machines

Review

Condition Monitoring of Electric Machines: Modern

Frameworks and Data-Driven Methodologies
Wesley Doorsamy

School of Electronic and Electrical Engineering, University of Leeds, Leeds LS2 9JT, UK; [email protected]

Abstract: Electrical machines are at the centre of most engineering processes, with rotating
electrical machines, in particular, becoming increasingly important in recent history due
to their growing applications in electric vehicles and renewable energy. Although the
landscape of condition monitoring in electrical machines has evolved over the past 50 years,
the intensification of engineering efforts towards sustainability, reliability, and efficiency,
coupled with breakthroughs in computing, has prompted a data-driven paradigm shift.
This paper explores the evolution of condition monitoring of rotating electrical machines
in the context of maintenance strategy, focusing on the emergence of this data-driven
paradigm. Due to the broad and varying nature of condition monitoring practices, a
framework is also offered here, along with other essential terms of reference, to provide
a concise overview of recent developments and to highlight the modern challenges and
opportunities within this area. The paper is purposefully written as a tutorial-style overview
for the benefit of practising engineers and researchers who are new to the field or not
familiar with the wider intricacies of modern condition monitoring systems.

Keywords: condition monitoring; data-driven; rotating electrical machines; mainte-

nance strategy

1. Introduction
Rotating electrical machines are ubiquitous in all manner of modern engineering
applications. Most of these applications employ machines, particularly at lower ratings,
Academic Editor: Davide Astolfi
that are adequately robust, effective, and reliable when compared to their availability
Received: 14 January 2025
requirements, such that condition monitoring is not required [1]. However, there are
Revised: 10 February 2025
many applications where the reliability of rotating electrical machines is critical to the
Accepted: 12 February 2025
Published: 13 February 2025
function they serve, necessitating condition monitoring. In recent decades, there has been
a significant increase in these types of applications in growth sectors such as renewable
Citation: Doorsamy, W. Condition
Monitoring of Electric Machines:
energy [2], industrial automation [3], and electric vehicles [4].
Modern Frameworks and Data-Driven This growth in demand for condition monitoring in rotating electrical machines, cou-
Methodologies. Machines 2025, 13, 144. pled with related technological advancements in sensing, communications, and computing,
https://doi.org/10.3390/ has led to increased research and development in this field, particularly with data-driven
machines13020144
approaches. Consequently, the field has garnered widespread interest and a corresponding
Copyright: © 2025 by the author. increase in the research literature, but much of this continues to be in directions that are
Licensee MDPI, Basel, Switzerland. not particularly useful to industry [5]. Although there have been some notable surveys in
This article is an open access article
recent history, such as [5–9], this issue extends to many other general surveys that have
distributed under the terms and
emerged more recently. A factor contributing to this misalignment with industry needs and
conditions of the Creative Commons
Attribution (CC BY) license
the inability to effectively position new developments is the lack of a common framework
(https://creativecommons.org/ and consistent ’terms of reference’ for condition monitoring of machines. The aim of this
licenses/by/4.0/). paper is twofold: to present a framework, within the context of maintenance strategy,

Machines 2025, 13, 144 https://doi.org/10.3390/machines13020144

Machines 2025, 13, 144 2 of 16

that systematically characterises modern condition monitoring systems; and to concisely

present some of the most recent developments in data-driven condition monitoring of
electrical machines, outlining the foremost opportunities and challenges. Thus, the study
contributes a framework to support research and development, as well as the selection
and evaluation of condition monitoring techniques for rotating electrical machines. Addi-
tionally, the review of the recent literature, including seminal works and other important
surveys, together with an analysis of opportunities and challenges, offers archival value for
researchers and practitioners and identifies potential avenues to catalyse future research in
this field.
This paper is organised as follows. The next section gives an overview of the evolu-
tion of condition monitoring, contextualises its role within a wider maintenance strategy,
and presents a generalised framework that encompasses systematic characteristics based
on this evolution and context. Thereafter, faults and failure mechanisms are discussed
together with a review of condition monitoring methods that focuses on data-driven meth-
ods. The future outlook for data-driven condition monitoring is then discussed before a
brief conclusion.

2. Background
2.1. Evolution of Condition Monitoring
Modern condition monitoring is a product of several decades of experience, inno-
vation, and technological advancement. The earliest generation of modern equipment
monitoring through telemetry systems dates back to the 1930–1940s, where the need for
progressive methods for maintenance planning, i.e., the ‘Waddington effect’, became appar-
ent [10]. This paved the way for the emergence of the concept of ‘continuous monitoring’
and supervisory control and data acquisition (SCADA) during the 1950–1970s, where
more emphasis was not only placed on automation but also on monitoring as part of
the maintenance strategy [11]. The proliferation of transducers and advancement of data
acquisition systems in the 1970–1980s [12] afforded further development and widespread
expansion of equipment monitoring systems. In recent decades, condition monitoring has
undergone another paradigm shift with advancements in communications and computing,
where the widespread adoption of the Internet-of-Things (IoT) and data analytics has seen
more industries modernising their asset management practices, not only to exploit strategic
business opportunities but also as an essential organisational function.
Condition monitoring of rotating electrical machines is distinct within the general
area of equipment condition monitoring, evolving in its own right, due to the ubiquity of
these machines and their vital importance in several sectors of industry, such as utilities,
transport, manufacturing, etc. In general, monitoring becomes less essential for lower-rated
machines, e.g., active power P < 20 kW, except for those serving critical functions [1].
Ultimately, the decision to monitor the machine is based on weighing the associated costs
against the significance of losing the machine and/or the function it provides. Although this
trade-off will always need to be evaluated for specific cases, monitoring costs have reduced
over time, leading to further uptake. The aforementioned advancements offer more cost-
effective approaches, whether through dedicated machine condition monitoring systems or
integrating machine monitoring into the wider plant-wide monitoring/expert system.

2.2. Asset Management Context

While the benefits of monitoring rotating electrical machines are often emphasised in
the literature, its role within the wider context of asset management is often overlooked.
Although a condition monitoring system itself may be highly effective in performing
its function, this does not equate to efficient asset management when viewed from a
Machines 2025, 13, 144 3 of 16

strategic perspective. For this reason, condition monitoring should be considered as

support to, and dependent on, the overall maintenance strategy [13]. Simply put, the
selection and evaluation of the condition monitoring approach to be deployed is based
on the requirements analysis arising from the maintenance strategy. The description
of asset management, offered by [14], refers to the “organisation’s objectives into asset-
related decisions, plans and activities risk-based approach”. Asset management thus
determines the requirements for condition monitoring based on the risk assessment [15],
e.g., production impact due to machine unavailability, and it is the condition monitoring
system that informs the maintenance decision making. Typically, a modern condition
monitoring system will comprise what are referred to here as monitoring and assessment
functions (as depicted in Figure 1). The asset management strategy determines the type
and level of assessment needed, which in turn defines the monitoring requirements. The
assessment function of the condition monitoring system does not only provide feedback on
the condition of the machine, but may also provide crucial information for failure mode and
effect analysis (FMEA), and defining safety levels [16]. In rotating electrical machines, the
assessment function may extend beyond diagnosing current problems to estimating future
degradation, that is, prognostics [17]. Furthermore, non-destructive evaluation (NDE),
which is inclusive of non-destructive inspection (NDI) and non-destructive testing (NDT),
needs mentioning here as a field that historically parallels condition monitoring in the
context of assessment management, particularly in the case of in-service NDE [18]. More
specifically, while in-service NDE plays a seemingly distinct role in identifying or locating
specific problems with the asset, there is an overlap with this role and the assessment
function of condition monitoring systems. The synonymity of in-service NDE with certain
condition monitoring techniques proposed for electrical machines is evident in some of the
literature, as in [19–21], where these terminology have been used interchangeably.

