Use of HFACS–FCM in fire prevention modelling on board ships
Omer Soner a,⇑
, Umut Asan b
, Metin Celik c
aDepartment of Maritime Transportation and Management Engineering, Istanbul Technical University, Tuzla 34940, Istanbul, Turkey
bDepartment of Industrial Engineering, Istanbul Technical University, Macka 34367, Istanbul, Turkey
cDepartment of Marine Engineering, Istanbul Technical University, Tuzla 34940, Istanbul, Turkey
article info
Article history:
Received 5 January 2015
Received in revised form 24 February 2015
Accepted 11 March 2015
Keywords:
Ship safety management
Root cause analysis
Fire prevention
FCM
HFACS
abstract
This research proposes a proactive modelling approach that combines Fuzzy Cognitive Mapping (FCM)
and Human Factors Analysis and Classification System (HFACS). Principally, the suggested model helps
predicting and eliminating the root causes behind the frequently repeating deficiencies on board ships.
Supported with qualitative simulations, the HFACS–FCM model is demonstrated on a fire related
deficiency sample database. The findings indicate that the root causes of a fire related deficiency on board
ship might be revealed in various levels such as unsafe acts, pre-conditions for unsafe acts, unsafe
supervision, and organization influences. Considering the determined root causes and their priorities,
the Safe Ship System Mechanism (SSSM), Safe Ship Operation Mechanism (SSOM), and Safe Ship
Execution Mechanism (SSEM) are constituted. Consequently, the paper has added value to both
predicting the root causes and enhancing fire-fighting potential which provides reasonable contributions
to safety improvements at sea.
2015 Elsevier Ltd. All rights reserved.
1. Introduction
Fire accident is one of the most challenging and fatal events on
board ships. The control of fire in such operational environments
requires immediate response and great effort. Thus, relevant
systems should be timely functioning under operator (ship crew)
control without any interruption (Kuo and Chang, 2003). In addition to firefighting systems on board ships, proactive approaches
against fire related non-conformities, accidents and hazardous
situations are required to avoid the operational problems.
Schröder-Hinrichs et al. (2011) argued that organizational factors
were usually not noticed in maritime accident investigations of
fires and explosions in machinery spaces, instead, lower echelons
such as unsafe acts are just considered.
In the current situation, Port State Control (PSC) organizations
have attempted to enable standard level of achievements in safety,
security, and environmental aspects. The strict control requirements at international level enforce the ship-owners and operators
and thereby ship management companies to ensure compliance
with the relevant rules and regulations. Li and Zheng (2008) made
suggestions on the improvement of the enforcement of PSC.
Otherwise, the lack of effective and systematic implementations
of the requirements might cause deficiencies, nonconformities or
major-nonconformities on board ships, which eventuate in detention. Besides its cost and unexpected catastrophic consequences, it
affects the reputation of ship management companies in the global
market. Above all, reoccurrence of the mentioned issues (i.e. fire
accidents) will threat the sustainable maritime transportation.
A brief review of PSC survey results on shipboard deficiencies
highlights the significance of the problem. For example, according
to the Tokyo MOU PSC annual report, fire safety measures constitute 18% of the total deficiencies (Tokyo MoU, 2013), while Paris
MOU statistics about fire safety related deficiencies illustrate 15%
(Paris MoU, 2012). These statistical reports serve also as key
documents for following the distribution of several deficiencies
at operational level. For instance, Det Norske Veritas (DNV), a prestigious member of International Association of Classification
Societies (IACS), reported that fire safety measures, with 19%, is
the most common category of deficiencies dealt with (DNV,
2012). Another report published by American Bureau Shipping
(ABS) illustrates that fire safety measures, as the most common
deficiency, has the ratio of 16% compared to other categories
(ABS, 2012). Finally, the report published by Nippon Kaiji Kyokai
(ClassNK or NK) emphasises the importance of fire safety measures
related deficiencies whose percentage is around 24% (ClassNK,
2012).
In order to prevent fire risk on board ships, the ship managers
and responsible decision makers should ensure effective implementation of a safety management system in accordance with
http://dx.doi.org/10.1016/j.ssci.2015.03.007
0925-7535/ 2015 Elsevier Ltd. All rights reserved.
⇑ Corresponding author. Tel.: +90 216 3951064; fax: +90 216 3954500.
E-mail address: soneromer023@gmail.com (O. Soner).
Safety Science 77 (2015) 25–41
Contents lists available at ScienceDirect
Safety Science
journal homepage: www.elsevier.com/locate/ssci
the International Safety Management (ISM) Code. The main objective of the system is to improve safety level on board ships while
preventing human injury, loss of life, and damage to marine
environment. According to recent amendments to the ISM Code,
identifying measures intended to prevent recurrence of deficiencies and near misses has become one of the core issues. It introduces a relatively new concept called preventive action planning,
which strictly requires detailed analysis in order to make
consistent decisions on actions to be taken (ISM Code, 2010).
There is no doubt that systematic analysis on causation is the most
essential aspect of preventive action practices along with the ship
operations and management.
This study proposes a novel preventive action planning
approach to enhance fire safety measures on board ships. The rest
of this paper is organized as follows: The current section discussed
the significance of controlling and monitoring the fire related
deficiencies on board ships. Then, a wide range of maritime safety
literature is reviewed in Section 2. Section 3 introduces the conceptual framework of the model which is based on combination of
Fuzzy Cognitive Mapping (FCM) and The Human Factors Analysis
and Classification System (HFACS). To demonstrate the suggested
model, a case study concerning a fire related deficiency sample
database is analysed in Section 4. In the final section, the research
outcomes and potential contributions through ship fire safety preparedness are extensively discussed.
2. Literature review
2.1. Maritime literature review
Maritime safety is a significantly important element of sustainability in world trade since maritime transportation has been carrying 80% of the global cargo (Asariotis et al., 2013). Furthermore,
maritime transportation system has long been monitored by
International Maritime Organization (IMO), whose primary
purpose is to maintain comprehensive regulatory framework at
international level (Wieslaw, 2012) Shipping might be considered
as one of the most dangerous and global industries of the world.
The shipping industry seeks for a modern and user friendly safety
system since the maritime accidents might cause catastrophic consequences (Hetherington et al., 2006). Hence, the contribution to
safety at sea are highly expected and appreciated by maritime
society. This section draws together a wide range of existing
literature on a range of issues on maritime safety. These issues
involve the following four important aspects of maritime safety:
(i) regulatory framework, (ii) human factor, (iii) technological
improvements and (iv) methodological approaches.
From the regulatory framework perspective, various conventions have been developed and adopted by IMO in order to promote the safety, security, and environmental sensitiveness in
shipping industry. However, the effects of the mentioned conventions on shipping industry have been argued and discussed by
maritime researchers, rule-makers, and responsible executives.
For instance, Vanem and Skjong (2006) criticized the regulation
requirements along with the evacuation procedures in which effective assessment is not possible to conduct. On the other hand,
Tzannatos and Kokotos (2009) investigated ship accident during
the pre- and post- ISM period so as to assess the effectiveness of
the ISM Code. In addition, Knapp and Franses (2009) studied on
the major international conventions regarding safety, pollution,
search and rescue measures. To strength the safety related
regulations, Celik (2009) proposed a systematic approach to evaluate the compliance level of the ISM code with the ISO 9001:2000 to
adopt an integrated quality and safety management system. The
study illustrated that safety management system implementations
on board ships can be enhanced via quality management
principles. Besides international conventions’ enforcement,
Knudsen and Hassler (2011) believed that there are additional
efforts required to challenge with the main causes of the ship accidents which have not been reduced to a desired ratio. Karahalios
et al. (2011) also conducted research to perform a cost-benefits
analysis along with the maritime regulations. Furthermore,
Schinas and Stefanakos (2012) investigated feasibility of the
environmental measurements defined within International
Convention for the Prevention of Pollution from Ships (MARPOL).
Human factor is another core topic in maritime safety studies.
To find out the role of human element in safety at sea, Hee et al.
(1999) conducted one of the pioneering researches on maritime
safety assessment. Furthermore, Hetherington et al. (2006) concluded a research that reviews a number of studies to eliminate
the human errors in ship accidents. As another study, Celik and
Er (2007) examined the potential role of design errors which trigger the human error in shipboard operations. To enhance human
factor analysis, HFACS was utilized in order to make quantitative
assessment of shipping accident (Celik and Cebi, 2009). To clarify
the exact reasons, Wang et al. (2013) proposed a new method in
order to enable accident causations. Recently, Akhtar and Utne
(2014) investigated human fatigue effects to bridge team management demonstrated with ship grounding case.
Besides human element, technological improvements and maritime innovations are one of the significant aspects of maritime
safety. At system safety level, Tzannatos (2005) investigated
probable equipment failures and their effects in terms of reliability
monitoring of the Greek coastal passenger fleet. Moreover, Eide
et al. (2007) developed an intelligent model to prevent oil spill as
another catastrophic event at sea. Beyond, Lun et al. (2008)
investigated the technological adoption to manage security
enhancement especially in container transport. In a further study,
Lambrou et al. (2008) introduced the Intelligent Maritime
Environment (i-MARE) framework and technological platform for
cargo shipping. Vanem and Ellis (2010) investigated the feasibility
of adapting a novel on board passenger monitoring and communication system based on RFID technology which provides a decision
support in emergency situations. Similarly, LiPing et al. (2011) took
the advantage of the video surveillance technology for safe navigation. It can be clearly seen that new technologies have potential
to enhance safety at sea; however, it is still a great deal to manage
the gaps among regulation implementations, human element and
recent technologies in order to increase the overall utility of such
attempts in safety improvements.
