Reliability Engineering and System Safety 234 (2023) 109204

Contents lists available at ScienceDirect

Reliability Engineering and System Safety

journal homepage: www.elsevier.com/locate/ress

Maintenance optimization in industry 4.0

Luca Pinciroli a, Piero Baraldi a, *, Enrico Zio a, b
a
Energy Department, Politecnico di Milano, Milan, Italy
b
MINES Paris, PSL, France

A R T I C L E I N F O A B S T R A C T

Keywords: This work reviews maintenance optimization from different and complementary points of view. Specifically, we
Maintenance optimization systematically analyze the knowledge, information and data that can be exploited for maintenance optimization
Industry 4.0 within the Industry 4.0 paradigm. Then, the possible objectives of the optimization are critically discussed,
Knowledge information and data
together with the maintenance features to be optimized, such as maintenance periods and degradation thresh­
Optimization approaches
Uncertain systems
olds. The main challenges and trends of maintenance optimization are, then, highlighted and the need is iden­
tified for methods that do not require a-priori selection of a predefined maintenance strategy, are able to deal
with large amounts of heterogeneous data collected from different sources, can properly treat all the un­
certainties affecting the behavior of the systems and the environment, and can jointly consider multiple opti­
mization objectives, including the emerging ones related to sustainability and resilience.

productivity, efficiency and sustainability [6]. On the other hand,

planning maintenance in practice is far from being a trivial task, as it
1. Introduction
requires to consider the components health state, future profiles of
system operation and aspects such as spare parts inventory and future
Modern society relies on highly technological and mechanized in­
production demand, with all associated uncertainties [7].
dustries and services to produce and distribute commodities. The in­
The increase interest in academia and demand in practice for
dustrial assets are inevitably affected by aging and degradation of their
maintenance optimization is demonstrated by the published works on
components, which strongly impact on production availability and
the topic, both theoretical and applied. Fig. 1 shows the evolution, in the
product quality. To counteract this, Operation and Maintenance (O&M)
time period between 2000 and 2020, of the number of published works
activities are carefully planned and carried out, at a significant fraction
on maintenance optimization. The data have been obtained considering
of the total business cost. For example, in food-related industries,
the publications indexed in the Scopus database and containing in the
average maintenance costs represent about 15% of the production cost
title, in the abstract or in the keywords, the terms “maintenance opti­
[1], in wind farms the O&M cost can reach 20%-25% of the life cycle
mization” and at least one among the following terms: “scheduled”,
cost [2], in mining industry the maintenance costs often account for over
“opportunistic”, “condition-based”, “predictive” and “prescriptive”. The
30% of the total production cost [3] and in nuclear power plants and
increase of the ratio between the number of works related to mainte­
heavy industries, the O&M cost is about 60%-70% of the total cost of
nance optimization and the total number of works indexed by Scopus
production [4].
(rescaled in the picture by dividing by 1000) confirms the increasing
The perception of maintenance and of maintenance management has
interest in maintenance optimization [8–10].
substantially changed through the years. Traditionally, it has been seen
Fig. 2 shows the repartition of the number of works on maintenance
as “a necessary evil”, i.e., a set of expensive activities to perform only if
optimization in the various industrial fields reviewed in this work. It is
unavoidable, because required or even mandatory, and that cannot
evident that maintenance optimization is pervasive in almost all in­
enhance profitability [5]. With the acquired field experience of the
dustries, with manufacturing and energy (wind power, oil & gas, power
negative impacts of failures and accidents, the role of maintenance has
generation and distribution) sectors accounting for more than 50% of
emerged as a strategic one, for anticipating components failures and
the total number of works considered.
system degradation of performance, and nowadays production man­
Since the initial stages of the research on maintenance optimization,
agers invest significant resources in the development and implementa­
which date to the 1960s, several issues had been identified as
tion of maintenance strategies for improving asset reliability,

* Corresponding author.
E-mail address: [email protected] (P. Baraldi).

https://doi.org/10.1016/j.ress.2023.109204
Received 22 June 2022; Received in revised form 17 February 2023; Accepted 24 February 2023
Available online 27 February 2023
0951-8320/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

Acronyms and symbols ML Machine Learning

MOGA Multi-Objective Genetic Algorithm
AHP Analytic Hierarchy Process MOM Maintenance Organization Model
AI Artificial Intelligence MSA Metaheuristic Search Algorithm
ANP Analytic Network Process NLP Natural Language Processing
BA Bayesian Approach O&M Operation & Maintenance
BCM Business Centered Maintenance RCM Reliability Centered Maintenance
CNN Convolutional Neural Network RL Reinforcement Learning
CPS Cyber Physical System RNN Recurrent Neural Network
DMG Decision Making Grid RUL Remaining Useful Life
DP Dynamic Programming SM Simulation Model
DSS Decision Support System SP Stochastic Programming
DT Decision Tree TOPSIS Technique for Order Preference by Similarity to Ideal
ELECTRE Elimination Et Choice Translating Reality Solution
FL Fuzzy Logic K Knowledge, information and data
GA Genetic Algorithm F Objective function
IoT Internet of Things f Vector of the optimization criteria
KID Knowledge, Information and Data x Vector of the features to be optimized
MA Mathematical Approach h Constraint function
MCDM Multiple Criteria Decision Making b Constraint function bound
MIP Mixed Integer Programming

challenging: maintenance strategies, which remarkably increases the complexity

of the optimization problem;
a the difficulty in retrieving data, information and knowledge for the c the difficulty of implementing the developed maintenance optimi­
development and validation of maintenance optimization models zation approaches in real-world applications, which generates a gap
and approaches for practical applications. This issue especially af­ between theory and practice. According to [11], this is mainly due
fects newly designed and safety-critical systems, for which scarce to: i) the difficulty of explaining to maintenance planners the main­
information is available related to the degradation and failure pro­ tenance optimization models, which are often seen as black-boxes
cesses of their components [7,10,11]; providing unintelligible maintenance recommendations, ii) the cost
b the difficulty of dealing with the complexity of industrial systems, in of developing maintenance optimization methods, which is not
terms of number of components and dependencies among them. a-priori guaranteed to be balanced by the benefit of implementing
Multi-unit systems had not been considered in the first surveys on the optimized maintenance policies.
maintenance optimization in the 1960s [12,13], and the lack of
effective methods to treat them emerged in the 1970s [14]. Actually, Nowadays, maintenance practice can benefit from the technological
even recent surveys have underlined that, although a lot of efforts developments driving the so called Industry 4.0 revolution. It was firstly
had been devoted to the development of maintenance optimization introduced in Germany in 2011 as a part of the country’s high-tech
models for multi-unit systems [7,15,16], the problem is still not fully strategy [17] and the topic was quickly adopted globally to lead the
solved. One key challenge is the proper identification and account of development and improvement of 21st century industry. In the Industry
functional dependencies of different nature, e.g., economic, sto­ 4.0 concept, production systems are built as smart systems in the form of
chastic, structural or logic, among system components, which are Cyber Physical Systems (CPS), which enable real-time communication
extremely difficult to handle when they co-exist [7]. Also, the and cooperation between humans and machines [18], and more efficient
different units of a multi-unit system most likely require different and flexible production to meet more challenging performance and

Fig. 1. Number of publications about maintenance optimization and total number of publications (divided by 1000) indexed in the Scopus database from 2000
to 2020.

2
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

safety goals. Industry 4.0 is based on the retrieval of large amount of maintained on a regular schedule basis, monitored and/or are suit­
data from the systems and the exploitation of the advancements in able for a run-to-failure strategy by means of the analysis of the
sensors, robotics and new technologies such as Artificial Intelligence failure modes and consequences [30], and according to the company
(AI), Machine Learning (ML), augmented reality, big data analytics, and objectives [31];
Internet of Things (IoT) [19]. b the approaches used to support the optimization procedure but not
With respect to issue a) above, new sources of information on com­ directly aimed at the optimization of the maintenance strategy, such
plex multi-unit systems have been made available, e.g., real-time data as Bayesian Approaches (BAs) [10], which are widely applied to
and images related to the operation and to the health state of the com­ identify the value of the unknown parameters of failure time distri­
ponents. The effective use of this information for maintenance optimi­ butions given some empirical data [32] and to estimate the failure
zation allows reducing the dependence on the subjective information probabilities [33,34], Simulation Models (SMs) relying on Petri net
from experts’ knowledge. Indeed, AI algorithms are able to effectively [35,36], Markov chain [37] or Monte Carlo simulation [38], which
exploit such information for detecting anomalies, diagnosing their are used to model the system behavior and to evaluate the goodness
causes and predicting failure times, operating conditions and demands of different maintenance strategies;
[20]. The outcomes of the AI algorithms can be used to adapt mainte­ c the frameworks employed to define an effective maintenance strat­
nance plans to the actual and predicted conditions of the components egy, such as Decision Support Systems (DSS), i.e., model-based sets of
and systems. With respect to issue b) above, deep learning algorithms procedures for processing data and judgments to support and
can effectively deal with the big data collected from the complex systems improve the decision-making [39], and Maintenance Organization
[21–23], from which they can identify the functional dependencies Models (MOMs), i.e., schemes to be followed during the organization
among their components [24]. Finally, with respect to issue c) above, of maintenance, which combine administrative, managerial and
methods have been developed to interpret the outcomes of AI and ML technical activities involved in maintenance strategy planning [40,
algorithms [25,26], and offer ways to visualize the maintenance strategy 41].
for understanding and explaining the maintenance strategies identified
by the optimization method [27]. The remainder of this review work is organized as follows: mainte­
The present survey reports and analyzes maintenance optimization nance strategies are presented in Section 2; Section 3 analyses the
within the Industry 4.0 paradigm. We firstly review systematically: i) the maintenance optimization problem in terms of the knowledge, infor­
knowledge, information and data available for maintenance optimiza­ mation and data available, optimization criteria and optimization out­
tion, ii) the optimization criteria typically considered, and iii) the out­ comes; Section 4 discusses the optimization approaches; in Section 5, the
comes of the optimization. The objective of the survey is to illustrate the main challenges related to maintenance optimization and the emerging
advancements already achieved in maintenance optimization and those trends are presented. Finally, Section 6 concludes the work.
that can be potentially obtained, the challenges to be addressed and the
most promising trends of methods development. The review considers 2. Maintenance strategies
also the recently developed approaches based on Reinforcement
Learning (RL) and the prescriptive maintenance strategy paradigm that, The concept of asset maintenance includes all actions aimed at
at the best of the authors’ knowledge, have not been considered in monitoring, restoring or enhancing the functionality of an asset, or at
previous surveys. This work is expected to be useful for maintenance preventing the asset to lose all or part of its functionality [42]. A
management professionals and researchers working on maintenance maintenance strategy is the set of rules according to which the different
optimization. maintenance actions are performed on the asset. It includes rules on the
Since the focus is on maintenance optimization, we purposely do not type of maintenance actions, on the maintenance instances, on the
consider: components or sub-systems priorities, on the spare parts flow on the
maintenance technical crew to involve [31,43]. Accordingly, mainte­
a the logical processes, such as Reliability-Centered Maintenance nance strategies are grouped into [44]:
(RCM) and Business-Centered Maintenance (BCM) [28,29], which
aim at identifying which components in a system should be

Fig. 2. Repartition of maintenance optimization into the different industrial fields of application, in the considered papers.

3
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

• Corrective maintenance. It restores the functionality of an asset after exploits failure projections to optimize the operational implica­
its failure. It involves only repairment or replacement procedures. It tions of maintenance tasks [57]. The recommended actions can be
is suitable for non-safety critical systems for which the maintenance maintenance actions or operational actions: for example, the
interventions can be performed quickly and at low costs, and whose repair of a pump or its running at a lower than nominal flow rate
failures do not induce severe consequences [45,46]. can be prescribed to slow down its degradation process for
• Preventive maintenance. It aims at preventing the asset to lose its allowing the timely delivery of a new piece of equipment. Pre­
functionality, by performing maintenance actions before failure oc­ scriptive maintenance requires the availability of historical and
curs. Five main types of preventive maintenance strategies are operational data collected in a wide variety of operating condi­
typically identified: tions [58], and advanced models of the considered system, e.g.
i Scheduled maintenance. It aims at preventing the asset to lose its digital mirrors and twins [59–62].
functionality through maintenance actions that are performed at
scheduled instances, both irregular or periodic. Typically, statis­ Fig. 3 shows the evolution of the number of publications relative to
tical data collected from assets, e.g., failure times and mainte­ the different maintenance strategies mentioned above. Notice that the
nance durations [47], are used to define the maintenance plan interest in scheduled maintenance has been decreasing, whereas the
[42]. The schedule optimization task is difficult since degradation interest in condition-based and predictive maintenance has been
mechanisms are complex and characterized by large uncertainty increasing, coherently with the development of the enabling technology.
[48]. Scheduled maintenance is suitable for high-risk systems, e. Some works about prescriptive maintenance have been recently pub­
g., systems whose failure may lead to severe safety consequences, lished [63]. This trend is confirmed by the results of the survey in [64],
can cause large production losses or whose maintenance planning according to which 79% of the interviewed professionals (mainly from
can provide economic advantages, e.g., because of not easily energy, transportation and manufacturing sectors) believe that predic­
available spare parts, which should be ordered in advance. tive and prescriptive maintenance of equipment will play a fundamental
ii Opportunistic maintenance. It aims at performing maintenance on role in their companies in the future. Nevertheless, the scheduled
more asset elements or sub-systems at the same time. This is done, maintenance strategy is still popular among systems managers and re­
for example, by combining the maintenance activities of compo­ searchers, with several works related to its optimization being still
nents characterized by similar failure rates and operation condi­ published in recent years. Also, mixed maintenance strategies have been
tions, or by exploiting a planned shutdown or an undesired adopted in some industrial applications. For example, a mixed mainte­
breakdown as an opportunity to maintain several different com­ nance strategy combining corrective, scheduled and opportunistic
ponents [49]. This maintenance strategy is suitable for systems maintenance has been developed to minimize the life cycle cost of
characterized by similar components or undergoing long, planned rolling bearings in [65]. Corrective, condition-based and predictive
shutdowns, e.g., nuclear power plants for refueling, and for sys­ maintenance strategies are mixed to minimize the maintenance cost and
tems whose maintenance activities require the rental of expensive maximize the reliability of nuclear power plant feeding pipes in [66]. To
equipment, e.g., a crane or a ship. counterbalance the fact that preventive maintenance interventions can
iii Condition-based maintenance. Similarly to scheduled maintenance, be imperfect, i.e., they are not able to restore equipment in
it aims at preventing the asset to lose its functionality, but the as-good-as-new condition, a mixed maintenance strategy composed of
planning of the maintenance interventions is based on the elab­ preventive actions, e.g., lubrication, cleaning, and adjustment, and
oration of data collected from the asset itself to evaluate its con­ corrective actions, e.g., replacement, is proposed and optimized in [67].
ditions. The application of condition-based maintenance requires When condition-based or predictive maintenance strategies are adopted,
the availability of a monitoring system to collect data of physical it can be useful to perform scheduled inspections to check the asset
quantities related to degradation of the asset [50]. Then, by health state during system shutdowns.
applying fault detection and diagnostic techniques [51], In practice, there is not an automatic way to select the most appro­
abnormal conditions are detected and diagnosed, calling for priate maintenance strategy for a specific system: each maintenance
specific maintenance actions to be performed [52]. strategy has its own particular characteristics and the maintenance en­
Condition-based maintenance is suitable for systems in which the gineer should take into account several aspects, e.g., company objec­
advantages of avoiding unplanned shutdowns caused by failures tives, type of system, failure consequences, maintenance costs,
overcome the costs of the monitoring system and of the devel­ availability of spare parts, etc.
opment of the detection and diagnostic tools [53].
iv Predictive maintenance. As an extension of condition-based main­ 3. Maintenance optimization
tenance, it processes further the monitoring data for prognostics
[54,55], i.e., to estimate the failure time, thus allowing planning In general, an optimization problem involves a vector of features of
the maintenance activities in advance [56]. It requires the the system to be optimized, x = [x1 , …, xn ], an objective function to be
development of the monitoring system and the prognostic tools. minimized (maximized), F(x) : Rn → R, which may involve q different
The variable and uncertain conditions under which the compo­ criteria, f (x) = [f1 (x), …, fq (x)], and possibly m constraint functions,
nents are operating can strongly influence the degradation pro­ hi (x) : Rn →R, i = 1, …, m, with associated bounds, bi , which limit the
cesses and failure mechanisms, thus requiring the proper choices on x because of physical, economic, environmental or other
treatment of data characterized by several sources of uncertainty. reasons. Then, the optimization problem can be mathematically framed
Predictive maintenance is suitable for systems which can benefit in terms of the identification of the vector x which satisfies [68]:
from the same advantages of condition-based maintenance, but
can also further benefit from planning in advance, e.g., due to not argminx (argmaxx ) F(x)
(1)
subject to hi (x) ≤ bi , i = 1, …, m
easily available spare parts which need to be ordered.
v Prescriptive maintenance. It goes beyond estimating the compo­ Specifically, in maintenance optimization, the features in the vector,
nents failure time by exploring hypothetical scenarios generated x, to be optimized relate to aspects of the maintenance planning, such as
by the O&M management. In fact, starting from the monitoring the interval between consecutive instances of maintenance intervention,
data collected from the asset, prescriptive maintenance provides a the degradation threshold for detection, the type of maintenance actions
recommendation of the actions to be taken by anticipating the to be performed, etc. The objective function, F(x), quantitatively de­
potential scenarios generated by such actions and evaluating their scribes objectives such as profit, reliability, safety and sustainability.
effects on the system. In other words, prescriptive maintenance The constraint functions, hi (x), and associated bounds, bi , are set