Figure 1. Condition monitoring in the context of maintenance strategies.

In asset management, the main maintenance strategies are condition-based mainte-

nance (CBM), reliability-centred maintenance (RCM), time-based maintenance (TCM), and
corrective maintenance (CM) [22]. These strategies are broadly categorised as shown in
Figure 2, where CBM and RCM are typically deployed in predictive maintenance strategies,
RCM and TCM are deployed in preventative maintenance strategies, and CM is deployed
in reactive maintenance strategies. The key factors that determine which strategy is most
appropriate are the process or system in which the machine is deployed and the machine
or component itself. For instance, where the condition of the machine or component is
Machines 2025, 13, 144 4 of 16

considered critical—e.g., due to low or no redundancy—and is deployed in a highly critical

process or system, RCM is the most suitable maintenance approach, where maintenance is
prioritised, and machine faults must be closely monitored, assessed, and managed; while
the maintenance strategy may specify distinct condition monitoring requirements, this
may adapt to changes in operational experience, priorities, and risk, which means that
organisations can draw tremendous benefit from condition monitoring systems that are
more flexible. This is a sought-after feature in modern condition monitoring systems and is
discussed later.

Figure 2. Categorisation of maintenance strategies.

2.3. Framework and Terms of Reference

The landscape for condition monitoring of electrical machines is vast and continu-
ously evolving. Currently, there is no modern framework that adequately encapsulates
this evolution and provides a means of systematically contextualising new research and
unambiguously positioning the specific area of incremental progress. Thus, the generalised
framework offered here (in Figure 3) proposes a basic system architecture for condition
monitoring comprising monitoring and assessment layers. These layers link to the asset
management strategy at the higher level (Figure 1) and represent different functions at the
level of the condition monitoring system. The monitoring layer encompasses the system
components that perform all of the functions from measurement to information extraction.
The subsequent layer utilises this information to carry out the assessment. It should be
highlighted that the framework is specific yet sufficiently generic such that it suitably char-
acterises the vast range of condition monitoring approaches and different combinations
thereof. For example, this framework characterises the condition monitoring approach,
Machines 2025, 13, 144 5 of 16

whether it is a technician using a handheld device to manually determine vibration or an

automated fault classifier based on motor current signature analysis.

Figure 3. Generalised framework for condition monitoring.

There is also some key terminology associated with the framework that enables
a common basis for developing condition monitoring systems for electrical machines.
Measurement modality refers to the primary source of information, that is, the machine
parameters or signals—e.g., current, temperature, and speed—that are being monitored.
This determines the pre-processing requirements, which poses limitations on how often
measurements can be taken, and whether monitoring is performed online or offline, ul-
timately influencing the type and extent of the assessment. In condition monitoring of
electrical rotating machines, the measurement modality can be broadly classified into
different types, such as thermal, chemical, mechanical, and electrical. The modality of the
detector, or transduction process, may in some cases be linked to the machine’s state that is
being monitored/measured through this categorisation, although these may not necessarily
be the same—e.g., vision sensing with an event-based camera [23] can be considered a
mechanical measurement modality, like an accelerometer-based method, when used to
monitor vibration.
The pre-processing component of the monitoring layer in the framework refers to
everything from signal acquisition to the conversion and extraction techniques used. This
makes an important distinction between the monitoring and assessment layer, where this
boundary is often blurred in the literature. This is because the assessment approach is,
in many instances, directly related to the pre-processing. However, in modern condition
monitoring, these must be differentiated for the purposes of distinctly identifying novelty
and comparatively evaluating new research and developments in the area. The assessment
layer can be thought of as the component of the condition monitoring system that carries
out the inference about the state of the electrical machine, or component/s thereof. In mod-
ern condition monitoring, the inference typically leads to three types of assessment—i.e.,
detection, diagnosis, or prognosis—which convey varying levels of fault information and,
in some prognostic assessments, the remaining useful life of machine components. This
layer thus carries out the ’intelligence’ function where the inference itself can range from a
simple checking of a parameter against a threshold or limit to a more complex prognosis.
The level of intelligence in this layer therefore varies according to the level of underlying
knowledge about the machine that is codified into the inference, where the inference, or
assessment, can be automated or carried out by a human. There are many instances where
existing terminology inadequately captures the aforementioned distinctions in modern
condition monitoring of electrical machines. For example, the spectrum-based methods
reviewed in [24] are all based on signals, utilise time, time–frequency, or frequency–domain
Machines 2025, 13, 144 6 of 16

methods, and employ models for inference about the state of the machine which are
based on underlying knowledge of the machine’s physics. Therefore, broadly classifying
these approaches as signal-based methods, and differentiating them from ’model’- and
’knowledge’-based approaches is somewhat imprecise. Furthermore, modern strategies
can and do use spectrum techniques to extract signal features for carrying out assess-
ments based on historical data and not on the modelled physics of the machine. Similarly,
Akbar et al. [25] separately categorise spectral and vibration analysis as ’conventional’
techniques, which fails to make the distinction between the modality and pre-processing
technique where conventional vibrational analysis utilises spectral techniques. Therefore,
the presented framework provides a more accurate way of characterising measurement,
pre-processing and assessment techniques, and a more sensible approach to classifying the
condition monitoring strategy according to the assessment method.
Assessment methods can be broadly classified as model-based or data-driven. The
main strength of model-based methods is perhaps the source of its main limitations. Al-
though a well-defined model of the physical component or process in the machine enables
a more effective assessment of its state, the availability of such models and domain knowl-
edge limits the application flexibility and assessment range of these methods. Therefore,
model-based methods are also key to the fundamental development of condition moni-
toring in electrical rotating machines as they typically offer new insights into modelling
fault mechanisms and, in some cases, novel measurement modalities and pre-processing
techniques. Data-driven assessment approaches overcome this dependence on physical
modelling by using historical data, potentially extending their application flexibility and
assessment range. However, this key strength over model-based methods is also a source of
limitations with data-driven approaches where historical data, particularly fault data, are
not readily available. Some of the key considerations and trade-offs between model-based
and data-driven assessment are paving the way for different avenues of research and
development in the condition monitoring of electrical rotating machines. Examples of these
are as follows:
• Model-based assessment relies on codified domain knowledge in well-defined repre-
sentations of physical processes/mechanisms, while data-driven assessment typically
depends on historical data.
• Research efforts have intensified into data-driven strategies more recently because
they seemingly offer the potential to further progress modern condition monitoring
goals such as incipient fault detection/diagnosis, holistic and integrated assessments,
as well as online, continuous, and real-time monitoring.
• While data proliferation, owing to related technological advancements, lends itself
to data-driven approaches, it also brings about several new challenges, which are
discussed later.
It should be noted that methods employing a combination of model-based and data-
driven techniques are sometimes referred to as hybrid methods. Although the framework
depicted in Figure 3 does not explicitly mention these methods under a different category,
they can be characterised according to how the actual assessment is carried out. For
example, physics-informed machine learning is an example of a hybrid method [26], where
the assessments are still based on available data but within the constraints of the physical
knowledge of the machine.
Machines 2025, 13, 144 7 of 16

