109
1 INTRODUCTION
As a result of continuous development of novel
marinesystems,inthelast20years,autonomousand
unmanned ships have become subject of intense
research in academia and industry. This resulted in
numerusresearchprojectsandscientific articles.The
motivation for introduction of autonomous and
unmanned ships
lies behind expected benefits, such
asimprovedsafetylevel,increasedenergyefficiency,
reduced operational and lifecycle costs, and
environmental footprint [1,2,3,4], however all these
claims need to be justified. Despite intense research,
theintroductionofautonomousandunmannedships
is associated with several challenges. One of main
challengesisrelatedto
thedesignofsafeautonomous
ships.Takingintoaccountthat,attimebeing,thereis
no detailed regulatory framework for autonomous
and unmanned ship, conventional maritime safety
approaches and tools cannot be used. New
approaches such as probabilistic risk assessment [5],
arerequiredforcarryingoutthesafetyassessmentof
next generation autonomous ships [6]. This is
connectedtoasetofotherchallenges,suchasthelack
of standardised assessment methodology, acceptable
risk levels, statistical data, hazardous events and
scenariosandsoon[7,8,9].
Inreality,theintroductionofautonomousshipsin
the maritime transport would result in disruptive
The Overview of Risk Analysis Methods and Discussion
on Their Applicability for Power System
of Autonomous Ships
I.Jovanović,N.Vladimir,H.Cajner&M.Perčić
UniversityofZagreb,Zagreb,Croatia
ABSTRACT:Theaimofsystemsafety,asasub‐disciplineofengineering,istoimplementscientific,engineering
andmanagementknowledgetoprovideidentification,evaluation,prevention,andcontrolofidentifiedhazards
throughout the life cycle and within the defined boundaries of operational effectiveness, time, and cost. By
utilizingriskanalysis,thesystemsafetyfunctioncanassignexpectedvaluestocertainhazardsand/orfailures
to determine the likelihood of their occurrence. Autonomous and unmanned shipping are emerging topics,
wheretechnologiesneededfortheirsuccessfulimplementationinglobalfleetalreadyexistsanditiscrucialto
demonstratethatthey
areassafeasconventionalships.Throughliteratureitissuggestedthatbyeliminating
human error as a cause of 53% of maritime accidents, autonomous and unmanned shipping will increase
maritime safety, but it is important to consider that new types of accidents can appear. Considering that
autonomousandunmannedships
needtooperatewithunattendedshipmachineryforextendedtimeperiods
andthatempiricaldataisnotavailable,newframeworkforreliabilityassessmentisneeded.Theaimofthis
paperistoprovideoverviewofriskapproachesthatcanbeappliedforreliabilityassessmentofautonomous
and unmanned ship. Within
this paper, literature review is performed where reliability methods and their
applicationtoautonomousshippingareoutlined.Furthermore,Bayesiannetworkisselectedasmostpromising
oneandfurtherdiscussed.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 18
Number 1
March 2024
DOI:10.12716/1001.18.01.09
110
changesinalllayersofmaritimeindustry.Thenovel
systems required for successful operation of
autonomous ships are highly complex, software‐
intensive and are composed of not only hardware
components but also numerous sensors and
communicationdevices.Althoughsomestudieshave
claimed that autonomous and unmanned ships can
increase
maritimetransportationsafety[10],thesafety
of autonomous systems has to be verified in detail.
Althoughtherearesomeprototypesthararetestedin
controlledenvironment,autonomousshipsarenotyet
commercially applied. It is important that the
minimum requirement for an autonomous vessel is
for it to be at least
as safe as conventional manned
ships [11], presenting an initial high‐level demand.
Therefor, all potential risks, hazards and disruptive
eventsneedtobecomprehendedandevaluated.
Survey conducted by van Cappelle et al. [12]
analysedtechnologyreadinessforremote,unmanned,
and autonomous operations. Results indicate that
technologyismature
enoughandthenextstepisto
successfully implement it on ships to increase
autonomy and reduce crew. Costs savings and
changesinthedesignofthe unmannedautonomous
bulk carrier are outlinedby Kretschmann et al. [13].