4
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

Fig. 3. Relative number of publications related to the optimization of different maintenance strategies from 2000 to 2020 [www.scopus.com].

according to specific physical limits of design and operation, e.g., the been used in the context of maintenance optimization to model com­
maximum power that can be produced by a system, and regulations, ponents degradation in nuclear [74] and manufacturing industries [78],
such as the maximum allowed interval of time between two consecutive availability and revenues in wind power industry [79] and maintenance
instances of inspection, or the minimum reliability accepted or costs in nuclear industry [80]. Numerical data, such as failure times and
maximum cost allowed. maintenance costs, are typically used to set the model parameters [48],
In practice, the definition of x, F(x), hi (x), bi depends on the avail­ to properly represent uncertainty [81,82] and the system health state
able knowledge, information and data, K, about the behavior of the [83]. In the context of Industry 4.0, signal measurements input to
system and its environment. models for fault detection, diagnostics and prognostics, in support to
The remaining part of this section will discuss the elements of the condition-based, predictive and prescriptive maintenance approaches.
optimization problem defined above, that are: i) knowledge, informa­ For example, the potential of using data for maintenance optimization
tion and data available, K, (Section 3.1), ii) optimization criteria, f (x), was shown in a manufacturing plant [84], in a wind farm [85], in
(Section 0), iii) optimization features, x, (Section 3.3). aeronautical systems [86] and in infrastructures [87]. Table 2 reports
the classification of the considered works in terms of type of KID and
3.1. Knowledge, information and data industrial field of application. It can be seen that independently from the
industrial field, models and data are the main sources of KID.
In practice, different sources of Knowledge, Information and Data Fig. 4 represents the maintenance strategies considering the KID
(KID) [69], K, concur to the definition of the optimization problem in typically used for their identification and development. It shows that
Eq. (1), depending on availability and on the input required by the each maintenance strategy requires different sources of KID to be
specific optimization method employed. They can be organized with properly developed. Reliability and availability models are used for
respect to: i) the type of KID, which is here classified as “expert developing scheduled [88] and opportunistic [89] maintenance strate­
knowledge”, “mathematical models” and “data”, where the latter can be gies. Degradation models and real time data about components health
in the form of numbers, texts and images, and ii) the topic, i.e., what the states are fundamental for the development of condition-based [83] and
KID represent. With respect to the latter, the KID typically involved in predictive [90] maintenance strategies. Data and models of the oper­
maintenance optimization represent characteristics of the maintenance ating conditions are employed to develop prescriptive maintenance
intervention, of the system and components reliability, availability and strategies, which require considering their influence on components
safety, of the degradation processes and health states of the system and degradation and failure [85].
components, e.g., the normal/abnormal condition outcome of an Even if some works have considered textual data for the development
anomaly detection module, the classification of the type of abnormal of reliability, availability and maintainability models [91], and images
condition, i.e., the outcome of a fault diagnostic module, and the pre­ have been used for the development of diagnostics models [92], these
diction of the component Remaining Useful Life (RUL), i.e., the outcome two types of data have not yet been used for maintenance optimization
of a fault prognostic module, of the system operating conditions and purposes. This is because text and image processing typically relies on
other information needed for the definition of the objective function. ML methodologies, such as Natural Language Processing (NLP) tech­
Table 1 reports the classification of some works about maintenance niques and Convolutional Neural Networks (CNN), which are difficult to
optimization in terms of type and topic of K. Expert knowledge is integrate within an optimization problem and whose functioning and
fundamental when new technologies, for which limited data are avail­ results are difficult to understand and interpret by maintenance plan­
able, are considered. It is mainly used for the definition of the objective ners. We expect that with the advancement of concepts of Industry 4.0
function [44], the set of feasible maintenance strategies [70] and the and Internet of Things (IoT) [93], the capability of ML methodologies in
setting of the values of model parameters and constraints [71]. It has dealing with large amounts of heterogeneous data and the development
been used in different sectors, such as in chemical [72], manufacturing of techniques for the interpretation of AI algorithm outcomes, textual
[70] and oil & gas [73] industries. Mathematical models are typically data and images, will become more and more relevant to the field of
used for describing component degradation processes [74,75] and the maintenance optimization.
effects of maintenance activities [76], and for monitoring the system
health state [77]. Stochastic models are typically exploited to deal with 3.2. Optimization criteria
the uncertainty inherent in stochastic processes such as degradation or
the evolution of operating and environmental conditions. They have The objective function F(x), which drives the maintenance strategy

5
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

Table 1
Classification of the knowledge, information and data involved in maintenance optimization, with respect to type and topic.
KID topic Maintenance Reliability, Degradation Health state Operating Objective
KID type availability and process Detection Diagnostics Prognostics conditions function
safety

Expert knowledge • Set of • Parameter Parameter Parameters of the • Weights to be

possible values, e.g., values, e.g., models for associated to
maintenance failure rate degradation and simulating the the different
alternative [96] and failure operating maintenance
strategies reliability thresholds [97]. conditions [85]. criteria [44];
[94]; threshold • Fitness of
• Set of [71]. alternatives
possible with respect
O&M actions to the criteria
[95]. [98].
Models Models of the • Models of • Models of Models for the Models for Models for the Models for the
duration [99], system system detection of fault prediction of predictions of the
effect [76] and reliability, degradation system diagnostics of the Remaining operating
cost [81] of the availability [74]. abnormal the system Useful Life environment
maintenance and safety, e. conditions, components (RUL) of the based on present
interventions. g., Bayesian based on data- based on data- system and past time
networks driven driven AI components, measurements, e.
[100], statistics and approaches based on g., wind speed,
Markov AI approaches [77]. model-based energy demand
models [101], [102]. and data- [89].
Petri nets driven
[36], Monte approaches
Carlo [77].
simulation
models [83].
Data Numerical • Maintenance • Failure times; • Simulation • Result of • Results of • Real time • Information
cost [103]; • Databases of parameters, e. inspections, inspections, signal related to
• Information component g., transition e.g., non- e.g., non- values system
about failure rates, probabilities destructive destructive [109]; quantities, e.g.,
previous e.g., OREDA among test and test and • Remaining production
inspections, [105]; degradation degradation degradation Useful Life level [110],
e.g., time • Availability states [107]; indicator indicators estimations buffer or spare
since last and • Parameters of values [108]; [81]. parts inventory
maintenance Reliability of the [108]; • Real time levels [111];
[104] or type the degradation • Real time signal values • Meteorological
of components models [48]; signal values [83]. data [89].
maintenance [106]; • Information [83].
intervention. • Transition related to
probabilities/ measurement
rates in uncertainty,
Markov e.g.,
models of measurement
system noise [81].
reliability and
availability.
Textual Maintenance • Maintenance Maintenance
reports and reports and reports and
work orders. work orders; work orders.
• Accident
investigation
reports.
Images • Televisual • Televisual • Televisual
inspections; inspections; inspections;
• Thermal • Thermal • Thermal
images; images; images;
• X-rays • X-rays • X-rays
images. images. images.

optimization, is often defined considering several different optimization 109], safety / risk [150,184] and resilience [154] are typical quan­
criteria, f (x), [182]. In this respect, it is possible to distinguish between titative metrics used.
approaches that consider a single criterion and approaches that consider
multiple criteria. The works which optimize multiple criteria jointly consider metrics
The works which optimize a single criterion employ performance quantifying:
metrics related to:
1 the effectiveness of personnel management and logistics; for
1 the economic benefit provided by the maintenance strategy, e.g., example, the quality of the shift schedule for the workers [185,186]
maintenance cost [71,103], life-cycle cost [157], profit [81,83,121], and of the management of the spare parts inventory [111,129] have
production loss and unmet demand [110,144]; been considered;
2 system safety and reliability, considering failure occurrences and
mitigation of failure consequences; availability [183], reliability [88,

6
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

Table 2 the field of application, the most used criteria are economic and safety-
Classification of the considered works in terms of type of KID and industrial field related [182], with the economic criteria mostly used in non-safety
of application. critical applications. Differently from what was pointed out in [182],
KID type Expert Mathematical Data it can be noticed that the trend is shifting towards multi-objective
Industrial field knowledge models optimization problems, in which several application-related criteria
Manufacturing [70,110, [78,84,110,111, [78,84,110–114, are jointly considered. This is due to the increasing complexity of in­
industry 112–120] 114–118,120–128] 116–120, dustrial systems, which are expected to simultaneously satisfy multiple
123–129] objectives. For example, in production plants, it is desirable to minimize
Wind power [81,85,100, [79,81,85,89,100, [79,81,85,89,100,
the maintenance cost while maximizing the machines availability and
industry 130–133] 131–144] 131–144]
Power production [145–148] [74,145–151] [74,145–151] the production output [120]. Furthermore, in recent years, maintenance
industry has become a key factor for sustainable operation and production,
Power distribution [152,153] [101,107,152–156] [101,107, leading to an increase in the number of research papers on sustainable
industry 152–156] maintenance management. The objectives are typically related to the
Infrastructures [87] [87,96,104, [87,96,104,
157–160] 157–160]
efficient management of resources and energy, the reduction of wastes
Aeronautical [86,161,162] [86,90,163–165] [86,90,162–165] produced by maintenance and of storage material [188], the reduction
industry of the maintenance environmental impact, including hazardous emis­
Oil & Gas industry [44,73, [73,168,169] [44,73,167–169] sions caused by system malfunctioning [189], and the increase of
166–168]
workers and public safety [190]. However, it has been pointed out that
Transportation [170–173] [171–174] [171–174]
industry research in maintenance optimization is still limited and mainly focused
Chemical industry [72,175] [108,176] [72,108,175,176] on conventional performance criteria [191]. Another criterion that has
Automation [177] [177] recently emerged for critical systems and infrastructures is resilience,
systems which is defined as the ability of a system to withstand potentially
Semiconductor [178]
high-impact disruptions, by mitigating impacts and quickly recovering
industry
Mining industry [88,179] [88,179] normal conditions [192]. Resilience is considered to be fundamental in
Military industry [94] [180] [180] the context of Industry 4.0, since nowadays systems can be affected by
Water distribution [181] [181] several potential disruptive events, such as natural events, pandemics,
industry
cyber-attacks [193], and their ability to quickly recover their function­
Data storage [71] [71] [71]
systems alities is of utmost importance. Then, it is reasonable to think that in the
next years more and more researchers and practitioners will consider
environmental impact, sustainability and resilience as criteria of main­
2 the effects of maintenance on the asset performance from the point of tenance optimization.
view of sustainability, environmental impact [158,174,187] and
production quality [44,166]; 3.3. Optimization outcomes
3 the time loss, e.g., the effects of time delays caused by maintenance
on other activities [96] and on the total maintenance time [128]; Maintenance optimization concerns different types of features (x in
4 the feasibility of performing the maintenance interventions [70,98] Eq. (1)) and considers them in different ways. They include the
and the accessibility of the components [167]. following, in relation to the optimization outcome:

Table 3 reports the classification of the considered works in terms of

optimization criteria and industrial field of application. Regardless of

Fig. 4. Representation of the maintenance strategies in relation to the type of required information.