3. Review and Analysis

3.1. Scope
Considering the extensive array of measurement modalities and pre-processing tech-
niques used in monitoring, as well as the proliferation of research into assessment methods,
recent developments in this area are too vast to concisely review. Therefore, the approach
taken concisely formulates a roadmap by focusing its scope on an illustrative selection of
data-driven methods that push the boundaries of the assessment layer. Further details of
the selection criteria for the examples is given Section 3.3. An overview of fault and failure
mechanisms in electrical rotating machines is given in the next section to better contextu-
alise the reviewed methods within the assessment layer. This is because the differentiation
of the capabilities, and the evaluation of the suitability, of the condition monitoring tech-
nique should be based on the faults and failure modes specific to machine for which the
technique is intended.

3.2. Faults and Failure Mechanisms

There are various types of faults and failure modes in electrical rotating machines.
Depending on the machine type, size, and application, the type, distribution, and frequency
of faults vary. Generally, faults and failure modes tend to be classified at the component
level, e.g., rotor and stator windings, air-gap, rotor bars, end rings, slip rings, permanent
magnets, brushes, shaft, cores and laminations, bearings, peripherals/assembly, load, and
auxiliaries [27]. Bearing and stator winding failures are the most common modes of failure
in machines; however, bearing failure is more prominent in machines rated up to 4 kV, and
stator winding failures account for the largest share among failure modes in higher-rated
machines [28]. This is because windings suffer insulation degradation due to thermal,
thermo-mechanical, and mechanical ageing, as well as partial discharges, particularly
in conventional machines rated 3.3 kV and above and 400 V and above for inverter-fed
motors [29]. Much of the research efforts on online, non-invasive, and incipient fault
diagnostics have focused on rotor faults, such as broken rotor bars, and bearing faults;
while some researchers have proposed online methods for assessing winding insulation
health, such as some of the promising new broadband frequency techniques [30], they
are yet to be proven in industry and therefore offline methods continue to be used more
widely [31]. Machine applications are also a driving force behind innovation in condition
monitoring, where advancements in fault detection and diagnosis extend beyond the
machine’s components. In recent history, this is especially true for wind energy and electric
vehicle applications where monitoring extends to faults and failure modes associated with
intertwined critical electrical and mechanical subsystems [32,33].
The trend in broadening the scope of the monitoring and assessment capabilities of
condition monitoring systems stems from long-standing goals of online and incipient fault
diagnosis in predictive maintenance. However, the latter goal is often overlooked in the lit-
erature, where failures can occur through cascading fault mechanisms. For example, while
bearing defects may cause eccentricity leading to failure, bearing wear and failure can also
be triggered through a succession of other faults, or ’fault tree’ [34,35], like insulation wear,
localised heating, etc. This makes incipient fault diagnosis quite challenging, where the
goal is not only to detect faults early but to diagnose them at an early stage of the fault tree.
Thus, although diagnosis may provide additional assessment information over detection
as to the fault location and type, it may not necessarily alleviate uncertainty around the
severity and root cause of the fault. It is worth again highlighting the essential role of
modality in constraining the subsequent layers of the overall condition monitoring system,
thereby ultimately determining what assessment is possible. For example, current [7,36]
and magnetic flux (axial and stray) [37] modalities offer the potential for assessing type,
Machines 2025, 13, 144 8 of 16

location, and severity for a wider variety of faults than is possible with mechanical and
thermal modalities. That being said, data-driven methods are being shown to better exploit
the information capacity of certain modalities, such as sound [38], to enhance overall as-
sessment capability. This further emphasises the need to distinguish approaches within
the layers of the presented framework where the overall fault assessment capabilities vary
with different combinations of techniques.

3.3. Data-Driven Methods

Data-driven methods essentially refer to the category of techniques that use statistical
or machine learning models to carry out assessment—i.e., detection, diagnosis, or prognosis.
Due to growing interest in data-driven condition monitoring, several reviews of related work
have recently been completed [25,39–43]. However, the proliferation of research in this area,
coupled with the lack of a suitable framework to systematically position and comparatively
evaluate new developments, has rendered concise and focused reviews in the area rarities.
This is evident in the recent groundswell of studies on machine learning techniques applied
to bearing fault diagnosis, as there are several open-source datasets available [44], where deep
learning ablation and optimisation studies have been used to improve classification accuracy;
while these studies may significantly contribute to setting new benchmarks for a particular
technique, they do not necessarily equate to new condition monitoring approaches or strategies.
Therefore, only a selection of examples is offered here with the aim of demonstrating some of
the most recent research offering progressive data-driven methods in the assessment layer of
condition monitoring systems for electrical rotating machines.
The examples presented in Table 1 were selected based on the following criteria. Only
research using data-driven techniques was considered, thereby excluding articles that propose
model-based methods. The search criteria required articles to be explicit about the applicability
of their proposed techniques to rotating electrical machines. The most recent articles were
considered, and the pool of potential research was limited to ten articles from 2023 up to the
time of writing. Although research articles that have garnered the most interest (top cited)
were used as a sorting mechanism, diversifying the set of examples according to measurement
modality, fault type, and data-driven method was given precedence. This approach is based
on the proliferation of bearing vibration monitoring studies mentioned earlier. As the vast
majority of recent articles focusing on this measurement modality are essentially ablation studies,
the selection of these articles was only considered where there was a significant difference in
monitoring and assessment approach—e.g., assessment layer using a physics-informed neural
network or transfer learning, as opposed to using a similar convolutional neural network
approach with slightly different parameters to demonstrate higher classification accuracy with a
specific benchmarked dataset.

Table 1. Summary of selected research in data-driven condition monitoring of rotating electri-

cal machines.

Method
Reference Fault
Monitoring Layer Assessment Layer
Event-based vision sensor,
Convolutional Neural
[23] Bearing image-shaping, data augmentation
Network (CNN)
and denoising
Accelerometer, vibration, discrete Physics-Informed Residual
[45] Bearing
Fourier transform (DFT) Network
Machines 2025, 13, 144 9 of 16

Table 1. Cont.

