See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/370002994

Dynamic Failure Analysis of Ship Energy Systems Using an Adaptive Machine

Learning Formalism

Article · March 2023

DOI: 10.47852/bonviewJCCE3202491

CITATIONS READS

0 49

5 authors, including:

Azubuike John Chuku Sidum Adumene

Rivers State University of Science and Technology Memorial University of Newfoundland
8 PUBLICATIONS   2 CITATIONS    51 PUBLICATIONS   306 CITATIONS   

SEE PROFILE SEE PROFILE

C.U. Orji Kombo Theophilus-Johnson

Rivers State University of Science and Technology Rivers State University of Science and Technology
25 PUBLICATIONS   56 CITATIONS    5 PUBLICATIONS   1 CITATION   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Gulf of Guinea Offshore Anti-Piracy Systems Analysis View project

Special Issue "Risk, Safety and Reliability Assessment in the Maritime and Offshore Industries" View project

All content following this page was uploaded by Samson Nitonye on 14 April 2023.

The user has requested enhancement of the downloaded file.

 Received: 31 October 2022 | Revised: 19 January 2023 | Accepted: 20 February 2023 | Published online: 8 March 2023

Journal of Computational and Cognitive Engineering

RESEARCH ARTICLE 2023, Vol. 00(00) 1–10
DOI: 10.47852/bonviewJCCE3202491

Dynamic Failure Analysis of Ship

Energy Systems Using an Adaptive
Machine Learning Formalism
Azubuike John Chuku1, Sidum Adumene2,* , Charles Ugochukwu Orji1, Kombo Theophilus Johnson1 and
Samson Nitonye1
1
Department of Marine Engineering, Rivers State University, Nigeria
2
School of Ocean Technology, Memorial University of Newfoundland, Canada

Abstract: The criticality of shipping operations in global trade requires a comprehensive understanding of its sustainability. This depends on
the integrity/performance of the ship structure and vital systems, such as the ship propulsion engine. The current research paper presents the
application of an adaptive machine learning formalism, the Bayesian network, for failure assessment of a ship propulsion engine considering
nonlinear and nonsequential failure interactions. The model captures critical failure influencing factors and their complex interactions to
predict the failure probability of the ship energy system. Sensitivity and uncertainty analysis was carried out to establish the degree of
influence of vital failure influencing factors as they affect the ship propulsion engine’s reliability and the associated uncertainty in the
prior data processing. The model is tested on the propulsion engine of an ocean going vessel to forecast the likelihood of failure based
on the logical dependencies among failure causative factors. Two scenarios were analyzed based on canonical probabilistic algorithms,
and the results show that upon evidence on the three critical failure modes, the ship propulsion engine failure likelihood increased by
11.8%, 8.2%, and 9.4%, respectively. The model shows an adaptive/dynamic capability to capture new failure information and update
the system’s failure probability. The proposed approach provides a condition monitoring tool and early warning guide for integrity
management of critical ship energy systems.
Keywords: ship propulsion engine, dynamic failure, machine learning formalism, energy systems, Bayesian network

1. Introduction importance. Marine diesel engine failure is an operational failure

that might result in damage both to the ship and crew members on
The tremendous gains from maritime transport have served as a board (Cicek & Celik, 2013).
catalyst for civilization and intercontinental trade’s lifeline. Though The complexity of the marine operations and energy system
numerous regional and international regulatory bodies have safety performance could affect their failure prediction and risk. There
conventions in place, major failures/accidents continue to be exist multidimensional dependency and interdependency among
reported every year for marine vessels (Adumene et al., 2022; its subsystems during operation. This results in nonlinear and
Aziz et al., 2019). The ship propulsion power system is a critical nonsequential failure interactions among subsystems. A better
system of the ship structure. As such, failure for any reason may understanding of these failure modes and their interrelationship is
result in substantial economic loss or loss of life. Heavy-duty crucial for the system’s time-dependent failure prediction.
diesel power plants and gas turbine plants are widely used in ship
Moreover, the system failure’s stochastic nature could require a
propulsion (Lion et al., 2019). Recently, diesel or diesel hybrid
dynamic and robust probabilistic tool that could capture and
engines have been used in marine propulsion systems to generate
integrate all causative factors for a reliable failure forecast. The
mechanical energy from thermal forces. Most vessel types, such
reviewed literature (Arzaghi et al., 2020; Aziz et al., 2019; Carroll
as small boats and recreational yachts, use a hybrid energy system
for propulsion. The common propulsion engine in the shipping et al., 2015; Hatti, 2020; Ossai et al., 2016; Zahraee et al., 2016)
industry is the diesel engine; the diesel engine converts thermal has demonstrated the application of probabilistic tools in energy
energy to mechanical energy used to exert kinetic force. It is systems reliability and failure analysis. However, they are limited
general knowledge that voyages from Port of departure to Port of to the dynamic modeling of ship energy systems under complex
destination should be completed on time; as such, reducing the dependencies.
potential for failure in the propulsion system is of great The Bayesian network (BN) is a machine learning (ML) tool
that captures the conditional dependencies among random
*Corresponding author: Sidum Adumene, School of Ocean Technology, variables to analyze a given phenomenon. Its inferential reasoning
Memorial University of Newfoundland, Canada. Email: [email protected] is based on Bayes’ theorem. The BN model has shown promise

© The Author(s) 2023. Published by BON VIEW PUBLISHING PTE. LTD. This is an open access article under the CC BY License (https://creativecommons.org/
licenses/by/4.0/).