Besides crew savings, improved energy efficiency,
safety and hull optimizationareexpected.Jovanovi
ć
et al. [14] simultaneously investigated the
applicabilityofautonomousshippingandalternative
power options for the Croatian ferry fleet. Both
economical and environmental benefits are outlined,
with the electricity‐powered autonomous ship being
mostattractivefrombothpointsofview.Peetersetal.
[15] provided a solution for road‐based
freight
transportinEuropebyemployingunmannedinland
cargo vessels. Guidelines are also given for design,
control, and interaction with other vessels and the
environment[15].
Thieme et al. [16] reviewed 64 risk models
publishedsince2005toinvestigatetheapplicabilityof
modelling approaches for autonomous ships. The
analysis results indicated
that most models use
historical or published data, and a combination of
thesetoobtaintheinputforriskapproaches.
Oneofthedrivingforcesbehindthedevelopment
ofautonomousandunmannedshipsisthattheyare
expected to decrease maritime accidents related to
human error. However, it should be
noted that
autonomy will bring out new types of accidents
related to the implementation of advanced
technologies, transitions between automatic and
manualcontrol,situationawareness,etc[10].Rødseth
and Tjora [17] presented system architecture for an
unmannedmerchantship,developedwithinMaritime
Unmanned Navigation through Intelligence in
Networks(MUNIN)project.For
MUNINautonomyis
constrainedandShoreControlCenter(SCC)iscrucial
forsuccessfuloperation.Unmannedshipsystemsare
classified into 10 functional groups and Hazard
Identification (HazId) method is suggested to assess
therisks[17].RødsethandBurmeister[3]performed
HazIdforanunmannedmerchantshipandidentified
65 main
hazards of which several were classified as
unacceptable (interaction with other ships; error in
detection of small objects; propulsion system
breakdown;heavyweathermanoeuvring;collisionin
low visibility). Fault Tree Analysis (FTA) and Event
TreeAnalysis(ETA)areemployedtoassesshazards
for unmanned underwater vehicles, with a focus on
human
and organisation factors [18]. The results
indicate that the risk in autonomous underwater
vehicleoperationcanbereducedbyapplyingtherisk
management framework. A risk model for
autonomous marine systems utilizing the Bayesian
belief network to assess the human–autonomy
collaborationperformance,wasdevelopedbyThieme
and Utne [19], outlining
that the reliability of
autonomousfunctionsandsituationalawarenesshave
the highest probability of malfunction. Starting with
the cause root of a potential accident, Wróbel et al.
[20],establishedathree‐levelsafetymodel.Beginning
with an accident event, to which unmanned vessels
aresusceptible,accidentsaredividedintonavigation,
engineering,
stabilityandotherrelated.Bothmanned
and unmanned systems with different autonomy
levels are considered by [21]. Emphasis is on safety
assessment that includes the whole lifecycle of an
unmanned ship, suggesting that uncertainties and
knowledgegapsshouldbetakenintoaccountrather
than probability. Also, online risk model, developed
as part of the unmanned ship, should provide
improved performance during the testing and
verification phase. Five categories (unsafe acts,
preconditions, unsafe supervision, organisational
influences,andexternalfactors)oftheaccidentcauses
areappliedinresearchconductedby[4]toassessthe
potential impact of unmanned vessels on maritime
transportation safety,
outlining the benefits and
drawbacks that unmanned vessels have regarding
maritimetransportationsafety.TheSystem‐Theoretic
ProcessAnalysis(STPA)frameworkisusedtocreatea
preliminary risk assessment of remotely‐controlled
merchantvesselstoprovidedesignrecommendations
[22].55riskinfluencingfactors,categorisedintofour
categories (human, technology, environment, ship),
thatcanaffectnavigationalsafetyofautonomylevel3
MASS are defined by [23]. Taking into account the
lack of knowledge and experience, complexity and
limitedabilityforverificationofautonomoussystems,
[24] presented an online risk model. The online risk
modelisdevelopedbycombiningSTPAandBBN.By
integrating
an online risk model and ship control
systems, Johansen and Utne [25] demonstrated that
improvements can be achieved for both safety and
costs.YangandUtne[26]showedthatacombination
of different risk analysis methods can contribute to
the improvement of an online risk model. Table 1
provides an
overview of risk analysismethods used
for the safety assessment of autonomous and
unmannedships.
