machines

Article
Systems Reliability and Data Driven Analysis for Marine
Machinery Maintenance Planning and Decision Making
Abdullahi Abdulkarim Daya 1,2, * and Iraklis Lazakis 1

1 Department of Naval Architecture Ocean and Marine Engineering, University of Strathclyde,

100 Montrose Street, Glasgow G4 0LZ, UK; [email protected]
2 Nigerian Navy, Naval Headquarters, Area 8 Garki, Abuja 900110, Nigeria
* Correspondence: [email protected]

Abstract: Understanding component criticality in machinery performance degradation is important

in ensuring the reliability and availability of ship systems, particularly considering the nature of ship
operations requiring extended voyage periods, usually traversing regions with multiple climate and
environmental conditions. Exposing the machinery system to varying degrees of load and operational
conditions could lead to rapid degradation and reduced reliability. This research proposes a tailored
solution by identifying critical components, the root causes of maintenance delays, understanding
the factors influencing system reliability, and recognising failure-prone components. This paper
proposes a hybrid approach using reliability analysis tools and machine learning. It uses dynamic
fault tree analysis (DFTA) to determine how reliable and important a system is, as well as Bayesian
belief network (BBN) availability analysis to assist with maintenance decisions. Furthermore, we
developed an artificial neural network (ANN) fault detection model to identify the faults responsible
for system unreliability. We conducted a case study on a ship power generation system, identifying
the components critical to maintenance and defects contributing to such failures. Using reliability
importance measures and minimal cut sets, we isolated all faults contributing over 40% of subsystem
failures and related events. Among the 4 MDGs, the lubricating system had the highest average
availability of 67%, while the cooling system had the lowest at 38% using the BBN availability outcome.
Citation: Daya, A.A.; Lazakis, I. Therefore, the BBN DSS recommended corrective action and ConMon as maintenance strategies due
Systems Reliability and Data Driven to the frequent failures of certain critical parts. ANN found overheating when MDG output was
Analysis for Marine Machinery
above 180 kVA, linking component failure to generator performance. The findings improve ship
Maintenance Planning and Decision
system reliability and availability by reducing failures and improving maintenance strategies.
Making. Machines 2024, 12, 294.
https://doi.org/10.3390/
Keywords: fault; components; criticality; maintenance; ship systems; system reliability
machines12050294

Academic Editors: Wilson Cesar

Sant’Ana, Helcio Francisco Villa-Nova
and Erik Leandro Bonaldi 1. Introduction
Received: 12 February 2024 The operations of marine platforms such as offshore platforms, ships, and boats
Revised: 15 April 2024 are increasingly being scrutinised for emission controls and safety of operations. This is
Accepted: 24 April 2024 more so because ships are central to global trade, responsible for over 90% of the global
Published: 27 April 2024 carriage of goods and services [1]. In addition, ships provide very essential services in the
installation and maintenance of all types of offshore renewable power generation systems,
as well as services in the oil and gas and security sectors. In order to ensure safety of life
and the reduced environmental impact of ships and other maritime platforms, original
Copyright: © 2024 by the authors.
equipment manufacturers (OEM) and operators rely on reliability analysis tools and data
Licensee MDPI, Basel, Switzerland.
driven methods onboard or handheld sensors to access machinery information. This way,
This article is an open access article
the health and performance of the machinery can be monitored in real time or recoded for
distributed under the terms and
future use. The essence is to enable documentation that can either be used in real time or
conditions of the Creative Commons
Attribution (CC BY) license (https://
for future diagnosis and prognosis to help with evidence-based machinery maintenance
creativecommons.org/licenses/by/
approach. Moreover, the complexity of systems onboard is such that there is a lot of
4.0/).
inter-operational dependencies either in a machinery unit or system.

Machines 2024, 12, 294. https://doi.org/10.3390/machines12050294 https://www.mdpi.com/journal/machines

Machines 2024, 12, 294 2 of 28

Ships in general are built for different purposes and applications; each of these has
its peculiarities with regard to maintenance and other regulations, which can vary from
state to state [2]. Notwithstanding, all ships constitute intricate connections systems and
equipment design and are constructed to move from one point to another and provide
support for human and machine inhabitation, as well as engage in designated services. In
accordance with Stopford’s [3] findings, it is generally observed that ships are constructed
with an intended operational lifespan of approximately 25 years, with the potential for
extension up to 35 years or beyond [4]. The life cycle management of ships and on-board
systems is frequently not adequately prioritised, resulting in significant maintenance costs
beyond the initial 5-year period [5,6].
In this regard, operators are burdened with substantial maintenance expenses and
frequently find themselves in forced marriages or alliances with equipment producers,
in some cases having to bear the burden of equipment upgrades and retrofits due to
regulations [7]. However, perception of maintenance activities differs among various
operators, with many considering it to be a routine task that is only given significant
attention in the case of a breakdown or when preparing for operations or exercises [8]. This
approach contradicts the fundamental purpose of maintenance in that maintenance can
be described as a comprehensive set of activities encompassing technical, administrative,
and managerial measures undertaken throughout the lifespan of an item with the aim
of preserving or reinstating its ability to fulfil the desired function [9]. It can therefore
be inferred that maintenance, as a fundamental engineering service, plays a crucial role
in operations, which significantly contributes to the achievement or failure of system
availability in a fleet that could translate to revenue generation.
On the other hand, performance degradation in machineries could result in increased
emissions in some machineries such as marine diesel generators and main propulsion
engines, while in other systems such as pumps, this could lead to high levels of power
consumption. Conversely, the presence of numerous equipment units that operate entirely
autonomously poses a significant challenge in terms of maintenance planning, which
cannot be effectively accomplished manually without some degree of data automation. In
this regard, operators must create a maintenance strategy for condition monitoring that
is appropriate for its intended purpose. When evaluating various aspects, it is important
to consider elements such as the capacity of maintenance staff, availability of spare parts,
environmental conditions, mission needs, future job predictions on the platform or fleet,
and data management and processing [10].
Thus, this paper will detail the impact of combining reliability analysis tools and a
machine learning approach on machinery degradation and reliability analysis. In this
regard, the paper will be presented in 5 sections: Section 1 includes the introduction to the
topic while the critical literature review focusing on machinery health degradation and
component criticality analysis is presented in Section 2. The methodology on the hybrid
approach is presented in Section 3, and Section 4 presents the case study. Finally, the results
and discussion are presented in Section 5, closely followed by the conclusions in Section 6.

2. Critical Literature Review

The performance degradation of mechanical systems diminishes with age and usage
due to component wear, tear, or fatigue. This degradation in performance is usually exhib-
ited through a reduction in the ability to carry normal loads or an inability to operate for
extended periods within the normal operational envelope and material conditions [11]. In
this regard, system reliability analysis is enabled using historical failure and repair data,
as well as the component and/or faults conditions responsible for machinery breakdown.
However, wear and tear as well as other causes of machinery failure can be delayed or
accelerated by some external or internal factors. External factors could include the en-
vironment, human factors such as technical skills, maintenance practice, and quality of
consumables; others could associate failure with faults from other machinery. Internal
factors include material failure, component failure as a result of shelf life, contamination,
Machines 2024, 12, 294 3 of 28

design issues, overload, etc. [12,13]. In this regard, research in the field of maintenance has
received a lot of attention, especially pertaining to the combination of reliability analysis
and data driven methods to enable comprehensive reliability and diagnostic and prog-
nostic analysis [14–16]. Moreover, condition-based maintenance techniques, which are
a viable replacement for planned or time-based maintenance concepts, largely underpin
the effectiveness of reliability-centred maintenance. [17]. Nonetheless, the application of
data-driven methods in predictive maintenance techniques were widely researched by
authors such as Li, Gebraeel [18,19] and Wang, Chen [16], providing efficient optimisation
models for implementation. Similarly, the use of supervised and non-supervised learning
for clustering and anomaly detection-based machinery performance degradation on naval
propulsion system was presented in [20,21].The use of data driven methods in machin-
ery reliability and degradation analysis have been research in [6,22] where the authors
have developed methodologies to improve reliability analysis outcomes using data-driven
methods, especially in machinery performance degradation analysis.

2.1. Degradation Analysis

Performance degradation analysis thrives on the strength of data-driven methods
enabled by machine learning. Maintenance strategies such as condition monitoring and
predictive maintenance are popular due to their ability to provide real machinery conditions
as well future machinery conditions based on current data. These achievement are all
thanks to intelligent sensors as well as real-time data transmission enabled by the internet,
commonly referred to as the internet of things of industry 4.0 [23]. Data-driven methods
have provided the biggest push in machinery degradation analysis due to the use of
qualitative sensor data, which can provide an unbiased image of machinery conditions.
The importance of data-driven methods is found more in the maintenance strategy of
critical machines and their components [24]. Therefore, data-driven methods have been
used to establish machinery conditions that can be responsible for certain failures through
indicators such as vibration, temperature, acoustics, pressure, speed, etc. [25–27].
The emergence of advance sensor technology gave rise to big data management, as
operators are able collect large amount of machinery health data at short intervals, enabled
by high-speed internet and wireless connectivity [28,29]. The combination of these tech-
nologies enables companies to monitor machinery in real time online, and in some cases,
provide control and diagnosis [23]. Therefore, big data or industry 4.0 enables real-time
insights into equipment performance, scheduling optimization, and downtime reduction,
hence revolutionising how maintenance operations are carried out across a variety of
industries [30]. Companies can foresee equipment failure and carry out proactive mainte-
nance by evaluating data from sensors and other sources, which lowers the possibility of
unscheduled downtime and boosts overall equipment efficiency [31].

Artificial Neural Networks

In general, there two types of machine learning approaches, namely, supervised and
unsupervised learning. The supervised machine learning is used to train a model with
labelled data, that is, the features to be looked out for are already known; therefore, the
algorithm is trained to look out for those features in the input data. On the other hand,
unsupervised learning deals with unlabelled data, which means the algorithm will identify
unique features in the data and partition it accordingly. Unsupervised learning is useful for
exploring data in order to understand the natural patten of the data, especially when there
is no specific information about significant incidents in the data that can easily point to
some fault indicators. Artificial neural networks (ANNs) have been applied in the field of
maintenance for machinery health analysis and the prediction of machinery conditions by
various authors. As an unsupervised learning method, SOMs are effective for data analysis
and clustering, as demonstrated by their use in identifying nonlinear latent features from
high dimensional data [32].
Machines 2024, 12, 294 4 of 28

Therefore, riding on the existing success and procedures in the use of ANNs for
machinery data analysis, this research will employ an ANN for fault classification and
detection, fault/condition prediction, and machinery remaining useful life analysis [33].
An ANN approach for fault detection was applied with FTA to identify the critical com-
ponent of a diesel generator in a research presented by [26,34]. In some cases, machinery
fault data are recorded without identifying the fault signals; therefore, this requires data
clustering. Clustering is a form of unclassified machine learning that is applied in machin-
ery diagnostics [26]. The advantages of using clustering models include helping identify
possible clusters as well as the most influential clusters in the data. In the research, ANN
self-organising maps (SOMs) were used for clustering machinery log data of DG. SOMs
consist of competitive layers which can classify a dataset of vectors with any number of
dimensions as the number neurons in the layer and are good for dimensionality reduction,
as presented in [19,35].
Accordingly, ANNs are widely employed for multiple tasks such as clustering, fore-
casting, prediction, pattern recognition, classification, and feature engineering [36]. ANNs
and Regression techniques were employed to estimate vessel power and fuel consumption
where the model was able to predict actual vessel fuel consumption in real time [37]. The
use of ANNs for fault classification has been employed by [34], using self-organising map
ANN clustering algorithm to analyse the health parameter of a marine diesel engine, look-
ing at the exhaust gas temperature, piston cooling outlet temperature, and piston cooling
inlet pressure. Therefore, the performance of ANNs in prediction and classification, as
reviewed in [38,39], was presented to be good in handling nonlinear, high-dimensional
data with a smaller dataset. In this regard, building on the success of ANNs, this work
will apply the use of an ANN on labelled data for diagnostic analysis on 4 sets of marine
diesel generators. Therefore, the feedback from the ANN is used in combination with the
reliability analysis output to identify the dominant faults and most affected components.
The success of a diagnostics analysis platform using artificial intelligence or machine
learning depends on an efficient and standardised data management system, in particular
when the information is needed for maintenance planning. To make sense of the informa-
tion available in a set of data collected from single or multiple machinery, it is important
that there is a common platform and procedure understood by all within and outside the or-
ganisation. Moreover, [40] defines data as any reinterpretable representation of information
in a formalised manner suitable for communication, hence the need to ensure that there is
standardisation of data such that there is little or no interpretation needed at the point of
use of analysis. While ship data as described in [41] is a measurement value from shipboard
machine and equipment to which a time stamp is added, maintenance engineers have
long depended on machinery data as their main source of information for understanding
the present and future health condition of the machinery. Similarly, a broad methodology
utilising various sensor data and technologies has been presented in the INCASS project,
which provides research data and methodology for both ship machinery and structural
risk analysis, enabled by the combination of sensor data and failure and repair data for
machinery health and reliability analysis [42,43]. Nonetheless, these technologies present
some challenges that companies need to be aware of, which are as follows [40]:
1. Data quality: Successful implementation depends on the quality of data; therefore, it
is important to ensure data accuracy, completeness, reflectiveness, and relevance to
the requirement of the ship.
2. Data Security: Appropriate security measures are needed to guard against cyberat-
tacks and unauthorised access when storing and transmitting significant amounts of
data from sensors and other sources.
3. Data Integration: In order to analyse and interpret big data from diverse sources, the
right tools and technologies must be used.
4. Competence: Companies must ensure they have the right competence and tools in
obtaining, analysing, and interpreting data so that they may make wise judgements;
hence, ships must have the requisite expertise or hire one.
 Machines 2024, 12, 294 5 of 29