01
Journal of Computational and Cognitive Engineering Vol. 00 Iss. 00 2023

for complex system failure modeling under uncertainty (Khakzad provides great insight into the system’s failure influential factors;
et al., 2013). It adopts a logical acyclic directed graph to represent however, they are limited to quantitatively predicting the failure
the cause–effect analysis of a given phenomenon. The reviewed likelihood and the complex interactions among basic elements of
literature (e.g., Cicek & Celik, 2013; Faturachman et al., 2018; the plant.
Golub Medvešek et al., 2014), modeled the ship energy reliability To further capture the logical dependency among the
based on qualitative and semiquantitative methods. These methods subsystems, a quantitative approach, such as the fault tree analysis
are limited to capturing the interactions among the subsystems. (FTA), has been demonstrated for marine system failure
This reveals the weaknesses of the existing model for ship energy assessment (Aziz et al., 2019; Golub Medvešek et al., 2014;
system modeling under complex failure modes interactions. Kabir, 2017; Nitonye et al., 2017). The FTA is a method that
Moreover, the interaction and complex dependence among critical captures possible failure root causes through a logical relationship.
failure influential factors in a dynamic ocean environment have It provides an in-depth structure for failure prediction that results
not been captured for ship propulsion engine failure. It is essential in loss of system integrity (Márquez et al., 2016). Typically, FTA
to adopt an algorithm that could capture the dynamic interactions is adapted to predict the top event (failure event) failure
among failure modes for a reliable system failure prediction under probability based on the associated system failure modes and
uncertainty. basic events. Furthermore, the basic failure events are not
The current study presents an approach that integrates the exhaustive; they are dependent on the analyst’s knowledge and
failure modes with the ML formalism (the BN) to model the interaction among the various subsystems (Ta et al., 2017). It is
nonlinearity and dependencies among the ship energy systems’ used to deduce the interaction between intermediate events; these
failure contributing factors. The model is able to i) apply a data interactions could be in the form of a combination of elements
updating algorithm that will update the failure data for real-time with the use of Boolean (“AND/OR”) logic gates. Márquez et al.
risk modeling of ship operations and ii) capture the complex (2016) used the failure modes technique to identify critical failure
dependencies among failure influential factors for a dynamic influential factors and the logical dependencies of marine wind
failure assessment framework. The energy system’s critical failure energy systems. The model captures the structural, wear,
modes are identified and classified based on their level of electrical, and mechanical causative factors as their interplay
importance and functionality. The essential predisposing factors affects the plant’s performance. Ta et al. (2017) recently
(causative factors) are described based on a logical framework to employed the FTA for the marine propulsion reliability analysis.
estimate their probability of occurrence. The formulated The intent is to predict the likelihood of propulsion system failure,
frameworks are mapped into the BN to dynamically capture the considering the basic causative factors. The model could predict
effects of the subsystem interactions and stochasticity on the ship the failure of the propulsion system upon the availability of the
power plant’s failure characteristic. This is intended to capture the failure characteristics (probability) of the basic events for the
variability and uncertainty in the failure initiating parameters and period under consideration. However, the Boolean logic gates
the effects of the dynamic ocean environments. The approach is present a static feature that limits the fault tree’s application,
demonstrated on a ship propulsion engine. It logically represents especially for a real-life ocean operation/scenarios where the
the various causative factors to capture their interdependency and failure characteristic is stochastic and dynamic.
stochasticity for a robust failure assessment. The likely impact of Adumene & Okoro (2020) applied a stochastic Markovian
the subsystems’ performance based on a sensitivity analysis was process for marine energy system reliability analysis. The authors
established to identify their vulnerability path for critical decision captured the stochasticity associated with the failure of the energy
making. systems through a time- and space-dependent framework. The
probability of failure over time was predicted, given the failure
2. Marine Energy Systems Failure Assessment rate as transition intensity based on the Markovian assumption.
Further application of artificial intelligence in ship systems and
The system’s reliability defines the performance characteristic offshore energy systems analysis has been demonstrated by the
of an engineering system for the period of operation. This can be researchers (Arzaghi et al., 2020; Carroll et al., 2015; Hatti, 2020;
expressed in terms of the failure rate or hazard function for a Ossai et al., 2016; Zahraee et al., 2016).
defined period. Failure of the marine energy system could be The increasing need for digitization in the marine industry and
catastrophic in extremely harsh ocean environments. Their failure data mining have created more opportunities to develop advanced
modes could classify the dependency/functionality of the data science and ML techniques for the maritime. Recent works
subsystems that characterized the overall system performance. that explore the application of ML for marine energy system
Several researchers have proposed various failure modes forecast and other aspects of the maritime operations are detailed
assessment techniques for marine energy systems in the open in the referenced literature (Cheliotis et al., 2020; Kim et al.,
sources (Banks et al., 2001; Faturachman et al., 2018; Hadiya, 2021; Peng et al., 2020; Planakis et al., 2022; Tay et al., 2021;
2011; Kang et al., 2017; Lau et al., 2012; Leimeister & Kolios, Uyanık et al., 2020). For instance, Peng et al. (2020) applied the
2018; Pfaffel et al., 2017). For example, Failure Mode and Effect ML formalism to predict ships’ energy consumption at Port. The
Analysis (FMEA), HAZOP, FMECA, and What If Analysis are authors adopt 15 modeling features, consisting of inherent ship
common qualitative approaches widely adopted to analyze the property and external Port features. The result shows that
severity, detection, and occurrence of failure events. Recently, deadweight tonnage, facilities efficiency, net tonnage, and actual
Kang et al. (2017) and Pfaffel et al. (2017) proposed the ship weight critically affect the amount of energy consumption of
application of FMEA and Correlations-FMEA for marine energy ships at Port. Similarly, Tay et al. (2021) highlighted the pros and
system failure assessment in the ocean environment. The authors cons of applying ML techniques, such as artificial neural
identify five fundamental subsystems that define the functionality networks, hidden Markov model, and Bayesian inference for ship
of the energy system. The critical subsystems could include the energy efficiency prediction. However, there is no conclusive
structural support system, electrical, turbine generator, study on the application of these models for marine energy system
transmission, and auxiliary systems. The qualitative approach failure prediction.