Methodological approaches, as the fourth important aspect of
maritime safety, have been playing a key role in transforming
operational data, facts, and figures into useful information along
with safety enhancement. With this purpose in mind, several
researchers, such as Rothblum (2000), O’Neil (2003), Darbra and
Casal (2004), and Toffoli et al. (2005), have conducted statistical
analyses, especially, on accidents and their prevention. On the
other hand Lee et al. (2001), Wang and Foinikis (2001), Wang
(2002), Lois et al. (2004) used formal safety assessment (FSA)
particularly supported with well-known techniques. Specifically,
Bayesian network modelling has been utilized in maritime safety
related studies (Antao et al., 2008; Trucco et al., 2008; Kelangath
et al., 2011; Zhang et al., 2013; Hänninen et al., 2014) in order to
deal with the inherent uncertainty and complexity in maritime
safety problems. Moreover, various methods derived from fuzzy
set theory have been cited (Sii et al., 2001; Balmat et al., 2009,
2011; Abou, 2012; John et al., 2014) in maritime safety literature.
There are also some hybrid quantified models (Celik and Cebi,
2009; Celik et al., 2010; Pam et al., 2013; Akyuz and Celik,
2014a,b; Karahalios, 2014; Wang et al., 2014) that provide satisfactory approaches to the specified operational problems in maritime
safety context.
26 O. Soner et al. / Safety Science 77 (2015) 25–41
Despite the various contributions at international level, the
numbers of ship accidents or detention rates have still not reached
their desired levels. Moreover, the number of maritime safety studies in the literature has been increasing at a relatively slow rate.
Researchers will need to focus more on operational fieldworks
and specific cases, as it appears to be the next phase of maritime
safety studies. This paper, hence, attempts to investigate the root
causes of fire safety related deficiencies in order to provide an
applicable proactive model for ship operation and management.
Considering both the theoretical and practical insights provided,
this study makes valuable contributions to the maritime safety
literature.
As the proposed model integrates Fuzzy Cognitive Mapping
(FCM) and The Human Factors Analysis and Classification System
(HFACS), the following sections introduce the theory of both
methods.
2.2. Human Factor Analysis and Classification System (HFACS)
HFACS is initiated from Swiss Cheese Model by Reason (1990).
HFACS is a comprehensive tool to analyse the human contribution
to catastrophic events, accidents, hazardous occurrences, and
deficiencies. Basically, HFACS investigates active failures and latent
conditions at four levels. Active failures are sets of inappropriate
actions by operators while latent conditions deal with the different
levels of organization (Chauvin et al., 2013). The described four
levels in HFACS are (i) unsafe acts, (ii) pre-conditions for unsafe
acts, (iii) unsafe supervision, and (iv) organization influences. If
the vulnerabilities in different levels cannot be controlled, the
occurrence probability of accidents might be arisen. HFACS was
first developed for the aviation accident investigation (Shappell
and Wiegmann, 2000, 2001). In the last decade, the HFACS model
was not only successfully applied in the aviation industry by
Shappell et al. (2007), but also in the railway (Reinach and Viale,
2006) and mining industry (Patterson and Shappell, 2010). For
instance, the study of Rothblum et al. (2002) was the pioneer scientific research that aimed to investigate human factor in maritime
accidents. Celik and Cebi (2009) combined HFACS with fuzzy set
theory in order to provide a quantitative approach to analyse a single accident case with error distribution in accordance with the
operational evidences given in accident reports. Recently, Akyuz
and Celik (2014a) used HFACS supported with cognitive mapping
approach to confirm the dependencies between causation factors.
2.3. Fuzzy Cognitive Map (FCM)
Fuzzy Cognitive Mapping, advanced by Kosko (1986) from the
classical cognitive mapping method, is an illustrative causative
representation of complex systems and can be used to model and
manipulate the dynamic behaviour of systems (Papakostas et al.,
2008). Combining elements of fuzzy logic and neural networks,
fuzzy cognitive mapping has been proven to be a promising
method for making inferences in cases with substantial uncertainty, imprecision and vagueness (Vasantha Kandasamy and
Smarandache, 2003; Tsadiras, 2008). Compared to expert systems,
fuzzy cognitive maps (FCMs) are relatively quicker and easier to
acquire knowledge (Papageorgiou and Stylios, 2008). FCMs have
been successfully applied in a variety of scientific areas, such
supervisory control systems (Stylios and Groumpos, 2000), distributed systems (Stylios et al., 1997), decision support system
(Tsadiras et al., 2003), organizational behaviour (Craiger et al.,
1996), medical informatics (Papageorgiou, 2011), marketing
(Nasserzadeh et al., 2008) and risk analysis (Lazzerini and
Mkrtchyan, 2011), among others.
Most of the FCM models are constructed basically by expert
knowledge and experience in the operation of the system.
Questionnaire survey, documentary coding and interviews are
the most common ways for this purpose. FCMs can be developed
for a single expert or a group of experts, where the latter has the
benefit of improving the reliability of the final model (Yaman
and Polat, 2009). The aggregation of knowledge from multiple
experts is a relatively simple process in fuzzy cognitive mapping
(Stach et al., 2005). Each expert describes every interconnection
with linguistic variables (weights) which are later composed (e.g.
by fuzzy arithmetic or defuzzification methods) to produce the
combined map. Several procedures have been proposed for
combining multiple FCM models into a single one (see e.g. Kosko,
1992; Stylios and Groumpos, 2000).
A FCM can be represented either as a graph, consisting of
concepts (e.g. entities, states, or characteristics of the system)
and weighted interconnections between these concepts, or as an
adjacency matrix, which has entries wij’s indicating the direct
relationship between concept i and concept j. Fig. 1 (Asan et al.,
2011) illustrates a simple FCM consisting of five concepts Ci
(i = 1, …, 5) where wij represents the influence degree from cause
Ci to effect Cj. FCM does not allow any direct connections between
a concept and itself, thus all wii elements equal to zero. All other wij
elements take values in [1, 1] and Papageorgiou (2011) explains
the meaning of these values as;
wij>0 indicates a causal increase (i.e., Cj increases as Ci increases,
and Cj decreases as Ci decreases).
wij<0 indicates causal decrease (i.e., Cj decreases as Ci increases,
and Cj increases as Ci decreases).
wij=0 indicates no causality.
Once the FCM is constructed it is used to perform qualitative
simulations in order to predict possible changes and to observe
whether the system converges toward a steady state. During the
simulations a model can reach three possible states that are listed
below (Kosko, 1997):
A steady state where the output values are stabilizing at fixed
numerical values.
A limit cycle behaviour where the concept values are falling in a
loop.
A chaotic behaviour where concept values wanders forever
without apparent structure or order.
A more formal definition of the iterative procedure can be
described as follows. The FCM should be first initialized. In other
words, the activation level of each concept takes a value based
on expert opinion about its current state or measurements from
the real system. Let each concept take its initial value as AðtÞ
i , where
Ai is the value of concept i at step t, and simulated iteratively. Then
the value of each concept in an iteration is calculated as
(Papageorgiou et al., 2009)
Aðtþ1Þ
i ¼ f AðtÞ
i þ Xn
j ¼ 1;
j–i
AðtÞ
j wji
0
BBBBB@
1
CCCCCA
ð1Þ
In Eq. (1), Aðtþ1Þ
i is the value of concept at step (t + 1), AðtÞ
i is the value
of concept at step (t), wji is the weight of interconnection between Cj
and Ci. f is the threshold function that reduces the result of the multiplication into a normalized range (within [0, 1] or [1, 1]). The
most common activation functions are (Tsadiras, 2008): bivalent,
trivalent, sigmoid, hyperbolic tangent.
O. Soner et al. / Safety Science 77 (2015) 25–41 27
3. Proposed model
3.1. Framework
Utilizing the HFACS and FCM model, a new framework on fire
safety related deficiency analysis is introduced. Conceptual framework of the proposed model is presented in Fig. 2. Principally, it
performs a great extent of proactive safety modelling through
deficiency causation, root cause identification, prioritization, and
preventive action generation. The database source in the model,
gathered from ship operational level, might incorporate PSC
inspections reports, company audits reports, near-miss reports,
hazardous occurrences reports, accident reports, and vetting surveys reports. Then, deficiency database are distributed to HFACS
to ensure satisfactory deficiency causation where FCM technique
highlights the relationships among the designated contributing
causes of fire related deficiencies on board ships. Considering the
initial results, it is decided whether a contributing cause is a root
DEFICIENCIES
(Fire safety)
PSC inspections
Company audits
Near-miss reports
Accident reports
Vetting survey results
Other
DATA
Ship management company
Ship fleet
Deficiency causation
(HFACS)
Root cause
identification
(FCM)
ANALYSIS
Preventive action
planning
INTEGRATION
Root cause prioritization
and verification
(Simulation)
Root cause analysis
Fig. 1. A hypothetical FCM model and the corresponding adjacency matrix (Asan et al., 2011).