7
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

Table 3
Classification of the considered works in terms of optimization criteria and industrial field of application.
Optimization criteria Economic Criteria Safety criteria Management Performance Temporal Practicality
Industrial field criteria criteria criteria criteria

Manufacturing industry [70,84,110,111,116–123,125, [70,113–118,120,124, [129] [113,115,120] [120] [70]

126,129] 126,127]
Wind power industry [79,81,85,89,100,130–132, [130–132] [130] [130]
134–144]
Power production [74,145–151] [145,150,151]
industry
Power distribution [101,107,152,153,155,156] [107,154]
industry
Oil & Gas industry [44,73,166–169] [44,73,166–168] [44,166] [44,167]
Infrastructures [87,96,157–160] [104,159] [158] [96]
Aeronautical industry [86,161,164,165] [90,109,162,164,165] [86]
Transportation industry [171,172,174] [170,172,173] [172] [174] [172]
Mining industry [88,179] [88]
Military industry [94] [180]
Automation systems [177]
Semiconductor industry [178] [178]
Water distribution [181] [181]
industry
Chemical industry [72,108,175,176] [72,175] [108]
Data storage systems [71]

• optimal maintenance strategy among several a-priori predefined al­ be maintained at the same time, assuming an a-priori maintenance
ternatives [44,108]; some works produce also a ranking of the al­ strategy, e.g., scheduled;
ternatives with respect to the optimization criteria; • optimized inventory management strategy [129,147]; the outcome
• optimized parameters values defining a single maintenance strategy consists in the optimized spare parts flow or the optimized spare
selected a-priori [76,88], e.g., the maintenance period or age parts or product inventory management strategy.
threshold for scheduled maintenance, the degradation threshold for
condition-based maintenance, or the type of action performed, e.g., Table 4 reports the classification of the considered works with
repairment, replacement; respect to the topic of the KID and the optimization outcome. Notice that
• optimal maintenance action to be performed for given data, such as the information provided by the anomaly detection, fault diagnostics
monitoring signals or operating conditions [104,133]; in this case, and prognostics, is not considered in the majority of the works on
the a-priori selection of the maintenance strategy is not needed since maintenance optimization. This is mainly due to the fact that researchers
the outcome is directly the action to be performed, e.g., repair, have been mainly focused on the development of detection, diagnostics
replace, order the spare parts, or decrease the production level to and prognostics methods for individual components of different types,
reduce the degradation rate; and only recently the interest has shifted towards the exploitation of the
• optimized grouping of components for opportunistic maintenance outcomes of these methods for maintenance optimization. The challenge
[194,195]; the outcome consists in the optimal set of components to is that diagnostics and prognostics methods typically consider single
components, whereas maintenance optimization requires to take

Table 4
Classification of the considered works in terms of required knowledge, information and data, topic and outcome.
Optimization Best maintenance strategy Optimized parameters for the a-priori Optimal maintenance Optimal grouping Optimized
outcome among several alternatives selected maintenance strategy strategy of components for inventory
KID topic maintenance management
strategy

Maintenance [44,70,72,94,98,100, [48,67,71,74,78–80,83,84,88–90,97,99, [73,81,85–87,95,104, [96,194,195] [99,111,129,147]

112–117,130,131,145,157, 101,106–111,118,119,121–126,128,129, 127,133,146–148,152,
161,162,166,167,170,173, 132,134–144,149,150,154–156,158–160, 153,168,171,172,203]
175,178,181] 164,165,169,174,176,177,179,180,185,
151,196–202]
System reliability, [72,113,157,167] [67,71,74,76,78–80,83,84,88,89,99,101, [87,95,146,148,171, [80,194,195] [99]
availability and safety 106,107,109,119,124–126,132,134, 172,203]
136–141,143,149,150,154–156,165,169,
177,185,197–200,202]
Degradation process [113,157,178] [48,67,74,78–80,83,88,89,97,101,106–110, [73,81,85–87,95,104, [80,194,195] [111,147]
118,119,122–126,137,140,141,143,149, 127,133,146–148,152,
156,164,165,169,174,176,177,179,180, 153,168,171,172,203]
185,151,196–201]
System Detection [100,157,178] [80,83,118,125,134,140,149,198] [87,171,172] [80]
health Diagnostics [83,125,144,149,198] [172]
state Prognostics [84,90,109,134,144,176,177] [81,85,86,95,133,147, [147]
148]
Operating condition [114,116] [89,90,99,110,111,118,119,121–125,128, [73,81,85,86,95,127, [96,195] [99,111,129]
129,134,137–140,144,155,159,164,165, 133,148,152,153,168,
176,177,198–200] 172]
Objective function [44,70,72,94,98,112–114, [194]
130,161,166,167,170,173,
175,181]

8
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

decisions considering the whole system. Furthermore, it can be noticed optimization of the maintenance strategy of an oil refinery [44] and a
that only few works, which try to achieve predictive or prescriptive wind farm [130]. The integration of MCDM with Fuzzy Logic (FL)
maintenance, use input data and information from most of the cate­ was extensively studied to cope with the uncertainty and the
gories listed above. This is due to the fact that, as already commented in subjectivity of the decision making process [112,166,205]. A
Section 2, these maintenance strategies require a large amount of data to generalized version of AHP, i.e., Analytic Network Process (ANP),
be properly implemented and deployed. has been applied to the selection of the best maintenance strategy for
The most popular outcomes of maintenance optimization are i) a chemical plant [175]. The main advantage of ANP is that the de­
ranking of different maintenance alternatives and ii) optimized parameters cision process is structured as a network instead of a hierarchy and
values for the a-priori given maintenance strategy: the two share the need of this makes it suitable to deal with the interdependencies among the
a-priori selecting the maintenance strategy. This way, the obtained criteria [206]. Other popular MCDM algorithms are the Technique
maintenance strategy is optimal with respect to a limited set of options. for Order Preference by Similarity to Ideal Solution (TOPSIS) and the
Actually, in the context of Industry 4.0, it can be expected that pre­ Elimination Et Choice Translating Reality (ELECTRE). TOPSIS is a
scriptive maintenance will become more and more popular [64] and, decision-making algorithm in which the best solution among the
therefore, approaches are expected to be developed, which give as possible alternatives is chosen by selecting the one which minimizes
outcome the optimal maintenance action considering the present state of the Euclidean distance from the ideal optimal option and maximizes
the system. They are expected to be advantageous since they do not the Euclidean distance from the worst possible option [207]. It was
require assuming a predetermined maintenance strategy. applied to select the optimal maintenance plan for military equip­
ment [94], manufacturing plants [70] and aeronautical systems
4. Optimization approaches [161]. In [113] and [72], a FL/TOPSIS-based approach was proposed
to deal with the selection of the most suitable maintenance strategy.
The optimization approaches are here presented considering the ELECTRE is based on the concept of outranking relations between
features, x, to be optimized. Section 4.1 will be dedicated to Multiple alternatives [208]. It was used to select the best set of elements to be
Criteria Decision Making, Decision Making Grid and Decision Tree, replaced at each scheduled maintenance in a multi-unit system [194]
which have been mainly applied to the identification of the best among and it was applied to the selection of the optimal maintenance
several alternatives of maintenance strategy. Section 4.2 will present strategy of compressors in the chemical industry, and in water dis­
Mathematical Approaches, Mixed Integer Programming, Dynamic Pro­ tribution networks [98,181]. The simplicity of these algorithms and
gramming and Metaheuristic Search Algorithms, which have been the high interpretability of their outcomes contributed to their
mainly used for optimizing the parameters of a predefined maintenance popularity for maintenance optimization, even if inter-dependence
strategy. Section 4.3 will introduce Reinforcement Learning to select the between alternatives and objectives can lead to inaccurate results
optimal maintenance actions to be performed. Table 5 reports the works [209].
in which the described optimization approaches have been applied to • Decision making grid (DMG): DMG is a graphical support tool used
obtain the different outcomes. to help the decision makers in selecting the most effective mainte­
nance strategy by considering multiple criteria, such as downtime,
4.1. Approaches for the identification of the best maintenance strategy failure frequency and failure cost. The main drawback is that it
among a predefined set of alternatives strongly relies on the user’s experience and can provide biased re­
sults. DMG was applied to automotive industry [170] and aero­
The following algorithms have been mainly applied to select the best nautical systems [162]. A fuzzy logic DMG was also proposed to
maintenance strategy among a predefined set of alternatives. They use consider equipment reliability and criticality [114]. DMG are useful
experts’ knowledge as one of the main sources of input, which allow when the available KID, K, is limited to expert knowledge.
considering both quantitative and qualitative aspects of maintenance: • Decision Tree (DT): DT is a decision-making support tool whose
outcomes are easy to interpret and which allows comparing the
• Multiple Criteria Decision Making (MCDM): A commonly applied performance of different alternatives in obtaining an established
MCDM method for O&M optimization is the Analytic Hierarchy goal, while considering random events, possible decisions and their
Process (AHP), which hierarchically structures the decision process consequences [210]. DTs are generally not suitable to deal with
into a series of pairwise comparisons and allows considering both complex problems and for long time-horizons [100]. DTs were
qualitative and quantitative aspects [204]. It was applied to the structured to identify the optimal maintenance strategy of gas [145]

Table 5
Classification of the considered works in terms of optimization approach and outcome.
Optimization Best maintenance strategy Optimized parameters for the a-priori Optimal maintenance Optimal grouping of Optimized inventory
outcome among several selected maintenance strategy strategy components for management strategy
Optimization alternatives maintenance
approach

MCDM [44,70,72,94,98,112,113, [194]

130,161,166,167,175,
181]
DMG [114,162] [170]
DT [100,115,145,178]
MA [121] [48,99,122,123,143,177,197] [171] [99]
MIP [84,89,90,124,134–136,149,154,155,165, [172]
176,202]
DP [107,160,164,179,201] [104] [96] [111]
MSA [94,120,156,169–171] [65,78–80,83,101,106,116–119,123,125, [195,200]
128,131,132,137–139,141,142,151,
156–159,173,195,199,200,219,227]
RL [180,198] [81,85–87,95,110,127, [147]
146,148,152,153,168,
169,203]

9
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

and wind turbines [100], and in the semiconductor [178] and equipment under parameters uncertainty [201]. The main issues
manufacturing [115] industries. with DP are the curse of dimensionality and the need of explicitly
defining the transition probabilities among all the possible system
4.2. Approaches to optimize the parameters of an a-priori selected states, which makes it unsuitable for complex systems [215].
maintenance strategy • Metaheuristic Search algorithms (MSAs): Metaheuristics are
computational procedures used to approximate the solution of an
With respect to the setting of the optimal parameters of an a-priori optimization problem by iteratively improving the candidate solu­
predefined maintenance strategy, the following approaches have been tions [216,217]. Genetic algorithms (GAs) are one of the most pop­
used: ular MSAs. They are based on the principles of genetics and natural
selection [218]. GAs were applied to set the degradation threshold
• Mathematical Approaches (MAs): MAs comprise all the ap­ for condition-based maintenance [83], to choose the best mainte­
proaches in which the optimization problem is formulated by means nance plan for a network of infrastructure facilities [157], to
of mathematical equations, which are then solved by means of dif­ schedule preventive maintenance interventions in the manufacturing
ferential calculus to identify the optimal parameters of the mainte­ and railway industries [106,173]. In [137,142], GAs were used to
nance strategy. In [121], a MA was used for optimizing maintenance optimize the scheduled maintenance strategy of a wind farm
profitability. In [122], a MA based on the Riccati equation was used considering the stochasticity of wind power production. GAs were
to identify a sub-optimal production and maintenance plan which applied to identify the opportunistic maintenance strategy for in­
maximizes the total profit of a manufacturing system. In [99], a MA dustrial components [195,200] and the condition-based mainte­
was proposed to optimize the inventory management and the nance strategy of degrading nozzles of gas turbines [151]. Also, GAs
scheduled maintenance strategy of a single unit. In [44] and [127] a were used to optimize the scheduled maintenance strategy of a
MA was used for scheduling maintenance considering uncertainty. In multi-unit system [199], the inspection and maintenance planning of
[143] a MA is developed to define the optimal imperfect preventive pressure vessels [108], the opportunistic maintenance plan of an
maintenance plan. Finally, a MA was used to optimize the prescrip­ onshore wind farm considering the dependencies among the com­
tive maintenance strategy of locomotive wheels in the railway in­ ponents [138] and the deterioration thresholds for condition-based
dustry [171]. The use of MAs for maintenance optimization is mainly maintenance [125]. Multi-Objective GAs (MOGAs) were adopted
limited to simple systems for which the optimization problem can be for the optimization of the preventive maintenance plan [120] and
solved analytically or numerically, but it is unfeasible for complex the inspection policy of a safety system [219], simultaneously
systems unless simplifications of the system behavior are introduced. considering several optimization criteria. Finally, in [118], a
• Mixed integer programming (MIP): MIP is the area of optimization GA-based framework was developed to analyze the advantages of
that addresses optimization problems with continuous and integer optimizing the scheduled maintenance plan starting from different
variables in the objective or in the constraints. The problems can be initial guesses of the maintenance policy in manufacturing industry.
linear (MILP) or present non-linearities (MINLP) [211]. The appli­
cation of MIP to maintenance optimization requires using integer Other MSAs used for maintenance optimization are: Grid search,
variables to represent the possible maintenance optima. MIP was Nelder-Mead algorithm [220], Harmony Search algorithm [221],
applied to the optimization of the maintenance schedule of a wind Differential Evolution [222], Particle Swarm Optimization algorithm
farm [89,134–136] and a power distribution system [154,155] and [223], Simulated Annealing [224], Artificial Colony Optimization
to the optimization of the condition-based maintenance strategy of a algorithms, e.g., ant colony optimization [225] and artificial bee
gas turbine considering the value of information [149]. The joint colony [226]. Grid search was used to set the optimal scheduled
optimization of the flight and maintenance plans of aircrafts was maintenance interval in the power distribution industry [101], to
performed using MIP in [90]. In [124], a practical integrated pro­ optimize the mixed maintenance strategy of battery packs [156], to
duction and scheduled maintenance planning was addressed devel­ compare several production, setup and maintenance policies of a
oping a MIP model, which considers the system’s manufacturing manufacturing system [116], and in the wind power industry [131,
capacity and its reliability. When some of the variables need to be 141]. Nelder-Mead algorithm was used to develop a simulation and
modeled by means of random variables, to deal with uncertainty, optimization platform to analyze the performance of several main­
Stochastic Programming (SP) is implemented [212]. SP was used to tenance policies in manufacturing industry [117]. Harmony Search
define the optimal maintenance schedule for a multi-unit system algorithm was used to find the best maintenance strategy for bridges
[165,202] and for the joint production and predictive maintenance infrastructures [158]. Particle Swarm Optimization algorithm was
optimization of a chemical plant [176]. Although its popularity, the applied to optimize the predictive maintenance interval of a
main drawback of MIP is that the computation time tends to rapidly manufacturing system [119] and the opportunistic maintenance
increase with the complexity of the systems [213]. strategy for a wind farm [139]. Simulated Annealing [224] was
• Dynamic Programming (DP): DP is a method for solving multi- applied to find the optimal scheduled maintenance plan of bridge
stage decision problems [214]. It is based on the concept of networks [159] and of multi-unit systems [126]. Ant Colony Opti­
breaking down complex problems into simpler sub-problems. For mization was used to optimize the maintenance scheduling of
example, for problems which involve long time horizons, DP con­ multi-unit systems [128] and offshore wind turbines [132], whereas
structs a sub-problem to be solved recursively, at each time step. DP an Artificial Bee Colony was applied to the optimization of the
was used to determine the optimal maintenance strategy of road opportunistic maintenance strategy of a wind farm [227]. MSAs are
networks considering budget constraints [96]. It was also applied to simple to understand and easily adaptable to different optimization
the optimization of the scheduled maintenance plans considering problems. On the other hand, they are slow to converge and do not
spare parts inventory management [111] and of the replacement guarantee convergence towards the global optimum.
strategy under uncertainty for assets in the mining industry [179]. In
[164], a DP-based methodology for the optimization of the mainte­ 4.3. Approaches for the selection of the optimal maintenance actions
nance check schedules in the aeronautical industry was presented.
DP was also proposed to determine the optimal maintenance strategy The most applied approach for the direct selection of the optimal
for power cables, considering the stochastic nature of cable failures actions to be performed is reinforcement learning:
[107], and used to deal with the optimization of lifetime-extending
maintenance decisions for aging infrastructures [160] and