02
Journal of Computational and Cognitive Engineering Vol. 00 Iss. 00 2023

The dynamic and variability in marine energy systems’ failure tool, such as FMEA, could be adopted to classify the subsystems’
variables require a probabilistic and ML formalism that is dynamic and level of importance and their role in the entire plant’s failure during
adaptive, such as the BN. The BN could capture the multidimensional operation. The interaction among these basic failure influential
complexity in the energy system configuration to predict its effect on factors is examined to develop a logical framework for the various
the overall plant performance and failure. Such a benefit is needed to failure modes.
develop a robust failure-based framework for critical ship energy
systems during operations. It could also provide a dynamic tool that 3.2 Logical interactive framework for failure modes
updates the failure probability upon the availability of new evidence
and/or the effect of maintenance on the system performance. The failure modes of marine propulsion engine failure study are
presented in Figure 2. Figure 2 shows a schematic of the logical
3. Proposed Methodology and Its Application framework of the significant failure modes. It captures the interaction
among the identified failure modes for the study scenario. Critical
This section presents the proposed integrated ML algorithm for factors that influence the system and their basic predisposing elements are
the failure analysis of marine energy systems. The algorithm is further studied for a robust failure framework for the ship propulsion
summarized in Figure 1, and the subsequent sections describe the engine systems. However, a comprehensive logical structure is
methodological procedure. presented in section 4 for the demonstration of the proposed approach.

3.1. System performance and failure modes 3.3. Mapping of logical framework into the ML
formalism
The marine system of study is defined and classified based on its
performance/functionality. This is followed by identifying the Figure 2 shows the developed logical representation of the vital
subsystems and their failure mode. The failure modes assessment failure modes. This is elaborated and mapped into the BN structure in

Figure 1
Algorithm for dynamic failure assessment methodology

Develop a The logical

Define the Assess the logical framework is
marine energy system failure interactive mapped into the
system and its modes & their framework AI (BN)
characteristic functionality among failure formalism
performance modes

Simulate the Predict the

formulated BN system failure
structure by and develop a
inputting the failure profile
basic events for decision
probabilities making

Figure 2
A schematic of the main and intermediate failure modes

03
Journal of Computational and Cognitive Engineering Vol. 00 Iss. 00 2023

section 4. The logical structure used by FTA mapped the basic maximizing the likelihood of the model (Beretta et al., 2017;
events, intermediate events, and top events as the root nodes, Chickering, 2002; Daly et al., 2011). In the constraint-based
intermediate nodes, and pivot node of the BN formalism, approach, the Incremental Association Markov Blanket (IAMB)
respectively. More information on the mapping procedures algorithm and PC algorithm have proven effective in learning the
adopted for the research analysis can be found in the referenced BN model, while the score-based approach adopted the maximum
literature (Bobbio et al., 2001; Khakzad et al., 2013). For more likelihood estimation algorithm. For instance, the learning process
information on the use of BN formalism, interested readers are based on the likelihood maximizer L of a set of failure data D for
referred to the work of Daly et al. (2011). a given model G can be expressed by equation (3) (Beretta et al.,
The BN is a directed probabilistic dependence graph that 2017).
captures or represents uncertain knowledge in ML (Bobbio et al., Q
2001; Pearl, 1988). It is applicable for discrete and continuous LLðG; DÞ ¼ PðdjGÞ (3)
random variables modeling of a given phenomenon. The BN dɛD

