data is available
Luo, 2019; D. Guo, Zhong, Ji, Liu, & Yang, in the target domain to train the model. In such a scenario, the param
2018) eters are trained primarily with the data samples available in the source
Between 2D Reshaped matrix (S. Li, Liu, Tang, Lu, & Hu, 2017; C. Lu,
& 1D Wang, & Zhou, 2017; S. Wang, Xiang, Zhong,
domain and then assigned to the model. After that, the available mi
& Zhou, 2018) nority target samples are used in comparison to match the required
1D ImagesRaw data (Ince, Kiranyaz, Eren, Askar, & Gabbouj, output accuracy. The parameter-based approach has the benefit of low
2016; G. Jiang, He, Yan, & Xie, 2019; Jiao, computation and fast results. Furthermore, this approach is directly
Zhao, Lin, & Zhao, 2018)
dependent on the number of samples available in the target domain for
13
Fig. 10. The architecture of the residual block.
beings. Moreover, deep learning and transfer learning algorithms have

TABLE 4
opened the FTC modeling for engineering problems which is normally a
Applications of transfer learning in fault-tolerant.
hard job to perform for the sake of machine health diagnosis.
Techniques References Procedures Advantages and Disadvantages of the Available Techniques.
Feature- (C. Chen, Li, Yang, & Liang, TCA and JDA In order to summarize the machine learning, neural networks, deep
based 2017; J. Xie, Zhang, Duan, & Deep transfer learningTransfer learning, and transfer learning techniques during this review, the ad
Wang, 2016; B. Yang, Lei, Jia, factor analysis vantages and disadvantages of each technique are given in Table 5.
& Xing, 2018; Xiang Li,
Zhang, & Ding, 2018; W. Lu
These advantages and disadvantages are summarized by considering the
et al., 2017; Wen, Gao, & Li, fault tolerant control for different applications.
2019; J.Wang, Xie, Zhang, & High-quality big data.
Duan, 2016; X. Wang, He, & The availability has significantly improved over the last few years
Li, 2019; Y. Xu, Sun, Liu, &
due to the improvement in electronics and technology. However, the
Zheng, 2019)
GAN based (X. Li, Zhang, & Ding, 2019; GAN and MMDTrAdaBoost quality of the data is still poor, it is low in density. This may be due to the
Y. Xie & Zhang, 2018; Zheng, working environment, and sensors malfunctioning. Hence, the data
Wang, Yang, Li, & Xu, 2020) quality needs to be improved to have accurate diagnosis models. Tech
Parameters (P. Cao, Zhang, & Tang, 2018; Deep learning and fine-tuning niques like clustering algorithms and crowd-sourcing databases can be
based S. Shao, McAleer, Yan, &
Baldi, 2019; Zhang, Tao, Wu,
an answer to the problem of high-quality big data.
& Guan, 2017) Deep learning models’ limitations.
The major limitations of the deep learning models are listed below.
its efficiency. • Deep learning models’ main flaw is that they can only learn through
observations. As a result, they are only aware of the knowledge
12. Discussion and future research directions contained in the training information. If a user simply has a small
amount of data, or if the data comes from an individual source that is
With the development of machine learning theories, the FTC model
has become more active resulting in less contribution from human
Fig. 11. Steps of feature-based transfer learning.
14
TABLE 5
Advantages and Disadvantages of AI Techniques for FTC.
Category Algorithm Advantage Disadvantage
Machine SVM • Requires less memory • Can’t perform well for large datasets
Learning Efficient in high dimensionsEffective when the number of dimensions is Prone to noise in dataSlow convergence
greater than the number of dimensions
KNN • No training period is required • Not suitable for large datasets
Implementation is easyThe addition of new data doesn’t affect the model Requires feature scalingAccuracy decreases with an
increase in dimensions
PGM • Works well with sequential data • Requires extensive mathematical modelingDoes not work
Can form a relation with all data pointsCan be used to build casual models well with missing data items
Decision • Does not require data normalization • Requires feature reduction
Trees Missing values do not affect the model’s accuracyData cleansing is not Prone to overfittingRequires data resampling
required
Neural ANN • Can perform parallel processing • Computationally expensiveLarge number of hyper
Networks Work well on non-linear dataCan eliminate the vanishing gradient parameters
Deep Learning AE • Flexible for all types of dataCan be used for pre-processing • Very sensitive to hyper parametersRequires a lot of memory
space
DBN • Can perform both classification and regressionInsensitive to training data • Heavy dependence on initial valuesComputationally
fragmentation expensive
ResNet • Fast convergence • Inappropriate for embedded system
Can be used to increase the depth of the networkIncrease the model Prone to overfittingLess interpretability
reusability
CNN • Automatic feature extraction • Cannot work with smaller datasetsCan’t be generalized in
Very useful for imagesSupports the transfer learning every case
Transfer Feature-Based • No need for feature extraction and trainingWorks well on images • Less accurateNeed a pre-trained model
Learning Parameter • Improved accuracyNo need for data labeling • Cannot be used for different types of dataRequires a large
Based amount of memory
Instance- • No need to estimate the entire instance setAdaptability to the new dataset • Needs fine-tuningRequires more expertise in setting the
Based hyper parameters
GAN Based • Can increase training dataCan reduce overfitting • Hard to trainNot suitable for textual data
not always typical of the greater functional area, the models are that becomes accessible in a sequential manner, if the model deals
unlikely to develop in a fashion that can be generalized. with streams of data.
• Another major issue with deep learning systems is biases. A model • As IoT devices increase and continuously produce data, this can lead
that has been trained with skewed data will reproduce the same to high data streams. High data streams alone do not, however,
biases in its forecasts. Since models can distinguish based on slight ensure improved deep learning-based models. Successful deep-
variations in data bits, deep learning experts have had difficulty learning models depend on the data’s quality, relevance to the
solving this problem. For the coder, the key deciding criteria are issue at hand, and speed at which they can be processed and used for
typically implicit. Therefore, a facial recognition system might form learning.
assumptions about a person’s appearance based on characteristics
like race or gender without the programmer’s understanding. Performance improvement of transfer-based learning.
• The learning rate may present deep learning models with substantial The drawback in the transfer learning-based FTC is that it is sub
challenges. If the speed of convergence is too high, the model will jected to the data received from multiple task nodes from alike machines
reach convergence too quickly and produce less-than-ideal results. If or different machines. In case the data provided contains negative in
the rate of progress is too slow, the process could get stopped, making formation, the model can get trained negatively, and hence the output
it much harder to come up with a solution. will be less accurate. This can be overcome by using GAN based
• The hardware requirements for deep learning models may poten approach of generating fake data samples and comparing them with the
tially have limitations. Multiple cores of outstanding graphics pro limited available data.
cessors (GPUs) along with additional processing units are required to Computational Cost Comparison.
ensure enhanced effectiveness and decreased time consumption. According to Table 6, the properties of the input to be processed are
These gadgets cost a lot of money and use a lot of energy, though. related to the model’s complexity. Systems with large input dimensions
Additional hardware requirements include an SSD along with exist demand modeling techniques that are more sophisticated and compu
ing RAM and a hard drive. tationally expensive. This does not, however, imply that the most