10
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

• Reinforcement Learning (RL): RL is a branch of machine learning, real-time [230]. Also, the extensive use of data-driven approaches in
based on DP, in which the learning agent learns the optimal set of Industry 4.0 requires to properly represent and treat model and data
actions to maximize a properly defined reward function by inter­ uncertainty since its wrong quantification can lead to sub-optimal or
acting, in a trial and error manner, with the environment [215]. even erroneous decisions [230].
Differently from DP, model-free RL does not require the definition of With respect to challenge 2), Industry 4.0 makes data acquisition and
the transition probabilities among the system states, which makes it processing technologies easily accessible. However, the tradeoff be­
suitable for dealing with maintenance optimization of complex sys­ tween the opportunities of exploiting new KID for maintenance opti­
tems. When an artificial neural network is employed as learning mization and the capital investments required to purchase the sensors
agent, all the available sources of information, including predictions and software necessary to perform ad hoc analyses and to properly train
about the future evolution of components health state and operating the operators to use the outcomes for their decisions on operation,
conditions, can be exploited as input. This can help the development control and maintenance [93], should be carefully evaluated [231]. For
of condition-based and predictive maintenance strategies, which safety critical systems, e.g., nuclear power plants, aeronautical systems,
receive data from the systems in real time. Also, the output can be the or for systems in which maintenance cannot be easily performed, e.g.,
best action to be performed at every time step, resulting in a pre­ offshore wind farms, the advantages of using new sources of KID have
scriptive maintenance strategy, without the need of a-priori selecting been intensively studied [232] and several approaches have been pro­
a maintenance strategy [81,110]. RL was used to select the best time posed. Notice that the approaches should, on the one hand, properly
to perform maintenance, assuming a condition-based maintenance manage the increasing KID becoming available during the system life
framework [198] and to optimize the time between consecutive cycle and, on the other hand, they should deal with the fact that the KID
maintenance interventions assuming a scheduled maintenance for new technologies and production processes may be, initially, not
strategy [180]. In [147], gas turbine parts flow was optimized by sufficient for the implementation of advanced maintenance strategies,
means of RL, and tabular [152] and neural network-based [153] RL such as predictive or prescriptive ones.
were applied to the O&M optimization of power grids. RL was For what concerns challenge 3), Industry 4.0 comes in a historical
applied to find the optimal maintenance strategy for a deteriorating time in which new challenges related to environment and modern so­
pumping system equipped with health monitoring capabilities [95] ciety are receiving an ever-increasing attention. The concepts of sus­
and to optimize the opportunistic maintenance strategy for a tainability and resilience are getting more and more critical and need to
manufacturing plant in [127]. Also, it was applied to the optimiza­ be considered by the companies, together with safety and economics-
tion of the maintenance strategy considering multi-state systems related objectives. Despite their importance, their consideration is not
[73], a wind farm [81,85], aeronautical systems [86], a steel widespread among practitioners and it is typically limited to qualitative
manufacturing line [110], infrastructures [87] and a generic aspects due to the lack of formal metrics for their evaluation [191].
multi-component system using ML. Finally, RL was applied to oil & Finally, with respect to challenge 4), Industry 4.0 is changing the
gas pipeline networks [168] and to nuclear power plants prescriptive perception of maintenance: from monitoring the degradation state of the
maintenance optimization [146,148]. Despite its advantages, RL components and anticipating their failures to prescribing the most
applications are limited by the large computation cost and by the suitable action to optimally manage the whole system considering the
non-guaranteed convergence of the solution to the optimal one dynamic production environment in which it is embedded [19]. This
[215]. requires the development of an optimization framework suitable to
• Other approaches, already commented in Sections 4.1 and 4.2, have process all sources of information available, with the associated un­
been applied for the selection of the optimal action to be performed. certainties, and manage the large number of system states and possible
For example, MIP was applied to the optimization of the prescriptive maintenance actions.
maintenance strategy of railway infrastructures [172] and DP was
used to determine the optimal maintenance strategy for bridge decks 5.2. Trends in maintenance optimization methods
[104].
In response to the challenges presented in the previous subsection,
5. Findings the emerging trends of maintenance optimization methods are here
analyzed. Table 6 reports the main trends and their impacts on the
In this Section the challenges related to maintenance optimization in definition of the optimization problem in terms of KID, K, optimization
Industry 4.0 are analyzed (Section 5.1) and the emerging trends in the criteria, f (x), and outcomes, x.
methods to address them are discussed (Section 5.2).
5.2.1. Complexity of the industrial systems
5.1. Challenges in maintenance optimization With respect to the joint optimization of multiple criteria (challenge
1.a) in Section 5.1), MCDM and MSAs are expected to gain attention for
The following practical challenges of maintenance optimization in application in the next years. The value of MCDM approaches lies in the
Industry 4.0 emerge from the previous sections: (1) complexity of the fact that they are suitable to deal also with qualitative aspects, that they
industrial systems, (2) data acquisition and processing, (3) new opti­ provide easily interpretable solutions and that they do not require any
mization criteria and (4) prescriptive maintenance. particular expertise in information technology. As pointed out in [233],
MCDM approaches, especially AHP, have been applied to solve problems
Challenge 1) calls for the development of methods able to deal with: of maintenance strategy selection in which the best maintenance strat­
1.a) multipurpose systems for which multiple criteria should be egy among several alternatives is to be selected considering re­
jointly optimized; quirements at the organizational level and company goals. Given the
1.b) the uncertainty of the complex system behavior and the sto­ subjectivity of the results, which is due to the qualitative nature of the
chasticity of the environment in which the system operates; criteria and the use of expert’s knowledge for the evaluation, it is ex­
1.c) unknown dependences and inter-dependencies among compo­ pected that the research in this area will move towards the combination
nents, subsystems, systems and even systems of systems [228,229]. of MCDM with methods to manage uncertainty, such as FL [234], and
the use of big data to extract more objective information.
These issues require to move away from static maintenance strate­ MSAs have been shown to provide optimal maintenance solutions for
gies, which are not suitable to deal with unexpected events, and develop complex systems characterized by significant uncertainty in their
dynamic maintenance strategies for adapting to the context changing in behavior. They are adaptable to many different problem formulations

11
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

Table 6
Findings of the present review.
Trends in Industry Consequences on the maintenance optimization problem Trends in optimization approaches
K f x

Complexity of the industrial systems: • Need to manage uncertainty in Need to jointly optimize • Interest in multi-objective optimiza­
• Multipurpose systems; data and models; multiple criteria. tion, e.g., MCDM, MSAs;
• Large uncertainty in the system • Need to extract information • Interest in AI and ML techniques to
behavior and stochasticity of the about components dependencies manage the inter-dependencies
environment; from data. among the components, e.g., MSA,
• Unknown dependences and inter- RL;
dependencies among components • Interest in AI and ML techniques
and subsystems. intrinsically able to treat uncertainty,
e.g., MSA, RL;
• Enhancement of MCDM with FL and
use of SP
Advancements in sensors and sensor Need to deal with: • AI and ML techniques to manage big
technology • Large amount of data; data and extract information from
• Heterogeneous data (numerical them;
signals, images and texts). • Reduced dependence on experts’
knowledge.
Availability of AI algorithms for data Possibility of exploiting: Integration of AI and ML techniques, e.
mining • Real time estimations of the g., NLP, CNN, within the optimization
present and future health states approach.
of system components;
• Estimation of present and future
operating conditions.
New and more challenging New criteria: • Multi-objective optimization;
performance and safety goals • sustainability/ • Methods to consider quantitative and
environmental qualitative criteria, e.g., MCDM.
impact;
• resilience.
Operation and maintenance Possibility of exploiting: Need to identify optimal Increasing interest in new optimization
considered as two sides of the • Detection, diagnostic and operation and maintenance approaches, i.e., RL
same coin (prescriptive prognostics information; actions to be performed.
maintenance) • Estimation of present and future
operating conditions.

and can be used also with non-differentiable objective functions. Despite has to interact with. Indeed, despite the learning agent can discover the
their popularity, GAs require the empirical setting of some hyper­ optimal maintenance policy by means of direct interactions with the
parameters, such as population size, crossover and mutation probabili­ real-world system, this turns out to be unfeasible for economic, safety
ties, whose setting can affect both the goodness of the solution and the and time issues [215]. Specifically, due to the trial-and-error nature of
convergence speed. In this context, Self-Organizing GAs, which auto­ the learning process, the agent would need to perform several times the
matically adapt the hyperparameters to the characteristics of the specific actions suggested by the algorithm to explore their outcomes, leading to
problem, are a promising research direction [235]. Limitations of MSAs economically inconvenient and unsafe system management in the early
are that they do not guarantee convergence to the global optimum and stages of the learning process, when the actions selected are not yet
the computation of the fitness value of the candidate solutions can be optimal. To improve this issue, the learning agent should be trained
very demanding. This latter problem can be tackled by developing fast using a white-box model of the system representative of the real-world
surrogate AI-based models for the computation of the fitness. environment.
With respect to the management of the uncertainty induced by the Other approaches, such as MCDM and MIP, have been shown to be
increased complexity of the systems in a stochastic environment (chal­ suitable to deal with uncertainty when combined with FL and SP,
lenge 1.b) in Section 5.1), new ML approaches, such as RL, are expected respectively, whereas some other approaches, such as MSAs and DP,
to further rise in popularity. In RL, the learning agent learns how to deal have already been successfully applied to maintenance optimization
with the stochasticity of the environment and the variability of the ef­ considering uncertainty [142,201,202,205].
fects of the performed actions by directly interacting with the environ­ With respect to the management of unknown dependencies and
ment [236]. Another major advantage of RL is that it tackles the problem inter-dependencies among components and subsystems (challenge 1.c)
of maintenance management dynamically, i.e., considering the effects of in Section 5.1), AI and ML algorithms have been used to identify de­
the O&M actions on the future system behavior and degradation evo­ pendencies among components from the information collected from the
lution, which allows identifying the actions to be performed at every system. For example, alarms signals have been used to identify de­
decision time [54]. Despite these potentialities, the application of RL to pendencies in complex technical infrastructures, allowing the reduction
maintenance optimization of complex systems is still limited by the of the computational burden of classical association rule mining ap­
following issues: i) the large computational effort, which is due to the proaches [238]. In the context of Industry 4.0, these approaches are
low convergence speed of RL and ii) the difficulty of explaining the expected to attract the interest of the researchers dealing with complex
rationale behind the selection of the maintenance actions, which, in systems of systems, since they can discover previously unknown de­
some cases, may appear counterintuitive and, therefore, obstacles its pendencies. Furthermore, being these methods able to identify de­
practical application. To overcome the latter limitation, studies are pendencies among the components of different nature, they can be
being devoted to the development of explainable RL [237], with the extremely useful when opportunistic maintenance is adopted, since they
objective of justifying the actions suggested by the RL agent. Finally, the help grouping different components to be maintained during the same
implementation of the most promising RL approaches, which are based maintenance opportunity [239].
on deep learning, require a great amount of data and the development of
an accurate and realistic model of the environment the learning agent

12
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

5.2.2. Data acquisition and processing true, given that the most suitable maintenance strategy should be
The second trend highlighted in Table 6 is related to the advance­ selected according to the characteristics of each component, e.g., func­
ments of Industry 4.0 in sensors and sensor technology, which makes tionality, costs, criticality, environmental legislations, and the company
available a large amount of heterogenous data containing valuable in­ objectives. Therefore, a dynamic and flexible maintenance strategy,
formation on the system state, the degradation of the components and adaptable to the specific conditions of the system and its environment
the environment. Specifically, the use of historical data, such as signal should be preferred. For this reason, we expect the developments of
values, images and maintenance reports, and of real time information methodologies that require maintenance engineers to list the possible
collected from the system is expected to reduce the dependence of the operation and maintenance actions, without a-priori selecting a main­
maintenance optimization on the subjectivity of experts’ knowledge tenance strategy for all components in all conditions. According to our
and, therefore, contributing to reduce the uncertainty and leading to a analysis, MIP, DP and RL emerge as possible ways to tackle this issue, but
more unbiased decision-making process. In this regard, one of the main it is expected that other approaches will be proposed to optimize pre­
challenges of maintenance optimization methods is to fully exploit all scriptive maintenance in the near future.
the available KID. To this aim, MAs and MCDM approaches are expected
to be replaced by MIP, MSAs and RL, which have been shown to be able 5.2.5. Further comments
to manage large amounts of data in optimization problems [240]. In Table 7 reports the classification of the considered papers according
particular, RL can be trained including feedbacks from on-field opera­ to the adopted optimization approaches and the industrial field
tors, allowing the learning agent to learn how they would act in a spe­ considered. It can be noticed that most of Industry 4.0 applications focus
cific situation [241]. on the use of MIP, MSAs and RL in manufacturing, energy and aero­
The integration of new AI algorithms (third trend in Table 6) in nautical industries. Also, few works consider real-world case studies,
maintenance optimization approaches is a necessary research direction whereas many works focus on generic multi-unit systems. This high­
to take into high consideration, especially considering the demand for lights that the gap between maintenance theory and practice is still
predictive and prescriptive maintenance. Also, that data-driven ap­ present and needs to be narrowed by means of capital investments by the
proaches are capable of dealing with uncertainty [242]. Yet, although companies and more realistic case studies by the researchers. Also,
applications of autoencoders to detect anomalies [243,244], DNN to noticed that, although many industrial fields were not explored, the
real-time estimate the present and future health states of components developed methods are general and can be easily transferred to other
[245], Recurrent Neural Networks (RNN) to catch the dynamic evolu­ field where the same sources of KID are available.
tion of the signals [246,247], CNN to classify images [248] and NLP to Finally, as already pointed out in several surveys on maintenance
extract information from texts [249,250] have been proposed, the optimization, it is important to mention the lack of benchmarks for the
effective integration of these algorithms into the methods for mainte­ evaluation of maintenance strategies on well-defined case studies. In
nance optimization is still in its infancy. The challenge is to pass from the [257], a benchmark has been proposed to compare the performance of
capability of performing fault detection, diagnostics and prognostics on different algorithms for the optimization of scheduled maintenance in a
a single component to optimize the maintenance of a complex system power plant. Although it limits the problem to the optimization of the
composed by interacting and dependent components using the infor­ maintenance schedule, it can be considered a starting point for bench­
mation provided in real time by fault detection, diagnostics and prog­ marking in maintenance optimization.
nostics algorithms [58].
6. Conclusions
5.2.3. New optimization criteria
Although Industry 4.0 includes some objectives in terms of energy In this work we have presented a survey on maintenance optimiza­
efficiency and environmental impact [251], its original concept focuses tion. The analysis has been focused on the identification of the knowl­
on enhancing performance, efficiency and safety of industry by means of edge, information and data available in the context of Industry 4.0, the
the possibilities provided by the technological advancements in AI, optimization criteria of interest and the possible outcomes of the opti­
cyber-physical systems, internet of things and robotics. In the last few mization. Maintenance management professionals and researchers
years, the interest of modern society has broadened to new challenges working on maintenance optimization can find in the present review
related to resilience and sustainability. Industry 5.0, which has been reference case studies and a guideline to select the maintenance opti­
proposed as an extension of Industry 4.0, focuses on the role of research mization approach given the characteristics of the industrial system and
and innovation to support industry in a long-term service to humanity the objectives. It emerges that the complexity of the modern systems
[251,252], taking into account the worldwide spread challenges that calls for the development of maintenance optimization methods able to
affect the society the most. Consequently, maintenance optimization jointly optimize several objectives and to properly treat the large un­
will definitely evolve to consider new criteria together with those certainty affecting the system behaviors and the environment in which
related to performance and safety. This requires the definition of they operate. Also, the advancements in sensors and sensor technology
measurable quantities to evaluate the performance of a specific main­ and the availability of new AI and ML algorithms offer the possibility to
tenance strategy with respect of system resilience and sustainability. For mine very useful information on the present and future health states of
example, a metric to quantify resilience has been defined in [253], system components, which need to be properly considered for mainte­
whereas a new metric based on return on investment was introduced to nance optimization at the system level. The analysis of the optimization
consider at the same time safety, sustainability, reliability, and resil­ criteria has shown that several industrial sectors are demanding to
ience [254]. It can be expected that many researchers will focus on consider new metrics related to the concepts of sustainability and
defining new ad-hoc metrics to integrate new criteria of interest in resilience within maintenance optimization. Also, there is an increasing
existing maintenance optimization approaches. interest towards prescriptive maintenance, which considers operation
and maintenance as two sides of the same coin, and overcomes the need
5.2.4. Prescriptive maintenance of a-priori selecting a maintenance strategy to be applied to the system
The last trend highlighted in Table 6 is prescriptive maintenance, during the time horizon of the optimization.
which is rapidly gaining popularity among researchers, even if the The capability of the different optimization methods to deal with the
literature works related to its optimization are still very limited [56, identified challenges has been reviewed. Although at the present state of
255]. This is due to the fact that it is common to think that complex the art there is not a single approach able to satisfactory tackle all the
maintenance strategies are always the best solutions and that corrective open issues, the analysis performed in this work allows concluding that
maintenance should always be avoided [93,256], which is not always multi-objective MSAs and RL-based approaches are among the most