Machines 2024, 12, 294 4. Competence: Companies must ensure they have the right competence and tools 5 of 28
in
obtaining, analysing, and interpreting data so that they may make wise judgements;
hence, ships must have the requisite expertise or hire one.
InInthis
thisregard,
regard,[19]
[19] discusseda system
discussed a system ofof predicting
predicting machineryhealth
machinery healthmonitoring
monitoring
using
using ANNsand
ANNs andFTA
FTAfor forreliability
reliability analysis,the
analysis, the methodology
methodology hashassuccessfully
successfully identified
identified
and
and defined
defined measurement
measurement for
for machinery
machinery data
data through
through step-by-step
step-by-step demonstration
demonstration ofof the
the
process
process and
and identificationofofthe
identification the criticalcomponent
critical componentusing usingFTA. FTA.Moreover,
Moreover,ISO ISO19847
19847[44]
[44]
andISO19845
and ISO19845[45] [45]provides
provides standard
standard guidelines
guidelines and anddefinition
definitionfor foronboard
onboard ship data
ship datacol-
lection, storage,
collection, storage,management,
management,and andtransmission
transmissionvia viathe
theinternet.
internet.Nonetheless,
Nonetheless,there thereal-
already
ready exists commonly
commonly understood
understood formats
formatsfor formanaging
managingand andcollecting
collectingdatadataonboard
onboard
ships
shipsthat
thatarearegenerated
generatedvia various
via varioussources
sourcesonon board,
board, asasshown
shown inin
Figure
Figure 1. 1.
Although
Although
there
theremaymaybebe nomenclature
nomenclaturedifferences,
differences,for forinstance,
instance,between
betweenmerchantmerchantand andnaval
navalships,
ships,
the
therecords
recordsmay maystill
stillbebereferring
referringtotothe same
the same objective.
objective.ItItisisa astandard
standardrequirement
requirementfor for
merchant
merchant ships
shipstoto
hold
hold historical
historicalrecords
records ofofship
ship repair
repairandand maintenance
maintenance being
beingmanaged
managed
bybyclassification
classificationsociety
societywho whoalso
alsoprovide
provideadditional
additionalstandards
standardsand andguidelines
guidelinesfor fordata
data
collection
collection and management [21,46]. These documents provide vital information onthe
and management [21,46]. These documents provide vital information on the
location,
location,speed,
speed,time,
time,engine
engine speed,
speed, andand generator(s)
generator(s) online,
online,asaswell wellasasother
other systems
systems op-
operational withina agiven
erational within giventime.time. The
The records
records provide
provide hourly
hourly updates
updates of the
of the operational
operational state
state
andand consumption
consumption ratesrates of important
of important machinery.
machinery. Typically,machinery
Typically, machineryhealth
healthconsists
consistsof
oftime
timeseries
seriesdata
datapoints
pointsofofsomesomeimportant
importantparameters,
parameters,such suchasastemperature,
temperature, pressure,
pressure, vi-
vibration, consumption rates, outputs, speed, load, deflections,
bration, consumption rates, outputs, speed, load, deflections, and clearances. and clearances.

Figure Ship
1. 1.
Figure Maintenance
Ship data
Maintenance Sources.
data Sources.
2.2. System Reliability Analysis
2.2. System Reliability Analysis
System reliability analysis is central to the successful implementation of any main-
tenance System reliability
strategy analysisclear
as it provides is central to the
insight on successful
machineryimplementation
behaviour andof any
the mainte-
impacts
of failure on the availability of machineries up to system levels. Accordingly, reliabilityof
nance strategy as it provides clear insight on machinery behaviour and the impacts
failure tools
analysis on theareavailability
widely used of machineries up to system
to support maintenance levels. selection
strategy Accordingly, reliability anal-
or implementation
ysis tools are widely used to support maintenance strategy selection
in line with organisational objectives. Therefore, various maintenance strategy such as or implementation
in line with organisational
reliability-centred maintenance objectives.
(RCM),Therefore,
risk-basedvarious maintenance
maintenance, strategy such
total productive as re-
main-
liability-centred
tenance, risk- and maintenance (RCM),
reliability-based risk-basedetc.,
maintenance, maintenance,
draw fromtotal productive
existing mainte-
maintenance
nance, risk- and reliability-based maintenance, etc., draw from existing
approaches using system reliability analysis to provide a tailored maintenance system [27]. maintenance ap-
proaches using system reliability analysis to provide a tailored maintenance
RCM was developed in the aviation industry and United States Navy in the 1970s [47] system [27].
RCM was
provides developed
clear intersectionin the aviation
on the industryofand
combination United
various States Navy
maintenance in theand
strategy 1970s
used[47]
provides clear intersection on the combination of various maintenance
of reliability tools. Moreover, maintainability analysis carried out at the design stage of strategy and used
of reliability
products tools. Moreover,
or platforms such shipsmaintainability
or aircraft and analysis carriedmachinery
other complex out at theare design stage
carried outof
products
with the useorofplatforms
reliabilitysuch shipstools
analysis or aircraft
such asand other
FTA, DFTA,complex
BBN,machinery
etc. [8,12]. are carried out
An overview
ofwith the use of
the adopted reliability
tools in this analysis
research tools such as FTA, DFTA, BBN, etc. [8,12]. An overview
is given.
of the adopted tools in this research is given.
2.2.1. Fault Tree Analysis
Fault tree analysis (FTA) is a graphical method of analysing and presenting how a
system or piece of equipment may fail. FTA was developed using a deductive approach that
began with the main fault, known as TOP EVENT, and progressed to the propagated events,
known as Primary or Basic Event [48]. Low-level events are further classified into basic,
house, conditional, and underdeveloped events, which can be thought of as fault conditions
that preside over the top event [49]. Simply put, the fault tree depicts the propagation
Machines 2024, 12, 294 6 of 28

of lower-level events that result in a system-wide undesirable or top event. As a result,

FTA uses logic gates and events to present failure patterns on systems or equipment for
both quantitative and qualitative analysis. Fault tree analysis (FTA) is a static method for
analysing component faults in systems or equipment by identifying all possible causes of
likely failures and impacts on the system through the logical analysis of dependencies of
basic events that lead to the undesired event, the top event of the fault tree [50].
The primary objective of fault tree analysis (FTA) is to assess the likelihood of the top
event occurring and illustrate the sequence of events that could lead to the occurrence of
the top event [51]. Before entering numerical data into a fault tree, it is advisable to conduct
a qualitative analysis using the FTA logic structure. Therefore, to calculate the likelihood
of the top event happening, it is necessary to generate quantitative information about the
system’s reliability and maintainability, such as the probability of failure, failure rate, or
repair rate. This information will then be used as inputs for the qualitative analysis of
failures. However, FTA has some shortcomings regarding sequence dependencies, temporal
order of occurrence, and redundancies due to standby systems; consequently, DFTA was
developed to overcome these constraints in the static FTA.

2.2.2. Dynamic Fault Tree Analysis

Dynamic FTA is an improvement over the static FTA tool; it is an equally important
tool for reliability and risk analysis as it provides critical information on system component
failure development. DFTA is so significantly improved over the static FTA such that
in addition to prioritising the importance of the contributors to the undesired events, it
also considers time and sequence dependencies [52]. Moreover, it uses Boolean law by
applying gates and events to describe faulty components and possible event(s) that could
develop a fault, as well as Markov Chains for the state probability occurrence among
multiple events [53], hence accounting for sequence- and time-dependent analysis. The
dynamic gates, which include the priority and gate (PAND), sequence enforcing gate (SEQ),
functional dependency gate (FDEP), spare gate (SPARE), and the spare event, when added
to the FTA structure, become dynamic FTA [50].
In the PAND gate, events are prioritized from left to right such that the left most event
(fault) is considered first before the next; similarly, SEQ considers events in left to right
fashion; however, rather than prioritizing, it enforces, thus ensuring that events follow the
expected failure mechanism [54]. On the other hand, FDEP evaluate events from left to
right considering the occurrence of primary or causal event, which is independent of other
faults to the right [55]. The SPARE gate and event have unique attributes and functions;
though events are evaluated from left to right as obtained in other gates, the dormancy
factor feature of the spare event makes lot of difference. The dormancy factor is a measure
of the ratio between failure and operational rate of the spare event in the standby mode [50].
A cold spare has dormancy factor 0, a hot spare has dormancy factor 1, and a warm spare
has a dormancy factor between 0 and 1 [52].
Overall, DFTA provides a platform that is capable of analysing reparable systems
while considering other factors such as dependencies and temporal behaviour or partially
operating state analysis [55–57]. Therefore, this makes it very relevant in analysing system
improvements as presented in [24,55–58]. Overall, these additional gates provided more
scope in DFT analysis [54,59], which can be used to factor repair or improvements due
to routine maintenance. Moreover, additional outputs such as the reliability importance
measures and minimal cuts sets in the DFTA are equally influenced by the logic structure
such of the model, meaning the output of static FT and dynamic in those output will be
significantly different and reflective of whatever dependencies exist in the model.

2.2.3. Bayesian Belief Network

Bayesian belief networks (BBNs) provide a good platform for dependability analysis,
and cause, effect, and inferential analysis in a wide range of sectors covering health care,
human reliability, machinery system reliability, and decision support systems. BBNs are
Machines 2024, 12, 294 7 of 28

represented as a direct acyclic graph (DAG) which consist of chance nodes (variables)
representing possible outcomes of system states and a given set of arrows (connections)
indicating dependability/relationships. The nodes can take variable inputs in BBN analysis
which can be continues or discreet and are not restricted to single top events, thus providing
great flexibility unlike fault tress or RBD [60]. BBNs can be used to represent cause and
effect between parts of systems or equipment by identifying potential causes of failure.
Authors have used BBNs for fault and diagnostic analysis as well decision support system
(DSS), for instance Jun et al. [61] presented a Bayesian-based fault identification system
for CBM by discretising continues parameters based on maximum likelihood estimation
(MLE) to identify failure conditions; the research used the discretised feature as binary
inputs for the BBN conditional probability table (CPT). Similarly, to address port energy
efficiency towards the reduction ships emission during port calls, a strategy using BBNs
was presented in [62]. This research also explains how BBN conditional probability can
efficiently in-cooperate to expert knowledge to provide vital inputs in decision making
variables in areas where there is inadequate data or literature.
Bayesian updating or inference provides bases for the use of influence diagrams in
decisions analysis by computing the impact of new evidence to the probability of events
and the influence on all related nodes [63]. As such, BNs provide a good platform for DSS,
especially in maintenance strategy when considering several dependent and independent
factors. Conducting system reliability and maintenance analysis demands inputs from
multiple sources which the BN platform can accommodate, as compared to other tools.
Papers by Jun et al. [61] and Li et al. [64] provide methodologies for the use of BBNs
in reliability analysis; however, while [61] focused on fault diagnosis, [64] emphasises
component reliability with limited analysis on factors affecting the reliability. Furthermore,
BBNs have been used to provide inferential analysis in conjunction with other tools such
as the Markov chain and Petri-nets, especially in risk and reliability analysis [60,65–67].
BBN-based DSS are widely applied in maritime industry to handle operational issues such
as human factors, and procedural issues such as maintenance [67,68]. Similarly, in the
field of ship system reliability analysis, Lazakis et al. [56] have presented on the use of
BBNs and FTA for ship machinery cooling system reliability analysis and DSS. Likewise,
BahooToroody et al. [69] applied a combination BBN and Markov chain Monte Carlo
simulation to analyse machinery reliability estimation onboard autonomous ships to help
maintenance planning and decision.
Overall, the critical literature review has established the prevalence of the use of
multiple tools in reliability and diagnostic analysis by several authors, especially in mainte-
nance planning, to overcome certain restrictions or provide further layers of validation [5].
Therefore, a combination of multiple tools for system reliability analysis has been in prac-
tice by many authors, this combination is necessitated by the complex dynamics of fault
development and organisational maintenance challenges [24,70]. In this regard, some of the
identified gaps in the literature addressed by this paper include assessing the importance
of components in relation to system availability, choosing maintenance decisions based on
the operator’s sentiment, and using IM obtained from DFTA to build a maintenance DSS
using BBN. Therefore, a novel methodology was developed which uses the DFTA outputs
as inputs for the BBN maintenance DSS. Similarly, a fault detection model was developed
to diagnose the prevalent causes of failure to aid maintenance DSS.