• After being trained, deep learning models become stiff and unable to difficult option is always the best one for solving the issue; in reality, it
multitask. They can only exactly and successfully handle one
particular problem. Even dealing with a similar problem would
require model retraining. TABLE 6
• Predictions may be incorrect if there is an imbalance in the data since Computational Cost Comparison.
the model may be biased in favor of the dominant class. In this sit Model Strengths Weaknesses
uation, cost-sensitive learning methods like AdaBoost and XGBoost Support Vector High accuracy, low Sensitive to noise, Slow
can be useful by modifying the weights of the instances while Machine (SVM) storage Computing for big data,
learning. There are additional methods that can be used to address Artificial Neural High accuracy, fault High computing cost,
this issue, such as oversampling the minority class, under sampling Network (ANN) tolerance Physical meaning,
Convolutional Neural Feature extraction is free, Long training time, High
the dominant class, or combining both (SMOTE, for instance). Network (CNN) efficient for big data computing costs
• When data is regularly generated, which can be problematic because Recurrent Neural Robust to input size, short Higher memory required
handling a large volume of data is difficult, that presents another Network (RNN) memory problem free and computing complexity
obstacle for a deep learning model. Consider using online learning or Generative Adversarial Complex structure Harder to train, physical
Network (GAN) detection, generalized meaning
incremental learning strategies, which are made to deal with data
training
15
frequently occurs that an algorithm’s effectiveness varies depending on measured stator currents and vibration signals. IEEE Transactions on Industry
Applications, 55(3), 2378–2391. https://doi.org/10.1109/TIA.2019.2895797
the input.
Ali, Y. H., H. Ahmed, F. Y., Abdelrhman, A. M., Ali, S. M., Borhana, A. A., & Ishak Raja
Interpreting deep learning models. Hamzah, R. (2022). Novel Spiking Neural Network Model for Gear Fault Diagnosis.
One major issue faced by researchers is the black-box nature of deep 2022 2nd International Conference on Emerging Smart Technologies and Applications
learning models. Researchers are currently unable to see the theory in (eSmarTA), 1–6. Ibb, Yemen: IEEE. 10.1109/eSmarTA56775.2022.9935414.
Alsuwian, T., Amin, A. A., Iqbal, M. S., Qadir, M. B., Almasabi, S., & Jalalah, M. (2022).
action. This is due to the reason that these models gain experience and Design of active Fault-Tolerant control system for Air-Fuel ratio control of internal
learn better based on experimental data. To overcome this gap, statis combustion engine using nonlinear regression-based observer model. PLoS One1, 17
tical theory can be used by researchers to completely understand the (12), e0279101.
Alsuwian, T., Iqbal, M. S., Amin, A. A., Qadir, M. B., Almasabi, S., & Jalalah, M. (2022).
FTC models based on deep learning. Also, the use of adaptive filter A comparative study of design of active Fault-Tolerant control system for Air-Fuel
theory can help in understanding and coming up with new approaches ratio control of internal combustion engine using particle swarm optimization,
to implementing deep learning-based models. genetic algorithm, and nonlinear Regression-Based observer model. Applied Sciences,
12(15), 7841. https://doi.org/10.3390/app12157841
To conclude, machine learning-based algorithms can be used for Alsuwian, T., Riaz, U., Amin, A. A., Qadir, M. B., Almasabi, S., & Jalalah, M. (2022).
developing FTC models for engineering problems. The unavailability of Hybrid Fault-Tolerant control for Air-Fuel ratio control system of internal
sufficient data can be compared through deep learning techniques. combustion engine using fuzzy logic and Super-Twisting sliding mode control
techniques. Energies, 15(19), 7010. https://doi.org/10.3390/en15197010
However, transfer learning can lead to efficient accuracy and conver Alsuwian, T., Tayyeb, M., Amin, A. A., Qadir, M. B., Almasabi, S., & Jalalah, M. (2022).
gence in such cases. Design of a hybrid Fault-Tolerant control system for Air-Fuel ratio control of internal
combustion engines using genetic algorithm and Higher-Order sliding mode control.
Energies, 15(15), 5666. https://doi.org/10.3390/en15155666
13. Conclusion
Amarnath, M., Sugumaran, V., & Kumar, H. (2013). Exploiting sound signals for fault
diagnosis of bearings using decision tree. Measurement, 46(3), 1250–1256. https://
This review paper presented a methodical investigation of the doi.org/10.1016/j.measurement.2012.11.011
Amin, A. A., & Hasan, K. M. (2019). A review of fault tolerant control systems:
conception of FTC using transfer learning, deep learning, and machine
Advancements and applications. Measurement, 143, 58–68. https://doi.org/
learning techniques. The IFTC model’s difficulties and restrictions are 10.1016/j.measurement.2019.04.083
listed, along with any potential solutions offered by machine learning Amin, A. A., & Mahmood-Ul-Hasan, K. (2019). Advanced fault tolerant Air-Fuel ratio
theories. The main types of machine learning algorithms and their control of internal combustion gas engine for sensor and actuator faults. IEEE Access,
7, 17634–17643. https://doi.org/10.1109/ACCESS.2019.2894796
advanced versions, such as deep learning and transfer learning, were Amin, A. A., & Mahmood-ul-Hasan, K. (2019a). Hybrid fault tolerant control for air–fuel
explained. This study provides a primer for FTC researchers on the many ratio control of internal combustion gasoline engine using kalman filters with
types of machine learning algorithms and their more complex varia advanced redundancy. Measurement and Control, 52(5–6), 473–492. https://doi.org/
10.1177/0020294019842593
tions, such as deep learning and transfer learning. The difficulties and Amin, A. A., & Mahmood-ul-Hasan, K. (2019b). Robust active fault-tolerant control for
constraints of each of them draw attention to the contrasts between internal combustion gas engine for air–fuel ratio control with statistical regression-
them. The work is notable in that it guides researchers who want to use based observer model. Measurement and Control, 52(9–10), 1179–1194. https://doi.
org/10.1177/0020294018823031
IFTC for certain industrial problems by discussing the majority of key Amin, A. A., & Mahmood-ul-Hasan, K. (2021). Robust passive fault tolerant control for
literature references, in particular those connected to significant elec air fuel ratio control of internal combustion gasoline engine for sensor and actuator
trical and mechanical industrial difficulties. Last but not least, pro faults. IETE Journal of Research, 1–16. https://doi.org/10.1080/
03772063.2021.1906767
spective research directions for the growth of IFTC are mentioned.
Amin, A. A., & Mahmood-ul-Hasan, K. (2022). Unified Fault-Tolerant control for Air-Fuel
ratio control of internal combustion engines with advanced analytical and hardware
Funding redundancies. Journal of Electrical Engineering & Technology, 17(3), 1947–1959.
https://doi.org/10.1007/s42835-021-00965-5
An, B., Wang, S., Zhao, Z., Qin, F., Yan, R., & Chen, X. (2022). Interpretable neural
This research project was funded by the faculty research support network via algorithm unrolling for mechanical fault diagnosis. IEEE Transactions on
grant (FRSG-2023) program of FAST National University of Computer Instrumentation and Measurement, 71, 1–11. https://doi.org/10.1109/
and Emerging Sciences Pakistan. TIM.2022.3188058
An, X., & Tang, Y. (2017). Application of variational mode decomposition energy
distribution to bearing fault diagnosis in a wind turbine. Transactions of the Institute
Declaration of Competing Interest of Measurement and Control, 39(7), 1000–1006. https://doi.org/10.1177/
0142331215626247
Araste, Z., Sadighi, A., & Jamimoghaddam, M. (2023). Fault diagnosis of a centrifugal
The authors declare that they have no known competing financial pump using electrical signature analysis and support vector machine. Journal of
interests or personal relationships that could have appeared to influence Vibration Engineering & Technologies, 11(5), 2057–2067. https://doi.org/10.1007/
the work reported in this paper. s42417-022-00687-6