2 BAYESIANNETWORK
Belief networks (also called Bayes’ networks or
Bayesian belief networks) are a way to depict the
independence assumptions made in a distribution.
Theirapplicationdomainiswidespread,rangingfrom
troubleshooting and expert reasoning
under
uncertainty to machine learning. Bayesian Network
(BN) is a graphical structure for representing
probabilistic relationships among a large number of
variables and making probabilistic inferences using
111
those variables. A BN is a DAG with the nodes
representingthevariablesandarcsrepresentingtheir
conditional dependencies [27]. One of the main
advantagesofBNisthattheyallowinterfacebasedon
observedevidence.FortherandomvariableX
1andX2
Bayesrulestates[27]:
Table1.Overviewofriskanalysismethodsandtheir
applications.
________________________________________________
Riskmethod LiteratureTypeofproblem
________________________________________________
Bayesian Wróbeletal.(2016)[20] Generaloverviewof
Networkrelationshipsbetween
(BN)safetyfeaturesof
unmannedvessels.
ThiemeandUtne Human‐autonomy
(2017)[19]collaborationassessment.
Utneetal.(2020)[24] Onlineriskmodellingfor
autonomousships.
JohansenandUtne Supervisoryriskcontrol
(2022)[25]of
autonomoussurface
ships.
Event Thiemeetal.(2015)[18] Riskmanagement
Treeframework(RMF)for
Analysisunmannedunderwater
(ETA)vehicles(UUV).
Wróbeletal.(2016)[20] Generaloverviewof
relationshipsbetween
safetyfeaturesof
unmannedvessels.
FaultTree Thiemeetal.(2015)[18] Riskmanagement
Analysisframework(RMF)
for
(FTA)unmannedunderwater
Hazardvehicles(UUV).
Identification RødsethandTjora Informationand
(HazId) (2014)[17]Communication
Technologies(ICT)
architectureforan
unmannedship.
RødsethandRiskAssessmentforan
Burmeister(2015)[3] UnmannedMerchant
Ship.
JohansenandUtne Supervisoryriskcontrol
(2022)[25]ofautonomoussurface
ships.
Preliminary YangandUtne(2022) Anonlineriskmodelfor
Hazard [26]autonomousmarine
Analysissystems.
(PHA)
Procedural YangandUtne(2022) Anonlineriskmodelfor
Hazardand [26]autonomousmarine
Operabilitysystems.
Analysis
(HAZOP)
Risk Thiemeetal.(2015)[18] Riskmanagement
managementframework(RMF)for
unmannedunderwater
vehicles(UUV).
System‐ Wróbeletal.(2018)[22] Safetyofremotely‐
Theoreticcontrolledmerchant
Processvessel.
Analysis Utneetal.(2020)[24] Onlineriskmodellingfor
(STPA)autonomousships.
JohansenandUtne Supervisoryriskcontrol
(2022)[25]ofautonomoussurface
ships.
YangandUtne(2022) An
onlineriskmodelfor
[26]autonomousmarine
systems.
Whatif Wróbeletal.(2017)[4] Potentialimpactof
unmannedvesselson
maritimetransportation
safety.
________________________________________________
21 1
12
21 1
|
|
ii
all i
PX X PX
PXX
PX X x PX x
│
(1)
The BN qualitative analysis determines the
relationshipsamongthenodes,whilethequantitative
analysismightbeperformedintwoways:apredictive
analysis or a diagnostic analysis. The predictive
analysiscalculatestheprobabilityofanynodebased
onparentnodesandconditionaldependencies,while
the diagnostic analysis calculates the
probability of
anysetof variablesgivensomeevidence.The nodes
and arcs are the qualitative components of the
networks and provide a set of conditional
independence assumptions that can be represented
through a graph notion called d‐separation, where
each arc built from variable X to Y is directly
dependent,
thatis,acause‐effectrelationship[28].