13
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

Table 7
Classification of the considered works in terms of industrial field of application and optimization approach.
Optimization approach MCDM DMG DT MA MIP DP MSA RL
Industrial field

Manufacturing industry [70,113] [114] [115] [121–123] [84,124] [111] [78,116–120,125,126,128,129] [110,127]
Wind power industry [130] [100] [143] [89,134–136,144] [79,131,132,137–142] [81,85]
Power production industry [145] [149] [74,150,151] [146–148]
Power distribution industry [154,155] [107] [101,156] [152,153]
Infrastructures [157–159] [96,104,160] [87]
Aeronautical industry [161] [162] [90,109] [164,165] [86]
Oil & Gas industry [44,166,167] [169] [73,168]
Transportation industry [170] [171] [172] [173,174]
Chemical industry [72,175] [176] [108]
Mining industry [179] [88]
Military industry [94] [180]
Automation system [177]
Semiconductor industry [178]
Water distribution industry [181]
Data storage system [71]

promising maintenance optimization approaches, given their capability [5] Ben-Daya M, Duffuaa SO. Maintenance and quality: the missing link. J Qual
Maint Eng 1995;1(1):20–6. https://doi.org/10.1108/13552519510083110.
of dealing with the joint optimization of several criteria and with the
[6] Marais KB, Saleh JH. Beyond its cost, the value of maintenance: An analytical
uncertainty of the system behavior and of the environment. framework for capturing its net present value. Reliab Eng Syst Saf 2009;94(2):
644–57. https://doi.org/10.1016/j.ress.2008.07.004.
[7] de Jonge B, Scarf PA. A review on maintenance optimization. Eur J Oper Res
CRediT authorship contribution statement 2020;285(3):805–24. https://doi.org/10.1016/j.ejor.2019.09.047.
[8] Sherif YS, Smith ML. Optimal maintenance models for systems subject to failure -
Luca Pinciroli: Conceptualization, Methodology, Validation, Visu­ a review. Nav Res Logist Q 1981;28(1):47–74. https://doi.org/10.1002/
nav.3800280104.
alization, Formal analysis, Investigation, Writing – original draft,
[9] Garg A, Deshmukh SG. Maintenance management: Literature review and
Writing – review & editing. Piero Baraldi: Conceptualization, Meth­ directions. J Qual Maint Eng 2006;12(3):205–38. https://doi.org/10.1108/
odology, Validation, Visualization, Formal analysis, Investigation, 13552510610685075.
[10] Sharma A, Yadava GS, Deshmukh SG. A literature review and future perspectives
Writing – original draft, Writing – review & editing, Supervision. Enrico
on maintenance optimization. J Qual Maint Eng 2011;17(1):5–25. https://doi.
Zio: Conceptualization, Methodology, Formal analysis, Writing – review org/10.1108/13552511111116222.
& editing, Supervision. [11] Dekker R. Applications of maintenance optimization models: A review and
analysis. Reliab Eng Syst Saf 1996;51(3):229–40. https://doi.org/10.1016/0951-
8320(95)00076-3.
Declaration of Competing Interest [12] McCall JJ. Maintenance policies for stochastically failing equipment : a survey.
Manage Sci 1965;11(5):493–524.
[13] Barlow RE, Proschan F. Mathematical Theory of Reliability. New York: John
The authors declare that they have no known competing financial Wiley, New York; 1965.
interests or personal relationships that could have appeared to influence [14] Pierskalla WP, Voelker JA. A survey of maintenance models: the control and
surveillance of deteriorating systems. Nav Res Logist Q 1976;23(3):353–88.
the work reported in this paper. https://doi.org/10.1002/nav.3800230302.
[15] Cho DI, Parlar M. Invited review a survey of maintenance models for multi- unit
Data availability systems. Eur J Oper Res 1991;51:1–23.
[16] Wang H. A survey of maintenance policies of deteriorating systems. Eur J Oper
Res 2002;139(3):469–89. https://doi.org/10.1016/S0377-2217(01)00197-7.
Data will be made available on request. [17] Vogel-Heuser B, Hess D. Guest editorial industry 4.0-prerequisites and visions.
IEEE Trans Autom Sci Eng 2016;13(2):411–3. https://doi.org/10.1109/
TASE.2016.2523639.
[18] Wang S, Wan J, Zhang D, Li D, Zhang C. Towards smart factory for industry 4.0: a
Acknowledgments self-organized multi-agent system with big data based feedback and coordination.
Comput Netw 2016;101:158–68. https://doi.org/10.1016/j.
comnet.2015.12.017.
iRel40 is a European co-funded innovation project that has been [19] Wang J, Gao RX. Chapter 7 - innovative smart scheduling and predictive
granted by the ECSEL Joint Undertaking (JU) under grant agreement No maintenance techniques. In: Mourtzis D, editor. Design and Operation of
876659. The funding of the project comes from the Horizon 2020 Production Networks for Mass Personalization in the Era of Cloud Technology.
Elsevier; 2022. p. 181–207.
research programme and participating countries. National funding is
[20] Candanedo IS, Nieves EH, González SR, Martín MTS, Briones AG. Machine
provided by Germany, including the Free States of Saxony and Thur­ learning predictive model for industry 4.0. Commun Comput Inf Sci 2018;877
ingia, Austria, Belgium, Finland, France, Italy, the Netherlands, (July):501–10. https://doi.org/10.1007/978-3-319-95204-8_42.
Slovakia, Spain, Sweden, and Turkey. [21] Bode G, Thul S, Baranski M, Müller D. Real-world application of machine-
learning-based fault detection trained with experimental data. Energy 2020;198:
117323. https://doi.org/10.1016/j.energy.2020.117323.
References [22] Gollapudi S. Practical machine learning. Packt Publishing Ltd; 2016.
[23] Zanero S. Cyber-Physical Systems. Computer (Long Beach Calif). 2017;50(4):
14–6. https://doi.org/10.1109/MC.2017.105.
[1] Mobley RK. Impact of maintenance. In: Engineering P, editor. An Introduction to
[24] Antonello F, Baraldi P, Shokry A, Zio E, Gentile U, Serio L. A novel association
Predictive Maintenance. Second. Burlington: Butterworth-Heinemann; 2002.
rule mining method for the identification of rare functional dependencies in
p. 1–22.
Complex Technical Infrastructures from alarm data. Expert Syst Appl 2021;170
[2] Li YF, Valla S, Zio E. Reliability assessment of generic geared wind turbines by
(January):114560. https://doi.org/10.1016/j.eswa.2021.114560.
GTST-MLD model and monte carlo simulation. Renew Energy 2015;83:222–33.
[25] Carvalho DV, Pereira EM, Cardoso JS. Machine learning interpretability: a survey
https://doi.org/10.1016/j.renene.2015.04.035.
on methods and metrics. Electron 2019;8(8):1–34. https://doi.org/10.3390/
[3] Komljenovic D, Paraszczak J, Kecojevic V. Potential for improvement of
electronics8080832.
reliability and maintenance in mining operations based on nuclear industry
[26] Zio E. Prognostics and Health Management (PHM): where are we and where do
know-how and experience. In: Mine Planning and Equipment Selection 2005;
we (need to) go in theory and practice. Reliab Eng Syst Saf 2022;218(PA):
2005. p. 143–52. https://doi.org/10.1201/noe0415401173.ch17.
108119. https://doi.org/10.1016/j.ress.2021.108119.
[4] J.B. Coble, P. Ramuhalli, L.J. Bond, W. Hines, and B. Upadhyaya, “Prognostics
[27] Pinciroli L, et al. Agent-based modeling and reinforcement learning for
and health management in nuclear power plants: a review of technologies and
optimizing energy systems operation and maintenance: the Pathmind solution. In:
applications,” 2012.

14
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

30th Eur. Saf. Reliab. Conf. ESREL 2020 15th Probabilistic Saf. Assess. Manag. [56] Zonta T, da Costa CA, da Rosa Righi R, de Lima MJ, da Trindade ES, Li GP.
Conf. PSAM 2020; 2020. p. 176–1480. https://doi.org/10.3850/978-981-14- Predictive maintenance in the Industry 4.0: a systematic literature review.
8593-0. Comput Ind Eng 2020;150(August):106889. https://doi.org/10.1016/j.
[28] Waeyenbergh G, Pintelon L. A framework for maintenance concept development. cie.2020.106889.
Int J Prod Econ 2002;77(3):299–313. https://doi.org/10.1016/S0925-5273(01) [57] Meissner R, Meyer H, Wicke K. Concept and economic evaluation of prescriptive
00156-6. maintenance strategies for an automated condition monitoring system. Int J
[29] Sajaradj Z, Huda LN, Sinulingga S. The application of reliability centered Progn Heal Manag 2021;12(3):1–17. https://doi.org/10.36001/ijphm.2021.
maintenance (RCM) methods to design maintenance system in manufacturing. v12i3.2911.
IOP Conf Ser Mater Sci Eng 2019;505(1). https://doi.org/10.1088/1757-899X/ [58] Choubey S, Benton R, Johnsten T. Prescriptive equipment maintenance: a
505/1/012058. framework. In: Proc. - 2019 IEEE Int. Conf. Big Data, Big Data 2019; 2019.
[30] Bloom NB. Reliability Centered Maintenance (RCM): Implementation Made p. 4366–74. https://doi.org/10.1109/BigData47090.2019.9006213.
Simple, 148. New York: McGraw-Hill Inc.; 2006. [59] Tao F, Zhang H, Liu A, Nee AYC. Digital twin in industry: state-of-the-art. IEEE
[31] Kelly A. Maintenance strategy: Business centred maintenance. Oxford: Trans Ind Informatics 2019;15(4):2405–15. https://doi.org/10.1109/
Butterworth-Heinemann; 1997. TII.2018.2873186.
[32] Sheu SH, Yeh RH, Lin YB, Juang MG. Bayesian approach to an adaptive [60] Fuller A, Fan Z, Day C, Barlow C. Digital twin: enabling technologies, challenges
preventive maintenance model. Reliab Eng Syst Saf 2001;71(1):33–44. https:// and open research. IEEE Access 2020;8:108952–71. https://doi.org/10.1109/
doi.org/10.1016/S0951-8320(00)00072-7. ACCESS.2020.2998358.
[33] Ben Said A, Shahzad MK, Zamai E, Hubac S, Tollenaere M. A Bayesian network [61] J. Franzen, J. Stecken, R. Pfaff, and B. Kuhlenkötter, “Using the digital shadow for
based approach to improve the effectiveness of maintenance actions in a prescriptive optimization of maintenance and operation,” vol. 1, no. Table 1,
Semiconductor Industry. PHM Soc Eur Conf 2013;2(1):1–11. pp. 265–276, 2019, doi: 10.1007/978-3-030-13535-5_19.
[34] Leoni L, BahooToroody A, De Carlo F, Paltrinieri N. Developing a risk-based [62] Worden K, Cross EJ, Gardner P, Barthorpe RJ, Wagg DJ. On digital twins, mirrors
maintenance model for a natural gas regulating and metering station using and virtualisations. In: Conference Proceedings of the Society for Experimental
Bayesian network. J Loss Prev Process Ind 2019;57(August 2018):17–24. https:// Mechanics Series. 3; 2020. p. 285–95. https://doi.org/10.1007/978-3-030-
doi.org/10.1016/j.jlp.2018.11.003. 12075-7_34.
[35] Lei Y, Liu J, Ni J, Lee J. Production line simulation using STPN for maintenance [63] M. Mulders and M. Haarman, “Predictive maintenance 4.0, predict the
scheduling. J Intell Manuf 2010;21(2):213–21. https://doi.org/10.1007/s10845- unpredictable,” 2017.
008-0194-1. [64] K.L. Lueth, C. Patsioura, Z.D. Williams, and Z.Z. Kermani, “Industrial analytics
[36] Eisenberger D, Fink O. Assessment of maintenance strategies for railway vehicles 2016/2017: the current state of data analytics usage in industrial companies,”
using Petri-nets. Transp Res Procedia 2017;27:205–14. https://doi.org/10.1016/ 2016.
j.trpro.2017.12.012. [65] Zhao X, Lv Z, He Z, Wang W. Reliability and opportunistic maintenance for a
[37] Liu B, Wu S, Xie M, Kuo W. A condition-based maintenance policy for degrading series system with multi-stage accelerated damage in shock environments.
systems with age- and state-dependent operating cost. Eur J Oper Res 2017;263 Comput Ind Eng 2019;137(August 2019):106029. https://doi.org/10.1016/j.
(3):879–87. https://doi.org/10.1016/j.ejor.2017.05.006. cie.2019.106029.
[38] de Angelis M, Patelli E, Beer M. Forced Monte Carlo simulation strategy for the [66] Bismut E, Pandey MD, Straub D. Reliability-based inspection and maintenance
design of maintenance plans with multiple inspections. ASCE-ASME J Risk planning of a nuclear feeder piping system. Reliab Eng Syst Saf 2022;224(August
Uncertain Eng Syst Part A Civ Eng 2017;3(2):1–9. https://doi.org/10.1061/ 2021):108521. https://doi.org/10.1016/j.ress.2022.108521.
ajrua6.0000868. [67] Li X, Ran Y, Zhang G. Optimization of equal-cycle maintenance strategy
[39] Turban E, Aronson JE, Liang T-P. Decision support systems and intelligent considering imperfect preventive maintenance. Proc Inst Mech Eng Part E J
systems. 7th Edition. Prentice Hall; 2005. Process Mech Eng 2022;236(4):1392–402. https://doi.org/10.1177/
[40] Sherwin D. A review of overall models for maintenance management. J Qual 09544089211063852.
Maint Eng 2000;6(3):138–64. https://doi.org/10.1108/13552510010341171. [68] Boyd S, Vandenberghe L. Convex optimization. Convex Optim Signal Process
[41] Mong SG, Mohamed SF, Misnan MS. Maintenance management model: an Commun 2004.
identification of key elements for value-based maintenance management by local [69] Zio E. Some challenges and opportunities in reliability engineering. IEEE Trans
authority. Int J Eng Technol 2018;7(3):35–43. https://doi.org/10.14419/ijet. Reliab 2016;PP(99):1769–82. https://doi.org/10.1109/TR.2016.2591504. M4 -
v7i3.25.17467. Citavi.
[42] Manzini R, Regattieri A, Pham H, Ferrari E. Maintenance for industrial System. [70] Ding S-H, Kamaruddin S. Selection of optimal maintenance policy by using fuzzy
London: Springer s. Springer-Verlag; 2010. multi criteria decision making method. In: Proceedings of the 2012 International
[43] Velmurugan RS, Dhingra T. Maintenance strategy selection and its impact in Conference on Industrial Engineering and Operations Management; 2012.
maintenance function: a conceptual framework. Int J Oper Prod Manag 2015;35 p. 435–43.
(12):1622–61. https://doi.org/10.1108/IJOPM-01-2014-0028. [71] Wang J, Zhu X, Yuan T. Cost-minimization preventive maintenance for the data
[44] Bevilacqua M, Braglia M. Analytic hierarchy process applied to maintenance storage system of a supercomputer. In: Proc. - 12th Int. Conf. Reliab. Maint.
strategy selection. Reliab Eng Syst Saf 2000;70(1):71–83. https://doi.org/ Safety, ICRMS 2018; 2018. p. 448–51. https://doi.org/10.1109/
10.1016/S0951-8320(00)00047-8. ICRMS.2018.00089.
[45] Erkoyuncu JA, Khan S, Eiroa AL, Butler N, Rushton K, Brocklebank S. [72] Carnero MC, Gómez A. Optimisation of maintenance in delivery systems for
Perspectives on trading cost and availability for corrective maintenance at the cytostatic medicines. BMC Health Serv Res 2021;21(1):1–26. https://doi.org/
equipment type level. Reliab Eng Syst Saf 2017;168(May):53–69. https://doi.org/ 10.1186/s12913-021-07093-w.
10.1016/j.ress.2017.05.041. [73] Liu Y, Chen Y, Jiang T. Dynamic selective maintenance optimization for multi-
[46] Zio E, Compare M. Evaluating maintenance policies by quantitative modeling and state systems over a finite horizon: a deep reinforcement learning approach. Eur J
analysis. Reliab Eng Syst Saf 2013;109:53–65. https://doi.org/10.1016/j. Oper Res 2020;283(1):166–81. https://doi.org/10.1016/j.ejor.2019.10.049.
ress.2012.08.002. [74] Lin YH, Li YF, Zio E. A framework for modeling and optimizing maintenance in
[47] Li G, et al. Development of a preventive maintenance strategy for an automatic systems considering epistemic uncertainty and degradation dependence based on
production line based on group maintenance method. Appl Sci 2018;8(10). PDMPs. IEEE Trans Ind Informatics 2018;14(1):210–20. https://doi.org/
https://doi.org/10.3390/app8101781. 10.1109/TII.2017.2743820.
[48] De Jonge B, Klingenberg W, Teunter R, Tinga T. Optimum maintenance strategy [75] Compare M, Baraldi P, Bani I, Zio E, Donnell DM. Development of a Bayesian
under uncertainty in the lifetime distribution. Reliab Eng Syst Saf 2015;133: multi-state degradation model for up-to-date reliability estimations of working
59–67. https://doi.org/10.1016/j.ress.2014.09.013. industrial components. Reliab Eng Syst Saf 2017;166(December 2016):25–40.
[49] Shafiee M, Sørensen JD. Maintenance optimization and inspection planning of https://doi.org/10.1016/j.ress.2016.11.020.
wind energy assets: models, methods and strategies. Reliab Eng Syst Saf 2019;192 [76] Hennequin S, Arango G, Rezg N. Optimization of imperfect maintenance based on
(November 2017). https://doi.org/10.1016/j.ress.2017.10.025. fuzzy logic for a single-stage single-product production system. J Qual Maint Eng
[50] Prajapati A, Bechtel J, Ganesan S. Condition based maintenance: a survey. J Qual 2009;15(4). 412–249.
Maint Eng 2012;18(4):384–400. https://doi.org/10.1108/13552511211281552. [77] Yin S, Ding SX, Zhou D. Diagnosis and prognosis for complicated industrial
[51] Abid A, Khan MT, Iqbal J. A review on fault detection and diagnosis techniques: systems - part I. IEEE Trans Ind Electron 2016;63(4):2501–5. https://doi.org/
basics and beyond. Artif Intell Rev 2021;54(5):3639–64. https://doi.org/ 10.1109/TIE.2016.2522944.
10.1007/s10462-020-09934-2. [78] Zhang C, Qi F, Zhang N, Li Y, Huang H. Maintenance policy optimization for
[52] Chiang LH, Russell EL, Braatz RD. Fault detection and diagnosis in industrial multi-component systems considering dynamic importance of components.
systems. London: Springer Science & Business Media; 2000. Reliab Eng Syst Saf 2022;226(January):108705. https://doi.org/10.1016/j.
[53] Pinciroli L, Baraldi P, Shokry A, Zio E, Seraoui R, Mai C. A semi-supervised ress.2022.108705.
method for the characterization of degradation of nuclear power plants steam [79] Wang J, Zhao X, Guo X. Optimizing wind turbine’s maintenance policies under
generators. Prog Nucl Energy 2021;131(December 2020):103580. https://doi. performance-based contract. Renew Energy 2019;135:626–34. https://doi.org/
org/10.1016/j.pnucene.2020.103580. 10.1016/j.renene.2018.12.006.
[54] Hu Y, Miao X, Si Y, Pan E, Zio E. Prognostics and health management: a review [80] Shi Y, Zhu W, Xiang Y, Feng Q. Condition-based maintenance optimization for
from the perspectives of design, development and decision. Reliab Eng Syst Saf multi-component systems subject to a system reliability requirement. Reliab Eng
2022;217(May 2021):108063. https://doi.org/10.1016/j.ress.2021.108063. Syst Saf 2020;202(May):107042. https://doi.org/10.1016/j.ress.2020.107042.
[55] Biggio L, Kastanis I. Prognostics and health management of industrial assets: [81] Pinciroli L, Baraldi P, Ballabio G, Compare M, Zio E. Deep reinforcement learning
current progress and road ahead. Front Artif Intell 2020;3(November):1–24. based on proximal policy optimization for the maintenance of a wind farm with
https://doi.org/10.3389/frai.2020.578613.