C1
C2
C5
C3
C4
w12
w23
w51
w42 w34
w54
w15
w25
Fig. 2. Conceptual framework of the model.
28 O. Soner et al. / Safety Science 77 (2015) 25–41
cause or not. Finally, the integration phase enables preventive
action adaptation at ship operational level. It is considered as a
phase to promote ship safety against fire related occurrences at
sea.
3.2. Modelling causes of fire related deficiencies
Before preventive actions are suggested against deficiencies, it
is crucial to identify the initiating causes of the current causal
chain that leads to fire related deficiencies on board ships.
Dealing with only a small number of these root causes will reasonably prevent many of the undesirable deficiencies. In order to identify root causes and their priorities, a fuzzy cognitive map is
constructed and analysed as summarized below (see Fig. 3).
3.2.1. Step 1: Identification of causal relationships
As previously explained, the causes (i.e. concepts in the fuzzy
cognitive map) are identified by reviewing diverse reports on
fire related deficiencies and employing the human error
framework HFACS. In this step, the causal relationships between
concepts are identified by providing domain experts ordered
pairs of concepts in a questionnaire format (see Fig. 4). This
allows systematic examination of all relationships. Here, a
causal relationship is characterized with vagueness, since
it represents the influence of one qualitative concept on another
one and will be determined using linguistic terms (Papageorgiou
and Stylios, 2008). In this way, an expert transforms his
knowledge and experience on the behaviour of the system into
a fuzzy weighted graph.
Identification of Causes
of Fire Related Deficiencies
Identification of Causal
Relationships
Analysis of Direct
Relationships a
Analysis of Indirect
Relationships b
Inference through
qualitative Simulations c
Aggregation of Weights from
Multiple Experts
Decision on the Final
List of Root Causes
c FCM Simulation
Algorithm, WhatIf Scenarios
Reports on Fire Related
Deficiencies and HFACS
Questionnaire Survey,
Linguistic Variables
b Reachability Matrix,
Normalization,
Outdgree, Indegree,
Impulse Index
Data Collection Analysis
List of
Potential
Root Causes
a Adjacency Matrix,
Outdegree, Indegree
Impulse Index
Defuzzification
Max/Sum Method
Center of Gravity
Tools and Techniques:
List of
Potential
Root Causes
Priorities
Fig. 3. The flow of the proposed FCM methodology.
Fig. 4. Ordered pairs of concepts in a questionnaire format.
O. Soner et al. / Safety Science 77 (2015) 25–41 29
The linguistic variable used in this study, expressing the degree
of the causal relationship between two concepts, takes values in
the universe U ¼ ½0; 1 and consists of the term set {does not affect
at all (N), affects weakly in a positive way (W), affects weaklymoderately in a positive way (WM), affects moderately in a positive way (M), affects moderately-strongly in a positive way (MS),
affects strongly in a positive way (S)}. Using these six linguistic
terms, an expert can describe in detail the influence of one concept
on another and can discern between different degrees of influence.
Fig. 5 depicts the membership functions of the fuzzy sets that are
used to characterize these terms.
3.2.2. Step 2: Aggregation of individual weights and defuzzification
In order to improve the reliability of the final model, a group
map is developed. The weights obtained from multiple experts
are combined to produce the overall linguistic weights and the
group adjacency matrix. The well-known SUM or MAX methods
in fuzzy logic can be employed for this purpose. Fig. 6 provides
an example of the MAX operation where the two linguistic terms
‘‘moderately-strongly’’ and ‘‘strongly’’ are aggregated. If the
experts have different priorities of importance, the different
influences are reflected in the results through multiplying the membership function l suggested by the kth expert by the corresponding
credibility weight (Stylios and Groumpos, 2004; Saaty, 2004).
Once the overall linguistic weights are obtained, they are transformed to numerical (crisp) values using the Center of Gravity
(CoG) defuzzification method. CoG is computed from the following
equation (Ross, 2004):
z ¼
R
lw~ ij
ðzÞz dz
R
lw~ ij
ðzÞdz ð2Þ
where R denotes an algebraic integration. The transformed numerical values will be within the range [0,1]. The same procedure is
applied to all the causal relationships among the n concepts of
the map.
3.2.3. Step 3: Identification of potential root causes (analysis of direct
relationships)
As we aim to identify root causes (i.e. initial causes in a causal
map), the role of each concept in the map needs to be carefully
examined. An adjacency matrix W allows analysing the
contribution of a concept in the map and articulates how this
concept is connected directly to other concepts (Kosko, 1986).
Each variable is defined by its outdegree (od) and indegree (id).
Outdegree shows the cumulative strengths of connections exiting
the concept and is expressed as the row sum of absolute weights
of a concept in the adjacency matrix (Nozicka et al., 1976;
Özesmi and Özesmi, 2004).
odðiÞ ¼ Xn
j¼1
jwjij ð3Þ
On the other hand, indegree shows the cumulative strength of
connections entering the concept and is expressed as the column
sum of absolute weights of a concept in the adjacency matrix
(Nozicka et al., 1976; Özesmi and Özesmi, 2004)
idðiÞ ¼ Xn
j¼1
jwijj ð4Þ
Thus, a concept (i.e. a cause) that is less affected by the rest of
the causal system than it has impact on it can be characterized
as a potential root cause. To identify the role of variables in a complex system different criteria or rules have been proposed in the
literature (Godet, 1994; Gausemeier et al., 1996; Asan et al.,
2004). Here, the rules defined by Asan et al. (2004) are adapted
to the root cause identification problem. Thus, a root cause i should
fulfil the following rules
odðiÞ P xod ð5aÞ
IPIi P 2 ð5bÞ
where xod denotes the average outdegree taken over the entire concept set and IPIi denotes the so called Impulse Index which is calculated for each concept i as follows
IPIi ¼ odðiÞ
idðiÞ ð6Þ
The maximum range of IPIi extends from 0 (no influence on the
system) to 1 where the concept is not influenced by other concepts, but has an impact on others.
3.2.4. Step 4: Identification of potential root causes (analysis of indirect
relationships)
Examining the adjacency matrix reveals only potential root
causes based on the direct relationships between concepts which
are represented by causal chains of length one. However, this is
not enough to reveal the hidden root causes which sometimes
greatly influence the problem under study. Therefore, the diffusion
of causal impacts through reaction paths and loops needs also to be
considered (Godet, 1994; Serdarasan and Asan, 2007). These
indirect relationships, which are represented by causal chains of
length greater than one, can be revealed by raising the adjacency
matrix to successive powers. The raw and column sums of each
resulting matrix (raised to a certain power) are normalized to
enable a comparison among the results of successive powers. The
normalization originally developed in this paper can be expressed
as follows:
Fig. 5. The membership functions describing the linguistic terms.
Fig. 6. MAX method and CoG.
30 O. Soner et al. / Safety Science 77 (2015) 25–41
NodðiÞ
q ¼ maxi¼1…nfodðiÞg odðiÞ
q
maxi¼1…nfodðiÞ
q
g ð7Þ
NidðiÞ
q ¼ maxi¼1…nfidðiÞg idðiÞ
q
maxi¼1…nfidðiÞ
q
g ð8Þ
where NodðiÞ
q and NidðiÞ
q denote the normalized outdegree and
indegree values of concept i for the adjacency matrix raised to the
qth power, respectively; and odðiÞ
q and idðiÞ
q denote the outdegree
and indegree values of concept i for the adjacency matrix raised to
the qth power, respectively. Thus, the total outdegree and indegree
values of concept i can be calculated by the following expressions
RodðiÞ ¼ odðiÞþNodðiÞ
2 þ…þNodðiÞ
n1 =Q ¼ X
Q
q¼1
NodðiÞ
q
!,Q
ð9Þ
RidðiÞ ¼ idðiÞþNidðiÞ
2 þ…þNidðiÞ
n1 =Q ¼ X
Q
q¼1
NidðiÞ
q
!,Q
ð10Þ
where RodðiÞ and RidðiÞ denotes the total outdegree and total
indegree of concept i, respectively and indicate the reachability of
concept i. In fact, the multiplication process continues until the
adjacency matrix is raised to a certain power ðQÞ in which the concepts’ order proves to be stable (for more detail see Godet, 1994).
The resulting indicators are used to reveal potential root causes
which might be assumed to be unimportant in the previous analysis
but play a leading role because of indirect relationships. For this
purpose the concepts are examined according to the same rules
described in Step 3.
3.2.5. Step 5: Inference through qualitative simulations
Once the direct and indirect relationships are examined, the
fuzzy cognitive map is used to perform qualitative simulations to
capture the transmission of influence along all paths and to
observe whether the system converges toward a steady state.
From the steady state calculation we can get an idea of the ranking
and thereby of the overall priorities of the variables in relation to
each other (Özesmi and Özesmi, 2004).
The simulation process is initialized through assigning a value in
[0, 1] to the activation level of each concept, based on experts’ opinion about a certain state. The value of zero indicates that a given
concept is not present in the system at a particular iteration, while
the value of one suggests that a given concept is present to its maximum degree (Papageorgiou and Kontogianni, 2012). In a particular
iteration, the value of each concept is determined by its previous
value and the preceding values of all concepts that exert influence
on it through non-zero relationships (Papageorgiou, 2011). This
iterative process does not produce exact numerical values; instead
it allows analysing the dynamic behaviour of the system.