Asr, M. Y., Ettefagh, M. M., Hassannejad, R., & Razavi, S. N. (2017). Diagnosis of
combined faults in rotary machinery by Non-Naive bayesian approach. Mechanical
Data availability Systems and Signal Processing, 85, 56–70. https://doi.org/10.1016/j.
ymssp.2016.08.005
Bangalore, P., & Tjernberg, L. B. (2015). An artificial neural network approach for early
No data was used for the research described in the article. fault detection of gearbox bearings. IEEE Transactions on Smart Grid, 6(2), 980–987.
https://doi.org/10.1109/TSG.2014.2386305
References Barakat, M., El Badaoui, M., & Guillet, F. (2013). Hard competitive growing neural
network for the diagnosis of small bearing faults. Mechanical Systems and Signal
Processing, 37(1–2), 276–292. https://doi.org/10.1016/j.ymssp.2012.11.002
Abdelsalam, M., Diab, H., Tennakoon, S., & Griffiths, A. (2016). Reliability enhancement
Barakat, M., Lefebvre, D., Khalil, M., Druaux, F., & Mustapha, O. (2013). Parameter
of modular multilevel converter by applying fault tolerant control. In 2016 51st
selection algorithm with self adaptive growing neural network classifier for
International Universities Power Engineering Conference (UPEC) (pp. 1–5). https://doi.
diagnosis issues. International Journal of Machine Learning and Cybernetics, 4(3),
org/10.1109/UPEC.2016.8114008
217–233. https://doi.org/10.1007/s13042-012-0089-5
Ahmad, S., Ahmad, Z., & Kim, J.-M. (2022). A centrifugal pump fault diagnosis
Belosludtsev, V., Borovik, S., Danilchenko, V., & Sekisov, Y. (2021). Wear diagnostics of
framework based on supervised contrastive learning. Sensors, 22(17), 6448. https://
the thrust bearing of NK-33 Turbo-Pump unit on the basis of Single-Coil eddy current
doi.org/10.3390/s22176448
sensors. Sensors, 21(10), 3463. https://doi.org/10.3390/s21103463
Ahmad, T., Madonski, R., Zhang, D., Huang, C., & Mujeeb, A. (2022). Data-driven
Bernier, J. L., Ortega, J., Ros, E., Rojas, I., & Prieto, A. (2000). A quantitative study of
probabilistic machine learning in sustainable smart energy/smart energy systems:
fault tolerance, noise immunity, and generalization ability of MLPs. Neural
Key developments, challenges, and future research opportunities in the context of
Computation, 12(12), 2941–2964. https://doi.org/10.1162/089976600300014782
smart grid paradigm. Renewable and Sustainable Energy Reviews, 160, Article 112128.
Berredjem, T., & Benidir, M. (2018). Bearing faults diagnosis using fuzzy expert system
https://doi.org/10.1016/j.rser.2022.112128
relying on an improved range overlaps and similarity method. Expert Systems with
Ahmed, S., Amin, A. A., Wajid, Z., & Ahmad, F. (2020). Reliable speed control of a
Applications, 108, 134–142. https://doi.org/10.1016/j.eswa.2018.04.025
permanent magnet DC motor using fault-tolerant h-bridge. Advances in Mechanical
Cai, C., Weng, X., & Zhang, C. (2017). A novel approach for marine diesel engine fault
Engineering, 12(10). https://doi.org/10.1177/1687814020970311
diagnosis. Cluster Computing, 20(2), 1691–1702. https://doi.org/10.1007/s10586-
Ali, M. Z., Shabbir, M. N. S. K., Liang, X., Zhang, Y., & Hu, T. (2019). Machine Learning-
017-0748-0
Based fault diagnosis for single- and Multi-Faults in induction motors using
16
Cao, P., Zhang, S., & Tang, J. (2018). Preprocessing-Free gear fault diagnosis using small Gelgele, H. L., & Wang, K. (1998). An expert system for engine fault diagnosis:
datasets with deep convolutional neural Network-Based transfer learning. IEEE Development and application. Journal of Intelligent Manufacturing, 9(6), 539–545.
Access, 6, 26241–26253. https://doi.org/10.1109/ACCESS.2018.2837621 https://doi.org/10.1023/A:1008888219539
Cao, X., Tian, Y., Ji, X., & Qiu, B. (2021). Fault-Tolerant controller design for path Gerdes, M., Galar, D., & Scholz, D. (2016). Decision trees and the effects of feature
following of the autonomous vehicle under the faults in braking actuators. IEEE extraction parameters for robust sensor network design. Eksploatacja i Niezawodnosc -
Transactions on Transportation Electrification, 7(4), 2530–2540. https://doi.org/ Maintenance and Reliability, 19(1), 31–42. 10.17531/ein.2017.1.5.
10.1109/TTE.2021.3071725 Gharavian, M. H., Almas Ganj, F., Ohadi, A. R., & Heidari Bafroui, H. (2013). Comparison
Cassottana, B., Roomi, M. M., Mashima, D., & Sansavini, G. (n.d.). Resilience analysis of of FDA-based and PCA-based features in fault diagnosis of automobile gearboxes.
cyber-physical systems: A review of models and methods. Risk Analysis, n/a(n/a). Neurocomputing, 121, 150–159. https://doi.org/10.1016/j.neucom.2013.04.033
10.1111/risa.14089. Glowacz, A., & Glowacz, Z. (2017). Diagnosis of stator faults of the single-phase
Cervantes-Bobadilla, M., García-Morales, J., Saavedra-Benítez, Y. I., Hernández- induction motor using acoustic signals. Applied Acoustics, 117, 20–27. https://doi.
Pérez, J. A., Adam-Medina, M., Guerrero-Ramírez, G. V., & Escobar-Jímenez, R. F. org/10.1016/j.apacoust.2016.10.012
(2023). Multiple fault detection and isolation using artificial neural networks in Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. Cambridge, MA, USA:
sensors of an internal combustion engine. Engineering Applications of Artificial The MIT Press. Retrieved from https://www.deeplearningbook.org/.
Intelligence, 117, Article 105524. https://doi.org/10.1016/j.engappai.2022.105524 Gu, J., Wang, Z., Kuen, J., Ma, L., Shahroudy, A., Shuai, B., … Chen, T. (2018). Recent
Chen, C., Li, Z., Yang, J., & Liang, B. (2017). A cross domain feature extraction method advances in convolutional neural networks. Pattern Recognition, 77, 354–377.
based on transfer component analysis for rolling bearing fault diagnosis. 2017 29th https://doi.org/10.1016/j.patcog.2017.10.013
Chinese Control And Decision Conference (CCDC), 5622–5626. 10.1109/ Gunerkar, R. S., Jalan, A. K., & Belgamwar, S. U. (2019). Fault diagnosis of rolling
CCDC.2017.7978168. element bearing based on artificial neural network. Journal of Mechanical Science and
Chen, J., & Shi, Y. (2021). Stochastic model predictive control framework for resilient Technology, 33(2), 505–511. https://doi.org/10.1007/s12206-019-0103-x
cyber-physical systems: Review and perspectives. Philosophical Transactions of the Guo, D., Zhong, M., Ji, H., Liu, Y., & Yang, R. (2018). A hybrid feature model and deep
Royal Society A: Mathematical, Physical and Engineering Sciences, 379(2207), learning based fault diagnosis for unmanned aerial vehicle sensors. Neurocomputing,
20200371. https://doi.org/10.1098/rsta.2020.0371 319, 155–163. https://doi.org/10.1016/j.neucom.2018.08.046
Chen, Jinglong. (n.d.). Wavelet transform based on inner product in fault diagnosis of Guo, L., Lei, Y., Xing, S., Yan, T., & Li, N. (2019). Deep convolutional transfer learning
rotating machinery_A review. Mechanical Systems and Signal Processing, 35. network: A new method for intelligent fault diagnosis of machines with unlabeled
Chen, X., Zhou, J., Xiao, H., Wang, E., Xiao, J., & Zhang, H. (2014). Fault diagnosis based data. IEEE Transactions on Industrial Electronics, 66(9), 7316–7325. https://doi.org/
on comprehensive geometric characteristic and probability neural network. Applied 10.1109/TIE.2018.2877090
Mathematics and Computation, 230, 542–554. https://doi.org/10.1016/j. Guo, X., Shen, C., & Chen, L. (2017). Deep fault recognizer: an integrated model to
amc.2013.12.122 denoise and extract features for fault diagnosis in rotating machinery. Applied
Chen, Z., Chen, Y., Wu, L., Cheng, S., & Lin, P. (2019). Deep residual network based fault Sciences, 7(1), 41. https://doi.org/10.3390/app7010041
detection and diagnosis of photovoltaic arrays using current-voltage curves and H. Saltzer, J., & Kaashoek, M. F. (2021, August 1). 2.3: Measures of Reliability and
ambient conditions. Energy Conversion and Management, 198, Article 111793. Failure Tolerance. Retrieved August 23, 2023, from Engineering LibreTexts website:
https://doi.org/10.1016/j.enconman.2019.111793 https://eng.libretexts.org/Bookshelves/Computer_Science/Programming_and_
Chikkam, S., & Singh, S. (2023). Condition monitoring and fault diagnosis of induction Computation_Fundamentals/Principles_of_Computer_System_Design_(Saltzer_and_