3. Methodology
The research methodology utilises the combined strength of reliability analysis tools for
system reliability and criticality, artificial neural network for diagnosis and fault prediction,
and Bayesian belief networks for maintenance decision support systems. Maintenance,
repair, overhaul, and machinery log data obtained from onboard ship system and machinery
were analysed so as to understand the courses of failure in diesel generators, identify the
most critical component, and provide possible ways to improve maintenance on board
ships. This process would assist both onboard technical staff as well as shore support units.
 The research methodology utilises the combined strength of reliability analysis tools
for system reliability and criticality, artificial neural network for diagnosis and fault pre-
diction, and Bayesian belief networks for maintenance decision support systems. Mainte-
nance, repair, overhaul, and machinery log data obtained from onboard ship system and
Machines 2024, 12, 294 machinery were analysed so as to understand the courses of failure in diesel generators, 8 of 28
identify the most critical component, and provide possible ways to improve maintenance
on board ships. This process would assist both onboard technical staff as well as shore
The process
support of The
units. the research
process ofinvolves the collection
the research of machinery
involves the collection ofdata from andata
machinery offshore
from
patrol vesselpatrol
an offshore (OPV)vessel
which(OPV)
were then
which analysed
were thento generate
analysed outputs relevant
to generate outputsto machinery
relevant to
health performance
machinery indicators. The
health performance research has
indicators. Thethree broad
research hasareas which
three broadareareas
usedwhich
as inputs
are
or in combination to analyse the condition of machinery health, as shown
used as inputs or in combination to analyse the condition of machinery health, as shownin the Figure 2
below. As shown in the figure
in the Figure 2 below. As shown inbelow, the three areas are the system reliability analysis using
figure below, the three areas are the system relia-
DFTA, fault detection
bility analysis and prediction
using DFTA, using ANNs,
fault detection and BBNsusing
and prediction for developing
ANNs, and a maintenance
BBNs for de-
decision
velopingsupport system. decision support system.
a maintenance

Figure2.2.Methodology.
Figure Methodology.

Thedata
The dataused
usedininthis
thisstudy
studywere
wereobtained
obtained through
through aa data
data collection
collection campaign
campaign con- con-
ducted onboard a Nigerian Navy OPV. It encompassed machinery
ducted onboard a Nigerian Navy OPV. It encompassed machinery health and historical health and historical
data,such
data, suchasasmaintenance
maintenanceand and overhaul
overhaul records,
records, repair
repair data,
data, andand machinery
machinery healthhealth rec-
records.
ords.
The The dataset
dataset covered covered the machinery
the machinery log forlog for a period
a period of up toof12upcalendar
to 12 calendar
months,months,
along
along
with with historical
historical recordsrecords
spanning spanning
up to 18 upmonths.
to 18 months. The collected
The collected machinery
machinery log and
log data data
aand a portion
portion of theofmaintenance
the maintenance and overhaul
and overhaul records
records werewere in manuscript
in manuscript form,form,
which which
ne-
necessitated
cessitated conversion
conversion intointo electronic
electronic format.
format. In In this
this regard,
regard, datadata pre-processing
pre-processing played
played a
a pivotal
pivotal rolerole in handling
in handling challenges
challenges related
related to missing
to missing valuesvalues and outliers.
and outliers. Various Various sta-
statistical
methods were used
tistical methods wereto used
interrogate the data;
to interrogate thetherefore, missingmissing
data; therefore, values were
valuesfilled
wereusing
filled
linear
usinginterpolation, as it provides
linear interpolation, a more aconsistent
as it provides approach
more consistent for timefor
approach series
timedata. Ondata.
series the
other
On the hand,
other outlier
hand,detection methodsmethods
outlier detection were applied werebased onbased
applied the specific
on theparameter under
specific parame-
analysis.
ter underInanalysis.
this regard, given
In this the multidimensionality
regard, of the data, aofflexible
given the multidimensionality approach
the data, to
a flexible
outlier
approachcleaning was adopted,
to outlier cleaninginvolving
was adopted,the use of the interquartile
involving the use of range (IQR) for natural
the interquartile range
number
(IQR) forvalues
natural and an iterative
number values approach for numbers
and an iterative approach less
forthan one. less
numbers Thisthan
ensured the
one. This
cleaned data remained consistent with the original dataset.
ensured the cleaned data remained consistent with the original dataset.

3.1. Maintenance and Machinery Health Data Preprocessing

Nonetheless, the approach in data pre-processing helped to address some challenges
with regard to missing values, outliers, or ‘not a number’ (NAN) entries. The main
approach that used outlier detection was the interquartile range (IQR). Missing values were
filled using linear interpolation and, in some cases, forward fill/backward fill was used,
especially on variables representing temperature as they can change in no particular patten.
Therefore, using this method would help retain the randomness in the time series as regard
fault development. These actions were carried out using the MATLAB R2023b software
Machines 2024, 12, 294 9 of 28

application. Additionally, the feature engineering process to identify the most relevant
variables to MDG faults was developed using correlation analysis and is presented in the
ANN diagnostics section.

3.2. Dynamic Fault Tree Analysis

Dynamic fault tree analysis (DFTA) is an extension of standard fault tree analysis
(FTA) that provides time or sequence dependent analysis and can also prioritise events
for analysis. Its flexibility calculation nature allows it to handle failure sequence and
time dependent events especially in reparable mechanical systems. Moreover, the spare
gates provide the DFTA a unique future that is useful for modelling failures in active,
inactive, or warm standby repairable systems. Overall, the addition of four gates and one
basic event in the FTA structure has provided a much more flexible way of modelling
faults/failures in complex systems with respect to sequence and dependencies in the DFTA.
These additions made it possible to consider the temporal order of the occurrence of events
including the dependencies in analysis. DFTA is selected for this study in order to utilise
its system-dependent relationship on the effect of component failures.
Accordingly, failure and maintenance data over a period of six calendar years, obtained
from the maintenance records, were processed to generate components failure rates (⋌)
based on Equation (4). The model structure was built suing both static and dynamic
FT gates and events to reflect the mode of failures and, in other cases, dependency and
sequence. Therefore, top events and sub-events were modelled using dynamic gates while
gates connecting to the main system were modelled using static FTs; this procedure is
necessary to reduce memory usage and improve calculation time. The probabilities for the
static gates used were generally AND gate Equation (5), OR gate Equation (6), and voting
gate Equitation (1):
n
⋌= (1)
τ
where n is number of failures (106 ) and τ is aggregated time in service of individual DG.
The results from the DFTA provides several outputs in addition to the reliability out-
comes; further manipulation of the gates such as the spare gates could also be used to model
the impact of standby systems and spares, while additional improvements in maintenance
activities can also be captured through these processes. Additionally, results that can be
obtained from the analysis include the minimal cut set and reliability importance measures.

3.2.1. Importance Measures

Reliability importance measures (IMs) are a means to identify the most critical com-
ponent or situation that contributes to the occurrence of the basic event leading up to
equipment failure or top event occurrence [48,71]. In essence, the IMs help the operators,
maintenance crew, and administrators including regulatory agency in the prioritisation of
actions that could improve equipment or system reliability. These IMs include Birnbaum
(Bir), Fussell-Vesely (F-V), and Criticality (Cri). The Bir IM evaluates the occurrence of the
top events based the probability of basic event occurring or not occurring, hence the higher
the probability of the basic events the high chances of top event occurring. Criticality (Cri)
IMs are calculated in a similar way to the Bir IM except that it considers the probability in
the occurrence of the basic event to the occurrences of the top event. On the other hand,
the F-V calculation adopts an entirely different approach in that it uses the minimal cut
set summation i.e., the minimum number of basic events that contribute to the top event.
Therefore, the F-V Ims consider the contribution of the basic event to occurrence of the top
event irrespective of how it contributes to the failure.
In this regard, using the Bir IM robust criticality analysis generates more than 290 basic
events modelled at component level failures per individual MDG. The Bir calculation
method provides more accurate results compared to the other 2 Ims; this is because it
considers all possible failures based on their individual contributions and occurrence.
Moreover, the use of dynamic gates also provides additional complexity to the calculation
Machines 2024, 12, 294 10 of 28

in that the location of the event and the type and position of the gates must be considered
to calculate the reliability of the component. For instance, some of the drawbacks with the
Cri and F-V is the possibility of overlooking or overemphasising faults, which might give
rise to high reliability or low reliability. The cut set approach used to determine criticality
in F-V method could give rise to false high reliability depending on the connection of the
events to the top gate, especially when using non dynamic gates, as it tends to consider
only the probability of occurrence against sequences and dependencies. On the other hand,
Bir IMs measure the increase in probability of the top event due to the occurrence of an
event A, which lies on the critical path of a likely failure event, as presented in Equation (2).
Equation (3) solves for local or sub-system level component criticality.

∂y( p(t))
I B ( i|t) = = h(1i, p(t)) − h(0i, p(t)) (2)
∂pi (t)

where
IB (i|t) = Birnbaum criticality at time t;
h (1i , p(t)) = system reliability when system is functioning.
h (0i , p(t)) = system reliability when system has failed.

liB ( A) = ( P{ X | A} − P{ | X | ∼ A}) (3)

where
liB ( A) = Birnbaum importance measures of for event A;
A = the event whose importance is being measured;
∼ A = the event did occur;
X = top event.

3.2.2. Minimal Cut Set

A minimal cut set (MCS) is the smallest set of events, which, if they all occur, cause
the top event to occur [72]. The qualitative analysis is conducted utilising the DFTA
structure, which relies on the logical properties of the gates. On the other hand, the
quantitative analysis is based on MRO data, including failure rate, MTBF, and frequency.
The quantitative analysis outputs are objective results that include system unreliability,
unavailability, and reliability importance measures which provide critical components
failures. However, the MCS evaluation is based on the output evaluated using the logic
combination of the top event occurrence usually from left to right. Therefore, to obtain
the MCS, the DFTA structure representing each MDG was built based on the functional
relationship and system boundary of the sub-systems on the respective marine MDGs.
Accordingly, the product of the MCS derived from the evaluated fault tree was used
to identify or isolate important failures in a system or sub-system. Moreover, considering
that a single basic event can equally form a cut set depending on the arrangement of the
fault tree, it goes to show how important the qualitative evaluation of fault trees can be.
Figure 3 provides some instance of MCS, such that sub-system 1 having an AND gate fails
only when all the events have occurred however the intermediate OR gate fail when any of
its BEs occur, while in the case of sub-system 2, the occurrence of BE 7 or BE 8 is an MCS.
On the other hand, sub-system 3 has all the BEs as MCS due to the AND top gate. This
highlights potential area where improvements can be achieved either through redesign or
simply altering the system to improve its reliability.
 Machines 2024, 12, 294 11 of 29

Machines 2024, 12, 294 This highlights potential area where improvements can be achieved either through
11 ofrede-
28
sign or simply altering the system to improve its reliability.

Figure 3. Example
Figure of of
3. Example alternative methods
alternative of minimal
methods cutcut
of minimal setset
formation.
formation.