The FCM simulation algorithm originally developed by (Kosko,
1988) utilizing Eq. (1) consists of the following five stages
(Papageorgiou and Kontogianni, 2012):
Stage 1. Define the initial vector A.
Stage 2. Multiply the initial vector A and the matrix W.
Stage 3. Update the resultant vector A at time step t + 1.
Stage 4. Consider the new vector Aðtþ1Þ as the initial vector in the
next iteration.
Stage 5. Steps 2 to 4 are repeated until Aðtþ1Þ AðtÞ 6 e ¼ 0:001
(where e is a residual describing the minimum error difference
among the subsequent concepts) or Aðtþ1Þ ¼ AðtÞ
. AðtÞ will be the
final vector.
Note that the iterative method applied here is not necessarily
concerned about the structure, but the outcome, or inference of
the map (Özesmi and Özesmi, 2004).
In order to prioritize the potential root causes, identified in Step
3 and 4, simulations are performed for different initial state
vectors. In each ‘‘what-if’’ scenario, only one particular concept
(i.e. cause) is activated by assigning a value of one to its activation
level. In this way, it is possible to observe the changes in the
activation levels of other concepts throughout the simulation.
The higher the number of concepts influenced (i.e. activated) in
the early iterations by a particular concept, the more likely the
concept is a root cause.
Consequently, the decision on the final list of root causes is
made by synthesizing the results of Step 3, 4 and 5. A case
study on a set of fire related deficiency data is conducted in
Section 4.
4. Case study
4.1. Fire related deficiency sample database
Supporting the fire related deficiency database, the research
tends to a great variety of maritime sources such as DNV’ annual
deficiency report (DNV, 2012), ABS’ Reducing the Port State
Detention Factor report (ABS, 2012), Paris Mou’ Taking PSC to the
Next Level Annual report (Paris Mou, 2012), Tokyo Mou’ Annual
report on PSC (Tokyo MoU, 2013), and ClassNK’ annual report on
PSC (ClassNK, 2013). The field investigation addressed the frequently encountered fire related deficiencies on board ships. The
specific deficiency items are categorized into twenty main groups
given as follows:
1. Fire-dampers.
2. Emergency fire pump.
3. Fire prevention.
4. Firefighting equipment and appliances.
5. Fire detection.
6. Fire doors within main vertical zone.
7. Fixed fire extinguishing installation.
8. Ready availability of firefighting equipment.
9. Ventilation.
10. Inert gas system.
11. Division – main zones.
12. Main vertical zone.
13. Personal equipment.
14. Means of control (opening, closure of skylights, pumps, etc.
machinery spaces.
15. Jacketed piping system for high pressure fuel lines.
16. Fire control plan – all ships.
17. International shore connection.
18. Main fire pumps.
19. Emergency Escape Breathing device (EEBD).
20. Other firefighting equipment.
The inoperable fire dampers might lead to minor or major
deficiencies. The causes of the inoperable fire dampers are inadequate familiarization, poor maintenance, adjustment mechanical
parts, functional malfunctions, corrosion, sealing materials, flap
positions, installations, etc. Another system related deficiency
on board ships is poor condition of emergency fire pump. In
detail, starting failures, self-priming issues, loss of pressure, leakages, remote control interruptions, electrical shortages, malfunctioned gauges, driven engine failures, fuel related matters are
the reasons for such deficiency item. Fire prevention measures,
might cause ship detention, are highly critical aspect of the fire
O. Soner et al. / Safety Science 77 (2015) 25–41 31
safety on board ship. Furthermore, oil leaks and improper storage
of combustible materials have great potential to spread of fire.
The poor condition of fire-fighting equipment and appliances
are also the key reason of firefighting vulnerabilities at sea in
terms of leaking line and fire hose, certification incompliances,
incomplete firemen’s outfits, etc. PSC audits state number of
causes evidenced with disconnected or covered alarm systems,
unavailable previous testing records. Marvellously, non-functional
fire doors, keeping the fire doors in an open position, unauthorized cuts in fire zone boundaries, blocking the emergency escape
routes might be observed as a result of substandard ship operations. The fixed fire extinguishing installation, addressing in company audits and vetting reports addressed especially empty or
blocked fire boxes, expired fire extinguisher and breathing
apparatus. Beyond all these, quite a number of ventilation
deficiencies are reported with critical failures such as corrosion
related malfunctioning. The surveys also pointed out performance
and condition matters in different equipment/system such as
inert gas system, personal protective equipment, control systems,
international shore connection, main fire pumps and emergency
escape breathing device.
4.2. Deficiency causation (HFACS)
Considering the four level of human factor such as unsafe acts,
pre-conditions for unsafe acts, unsafe supervision, and organization influences, the next step is to determine causes leading to fire
related deficiencies. Reviewing the fire related deficiency sample
database, the possible causes in ‘‘unsafe acts level’’ are determined
as follows:
C1. Late responding to the sudden operational failure in critical
components of fire system.
C2. Misperception of fire related emergency situations
complexity.
C3. Responding to emergency fire related situations in panic.
C4. Omitted step in fire safety related procedure.
C5. Neglected items in fire frightening equipment routine
inspection checklist.
C6. Failed to prioritize actions to be taken during firefighting
drills.
C7. Misunderstanding of fire safety procedures.
C8. Unorganized responding to fire related emergency situation.
C9. Violated firefighting training and practice.
C10. Failed to fulfilment of designated responsibilities in fire
prevention.
C11. Failed to properly use of firefighting equipment and
appliances.
C12. Failed to test and maintain standby arrangements of fire
frightening alarm and equipment.
C13. Use of fire firefighting’s tool/equipment with a known
defect.
C14. Incorrect placement of portable tools, equipment or material in firefighting system.
C15. Lack of safety culture about the use of personnel protective
equipment.
C16. Disable or remove safe guards, warning system or safety
devices.
C17. Inappropriate team integration and discipline in
firefighting.
C18. Lack of information due to poor emergency
communication.
C19. Distributed storage of materials and spares on board ship.
C20. Missing and wrong labelling on firefighting equipment and
appliances.
C21. Violation of drugs and alcohol policies on board ship.
Reviewing the fire related deficiency sample database, the possible
causes in ‘‘preconditions for unsafe acts level’’ are determined as
follows:
C22. Loss of situational awareness in fire safety on board.
C23. Misplaced motivation of crews on board.
C24. Impaired physiological states of crews on board.
C25. Lack of familiarization about fire safety.
C26. Excessive self-confidence of crew members.
C27. Physical fatigues of crews on board.
C28. Time constraints on crew members’ reaction in operational
level.
C29. Incompatible intelligence/aptitude of crews.
C30. Insufficient physical capability in emergency response and
actions.
C31. Failed to communicate among ship and shore based
organization in emergency situations.
C32. Failed to coordinate the actions during the fire related
emergency situations.
C33. Failed to conduct adequate operational planning and
briefing.
C34. Failed to use all available firefighting resources.
C35. Poor coordination of fire frightening equipment and
system.
C36. Lack of warnings and signals about fire safety.
C37. Increased concentration demands in fire related emergency situations.
C38. Poor safety attitudes of crew due to basic health problems
and illness.
C39. Impairment of crew due to drug, alcohol or medication.
C40. Other emotional overload of crews.
Reviewing the fire related deficiency sample database, the
possible causes in ‘‘unsafe supervision level’’ are determined as
follows:
C41. Failed to provide guidance’s about fire prevention system.
C42. Failed to provide dynamic operational plans against fire
situation on board.
C43. Failed to provide considerable level of supervision on
board.
C44. Failed to provide specific firefighting training along with
different scenario.
C45. Failed to ensure about qualification of crews embarkation
on board ship.
C46. Failed to continuously monitoring crew performance on
board.
C47. Failed to provide adequate audit time.
C48. Ill-defined rules and responsibilities in fire safety plans.
C49. Ignored crew resting hours.
C50. Failed to update/revise documentation in fire safety plans.
C51. Failed to identify fire related hazards on board ship.
C52. Failed to initiate fire safety related corrective actions.
C53. Failed to report unsafe fire prevention tendencies.
C54. Failed to comply with fire safety rules and regulations.
C55. Failed to collect data and evidences about fire safety
measurement.
C56. Lack of conditions assessment program for firefighting
equipment.
32 O. Soner et al. / Safety Science 77 (2015) 25–41
C57. Failure to correct repeating unsafe occurrences on board
ship.
C58. Lack of methodological tools and background to perform
technical safety analysis.
C59. Poor communication between supervisor and crew
members on board ship.
Reviewing the fire related deficiency sample database, the possible
causes in ‘‘organizational influences level’’ are determined as
follows:
C60. The integration problem of company safety policy into
operational level.
C61. Poor design of ship fire system components.
C62. Ergonomic design errors in fire safety installations.
C63. Lack of systematic personnel selection and recruitment
procedures.
C64. Lack of managerial skill in shore-based personnel.
C65. Financial resourcing/budget constraints to timely meet
running/operational expenses.
C66. Insufficient scope of crew training program.
C67. Ineffective promotion system for crews.
C68. Purchasing of firefighting equipment and appliances in low
quality.
C69. High level of documentation bureaucracy.
C70. Lack of policy to monitor the required revisions in safety
procedures.
C71. Inefficient fire safety communication planning.
C72. Fire control plans’ inconsistencies.