motor using DWT and ANN. Arabian Journal for Science and Engineering, 48(5), Kaashoek)/02%3A_Fault_Tolerance_-_Reliable_Systems_from_Unreliable_
6237–6252. https://doi.org/10.1007/s13369-022-07294-3 Components/2.03%3A_Measures_of_Reliability_and_Failure_Tolerance.
Choudhary, A., Goyal, D., & Letha, S. S. (2021). Infrared Thermography-Based fault Hallajian, B., Motameni, H., & Akbari, E. (2022). Ensemble feature selection using
diagnosis of induction motor bearings using machine learning. IEEE Sensors Journal, distance-based supervised and unsupervised methods in binary classification. Expert
21(2), 1727–1734. https://doi.org/10.1109/JSEN.2020.3015868 Systems with Applications, 200, Article 116794. https://doi.org/10.1016/j.
Choudhary, A., Mian, T., & Fatima, S. (2021). Convolutional neural network based eswa.2022.116794
bearing fault diagnosis of rotating machine using thermal images. Measurement, 176, Han, D., Zhao, N., & Shi, P. (2017). A new fault diagnosis method based on deep belief
Article 109196. https://doi.org/10.1016/j.measurement.2021.109196 network and support vector machine with Teager-Kaiser energy operator for
Dai, J., Tang, J., Huang, S., & Wang, Y. (2019). Signal-Based intelligent hydraulic fault bearings. Advances in Mechanical Engineering, 9(12). https://doi.org/10.1177/
diagnosis methods: Review and prospects. Chinese Journal of Mechanical Engineering, 1687814017743113
32(1), 75. https://doi.org/10.1186/s10033-019-0388-9 Han, S., Zhou, J., Chen, Y., Zhang, Y., Tang, G., & Wang, L. (2020). active Fault-Tolerant
Dai, W., Yang, Q., Xue, G.-R., & Yu, Y. (2007). Boosting for transfer learning. Proceedings control for discrete vehicle active suspension via Reduced-Order observer. IEEE
of the 24th International Conference on Machine Learning - ICML ’07, 193–200. Transactions on Systems, Man, and Cybernetics: Systems, 1–11. https://doi.org/
Corvalis, Oregon: ACM Press. 10.1145/1273496.1273521. 10.1109/TSMC.2020.2964607
Deep Learning Enabled Fault Diagnosis Using Time-Frequency Image Analysis of Rolling Han, T., Liu, C., Yang, W., & Jiang, D. (2019). A novel adversarial learning framework in
Element Bearings. (n.d.). Retrieved February 19, 2021, from https://www.hindawi. deep convolutional neural network for intelligent diagnosis of mechanical faults.
com/journals/sv/2017/5067651/. Knowledge-Based Systems, 165, 474–487. https://doi.org/10.1016/j.
Diagnosis and Fault-tolerant Control 2: From Fault Diagnosis to Fault-tolerant Control (1st knosys.2018.12.019
ed.). (2021). Wiley. 10.1002/9781119882350. Han, Y., Tang, B., & Deng, L. (2018). Multi-level wavelet packet fusion in dynamic
Diaz, M., Henriquez, P., Ferrer, M. A., Pirlo, G., Alonso, J. B., Carmona-Duarte, C., & ensemble convolutional neural network for fault diagnosis. Measurement, 127,
Impedovo, D. (2017). Stability-based system for bearing fault early detection. Expert 246–255. https://doi.org/10.1016/j.measurement.2018.05.098
Systems with Applications, 79, 65–75. https://doi.org/10.1016/j.eswa.2017.02.030 Hasan, M. J., & Kim, J.-M. (2018). Bearing fault diagnosis under variable rotational
Ding, X., & He, Q. (2017). Energy-Fluctuated multiscale feature learning with deep speeds using stockwell Transform-Based vibration imaging and transfer learning.
ConvNet for intelligent spindle bearing fault diagnosis. IEEE Transactions on Applied Sciences, 8(12), 2357. https://doi.org/10.3390/app8122357
Instrumentation and Measurement, 66(8), 1926–1935. https://doi.org/10.1109/ He, K., Zhang, X., Ren, S., & Sun, J. (2016). In Deep Residual Learning for Image Recognition
TIM.2017.2674738 (pp. 770–778). Las Vegas, NV, USA: IEEE. https://doi.org/10.1109/CVPR.2016.90.
Dogan, A., & Birant, D. (2021). Machine learning and data mining in manufacturing. He, Y.-L., Wang, R., Kwong, S., & Wang, X.-Z. (2014). Bayesian classifiers based on
Expert Systems with Applications, 166, Article 114060. https://doi.org/10.1016/j. probability density estimation and their applications to simultaneous fault diagnosis.
eswa.2020.114060 Information Sciences, 259, 252–268. https://doi.org/10.1016/j.ins.2013.09.003
Dong, L., Li, W., Wang, C.-H., & Lin, K.-P. (2020). Gyro motor fault classification model Hendrickx, K., Meert, W., Mollet, Y., Gyselinck, J., Cornelis, B., Gryllias, K., & Davis, J.
based on a coupled hidden Markov model with a minimum intra-class distance (2020). A general anomaly detection framework for fleet-based condition
algorithm. Proceedings of the Institution of Mechanical Engineers, Part I: Journal of monitoring of machines. Mechanical Systems and Signal Processing, 139, Article
Systems and Control Engineering, 234(5), 646–661. 10.1177/0959651819866281. 106585. https://doi.org/10.1016/j.ymssp.2019.106585
Dong, S., Luo, T., Zhong, L., Chen, L., & Xu, X. (2017). Fault diagnosis of bearing based Heo, S., & Lee, J. H. (2018). Fault detection and classification using artificial neural
on the Kernel principal component analysis and optimized k -nearest neighbour networks. IFAC-PapersOnLine, 51(18), 470–475. https://doi.org/10.1016/j.
model. Journal of Low Frequency Noise, Vibration and Active Control, 36(4), 354–365. ifacol.2018.09.380
https://doi.org/10.1177/1461348417744302 Hongyan dui. (2015). Reliability optimization of automatic control systems based on
Dong, S., Xu, X., & Chen, R. (2016). Application of fuzzy c-means method and importance measures: A framework. International Journal of Performability
classification model of optimized k-nearest neighbor for fault diagnosis of bearing. Engineering, 12(3), 297–300.
Journal of the Brazilian Society of Mechanical Sciences and Engineering, 38(8), Ince, T., Kiranyaz, S., Eren, L., Askar, M., & Gabbouj, M. (2016). Real-Time motor fault
2255–2263. https://doi.org/10.1007/s40430-015-0455-9 detection by 1-D convolutional neural networks. IEEE Transactions on Industrial
Du, D., & Li, Z. (2023). Passive fault-tolerant control for discrete parameter system. IET Electronics, 63(11), 7067–7075. https://doi.org/10.1109/TIE.2016.2582729
Power Electronics, 16(12), 1969–1983. https://doi.org/10.1049/pel2.12515 Iqbal, M. S., & Amin, A. A. (2022). Genetic algorithm based active fault-tolerant control
Duan, Z., Wu, T., Guo, S., Shao, T., Malekian, R., & Li, Z. (2018). Development and trend system for air fuel ratio control of internal combustion engines. Measurement and
of condition monitoring and fault diagnosis of multi-sensors information fusion for Control, 55(7–8), 703–716. https://doi.org/10.1177/00202940221115233
rolling bearings: A review. The International Journal of Advanced Manufacturing Islam, M. M. M., & Kim, J.-M. (2019). Automated bearing fault diagnosis scheme using
Technology, 96(1–4), 803–819. https://doi.org/10.1007/s00170-017-1474-8 2D representation of wavelet packet transform and deep convolutional neural
Dubrova, E. (2013). Dependability evaluation techniques. In Fault-Tolerant Design (pp. network. Computers in Industry, 106, 142–153. https://doi.org/10.1016/j.
19–42). New York: Springer-Verlag. compind.2019.01.008
Shen, F., Chen, C., Yan, R., & Gao, R. X. (2015). In Bearing fault diagnosis based on SVD James, A. T., Gandhi, O. P., & Deshmukh, S. G. (2018). fault diagnosis of automobile
feature extraction and transfer learning classification (pp. 1–6). Beijing, China: IEEE. systems using fault tree based on digraph modeling. International Journal of System
https://doi.org/10.1109/PHM.2015.7380088.
17
Assurance Engineering and Management, 9(2), 494–508. https://doi.org/10.1007/ Li, Y., Yang, Y., Wang, X., Liu, B., & Liang, X. (2018). Early fault diagnosis of rolling
s13198-017-0693-6 bearings based on hierarchical symbol dynamic entropy and binary tree support
Jia, F., Lei, Y., Guo, L., Lin, J., & Xing, S. (2018). A neural network constructed by deep vector machine. Journal of Sound and Vibration, 428, 72–86. https://doi.org/
learning technique and its application to intelligent fault diagnosis of machines. 10.1016/j.jsv.2018.04.036
Neurocomputing, 272, 619–628. https://doi.org/10.1016/j.neucom.2017.07.032 Li, Z., Yan, X., Tian, Z., Yuan, C., Peng, Z., & Li, L. (2013). Blind vibration component
Jia, Y., Xu, M., & Wang, R. (2018). Symbolic important point perceptually and hidden separation and nonlinear feature extraction applied to the nonstationary vibration