Ifthevariablesarediscrete,thenthe probabilistic
relationshipofeachnodeXwithitsrespectiveparents
pa(X)isdefinedusingaconditionalprobabilitytable
(CPT). For continuous variables, the conditional
probability distribution (CPD), which represents
conditionalprobabilitydensityfunctions,definesthis
probabilistic
relationship, and the quantitative
analysis is based on a conditional independence
assumption.ConsideringthreerandomvariablesX,Y,
andZ,XisconditionallyindependentofYgivenZif
P(X,Y|Z) = P(X|Z)P(Y|Z) [28]. The joint probability
distribution of a set of variables, based on their
conditionalindependence,canbefactorizedas
shown
inEquation(1):
12
1
,,,
n
nii
i
Px x x PxParentx
(2)
Thegraphicalrepresentationisthebridgingofthe
gap between (high‐level) conditional independence
statements encoded in the model and (low‐level)
constraints, which enforce the CPD. Given some
evidence,thebeliefsarerecalculatedtoindicatetheir
impact on the network. The possibility of using
evidencefromthesystem
toreassesstheprobabilities
of network events is another important feature of
BNs,whichisusefultodeterminecriticalpointsinthe
system.ClassicalmethodsofinferenceofaBNforthis
purpose involve the computation of the posterior
marginalprobabilitydistributionofeachcomponent,
the posterior joint probabilitydistribution of
subsets
of components, and the posterior joint probability
distributionofthesetofallnodes.
3 DISCUSSION
Inthissubsectionthreerelevantarticles,[1],[29],[30],
are selected for further discussion of applicability of
BNforreliabilityassessmentofautonomousships.In
all three articles focus is on ship machinery
system,
Figure1.
112
Figure1. A schematicarrangementofthe machinery plant
[1],[30].
Abaeietal.[1]proposedandverifiedmethodology
topredictfailureprobabilitiesinthesystemthatwill
lead to breakdown of unattended machinery plant.
Theframeworkconsistsoffoursteps:
1. Identifying failure sensitive components (selecting
sensitive components according to severity and
risk index, determining type of human activities,
and observing
data for critical and non‐critical
failures),
2. Multinominal process tree (constructing branch
trees, defining probability function, and
developingcategoricalfailurefunction)
3. Hierarchical Bayesian interface (constructing
Bayesian network, setting non‐informative prior
function for unknown parameters, deriving
likelihoodfunction,runningMCMCforpredicting
marginalizedposteriorifunction),
4. Monte Carlo
simulations (estimating number of
critical and non‐critical failures in consecutive
intervals).
Abaeietal.[29]updatedpreviousframeworkand
also considered redundancy of autonomous
machinerytogainresilience.Thisstudyresultsshow
that with adding redundancy significant advantages
can be achieved regarding costly unplanned
interruptions and repairs. BahooToroody et al.
[30]
employedBNtoestimatethetrustedoperationaltime
of the shipmachinery system throughfour different
autonomy degrees (conventional ship, remotely
controlled ship with crew onboard, remotely
controlled ship, and fully autonomousship). A two‐
parameterWeibulldistributionisgeneratedtomodel
thetrustedtime.MCMCsimulationthroughBayesian
inference was adopted to formulate an appropriate
likelihood function for obtaining the joint posterior
distributionofhyper‐parameters.
4 CONCLUSIONS
Themodellingpoweroftraditionalriskanalysissuch
as fault tree and event tree analysis are clearly
surpassedbyBN.Bothfaulttreesandeventtreescan
easilybeconvertedto
aBayesiannetwork.
The main advantages of BN, with respect to risk
analysis,aresummarizedasfollows:
Theycanrepresentuncertainknowledgewhichis
necessary for novel systemswhit nodocumented
failurehistory,
Theyenablemodellingofcontinuousvariables,
They offer possibility of insertion of evidence for
systemreassessmentandupdating,
They provide combination of qualitative and
quantitativevariables,
Theyofferidentificationofrelevantandirrelevant
information.
Taking into account that with higher degree of
autonomy, complexity of marine systems will
increase, employment of BN in risk analysis has
immense potential. BN enable quicker and more
intuitivemodelling.BNhasbeensuccessfullyapplied
for risk assessment of autonomous ships, providing
useful tool to model uncertainty and overcome data
scarcity.