modeling can identify both the qualitative and quantitative

The common algorithm for the score-based approach is the Bayesian
topologies, which in most cases are based on the d-separation
information criteria (BIC). The algorithm adopts the log-likelihood
notion and direct dependence among random variables. Its
and regularization terms to define the model structure from the
advantages range from its graphical decoding of conditional
failure/observed dataset, D: It can be expressed by equation (4).
independence to joint probability representation among random
variables. The BN formalism presents modeling flexibility that log m
accommodates various statistical dependencies and dynamic BIC ðG; DÞ ¼ LLðG; DÞ  dimðGÞ (4)
2
interdependencies among failure causative factors. The BN
formalism shows great merits over the fault tree technique. where D denotes the dataset, m indicates the number of samples, and
For a given set of basic random variables ðX1 ; X2 ; . . . . . . :; Xn Þ, dimðGÞ is the number of parameters in the BN model.
the joint conditional probability distribution, PðU Þ, gives The formulated ML formalism is learned based on the score-
based approach for the basic events (root nodes) probability of
Y
n
occurrence and the conditional probabilities that describe the
PðU Þ ¼ P½X1 ; X2 ; X3 ; . . . ::; Xn  ¼ P½ðXi jParent ðXi ÞÞ (1) dependencies. This is quantified as a conditional probability table
i¼1
(CPT). The root nodes’ probabilities are numerical values from
where PðU Þ is the joint probability distribution, and Parent ðXi Þ is the historical failure data, expert opinions, and literature. This defined
parent of the set of random variables, Xi . the initial state of the systems and the belief about their
The BN consists of various algorithms for inference and performance. For this analysis, n-array variables are adopted to
computing of the posterior probability distribution on a set of depict the component behavior in a real-life case. This captures
query variables designated as Q, for a given evidence called E the possible presence of multifailure modes and their probabilities
(i.e., PðQjEÞÞ. Figure 3 represents a conventional BN structure for between 0 and 1. The CPT is defined based on the Noisy gates’
random variables Y1 ; Y2 ; Y3 ; Y4 . Also, the BN formalism can update principles for the likely multifailure. When changes occur due to
the probabilities for new information or change in the characteristic the system’s environmental dynamics and stochasticity, the BN
variables. For example, given an observed variable Y3 to be in state e, can be updated based on the new information. For the parametric
the joint probability distribution can be updated based on the Bayes’ learning of the BN structure, the nodes are categorized into
theorem, as shown in equation (2). success and failure. The probability of failure of the pivot node
(ship propulsion engine failure) can be simulated for multiple
PðY1 ; Y2 ; Y4 jeÞ scenarios for the period under consideration.
PðY1 ; Y2 ; Y4 jeÞ ¼ P (2) Furthermore, the sensitivity analysis examines the degree of
Y1 ;Y2 ;Y4 P ðY1 ; Y2 ; Y4 ; eÞ
influence of the basic failure influencing factors on the pivot
node. The essence is to identify the critical input parameters that
significantly impact the energy system’s failure profile and
3.4. Structural and parametric learning of BN provide safety barriers/measures. It also propagates the uncertainty
structure for decision making for the random input dataset to the output of the model. The
variance reduction technique is adopted for the sensitivity analysis
Structural learning methods of the BN from the dataset have
of the BN model’s output for this research. The technique is based
been demonstrated in the literature. The two common learning
on the expected reduction in the variance of the characterized
methods include the score-based and constraint-based approaches.
value Q given the evidence R. It is mathematically represented by
These methods considered learning the dataset based on the
equation (5) (Pearl, 1988; Shabarchin & Tesfamariam, 2016).
conditional independence relationship among the data points and
X  
V ðqjr Þ ¼ Pðqjr Þ Xq  EðQjr Þ 2 (5)
q
Figure 3
Schematic of a conventional BN structure where q is the state of the query node Q; r is the state of varying node
R, Pðqjr Þ is the conditional probability of q when node R is given to
Y1 be in state r, Xq is the numerical value corresponding to state q, and
Y3 EðQjr Þ is the expected real value of Q due to a finding of the state r in
node R.
The quantification and model learning of the failure data present
Y2 Y4 some uncertainty due to randomness and errors. To evaluate the
associated error and uncertainty in the dataset due to the