markov model based hydraulic pump fault diagnosis method. Sensors, 18(12), 4460. signals for the gearbox multi-fault diagnosis. Measurement, 46(1), 259–271. https://
https://doi.org/10.3390/s18124460 doi.org/10.1016/j.measurement.2012.06.013
Jiang, F., Zhu, Z., Li, W., Ren, Y., Zhou, G., & Chang, Y. (2018). A fusion feature Lin, T. R., Tan, A. C., Ma, L., & Mathew, J. (2015). Condition monitoring and fault
extraction method using EEMD and correlation coefficient analysis for bearing fault diagnosis of diesel engines using instantaneous angular speed analysis. Proceedings of
diagnosis. Applied Sciences, 8(9), 1621. https://doi.org/10.3390/app8091621 the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering
Jiang, G., He, H., Yan, J., & Xie, P. (2019). Multiscale convolutional neural networks for Science, 229(2), 304–315. https://doi.org/10.1177/0954406214533253
fault diagnosis of wind turbine gearbox. IEEE Transactions on Industrial Electronics, 66 Long, M., Wang, J., Cao, Y., Sun, J., & Yu, P. S. (2016). Deep learning of transferable
(4), 3196–3207. https://doi.org/10.1109/TIE.2018.2844805 representation for scalable domain adaptation. IEEE Transactions on Knowledge and
Jiang, L., Shi, T., & Xuan, J. (2014). Fault diagnosis of rolling bearings based on marginal Data Engineering, 99, 14.
fisher analysis. Journal of Vibration and Control, 20(3), 470–480. https://doi.org/ Long, M., Wang, J., Ding, G., Sun, J., & Yu, P. S. (2013). Transfer feature learning with
10.1177/1077546312463747 joint distribution adaptation. In 2013 IEEE International Conference on Computer
Jiang, L., Xuan, J., & Shi, T. (2013). Feature extraction based on semi-supervised kernel Vision (pp. 2200–2207). Sydney, Australia: IEEE. https://doi.org/10.1109/
marginal fisher analysis and its application in bearing fault diagnosis. Mechanical ICCV.2013.274.
Systems and Signal Processing, 41(1–2), 113–126. https://doi.org/10.1016/j. Long, M., Zhu, H., Wang, J., & Jordan, M. I. (2017). Deep transfer learning with joint
ymssp.2013.05.017 adaptation networks. Proceedings of the 34th International Conference on Machine
Jiao, J., Zhao, M., Lin, J., & Zhao, J. (2018). A multivariate encoder information based Learning - Volume 70, 2208–2217. Sydney, NSW, Australia: JMLR.org.
convolutional neural network for intelligent fault diagnosis of planetary gearboxes. Lu, C., Wang, Z., & Zhou, B. (2017). Intelligent fault diagnosis of rolling bearing using
Knowledge-Based Systems, 160, 237–250. https://doi.org/10.1016/j. hierarchical convolutional network based health state classification. Advanced
knosys.2018.07.017 Engineering Informatics, 32, 139–151. https://doi.org/10.1016/j.aei.2017.02.005
Jin, Y., Hou, L., Chen, Y., & Lu, Z. (2022). An effective crack position diagnosis method Lu, C., Wang, Z.-Y., Qin, W.-L., & Ma, J. (2017). Fault diagnosis of rotary machinery
for the hollow shaft rotor system based on the convolutional neural network and components using a stacked denoising autoencoder-based health state identification.
deep metric learning. Chinese Journal of Aeronautics, 35(9), 242–254. https://doi. Signal Processing, 130, 377–388. https://doi.org/10.1016/j.sigpro.2016.07.028
org/10.1016/j.cja.2021.09.010 Lu, W., Liang, B., Cheng, Y., Meng, D., Yang, J., & Zhang, T. (2017). Deep model based
Kafeel, A., Aziz, S., Awais, M., Khan, M. A., Afaq, K., Idris, S. A., … Mostafa, S. M. (2021). domain adaptation for fault diagnosis. IEEE Transactions on Industrial Electronics, 64
An expert system for rotating machine fault detection using vibration signal analysis. (3), 2296–2305. https://doi.org/10.1109/TIE.2016.2627020
Sensors, 21(22), 7587. https://doi.org/10.3390/s21227587 Ma, J., Xu, F., Huang, K., & Huang, R. (2017). GNAR-GARCH model and its application in
Khan, S., & Yairi, T. (2018). A review on the application of deep learning in system feature extraction for rolling bearing fault diagnosis. Mechanical Systems and Signal
health management. Mechanical Systems and Signal Processing, 107, 241–265. Processing, 93, 175–203. https://doi.org/10.1016/j.ymssp.2017.01.043
https://doi.org/10.1016/j.ymssp.2017.11.024 Ma, M., Sun, C., & Chen, X. (2018). Deep coupling autoencoder for fault diagnosis with
Khazaee, M., Banakar, A., Ghobadian, B., Mirsalim, M., Minaei, S., Jafari, M., & multimodal sensory data. IEEE Transactions on Industrial Informatics, 14(3),
Sharghi, P. (2016). Fault detection of engine timing belt based on vibration signals 1137–1145. https://doi.org/10.1109/TII.2018.2793246
using data-mining techniques and a novel data fusion procedure. Structural Health Ma, S., Chu, F., & Han, Q. (2019). Deep residual learning with demodulated time-
Monitoring, 15(5), 583–598. https://doi.org/10.1177/1475921716652582 frequency features for fault diagnosis of planetary gearbox under nonstationary
Kim, J.-H., Choi, J.-H., Park, Y.-H., Leung, C.-K.-S., & Nasridinov, A. (2021). KNN-SC: running conditions. Mechanical Systems and Signal Processing, 127, 190–201. https://
Novel spectral clustering algorithm using k-Nearest neighbors. IEEE Access, 9, doi.org/10.1016/j.ymssp.2019.02.055
152616–152627. https://doi.org/10.1109/ACCESS.2021.3126854 Machine learning, explained. (n.d.). Retrieved December 20, 2022, from MIT Sloan
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017). ImageNet classification with deep website: https://mitsloan.mit.edu/ideas-made-to-matter/machine-learning-
convolutional neural networks. Communications of the ACM, 60(6), 84–90. https:// explained.
doi.org/10.1145/3065386 Matasci, G., Volpi, M., Kanevski, M., Bruzzone, L., & Tuia, D. (2015). Semisupervised
Kumar, V., Singh, L., & Tripathi, A. K. (2018). Reliability analysis of safety-critical and transfer component analysis for domain adaptation in remote sensing image
control systems: A state-of-the-art review. IET Software, 12(1), 1–18. https://doi.org/ classification. IEEE Transactions on Geoscience and remote sensing, 53(7), 3550–3564.
10.1049/iet-sen.2017.0053 https://doi.org/10.1109/TGRS.2014.2377785
LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444. Mayer-Schönberger, V., & Cukier, K. (2013). Big data: A revolution that will transform how
https://doi.org/10.1038/nature14539 we live, work, and think. Houghton Mifflin Harcourt.
Li, C., Xiong, G., Fu, X., Mohamed, A. W., Yuan, X., Al-Betar, M. A., & Suganthan, P. N. Mihály, N.-B., Luca, A.-V., & Cristea, V. M. (2023). Artificial neural networks-based
(2022). Takagi-Sugeno fuzzy based power system fault section diagnosis models via identification of the WWTP DO sensor types of faults. In Computer Aided Chemical
genetic learning adaptive GSK algorithm. Knowledge-Based Systems, 255, Article Engineering (Vol. 52, pp. 1879–1884). Elsevier. https://doi.org/10.1016/B978-0-
109773. https://doi.org/10.1016/j.knosys.2022.109773 443-15274-0.50298-5.
Li, Chuan. (n.d.). Gearbox fault diagnosis based on deep random forest fusion of acoustic Naha, A., Samanta, A. K., Routray, A., & Deb, A. K. (2017). Low complexity motor
and vibratory signals. Mechanical Systems and Signal Processing, 11. current signature analysis using Sub-Nyquist strategy with reduced data length. IEEE
Li, G., Gu, Y., & Ding, J. (2020, October 2). The Efficacy of $L_1$ Regularization in Two- Transactions on Instrumentation and Measurement, 66(12), 3249–3259. https://doi.
Layer Neural Networks. arXiv. Retrieved from http://arxiv.org/abs/2010.01048. org/10.1109/TIM.2017.2737879
Li, H., & Zhao, Q. (2005). Reliability modeling of fault tolerant control systems. In Namigtle-Jiménez, A., Escobar-Jiménez, R. F., Gómez-Aguilar, J. F., García-
Proceedings of the 44th IEEE Conference on Decision and control (pp. 2397–2402). Beltrán, C. D., & Téllez-Anguiano, A. C. (2020). Online aNN-based fault diagnosis
https://doi.org/10.1109/CDC.2005.1582521 implementation using an FPGA: Application in the EFI system of a vehicle. ISA
Li, J., Qu, Y., Hong, L., & He, D. (2018). PSO optimized ANN diagnosis of early gear Transactions, 100, 358–372. https://doi.org/10.1016/j.isatra.2019.11.003
pitting. In 2018 Prognostics and System Health Management Conference (PHM- Nezamivand Chegini, S., Haghdoust Manjili, M. J., Ahmadi, B., Amirmostofian, I., &