15
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

multiple crews. Energies 2021;14(20):1–17. https://doi.org/10.3390/ [106] Briš R. Parallel simulation algorithm for maintenance optimization based on
en14206743. directed Acyclic Graph. Reliab Eng Syst Saf 2008;93(6):874–84. https://doi.org/
[82] Baraldi P, Compare M, Zio E. Maintenance policy performance assessment in 10.1016/j.ress.2007.03.036.
presence of imprecision based on Dempster-Shafer Theory of Evidence. Inf Sci [107] Sachan S, Zhou C. Probabilistic dynamic programming algorithm: a solution for
(Ny). 2013;245:112–31. https://doi.org/10.1016/j.ins.2012.11.003. optimal maintenance policy for power cables. Life Cycle Reliab Saf Eng 2019;8
[83] Marseguerra M, Zio E, Podofillini L. Condition-based maintenance optimization (2):117–27. https://doi.org/10.1007/s41872-019-00074-3.
by means of genetic algorithms and Monte Carlo simulation. Reliab Eng Syst Saf [108] Haladuick S, Dann MR. Genetic algorithm for inspection and maintenance
2002;77(2):151–65. https://doi.org/10.1016/S0951-8320(02)00043-1. planning of deteriorating structural systems: application to pressure vessels.
[84] Yeardley AS, Ejeh JO, Allen L, Brown SF, Cordiner J. Integrating machine Infrastructures 2018;3(3). https://doi.org/10.3390/infrastructures3030032.
learning techniques into optimal maintenance scheduling. Comput Chem Eng [109] Li Z, Guo J, Zhou R. Maintenance scheduling optimization based on reliability and
2022;166(August):107958. https://doi.org/10.1016/j. prognostics information. In: Proc. - Annu. Reliab. Maintainab. Symp. 2016-April;
compchemeng.2022.107958. 2016. https://doi.org/10.1109/RAMS.2016.7448069.
[85] Pinciroli L, Baraldi P, Ballabio G, Compare M, Zio E. Optimization of the [110] Neto WF, Cavalcante C, Do P. Deep reinforcement learning-based maintenance
operation and maintenance of renewable energy systems by deep reinforcement decision-making for a steel production line. In: Proceedings of the 31st European
learning. Renew Energy 2021;183:752–63. https://doi.org/10.1016/j. Safety and Reliability Conference; 2021. p. 2611–8. https://doi.org/10.3850/
renene.2021.11.052. 978-981-18-2016-8_600-cd.
[86] Hu Y, Miao X, Zhang J, Liu J, Pan E. Reinforcement learning-driven maintenance [111] Wang W. A stochastic model for joint spare parts inventory and planned
strategy: a novel solution for long-term aircraft maintenance decision maintenance optimisation. Eur J Oper Res 2012;216(1):127–39. https://doi.org/
optimization. Comput Ind Eng 2021;153. https://doi.org/10.1016/j. 10.1016/j.ejor.2011.07.031.
cie.2020.107056. [112] Azadeh A, Abdolhossein Zadeh S. An integrated fuzzy analytic hierarchy process
[87] Andriotis CP, Papakonstantinou KG. Managing engineering systems with large and fuzzy multiple-criteria decision-making simulation approach for maintenance
state and action spaces through deep reinforcement learning. Reliab Eng Syst Saf policy selection. Simulation 2016;92(1):3–18. https://doi.org/10.1177/
2019;191(November 2018):106483. https://doi.org/10.1016/j. 0037549715616686.
ress.2019.04.036. [113] Chan FTS, Prakash A. Maintenance policy selection in manufacturing firms using
[88] Javanmard H, al W. Koraeizadeh A. Optimizing the preventive maintenance the fuzzy MCDM approach. Int J Prod Res 2012;50(23):70044–7056. https://doi.
scheduling by genetic algorithm based on cost and reliability in National Iranian org/10.1080/00207543.2011.653451.
Drilling Company. J Ind Eng Int 2016;12(4):509–16. https://doi.org/10.1007/ [114] Labib A, Yuniarto MN. Maintenance strategies for changeable manufacturing.
s40092-016-0155-9. Changeable and reconfigurable manufacturing systems. London: Springer; 2009.
[89] Besnard F, Patriksson M, Strömberg AB, Wojciechowski A, Bertling L. An p. 327–51. Springer S.
optimization framework for opportunistic maintenance of offshore wind power [115] Waeyenbergh G, Pintelon L. Maintenance concept development : a case study. Int
system. In: 2009 IEEE Bucharest PowerTech Innov. Ideas Towar. Electr. Grid J Prod Econ 2004;89:395–405. https://doi.org/10.1016/j.ijpe.2003.09.008.
Futur.; 2009. p. 1–7. https://doi.org/10.1109/PTC.2009.5281868. [116] Assid M, Gharbi A, Hajji A. Joint production, setup and preventive maintenance
[90] Kozanidis G, Gavranis A, Kostarelou E. Mixed integer least squares optimization policies of unreliable two-product manufacturing systems. Int J Prod Res 2015;53
for flight and maintenance planning of mission aircraft. Nav Res Logist 2012;59 (15):4668–83. https://doi.org/10.1080/00207543.2015.1030468.
(3–4):212–29. https://doi.org/10.1002/nav.21483. [117] Roux O, Jamali MA, Kadi DA, Châtelet E. Development of simulation and
[91] Blanco MA, Marti-Puig P, Gibert K, Cusidó J, Solé-Casals J. A text-mining optimization platform to analyse maintenance policies performances for
approach to assess the failure condition of wind turbines using maintenance manufacturing systems. Int J Comput Integr Manuf 2008;21(4):407–14. https://
service history. Energies 2019;12(10). https://doi.org/10.3390/en12101982. doi.org/10.1080/09511920701574214.
[92] Li Y, Xi Gu J, Zhen D, Xu M, Ball A. An evaluation of gearbox condition [118] Linnéusson G, Ng AHC, Aslam T. Quantitative analysis of a conceptual system
monitoring using infrared thermal images applied with convolutional neural dynamics maintenance performance model using multi-objective optimisation.
networks. Sensors (Switzerland) 2019;19(9). https://doi.org/10.3390/ J Simul 2018;12(2):171–89. https://doi.org/10.1080/17477778.2018.1467849.
s19092205. [119] Loganathan MK, Gandhi OP. Maintenance cost minimization of manufacturing
[93] Compare M, Baraldi P, Zio E. Challenges to IoT-enabled predictive maintenance systems using PSO under reliability constraint. Int J Syst Assur Eng Manag 2016;7
for industry 4.0. IEEE Internet Things J 2020;7(5):4585–97. https://doi.org/ (1):47–61. https://doi.org/10.1007/s13198-015-0374-2.
10.1109/JIOT.2019.2957029. [120] Yulan J, Zuhua J, Wenrui H. Multi-objective integrated optimization research on
[94] Zhou F, Zhu X, Chen H. Research on optimization of equipment maintenance plan preventive maintenance planning and production scheduling for a single
based on entropy and TOPSIS. In: Proceedings of the 2011 International machine. Int J Adv Manuf Technol 2008;39(9-10):954–64. https://doi.org/
Conference on Informatics, Cybernetics, and Computer Engineering (ICCE2011) 10.1007/s00170-007-1268-5.
November 19-20; 2011. p. 145–50. [121] Oke SA. An analytical model for the optimisation of maintenance profitability. Int
[95] Bellani L, Compare M, Baraldi P, Zio E. Towards developing a novel framework J Product Perform Manag 2005;54(2):113–36. https://doi.org/10.1108/
for practical PHM: a sequential decision problem solved by reinforcement 17410400510576612.
learning and artificial neural networks. Int J Progn Heal Manag 2020;31:1–15. [122] Kianfar F. A numerical method to approximate optimal production and
[96] Ma J, Cheng L, Li D. Road maintenance optimization model based on dynamic maintenance plan in a flexible manufacturing system. Appl Math Comput 2005;
programming in urban traffic network. J Adv Transp 2018;2018:11. https://doi. 170(2):924–40. https://doi.org/10.1016/j.amc.2004.12.030.
org/10.1155/2018/4539324. [123] Jin X, Li L, Ni J. Option model for joint production and preventive maintenance
[97] Caballé NC, Castro IT, Pérez CJ, Lanza-Gutiérrez JM. A condition-based system. Int J Prod Econ 2009;119(2):347–53. https://doi.org/10.1016/j.
maintenance of a dependent degradation-threshold-shock model in a system with ijpe.2009.03.005.
multiple degradation processes. Reliab Eng Syst Saf 2015;134:98–109. https:// [124] Aghezzaf EH, Khatab A, Tam PLe. Optimizing production and imperfect
doi.org/10.1016/j.ress.2014.09.024. preventive maintenance planning’s integration in failure-prone manufacturing
[98] Li C, Xu M. ELECTRE III based on ranking fuzzy numbers for deterministic and systems. Reliab Eng Syst Saf 2016;145:190–8. https://doi.org/10.1016/j.
fuzzy maintenance strategy decision problems. In: International Conference on ress.2015.09.017.
Automation and Logistics; 2007. p. 309–12. [125] Goti A, Oyarbide-Zubillaga A, Sanchez A, Akyazi T, Alberdi E. Multi equipment
[99] Rezg N, Chelbi A, Xie X. Modeling and optimizing a joint inventory control and condition based maintenance optimization using multi-objective evolutionary
preventive maintenance strategy for a randomly failing production unit: algorithms. Appl Sci 2019;9(22):1–11. https://doi.org/10.3390/app9224849.
analytical and simulation approaches. Int J Comput Integr Manuf 2005;18(2–3): [126] Franty YK, Handoko B. The determination of preventive maintenance using
225–35. https://doi.org/10.1080/0951192052000288152. simulated annealing algorithm based on weighted fitness function. J Tek Ind
[100] Nielsen JS, Sørensen JD. Methods for risk-based planning of O&M of wind 2019;20(1):53–61.
turbines. Energies 2014;7(10):6645–64. https://doi.org/10.3390/en7106645. [127] Kuhnle A, Jakubik J, Lanza G. Reinforcement learning for opportunistic
[101] Bajestani MA, Banjevic D. Calendar-based age replacement policy with dependent maintenance optimization. Prod Eng 2019;13(1):33–41. https://doi.org/
renewal cycles. IIE Trans (Institute Ind Eng 2016;48(11):1016–26. https://doi. 10.1007/s11740-018-0855-7.
org/10.1080/0740817X.2016.1163444. [128] Tran LV, Huynh BH, Akhtar H. Ant colony optimization algorithm for
[102] Chandola V, Banerjee A, Kumar V. Anomaly detection : a survey. ACM Comput maintenance, repair and overhaul scheduling optimization in the context of
Surv 2009;41(3):1–72. industrie 4.0. Appl Sci 2019;9(22):4815.
[103] De Carlo F, Arleo MA. Maintenance cost optimization in condition based [129] Ilgin MA, Tunali S. Joint optimization of spare parts inventory and maintenance
maintenance: a case study for critical facilities. Int J Eng Technol 2013;5(5): policies using genetic algorithms. Int J Adv Manuf Technol 2007;34(5–6):
4296–302. 594–604. https://doi.org/10.1007/s00170-006-0618-z.
[104] Robelin CA, Madanat SM. Dynamic programming based maintenance and [130] Gassner S. Deriving maintenance strategies for cooperative alliances – a Value
replacement optimization for bridge decks using history-dependent deterioration Chain approach. In: Managing Operations in Service Economies International
models. In: Proceedings of the Ninth International Conference on Applications of Conference Proceeding; 2010. p. 1–10.
Advanced Technology in Transportation. 40799; 2006. p. 13–8. https://doi.org/ [131] Santos FP, Teixeira ÂP, Soares CG. Modeling, simulation and optimization of
10.1061/40799(213)3. maintenance cost aspects on multi-unit systems by stochastic Petri nets with
[105] Langseth H, Haugen K, Sandtorv H. Analysis of OREDA data for maintenance predicates. Simulation 2019;95(5):461–78. https://doi.org/10.1177/
optimisation. Reliab Eng Syst Saf 1998;60(2):103–10. https://doi.org/10.1016/ 0037549718782655.
S0951-8320(98)83003-2. [132] Wang Y, Deng Q. Optimization of maintenance scheme for offshore wind turbines
considering time windows based on hybrid ant colony algorithm. Ocean Eng
2022;263(August):112357. https://doi.org/10.1016/j.oceaneng.2022.112357.