04
Journal of Computational and Cognitive Engineering Vol. 00 Iss. 00 2023

randomness for the study phenomenon, equations (6) and (7) are trip failure, and failure to start modes. The interdependency
adopted demonstrated in parametric learning shows that one failure mode’s
occurrence causes the ship engine’s failure. However, in many
S real-case scenarios, there exist different failure states or latent
SEx̄ ¼ pffiffiffi (6) failure/faults that may influence the CPT formation based on the
n
belief of the degree of interactions/dependency among failure
tS causative factors. To capture this scenario, the canonical noisy-OR
UD̄ ¼  pffiffiffi (7) gate is introduced, as shown in Figure 5.
n
Figure 5 shows the result of the ship propulsion engine failure
where S is the deviation of the mean, n is the total number of ele- profile under noisy-OR configuration. The essence is to capture
ments, SEx̄ is the standard error of the sample mean, UD̄ is the uncer- scenarios when latent failure/faults in the system affects the
tainty randomness in the dataset, and t is the t-test statistic at 95% system’s functionality but does not result in total failure. This
confidence level. represents cases where there exists a nonsequential failure
For the current study, multiple scenarios are developed to process as well (Adedigba et al., 2016). The noisy-OR model
demonstrate different failure profiles and vulnerability paths under demonstrates the ability to consider a reconfigured system
various operational decision-making scenarios. All BN structures performance in the presence of latent failure/faults. As shown
in this study are modeled in the GeNIe™ software environment. in Figure 5, the likelihood of failure for the propulsion engine
fails start node in comparison with the result in Figure 4
3.5. Methodology application increases by 55.6% under the noisy-OR gate configuration.
While that of the main engine failure and the propulsion engine
The proposed approach is demonstrated with a case study of trip failure nodes probabilities decreases by 89.9% and 20%,
ship propulsion engine failure (Aziz et al., 2019; Kum & Sahin, respectively.
2015; Lion et al., 2019). This provides validation and To further analyze the failure state’s effect of the failure modes
demonstrates the applicability of the ML formalism in ship energy on the ship propulsion engine failure likelihood, the evidence is
system failure analysis. As presented in the referenced literature placed on the “Yes” state for the main engine node, propulsion
(Aziz et al., 2019), the logical framework is used to apply ML engine trip node, and propulsion engine fail to start node.
formalism in this research. The probabilistic failure dataset Learning the structure under evidence shows that the ship
presented by (Kum & Sahin 2015) was used for the proposed propulsion engine failure probability increases by 11.8%, 8.2%,
model’s structural learning (training), while Table 1 shows the and 9.4%, respectively. The work demonstrates the BN’s adaptive
prior probabilities dataset of the principal events for the model test nature to capture any change(s) in belief or information on the
and validation (Aziz et al., 2019). state of failure of the influential critical factors to predict the pivot
The proposed approach’s computational procedure is applied to node’s performance state. This reflects the real case scenarios
the case study based on the system’s data, as shown in Table 1. The where common cause failure and latent faults are predominant.
results outcome is shown in section 4. Sensitivity analysis examines the influence of the most critical
failure influential parameters on the pivot node failure state. In this
4. Results and Discussion case, the nodes’ failure data variance is evaluated based on the
method presented in Shabarchin & Tesfamariam (2016). The
The current research objective is to demonstrate the application sensitivity analysis result is shown in Figure 6.
of the ML formalism (the BN), for failure prediction of the ship The normalized percentage of the ship propulsion engine failure
propulsion energy system under uncertainty. This is shown by i) parameters, as shown in Figure 6, indicates that the pneumatic start
utilizing the data updating algorithm in the ML formalism to system failures show more significant effects. This represents over
update the failure data for real-time failure modeling in ship 15% influence on the frequency of failure of the ship propulsion
operations and ii) capturing the complex dependencies among plant. This is followed by the air system failure that results in a
failure influential factors for dynamic failure assessment. The BN lack of starting air for the demonstrated case study. The drive
captures the various failure influencing factors dynamically and (shaft) line functionality is crucial in the sustainable operation of
predicts their dependencies’ effect on the ship power plant’s the power plant. Its frequency of failure accounted for 4.7% of the
probability of failure. The critical failure modes were represented total system failure. The degree of inference of the other
and evaluated based on the prior probabilities, as shown in parameters can be deduced from Figure 6. Understanding the
section 3. The ML formalism captures both the parametric and effect of the variations in the vital basic elements on ship
structural learning of the parametric interactions among crucial propulsion engine performance will aid parametric integrity
failure influential parameters of the ship energy system. The management, especially in harsh ocean conditions, where
parameter learning result of the BN structure is shown in environmental constraints pose additional load effect on the ship
Figure 4. The result shows that for the given failure data, as propulsion. The energy need in harsh weather ship operation and
indicated in Table 1, the ship propulsion engine failure’s the interdependencies among the critical subcomponents could
likelihood is 0.9999 based on the Boolean logic (0 1]. This is create degradation and fatigue-related failure of the system. This
adopted for the CPT formation based on a deterministic logical- provides an initial validation to the model application.
OR gate analysis. It implies that for a state of affirmation on the The modeling of the uncertainty in the dataset and the proposed
subsystems’ failure states, the learning of the structure confirmed models’ epistemic uncertainty is based on the test statistic. The
that the ship propulsion engine would fail. This is due to epistemic is inherent in the assigned/learned CPT of the BN
the logical interrelationship of the main engine failure, propulsion structure and is propagated to the output. The analysis result