C73. Incorrect behaviour enforced by shipping companies.
C74. Excessive time pressure due to improper operational
scheduling.
C75. Lack of management tools to implement suitable preventive action planning on board.
C76. Lack of effective system to determine adequate risk control
options on board.
C77. Management review input data incompleteness.
C78. Lack of consistent improvement decisions and follow-up
actions in management review output.
4.3. Root cause analysis
In order to identify the initiating causes of the causal system
described above, a fuzzy cognitive map is constructed and analysed. First, the causal relationships between concepts are specified
using a self-administered questionnaire where domain experts are
asked to indicate for each ordered pair of distinct concepts (Ci, Cj)
whether, ceteris paribus, a change in Ci has a significant impact
on Cj. To express the degree of the causal relationship (weights)
between two concepts the experts use the linguistic scale given
in Fig. 5. Since the number of causes considered in our FCM model
is very high (i.e. 78 distinct causes), it becomes a difficult and
tedious task for experts to answer all pairwise questions (i.e.
78 ⁄ (78–1) = 6006) and the likelihood of the experts to introduce
erroneous data increases (Asan and Soyer, 2009). To overcome this
drawback and make the administration of the questionnaire more
manageable, the adjacency matrix is divided into 16 distinct
regions with respect to the four levels in HFACS, as shown in
Fig. 7. Seven different groups of experts from the academia and
industry are, then, assigned to one or more of these regions consistent with their area of expertise (see Fig. 7). These groups consists
of (i) Maritime researchers (Group #1), (ii) Maritime stakeholders
(Group #2), (iii) Port state control officers (Group #3), (iv) Ship
management executives (Group #4), (v) Safety researchers
(Group #5), (vi) Industrial engineers (Group #6), and (vii)
Experienced seagoing officers/engineers (Group #7). This approach
not only reduces the number of questions for each group of
experts, it also improves the accuracy of judgments and the overall
efficiency.
Next, weights obtained from a group of experts are combined
using the Max operator to produce the overall linguistic weights
and, thus, the group adjacency matrix. The overall linguistic
weights are then transformed to crisp values using the CoG
defuzzification method. The calculations involved in the aggregation and defuzzification process regarding the impact of
‘‘Insufficient scope of crew training program (C66)’’ on ‘‘Failed to
provide specific firefighting training along with different scenarios
(C44)’’ in Region 15 are illustrated below. The linguistic weights
obtained from two experts regarding this causal relationship are
‘‘moderately-strongly’’ and ‘‘strongly’’ (see Fig. 6). Thus, the CoG
for the overall linguistic weight is calculated as follows
z ¼
R 0:8
0:6
ðz0:6Þ
0:2 zdz þ R 0:9
0:8
ðz1Þ
0:2 zdz þ R 1
0:9
z0:8
0:2 zdz
R 0:8
0:6
z0:6
0:2 dz þ R 0:9
0:8
ðz1Þ
0:2 dz þ R 1
0:9
z0:8
0:2 dz ¼ 0:834
Note that, in this study, it is assumed that the experts have equal
weights of credibility. The adjacency matrices (for only Region 15)
of both experts’ and the resulting matrix with crisp values are
shown in Figs. 8–10, respectively.
In the following step, the direct relationships represented in the
aggregated adjacency matrix are examined. Using Eqs. (3) and (4),
the outdegree and indegree, in other words the cumulative
strengths of connections entering and exiting the concepts are calculated. These values, which serve to identify the role of each concept in the system, are depicted in Fig. 11. For example, it can be
suggested that the concepts C66, C73 and C74 are highly influential
causes, while C8, C10 and C11 are highly dependent on the rest of
Fig. 7. The partitioned adjacency matrix and the assigned expert groups.
O. Soner et al. / Safety Science 77 (2015) 25–41 33
the system. Since we are basically interested in identifying the
potential root causes, concepts deserving this characteristic should
fulfil the rules provided in Eqs. (5a) and (5b). In this particular case,
the average outdegree taken over the entire concept set is found to
be approximately 13.6. Accordingly, for concept i, if odðiÞ P 13:6
and IPIi P 2 then the concept will be considered as a potential root
cause. These rules are depicted in Fig. 11, where the dashed line
represents IPI ¼ 2 and the solid line represents the average outdegree. For example, C66 is a potential root cause, since
odð66Þ ¼ 28:1, idð66Þ ¼ 4:8 and IPI66 ¼ 28_1=4:8 ¼ 5:85. Table 1
summarizes the results of the direct relationships analysis.
Finally, according to the direct relationship analysis the potential
root causes identified are C21, C26, C48, C49, C58, C62, C64, C66,
C67, C71, C72, C73, C74, and C75.
A similar classification is performed in the analysis of indirect
relationships. Here, the diffusion of causal impacts through reaction paths and loops are considered to explore hidden root causes.
To do this, the adjacency matrix is raised to successive powers. In
this study, the adjacency matrix is raised to the sixth power ðq ¼ 6Þ
where the concepts’ order proves to be stable. The outdegree and
indegree values in each resulting matrix ðq ¼ 1; … ; 6Þ are then
normalized to enable a comparison among the results of successive
powers. For example, the normalized values of C66 for q ¼ 2 are
calculated using Eqs. (7) and (8) as follows
Nodð66Þ
2 ¼ maxi¼1…78fodðiÞg odð66Þ
2
maxi¼1…78fodðiÞ
2
g ¼ 33:3 342:3
513:2 ¼ 22:2
Nidð66Þ
2 ¼ maxi¼1…78fidðiÞg idð66Þ
2
maxi¼1…78fidðiÞ
2
g ¼ 37:3 56:1
496:4 ¼ 4:2
Consequently, the total outdegree and indegree values of C66,
which indicate the reachability of this concept, is calculated as
follows
Rodð66Þ ¼
P6
q¼1Nodð66Þ
q
6
¼ ð Þ 28:1 þ 22:2 þ 22:7 þ 22:4 þ 22:4 þ 22:4 =6 ¼ 23:4
C41 C42 C43 C44 C45 C46 C47 C48 C49 C50 C51 C52 C53 C54 C55 C56 C57 C58 C59
C60 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
C61 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.8 1.0 1.0 0.0 0.0 0.0
C62 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0 0.4 0.6 0.8 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0
C63 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.8 1.0 1.0 0.0 0.2 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8
C64 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0
C65 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.2 0.4 0.6 0.4 0.6 0.8 0.4 0.6 0.8 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0
C66 0.2 0.4 0.6 0.2 0.4 0.6 0.0 0.0 0.0 0.8 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.2 0.4 0.6 0.4 0.6 0.8 0.6 0.8 1.0 0.0 0.0 0.2 0.0 0.0 0.0 0.4 0.6 0.8 0.4 0.6 0.8 0.0 0.0 0.0
C67 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.8 1.0
C68 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.2 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0
C69 0.2 0.4 0.6 0.2 0.4 0.6 0.0 0.0 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.2 0.4 0.6 0.0 0.0 0.0 0.8 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.2 0.4 0.0 0.0 0.0 0.0 0.0 0.0
C70 0.0 0.0 0.0 0.0 0.2 0.4 0.2 0.4 0.6 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
C71 0.4 0.6 0.8 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.8 1.0 1.0
C72 0.2 0.4 0.6 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
C73 0.8 1.0 1.0 0.4 0.6 0.8 0.8 1.0 1.0 0.6 0.8 1.0 0.4 0.6 0.8 0.4 0.6 0.8 0.6 0.8 1.0 0.8 1.0 1.0 0.8 1.0 1.0 0.2 0.4 0.6 0.2 0.4 0.6 0.4 0.6 0.8 0.6 0.8 1.0 0.6 0.8 1.0 0.2 0.4 0.6 0.4 0.6 0.8 0.4 0.6 0.8 0.4 0.6 0.8 0.0 0.2 0.4
C74 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.8 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0
C75 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0
C76 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
C77 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0
C78 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0
Fig. 8. The adjacency matrix of Expert 1 for Region 15.
C41 C42 C43 C44 C45 C46 C47 C48 C49 C50 C51 C52 C53 C54 C55 C56 C57 C58 C59
C60 0.8 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
C61 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.8 1.0 1.0 0.0 0.0 0.0
C62 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.8 1.0 1.0 0.0 0.0 0.0 0.0 0.2 0.4 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0
C63 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.8 1.0 1.0 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.8 1.0 1.0
C64 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.0 0.0 0.2 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.2 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0
C65 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.2 0.2 0.4 0.6 0.6 0.8 1.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0
C66 0.4 0.6 0.8 0.4 0.6 0.8 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.4 0.6 0.8 0.4 0.6 0.8 0.4 0.6 0.8 0.4 0.6 0.8 0.0 0.0 0.0 0.4 0.6 0.8 0.4 0.6 0.8 0.0 0.0 0.0
C67 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.8 1.0 1.0
C68 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0
C69 0.0 0.0 0.2 0.0 0.0 0.2 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.2 0.4 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.0 0.0 0.0
C70 0.0 0.0 0.0 0.0 0.0 0.2 0.2 0.4 0.6 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
C71 0.4 0.6 0.8 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.8 1.0 1.0
C72 0.0 0.2 0.4 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
C73 0.8 1.0 1.0 0.6 0.8 1.0 0.8 1.0 1.0 0.8 1.0 1.0 0.6 0.8 1.0 0.6 0.8 1.0 0.6 0.8 1.0 0.8 1.0 1.0 0.8 1.0 1.0 0.2 0.4 0.6 0.0 0.2 0.4 0.2 0.4 0.6 0.6 0.8 1.0 0.6 0.8 1.0 0.4 0.6 0.8 0.4 0.6 0.8 0.6 0.8 1.0 0.4 0.6 0.8 0.0 0.0 0.2
C74 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.2 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0
C75 0.8 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.0 0.0 0.0
C76 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
C77 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.0 0.0 0.0 0.0 0.0 0.0
C78 0.0 0.0 0.0 0.4 0.6 0.8 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.6 0.8 1.0 0.0 0.0 0.0
Fig. 9. The adjacency matrix of Expert 2 for Region 15.