16
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

[133] Pinciroli L, Baraldi P, Ballabio G, Compare M, Zio E. Deep Reinforcement learning [157] Morcous G, Lounis Z. Maintenance optimization of infrastructure networks using
for optimal operation and maintenance of energy systems equipped with PHM genetic algorithms. Autom Constr 2005;14(1):129–42. https://doi.org/10.1016/j.
capabilities. In: 30th European Safety and Reliability Conference and the 15th autcon.2004.08.014.
Probabilistic Safety Assessment and Management Conference; 2020. [158] García-Segura T, Yepes V, Frangopol DM, Yang DY. Lifetime reliability-based
[134] Fallahi F, Bakir I, Yildirim M, Ye Z. A chance-constrained optimization framework optimization of post-tensioned box-girder bridges. Eng Struct 2017;145:381–91.
for wind farms to manage fleet-level availability in condition based maintenance https://doi.org/10.1016/j.engstruct.2017.05.013.
and operations. Renew Sustain Energy Rev 2022;168(August). https://doi.org/ [159] Mao X, Jiang X, Yuan C, Zhou J. Modeling the optimal maintenance scheduling
10.1016/j.rser.2022.112789. strategy for bridge networks. Appl Sci 2020;10(2). https://doi.org/10.3390/
[135] Irawan CA, Ouelhadj D, Jones D, Stålhane M, Sperstad IB. Optimisation of app10020498.
maintenance routing and scheduling for offshore wind farms. Eur J Oper Res [160] van den Boomen M, van den Berg PL, Wolfert ARM. A dynamic programming
2017;256(1):76–89. https://doi.org/10.1016/j.ejor.2016.05.059. approach for economic optimisation of lifetime-extending maintenance,
[136] Schouten TN, Dekker R, Hekimoğlu M, Eruguz AS. Maintenance optimization for renovation, and replacement of public infrastructure assets under differential
a single wind turbine component under time-varying costs. Eur J Oper Res 2022; inflation. Struct Infrastruct Eng 2019;15(2):193–205. https://doi.org/10.1080/
300(3):979–91. https://doi.org/10.1016/j.ejor.2021.09.004. 15732479.2018.1504803.
[137] Carlos S, Sánchez A, Martorell S, Marton I. Onshore wind farms maintenance [161] Ahmadi A, Gupta S, Karim R, Kumar U. Selection of maintenance strategy for
optimization using a stochastic model. Math Comput Model 2013;57(7–8): aircraft systems using multi-criteria decision making methodologies. Int J Reliab
1884–90. https://doi.org/10.1016/j.mcm.2011.12.025. Qual Saf Eng 2010;17(3):223–43. https://doi.org/10.1142/
[138] Izquierdo J, Erguido A, Zubizarreta PX, Uribetxebarria J. Framework for S0218539310003779.
managing the operations and maintenance of wind farms. J Phys Conf Ser 2019; [162] J. Khalil, S.M. Saad, N. Gindy, and K. Mackechnie, A maintenance policy selection
1222(1). https://doi.org/10.1088/1742-6596/1222/1/012046. tool for industrial machine parts., vol. 159. 2005.
[139] Atashgar K, Abdollahzadeh H. Reliability optimization of wind farms considering [163] Li Z, Guo J, Zhou R. Maintenance scheduling optimization based on reliability and
redundancy and opportunistic maintenance strategy. Energy Convers Manag prognostics information. In: Proc. - Annu. Reliab. Maintainab. Symp. 2016-April;
2016;112:445–58. https://doi.org/10.1016/j.enconman.2016.01.027. 2016. https://doi.org/10.1109/RAMS.2016.7448069.
[140] Yang L, Peng R, Li G, Lee CG. Operations management of wind farms integrating [164] Deng Q, Santos BF, Curran R. A practical dynamic programming based
multiple impacts of wind conditions and resource constraints. Energy Convers methodology for aircraft maintenance check scheduling optimization. Eur J Oper
Manag 2020;205(September 2019):112162. https://doi.org/10.1016/j. Res 2020;281(2):256–73. https://doi.org/10.1016/j.ejor.2019.08.025.
enconman.2019.112162. [165] Ghorbani M, Nourelfath M, Gendreau M. A two-stage stochastic programming
[141] Zhao X, Huang X, Sun J. Reliability modeling and maintenance optimization for model for selective maintenance optimization. Reliab Eng Syst Saf 2022;223
the two-unit system with preset self-repairing mechanism. Proc Inst Mech Eng (March):108480. https://doi.org/10.1016/j.ress.2022.108480.
Part O J Risk Reliab 2020;234(2):221–34. https://doi.org/10.1177/ [166] Wang L, Chu J, Wu J. Selection of optimum maintenance strategies based on a
1748006X19890739. fuzzy analytic hierarchy process. Int J Prod Econ 2007;107(1):151–63. https://
[142] Li M, Jiang X, Carroll J, Negenborn RR. A multi-objective maintenance strategy doi.org/10.1016/j.ijpe.2006.08.005.
optimization framework for offshore wind farms considering uncertainty. Appl [167] Tan Z, Li J, Wu Z, Zheng J, He W. An evaluation of maintenance strategy using
Energy 2022;321(January):119284. https://doi.org/10.1016/j. risk based inspection. Saf Sci 2011;49(6):852–60. https://doi.org/10.1016/j.
apenergy.2022.119284. ssci.2011.01.015.
[143] Aafif Y, Chelbi A, Mifdal L, Dellagi S, Majdouline I. Optimal preventive [168] Fan L, et al. A systematic method for the optimization of gas supply reliability in
maintenance strategies for a wind turbine gearbox. Energy Rep 2022;8(May): natural gas pipeline network based on Bayesian networks and deep reinforcement
803–14. https://doi.org/10.1016/j.egyr.2022.07.084. learning. Reliab Eng Syst Saf 2022;225(December 2021):108613. https://doi.org/
[144] Kovács A, Erds G, Viharos ZJ, Monostori L. A system for the detailed scheduling of 10.1016/j.ress.2022.108613.
wind farm maintenance. CIRP Ann - Manuf Technol 2011;60(1):497–501. [169] Liu B, Yeh RH, Xie M, Kuo W. Maintenance scheduling for multicomponent
https://doi.org/10.1016/j.cirp.2011.03.049. systems with hidden failures. IEEE Trans Reliab 2017;66(4):1280–92. https://doi.
[145] Carazas FG, Souza GFM. Risk-based decision making method for maintenance org/10.1109/TR.2017.2740562.
policy selection of thermal power plant equipment. Energy 2010;35(2):964–75. [170] Labib AW. World-class maintenance using a computerised maintenance
https://doi.org/10.1016/j.energy.2009.06.054. management system. J Qual Maint Eng 1998;4(1):66–75. https://doi.org/
[146] Zhao Y, Smidts C. Reinforcement learning for adaptive maintenance policy 10.1108/13552519810207470.
optimization under imperfect knowledge of the system degradation model and [171] Liu B, Lin J, Zhang L, Kumar U. A dynamic prescriptive maintenance model
partial observability of system states. Reliab Eng Syst Saf 2022;224(April). considering system aging and degradation. IEEE Access 2019;7:94941–3. https://
https://doi.org/10.1016/j.ress.2022.108541. doi.org/10.1109/ACCESS.2019.2928587.
[147] Compare M, Bellani L, Cobelli E, Zio E. Reinforcement learning-based flow [172] Consilvio A, et al. Prescriptive maintenance of railway infrastructure: From data
management of gas turbine parts under stochastic failures. Int J Adv Manuf analytics to decision support. In: MT-ITS 2019 - 6th Int. Conf. Model. Technol.
Technol Springer Verlag 2018;99(9–12):2981–92. Intell. Transp. Syst.; 2019. p. 1–10. https://doi.org/10.1109/
[148] Hao Z, Di Maio F, Pinciroli L, Zio E. Optimal prescriptive maintenance of nuclear MTITS.2019.8883331.
power plants by deep reinforcement learning. In: Proceedings of the 32nd [173] Hani Y, Amodeo L, Yalaoui F, Chen H. Simulation based optimization of a train
European Safety and Reliability Conference (ESREL 2022); 2022. p. 689–97. maintenance facility. J Intell Manuf 2008;19(3):293–300. https://doi.org/
https://doi.org/10.3850/978-981-18-5183-4. 10.1007/s10845-008-0082-8.
[149] Mancuso A, Compare M, Salo A, Zio E. Optimal prognostics and health [174] Afrinaldi F, Taufik AMT, Zhang HC, Hasan A. Minimizing economic and
management-driven inspection and maintenance strategies for industrial systems. environmental impacts through an optimal preventive replacement schedule:
Reliab Eng Syst Saf 2021;210(January):107536. https://doi.org/10.1016/j. model and application. J Clean Prod 2017;143:882–93. https://doi.org/10.1016/
ress.2021.107536. j.jclepro.2016.12.033.
[150] Martorell S, et al. RAMS+C informed decision-making with application to multi- [175] Hemmati N, Rahiminezhad Galankashi M, Imani DM, Farughi H. Maintenance
objective optimization of technical specifications and maintenance using genetic policy selection: a fuzzy-ANP approach. J Manuf Technol Manag 2018;29(7):
algorithms. Reliab Eng Syst Saf 2005;87(1):65–75. https://doi.org/10.1016/j. 1253–68. https://doi.org/10.1108/JMTM-06-2017-0109.
ress.2004.04.009. [176] Leo E, Engell S. Condition-based maintenance optimization via stochastic
[151] Compare M, Martini F, Zio E. Genetic algorithms for condition-based maintenance programming with endogenous uncertainty. Comput Chem Eng 2022;156:
optimization under uncertainty. Eur J Oper Res 2015;244(2):611–23. https://doi. 107550. https://doi.org/10.1016/j.compchemeng.2021.107550.
org/10.1016/j.ejor.2015.01.057. [177] H.C. Vu, P. Do, and A. Barros, “Mean residual life and the Birnbaum importance
[152] Rocchetta R, Compare M, Patelli E, Zio E. A reinforcement learning framework for measure for complex structures,” vol. 65, no. 1, pp. 217–234, 2016.
optimisation of power grid operations and maintenance. In: 8th International [178] Ding S, Goh T-T, Tan P, Wee S. Implementation of decision tree for maintenance
Workshop on Reliable Computing, “Computing with Confidence; 2018. policy decision making- a case study in semiconductor industry. Adv Mater Res
p. 223–32. https://doi.org/10.1016/j.apenergy.2019.03.027. 2012;593:704–7. https://doi.org/10.4028/www.scientific.net/AMR.591-
[153] Rocchetta R, Bellani L, Compare M, Zio E, Patelli E. A reinforcement learning 593.704.
framework for optimal operation and maintenance of power grids. Appl Energy [179] Richardson S, Kefford A, Hodkiewicz M. Optimised asset replacement strategy in
2019;(May):291–301. https://doi.org/10.1016/j.apenergy.2019.03.027. the presence of lead time uncertainty. Int J Prod Econ 2013;141(2):659–67.
[154] Dehghani NL, Darestani YM, Shafieezadeh A. Optimal life-cycle resilience https://doi.org/10.1016/j.ijpe.2012.10.006.
enhancement of aging power distribution systems: A MINLP-Based preventive [180] Barde SRA, Yacout S, Shin H. Optimal preventive maintenance policy based on
maintenance planning. IEEE Access 2020;8:22324–34. https://doi.org/10.1109/ reinforcement learning of a fleet of military trucks. J Intell Manuf 2016. https://
ACCESS.2020.2969997. doi.org/10.1007/s10845-016-1237-7.
[155] Matin SAA, Mansouri SA, Bayat M, Jordehi AR, Radmehr P. A multi-objective bi- [181] Trojan F, Morais DC. Prioritising alternatives for maintenance of water
level optimization framework for dynamic maintenance planning of active distribution networks : a group decision approach. Water SA 2012;38(4):555–64.
distribution networks in the presence of energy storage systems. J Energy Storage [182] Van Horenbeek A, Pintelon L, Muchiri P. Maintenance optimization models and
2022;52(PA):104762. https://doi.org/10.1016/j.est.2022.104762. criteria. Int J Syst Assur Eng Manag 2010;1(3):189–200. https://doi.org/
[156] Wang X, Zhao X, Wang S, Sun L. Reliability and maintenance for performance- 10.1007/s13198-011-0045-x.
balanced systems operating in a shock environment. Reliab Eng Syst Saf 2020;195 [183] Laggoune R, Mokhtar WA, Kheloufi K. Preventive maintenance optimization
(January 2019). https://doi.org/10.1016/j.ress.2019.106705. based on both cost and availability measures. A case study. In: ESReDA Conf.
2011 Adv. Reliab. Maint. Policies; 2011.