Using
UsingFigure
Figure3 3above,
above,the
thecut
cutsets
sets associated
associated with each of
with each of the
thetop
topevents
eventscan
canbebe
cal-
calculated as follows:
culated as follows:
C1 = {C1,1 , C1,2 , . . . , C1,n2 } = (4)
𝐶1 = {𝐶1,1 , 𝐶1,2 , … , 𝐶1,𝑛2n }= o (4)
C2 = {C2,1 , C2,2 , . . . , C2,n2 } = Ujn2 =1 C2,j (5)
𝑛2
𝐶2 = {𝐶2,1 , 𝐶2,2 , … , 𝐶2,𝑛2 } =n{𝑈𝑗=1 𝐶2,𝑗 }o (5)
nm
Cm = {Cm,1 , Cm,2 , . . . , Cm,nm } = Uj=1 Cm,j (6)
𝑛𝑚
𝐶𝑚 = {𝐶𝑚,1 , 𝐶𝑚,2 , … , 𝐶𝑚,𝑛𝑚 } = {𝑈𝑗=1 𝐶𝑚,𝑗 } (6)
where
where
Cm,j = the basic event in the group of minimal cut set.
𝐶𝑚,𝑗 = the basic
Thereafter, event
the in the group
probabilities of minimal of
of occurrence cutthe
set.top events can be obtained using
Equation (7):
Thereafter, the probabilities
n of occurrence of o the top
n eventsocan be obtained using
[ [ [m
Equation (7): P( TE) = P C1 C2 . . . . Cm =P Ci (7)
i =1

where 𝑃(𝑇𝐸) = 𝑃{𝐶1 ⋃ 𝐶2 … . ⋃ 𝐶𝑚 } = 𝑃{⋃𝑚

𝑖=1 𝐶𝑖 } (7)
P(TE)
where = the probability of the occurrence of the top event;
(Ci . . .m, i ̸= 0) = cumulative summation of all minimal cut set.
P(TE) = the probability of the occurrence of the top event;
(Ci …m, i ≠ 0) = cumulative summation of all minimal cut set.
3.3. ANN Diagnostics
3.3.Machinery failures frequently exhibit early indicators, enabling operators to foresee
ANN Diagnostics
potential problems, which, if addressed, could enhance the availability, reliability, and
Machinery failures frequently exhibit early indicators, enabling operators to foresee
cost-effectiveness of maintenance management. Therefore, gaining comprehension of these
potential problems, which, if addressed, could enhance the availability, reliability, and
indicators can result in enhanced maintenance and operational procedures. A neural
cost-effectiveness
network analysis wasofperformed
maintenance management.
in conjunction withTherefore, gaininganalysis
system reliability comprehension
on MDGs of
these indicators can result in enhanced maintenance and operational procedures. A neural
to detect faults associated with system failures. The ANN fault detection model, utilising
network analysis was performed in conjunction with system reliability
FFNN, offers comprehensive insights into significant causes of failures. analysis on MDGs
to detect faults associated with system failures. The ANN fault detection model, utilising
FFNN, offers
Feedforward comprehensive
Neural Network insights into significant causes of failures.
Diagnostics analysis involves recognising patterns in the data that indicates the pres-
Feedforward
ence of variations Neural Network
pointing to a change in the normal health parameters of the system or
machineryDiagnostics analysis
of interest. involvesANN
A supervised recognising patterns
feedforward in thenetwork
neural data thatwasindicates the pres-
implemented
forence of variationsanalysis.
the classification pointingFeedforward
to a change inANNthe is
normal
a timehealth parameters
series algorithm of can
that the system
be used or
formachinery
both functionof interest. A supervised
fitting and ANN feedforward
pattern recognition [73]. Theneural network
feedforward was implemented
networks usually
for single
have the classification
or multilayer analysis.
hidden Feedforward
sigmoid neuronsANN is a time
followed by series
a seriesalgorithm
of outputthat can be
neurons.
used forlayers
Multiple both function
of neurons fitting
withand pattern transfer
nonlinear recognition [73]. The
functions feedforward
enable networks
the network usu-
to learn
ally have
nonlinear single or multilayer
relationships betweenhidden sigmoid
input and outputneurons
vectorsfollowed
[74]. by a series of output neu-
A two-layer
rons. feedforward
Multiple layers network
of neurons withwith sigmoid
nonlinear activation
transfer and SoftMax
functions enable theoutput neu-to
network
rons were
learn adoptedrelationships
nonlinear for the studybetween
based on Equation
input (8). The
and output sigmoid
vectors [74].activation function,
Equation (9), helps to improve the prediction capability of the neurons by adding bias and
non-linearity while the SoftMax activation function, Equation (10), is a probability function
with values between 0 and 1, that determines the similarity of the output to the input class.
Machines 2024, 12, 294 12 of 28

Therefore, highest probability being 1 and vice-versa, both the sigmoid and SoftMax are
used for classification problems, and they help improve the model’s capability [36].
   
M D
yk ( x, w) = σ ∑ j=i wkj h ∑i=1 w ji + w j0 + wk0
(2) (1) (1) (2)
(8)

1
σ( x) = (9)
1 + e− x
exp( ak )
(10)
∑ j exp a j

4. Case Study
Machinery repair and maintenance data, as well as operational health monitoring data,
over a period of six calendar years, were obtained from the power generation system (PGS)
of an offshore patrol vessel (OPV). The PGS is only used for electricity generation onboard
and is the main source of power to both at sea and at the harbour. The system was design
without an emergency MDG; therefore, the four generators are run as main MDGs. These
MDGs are rated at 440 Volts, 60 Hertz 3 Phase with a steady load speed of 1800 RPM. The
MDGs are 12-cylinder V arrangement and turbo charged with direct sea water intercooler.
The main type of fuel type used onboard is low-sulphur fuel oil bunkered via fuel tankers.

4.1. Case Study: Vessel Mission Profile

The vessel selected for this research belongs to the Centenary Class OPVs of the
Nigerian Navy. The vessel is mainly engaged in patrol duties typically lasting 3–4 weeks at
sea and 2 weeks at harbour. In addition to normal patrols, the vessel can conduct search
and rescue operations, oil spill clean-up, and helicopter recovery and launching. Therefore,
the MDGs play a central role in the vessel’s operation safety and efficiency, especially
during rescue and helicopter operations. As such, the MDGs play a pivotal in the vessel’s
overall ability to achieve its operational objectives. Consequently, it is a standard practice
in the Navy to run on parallel MDGs while transiting through a channel or in areas with
restricted traffic and during close-up situations; hence, this explains the accrued long hours
of operation of the MDGs.
It is therefore clear that the MDGs were in high demand, making the PGS the most
critical system onboard. Overall, the MDGs have an average monthly running hour of
about 160 h per generator. It therefore becomes important to ensure their availability while
Machines 2024, 12, 294 efficiently putting in place a maintenance strategy that considers the environment. Figure
13 of 4
29
presents utility obtained from the PGS to some major users onboard.

Figure4.4.Power
Figure Powergeneration
generationsystem
systemutility
utilityservice
serviceconsumers.
consumers.

4.2.
4.2.Case
CaseStudy:
Study:Assumptions
Assumptionsand
andLimitations
Limitations
Some
Some necessary assumptionswere
necessary assumptions weremade
madeto toenable
enableaamore
moregeneralised
generalisedimplementa-
implementa-
tion
tion of the methodology on the collected data. In particular, thereliability
of the methodology on the collected data. In particular, the reliabilityanalysis
analysiswas
was
conducted using failure rate obtained from MRO of the MDGs over a period of 78 calendar
months; hence, the analysis assumes equal operational time distribution among the
MDGs. Therefore, failure rates for individual MDGs were assumed to be constant withing
the period, while multiple components such as the injector nozzles, valves and tappets
were assumed to have identical failure rates. In this regard, these components were model
Machines 2024, 12, 294 13 of 28

conducted using failure rate obtained from MRO of the MDGs over a period of 78 calendar
months; hence, the analysis assumes equal operational time distribution among the MDGs.
Therefore, failure rates for individual MDGs were assumed to be constant withing the
period, while multiple components such as the injector nozzles, valves and tappets were
assumed to have identical failure rates. In this regard, these components were model as re-
peated faults not as individual distinct failures. The quality of maintenance and repair was
assumed to be the same irrespective of whether the component was repaired or replaced.
As stated earlier, data collected include MRO and sensor data on machinery health
parameters. The MRO was used to generate failure rate for individual MDGs, while the
sensor data provided information for machinery diagnostic analysis. Some other details
extracted from MRO were on maintenance or repair action taken and possible causes of
certain failures; however, these data lack details on routine services and the extent of work
performed. Therefore, based on the data collected, the analysis did not explicitly consider
issues such as human factors, quality of spares, and environmental impacts on machinery
performance or fault. Additionally, performance degradation to establish machinery life
was difficult to establish. In this regard, the research only considered machinery health
parameters for faut identification in relation to component reliability analysis limited to the
case study data.

4.3. Case Study Data Presentation

The case study data obtained from individual MDGs provide details on failure rates,
date failure occurred, and action taken as regards either repair or replacement, while the
machinery health data contain information on temperature, pressure, and power output.
On the other hand, the data did not provide evidence on performance degradation on the
MDGs due to the frequency of repair and replacement that are often not clearly captured
by the maintenance crew. Consequently, the case study covers mainly reliability and fault
identification due to data constraints. In this regard, using the generated data, system
reliability and criticality measures are obtained from the DFTA results. The failure rate data
obtained from the maintenance records covering about six calendar years were used as
input for the DFTA analysis; summary of the failure rates for 4 MDGs is shown Table 1.
Using the failure rate data, the unreliability in the system as well as the machinery
were established, while identifying the most critical component in the system or machinery.
In addition, the MCS function of the DFTA is relevant in analysing the failure path and
possible ways of mitigating them. Furthermore, the output from the DFTA tool, namely,
component criticality and cut set, are going to form a very important input for the BBN
analysis while the ANN uses parametric data for fault detection analysis. Overall, the
MDGs have an average monthly running hours of about 160 h per generator; consequently,
it is important to ensure their availability while efficiently putting in place a maintenance
strategy that ensures reduced emission from both primary and secondary.

Table 1. Component failure rate per 10,000/hours for the case study MDGs.

Components Frequency
Failure Type Action Taken MDG1 MDG2 MDG3 MDG4
Turbo charger Black smoke Replaced, Repaired 8 10 12 12
1. Replaced
Oil leakage 2. Cleaned and zinc anode 16 18 15 16
Lub oil cooler
replaced
External leakage 10 8 8 12
Oil cooler valve failed Remove/repaired 1 1 2 1
Machines 2024, 12, 294 14 of 28

Table 1. Cont.

Components Frequency
Failure Type Action Taken MDG1 MDG2 MDG3 MDG4
1.Liner, O-ring replaced 1 × (A1 & A2) 3 ×
(G1 & G3) (A2, liner) 2 × (A2
1. Oil leakage 20 19 21
2. Cylinder replaced (G3 & head) 1 × (A3 &B 2
Cylinder head 2. Fresh water leakage
G2) replaced gasket (G3) gskt)
from A2 exhaust
Guide bushing 20 14 20 20
3. Unable to start
O-ring 28 32 23 23
Holding bolts 18 17 17 16
Cylinder 1.Scuffed × 4
Replaced 11 12 11 12
jacket/sleeve 2. Cracked × 2
Rings Replaced 12 13 13 14
Piston
cooling/crown 8 13 15 14
bent 7 9 8 9
ConRod
Gudgeon pin 8 6 8 6
failed Replaced 8 8 9 11
Drive belt
Torn(wear) Replaced 11 5 9 3
1. Replaced bolt and drive
(G1 & G3)
1. Cracked bolts
2. Replaced bolt, pulley,
Mech Injector 2. Broken bolts 16 12 12 13
and set injector timing (G1
pump 3. Broken shims
& G2)
3. Replaced shims
Drive defects 22 20 21 24

Case Study Diagnostic Data Analysis

In addition to the MRO data, which provide data for the DFTA, machinery health log
data for the four MDGs was accessed for the purpose of diagnostics analysis. In this regard,
fault identification analysis was carried out using ANN feedforward networks. Machinery
health parameters and their limits are presented in Table 2. However, to improve model
quality, a correlation analysis was conducted to extract the most relevant futures for the
diagnostics analysis; details are provided in Section 5.4.

Table 2. MDG safe health parameters and alarm limit range used for diagnostics.

No Parameter Abbreviation Operating Ranges Alarm

Min Max
1 Lubricating Oil Pressure LoP 0.4 Mpa 0.55 Mpa >0.6
2 Cooling Fresh Water Temperature FWT(A/B) 75 ◦ C 80 ◦ C >85 ◦ C
3 Lubricating Oil Temperature LoT 30 ◦ C 110 ◦ C >120 ◦ C
4 Fresh Water Pressure FWP 0.02 Mpa 0.25 Mpa >0.3
5 Exhaust Gas Temperature EGT(A/B) 220 ◦ C 400 ◦ C >520
6 Engine Speed RPM 1789 RPM 1850 RPM 2052 RPM
7 Power Output KW 0 440 KVA 440 Kva
8 Generator Running Hours HRS ≥2000 h

Therefore, machinery log data collected from the case study ship were used to develop
a diagnostic model. Input and response variables were obtained based on the outcome of
feature engineering, Table 3 presents the response and predictor variables. The predictor
variables represent the most sensitive parameters to the response variable considering the
thermodynamics behaviour of diesel engines.
Machines 2024, 12, 294 15 of 28

Table 3. Diagnostics input Variables.