34 O. Soner et al. / Safety Science 77 (2015) 25–41
Ridð66Þ ¼
P6
q¼1Nidð66Þ
q
6
¼ ð Þ 4:8 þ 4:2 þ 3:8 þ 3:7 þ 3:7 þ 3:7 =6 ¼ 4:0
The total outdegree and indegree values of the concepts are
depicted in Fig. 12. To identify the potential root causes, the same
rules suggested in the direct relationships analysis are employed
here. In other words, those concepts fulfilling the rules
RodðiÞ P 11:9 and IPIi P 2 are considered as potential root causes
(see Fig. 12). Table 2 summarizes the results of the indirect relationships analysis, where C48, C58, C64, C65, C66, C67, C71, C72, C73,
and C74 are labeled as potential root causes. Notice that C65 is
one hidden root cause which is supposed to be unimportant with
respect to the direct relationships. Therefore, comparing the results
of the two analyses can help to confirm the importance of certain
concepts as potential root causes and can reveal hidden root causes
which are previously thought to be unimportant but play a critical
role because of indirect impacts.
In the final step of root cause analysis, qualitative simulations
are performed to analyze the transmission of influence along all
paths and observe changes initiated by the root causes. These simulations give an idea of the overall priorities of potential root
causes determined in the previous two steps. In this study, 78
alternative what-if scenarios are considered for the simulation of
the causal system. In each scenario, a FCM is first initialized, i.e.
the activation level of each concept in the map takes on a value
in the set {0,1} based on the choice of the concept to be analysed
for its initiating role in the causal system. For example, Að0Þ
1 ¼ [0
0000000000000000000000000000000000
000000000000000000000000000000 1 000
0 0 0 0 0 0 0 0 0] represents the initial vector state where only
the concept ‘‘Insufficient scope of crew training program’’ is
activated/fired. Then the concepts are set free to interact according
to Eq. (1); here, the hyperbolic tangent is used as the threshold
function. The iterations are repeated until the t = 4, where
Aðtþ1Þ
i AðtÞ
i 6 e ¼ 0:000001 for all i. Fig. 13 depicts the dynamic
behaviour of the concepts for scenario 66, where only C66 is activated in the initial state.
The results of the 78 different scenarios suggest that the system
converges toward a steady state in maximum four iterations, and
in all scenarios only 25 causes reach an activation level of exact
1 (the rest ends up between 0.98–0.99). A critical indicator in these
simulations is the number of concepts influenced in the early iterations by a particular concept. Those concepts influencing a higher
number of concepts in the early iterations are supposed to be more
likely a root cause. Consequently, the decision on the final list of
root causes is made by synthesizing the results of Step 3, 4 and
5. The results are provided in Table 3. The priorities are determined
based on the averages of rank orders of the scenarios with respect
to iterations one and two. For example, the most influential root
cause is ‘‘Incorrect behaviour enforced by shipping companies’’
which activates 41 concepts in the first iteration and 77 concepts
in the second. Finally, the root causes listed according to their
priorities are C73, C74, C66, C48, C58, C64, C72, C21, C65, C62,
C67, C71, C26, C75 and C49.
4.4. Integration (preventive action planning)
The model clearly reveals the common root causes of fire
related deficiencies based on their priorities stated as follows:
C41 C42 C43 C44 C45 C46 C47 C48 C49 C50 C51 C52 C53 C54 C55 C56 C57 C58 C59
C60 0.834 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
C61 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.166 0.0 0.0 0.7 0.0 0.0 0.0 0.933 0.0
C62 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.834 0.0 0.4 0.7 0.0 0.0
C63 0.0 0.0 0.0 0.0 0.933 0.5 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.711
C64 0.0 0.0 0.0 0.3 0.6 0.0 0.0 0.3 0.289 0.5 0.0 0.0 0.0 0.7 0.3 0.0 0.0 0.6 0.0
C65 0.0 0.5 0.0 0.289 0.5 0.7 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.289 0.7 0.0 0.0
C66 0.5 0.5 0.0 0.834 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.5 0.6 0.7 0.422 0.0 0.6 0.6 0.0
C67 0.0 0.0 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.834
C68 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.7 0.166 0.0 0.0 0.6 0.0
C69 0.289 0.289 0.289 0.0 0.0 0.0 0.422 0.3 0.0 0.834 0.0 0.0 0.289 0.0 0.4 0.0 0.166 0.0 0.0
C70 0.0 0.166 0.4 0.5 0.0 0.0 0.4 0.166 0.0 0.0 0.0 0.0 0.0 0.6 0.0 0.0 0.0 0.0 0.0
C71 0.6 0.5 0.0 0.0 0.0 0.0 0.0 0.6 0.0 0.0 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.0 0.933
C72 0.3 0.7 0.0 0.0 0.0 0.0 0.0 0.7 0.0 0.0 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.0 0.0
C73 0.933 0.7 0.933 0.834 0.7 0.7 0.8 0.933 0.933 0.4 0.3 0.5 0.8 0.8 0.5 0.6 0.7 0.6 0.166
C74 0.0 0.5 0.0 0.0 0.0 0.166 0.5 0.0 0.3 0.0 0.0 0.0 0.3 0.0 0.4 0.0 0.5 0.0 0.0
C75 0.834 0.0 0.0 0.3 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.5 0.0 0.289 0.0 0.0 0.289 0.0 0.0
C76 0.0 0.7 0.0 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.0 0.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0
C77 0.8 0.0 0.0 0.0 0.0 0.166 0.0 0.5 0.0 0.8 0.0 0.0 0.6 0.0 0.0 0.0 0.4 0.0 0.0
C78 0.0 0.7 0.0 0.0 0.7 0.0 0.0 0.0 0.0 0.0 0.166 0.0 0.0 0.0 0.0 0.0 0.0 0.8 0.0
Fig. 10. The aggregated adjacency matrix for Region 15.
Fig. 11. Influence-dependence chart for direct relationships (dashed line represents
IPI = 2, solid line represents xod ¼ 13:6).
O. Soner et al. / Safety Science 77 (2015) 25–41 35
Table 1
Results of the analysis of direct relationships.