ACKNOWLEDGEMENT
This research was supported by the Croatian Science
Foundation under the project Green Modular Passenger
VesselforMediterranean(GRiMM),(ProjectNo.UIP‐2017‐
05‐1253). Ivana Jovanović, Ph.D. student, is supported
throughthe“Youngresearchers’careerdevelopmentproject
– training of doctoral students” of the Croatian Science
Foundation.
REFERENCES
[1]Abaei, M. M., Hekkenberg, R., & BahooToroody, A.
(2021). A multinomial process tree for reliability
assessment of machinery in autonomous ships.
Reliability Engineering and System Safety, 210.
https://doi.org/10.1016/j.ress.2021.107484.
[2]de Vos, J., Hekkenberg, R. G., & Valdez Banda, O. A.
(2021). The Impact of Autonomous Ships on Safety at
Sea
–AStatisticalAnalysis.ReliabilityEngineeringand
System Safety, 210.
https://doi.org/10.1016/j.ress.2021.107558.
[3]Rødseth, Ø. J. and H. C. Burmeister (2015). Risk
assessmentforanunmannedmerchantship.TransNav:
International Journal on Marine Navigation and Safety
ofSeaTransportation,9(3),357‐364.
[4]Wróbel, K., Montewka, J., & Kujala,P.(2017).
Towards
theassessmentofpotentialimpactofunmannedvessels
on maritime transportation safety. Reliability
Engineering and System Safety, 165(March), 155–169.
https://doi.org/10.1016/j.ress.2017.03.029.
[5]Rozell, D. J. (2018). The ethical foundations of risk
analysis.RiskAnalysis,38(8),1529‐1533.
[6]Nzengu,W.,Faivre,J.,Pauwelyn,A.S.,Bolbot,V.,Lien
Wennersberg, L. A.,
& Theotokatos, G. (2021).
Regulatory framework analysis for the unmanned
inland waterway vessel. WMU Journal of Maritime
Affairs,20(3),357‐376.
[7]Bolbot, V., Theotokatos, G., Andreas Wennersberg, L.,
Faivre,J.,Vassalos,D.,Boulougouris,E.,&VanCoillie,
A.(2021).Anovelriskassessmentprocess:Application
to an autonomous inland
waterways ship.Proceedings
of the Institution of Mechanical Engineers, Part O:
JournalofRiskandReliability,1748006X211051829.
[8]Chang, C. H., Kontovas, C., Yu, Q., & Yang, Z. (2021).
Risk assessment of the operations of maritime
autonomous surface ships. Reliability Engineering and
System Safety, 207(November 2020), 107324.
https://doi.org/10.1016/j.ress.2020.107324.
113
[9]Montewka,J.,Wróbel,K.,Heikkilä,E.,ValdezBanda,O.,
Goerlandt, F., & Haugen, S. (2018, September).
Challenges, solution proposals and research directions
insafetyandrisk assessmentofautonomous shipping.
InPSAM14thProbabilisticSafAssessManagConf.
[10]Porathe,T.,Å.Hoem,Ø.Rødseth,K.Fjørtoft,andS.O.
Johnsen (2018). At least as safe as manned shipping?
Autonomousshipping,safetyand“humanerror”.Safety
andReliability–SafeSocietiesinaChangingWorld(pp.
417‐425).CRCPress.
[11]Zhou,B.,Gao,F.,Wang, L.,Liu,C., &Shen,S.(2019).
Robustandefficientquadrotortrajectorygenerationfor
fastautonomous
flight.IEEERoboticsandAutomation
Letters,4(4),3529‐3536.
[12]vanCappelle,L.E.,Chen,L.,&Negenborn,R.R.(2018).
Survey on short‐term technology developments and
readiness levels for autonomous shipping. In Lecture
NotesinComputerScience(includingsubseriesLecture
Notes in Artificial Intelligence and Lecture Notes in
Bioinformatics):
Vol.11184LNCS.SpringerInternational
Publishing.https://doi.org/10.1007/978‐3‐030‐00898‐7_7.
[13]Kretschmann, L., Burmeister, H. C., & Jahn, C. (2017).
Analyzing the economic benefit of unmanned
autonomous ships: An exploratory cost‐comparison
betweenanautonomousandaconventionalbulkcarrier.
Researchintransportationbusiness&management,25,
76‐86.