Variable Abbreviations Remarks

Fresh Water Temperature A-Bank FWTA Response Variable
Fresh Water Temperature B-Bank FWTB Response Variable
Exhaust Gas Temperature B-Bank ETB(EGTB) Response Variable
Exhaust Gas Temperature A-Banks ETA (EGTA) Response Variable
Lubricating Oil Temperature LoT Response Variable
Lubricating Oil Pressure LoP Response Variable
Power Output Kw Predictor Variable

In this regard, the performance of the training process to develop the diagnostic
using FFNN is presented in the below figures. Overall, the predictor variables consist
of 1090 observations from 7 features, from which only one was used, while the response
variable includes 1090 observations from 3 classes. The data was then split into 70% training,
15% validation, and 15% test.
Accordingly, details on the training performance are given in the confusion matrix of
the training, validation, and test partitions. The matrices provide the percentage accuracy
at each level of the model development, as well as a combined or generic matrix for the
3 levels. Overall, each of the classes had an over 80% score in matrices, which suggest a
strong model performance. Furthermore, the combined output of the matrices, as shown
in Figure 5, show a collative score for the classes at 83.7%, which proves the quality of the
Machines 2024, 12, 294 data and the choice of response predictor variables. The overall picture on model accuracy
16 of 29
based on the confusion matrix for the diagnostic training model is presented Figure 5.

Figure
Figure 5.
5. Training confusion matrix.
Training confusion matrix.

5. Results and Discussion

This section discusses the results of the case study findings as developed in the meth-
odology section, which began by presenting the DFTA results of the reliability importance
measures and the minimal cut set (MCS). The MCSs were used as input for the building
maintenance decision support platform using BBN. Thereafter, feature selection and fault
identification analysis using FFNN was presented.
Machines 2024, 12, 294 16 of 28

5. Results and Discussion

This section discusses the results of the case study findings as developed in the method-
ology section, which began by presenting the DFTA results of the reliability importance
measures and the minimal cut set (MCS). The MCSs were used as input for the building
maintenance decision support platform using BBN. Thereafter, feature selection and fault
identification analysis using FFNN was presented.

5.1. Importance Measures

Reliability importance provide significant insight on component that greatly impacts
system failures. The information from the IM can be considered collectively as all subsys-
tems are analysed individually, hence helping to isolate the most critical component to each
system, though with the downside of common cause failures. Therefore, using the infor-
mation from the IM data the maintenance department can help us glean very important
aspects on component prone to certain failures, which can help generate spares holding
based likely usage patten for onboard usage and shore technical stores. Therefore, instead
of considering a single MDG, a collection of all critical components can be made at the fleet
level while at the ship or platform MDGs can be considered individually. This brings to
light the importance of component criticality to faults mapping, as implemented in this
research. Moreover, the overall maintenance platform being developed is geared towards
a more flexible maintenance approach that can be updated based on actual machinery
operating condition either through manual or automated data imputation. Overall, the IMs
in Table 4 represent components that have at least contributed more than 50% of all failures
within the period analysed that are common to all the MDGs in the PGS.

Table 4. Component High Criticality Across MDGs.

Component Criticality
Valve Clearance 0.50
Oil Inlet Hose 0.52
Primary Fuel Lift Pump 0.53
Air Filter 0.55
Primary Fuel Filter 0.56
Pulley Bolts 0.58
Fuel Injection Pump Erratic 0.60
TBC Seal Lub 0.63
Crank Shaft Main Bearing 0.63
Top Cylinder Gasket 0.71
Top Cylinder Bolts 0.73
Fresh Water Heat Exchanger Tubes (fouled) 0.78
Crankshaft Journal Failure 0.82
High Pressure Fuel Pipe 0.82
Cylinder Block Damage 0.88
Lub Oil Pump 0.99
Cylinder Damage 1
Fresh Water Circulation Pump 1
Fresh Water Heat Exchanger Tubes (leakages) 1

The results in the above table provide a list of 19 components that contribute to most
failures. Additionally, the components are equally among the most challenging with regard
to cost and repair time, except for the oil filter, which is easily replaceable. However,
missing the replacement of oil filters could result in undesirable impacts on the condition of
the filter that may lead collapse or constriction. These situations can bring about disruption
in oil supply to other major parts of the MDG. Consequently, this points to additional risk
factors associated to component failures. Therefore, it also becomes important that failures
associated with this component are identified, which can be achieved through the MCS.
Machines 2024, 12, 294 17 of 28

5.2. Minimal Cut Sets

MCSs obtained through DFTA for individual subsystems had dual purposes: the
first was its relevance within the DFTA framework and the second was its use as input to
build the BBN probability analysis. The adoption of the MCSs as inputs for the BBN was
to ensure that the maintenance DSS is consistent or to translate findings of the DFTA to
actionable maintenance decisions. Having this in mind, the MCSs provide very valuable
inputs to the maintenance decision support, especially considering the failure relations,
such as failures triggered by a fault in another system. Hence, the relevance of investigating
interrelationships in component failures becomes critical. Moreover, a lot of failures in
main machineries are as result of primary failures in a performance supporting system.
An example of this is the crank case failure of the MDGs investigated, which exhibited a
concerning low level of reliability due to faults initiated in other system and the crankcase
seemed not designed to handle such faults, particularly faults originating from the lubrica-
tion and freshwater cooling as well as the air distribution system. This makes it difficult to
isolate failure to faults, so the approach in this research is to identify the MCSs, and link
the components and their percentage of occurrence, as shown in Table 5. This way, the
operators can prioritise maintenance and identify spare parts shortages, as necessary.

Table 5. Most critical MCSs to MDG availability.

DG1 % DG2 % DG3 % DG4 %

Crankshaft Journal Fuel Injection Pump Crankshaft Journal
49 82 78 HP Fuel Pipe Leakages 85
Failure Mechanical Failure Failure
FW Heat Exchanger FW Heat Exchanger FW Heat Exchanger
Fuel Filter (1&2) 87 67 94 70
Fouling Fouling Tube Fouling
Tappet Clearance
Rocker Arm and
Sea Chest Blockage 71 (Inlet and Exhaust 82 RW Impeller Damage 84 86
Tappets Clearance
Valves)
Tappet Clearance
Burnt Top Cylinder Turbo Charger Lub
(Inlet and Exhaust 52 86 75 Governor Drive 77
Gasket Failure
Valves)
Cylinder Head Gasket Intercooler Fins
Cylinder Head Sealing 75 Clogged Air Filter 75 72 53
Damage Fouling
Injector Nozzles
Fuel Lift Pump Defects 82 Injector Nozzle Faults 74 72 Turbo Charger 52
Cylinder
Turbo Charger Cylinder Head Gasket
54 Clogged Air Filter 76 Blacked Fuel Filter 76 73
Leakages Damage
Loose Cylinder Head
Cylinder Jacket Cracks 50 Oil Filter 46 Piston Crown Damage 87 64
Bolts
Low Fuel Pressure 63 No Fuel Supply 78 Tappet Clearance 80 Clogged Air Filter 65
Loose Cylinder Head Injector Camshaft
RW Water Impeller 84 Defective Fuel Pump 82 68 54
Bolts Failure

Overall, MCSs capture fault formations based on the qualitative structure of the DFTA,
thus providing a more structured understanding of failure and fault relationships. For
instance, in Table 5, heat exchanger fouling or scaling remains one of the dominant faults,
which can be linked to reduced cooling efficiency and possibly failures due to overheating.
Similarly, failures related to the top cylinder gasket can be associated with overheating
problems due to sea water or freshwater cooling problems. Therefore, we can establish the
relationship between a fault and failure based on the percentage influence of the cut set.
Failures do not always result in components becoming critical to maintenance, but once
they are, these kinds of critical faults can be well planned for, either by adding more repair
or inspection or by offering backup systems. Emphasising these kinds of measures could
improve monitoring and quick intervention. Moreover, the ability to efficiently identify
faults and their possible courses can help address the challenges associated with extended
downtime, the cost of repairs, and repair capability, which are major concerns for operators.
Machines 2024, 12, 294 18 of 28

5.3. BBN Results

The BBN model for each of the four MDGs was created in two stages. The first stage
involved analysing the availability of the subsystem, while the second stage was developing
the maintenance decision support system (DSS). The analysis of component availability was
conducted using data from MCSs for components that contribute more than 50% chance to
the top events occurrence as identified in the DFTA. The CPT tables were filled with data
derived from the MCS percentage rates obtained from the DFTA and the corresponding
influence of those faults on the components. The BBN also offers the ability to simulate and
quantify the influence of CCF on various parts and systems. Thus, minimising redundancy
and improving the evaluation of faults’ consequences.
The BBN availability model’s purpose is to investigate the availability of key com-
ponents in the PGS. This is particularly important because it directly affects the safety of
operators, passengers, equipment, and cargo. As a result, the model investigated various
failure types and their influence on component availability and MDG. We obtained the
components used in the model from DFTA MCS and determined their probability of failure
based on the collected MRO data. This input was used to determine the availability of
individual MDGs, as presented in Table 6. The overall MDG availability is determined by
subsystem availability; this explains the low availability of MDG 3 and 4, which seemed to
be influenced by low numbers in fuel and lubricating systems, both of which would have
a significant impact on the MDG’s performance and ability to take significant loads. In
general, subsystem availability serves as a guide for determining the focus of maintenance
efforts. However, to enhance maintenance decision-making, it is necessary to take into
account additional factors that impact on the delivery and quality of maintenance.

Table 6. BBN MDG and component availability.

MDG MDG1 MDG2 MDG3 MDG4

Individual Availability 50% 53% 48% 47%
Subsystem Availability
Cylinder Block 47% 43% 44% 44%
PTO 60% 56% 50% 60%
Cooling 37% 39% 39% 37%
Fuel System 43% 45% 44% 44%
Air Distribution 50% 52% 52% 42%
Lubrication 62% 75% 56% 55%
Inlet and Exhaust 60% 63% 62% 58%
Alternator 59% 52% 59% 57%

The maintenance DSS was developed using two major input sources, namely, the
availability for the BBN and RPN from the FMECA. It is pertinent to note that the FMECA
analysis and results used for the DSS were presented in earlier research by the author.
Therefore, details of the FMECA can be obtained via [75]. The relevance of the FMECA is
in providing subjective inputs regarding operator sentiment on mission critical component
failures to help provide intuition to the overall DSS. Accordingly, four maintenance strategy
options were adopted for developing the maintenance DSS; these are corrective action,
condition monitoring (ConMon), planned maintenance system (PMS), and delay action.
The allocation of an MDG to any of the maintenance strategies is determined by the
cumulative availability of the subsystems measured against RPN with a value between
0–100, as defined in Table 7.
The assigned target value for each of the strategies determines how maintenance
action, planning, and monitoring should be prioritised. This allows for flexibility regarding
the distribution of resources such as personnel, spare parts, logistics, and operational
deployment. Furthermore, the high criticality ranking for ConMon indicates the need for
additional monitoring approaches, which can include the addition of sensors, increased
inspection frequency, or watchkeeping attention. The strategy prioritisation enables place-
Machines 2024, 12, 294 19 of 28

ment of the MDGs in the maintenance DSS as a representation of the MDGs cumulative
sub-system non-availability. Consequently, the higher the MDG score in a certain strategy,
the greater its position within it and the level of maintenance or monitoring required. In
this regard, all the MDGs belong to one or all of the four strategy types based on which the
maintenance department can plan and make projections.

Table 7. Developed maintenance strategy definitions and RPN ranges.