Chongqing), 1043–1048. Chongqing: IEEE. https://doi.org/10.1109/PHM- Bagheri, A. (2021). New bearing slight degradation detection approach based on the
Chongqing.2018.00185. periodicity intensity factor and signal processing methods. Measurement, 170, Article
Li, S., Liu, G., Tang, X., Lu, J., & Hu, J. (2017). An ensemble deep convolutional neural 108696. https://doi.org/10.1016/j.measurement.2020.108696
network model with improved d-S evidence fusion for bearing fault diagnosis. Oh, H., Lee, Y., Lee, J., Joo, C., & Lee, C. (2022). Feature selection algorithm based on
Sensors, 17(8), 1729. https://doi.org/10.3390/s17081729 density and distance for fault diagnosis applied to a roll-to-roll manufacturing
Li, W., Monti, A., & Ponci, F. (2014). Fault detection and classification in medium voltage system. Journal of Computational Design and Engineering, 9(2), 805–825. https://doi.
DC shipboard power systems with wavelets and artificial neural networks. IEEE org/10.1093/jcde/qwac028
Transactions on Instrumentation and Measurement, 63(11), 2651–2665. https://doi. Oroojlooy, A., & Hajinezhad, D. (2022). A review of cooperative multi-agent deep
org/10.1109/TIM.2014.2313035 reinforcement learning. Applied Intelligence, 1–46. https://doi.org/10.1007/s10489-
Li, X., Zhang, W., & Ding, Q. (2019). Cross-Domain fault diagnosis of rolling element 022-04105-y
bearings using deep generative neural networks. IEEE Transactions on Industrial Pan, S. J., & Yang, Q. (2010). A survey on transfer learning. IEEE Transactions on
Electronics, 66(7), 5525–5534. https://doi.org/10.1109/TIE.2018.2868023 Knowledge and Data Engineering, 22(10), 1345–1359. https://doi.org/10.1109/
Li, X., Zhang, W., & Ding, Q. (2018). A robust intelligent fault diagnosis method for TKDE.2009.191
rolling element bearings based on deep distance metric learning. Neurocomputing, Pan, S. J., Tsang, I. W., Kwok, J. T., & Yang, Q. (2011). Domain adaptation via transfer
310, 77–95. https://doi.org/10.1016/j.neucom.2018.05.021 component analysis. IEEE Transactions on Neural Networks, 22(2), 199–210. https://
Li, Y., Xu, M., Zhao, H., & Huang, W. (2016). Hierarchical fuzzy entropy and improved doi.org/10.1109/TNN.2010.2091281
support vector machine based binary tree approach for rolling bearing fault Pang, B., Tang, G., Zhou, C., & Tian, T. (2018). Rotor fault diagnosis based on
diagnosis. Mechanism and Machine Theory, 98, 114–132. https://doi.org/10.1016/j. characteristic frequency band energy entropy and support vector machine. Entropy,
mechmachtheory.2015.11.010 20(12), 932. https://doi.org/10.3390/e20120932
Li, Y., Yang, Y., Li, G., Xu, M., & Huang, W. (2017). A fault diagnosis scheme for Park, S., Kim, S., & Choi, J.-H. (2018). Gear fault diagnosis using transmission error and
planetary gearboxes using modified multi-scale symbolic dynamic entropy and ensemble empirical mode decomposition. Mechanical Systems and Signal Processing,
mRMR feature selection. Mechanical Systems and Signal Processing, 91, 295–312. 108, 262–275. https://doi.org/10.1016/j.ymssp.2018.02.028
https://doi.org/10.1016/j.ymssp.2016.12.040
18
Pasqualotto, D., & Zigliotto, M. (2022). increasing feasibility of neural Network-Based Sun, L., Wang, T., Ding, W., Xu, J., & Lin, Y. (2021). Feature selection using fisher score
early fault detection in induction motor drives. IEEE Journal of Emerging and Selected and multilabel neighborhood rough sets for multilabel classification. Information
Topics in Power Electronics, 10(2), 2042–2051. https://doi.org/10.1109/ Sciences, 578, 887–912. https://doi.org/10.1016/j.ins.2021.08.032
JESTPE.2021.3115170 Sun, L., Yin, T., Ding, W., Qian, Y., & Xu, J. (2022). Feature selection with missing labels
Peebles, P. Z. (2002). Probability. In Probability, Random Variables, and Random Signal using multilabel fuzzy neighborhood rough sets and maximum relevance minimum
Principles, 4th ed. (4th Edition edition). TMH. redundancy. IEEE Transactions on fuzzy Systems, 30(5), 1197–1211. https://doi.org/
Peña, M., Cerrada, M., Alvarez, X., Jadán, D., Lucero, P., Milton, B., … Sánchez, R.-V. 10.1109/TFUZZ.2021.3053844
(2018). Feature engineering based on ANOVA, cluster validity assessment and KNN Tang, B., Song, T., Li, F., & Deng, L. (2014). Fault diagnosis for a wind turbine
for fault diagnosis in bearings. Journal of Intelligent & Fuzzy Systems, 34(6), transmission system based on manifold learning and shannon wavelet support vector
3451–3462. https://doi.org/10.3233/JIFS-169525 machine. Renewable Energy, 62, 1–9. https://doi.org/10.1016/j.renene.2013.06.025
Pudjihartono, N., Fadason, T., Kempa-Liehr, A. W., & O’Sullivan, J. M. (2022). A review Tang, G., Pang, B., Tian, T., & Zhou, C. (2018). Fault diagnosis of rolling bearings based
of feature selection methods for machine Learning-Based disease risk prediction. on improved fast spectral correlation and optimized random forest. Applied Sciences,
Frontiers in Bioinformatics, 2, Article 927312. https://doi.org/10.3389/ 8(10), 1859. https://doi.org/10.3390/app8101859
fbinf.2022.927312 Transfer Learning With Neural Networks for Bearing Fault Diagnosis in Changing
Qiao, Z., Lei, Y., & Li, N. (2019). Applications of stochastic resonance to machinery fault Working Conditions—IEEE Journals & Magazine. (n.d.). Retrieved February 18,
detection: A review and tutorial. Mechanical Systems and Signal Processing, 122, 2021, from https://ieeexplore.ieee.org/abstract/document/7961149.
502–536. https://doi.org/10.1016/j.ymssp.2018.12.032 Urbanowicz, R. J., Meeker, M., La Cava, W., Olson, R. S., & Moore, J. H. (2018). Relief-
Qu, Y., He, M., Deutsch, J., & He, D. (2017). Detection of pitting in gears using a deep based feature selection: Introduction and review. Journal of Biomedical Informatics,
sparse autoencoder. Applied Sciences, 7(5), 515. https://doi.org/10.3390/ 85, 189–203. https://doi.org/10.1016/j.jbi.2018.07.014
app7050515 Van, M., & Kang, H.-J. (2016). Two-stage feature selection for bearing fault diagnosis
Rai, A., & Upadhyay, S. H. (2016). A review on signal processing techniques utilized in based on dual-tree complex wavelet transform and empirical mode decomposition.
the fault diagnosis of rolling element bearings. Tribology international, 96, 289–306. Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical
https://doi.org/10.1016/j.triboint.2015.12.037 Engineering Science, 230(2), 291–302. https://doi.org/10.1177/0954406215573976
Research on Degradation State Recognition of Planetary Gear Based on Multiscale Vanraj, Dhami, S., & Pabla, B. (2018). Hybrid data fusion approach for fault diagnosis of
Information Dimension of SSD and CNN. (n.d.). Retrieved February 19, 2021, from fixed-axis gearbox. Structural Health Monitoring, 17(4), 936–945. 10.1177/
https://www.hindawi.com/journals/complexity/2019/8716979/. 1475921717727700.
Reyes, O., Morell, C., & Ventura, S. (2015). Scalable extensions of the ReliefF algorithm Venkateswara, H., Chakraborty, S., & Panchanathan, S. (2017). Deep-Learning systems
for weighting and selecting features on the multi-label learning context. for domain adaptation in computer vision: Learning transferable feature
Neurocomputing, 161, 168–182. https://doi.org/10.1016/j.neucom.2015.02.045 representations. IEEE Signal Processing Magazine, 34(6), 117–129. https://doi.org/
Rezvani, S., Wang, X., & Pourpanah, F. (2019). Intuitionistic fuzzy twin support vector 10.1109/MSP.2017.2740460
machines. IEEE Transactions on fuzzy Systems, 27(11), 2140–2151. https://doi.org/ Wang, F., Qian, Z., Yan, Z., Yuan, C., & Zhang, W. (2020). A novel resilient robot:
10.1109/TFUZZ.2019.2893863 Kinematic analysis and experimentation. IEEE Access, 8, 2885–2892. https://doi.
Riaz, U., Amin, A. A., & Tayyeb, M. (2022). Design of active fault-tolerant control system org/10.1109/ACCESS.2019.2962058
for air-fuel ratio control of internal combustion engines using fuzzy logic controller. Wang, J., Xie, J., Zhang, L., & Duan, L. (2016). A factor analysis based transfer learning
Science Progress, 105(2). https://doi.org/10.1177/00368504221094723 method for gearbox diagnosis under various operating conditions. International
Rostaghi, M., Khatibi, M. M., Ashory, M. R., & Azami, H. (2021). Bearing fault diagnosis Symposium on Flexible Automation (ISFA), 2016, 81–86. https://doi.org/10.1109/