05
06

Journal of Computational and Cognitive Engineering

Table 1
Principal events of ship propulsion engine failure
Main event Main intermediate events Codes Subintermediate events Codes Basic events Codes Basic events probability
Ship propulsion Engine failure Main engine (M/E) failure XA Mechanical failure X1 Scaled or worn valves X001 0.133
Propulsion engine trip XB Drive malfunction X2 Cylinder wear X002 0.128
Propulsion engine fails to start XC Mechanical seizure X3 Bearing overheat X003 0.075
Human failures X4 Loss of lubrication X004 0.349
External failures X5 Stern tube and gearbox fails X005 0.409
Main engine compressor X6 Coupling failure X006 0.017
Main engine control system X7 ECU/hydraulics X007 0.173
Combustion air system failure X8 FIVA and speed governor X008 0.249
Cooling system failure X9 Air compressor failure X009 0.393
Fuel oil system failure X10 Control air supply failure X010 0.209
Lube oil system failure X11 CCU/ACU control X011 0.370
Electrical system failure X12 M/E blower fail X012 0.487
Power outage/blackout X13 M/E Turbocharging fail X013 0.133
Pneumatic start system failures X14 Scaled cooler X014 0.982
Lack of start air X15 Fuel injector failure X015 0.474
Start-up hydraulic pressure X16 Poor quality of fuel X016 0.400
Low flow of lube oil X17 Fuel filter block X017 0.221
Lube oil pressure drop X18 F/O booster fails X018 0.160
Fuel oil supply X19 L/O leakage X019 0.002
Poor fuel oil quality X20 L/O purifier fails X020 0.623

Vol. 00
Low flow of cooling water X21 L/O pump fails X021 0.077
Pneumatic controller failure X22 Pneumatic L/O supply fails X022 0.213
Cylinder mechanics failure X23 Short circuit X023 0.010
M/E sensor failures X024 0.133

Iss. 00 2023
Trip signal malfunction X025 0.503
Service air compressor failure X026 0.632
Start-up hydraulics X027 0.865
Bottled air pressure low X028 0.057
Compressor fails X029 0.300
Cracking of cylinder block X030 0.020
Deadlocking of piston ring X031 0.050
Breakage of piston X032 0.030
cracking of liner X033 0.112
M/E, main engine; CCU, cylinder control unit; ECU, engine control unit; ACU, actuator control unit; F/O, fuel oil; L/O, lube oil.
Journal of Computational and Cognitive Engineering Vol. 00 Iss. 00 2023

Figure 4
Result of the parametric learning of the BN structure

indicates that the standard error associated with the deviation from the maritime industry as a tool for condition monitoring and
the mean is 0:0418. However, the deviation effects for each data integrity management of critical marine energy systems against
point may show some diversity in the errors as predicted. This is total failure.
crucial considering the failure data learning/partitioning process
for prior probability estimation of the various ship engine subsys- 5. Conclusions
tems. The uncertainty due to dataset randomness at a 95% confi-
dence level is 0:0855. The essence is to understand the The current study has shown the application of an adaptive ML
characteristic nature of the input data (prior probability) and the formalism, the BN, for failure analysis of a ship propulsion energy
assigned/learned CPT of the basic influential factors on the model system. The approach explores the vital influencing factors and their
outcome. Upon the prediction of the uncertainty, it is propagated criticality for the overall plant failure prediction. The model captures
into the predicted output. This process of uncertainty propagation the effects of dependencies and interdependencies among failure
or casual error propagation is crucial in the model validity. This is influential factors and predicts the ship energy system’s dynamic
an important measure of the proposed model’s performance in fail- state. It is observed that the pneumatic start system, starting air
ure analysis. The proposed method offers better performance in system, and drive system significantly affect the frequency of
ship energy system modeling in comparison with the works of failure of the plant. The model captures the uncertainty in the
Cicek & Celik (2013), Faturachman et al. (2018), and Golub dataset and propagates through the structural learning process for
Medvešek et al. (2014). The ability to capture multidimensional a reliable output prediction. This provides validation to the model
dependencies enhances the model application in complex system application in ship energy system modeling. The following are
analysis. vital findings of the current research study:
The proposed approach has shown the capacity to capture
influential critical factors and the various failure modes for ship • The applicable model provides a useful tool for dynamic failure
propulsion engine failure analysis. The dynamic and updating analysis of ship energy systems.
strength of the ML formalism is of practical importance for ship • The approach captures the nonlinear and nonsequential inter-
operations in unstable ocean environments. The approach benefits actions among key failure influential factors for ship energy

07
Journal of Computational and Cognitive Engineering Vol. 00 Iss. 00 2023

Figure 5
Parameter learning of the BN structure under noisy-OR gate configuration

Figure 6
Sensitively analysis of influential parameters on the pivot node

system failure prediction and demonstrates it updating/adapting • That the model could propagate the casual error and uncertainty in
capacity upon the availability of new failure information. the prior probability estimation of the basic events to the top event
• The approach could capture the effect of critical systems’ soft failure prediction.
failure and their interaction on the overall plant failure under • The approach captures the effect of the variation in variance
diverse belief systems based on the OR-gate and the noisy-OR through the sensitivity analysis to predict the parametric degree
gate configurations. of influence on the overall failure prediction.