Maintenance RPN Range

Definition
Strategy (0–100)
This is recommended for very high to high mission critical component or faults for
Corrective Action example sea water supply pump impeller, fuel supply pump, automatic voltage regulator 75–100
faults, etc.
Condition This strategy serves as intervention to ensure system availability targeted at component or
55–75
Monitoring failures whose early identification could avert major operational delays.
Planned
The PMS maintenance choices prioritise time dependent component failures with no
Maintenance 35–55
immediate impacts
Machines 2024, 12, 294 to availability repair requirements. 20 of 29
System
Delay action maintenance choice is directed at those components with good resilience or
Delay Action 0–35
sufficient redundancy such that there is little or no danger personnel and system safety.
inspection frequency, or watchkeeping attention. The strategy prioritisation enables place-
ment of the MDGs in the maintenance DSS as a representation of the MDGs cumulative
sub-system non-availability. Consequently, the higher the MDG score in a certain strat-
Figure 6 presents maintenance DSS allocation for the four MDGs, which shows all
egy, the greater its position within it and the level of maintenance or monitoring required.
the MDGs having In some form all
this regard, of the
presence in alltoof
MDGs belong one the four
or all strategy
of the types.
four strategy typesOverall,
based onthe which
results indicate that
thecorrective
maintenance action and ConMon
department can plan and aremake
the projections.
prepared options for all of the
MDGs, with the exception Figureof 6MDG
presents maintenance
1, which DSSConMon
has low allocation values
for the four MDGs,PMS.
but high whichOn shows
the all
other hand, MDG1the and MDGs
MDG2 having
were some
theform
onlyofones
presence in all ofwithin
to appear the four
thestrategy
delaytypes.
actionOverall,
while the
results indicate that corrective action and ConMon are the prepared options for all of the
showing elevated values for both corrective action and ConMon. These findings indicate
MDGs, with the exception of MDG 1, which has low ConMon values but high PMS. On
that the two generators require
the other hand,aMDG1
significant
and MDG2 amount
were of themaintenance.
only ones to appearAdditionally, MDG1
within the delay action
has a corrective action
while rate of approximately
showing elevated values for 54%,bothwhile MDG2
corrective actionhas
anda ConMon.
rate of about 58%
These findings
in ConMon. However, MDG3
indicate that theand
twoMDG4 have
generators similar
require levels of
a significant priority,
amount except in Addition-
of maintenance. PMS,
where the numbers for MDG4 are significantly higher than those for MDG3.MDG2
ally, MDG1 has a corrective action rate of approximately 54%, while has a rate of
One possible
about 58% in ConMon. However, MDG3 and MDG4 have similar levels of priority, except
explanation for this is that MDG1 and MDG2 are situated in the same engine room, just
in PMS, where the numbers for MDG4 are significantly higher than those for MDG3. One
like MDG3 and MDG4. possible Consequently,
explanation for owing to the
this is that MDG1 utilisation
and MDG2 ofare
common
situated resources
in the samelikeengine
sea chest, ventilation, fuel line, and local stress such as vibration, the
room, just like MDG3 and MDG4. Consequently, owing to the utilisation of generators arecommon
prone re-
to experiencing similar
sourcespatterns
like sea of failure.
chest, Though
ventilation, fuelsome
line, andof local
thesestress
findings
such aswere not evident
vibration, the genera-
tors are prone to experiencing similar patterns of failure.
to the operators before this research, they were nevertheless consistent with comparable Though some of these findings
were not evident to the operators before this research, they were nevertheless consistent
research findings in the shipping industry.
with comparable research findings in the shipping industry.

Figure 6. Maintenance DSS choice for all MDGs.

Figure 6. Maintenance DSS choice for all MDGs.
Generally, the analysis indicates how each of the MDGs aligned to a certain mainte-
nance strategy regime as a reflection of its reliability or failure pattens. Therefore, the main
determinants are to identify what influences faults leading to failures or high level of un-
availability in subsystems. Similarly, the ability to repair the MDGs when failed is an im-
portant consideration to make, hence the need to carry out further investigation regarding
MDG health parameters to conduct machine learning analysis for fault detection.
Machines 2024, 12, 294 20 of 28

Generally, the analysis indicates how each of the MDGs aligned to a certain main-
tenance strategy regime as a reflection of its reliability or failure pattens. Therefore, the
main determinants are to identify what influences faults leading to failures or high level
of unavailability in subsystems. Similarly, the ability to repair the MDGs when failed is
an important consideration to make, hence the need to carry out further investigation
regarding MDG health parameters to conduct machine learning analysis for fault detection.

5.4. Fault Detection

The fault identification analysis using a feed-forward ANN with two layers based
on sigmoid and SoftMax activation functions was used for the classification analysis. A
time-series data of about 1000 data points was used; the data were divided into three
categories: 70% for training, 15% for validation, and 15% for testing. The anomaly data
labels were used for the initial training using MDG 1; this was executed to develop a single
model for all four MDGs. Hence, the labelled fault data using temperature fault codes
(Temp) was used for fault detection, which contains three fault classes. Accordingly, overall
training data utilised 20% of the data from all MDGS added to MDG1 data before splitting,
as highlighted earlier. To enable the fault detection model, data threshold values were
established, as shown in Table 8.

Table 8. Limits of data labels used for fault identification.

Temperature Ranges
Fault Fault Identity Fault Parameter Operating State
(◦ C)
Normal Lubricating Oil
Normal Temperature NTM 80–110 Normal
Temperature
High Lubricating Oil
High Temperature HTM 110–115 Abnormal
Temperature
Overheating OVH Engine Overheating Max 120 Fault/Failure

For the initial training using MDG 1, we used the anomaly data labels presented in
Table 9, aiming to develop a single model for all four MDGs. Hence, the labelled fault data
using temperature fault codes (Temp) was used for fault detection, which contains three
fault classes. Accordingly, overall training data utilised 20% of the data from all MDGS
added to MDG1 data before splitting, as highlighted earlier.

Table 9. Fault labels.

RPM LoP FWTA FWTB LoT FWP EGTA EGTB RH KW Fault Temp
1800 0.458 72.9 75.4 90 0.067 332.1 319.5 5234 115 Normal NML
1800 0.465 72.8 75.3 89.9 0.068 335.3 323.9 5235 120 Normal NML
1800 0.59 72.01 74.06 89.3 0.068 329.5 316.7 5236 115 Fault HTM
1800 0.53 70.7 73.2 87.6 0.068 310.2 29.4 5262 100 Normal NML
1800 0.58 78 80.68 96.2 0.066 366.1 355.9 5294 150 Abnormal OVH
1801 0.58 75.8 78.6 94.6 0.067 360.4 351.7 5298 140 Abnormal HTM
1800 0.504 76.2 79.1 95 0.067 361.2 353.1 5299 140 Normal HTM
1800 0.58 78.6 78.7 94.5 0.067 359.1 350.1 5300 140 Abnormal HTM
1800 0.502 76.2 79.1 94.8 0.067 358.3 351 5201 140 Normal HTM
1800 0.499 75.8 78.8 95.6 0.067 360.1 353.7 5302 150 Normal NML
1800 0.488 77.8 80.5 96.1 0.066 374.2 363.3 5203 140 Normal OVH
1800 0.498 77.3 80 95.8 0.066 364.3 354.3 5204 150 Normal HTM

Therefore, having established the fault labels for the diagnostic analysis, it is necessary
to identify which of the variables can be used a good health indicator. Accordingly, a
correlations analysis was conducted for feature selection using the R-values with strongest
correlation; this was also backed by personal experience and established industry practice
Machines 2024, 12, 294 21 of 28

Machines 2024, 12, 294 22 of 29

Machines 2024, 12, 294 for diagnostic feature selection on MDG. Hence, this provides the foundation for29the
22 of
variable used for the diagnostic analysis, as seen in Figure 7.

Figure7.
Figure
Figure 7.7.Correlation
CorrelationMatrix.
Correlation Matrix.
Matrix.

InInview
viewof ofthe
the above,
above, fault identification
identificationlimits
limitswere
weredetermined
determined bybyutilising
utilising thethe
thresholdvalues
threshold
threshold valuesspecified
values specifiedinin
specified in Table
Table
Table 8. 8.
8. Therefore,
Therefore,
Therefore, aamodel
a model was
model was trained
trained
was using
using
trained data
data
using fromfrom
data MDG
from
MDG
1MDG
and was 1 and was partitioned
partitioned as as
earlier earlier explained.
explained. Among Among
the the
six six response
response
1 and was partitioned as earlier explained. Among the six response variables, LoT variables,
variables, LoT LoT
had a
had
good a good
R-score R-score
against against
KW, whichKW, which
is the is the
predictorpredictor variable,
variable, as can
had a good R-score against KW, which is the predictor variable, as can be seen in Figureas
be can
seenbeinseen in
Figure Figure
7. In this
7.7.InInthis
regard, this
theregard,
original
regard, themodel
the original
original model was
training
model training was
wasconducted
conducted
training with LoT
conducted with
as LoT
with the asas
thethe
response
LoT response
variable
response var-and
var-
iable
KW as and
theKWKW as the
predictor. predictor.
The modelThe The model
accuracy accuracy based on the true positive rate (TPR)
iable and as the predictor. model based on the
accuracy basedtrue
onpositive
the truerate (TPR)rate
positive and(TPR)
false
and false
negative negative rate (FNR) classes isinpresented
Figure 8. in Figurethe8. Overall, theachieved
model has
and falserate (FNR) rate
negative classes is presented
(FNR) classes is presented Overall,
in Figure 8.model has
Overall, the modelmore has
achieved
than 97% more than
accuracy 97% accuracy
between the true between
and the trueclasses
predicted and predicted
in classes the
identifying in identifying
three classes,
achieved more than 97% accuracy between the true and predicted classes in identifying
the three
namely classes, namely NML, HTM, and OVH.
the threeNML, HTM,
classes, and NML,
namely OVH. HTM, and OVH.

Figure 8. Training model accuracy.

Figure The
8. Training model
results model accuracy.
of the original
Figure 8. Training accuracy.model using MDG 1 are shown in Figure 9. The fault iden-
tification scatter plot indicates that the MDG was operating with relatively elevated
The results of the original model using MDG 1 are shown in Figure 9. The fault iden-
tification scatter plot indicates that the MDG was operating with relatively elevated
Machines 2024, 12, 294 22 of 28

Machines 2024, 12, 294 23 of 29

Machines 2024, 12, 294 23 of 29
The results of the original model using MDG 1 are shown in Figure 9. The fault
identification scatter plot indicates that the MDG was operating with relatively elevated
temperatures, usually
temperatures, usually above
usually above
above 8080 degrees.
80 degrees.Similarly,
Similarly,an anindication
indicationofofabnormally
abnormallyoperating
operating
temperatures,
condition is the region of degrees.situation
overheating Similarly,around
an indication
60 kw of
to abnormally
100 kw, as operating in
indicated
condition is
condition is the
the region
region of
of overheating
overheating situation
situation around
around 6060 kw
kw toto 100
100 kw,
kw, as
as indicated
indicated in in
Figure
Figure 9.
9. Moreover,
Moreover, the selection
the selection of
of the
theLoT
LoTas asaapredictor
predictor isispremised
premised ononitsitsfidelity
fidelity
toto
Figure 9.performance
indicate Moreover, the selection of the
degradation, as LoT as
well as the
a predictor
overall is premised
health of on its combustion
internal fidelity to
indicate performance
indicate performance degradation,
degradation, asas well
well as
as the
the overall
overall health
health of of internal
internal combustion
combustion
engines
engines especially
especially within
within the
the normal operating
normal operating rangeas
range aswell
wellininoverload
overload conditions.
conditions.
engines especially within the normal operating range as well in overload conditions.

Figure 9. Original
Figure Original Data
Data Model.
Model.
Figure 9.
9. Data Model.

Following the
Following
Following the original
the original dataset
original dataset diagnostics
datasetdiagnostics outcome
diagnosticsoutcome using
outcomeusing the
usingthe LoT,
theLoT, the
LoT,the model
themodel
model was
was
was
tested
tested using
using the
the held
held partitioned
partitioned data
data for
for the
the test.
test. The
The test
test model
model prediction
prediction of
tested using the held partitioned data for the test. The test model prediction of the fault of
the the fault
fault class
class performed
class performed
performed well above
well above
well above 96%,
96%,96%, as presented
presented
as presented
as in Figure
in Figure
in Figure 10.
10. 10. Going
Going byby
Going by the
thethe test
test model
model
test model accu-
accuracy,
accu-
racy,
the
racy, the
model model
seemed
the model seemed to generalise
to generalise
seemed well
well well
to generalise with
withwith
majormajor fault
faultfault
major classes
classes and
andand
classes is notis not very
veryvery
is not different
different from
different
from was
what
from whatpresented
what was presented
was presented
in theinin the original
original
original
the modelmodel prediction.
prediction.
model prediction.

Figure 10. Test model accuracy.

Figure 10.
Figure 10. Test
Test model
model accuracy.
accuracy.
The model was
The was deployed onon the combined
combined data ofof the MDGs,
MDGs, and thethe result
result remains
remains
The model was deployed
deployed onthethe combineddatadata ofthe
the MDGs,and and the result remains
consistent with both the
consistent with both the validation and test data results earlier presented. The prediction
consistent the validation
validationandandtest
testdata
dataresults
resultsearlier
earlierpresented.
presented.TheTheprediction
prediction
model shows
model shows more
more fault
fault detections
detections with
with improved
improved accuracy,
accuracy, mainly
mainly because
because ofof improved
improved
model fault detections with improved accuracy, mainly because of improved
data presented
data presented during
during both
both the
the test
test and
and validation
validation phases.
phases. Figure
Figure 11
11 shows
shows predictions
predictions
Machines 2024, 12, 294 23 of 28
Machines 2024, 12, 294 24 of 29
Machines 2024, 12, 294 24 of 29

data presented during both the test and validation phases. Figure 11 shows predictions
using
using the
the combined data from
combined data from the
the MDGs;
MDGs;thetheplot
plotindicate
indicatechallenges
challengeswithwiththe
the MDGs
MDGs
using the temperature
operating combined data from thearound
especially MDGs;>100
the plot
kw. indicate
This challenges
information with theimportant
provides MDGs
operating temperature especially around >100 kw. This information provides important
operating temperature
confirmation on the especially temperature
theoperational
operational around >100 kw. This information
disparity between theprovides
OEM important
and operator.
confirmation on temperature disparity between the OEM and operator. In
confirmation
In on the operational temperature disparity between the OEM and operator. In
this regard, additional investigations were carried out with other response variable to pro- to
this regard, additional investigations were carried out with other response variable
this regard,
provide someadditional
level of investigations
validation were carried
between the out with
different other response
response variables.variable to pro-
vide some level of validation between the different response variables.
vide some level of validation between the different response variables.