Cause odðiÞ idðiÞ IPIi odðiÞ P 13:6 IPIi P 2 Potential root cause
C1 5.1 31.9 0.16 No No No
C2 8.7 14.7 0.59 No No No
C3 9.4 22.2 0.42 No No No
C4 11.0 15.1 0.73 No No No
C5 13.5 10.8 1.25 No No No
C6 7.1 19.1 0.37 No No No
C7 7.5 9.9 0.76 No No No
C8 6.5 37.3 0.17 No No No
C9 20.1 15.5 1.30 Yes No No
C10 5.5 35.5 0.15 No No No
C11 9.8 33.8 0.29 No No No
C12 10.9 17.9 0.61 No No No
C13 9.6 11.1 0.87 No No No
C14 10.9 12.1 0.90 No No No
C15 10.3 6.5 1.60 No No No
C16 14.6 10.1 1.45 Yes No No
C17 11.8 26.6 0.44 No No No
C18 9.5 9.6 0.99 No No No
C19 5.4 10.3 0.52 No No No
C20 9.5 10.1 0.94 No No No
C21 24.7 10.1 2.46 Yes Yes Yes
C22 9.9 19.3 0.52 No No No
C23 15.7 22.3 0.70 Yes No No
C24 14.7 15.8 0.93 Yes No No
C25 16.5 14.9 1.11 Yes No No
C26 14.6 6.9 2.13 Yes Yes Yes
C27 6.2 18.4 0.34 No No No
C28 13.2 6.2 2.13 No Yes No
C29 9.5 4.8 1.99 No No No
C30 4.1 7.4 0.55 No No No
C31 1.7 26.2 0.07 No No No
C32 7.9 30.4 0.26 No No No
C33 11.0 18.4 0.60 No No No
C34 9.7 28.6 0.34 No No No
C35 8.5 15.3 0.55 No No No
C36 7.6 10.1 0.75 No No No
C37 2.8 5.0 0.56 No No No
C38 6.9 8.5 0.81 No No No
C39 11.0 7.9 1.40 No No No
C40 11.9 8.6 1.38 No No No
C41 23.4 20.1 1.16 Yes No No
C42 12.0 19.5 0.62 No No No
C43 30.5 20.6 1.48 Yes No No
C44 13.9 16.3 0.85 Yes No No
C45 11.3 20.8 0.54 No No No
C46 9.0 13.8 0.65 No No No
C47 21.0 23.8 0.88 Yes No No
C48 24.7 7.6 3.27 Yes Yes Yes
C49 18.3 5.5 3.31 Yes Yes Yes
C50 8.0 10.3 0.78 No No No
C51 15.8 14.9 1.06 Yes No No
C52 24.8 16.1 1.54 Yes No No
C53 17.0 21.1 0.81 Yes No No
C54 14.3 24.6 0.58 Yes No No
C55 23.7 23.8 1.00 Yes No No
C56 12.4 7.9 1.57 No No No
C57 28.8 23.8 1.21 Yes No No
C58 18.4 7.9 2.34 Yes Yes Yes
C59 20.7 15.4 1.35 Yes No No
C60 15.6 17.8 0.88 Yes No No
C61 13.1 1.8 7.29 No Yes No
C62 15.5 4.5 3.43 Yes Yes Yes
C63 9.5 3.1 3.02 No Yes No
C64 16.6 3.6 4.61 Yes Yes Yes
C65 12.8 6.3 2.02 No Yes No
C66 28.1 4.8 5.85 Yes Yes Yes
C67 21.0 3.2 6.57 Yes Yes Yes
C68 12.8 7.1 1.79 No No No
C69 9.1 2.6 3.55 No Yes No
C70 12.9 3.4 3.75 No Yes No
C71 18.2 5.8 3.12 Yes Yes Yes
C72 20.7 6.9 3.01 Yes Yes Yes
C73 33.3 7.0 4.73 Yes Yes Yes
C74 26.3 4.1 6.34 Yes Yes Yes
36 O. Soner et al. / Safety Science 77 (2015) 25–41
1. Incorrect behaviour enforced by shipping companies (C73).
2. Excessive time pressure due to improper operational
scheduling (C74).
3. Insufficient scope of crew training program (C66).
4. Ill-defined rules and responsibilities in fire safety plans
(C48).
5. Lack of methodological tools and background to perform
technical safety analysis (C58).
6. Lack of managerial skill in shore-based personnel (C64).
7. Fire control plans’ inconsistencies (C72).
8. Violation of drugs and alcohol policies on board ship (C21).
9. Financial resourcing/budget constraints to timely meet running/operational expenses (C65).
10. Ergonomic design errors in fire safety installations (C62).
11. Ineffective promotion system for crews (C67).
12. Inefficient fire safety communication planning (C71).
13. Excessive self-confidence of crew members (C26).
14. Lack of management tools to implement suitable preventive
action planning on board (C75).
15. Ignored crew resting hours (C49).
Now, it is an onerous task to successfully eliminate the featured
factors via suggesting effective preventive actions systematically.
In this stage, so-called integration, preventive actions are explored
rather than relatively simple corrective actions which are mainly
preferred in majority of ship fleets. Hence, it requires a genius
approach to produce comprehensive solutions especially along
with the latent error sources. For instance, in unsafe act level, violation of drugs and alcohol policies on board ship (C21) is found as
the most contributing factor since it directly influences the
cognitive, physical and mental performance during operations.
The nature of fire events on board ships extremely require team
integrity which might be influenced by excessive self-confidence
of crew members (C26). On the other hand, the combination of
inefficient emergency communication (C71), ill-defined rules and
responsibilities in fire safety plans (C48) and relevant inconsistencies (C72) will reduce the response level to the complex situations.
In addition to the key operational challenges that many
organizations are recently facing, such as crew resting hours
(C49), excessive time constraints (C74), and running costs (C65),
the design phase should also consider additional ergonomic
aspects (C62) to support safety level at sea. Furthermore, a shipowner tendency in terms of affecting organizational behaviour of
company (C73) is one of the dominating factors whose effects
are widely seen in each level. As another core aspect, operating
crew qualifications, highly depend on consistent promotion (C66)
and effective training (C67), should be enough to overcome
complex hazardous occurrences. Finally, managerial capabilities
(C64) supporting with advance analysis and execution tools (C58,
C75) have been playing a crucial role to ensure continuous
improvement of fire-fighting capability on board ships.
It can be easily seen that the determined root causes directly fall
within the scope of prevention, mitigation, preparedness, response
and recovery strategies against fire related emergency situations.
Establishing a continuous proactive system is the final step of this
research. Considering the determined root causes, ship operating
environment and recent marine technologies, it is a great necessity
to produce effective prevention actions comprehensively. That
means, a preventive action proposal should be in generic form,
originated from certain root causes, and applicable to various types
of fleet in order to prevent recurrence of focused operational facts.
Table 4 provides the suggested mechanisms as preventive action
proposals in accordance with their priorities.
The suggested mechanism includes Safe Ship System
Mechanism (SSSM), Safe Ship Operation Mechanism (SSOM) and
Safe Ship Execution Mechanism (SSEM). The priorities of root
causes derived from qualitative simulation application are used
to scheduling of preventive action proposals produced within the
suggested mechanism. Besides ensuring a systematic approach,
the mechanisms also classify the root causes and corresponding
solutions into design, operation or management perspectives.
For instance, the function of SSSM is to ensure system operability, maintainability, equipment reliability via testing arrangements, design and installation monitoring. Especially, the
ergonomics of user interfaces are the main concern of SSSM. In
practice, the SSSM provides invaluable feedbacks to the design
and construction phases of fire control and safety system at new
buildings as well as to the reconstruction, conversion, repair and
maintenance processes of existing ships in fleet. Hence, SSSM provides continuous response against defective elements in system
level, leading to firefighting vulnerabilities. To achieve that, the
SSSM requires establishing and integrating the following subsystems: (i) safe design and installation feedback system, (ii) operability monitoring system, (iii) equipment reliability assessment
system.
On the other hand, SSOM mainly targets to improve
safety awareness in shipboard operations. For example, an advance
scheduling and responsibility allocation system should be
Table 1 (continued)
Cause odðiÞ idðiÞ IPIi odðiÞ P 13:6 IPIi P 2 Potential root cause
C75 14.0 3.0 4.65 Yes Yes Yes
C76 10.9 4.7 2.33 No Yes No
C77 10.0 11.0 0.91 No No No
C78 11.3 4.6 2.45 No Yes No
Fig. 12. Influence-dependence chart for indirect relationships (dashed line
represents IPI = 2, solid line represents xRod ¼ 11:9).
O. Soner et al. / Safety Science 77 (2015) 25–41 37
Table 2
Results of the analysis of indirect relationships.
Cause Rod(i) Rid(i) IPIi RodðiÞ P 11:9 IPIi P 2 Potential root cause
C1 3.4 29.4 0.12 No No No
C2 5.8 13.8 0.42 No No No
C3 5.9 21.2 0.28 No No No
C4 9.4 15.9 0.59 No No No
C5 14.6 11.2 1.30 Yes No No
C6 5.5 17.9 0.31 No No No
C7 5.4 6.1 0.89 No No No
C8 5.0 37.3 0.13 No No No
C9 16.6 14.5 1.14 Yes No No
C10 4.0 36.3 0.11 No No No
C11 8.1 32.8 0.25 No No No
C12 10.2 18.1 0.56 No No No
C13 10.4 12.1 0.86 No No No
C14 9.9 11.4 0.87 No No No
C15 8.9 7.5 1.19 No No No
C16 13.9 12.0 1.17 Yes No No
C17 9.2 28.9 0.32 No No No
C18 5.5 9.6 0.57 No No No
C19 5.5 11.8 0.47 No No No
C20 9.3 10.6 0.88 No No No
C21 20.5 11.4 1.80 Yes No No
C22 9.0 16.7 0.54 No No No
C23 15.7 21.2 0.74 Yes No No
C24 15.9 12.0 1.33 Yes No No
C25 14.0 14.7 0.96 Yes No No
C26 14.3 9.7 1.47 Yes No No
C27 4.6 10.7 0.43 No No No
C28 11.2 4.7 2.40 No Yes No
C29 8.0 4.0 2.02 No Yes No
C30 3.5 6.4 0.54 No No No
C31 1.6 27.8 0.06 No No No
C32 6.8 34.2 0.20 No No No
C33 8.9 18.8 0.47 No No No
C34 9.9 30.3 0.33 No No No
C35 7.5 12.8 0.59 No No No
C36 6.8 10.5 0.64 No No No
C37 2.6 4.3 0.61 No No No
C38 6.2 9.4 0.66 No No No
C39 9.0 7.9 1.14 No No No
C40 10.1 9.9 1.02 No No No
C41 19.9 22.4 0.89 Yes No No
C42 8.9 23.6 0.38 No No No
C43 28.2 23.1 1.22 Yes No No
C44 10.4 21.6 0.48 No No No
C45 10.0 24.1 0.42 No No No
C46 11.7 14.9 0.79 No No No
C47 19.9 25.4 0.78 Yes No No
C48 21.6 7.7 2.80 Yes Yes Yes
C49 12.5 6.2 2.00 Yes No No
C50 9.6 12.8 0.75 No No No
C51 14.2 18.1 0.78 Yes No No
C52 23.5 20.6 1.14 Yes No No
C53 15.0 25.8 0.58 Yes No No
C54 16.5 26.4 0.63 Yes No No
C55 23.0 28.2 0.82 Yes No No
C56 11.5 9.0 1.28 No No No
C57 26.1 25.9 1.01 Yes No No
C58 15.2 5.9 2.60 Yes Yes Yes
C59 21.7 14.2 1.53 Yes No No
C60 10.6 16.4 0.65 No No No
C61 9.1 1.3 7.05 No Yes No
C62 11.5 2.7 4.22 No Yes No
C63 9.5 2.1 4.51 No Yes No
C64 16.0 2.6 6.15 Yes Yes Yes
C65 13.2 3.9 3.40 Yes Yes Yes
C66 23.4 4.0 5.89 Yes Yes Yes
C67 14.4 2.1 6.70 Yes Yes Yes
C68 9.7 5.0 1.93 No No No
C69 9.0 1.9 4.76 No Yes No
C70 11.2 3.1 3.65 No Yes No
C71 13.9 3.4 4.10 Yes Yes Yes
C72 15.3 4.0 3.80 Yes Yes Yes
C73 33.3 5.2 6.35 Yes Yes Yes
C74 21.6 4.2 5.09 Yes Yes Yes
38 O. Soner et al. / Safety Science 77 (2015) 25–41
established to control the time constraints on operational processes. Identification of suitable training items is another critical
point of interest which can be supported by problem-based training including regulatory amendments rather than traditional
safety trainings. The experienced operational safety cases should
be analysed and shared as critical lessons to be learned. The
motivating factors (i.e. resting hours’ compliance, fair promotion
process, etc.) should be considered and systematically actualized
in order to increase the number of good practices on board. To
address the mentioned issues, the SSOM should be supported with
the following sub-systems: (i) safe operation verification system,
(ii) crew improvement program, (iii) safety regulation compliance
system.