Method
Reference Fault
Monitoring Layer Assessment Layer
Infrared thermal camera
CNN with Transfer
[46] Bearing (thermography), image-shaping and
Learning (TL)
pre-processing
Stator currents and speed, Digital twin with health
[47] Rotor bar, stator winding
Ramanujan Periodic Transform indicator
Infrared thermal camera,
[48] Stator winding CNN
image-shaping and pre-processing
k-nearest neighbours
Stator current, short-time Fourier
[49] Rotor permanent magnet (kNN) and multilayer
transform (STFT)
perceptron (MLP)
Torque, stator current and voltage, Graph Neural Network
[50] Bearing
normalisation and cosine similarity (GNN)
Driven-equipment faults Torque, stator current and voltage, CNN–Long short-term
[51]
(pump) frequency spectrum estimation memory (LSTM)
Stator winding, eccentricity, Search coil, magnetic flux, frequency
[52] Random Forest
permanent magnets spectrum estimation
Stator currents, Fast Fourier
[53] Rotor permanent magnets CNN with TL
Transform (FFT)

The selected research given in Table 1 is organised here in terms of the presented
framework where the various components or layers of the condition monitoring system are
separated. This is to demonstrate how framing of these condition monitoring approaches
enables better comparison and assessment of their suitability, applicability, capability, and
even novelty. For example, although the proposed method presented in [23] is similar in
terms of the bearing fault type and CNN assessment layer to many other recent research
articles, its monitoring layer uses event-based vision sensing that produces very different
data to the typical accelerometer-based vibration monitoring techniques. Similarly, both
of the proposed approaches in [46,48] employ thermal monitoring, or more specifically
thermography in these cases, together with CNN-based techniques in their assessment
layer, but they each focus on very different machine components and fault types—i.e.,
bearing and stator winding faults.

4. Prospects for Driven-Driven Condition Monitoring

4.1. Opportunities
4.1.1. Online, Real-Time, and Automated Assessment
Condition monitoring of rotating electrical machines has always had some level of
automation, where even the earliest systems could automatically take a measurement
and provide a basic assessment. Modern condition monitoring systems are automated in
this sense, but their layers, particularly the assessment layer, can have varying degrees
of automation. As mentioned above, the assessment can range from basic threshold
checking, which is still relevant in modern condition monitoring [54], to a more advanced
prognosis [55], with different levels of information pertaining to the type, location, and
severity of the fault, as well as the remaining useful life of the components. Therefore, the
goal of fully automating assessments—many of which continue to be carried out by human
experts in practice—is a catalyst for further research and development in data-driven
methods. For example, large language models (LLMs) are fast becoming an interesting
prospect for building large-scale foundation models in industrial settings [56]. LLMs
offer a practical means, when used in conjunction with data-driven condition monitoring
Machines 2025, 13, 144 10 of 16

techniques in the assessment layer, to further automate assessment tasks, as presented

in [57]. Automated data-driven assessment has the potential to unlock several other benefits
such as improved scalability and flexibility [58], and real-time assessment, particularly
when deployed in online and continuous monitoring systems [59].

4.1.2. Fault Detection, Diagnostics, and Prognostics

Much has been discussed about the different levels of assessment and fault types.
This is because recent research has demonstrated the potential of data-driven methods to
enhance assessment at each level—i.e., detection, diagnosis, and prognosis—as well as
widen the range of fault types that can be assessed. Model-based assessments typically
developed around specific modalities, whereas data-driven assessments are based on
features of the available data. Data-driven methods therefore have the flexibility to combine
modalities/sensory data and provide multimodal assessment, thereby extending the range
of fault types that can be assessed by a single model-based system [39,60].
While the fault detection category of assessment has undergone a wide range of
advancements through different data-centric methods, the most recent strides in this area
have been with the provision of online, real-time, and continuous assessment. The change in
terminology used in recent literature from fault detection to anomaly detection delineates
this progressive shift towards a particular set of characteristics within this assessment
category. Real-time anomaly detection has featured in other applications [61,62], but is now
also emerging in fault detection as these methods, unsupervised learning techniques in
particular [63], are equipped to handle continuous data streams and make fast automated
assessments without the need for extensive offline processing. The concept of digital twins
in machine condition monitoring has also received a lot of attention lately as it promises to
take advantage of several of the aforementioned benefits of data-driven methods, coupled
with breakthroughs in computing capacity to develop high-fidelity models for online, real-
time, and continuous monitoring and fault detection [47]. Research in this area is expected
to continue to grow with prospects for improved digital twin models and architectures for
interoperable digital twins, where monitoring may benefit from machine models seamlessly
interacting with models of other equipment in its operating context. Fault diagnosis is the
category of assessments that has arguably benefited the most from data-driven methods,
where supervised machine learning techniques have been leveraged to classify faults using
labelled historical data [64]. Similarly, data-driven techniques have opened new possibilities
in the fault prognosis assessment category with extensive progress in modelling component
degradation to predict remaining useful life and forecast failures [65].
Despite these recent strides in data-driven assessment methods, an ongoing challenge,
which will be discussed further in the next section, is to extend the online, continuous, and
real-time capabilities of modern fault detection to diagnostic and prognostic assessments
that can help with root cause analysis and FMEA. This is an area of future research and
development that is expected to bring about assessment depth, such as fault type, severity,
and location, in online, real-time, and streaming applications.

4.1.3. System Integration

The flexibility, scalability, and increasing ubiquity of data-driven methods are seem-
ingly converging towards the development of systems that can exploit all the aforemen-
tioned opportunities. This offers the chance to truly integrate condition monitoring into
plant management, where fleet-wide monitoring [66], and even multi-plant management
requiring assessment of a combination of factors, such as maintenance, production, safety,
resources, etc. [15], are becoming more realistic prospects. An example of recent work
that represents this latest direction in data-driven methods is given in [67], which seeks to
Machines 2025, 13, 144 11 of 16

integrate monitoring and control systems to optimise overall operational efficiency using
data-driven approaches. Figure 4 provides a simplified example of system-level integration
to illustrate this concept that combines monitoring, control, and analytics across flexible
and scalable layers of intelligence—i.e., a lower monitoring/control layer for x to n ma-
chines/components, an analytics layer for y to p processes/functions, and an upper layer
for system-level analytics.

Figure 4. Example of modern condition monitoring architecture for exploiting system-

level integration.