Figure 11.
11. Model prediction
prediction with
with LoT.
Figure 11. Model
Figure Model prediction with LoT.
LoT.

Figure 12 is a diagnostic
Figure diagnostic plot
plot using
usingETA
ETAwith
withthetheKWKWpredictor
predictorvariable.
variable.The
Thescatter
scatter
Figure 12 is a diagnostic plot using ETA with the KW predictor variable. The scatter
positionsare
positions aregenerally
generallysimilar
similarto
tothat
thatofofthe
theLoT,
LoT,ininparticular
particularwith
withHTM
HTMpositions
positionsdominate
domi-
positions are generally similar to that of the LoT, in particular with HTM positions domi-
nate around
around 100–150
100–150 KWKW output.
output. OnOn thethe otherhand,
other hand, thereseem
seem to bebeno
nohealth
healthindications
nate around 100–150 KW output. On the other hand,there
there seem to be no health indications
indications
beyond 180 KW, suggesting
beyond suggesting that the MDGs hardly produce up toto200 KW. Moreover, the
beyond 180 KW, suggesting that the MDGs hardly produce up to 200 KW. Moreover, thethe
that the MDGs hardly produce up 200 KW. Moreover,
only
only few scatter positions above 200 KW are that of OVH and possibly misclassified NML.
only few positions above
few scatter positions above200200KWKWarearethat
thatofofOVH
OVHand and possibly
possibly misclassified
misclassified NML.
NML.

Figure 12. Combined MDG data prediction using ETA.

Figure 12.
Figure 12. Combined
Combined MDG
MDG data
data prediction
predictionusing
usingETA.
ETA.
Overall, scatter plots using the LoT and ETA diagnostic models, as presented in this
Overall, scatter
Overall, scatter plots
plots using
using the LoT
LoTand
andETA
thedetection. ETAdiagnostic
diagnosticmodels,
models,asaspresented
presentedinin
this
this
research, have good fitness for fault Though each of the plots presented varying
research,
research, have good
have good fitness for
fitness for fault detection. Though each of the plots presented varying
levels of intensity regarding thefault
faultdetection. Though
classes and commoneach of the that
regions plotsthe
presented
MDGs tendvarying
to
levels of
levels of intensity
intensity regarding
regarding the
the fault classes
fault and
classes common
and common regions thatthat
regions the MDGs
the tendtend
MDGs to
deviate from an acceptable or normal operating range, the two plots clearly identified the
deviate from an acceptable or normal operating range, the two plots clearly identified the
Machines 2024, 12, 294 24 of 28

to deviate from an acceptable or normal operating range, the two plots clearly identified
the HTM positions at nearly the same points in the plot, though with varying clarity.
Accordingly, the findings confirm the low reliability and criticality of both the cooling and
lubricating systems, thereby establishing a reasonable ground for relatively MCS outcomes
as well as concerns regarding the cooling system and cylinder head bolts. Therefore, the
link between component reliability and fault can be established based on the fact that the
MDGs operate most of the time at relatively high temperatures, above the normal operating
range but below alarm levels.
Furthermore, the component reliability analysis identified components such as the sea
chest, FW heat exchanger, tappet clearance, and turbo charger as the most critical to MDG
reliability. In the MDG, all of the stated components can be associated with temperature
increases and performance degradation. However, since the MDGs are run most of the
time at the harbour or when the ship is alongside, this could explain the reliability issues
with the sea chest and due to objects in the water.
The fault identification analysis has played a crucial role in pinpointing failure con-
ditions that are associated with reliability outcomes. These results would significantly
reinforce the reliability analysis outcome but would also raise the issue of the performance
of the MDGs above 50% of the rated output. In fact, the LoT diagnostics analysis and ETA
fault identification analysis have both identified the maximum loads the MDGs overheat.
In general, using both ETA and LoT, the safe working load for the MDGs was between
160 kw and 200 kw, suggesting the MDGs could have been overrated. Therefore, with this
finding, it is safe to say that the MDGs are overrated; hence, the operator can decide to take
this up with the OEM.

6. Conclusions
This paper presents a hybrid approach that combines dynamic fault tree analysis
DFTA, BBNs, and ANNs to analyse system reliability and machinery health degradation.
The methodology introduced in the paper serves as a valuable tool for operators and
asset managers to enhance maintenance strategies by offering a detailed assessment of
component criticality and related faults based on collected data. As the shipping industry
faces increasing scrutiny and regulations related to emissions control, efficient maintenance
approaches that ensure equipment availability and system reliability while minimizing
emissions become crucial. Traditional maintenance methods may not suffice to address
these complex challenges. In this regard, to tackle these issues, the study highlights the
potential of sensor technologies, reliability, and data analysis in offering efficient solutions
for ship maintenance. The hybrid methodology employing DFTA, BBNs, and ANNs was
developed to determine component criticality and identify faults.
A case study was conducted on the power generation system of an OPV, which
identified critical components and their corresponding faults. These faults had a significant
impact on vessel availability and overall system performance. The utilisation of BBN
analysis, which includes inputs from the DFTA criticality and MSC, enabled the analysis of
sub-system availability and the development of a maintenance decision support system
(DSS). The study also assessed the accessibility of MDGs by analysing their individual
components and suggesting strategies for upkeep. Overall, corrective action and ConMon
were the most preferred options among most MDGs, although their availability differed.
Ultimately, an artificial neural network (ANN) was utilised to detect faults, with LoT and
ETA serving as a response and KW as predictor. The results indicated that the majority of
malfunctions took place when the power output was slightly above 150 KW and associated
to high temperature faults. Overall, the main issues identified are overheating and low
load carrying capacity of MDGs, contributing to frequent failures affecting key components
such as heat exchanger tubes, crankshaft journals, cylinder head bolts and freshwater
circulation pump.
Machines 2024, 12, 294 25 of 28

Author Contributions: Conceptualisation, A.A.D. and I.L.; methodology, A.A.D. and I.L.; software,
A.A.D.; validation, A.A.D.; formal analysis, A.A.D. and I.L.; investigation, I.L. and A.A.D.; resources,
A.A.D.; data curation, A.A.D.; writing—original draft preparation, A.A.D.; writing—review and edit-
ing, A.A.D. and I.L.; visualisation, A.A.D.; supervision, I.L.; project administration, A.A.D.; funding
acquisition, A.A.D. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by the Petroleum Technology Development Fund (PTDF) of
Nigeria, grant number 17PHD178.
Data Availability Statement: The data presented in this study are available on request from the
corresponding author.
Acknowledgments: The author wants to acknowledge and appreciate the support and scholarship
funding received from the Federal Government of Nigeria via the Petroleum Technology Development
Fund (PTDF) and the Nigerian Navy for providing access to ship data used in the research.
Conflicts of Interest: The funders as well as the data providers had no role in the design of the study,
in the collection, analysis, or interpretation of data, in the writing of the manuscript, or in the decision
to publish the results.

Abbreviations

ABS(NS) American Bureau of Shipping (Nautical System) ISM code International Safety Management
ANN Artificial Neural Network MCS Minimal Cut Set
BBN Bayesian Belief Network MTTF Mean Time to Failure
BE Basic Event MTBF Mean Time Between Failure
BSI British Standards Institution MDT Mean Down Time
CBM Condition-Based Maintenance MRO Maintenance, Repair, and Overhaul
CMMS Computerised Maintenance Management System NASA National Aeronautics and Space Administration
CPT Conditional Probability Table ISO International Standard Organisation
RPN Risk Priority Number OEM Original Equipment Manufacturer
OREDA Offshore and Onshore Reliability Data OPV Offshore Patrol Vessel
MDG Marine Diesel Generator PAND Priority-AND
ETA Event Tree Analysis DFTA Dynamic Fault Tree Analysis
DSS Decision Support System PMS Planned Maintenance System
GHG Green House Gas RCM Reliability-Centred Maintenance
CII Carbon Intensity Index UN United nations
EEXI Energy Efficiency Existing Ship Index RPM Revolution Per Minute
SOM Self-Organising Maps LoP Lubricating Oil Pressure
FFNN Feedforward Neural Network FWT(A/B) Fresh Water Temperature (Bank A/B)
FDEP Functional Dependency LoT Lubricating Oil Temperature
FMEA Failure Mode and Effect Analysis FWP Fresh Water Pressure
FMECA Failure Mode Effect and Criticality Analysis EGT(A/B) Exhaust Gas Temperature (Bank A/B)
FTA Fault Tree Analysis RH Running Hours
IM Importance Measure KW Kilo Watt
IMO International Maritime Organisation HRS Hours

References
1. UNCTAD. Review of Maritime Transport 2022; UNCTAD: Geneva, Switzerland, 2022.
2. MEPC.328(76); 2021 Revised MARPOL Annex VI. IMO: London, UK, 2021.
3. Stopford, M. Maritime Economics, 3rd ed.; Routledge: London, UK, 2009.
4. Horvath, S.; Fasihi, M.; Breyer, C. Techno-economic analysis of a decarbonized shipping sector: Technology suggestions for a fleet
in 2030 and 2040. Energy Convers. Manag. 2018, 164, 230–241. [CrossRef]
5. Kang, Y.-J.; Noh, Y.; Jang, M.-S.; Park, S.; Kim, J.-T. Hierarchical level fault detection and diagnosis of ship engine systems. Expert
Syst. Appl. 2023, 213, 118814. [CrossRef]
6. Karatuğ, Ç.; Arslanoğlu, Y.; Soares, C.G. Design of a decision support system to achieve condition-based maintenance in ship
machinery systems. Ocean Eng. 2023, 281, 114611. [CrossRef]
7. Bouman, E.A.; Lindstad, E.; Rialland, A.I.; Strømman, A.H. State-of-the-art technologies, measures, and potential for reducing
GHG emissions from shipping—A review. Transp. Res. Part D Transp. Environ. 2017, 52, 408–421. [CrossRef]
Machines 2024, 12, 294 26 of 28