using refined composite generalized multiscale dispersion Entropy-Based skewness ISFA.2016.7790140
and variance and multiclass FCM-ANFIS. Entropy, 23(11), 1510. https://doi.org/ Wang, R., Chen, H., & Guan, C. (2021). A bayesian inference-based approach for
10.3390/e23111510 performance prognostics towards uncertainty quantification and its applications on
Safizadeh, M. S., & Latifi, S. K. (2014). Using multi-sensor data fusion for vibration fault the marine diesel engine. ISA Transactions, 118, 159–173. https://doi.org/10.1016/j.
diagnosis of rolling element bearings by accelerometer and load cell. Information isatra.2021.02.024
Fusion, 18, 1–8. https://doi.org/10.1016/j.inffus.2013.10.002 Wang, S., Xiang, J., Zhong, Y., & Zhou, Y. (2018). Convolutional neural network-based
Safizadeh, M. S., & Yari, B. (2017). Pump cavitation detection through fusion of support hidden markov models for rolling element bearing fault identification. Knowledge-
vector machine classifier data associated with vibration and motor current signature. Based Systems, 144, 65–76. https://doi.org/10.1016/j.knosys.2017.12.027
59(12), 5. Wang, T., Wei, X., Wang, J., Huang, T., Peng, H., Song, X., … Pérez-Jiménez, M. J.
Saud, A. S., & Shakya, S. (2021). Analysis of L2 regularization hyper parameter for stock (2020). A weighted corrective fuzzy reasoning spiking neural P system for fault
price prediction. Journal of Institute of Science and Technology, 26(1), 83–88. https:// diagnosis in power systems with variable topologies. Engineering Applications of
doi.org/10.3126/jist.v26i1.37830 Artificial Intelligence, 92, Article 103680. https://doi.org/10.1016/j.
Saufi, S. R., Ahmad, Z. A. bin, Leong, M. S., & Lim, M. H. (2018). Differential evolution engappai.2020.103680
optimization for resilient stacked sparse autoencoder and its applications on bearing Wang, X. (2020). Active fault tolerant control for unmanned underwater vehicle with
fault diagnosis. Measurement Science and Technology, 29(12), 125002. 10.1088/1361- sensor faults. IEEE Transactions on Instrumentation and Measurement, 69(12),
6501/aae5b2. 9485–9495. https://doi.org/10.1109/TIM.2020.3003108
Savchynskyy, B. (2019). Discrete graphical models: An optimization perspective (B. Curless, Wang, X., He, H., & Li, L. (2019). A hierarchical deep domain adaptation approach for
W. T. Freeman, & L. V. Gool, Eds.). Boston Delft: Now. fault diagnosis of power plant thermal system. IEEE Transactions on Industrial
Schmidhuber, J. (2015). Deep learning in neural networks: an overview. Neural Informatics, 15(9), 5139–5148. https://doi.org/10.1109/TII.2019.2899118
Networks, 61, 85–117. https://doi.org/10.1016/j.neunet.2014.09.003 Wang, Y., Xiang, J., Markert, R., & Liang, M. (2016). Spectral kurtosis for fault detection,
Shahbaz, M. H., & Amin, A. A. (2021). Design of active fault tolerant control system for diagnosis and prognostics of rotating machines: A review with applications.
air fuel ratio control of internal combustion engines using artificial neural networks. Mechanical Systems and Signal Processing, 66–67, 679–698. https://doi.org/10.1016/
IEEE Access, 9, 46022–46032. https://doi.org/10.1109/ACCESS.2021.3068164 j.ymssp.2015.04.039
Shahbaz, M. H., & Amin, A. A. (2023). Design of hybrid fault-tolerant control system for Wang, Y., Zhou, D., Qin, S. J., & Wang, H. (2008). Active Fault-Tolerant control for a
air-fuel ratio control of internal combustion engines using artificial neural network class of nonlinear systems with sensor faults. International Journal of control,
and sliding mode control against sensor faults. Advances in Mechanical Engineering, 15 Automation, and systems, 6(3), 339–350.
(3). https://doi.org/10.1177/16878132231160729 Wang, Z., & Fan, J. (2015). Fault early recognition and health monitoring on aeroengine
Shao, H., Jiang, H., Wang, F., & Wang, Y. (2017). Rolling bearing fault diagnosis using rotor system. Journal of Aerospace Engineering, 28(2), 04014065. https://doi.org/
adaptive deep belief network with dual-tree complex wavelet packet. ISA 10.1061/(ASCE)AS.1943-5525.0000386
Transactions, 69, 187–201. https://doi.org/10.1016/j.isatra.2017.03.017 Wang, Z., Zhang, Q., Xiong, J., Xiao, M., Sun, G., & He, J. (2017). Fault diagnosis of a
Shao, H., Jiang, H., Zhao, H., & Wang, F. (2017). A novel deep autoencoder feature rolling bearing using wavelet packet denoising and random forests. IEEE Sensors
learning method for rotating machinery fault diagnosis. Mechanical Systems and Journal, 17(17), 5581–5588. https://doi.org/10.1109/JSEN.2017.2726011
Signal Processing, 95, 187–204. https://doi.org/10.1016/j.ymssp.2017.03.034 Warde-Farley, D., & Xu, B. (2014). Generative adversarial nets. Advances in Neural
Shao, S., McAleer, S., Yan, R., & Baldi, P. (2019). Highly accurate machine fault diagnosis Information Processing Systems (NIPS).
using deep transfer learning. IEEE Transactions on Industrial Informatics, 15(4), Weiser, C., & Ossmann, D. (2022). Fault-Tolerant control for a high altitude long
2446–2455. https://doi.org/10.1109/TII.2018.2864759 endurance aircraft. IFAC-PapersOnLine, 55(6), 724–729. https://doi.org/10.1016/j.
Shi, P., Liang, K., Han, D., & Zhang, Y. (2017). A novel intelligent fault diagnosis method ifacol.2022.07.213
of rotating machinery based on deep learning and PSO-SVM. Journal of Weiss, K., Khoshgoftaar, T. M., & Wang, D. (2016). A survey of transfer learning. Journal
Vibroengineering, 19(8), 5932–5946. 10.21595/jve.2017.18380. of Big Data, 3(1), 9. https://doi.org/10.1186/s40537-016-0043-6
Shi, Q., & Zhang, H. (2021). Fault diagnosis of an autonomous vehicle with an improved Wen, L., Gao, L., & Li, X. (2019). A new deep transfer learning based on sparse Auto-
SVM algorithm subject to unbalanced datasets. IEEE Transactions on Industrial Encoder for fault diagnosis. IEEE Transactions on Systems, Man, and Cybernetics:
Electronics, 68(7), 6248–6256. https://doi.org/10.1109/TIE.2020.2994868 Systems, 49(1), 136–144. https://doi.org/10.1109/TSMC.2017.2754287
Shifat, T. A., & Hur, J.-W. (2021). ANN assisted multi sensor information fusion for BLDC Wong, P. K., Zhong, J.-H., Yang, Z.-X., & Vong, C. M. (n.d.). A new framework for
motor fault diagnosis. IEEE Access, 9, 9429–9441. https://doi.org/10.1109/ intelligent simultaneous-fault diagnosis of rotating machinery using pairwise-
ACCESS.2021.3050243 coupled sparse Bayesian extreme learning committee machine. 16.
Singh, M., & Shaik, A. G. (2019). Faulty bearing detection, classification and location in a Xiao, Y., Chen, L., Chen, S., Hu, Z., & Tang, J. (2023). Intelligent fault diagnosis of gear
three-phase induction motor based on stockwell transform and support vector crack based on side frequency feature under different working conditions.
machine. Measurement, 131, 524–533. https://doi.org/10.1016/j. Measurement Science and Technology, 34(9), Article 095013. https://doi.org/
measurement.2018.09.013 10.1088/1361-6501/acd9df
19
Xie, J., Du, G., Shen, C., Chen, N., Chen, L., & Zhu, Z. (2018). An End-to-End model based classifier. Journal of Mechanical Science and Technology, 32(1), 37–47. https://doi.
on improved adaptive deep belief network and its application to bearing fault org/10.1007/s12206-017-1205-y
diagnosis. IEEE Access, 6, 63584–63596. https://doi.org/10.1109/ Yu, J., Ding, B., & He, Y. (2018). Rolling bearing fault diagnosis based on mean
ACCESS.2018.2877447 multigranulation decision-theoretic rough set and non-naive bayesian classifier.
Xie, J., Zhang, L., Duan, L., & Wang, J. (2016). on cross-domain feature fusion in gearbox Journal of Mechanical Science and Technology, 32(11), 5201–5211. https://doi.org/
fault diagnosis under various operating conditions based on transfer component 10.1007/s12206-018-1018-7
analysis. IEEE International Conference on Prognostics and Health Management Yu, J., Huang, W., & Zhao, X. (2016). Combined flow graphs and normal naive bayesian
(ICPHM), 2016, 1–6. https://doi.org/10.1109/ICPHM.2016.7542845 classifier for fault diagnosis of gear box. Proceedings of the Institution of Mechanical
Xie, Y., & Zhang, T. (2018). A transfer learning strategy for rotation machinery fault Engineers, Part C: Journal of Mechanical Engineering Science, 230(2), 303–313. https://