4.2. Challenges
4.2.1. Data Availability and Reliability
Unlike model-based methods, the reliability of data-driven models is highly dependent
on the availability of reliable data. Supervised learning techniques can be used to train
models to accurately classify a wide variety of incipient faults, but the historical data
must be suitably labelled and contain all of the nuanced patterns that characterise these
faults. The accuracies and ranges of fault classification models are also dependent on the
suitability of the extracted features. Furthermore, the performance of models trained in one
context cannot be guaranteed, especially when there are significant differences in the design,
manufacture, and operating regimes of certain machines. The availability of fault-type
data is also often limited in practice, where fault conditions occur sparsely during the life
of a machine, leading to the challenge of data imbalance. The focus of recent research in
this area has thus shifted away from conventional supervised learning techniques, turning
to ensemble, resampling, transfer learning, semi-supervised, and unsupervised learning
approaches, amongst others, that may address these ongoing challenges [68–73]. However,
there is still much work to be performed in terms of improving the assessment capabilities
of these methods, particularly with fault severity and root cause analysis. Research aimed
at addressing data availability and reliability is therefore expected to continue to grow, as
these issues form an especially critical stumbling block to the practical implementation and
widespread adoption of data-driven assessment methods.
Machines 2025, 13, 144 12 of 16

4.2.2. Model Transparency and Interpretability

With all the advancements in condition monitoring of electrical machines, a vast
range of model-based and data-driven assessment methods are now available. These
methods are based on models that differ in complexity and transparency. Traditional
model-based approaches typically offer transparent, or ’white-box’, models, whereas most
data-driven models are not transparent, or ’black-box’ models, having been built using ma-
chine learning techniques. This lack of transparency in models poses a significant challenge
to interpretability, which in turn may hinder trust, serviceability, compatibility, and adapt-
ability. An understanding of the inner workings of the assessment model plays a crucial
role in evaluating how well it is able to capture fault characteristics and degradation trends
in its decision-making process, making the results more intelligible and trustworthy [74].
Similarly, models with low interpretability make it difficult to troubleshoot problems with
the condition monitoring system [75], especially if several of these models are intercon-
nected. The black-box nature of data-driven models also complicates the assessment of how
compatible they are with existing components/subsystems, or how adaptable they will
be to new components/subsystems; while there have been some recent efforts to improve
model interpretability, such as model-agnostic methods and model embedding, it still
remains a significant obstacle to the wider uptake of data-driven methods for condition
monitoring [75,76].

4.2.3. Systems Design, Deployment, and Operation

Fully exploiting the highlighted opportunities offered by modern data-driven con-
dition monitoring means confronting several practical challenges with system design,
deployment, and operation, particularly when adopting these approaches at scale. Despite
the vast amount of research literature on data-driven condition monitoring, there is still a
lack of technical literature, including standards, that deal with these practical issues. This
is owing not only to a lack of maturity with some of these technologies but also to the
aforementioned challenges where comparative evaluations with ’legacy’ systems are not
yet available. Consequently, the uncertainties surrounding the design, deployment, and op-
eration of data-driven condition monitoring systems may hinder uptake, as there is a need
to consider all of the issues to properly assess initial investment versus long-term gains.
The issue of data ontology relates to other challenges discussed here, such as data
availability and reliability, but warrants several design, deployment, and operational con-
siderations in its own right [77]. Due to the data-centric nature of modern condition
monitoring approaches, industrial-scale implementations typically require a complete re-
think of how data are managed [78]. This includes considering the design architecture of
the data pipeline based on the current and future asset management needs. For example,
the decision of which functions within the different layers of the condition monitoring
system—such as storage, pre-processing, and assessment of sensor data—should be im-
plemented locally to the machine through edge computing, or via cloud computing, or
some combination of both; this not only affects costs, efficiency, and performance in the
short term but can also determine scalability, flexibility, and compatibility in the long term.
These decisions are largely based on the condition monitoring requirements in the context
of the maintenance strategy as discussed in Section 2.2. The data transmission (latency,
throughput, etc.), data storage (local and/or remote capacities), and data handling (com-
pute resources) needs therefore follow a comprehensive definition of these requirements
through considering, inter alia, the number, types, and frequency of measurements; where
and what quantity of raw data will be stored, pre-processed, and processed; whether the
processing of data will be carried out online, continuously, and in real-time; assessment
Machines 2025, 13, 144 13 of 16

model workflows including training, retraining, visualisation, analytics, etc.; integration

with other control or management systems.
While there is some general guidance available on the management of data assets
within the area of asset management [79], and some good examples of recent research that
consider practical issues with data in condition monitoring [80–82], this is a challenge area
with many opportunities for future interdisciplinary research. Similarly, there are also
several other wider issues surrounding design, deployment, and operation challenges in
data-driven condition monitoring that may benefit from future interdisciplinary research—
e.g., human factors in engineering [83,84], and efficiencies [85].

5. Conclusions
The demand for condition monitoring of rotating electrical machines has risen together
with their ever-widening role in engineering applications, particularly in areas such as
renewable energy, industrial automation, and electric vehicles. This groundswell of interest
in condition monitoring, coupled with advancements in sensing, communications, and
computing, has led to a rapid increase in research and development in the field. Despite
the resulting proliferation in the literature on the topic, there are very few resources that
bridge the gap between academic research and industrial needs. Therefore, this paper gives
a tutorial-style overview of the field that discusses the evolution of condition monitoring,
offers a structured framework for modern condition monitoring systems in the context of
maintenance strategy, and concisely examines recent advancements, challenges, and op-
portunities. The generalised framework links maintenance strategy to a modern condition
monitoring system architecture, thereby enabling systematic characterisation, selection,
analysis, and evaluation of techniques in line with the requirements of the monitoring
application. Focusing on data-driven methods, the selective review of recent progress illus-
trated how the framework can be used to systematically characterise emerging techniques
and provide insights into areas of incremental progress. Building on this, the research
and development prospects in data-driven condition monitoring were discussed. The
assessment layer was identified as a key area for exploiting the benefits of data-driven
methods, with opportunities for enhancing flexibility, scalability, integrability, and inter-
operability of condition monitoring systems. Some key future research and development
challenges were also identified, specifically in the areas of data availability and reliability,
model transparency and interpretability, and systems design, deployment, and operation.
Funding: This research received no external funding.

Data Availability Statement: No new data were created or analysed in this study. Data sharing is
not applicable to this article.

Conflicts of Interest: The author declares no conflicts of interest.

References
1. Tavner, P.; Ran, L.; Penman, J.; Sedding, H. Condition Monitoring of Rotating Electrical Machines; The Institution of Engineering and
Technology: London, UK, 2008.
2. Hussain, M.; Mirjat, N.H.; Shaikh, F.; Dhirani, L.L.; Kumar, L.; Sleiti, A.K. Condition Monitoring and Fault Diagnosis of Wind
Turbine: A Systematic Literature Review. IEEE Access 2024, 12, 190220–190239. [CrossRef]
3. Carratù, M.; Gallo, V.; Iacono, S.D.; Sommella, P.; Grasso, A.B.F.; Ciani, L.; Patrizi, G. A Novel Methodology for Unsupervised
Anomaly Detection in Industrial Electrical Systems. IEEE Trans. Instrum. Meas. 2023, 72, 17–24. [CrossRef]
4. Choudhary, A.; Fatima, S.; Panigrahi, B.K. State-of-the-Art Technologies in Fault Diagnosis of Electric Vehicles: A Component-
Based Review. IEEE Trans. Transp. Electrif. 2023, 9, 2324–2347. [CrossRef]
5. Bellini, A.; Filippetti, F.; Tassoni, C.; Capolino, G.A. Advances in Diagnostic Techniques for Induction Machines. IEEE Trans. Ind.
Electron. 2008, 55, 4109–4126. [CrossRef]
Machines 2025, 13, 144 14 of 16