8. Li, Z.; Zhang, D.; Han, B.; Wan, C. Risk and reliability analysis for maritime autonomous surface ship: A bibliometric review of
literature from 2015 to 2022. Accid. Anal. Prev. 2023, 187, 107090. [CrossRef] [PubMed]
9. BS EN 13306:2010; Maintenance-Maintenance Terminology. BSI: London, UK, 2010.
10. ISO. Condition monitoring and Diagnostics of Machines—General Guidelines; ISO: Geneva, Switzerland, 2018.
11. Jimenez, V.J.; Bouhmala, N.; Gausdal, A.H. Developing a predictive maintenance model for vessel machinery. J. Ocean Eng. Sci.
2020, 5, 358–386. [CrossRef]
12. Hirzinger, B.; Nackenhorst, U. Efficient model-correction-based reliability analysis of uncertain dynamical systems. Acta Mechanica
2023, 235, 1419–1436. [CrossRef]
13. Zhang, P.; Gao, Z.; Cao, L.; Dong, F.; Zou, Y.; Wang, K.; Zhang, Y.; Sun, P. Marine Systems and Equipment Prognostics and Health
Management: A Systematic Review from Health Condition Monitoring to Maintenance Strategy. Machines 2022, 10, 72. [CrossRef]
14. BahooToroody, A.; Abaei, M.M.; Banda, O.V.; Montewka, J.; Kujala, P. On reliability assessment of ship machinery system in
different autonomy degree; A Bayesian-based approach. Ocean Eng. 2022, 254, 111252. [CrossRef]
15. Aizpurua, J.I.; Knutsen, K.E.; Heimdal, M.; Vanem, E. Integrated machine learning and probabilistic degradation approach for
vessel electric motor prognostics. Ocean Eng. 2023, 275, 114153. [CrossRef]
16. Wang, R.; Chen, H.; Guan, C. A self-supervised contrastive learning framework with the nearest neighbors matching for the fault
diagnosis of marine machinery. Ocean Eng. 2023, 270, 113437. [CrossRef]
17. Cort, S.J. Moving from Planned to Condition Based Maintenance; IMDEX: Singapore, 2017.
18. Li, N.; Gebraeel, N.; Lei, Y.; Bian, L.; Si, X. Remaining useful life prediction of machinery under time-varying operating conditions
based on a two-factor state-space model. Reliab. Eng. Syst. Saf. 2019, 186, 88–100. [CrossRef]
19. Lazakis, I.; Gkerekos, C.; Theotokatos, G. Investigating an SVM-driven, one-class approach to estimating ship systems condition.
Ships Offshore Struct. 2018, 14, 432–441. [CrossRef]
20. Cipollini, F.; Oneto, L.; Coraddu, A.; Murphy, A.J.; Anguita, D. Condition-Based Maintenance of Naval Propulsion Systems with
supervised Data Analysis. Ocean Eng. 2018, 149, 268–278. [CrossRef]
21. Cipollini, F.; Oneto, L.; Coraddu, A.; Murphy, A.J.; Anguita, D. Condition-based maintenance of naval propulsion systems: Data
analysis with minimal feedback. Reliab. Eng. Syst. Saf. 2018, 177, 12–23. [CrossRef]
22. Velasco-Gallego, C.; Lazakis, I. RADIS: A real-time anomaly detection intelligent system for fault diagnosis of marine machinery.
Expert Syst. Appl. 2022, 204, 117634. [CrossRef]
23. Galar, D.; Kumar, U. Sensors and Data Acquisition. In eMaintenance; Academic Press: Cambridge, MA, USA, 2017; pp. 1–72.
24. Chen, R.; Zhang, C.; Wang, S.; Zio, E.; Dui, H.; Zhang, Y. Importance measures for critical components in complex system based
on Copula Hierarchical Bayesian Network. Reliab. Eng. Syst. Saf. 2023, 230, 108883. [CrossRef]
25. Velasco-Gallego, C.; Lazakis, I. A real-time data-driven framework for the identification of steady states of marine machinery.
Appl. Ocean Res. 2022, 121, 103052. [CrossRef]
26. Gkerekos, C.; Lazakis, I.; Theotokatos, G. Machine learning models for predicting ship main engine Fuel Oil Consumption: A
comparative study. Ocean Eng. 2019, 188, 106282. [CrossRef]
27. Cheliotis, M.; Lazakis, I.; Theotokatos, G. Machine learning and data-driven fault detection for ship systems operations. Ocean
Eng. 2020, 216, 107968. [CrossRef]
28. Bousdekis, A.; Papageorgiou, N.; Magoutas, B.; Apostolou, D.; Mentzas, G. Enabling condition-based maintenance decisions with
proactive event-driven computing. Comput. Ind. 2018, 100, 173–183. [CrossRef]
29. Galar, D.; Kumar, U. Data Collection. In eMaintenance; Academic Press: Cambridge, MA, USA, 2017; pp. 73–128.
30. Fuller, A.; Fan, Z.; Day, C.; Barlow, C. Digital Twin: Enabling Technologies, Challenges and Open Research. IEEE Access 2020, 8,
108952–108971. [CrossRef]
31. Mihanovic, L.R.; Belamaric, G.P. Use of new informtion technologies in the maintenance of ship systems. Pomorstvo 2016, 30,
38–44. [CrossRef]
32. Patil, C.; Theotokatos, G. Comparative Analysis of Data-Driven Models for Marine Engine In-Cylinder Pressure Prediction.
Machines 2023, 11, 926. [CrossRef]
33. Wu, S.-J.; Gebraeel, N.; Lawley, M.A.; Yih, Y. A Neural Network Integrated Decision Support System for Condition-Based Optimal
Predictive Maintenance Policy. IEEE Trans. Syst. Man Cybern. Part A Syst. Hum. 2007, 37, 226–236. [CrossRef]
34. Raptodimos, Y.; Lazakis, I. Using artificial neural network-self-organising map for data clustering of marine engine condition
monitoring applications. Ships Offshore Struct. 2018, 13, 649–656. [CrossRef]
35. Cheliotis, M.; Gkerekos, C.; Lazakis, I.; Theotokatos, G. A novel data condition and performance hybrid imputation method for
energy efficient operations of marine systems. Ocean Eng. 2019, 188, 106220. [CrossRef]
36. Gurney, K. An Introduction to Neural Networks; Taylor and Francis Group: London, UK, 1997.
37. Farag, Y.B.; Ölçer, A.I. The development of a ship performance model in varying operating conditions based on ANN and
regression techniques. Ocean Eng. 2020, 198, 106972. [CrossRef]
38. Stetco, A.; Dinmohammadi, F.; Zhao, X.; Robu, V.; Flynn, D.; Barnes, M.; Keane, J.; Nenadic, G. Machine learning methods for
wind turbine condition monitoring: A review. Renew. Energy 2019, 133, 620–635. [CrossRef]
39. Velasco-Gallego, C.; Lazakis, I. Real-time data-driven missing data imputation for short-term sensor data of marine systems. A
comparative study. Ocean Eng. 2020, 218, 108261. [CrossRef]
40. DNV. Data Collection Infrastructure; DNV: Bærum, Norway, 2020.
Machines 2024, 12, 294 27 of 28

41. ISO-19848-2018(E); Ships and Marine Technology—Standard Data for Shipboard Machinery and Equipment. ISO: Geneva,
Switzerland, 2018.
42. Lazakis, I.; Theotokatos, G.; Barltrop, N.; Dikis, K. Inspection Capabilities for Enhanced Ship Safety. In Proceedings of the 16th
International Congress of the International Maritime Association of the Mediterranean, Pula, Croatia, 21–24 September 2015;
pp. 775–780.
43. Eriksen, S.; Utne, I.B.; Lützen, M. An RCM approach for assessing reliability challenges and maintenance needs of unmanned
cargo ships. Reliab. Eng. Syst. Saf. 2021, 210, 107550. [CrossRef]
44. ISO19847; Ships and Marine Technology-Shipboard Data Servers to Share Field Data at Sea. ISO: Geneva, Switzerland, 2018.
45. ISO/IEC 19845; Information Technology-Universal Business Language Version 2.1. ISO/IEC: Geneva, Switzerland, 2015.
46. Raptodimos, Y. Collection and Analysis of Data for Ship Condition Monitoring aiming at Enhanced Reliability and Safety. In
Proceedings of the 26th International Ocean and Polar Engineering Conference, Rhodes, Greece, 26 June–2 July 2016; Volume 4.
47. NAVSEA. RCM Handbook 2007; NAVSEA: Washington, DC, USA, 2007.
48. Jakkula, B.; Mandela, G.R.; N, M.C.S. Reliability block diagram (RBD) and fault tree analysis (FTA) approaches for estimation of
system reliability and availability—A case study. Int. J. Qual. Reliab. Manag. 2020, 38, 682–703. [CrossRef]
49. Chybowski, L.; Gawdzińska, K.; Laskowski, R. Assessing the Unreliability of Systems during the Early Operation Period of a
Ship—A Case Study. J. Mar. Sci. Eng. 2019, 7, 213. [CrossRef]
50. NASA. Fault Tree handbook with Aerospace Applications; NASA: Washington, DC, USA, 2002.
51. Byun, S.; Papaelias, M.; Márquez, F.P.G.; Lee, D. Fault-Tree-Analysis-Based Health Monitoring for Autonomous Underwater
Vehicle. J. Mar. Sci. Eng. 2022, 10, 1855. [CrossRef]
52. Relex. Reliability: Practitioner’s Guide; Relex Software Corpoaration: Greensburg, Pennsylvania, 2003; p. 350.
53. Boudali, H.; Crouzen, P.; Stoelinga, M. Dynamic Fault Tree analysis using Input Output Interactive Markov Chains. In Proceedings
of the 37th Annual IEEE/IFIP International Conference on Dependable Systems and Networks (DSN’07), Edinburgh, UK, 25–28
June 2007.
54. Kabir, S. An overview of fault tree analysis and its application in model based dependability analysis. Expert Syst. Appl. 2017, 77,
114–135. [CrossRef]
55. Zhou, S.; Ye, L.; Xiong, S.; Xiang, J. Reliability analysis of dynamic fault trees with Priority-AND gates based on irrelevance
coverage model. Reliab. Eng. Syst. Saf. 2022, 224, 108553. [CrossRef]
56. Lazakis, I.; Dikis, K.; Michala, A.L.; Theotokatos, G. Advanced Ship Systems Condition Monitoring for Enhanced Inspection,
Maintenance and Decision Making in Ship Operations. Transp. Res. Procedia 2016, 14, 1679–1688. [CrossRef]
57. Čepin, M.; Mavko, B. A dynamic fault tree. Reliab. Eng. Syst. Saf. 2002, 75, 83–91. [CrossRef]
58. Turan, O.; Lazakis, I.; Judah, S. Establishing the Optimum Vessel Maintenance Approach Based on System Reliability and
Criticality Analysis. In Managing Reliability & Maintainability in the Maritime Industry; RINA SaRS Conference, London, UK, 25
January 2012; Royal Institution of Naval Architects: London, UK, 2012; pp. 59–70.
59. Ruijters, E.; Stoelinga, M. Fault tree analysis: A survey of the state-of-the-art in modeling, analysis and tools. Comput. Sci. Rev.
2015, 15–16, 29–62.
60. Kabir, S.; Papadopoulos, Y. Applications of Bayesian networks and Petri nets in safety, reliability, and risk assessments: A review.
Saf. Sci. 2019, 115, 154–175. [CrossRef]
61. Jun, H.-B.; Kim, D. A Bayesian network-based approach for fault analysis. Expert Syst. Appl. 2017, 81, 332–348. [CrossRef]
62. Canbulat, O.; Aymelek, M.; Turan, O.; Boulougouris, E. An application of BBNs on the integrated energy efficiency of ship–port
interface: A dry bulk shipping case. Marit. Policy Manag. 2019, 46, 845–865. [CrossRef]
63. BayesFusion. GeNIe Modeler; BayesFusion: Pittsburgh, PA, USA, 2020; p. 614.
64. Li, H.; Soares, C.G.; Huang, H.-Z. Reliability analysis of a floating offshore wind turbine using Bayesian Networks. Ocean Eng.
2020, 217, 107827. [CrossRef]
65. Don, M.G.; Khan, F. Dynamic process fault detection and diagnosis based on a combined approach of hidden Markov and
Bayesian network model. Chem. Eng. Sci. 2019, 201, 82–96.
66. Khakzad, N.; Khan, F.; Amyotte, P. Safety analysis in process facilities: Comparison of fault tree and Bayesian network approaches.
Reliab. Eng. Syst. Saf. 2011, 96, 925–932. [CrossRef]
67. Kampitsis, D.; Panagiotidou, S. A Bayesian condition-based maintenance and monitoring policy with variable sampling intervals.
Reliab. Eng. Syst. Saf. 2022, 218, 108159. [CrossRef]
68. Ahn, S.I.; Kurt, R.E. Application of a CREAM based framework to assess human reliability in emergency response to engine room
fires on ships. Ocean Eng. 2020, 216, 108078. [CrossRef]
69. BahooToroody, A.; Abaei, M.M.; Banda, O.V.; Kujala, P.; De Carlo, F.; Abbassi, R. Prognostic health management of repairable
ship systems through different autonomy degree; From current condition to fully autonomous ship. Reliab. Eng. Syst. Saf. 2022,
221, 108355. [CrossRef]
70. New, C.; Rob, G. RCM In The Royal Navy—Developing A Risk Based Policy For Integrating Safety And Maintenance Management.
In Proceedings of the Managing Reliability and Maintainability in the Maritime Industry, London, UK, 25–26 January 2012.
71. Daya, A.A.; Lazakis, I. Application of Artifical Neural Network and Dynamic Fault Tree Analysis to Enhance Reliability in Predicitive Ship
Machinery Health Condintion Monitoring; GMO-SHIPMAR: Istanbul, Turkey, 2021.
Machines 2024, 12, 294 28 of 28

72. Windchill, P. PTC Windchill Quality Solutions Getting Started Guide, 11th ed.; PTC Winchill Quality Solutions: Boston, MA, USA,
2015; p. 161.
73. Sazli, M.H. A brief review of feed-forward neural networks. Commun. Fac. Sci. Univ. Ank. 2006, 50, 11–17. [CrossRef]
74. Sarwar, U.; Muhammad, M.B.; Karim, Z.A.A. Time Series Method for Machine Performance Prediction Using Condition
Monitoring Data. In Proceedings of the 2014 International Conference on Computer, Communications, and Control Technology
(I4CT), Langkawi, Malaysia, 2–4 September 2014.
75. Daya, A.A.; Lazakis, I. Developing an advanced reliability analysis framework for marine systems operations and maintenance.
Ocean Eng. 2023, 272, 113766. [CrossRef]

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