[14]Jovanović, I., Vladimir, N., Perčić, M., & Koričan, M.
(2022).Thefeasibilityofautonomouslow‐emissionro‐ro
passenger shipping in the Adriatic Sea. Ocean
Engineering, 247.
https://doi.org/10.1016/j.oceaneng.2022.110712.
[15]Peeters, G., M. Kotzé, M. R. Afzal, T Catoor, S. Van
Baelen, P. Geenen, M. Vanierschot, R.
Boonen, and P.
Slaets(2020).Anunmannedinlandcargovessel:Design,
build, and experiments. Ocean Engineering, 201,
p.107056.
[16]Thieme, C.A., Utne, I.B., Haugen, S., 2018. Assessing
ship risk model applicability to Marine Autonomous
Surface Ships. Ocean Engineering 165, 140–154.
https://doi.org/10.1016/j.oceaneng.2018.07.040.
[17]Rødseth,Ø.J.andÅ.Tjora(2014).A
systemarchitecture
for an unmanned ship. Proceedings of the 13th
international conference on computer and IT
applications in the maritime industries (COMPIT).
VerlagSchriftenreiheSchiffbau,2014Redworth,UK.
[18]Thieme,C.A.,I.B.Utne,andI.Schjølberg(2015).Arisk
management framework for unmanned underwater
vehiclesfocusingonhuman
andorganizationalfactors.
International Conference on Offshore Mechanics and
Arctic Engineering, vol. 56499, p. V003T02A075.
AmericanSocietyofMechanicalEngineers.
[19]Thieme, C. A. and I. B. Utne (2017). A risk model for
autonomousmarinesystemsandoperationfocusingon
human–autonomy collaboration. Proceedings of the
Institution of Mechanical Engineers, Part O: Journal of
RiskandReliability,231(4),446‐464.
[20]Wróbel,K.,P.Krata,J.Montewka,andT.Hinz(2016).
Towardsthedevelopmentofariskmodelforunmanned
vessels design and operations. TransNav: International
Journal on Marine Navigation and Safety of Sea
Transportation,10(2).
[21]Utne,I.B.,A.J.
Sørensen,andI.Schjølberg(2017,June).
Risk management of autonomous marine systems and
operations. International conference on offshore
mechanics and arctic engineering (Vol. 57663, p.
V03BT02A020). American Society of Mechanical
Engineers.
[22]Wróbel,K.,J.Montewka,andP.Kujala(2018).System‐
theoretic approach to safety of remotely‐controlled
merchantvessel.Ocean
Engineering,152,334‐345.
[23]Fan,C.,K.Wróbel,J.Montewka,M.Gil,C.Wan,andD.
Zhang (2020). A framework to identify factors
influencingnavigationalriskforMaritimeAutonomous
SurfaceShips.OceanEngineering,202,107188.
[24]Utne,I.B.,B.Rokseth,A.J.Sørensen,andJ.E.Vinnem
(2020).
Towardssupervisoryriskcontrolofautonomous
ships. Reliability Engineering & System Safety 196:
106757.
[25]Johansen, T., and I. B. Utne (2022). Supervisory risk
controlofautonomoussurfaceships.OceanEngineering
251:111045.
[26]Yang,R.,andI.B.Utne(2022).Towardsanonlinerisk
model for autonomous marine systems (AMS).
Ocean
Engineering251:111100.
[27]Barber, D. (2012). Bayesian reasoning and machine
learning.CambridgeUniversityPress.
[28]Friis‐Hansen,A.(2000).Bayesiannetworksasadecision
support tool in marine applications. Denmark:
Department of Naval Architecture and Offshore
Engineering,TechnicalUniversityofDenmark.
[29]Abaei, M. M., Hekkenberg, R., BahooToroody,
A.,
Banda, O. V., & van Gelder, P. (2022). A probabilistic
model to evaluate the resilience of unattended
machinery plants in autonomous ships. Reliability
Engineering&SystemSafety,219,108176.
[30]BahooToroody, A., Abaei, M. M., Banda, O. V.,
Montewka, J., & Kujala, P. (2022). On reliability
assessment of ship machinery
system in different
autonomy degree; A Bayesian‐based approach. Ocean
Engineering,254,111252.