6. Nandi, S.; Toliyat, H.A.; Li, X. Condition Monitoring and Fault Diagnosis of Electrical Motors—A Review. IEEE Trans. Energy
Convers. 2005, 20, 719–729. [CrossRef]
7. El Hachemi Benbouzid, M. A review of induction motors signature analysis as a medium for faults detection. IEEE Trans. Ind.
Electron. 2000, 47, 984–993. [CrossRef]
8. Riera-Guasp, M.; Antonino-Daviu, J.A.; Capolino, G.A. Advances in Electrical Machine, Power Electronic, and Drive Condition
Monitoring and Fault Detection: State of the Art. IEEE Trans. Ind. Electron. 2015, 62, 1746–1759. [CrossRef]
9. Zhang, P.; Du, Y.; Habetler, T.G.; Lu, B. A Survey of Condition Monitoring and Protection Methods for Medium-Voltage Induction
Motors. IEEE Trans. Ind. Appl.s 2011, 47, 34–46. [CrossRef]
10. Ignizio, J.P. The Waddington Effect, C4 U-Compliance, and Subsequent Impact on Force Readiness. Phalanx 2010, 46, 17–24.
11. Fiedler, H.J.; Swarthout, R.W. Exploration of Utility Automation Applications Trough Supervisory Control. IEEE Trans. Ind.
Electron. Control. Instrum. 1973, IECI-20, 12–20. [CrossRef]
12. National Semiconductor. Data Aquisition; Handbook; National Semiconductor: Santa Clara, CA, USA, 1978.
13. IEC 60050-192-06-28; International Electrotechnical Vocabulary (IEV)—Part 192: Dependability. International Electrotechnical
Commission: Geneva, Switzerland, 2015.
14. ISO 55000; Asset Management—Vocabulary, Overview and Principles. International Standards Organisation: Geneva, Switzer-
land, 2024.
15. Hesla, E.; Fowler, C.; Huber, J. Maintenance management in multiple plants: Examining various factors. IEEE Ind. Appl. Mag.
2006, 26, 88–94. [CrossRef]
16. Igder, M.A.; Rafiei, M.; Boudjadar, J.; Khooban, M.H. Reliability and safety improvement of emission-free ships: Systemic
reliability-centered maintenance. IEEE Trans. Transp. Electrif. 2021, 21, 256–266. [CrossRef]
17. Yildirim, M.; Sun, X.A.; Gebraeel, N.Z. Sensor-driven condition-based generator maintenance scheduling—Part I: Maintenance
problem. IEEE Trans. Power Syst. 2016, 31, 4253–4262. [CrossRef]
18. Papadakis, E.P. Future Growth of Nondestructive Evaluation. IEEE Trans. Sonics Ultrason. 1976, 23, 284–286. [CrossRef]
19. Qiao, W.; Lu, D. A Survey on Wind Turbine Condition Monitoring and Fault Diagnosis—Part I: Components and Subsystems.
IEEE Trans. Ind. Electron. 2015, 62, 6536–6545. [CrossRef]
20. Shin, S.M.; Choi, B.H.; Kang, H.G. Motor Health Monitoring at Standstill Through Impedance Analysis. IEEE Trans. Ind. Electron.
2016, 63, 4422–4431. [CrossRef]
21. Peng, Y.; Huang, S.; Deng, B.; He, Y.; Guo, X.; Wang, H. Joint Scanning Electromagnetic Thermography for Industrial Motor
Winding Defect Inspection and Quantitative Evaluation. IEEE Trans. Ind. Inform. 2021, 17, 6832–6841. [CrossRef]
22. Schneider, J.; Gaul, A.J.; Neumann, C.; Hogräfer, J.; Wellßow, W.; Schwan, M.; Schnettler, A. Asset management techniques. Int. J.
Electr. Power Energy Syst. 2006, 28, 643–654. [CrossRef]
23. Li, X.; Yu, S.; Lei, Y.; Li, N.; Yang, B. Intelligent Machinery Fault Diagnosis with Event-Based Camera. IEEE Trans. Ind. Inform.
2024, 20, 380–389. [CrossRef]
24. Bossio, G.R.; Meira, M.; Bossio, J.M.; Verucchi, C.J.; Ruschetti, C.R. Full Spectrum for Rotating Electrical Machines Condition
Monitoring and Fault Diagnosis: A Review. In Proceedings of the IEEE XX Workshop on Information Processing and Control
(RPIC), Obera, Argentina, 1–3 November 2023 ; pp. 1–8.
25. Akbar, S.; Vaimann, T.; Asad, B.; Kallaste, A.; Sardar, M.U.; Kudelina, K. State-of-the-art techniques for fault diagnosis in electrical
machines: Advancements and future directions. Energies 2023, 16, 6345. [CrossRef]
26. Wu, Y.; Sicard, B.; Gadsden, S.A. Physics-informed machine learning: A comprehensive review on applications in anomaly
detection and condition monitoring. Expert Syst. Appl. 2024, 255, 124678. [CrossRef]
27. Frosini, L. Novel diagnostic techniques for rotating electrical machines—A review. Energies 2000, 13, 5066. [CrossRef]
28. Haq, S.U.; Trivedi, A.; Rochon, S.; Moorthy, M.T. Alternative Methods of Machine Online Condition Monitoring: Recommenda-
tions for Rotating Machines in the Petroleum and Chemical Industry. IEEE Ind. Appl. Mag. 2024, 30, 19–31. [CrossRef]
29. Lee, S.B.; Stone, G.C.; Antonino-Daviu, J.; Gyftakis, K.N.; Strangas, E.G.; Maussion, P.; Platero, C.A. Condition monitoring of
industrial electric machines: State of the art and future challenges. IEEE Ind. Electron. Mag. 2020, 14, 158–167. [CrossRef]
30. Ruiz-Sarrio, J.E.; Antonino-Daviu, J.A.; Navarro-Navarro, A.; Biot-Monterde, V. A Review of Broadband Frequency Techniques
for Insulation Monitoring and Diagnosis in Rotating Electrical Machines. IEEE Trans. Ind. Appl. 2024, 60, 6092–6102. [CrossRef]
31. Henao, H.; Capolino, G.A.; Fernandez-Cabanas, M.; Filippetti, F.; Bruzzese, C.; Strangas, E. Trends in Fault Diagnosis for Electrical
Machines: A Review of Diagnostic Techniques. IEEE Ind. Electron. Mag. 2014, 8, 31–42. [CrossRef]
32. Badihi, H.; Zhang, Y.; Jiang, B.; Pillay, P.; Rakheja, S. A Comprehensive Review on Signal-Based and Model-Based Condition
Monitoring of Wind Turbines: Fault Diagnosis and Lifetime Prognosis. Proc. IEEE 2022, 110, 754–806. [CrossRef]
33. Zhang, X.; Hu, Y.; Gong, C.; Deng, J.; Wang, G. Artificial Intelligence Technique-Based EV Powertrain Condition Monitoring and
Fault Diagnosis: A Review. IEEE Sens. J. 2023, 23, 16481–16500. [CrossRef]
34. Shu, X.; Guo, Y.; Yang, W.; Wei, K.; Zhu, Y.; Zou, H. A Detailed Reliability Study of the Motor System in Pure Electric Vans by the
Approach of Fault Tree Analysis. IEEE Access 2020, 8, 5295–5307. [CrossRef]
Machines 2025, 13, 144 15 of 16