diagnosis based on Cycle-Consistent generative adversarial networks. Chinese doi.org/10.1177/0954406215575582
Automation Congress (CAC), 2018, 1309–1313. https://doi.org/10.1109/ Yu, Z., Zhang, Y., Jiang, B., Fu, J., & Jin, Y. (2022). A review on fault-tolerant
CAC.2018.8623346 cooperative control of multiple unmanned aerial vehicles. Chinese Journal of
Xiong, C., Su, M., Jiang, Z., & Jiang, W. (2019). Prediction of hemodialysis timing based Aeronautics, 35(1), 1–18. https://doi.org/10.1016/j.cja.2021.04.022
on LVW feature selection and ensemble learning. Journal of Medical Systems, 43(1), Zhai, Y.-J., Yu, D.-L., Qian, K.-J., Lee, S., & Theera-Umpon, N. (2017). A soft Sensor-
18. https://doi.org/10.1007/s10916-018-1136-x Based Fault-Tolerant control on the air fuel ratio of Spark-Ignition engines. Energies,
Xu, Q., Zhang, P., Liu, W., Liu, Q., Liu, C., Wang, L., … Joe Qin, S. (2018). A platform for 10(1), 131. https://doi.org/10.3390/en10010131
fault diagnosis of High-Speed train based on big data. IFAC-PapersOnLine, 51(18), Zhang, S., Yang, P., Kong, L., Chen, W., Fu, Q., & Peng, K. (2021). Neural Networks-Based
309–314. https://doi.org/10.1016/j.ifacol.2018.09.318 fault tolerant control of a robot via fast terminal sliding mode. IEEE Transactions on
Xu, X., Lei, Y., & Li, Z. (2020). An incorrect data detection method for big data cleaning Systems, Man, and Cybernetics: Systems, 51(7), 4091–4101. https://doi.org/10.1109/
of machinery condition monitoring. IEEE Transactions on Industrial Electronics, 67(3), TSMC.2019.2933050
2326–2336. https://doi.org/10.1109/TIE.2019.2903774 Zhang, W. J., & Lin, Y. (2010). On the principle of design of resilient systems –
Xu, Y., Sun, Y., Liu, X., & Zheng, Y. (2019). A Digital-Twin-Assisted fault diagnosis using application to enterprise information systems. Enterprise Information Systems, 4(2),
deep transfer learning. IEEE Access, 7, 19990–19999. https://doi.org/10.1109/ 99–110. https://doi.org/10.1080/17517571003763380
ACCESS.2018.2890566 Zhang, W. J., & Wang, J. W. (2016). Design theory and methodology for enterprise
Yadav, A. K., Shukla, R., & Singh, T. R. (2021). Machine learning in expert systems for systems. Enterprise Information Systems, 10(3), 245–248. https://doi.org/10.1080/
disease diagnostics in human healthcare. In Machine Learning, Big Data, and IoT for 17517575.2015.1080860
Medical Informatics (pp. 179–200). Elsevier. https://doi.org/10.1016/B978-0-12- Zhang, W. J., Yang, G., Lin, Y., Ji, C., & Gupta, M. M. (2018). On Definition of Deep
821777-1.00022-7. Learning. 2018 World Automation Congress (WAC), 1–5. Stevenson, WA: IEEE.
Yang, B., Lei, Y., Jia, F., & Xing, S. (2018). A Transfer Learning Method for Intelligent 10.23919/WAC.2018.8430387.
Fault Diagnosis from Laboratory Machines to Real-Case Machines. 2018 International Zhang, W., Li, C., Peng, G., Chen, Y., & Zhang, Z. (2018). A deep convolutional neural
Conference on Sensing,Diagnostics, Prognostics, and Control (SDPC), 35–40. 10.1109/ network with new training methods for bearing fault diagnosis under noisy
SDPC.2018.8664814. environment and different working load. Mechanical Systems and Signal Processing,
Yang, B., Lei, Y., Jia, F., & Xing, S. (2019). An intelligent fault diagnosis approach based 100, 439–453. https://doi.org/10.1016/j.ymssp.2017.06.022
on transfer learning from laboratory bearings to locomotive bearings. Mechanical Zhang, W., Li, X., & Ding, Q. (2019). Deep residual learning-based fault diagnosis method
Systems and Signal Processing, 122, 692–706. https://doi.org/10.1016/j. for rotating machinery. ISA Transactions, 95, 295–305. https://doi.org/10.1016/j.
ymssp.2018.12.051 isatra.2018.12.025
Yang, B.-S., Di, X., & Han, T. (2008). Random forests classifier for machine fault Zhang, Y., Li, X., Gao, L., Wang, L., & Wen, L. (2018). Imbalanced data fault diagnosis of
diagnosis. Journal of Mechanical Science and Technology, 22(9), 1716–1725. https:// rotating machinery using synthetic oversampling and feature learning. Journal of
doi.org/10.1007/s12206-008-0603-6 Manufacturing Systems, 48, 34–50. https://doi.org/10.1016/j.jmsy.2018.04.005
Yang, D., Liu, Y., Li, S., Li, X., & Ma, L. (2015). Gear fault diagnosis based on support Zhao, M., Kang, M., Tang, B., & Pecht, M. (2018). Deep residual networks with
vector machine optimized by artificial bee colony algorithm. Mechanism and Machine dynamically weighted wavelet coefficients for fault diagnosis of planetary
Theory, 90, 219–229. https://doi.org/10.1016/j.mechmachtheory.2015.03.013 gearboxes. IEEE Transactions on Industrial Electronics, 65(5), 4290–4300. https://doi.
Yang, T., Pen, H., Wang, Z., & Chang, C. S. (2016). Feature knowledge based fault org/10.1109/TIE.2017.2762639
detection of induction motors through the analysis of stator current data. IEEE Zhao, M., Kang, M., Tang, B., & Pecht, M. (2019). Multiple wavelet coefficients fusion in
TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, 65(3), 11. deep residual networks for fault diagnosis. IEEE Transactions on Industrial Electronics,
Yang, Z.-X., Wang, X.-B., & Zhong, J.-H. (2016). Representational learning for fault 66(6), 4696–4706. https://doi.org/10.1109/TIE.2018.2866050
diagnosis of wind turbine equipment: A Multi-Layered extreme learning machines Zhao, R., Yan, R., Chen, Z., Mao, K., Wang, P., & Gao, R. X. (2019). Deep learning and its
approach. Energies, 9(6), 379. https://doi.org/10.3390/en9060379 applications to machine health monitoring. Mechanical Systems and Signal Processing,
Yao, B., Zhen, P., Wu, L., & Guan, Y. (2017). Rolling element bearing fault diagnosis 115, 213–237. https://doi.org/10.1016/j.ymssp.2018.05.050
using improved manifold learning. IEEE Access, 5, 6027–6035. https://doi.org/ Zhao, X., & Jia, M. (2018). Fault diagnosis of rolling bearing based on feature reduction
10.1109/ACCESS.2017.2693379 with global-local margin fisher analysis. Neurocomputing, 315, 447–464. https://doi.
Yoon, J., & He, D. (2015). Planetary gearbox fault diagnostic method using acoustic org/10.1016/j.neucom.2018.07.038
emission sensors. IET Science, Measurement & Technology, 9(8), 936–944. https://doi. Zhao, Y., Li, T., Zhang, X., & Zhang, C. (2019). Artificial intelligence-based fault
org/10.1049/iet-smt.2014.0375 detection and diagnosis methods for building energy systems: Advantages,
Yu, H., Wang, K., & Li, Y. (2018). Multiscale representations fusion with joint multiple challenges and the future. Renewable and Sustainable Energy Reviews, 109, 85–101.
reconstructions autoencoder for intelligent fault diagnosis. IEEE Signal Processing https://doi.org/10.1016/j.rser.2019.04.021
Letters, 25(12), 1880–1884. https://doi.org/10.1109/LSP.2018.2878356 Zheng, H., Wang, R., Yang, Y., Li, Y., & Xu, M. (2020). Intelligent fault identification
Yu, J., Bai, M., Wang, G., & Shi, X. (2018). Fault diagnosis of planetary gearbox with based on multisource domain generalization towards actual diagnosis scenario. IEEE
incomplete information using assignment reduction and flexible naive bayesian Transactions on Industrial Electronics, 67(2), 1293–1304. https://doi.org/10.1109/
TIE.2019.2898619
20

IFTCS

Uploaded by

Copyright:

Available Formats

IFTCS

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

IFTCS

Uploaded by

Copyright:

Available Formats

Expert Systems With Applications 238 (2024) 121956

Contents lists available at ScienceDirect

Expert Systems With Applications

Development of Intelligent Fault-Tolerant Control Systems with Machine

Fig. 1. Hierarchy and classification of intelligent fault diagnosis algorithms.

Our contribution to this review paper is a methodical investigation of

Fig. 3. Intelligent fault diagnosis (Krizhevsky et al., 2017).

Fig. 4. The architecture of expert-based IFTC.

Fig. 5. The decision tree (J. Yu et al., 2016).

Fig. 6. Diagnosis procedure of IFTC (Mayer-Schönberger & Cukier, 2013).

Fig. 8. RBM architecture (D. Han, Zhao, & Shi, 2017).

Where Z(θ) is the partition function. The activation condition of

Fig. 9. Convolution and pooling process (Gu et al., 2018).

Fig. 10. The architecture of the residual block.

beings. Moreover, deep learning and transfer learning algorithms have

Fig. 11. Steps of feature-based transfer learning.

You might also like