08
Journal of Computational and Cognitive Engineering Vol. 00 Iss. 00 2023

• The maritime industry stands to benefit from the approach’s on-board ship. Safety Science, 51(1), 6–10. https://doi.org/
capability to monitor and manage critical energy systems during 10.1016/j.ssci.2012.06.003
marine operations. This provides a guide/early warning against Daly, R., Shen, Q., & Aitken, S. (2011). Learning Bayesian
total failure of the critical ship systems. networks: Approaches and issues. Knowledge Engineering
Review, 26(2), 99–157. https://doi.org/10.1017/S02698
The application of the adaptive/dynamic model confirms its
88910000251
advantages in failure assessment of ship energy systems under
Faturachman, D., Mustafa, S., Octaviany, F., & Novita, T. D. (2018).
uncertainty. Nevertheless, the proposed approach could be
Failure mode and effects analysis (FMEA) of diesel engine
improved in future study with the inclusion of safety barriers
maritime transportation for ship navigation system
modeling and the integration of loss function technique for
improvement. In the International Conference on Business
economic risks prediction in shipping/marine operations.
Innovation, Entrepreneurship and Engineering.
Golub Medvešek, I., Šoda, J., & Perić, T. (2014). Fault tree analysis
Conflicts of Interest
in the reliability of heavy fuel oil supply. Transactions on
The authors declare that they have no conflicts of interest to this Maritime Science, 3(2), 131–136. https://doi.org/10.7225/
work. toms.v03.n02.004
Hadiya, M. (2011). Case study of offshore wind farm integration to
References offshore oil and gas platforms as an isolated system-system
topologies, Steady State and Dynamic Aspects. Master’s
Adedigba, S. A., Khan, F., & Yang, M. (2016). Dynamic safety thesis, Norwegian University of Science and Technology.
analysis of process systems using nonlinear and non- Hatti, M. (2018). Renewable energy for smart and sustainable cities:
sequential accident model. Chemical Engineering Research Artificial intelligence in renewable energetic systems.
and Design, 111, 169–183. https://doi.org/10.1016/j.cherd. Germany: Springer Nature.
2016.04.013 Kabir, S. (2017). An overview of fault tree analysis and its
Adumene, S., Afenyo, M., Salehi, V., & William, P. (2022). An application in model based dependability analysis. Expert
adaptive model for human factors assessment in maritime Systems with Applications, 77, 114–135. https://doi.org/10.
operations. International Journal of Industrial Ergonomics, 1016/j.eswa.2017.01.058
89, 103293. Kang, J., Sun, L., Sun, H., & Wu, C. (2017). Risk assessment of
Adumene, S., & Okoro, A. (2020). A Markovian reliability approach floating offshore wind turbine based on correlation-FMEA.
for offshore wind energy system analysis in harsh Ocean Engineering, 129(154), 382–388. https://doi.org/10.
environments. Engineering Reports, 2(3), e12128. 1016/j.oceaneng.2016.11.048
Arzaghi, E., Mahdi, M., Abbassi, R., Reilly, M. O., Garaniya, V., & Khakzad, N., Khan, F., & Amyotte, P. (2013). Dynamic safety
Penesis, I. (2020). A Markovian approach to power generation analysis of process systems by mapping bow-tie into
capacity assessment of floating wave energy converters. Bayesian network. Process Safety and Environmental
Renewable Energy, 146, 2736–2743. https://doi.org/10.1016/ Protection, 91(1–2), 46–53. https://doi.org/10.1016/j.psep.
j.renene.2019.08.099 2012.01.005
Aziz, A., Ahmed, S., Khan, F., Stack, C., & Lind, A. (2019). Kim, Y. R., Jung, M., & Park, J. B. (2021). Development of a fuel
Operational risk assessment model for marine vessels. consumption prediction model based on machine learning
Reliability Engineering and System Safety, 185, 348–361. using ship in-service data. Journal of Marine Science and
https://doi.org/10.1016/j.ress.2019.01.002 Engineering, 9(2), 1–25. https://doi.org/10.3390/jmse9020137
Banks, J., Hines, J., Lebold, M., Campbell, R., & Begg, C. (2001). Kum, S., & Sahin, B. (2015). A root cause analysis for arctic marine
Failure modes and predictive diagnostics considerations for accidents from 1993 to 2011. Safety Science, 74, 206–220.
diesel engines. Technical report, The Pennsylvania State https://doi.org/10.1016/j.ssci.2014.12.010
University Park. Lau, B. C. P., Ma, E. W. M., & Pecht, M. (2012). Review of offshore
Beretta, S., Castelli, M., Gonçalves, I., Henriques, R., & Ramazzotti, D. wind turbine failures and fault prognostic methods.
(2018). Learning the structure of bayesian networks: A Proceedings of IEEE 2012 Prognostics and System Health
quantitative assessment of the effect of different algorithmic Management Conference, 1–5. https://doi.org/10.1109/PHM.
schemes. Complexity, 2018. 2012.6228954
Bobbio, A., Portinale, L., Minichino, M., & Ciancamerla, E. (2001). Leimeister, M., & Kolios, A. (2018). A review of reliability-based
Improving the analysis of dependable systems by mapping methods for risk analysis and their application in the offshore
fault trees into Bayesian networks. Reliability Engineering wind industry. Renewable and Sustainable Energy Reviews,
and System Safety, 71, 249–260. 91, 1065–1076. https://doi.org/10.1016/j.rser.2018.04.004
Carroll, J., McDonald, A., McMillan, D., & Bakhshi, R. (2015). Lion, S., Taccani, R., Vlaskos, I., Scrocco, P., Vouvakos, X., &
Offshore wind turbine sub-assembly failure rates through Kaiktsis, L. (2019). Thermodynamic analysis of waste heat
time. EWEA 2015 Annual Event. recovery using organic Rankine cycle (ORC) for a two-
Cheliotis, M., Lazakis, I., & Theotokatos, G. (2020). Machine stroke low speed marine diesel engine in IMO tier II and tier
learning and data-driven fault detection for ship systems III operation. Energy, 183, 48–60. https://doi.org/10.1016/j.
operations. Ocean Engineering, 216, 107968. https://doi.org/ energy.2019.06.123
10.1016/j.oceaneng.2020.107968 Márquez, F. P. G., Pérez, J. M. P., Marugán, A. P., & Papaelias, M.
Chickering, D. M. (2002). Learning equivalence classes of Bayesian- (2016). Identification of critical components of wind turbines
network structures. Journal of Machine Learning Research, using FTA over the time. Renewable Energy, 87(2), 869–883.
2(3), 445–498. https://doi.org/10.1162/153244302760200696 https://doi.org/10.1016/j.renene.2015.09.038
Cicek, K., & Celik, M. (2013). Application of failure modes and Nitonye, S., Adumene, S., & Howells, U. U. (2017). Numerical
effects analysis to main engine crankcase explosion failure design and performance analysis of a tug boat propulsion