35. Filippetti, F.; Franceschini, G.; Tassoni, C.; Vas, P. Recent developments of induction motor drives fault diagnosis using AI
techniques. IEEE Trans. Ind. Electron. 2000, 47, 994–1004. [CrossRef]
36. Thomson, W.T.; Fenger, M. Current signature analysis to detect induction motor faults. IEEE Ind. Appl. Mag. 2001, 7, 26–34.
[CrossRef]
37. Zamudio-Ramirez, I.; Osornio-Rios, R.A.; Antonino-Daviu, J.A.; Razik, H.; de Jesus Romero-Troncoso, R. Magnetic Flux Analysis
for the Condition Monitoring of Electric Machines: A Review. IEEE Trans. Ind. Inform. 2022, 18, 2895–2908. [CrossRef]
38. Kiranyaz, S.; Devecioglu, O.C.; Alhams, A.; Sassi, S.; Ince, T.; Avci, O. Exploring Sound Versus Vibration for Robust Fault
Detection on Rotating Machinery. IEEE Sens. J. 2024, 24, 2895–2908. [CrossRef]
39. Gawde, S.; Patil, S.; Kumar, S.; Kamat, P.; Kotecha, K.; Abraham, A. Multi-fault diagnosis of Industrial Rotating Machines using
Data-driven approach: A review of two decades of research. Eng. Appl. Artif. Intell. 2023, 123, 106139. [CrossRef]
40. AlShorman, O.; Irfan, M.; Masadeh, M.; Alshorman, A.; Sheikh, M.A.; Saad, N.; Rahman, S. Advancements in condition
monitoring and fault diagnosis of rotating machinery: A comprehensive review of image-based intelligent techniques for
induction motors. Eng. Appl. Artif. Intell. 2024, 130, 107724. [CrossRef]
41. Das, O.; Das, D.B.; Birant, D. Machine learning for fault analysis in rotating machinery: A comprehensive review. Eng. Appl.
Artif. Intell. 2023, 9, e17584. [CrossRef] [PubMed]
42. Qi, R.; Zhang, J.; Spencer, K. A review on data-driven condition monitoring of industrial equipment. Eng. Appl. Artif. Intell. 2023,
16, 9. [CrossRef]
43. Surucu, O.; Gadsden, S.A.; Yawney, J. Condition Monitoring using Machine Learning: A Review of Theory, Applications, and
Recent Advances. Energy Syst. Appl. 2023, 221, 119738. [CrossRef]
44. Zhang, X.; Zhao, B.; Lin, Y. Machine Learning Based Bearing Fault Diagnosis Using the Case Western Reserve University Data: A
Review. IEEE Access 2021, 9, 155598–155608. [CrossRef]
45. Ni, Q.; Ji, J.; Halkon, B.; Feng, K.; Nandi, A.K. Physics-Informed Residual Network (PIResNet) for rolling element bearing fault
diagnostics. Mech. Syst. Signal Process. 2023, 200, 110544. [CrossRef]
46. Choudhary, A.; Mian, T.; Fatima, S.; Panigrahi, B.K. Passive Thermography Based Bearing Fault Diagnosis Using Transfer
Learning with Varying Working Conditions. IEEE Sens. J. 2023, 23, 4628–4637. [CrossRef]
47. Hu, W.; Wang, T.; Chu, F. Novel Ramanujan Digital Twin for Motor Periodic Fault Monitoring and Detection. IEEE Trans. Ind.
Inform. 2023, 19, 11564–11572. [CrossRef]
48. Attallah, O.; Ibrahim, R.A.; Zakzouk, N.E. CAD system for inter-turn fault diagnosis of offshore wind turbines via multi-CNNs
and feature selection. Renew. Energy 2023, 203, 870–880. [CrossRef]
49. Pietrzak, P.; Wolkiewicz, M. Demagnetization Fault Diagnosis of Permanent Magnet Synchronous Motors Based on Stator Current
Signal Processing and Machine Learning Algorithms. Sensors 2023, 23, 1757. [CrossRef] [PubMed]
50. Li, T.; Sun, C.; Li, S.; Wang, Z.; Chen, X.; Yan, R. Explainable Graph Wavelet Denoising Network for Intelligent Fault Diagnosis.
IEEE Trans. Neural Netw. Learn. Syst. 2024, 35, 8535–8548. [CrossRef] [PubMed]
51. Han, Y.; Zou, J.; Gong, B.; Luo, Y.; Wang, L.; Batlló, A.P.; Yuan, J.; Wang, C. The use of model-based voltage and current analysis
for torque oscillation detection and improved condition monitoring of centrifugal pumps. Mech. Syst. Signal Process. 2025,
222, 111781. [CrossRef]
52. Du, B.; Huang, W.; Cheng, Y.; Chen, J.; Tao, R.; Cui, S. Fault Diagnosis and Separation of PMSM Rotor Faults Using Search Coil
Based on MVSA and Random Forests. IEEE Trans. Ind. Electron. 2024, 71, 15089–15099. [CrossRef]
53. Skowron, M. Transfer Learning-Based Fault Detection System of Permanent Magnet Synchronous Motors. IEEE Access 2024,
12, 135372–135389. [CrossRef]
54. Chang, H.C.; Jheng, Y.M.; Kuo, C.C.; Hsueh, Y.M. Induction Motors Condition Monitoring System with Fault Diagnosis Using a
Hybrid Approach. Energies 2019, 12, 1471. [CrossRef]
55. Magadán, L.; Suárez, F.; Granda, J.; delaCalle, F.; García, D. Robust prediction of remaining useful lifetime of bearings using deep
learning. Eng. Appl. Artif. Intell. 2024, 130, 107690. [CrossRef]
56. Li, Y.; Wang, H.; Sun, M. ChatGPT-like large-scale foundation models for prognostics and health management: A survey and
roadmaps. Reliab. Eng. Syst. Saf. 2024, 243, 109850. [CrossRef]
57. Wang, H.; Li, C.; Li, Y.-F.; Tsung, F. An Intelligent Industrial Visual Monitoring and Maintenance Framework Empowered by
Large-Scale Visual and Language Models. IEEE Trans. Ind.-Cyber-Phys. Syst. 2024, 2, 166–175. [CrossRef]
58. Cheng, J.H.; Lu, C.L.; Zhang, G.; Wang, B.; Fang, J. Design of Motor Intelligent Monitoring and Fault Diagnosis System Based on
LoRa. IEEE Trans. Appl. Supercond. 2021, 31, 0601904. [CrossRef]
59. Dong, H.; Ma, H.; Wang, Z.; Man, J.; Jia, L.; Qin, Y. An Online Health Monitoring Framework for Traction Motors in High-Speed
Trains Using Temperature Signals. IEEE Trans. Ind. Inform. 2023, 19, 1389–1400. [CrossRef]
60. Liu, D.; Cui, L.; Wang, H. Rotating Machinery Fault Diagnosis Under Time-Varying Speeds: A Review. IEEE Access 2023,
23, 29969–29990. [CrossRef]
Machines 2025, 13, 144 16 of 16