17
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

[184] Bucher C, Frangopol DM. Optimization of lifetime maintenance strategies for for solving unit commitment. Int J Electr Power Energy Syst 2020;125(April
deteriorating structures considering probabilities of violating safety, condition, 2020):106426. https://doi.org/10.1016/j.ijepes.2020.106426.
and cost thresholds. Probab Eng Mech 2006;21(1):1–8. https://doi.org/10.1016/ [214] Bellman RE. Dynamic programming. Science (80-.) 1966;153(3731):34–7.
j.probengmech.2005.06.002. [215] Sutton RS, Barto A. Reinforcement Learning: An Introduction. Cambridge, MA:
[185] Quan G, Greenwood GW, Liu D, Hu S. Searching for multiobjective preventive MIT Press; 2018.
maintenance schedules: combining preferences with evolutionary algorithms. Eur [216] Wang FS, Chen LH. Heuristic Optimization, Encycloped. New York: Springer; 2013.
J Oper Res 2007;177(3):1969–84. https://doi.org/10.1016/j.ejor.2005.12.015. [217] Nesmachnow S. An overview of metaheuristics: accurate and efficient methods
[186] Ni J, Jin X. Decision support systems for effective maintenance operations. CIRP for optimisation. Int J Metaheuristics 2014;3(4):320. https://doi.org/10.1504/
Ann - Manuf Technol 2012;61(1):411–4. https://doi.org/10.1016/j. ijmheur.2014.068914.
cirp.2012.03.065. [218] Mitchell M. Introduction to genetic algorithms. Cambridge, MA: MIT Press.; 1996.
[187] Van Horenbeek A, Kellens K, Pintelon L, Duflou JR. Economic and environmental ISBN 9780585030944.
aware maintenance optimization. Procedia CIRP 2014;15:343–8. https://doi.org/ [219] Giuggioli Busacca P, Marseguerra M, Zio E. Multiobjective optimization by
10.1016/j.procir.2014.06.048. genetic algorithms: application to safety systems. Reliab Eng Syst Saf 2001;72:
[188] Kazemi S. Proposing a green maintenance model in order to analyses the effects of 59–74.
influential criteria on the environment and green maintenance index, using [220] Nelder JA, Mead R. A simplex method for function minimization. Comput J 1965;
system dynamics method. Adv Environ Biol 2013;7(11):3529–34. 7(4):308–13. https://doi.org/10.1093/comjnl/7.4.308.
[189] J.P. Liyanage, F. Badurdeen, and R.M.C. Ratnayake, “Industrial asset maintenance [221] Advances i. Askarzadeh A, Rashedi E. Harmony search algorithm: basic concepts
and sustainability performance: Economical, environmental, and societal and engineering applications. In: Patnaik S, editor. Recent Developments in
implications,” in Handbook of Maintenance Management and Engineering, 2009, pp. Intelligent Nature-Inspired Computing. Hershey, PA 17033, USA: IGI GLobal;
665–693. 2017. p. 1–36. Advances i.
[190] Lee J, Holgado M, Kao HA, Macchi M. New thinking paradigm for maintenance [222] Wang X, Guo S, Shen J, Liu Y. Optimization of preventive maintenance for series
innovation design. In: Proceedings of the 19th World Congress The International manufacturing system by differential evolution algorithm. J Intell Manuf 2020;31
Federation of Automatic Control. 19; 2014. p. 7104–9. https://doi.org/10.3182/ (3):745–57. https://doi.org/10.1007/s10845-019-01475-y.
20140824-6-za-1003.02519. [223] Kennedy J, Eberhart R. Particle swarm optimization. In: ICNN’95- International
[191] Franciosi C, Iung B, Miranda S, Riemma S. Maintenance for Sustainability in the Conference on Neural Networks; 1995. p. 1942–8.
Industry 4.0 context: a Scoping Literature Review. IFAC-PapersOnLine 2018;51 [224] van Laarhoven PJM, Aarts EHL. Simulated Annealing: Theory and Applications.
(11):903–8. https://doi.org/10.1016/j.ifacol.2018.08.459. Dordrecht: Springer, Dordrecht; 1987. Mathematic.
[192] Youn BD, Hu C, Wang P. Resilience-driven system design of complex engineered [225] Dorigo M, Birattari M, Stützle T. Ant colony optimization. IEEE Comput Intell
systems. J Mech Des 2011;133(10). Mag 2006;1(4):28–39. https://doi.org/10.1007/978-3-319-93025-1_3.
[193] Rauch E. Back to the future: Emerging trends for long-term resilience in [226] Karaboga D, Akay B. A comparative study of Artificial Bee Colony algorithm. Appl
manufacturing. World Munufacturing Foundation; 2021. Math Comput 2009;214(1):108–32. https://doi.org/10.1016/j.amc.2009.03.090.
[194] Certa A, Enea M, Lupo T. ELECTRE III to dynamically support the decision maker [227] Yang L, Li G, Zhang Z, Ma X. Operations & maintenance optimization of wind
about the periodic replacements configurations for a multi-component system. turbines integrating wind and aging information. IEEE Trans Sustain Energy
Decis Support Syst 2013;55(1):126–34. https://doi.org/10.1016/j. 2020;3029(c). https://doi.org/10.1109/tste.2020.2986586. 1–1.
dss.2012.12.044. [228] Zio E, Sansavini G. Vulnerability of smart grids with variable generation and
[195] Vu HC, Do P, Barros A, Bérenguer C. Maintenance grouping strategy for multi- consumption: a system of systems perspective. IEEE Trans Syst Man, Cybern Part
component systems with dynamic contexts. Reliab Eng Syst Saf 2014;132:233–49. ASystems Humans 2013;43(3):477–87. https://doi.org/10.1109/
https://doi.org/10.1016/j.ress.2014.08.002. TSMCA.2012.2207106.
[196] Huang JY. New search algorithm for solving the maintenance scheduling problem [229] Kröger W, Zio E. Vulnerable Systems. 1st ed. London: Springer; 2011.
for a family of machines. Optim Methods Softw 2006;21(3):461–77. https://doi. [230] Bakon K, Holczinger T, Sule Z, Jasko S, Abonyi J. Scheduling under uncertainty
org/10.1080/10556780500104090. for industry 4.0 and 5.0. IEEE Access 2022;10(June):74977–5017. https://doi.
[197] Cha JH, Finkelstein M. Optimal long-run imperfect maintenance with asymptotic org/10.1109/ACCESS.2022.3191426.
virtual age. IEEE Trans Reliab 2016;65(1):187–96. https://doi.org/10.1109/ [231] Compare M, Antonello F, Pinciroli L, Zio E. A general model for life-cycle cost
TR.2015.2451612. analysis of Condition-Based Maintenance enabled by PHM capabilities. Reliab
[198] Knowles M, Baglee D, Wermter S. Reinforcement learning for scheduling of Eng Syst Saf 2022;224(April):108499. https://doi.org/10.1016/j.
maintenance. In: Research and Development in Intelligent Systems XXVII; 2011. ress.2022.108499.
p. 409–22. https://doi.org/10.1007/978-0-85729-130-1. [232] Rinaldi G, Thies PR, Johanning L. Current status and future trends in the
[199] Dao CD, Zuo MJ. Selective maintenance of multi-state systems with structural operation and maintenance of offshore wind turbines: a review. Energies 2021;14
dependence. Reliab Eng Syst Saf 2017;159(November):184–95. https://doi.org/ (9). https://doi.org/10.3390/en14092484.
10.1016/j.ress.2016.11.013. [233] Syan CS, Ramsoobag G. Maintenance applications of multi-criteria optimization: a
[200] Shen Y, Zhang X, Shi L. Joint optimization of production and maintenance for a review. Reliab Eng Syst Saf 2019;190(May):106520. https://doi.org/10.1016/j.
serial–parallel hybrid two-stage production system. Reliab Eng Syst Saf 2022;226 ress.2019.106520.
(December 2021):108600. https://doi.org/10.1016/j.ress.2022.108600. [234] Çalık A. A novel Pythagorean fuzzy AHP and fuzzy TOPSIS methodology for green
[201] Dursun İ, Akçay A, van Houtum GJ. Age-based maintenance under population supplier selection in the Industry 4.0 era. Soft Comput 2021;25(3):2253–65.
heterogeneity: optimal exploration and exploitation. Eur J Oper Res 2022;301(3): https://doi.org/10.1007/s00500-020-05294-9.
1007–20. https://doi.org/10.1016/j.ejor.2021.11.038. [235] Katoch S, Chauhan SS, Kumar V. A review on genetic algorithm: past, present, and
[202] Zhu Z, Xiang Y, Zeng B. Multicomponent maintenance optimization: a stochastic future. Multimedia Tools Appl 2021;80(5).
programming approach. INFORMS J Comput 2021;33(3):898–914. https://doi. [236] Cao D, et al. Reinforcement learning and its applications in modern power and
org/10.1287/ijoc.2020.0997. energy systems: a review. J Mod Power Syst Clean Energy 2020;8(6):1029–42.
[203] Nguyen V, Do P, Vosin A, Iung B. Artificial-intelligence-based maintenance https://doi.org/10.35833/MPCE.2020.000552.
decision-making and optimization for multi-state component systems. Reliab Eng [237] Zelvelder AE, Westberg M, Främling K. Assessing explainability in reinforcement
Syst Saf 2022;228(August):108757. https://doi.org/10.1016/j.ress.2022.108757. learning. In: 3rd International Workshop on EXplainable and TRAnsparent AI and
[204] Mishra SK, Mahapatra D, Making D. Maintenance strategy and decision making - Multi-Agent Systems (EXTRAAMAS 2021); 2021. p. 223–40.
AHP method. Int J Adv Eng Res Stud 2015;IV(II):256–8. [238] Antonello F, Baraldi P, Zio E. Multi-objective evolutionary pattern mining
[205] Asemi A, Baba MS, Asemi A, Abdullah RBH, Idris N. Fuzzy multi criteria decision algorithm for the identification of functional dependencies in complex technical
making applications : a review study. ResearchGate 2014;(December):344–51. infrastructures from alarm data. In: Proceedings of the 29th European Safety and
[206] Shyur HJ, Shih HS. A hybrid MCDM model for strategic vendor selection. Math Reliability Conference (ESREL); 2020.
Comput Model 2006;44(7–8):749–61. https://doi.org/10.1016/j. [239] Geng J, Azarian M, Pecht M. Opportunistic maintenance for multi-component
mcm.2005.04.018. systems considering structural dependence and economic dependence. J Syst Eng
[207] C.-L. Hwang and K. Yoon, Multiple attribute decision making: methods and Electron 2015;26(3):493–501. https://doi.org/10.1109/JSEE.2015.00057.
applications a state-of-the-art survey. 1981. [240] Di Pasquale R, Marenco J. Optimization meets big data: a survey. In: IEEE
[208] Roy B. ELECTRE III: Algorithme de Classement Basé sur une Représentation Floue Congress on Evolutionary Computation (CEC) 2017; 2017.
des Préférences en Présence de Critères Multiples. Cah du CERO 1978;20(1):3–24. [241] Yamagata T, McConville R, Santos-Rodriguez R. Reinforcement learning with
[209] Siksnelyte-Butkiene I, Zavadskas EK, Streimikiene D. Multi-criteria decision- feedback from multiple humans with diverse skills. In: NeurIPS 2021 Workshop
making (MCDM) for the assessment of renewable energy technologies in a on Safe and Robust Control of Uncertain Systems; 2021.
household: a review. Energies 2020;13(5). https://doi.org/10.3390/en13051164. [242] Ning C, You F. Optimization under uncertainty in the era of big data and deep
[210] Kamiński B, Jakubczyk M, Szufel P. A framework for sensitivity analysis of learning: when machine learning meets mathematical programming. Comput
decision trees,” Cent. Eur J Oper Res 2018;26(1):135–59. https://doi.org/ Chem Eng 2019;125:434–48. https://doi.org/10.1016/j.
10.1007/s10100-017-0479-6. compchemeng.2019.03.034.
[211] Sahinidis NV. Mixed-integer nonlinear programming 2018. Optim Eng 2019;20 [243] Wang C, Wang B, Liu H, Qu H. Anomaly detection for industrial control system
(2):301–6. https://doi.org/10.1007/s11081-019-09438-1. based on autoencoder neural network. Wirel Commun Mob Comput 2020;2020.
[212] Higle JL. Stochastic programming: optimization when uncertainty matters. https://doi.org/10.1155/2020/8897926.
INFORMS TutORials Oper Res 2005;(September):30–53. https://doi.org/ [244] Yang Z, Baraldi P, Zio E. A method for fault detection in multi-component systems
10.1287/educ.1053.0016. based on sparse autoencoder-based deep neural networks. Reliab Eng Syst Saf
[213] Putz D, Schwabeneder D, Auer H, Fina B. A comparison between mixed-integer 2022;220(December 2021):108278. https://doi.org/10.1016/j.
linear programming and dynamic programming with state prediction as novelty ress.2021.108278.

18
L. Pinciroli et al. Reliability Engineering and System Safety 234 (2023) 109204

[245] Fan Y, Xiao F, Li C, Yang G, Tang X. A novel deep learning framework for state of [251] M. Breque, L. De Nul, and A. Petrides, “Industry 5.0 - Towards a sustainable,
health estimation of lithium-ion battery. J Energy Storage 2020;32(June): human-centric and resilient European industry,” 2021.
101741. https://doi.org/10.1016/j.est.2020.101741. [252] Xu X, Lu Y, Vogel-Heuser B, Wang L. Industry 4.0 and industry 5.0—inception,
[246] Rigamonti M, Baraldi P, Zio E, Roychoudhury I, Goebel K, Poll S. Ensemble of conception and perception. J Manuf Syst 2021;61(September):530–5. https://doi.
optimized echo state networks for remaining useful life prediction. org/10.1016/j.jmsy.2021.10.006.
Neurocomputing 2018;281:121–38. https://doi.org/10.1016/j. [253] Sun H, Yang M, Wang H. Resilience-based approach to maintenance asset and
neucom.2017.11.062. operational cost planning. Process Saf Environ Prot 2022;162:987–97. https://
[247] Nguyen HP, Baraldi P, Zio E. Ensemble empirical mode decomposition and long doi.org/10.1016/j.psep.2022.05.002.
short-term memory neural network for multi-step predictions of time series [254] Moreno-Sader K, Jain P, Tenorio LCB, Mannan MS, El-Halwagi MM. Integrated
signals in nuclear power plants. Appl Energy 2021;283(November 2020):116346. approach of safety, sustainability, reliability, and resilience analysis via a return
https://doi.org/10.1016/j.apenergy.2020.116346. on investment metric. ACS Sustain Chem Eng 2019;7(24):19522–36. https://doi.
[248] Staar B, Lütjen M, Freitag M. Anomaly detection with convolutional neural org/10.1021/acssuschemeng.9b04608.
networks for industrial surface inspection. Procedia CIRP 2019;79:484–9. https:// [255] Matyas K, Nemeth T, Kovacs K, Glawar R. A procedural approach for realizing
doi.org/10.1016/j.procir.2019.02.123. prescriptive maintenance planning in manufacturing industries. CIRP Ann -
[249] Tanguy L, Tulechki N, Urieli A, Hermann E, Raynal C. Natural language Manuf Technol 2017;66(1):461–4. https://doi.org/10.1016/j.cirp.2017.04.007.
processing for aviation safety reports: From classification to interactive analysis. [256] Hodkiewicz M, Lukens S, Brundage MP, Sexton T. Rethinking maintenance
Comput Ind 2016;78:80–95. https://doi.org/10.1016/j.compind.2015.09.005. terminology for an industry 4.0 future. Int J Progn Heal Manag 2021;12:1–14.
[250] Yang Z, Baraldi P, Zio E. A novel method for maintenance record clustering and [257] Almakhlafi A, Knowles J. Benchmarks for maintenance scheduling problems in
its application to a case study of maintenance optimization. Reliab Eng Syst Saf power generation. In: 2012 IEEE Congr. Evol. Comput. CEC 2012; 2012. p. 10–5.
2020;203:107103. https://doi.org/10.1016/j.ress.2020.107103. https://doi.org/10.1109/CEC.2012.6252988.

19