09
Journal of Computational and Cognitive Engineering Vol. 00 Iss. 00 2023

system. Journal of Power and Energy Engineering, 5(11), 80– Industries, 40, 479–495. https://doi.org/10.1016/j.jlp.2016.
98. https://doi.org/10.4236/jpee.2017.511007 02.001
Ossai, C. I., Boswell, B., & Davies, I. J. (2016). A Markovian Ta, T. V., Thien, D. M., & Cang, V. T. (2017). Marine propulsion
approach for modelling the effects of maintenance on system reliability assessment by fault tree analysis.
downtime and failure risk of wind turbine components. International Journal of Mechanical Engineering and
Renewable Energy, 96, 775–783. https://doi.org/10.1016/j. Applications, 5(4), 1–7. https://doi.org/10.11648/j.ijmea.s.
renene.2016.05.022 2017050401.11
Pearl, J. (1988). Probabilistic reasoning in intelligent systems: Tay, Z. Y., Hadi, J., Chow, F., Loh, D. J., & Konovessis, D. (2021).
Networks of plausible inference. USA: Morgan Kaufmann Big data analytics and machine learning of harbour craft
Publishers. vessels to achieve fuel efficiency: A review. Journal of
Peng, Y., Liu, H., Li, X., Huang, J., & Wang, W. (2020). Machine Marine Science and Engineering, 9(12), 1351. https://doi.
learning method for energy consumption prediction of ships in org/10.3390/jmse9121351
port considering green ports. Journal of Cleaner Production, Uyanık, T., Karatuğ, Ç., & Arslanoğlu, Y. (2020). Machine learning
264, 121564. https://doi.org/10.1016/j.jclepro.2020.121564 approach to ship fuel consumption: A case of container vessel.
Pfaffel, S., Faulstich, S., & Rohrig, K. (2017). Performance and Transportation Research Part D: Transport and Environment,
reliability of wind turbines: A review. Energies, 10(11), 84, 10238. https://doi.org/10.1016/j.trd.2020.102389
1904. https://doi.org/10.3390/en10111904 Zahraee, S. M., Khalaji Assadi, M., & Saidur, R. (2016). Application
Planakis, N., Papalambrou, G., & Kyrtatos, N. (2022). Ship energy of artificial intelligence methods for hybrid energy system
management system development and experimental evaluation optimization. Renewable and Sustainable Energy Reviews,
utilizing marine loading cycles based on machine learning 66, 617–630. https://doi.org/10.1016/j.rser.2016.08.028
techniques. Applied Energy, 307, 118085. https://doi.org/10.
1016/j.apenergy.2021.118085 How to Cite: Chuku, A. J., Adumene, S., Orji, C. U., Johnson, K. T., & Nitonye, S.
(2023). Dynamic Failure Analysis of Ship Energy Systems Using an Adaptive
Shabarchin, O., & Tesfamariam, S. (2016). Internal corrosion hazard
Machine Learning Formalism. Journal of Computational and Cognitive
assessment of oil & gas pipelines using Bayesian belief Engineering https://doi.org/10.47852/bonviewJCCE3202491
network model. Journal of Loss Prevention in the Process

10

View publication stats