61. Flusser, M.; Somol, P. Efficient anomaly detection through surrogate neural networks. Neural Comput. Appl. 2022, 34, 20491–20505.
[CrossRef]
62. Razzak, M.I.; Imran, M.; Xu, G. Big data analytics for preventive medicine. Neural Comput. Appl. 2022, 9, 4417–4451. [CrossRef]
[PubMed]
63. Ahmad, S.; Lavin, A.; Purdy, S.; Agha, Z. Unsupervised real-time anomaly detection for streaming data. Neurocomputing 2017,
262, 134–147. [CrossRef]
64. Kudelina, K.; Asad, B.; Vaimann, T.; Rassõlkin, A.; Khang, A.K.H.V. Methods of condition monitoring and fault detection for
electrical machines. Energies 2021, 14, 7459. [CrossRef]
65. Magadán, L.; Suárez, F.J.; Granda, J.C.; delaCalle, F.J.; García, D.F. A Robust Health Prognostics Technique for Failure Diagnosis
and the Remaining Useful Lifetime Predictions of Bearings in Electric Motors. Appl. Sci. 2023, 13, 2220. [CrossRef]
66. Kande, M.; Isaksson, A.J.; Thottappillil, R.; Taylor, N. Rotating electrical machine condition monitoring automation—A review.
Machines 2017, 5, 24. [CrossRef]
67. Wang, H.Z.Z.; Liu, X.; Gaudoin, O.; Xie, M. Joint optimization of condition-based production and maintenance with mutual
production-deterioration dependencies. Reliab. Eng. Syst. Saf. 2025, 256, 110679.
68. Ali, M.Z.; Shabbir, M.N.S.K.; Zaman, S.M.K.; Liang, X. Single- and Multi-Fault Diagnosis Using Machine Learning for Variable
Frequency Drive-Fed Induction Motors. IEEE Trans. Ind. Appl. 2020, 56, 2324–2337. [CrossRef]
69. Swana, E.F.; Doorsamy, W.; Bokoro, P. Tomek Link and SMOTE Approaches for Machine Fault Classification with an Imbalanced
Dataset. Sensors 2022, 22, 3246. [CrossRef]
70. Zhang, C.; Tan, K.C.; Li, H.; Hong, G.S. A Cost-Sensitive Deep Belief Network for Imbalanced Classification. IEEE Trans. Neural
Netw. Learn. Syst. 2019, 30, 109–122. [CrossRef] [PubMed]
71. Guo, L.; Lei, Y.; Xing, S.; Yan, T.; Li, N. Deep Convolutional Transfer Learning Network: A New Method for Intelligent Fault
Diagnosis of Machines with Unlabeled Data. IEEE Trans. Ind. Electron. 2019, 66, 7316–7325. [CrossRef]
72. Chen, X.; Yang, R.; Xue, Y.; Huang, M.; Ferrero, R.; Wang, Z. Deep Transfer Learning for Bearing Fault Diagnosis: A Systematic
Review Since 2016. IEEE Trans. Instrum. Meas. 2023, 72, 3508221. [CrossRef]
73. Russell, M.; Wang, P.; Liu, S.; Jawahir, I.S. Mixed-Up Experience Replay for Adaptive Online Condition Monitoring. IEEE Trans.
Ind. Electron. 2024, 71, 1979–1986. [CrossRef]
74. Li, Y.; Sun, Y.; Li, Z.; Chen, X.; Yang, L. Interpretable Spectra PCA for Condition Monitoring of Rotating Machinery: Theoretical
and Experimental Investigations. IEEE Trans. Instrum. Meas. 2024, 73, 35387122. [CrossRef]
75. Chen, G.; Yuan, J.; Zhang, Y.; Zhu, H.; Huang, R.; Wang, F. Enhancing Reliability Through Interpretability: A Comprehensive
Survey of Interpretable Intelligent Fault Diagnosis in Rotating Machinery. IEEE Sens. J. 2024, 12, 103348–103379. [CrossRef]
76. Sharma, J.; Mittal, M.L.; Soni, G. Condition-based maintenance using machine learning and role of interpretability: A review. Int.
J. Syst. Assur. Eng. Manag. 2024, 15, 1345–1360. [CrossRef]
77. Hendriks, J.; Azarm, M.; Dumond, P. Structured Data Ontology for AI in Industrial Asset Condition Monitoring. J. Sens. Actuator
Netw. 2022, 13, 23. [CrossRef]
78. Cocconcelliand, M.; Capelli, L.; Cavalaglio Camargo Molano, J.; Borghi, D. Development of a Methodology for Condition-Based
Maintenance in a Large-Scale Application Field. Machines 2018, 6, 17. [CrossRef]
79. ISO 55013; Asset Management—Guidance on the Management of Data Assets. International Standards Organisation: Geneva,
Switzerland, 2024.
80. Zhou, B.; Svetashova, Y.; Gusmao, A.; Soylu, A.; Cheng, G.; Mikut, R.; Waaler, A.; Kharlamov, E. SemML: Facilitating development
of ML models for condition monitoring with semantics. J. Web Semant. 2021, 71, 100664. [CrossRef]
81. Saki, M.; Abolhasan, M.; Lipman, J. A Novel Approach for Big Data Classification and Transportation in Rail Networks. IEEE
Trans. Intell. Transp. Syst. 2020, 21, 1239–1249. [CrossRef]
82. Chen, Q.; Cao, J.; Zhu, S. Data-Driven Monitoring and Predictive Maintenance for Engineering Structures: Technologies,
Implementation Challenges, and Future Directions. IEEE Internet Things J. 2023, 10, 14527–14551. [CrossRef]
83. van Oudenhoven, B.; Van de Calseyde, P.; Basten, R.; Demerouti, E. Predictive maintenance for industry 5.0: Behavioural inquiries
from a work system perspective. Int. J. Prod. Res. 2022, 66, 7846–7865. [CrossRef]
84. Khamaj, A.; Ali, A.M.; Saminathan, R.; Shanmugasundaram, M. Human factors engineering simulated analysis in administrative,
operational and maintenance loops of nuclear reactor control unit using artificial intelligence and machine learning techniques.
Heliyon 2024, 10, e30866. [CrossRef] [PubMed]
85. Frederiksen, R.D.; Bocewicz, G.; Radzki, G.; Banaszak, Z.; Nielsen, P. Cost-Effectiveness of Predictive Maintenance for Offshore
Wind Farms: A Case Study. Energies 2024, 17, 3147. [CrossRef]

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