At the organizational level, the SSEM deals with governing the
overall process of fire safety improvement at sea. The mechanism
might require organizational redesign to avoid the incorrect
organizational behaviour, ordinary policies and management
practices. It is the most significant issue to be addressed. It might
Table 2 (continued)
Cause Rod(i) Rid(i) IPIi RodðiÞ P 11:9 IPIi P 2 Potential root cause
C75 10.2 2.3 4.43 No Yes No
C76 8.7 4.4 1.99 No No No
C77 9.9 11.2 0.89 No No No
C78 8.1 4.3 1.87 No No No
Fig. 13. Results of the FCM simulation process for Scenario 66.
Table 3
Priorities of potential root causes.
Scenario Activated cause # of Activated concepts (>0.5) Rank order w.r.t. I1 Rank order w.r.t. I2 Priority
Iteration 1 (I1) Iteration 2 (I2) Iteration 3 (I3)
21 C21 29 65 77 4.5 10.5 8
26 C26 16 65 77 14 10.5 13
48 C48 29 72 77 4.5 6 4
49 C49 19 52 77 11 15 15
58 C58 22 74 77 8 5 5
62 C62 20 66 77 10 9 10
64 C64 17 77 77 12.5 1.5 6
65 C65 14 76 77 15 3.5 9
66 C66 35 71 77 2.5 7 3
67 C67 26 62 77 7 13.5 11
71 C71 21 62 77 9 13.5 12
72 C72 27 70 77 6 8 7
73 C73 41 77 77 1 1.5 1
74 C74 35 76 77 2.5 3.5 2
75 C75 17 63 77 12.5 12 14
O. Soner et al. / Safety Science 77 (2015) 25–41 39
critically affect the management review decisions, utilization of
methodological tools, even existing managerial skills, and specific
policies. In this cycle, the mentioned attempts will improve the
proactive fleet management capability. Since the majority of
deficiencies are caused by a combination of root causes, strict
safety barriers at organizational level have the potential to prevent
various root causes to develop and interact. At organization level,
the SSEM necessitates to adopt the following sub-systems: (i)
safety performance measurement system, (ii) safety management
database, (iii) operational problem analysis and solution system.
5. Conclusion
Managing safety at sea is a complex problem which requires
genius onsite solutions. In fact, the safety level on board ships
can be enhanced via two key approaches: (i) eliminate the potential causes leading to event, (ii) strength the response and preparedness level. It can be assumed that if the operators (crew)
and organizations (executives) can manage both targets, the
probability and severity of accidents (i.e. fire on board) would be
minimized. This study proposes a novel proactive modelling
approach that intends to prevent reoccurrence of the fire related
deficiencies or accidents. It utilizes HFACS and FCM model to
scientifically analyse the fire related deficiency database. In addition, qualitative simulations are performed to verify and prioritize
the derived root causes. The integration phase of the proactive
model substantially reveals three mechanisms such as SSSM,
SSOM, SSEM. Various sub-systems (i.e. equipment reliability
assessment system, crew improvement program, operational problem analysis and solution system, etc.) are suggested in detail. The
main issues addressed in this paper can be summarized as follows:
(i) provide proactive safety modelling towards fire related deficiencies on board ship, (ii) identify and prioritize the consistent root
causes, (iii) apply HFACS–FCM to maritime safety literature, (iv)
promote human element on board ships, and (v) encourage the
maritime researchers to produce genius fire safety systems. The
main idea behind the paper is to analyse the operational data to
strengthen the organizational safety barriers. Besides active reasons, identification of the latent factors is recognized as an onerous
task. Hence, the methodological background in this research
addresses the mentioned expectations. Consequently, the paper
contributes to consistent prediction of the root causes while the
proposed proactive model strengths the safety barriers and firefighting capability in ship fleets. Hence, this study provides
reasonable contributions to safety improvements at sea.
Furthermore, detailed projections of suggested safety mechanisms
and sub-systems can be studied in order to manage the integration
in operational level.
Acknowledgements
This article is produced from MSc thesis research entitled ‘‘A
human factor analysis approach to prevent fire safety related
deficiencies on board ships’’ which has been executed in MSc
Program in Maritime Transportation Engineering of Istanbul
Technical University Graduate School of Science, Engineering and
Technology. The authors would like to express their gratitude to
various expert professionals (i.e. Maritime researchers (Group
#1), (ii) Maritime stakeholders (Group #2), (iii) Port state control
officers (Group #3), (iv) Ship management executives (Group #4),
(v) Safety researchers (Group #5), (vi) Industrial engineers
(Group #6), and (vii) Experienced seagoing officers/engineers
(Group #7)) for providing technical knowledge support to the
research.
References
Abou, S.C., 2012. Fuzzy-logic-based network for complex systems risk assessment:
application to ship performance analysis. Accident; Anal. Prevent. 45, 305–316.
ABS, 2012. Reducing the Port State Detention Factor, Pre-Port Arrival Quick
Reference Guide, October 2012.
Akhtar, M.J., Utne, I.B., 2014. Human fatigue’s effect on the risk of maritime
groundings – a Bayesian network modeling approach. Saf. Sci. 62, 427–440.
Akyuz, E., Celik, M., 2014a. Utilisation of cognitive map in modelling human error in
marine accident analysis and prevention. Saf. Sci. 70, 19–28.
Akyuz, E., Celik, M., 2014b. A hybrid decision-making approach to measure
effectiveness of safety management system implementations on-board ships.
Saf. Sci. 68, 169–179.
Antao, P., Guedes Suares, C., Grande, O., Trucco, P., 2008. Analysis of maritime
accident data with BBN models. Safety, Reliability and Risk Analysis: Theory,
Methods and Applications 2, 3265–3273.
Asan, U., Soyer, A., 2009. Identifying strategic management concepts: an analytic
network process approach. Comput. Ind. Eng. 56, 600–615.
Asan, U., Bozdag˘, C.E., Polat, S., 2004. A fuzzy approach to qualitative cross-impact
analysis. Omega 32, 443–458.
Asan, U., Kutlu, A.C., Kadaifci, Ç., 2011. Analysis of critical factors in energy service
contracting using fuzzy cognitive mapping. In: Proceedings of the 41st
International Conference on Computers and Industrial Engineering, October
23–26, 2011, Los Angeles, USA.
Asariotis, R., Benamara, H., Hoffmann, J., Jaimurzina, A., Premti, A., María Rubiato, J.,
Youssef, F., 2013. Review of Maritime Transport, 2013 (No. UNCTAD/RMT/
2013).
Balmat, J.-F., Lafont, F., Maifret, R., Pessel, N., 2009. MAritime RISk Assessment
(MARISA), a fuzzy approach to define an individual ship risk factor. Ocean Eng.
36 (15–16), 1278–1286.
Balmat, J.-F., Lafont, F., Maifret, R., Pessel, N., 2011. A decision-making system to
maritime risk assessment. Ocean Eng. 38, 171–176.
Celik, M., 2009. Designing of integrated quality and safety management system
(IQSMS) for shipping operations. Saf. Sci. 47, 569–577.
Celik, M., Cebi, S., 2009. Analytical HFACS for investigating human errors in shipping
accidents. Accident; Anal. Prevent. 41, 66–75.
Celik, M., Er, I.D., 2007. Identifying the potential roles of design-based failures on
human errors in shipboard operations. In: 7th Navigational Symposium on
Marine Navigation and Safety of Sea Transportation, 20–22 June, Gdynia,
Poland, pp. 617–621.
Celik, M., Miri, S., Wang, J., 2010. A risk-based modelling approach to enhance
shipping accident investigation. Saf. Sci. 48, 18–27.
Chauvin, C., Lardjane, S., Morel, G., Clostermann, J., Langard, B., 2013. Human and
organisational factors in maritime accidents: analysis of collisions at sea using
the HFACS. Accid. Anal. Prev. 59, 26–37.
ClassNK, 2013. Port State Control Annual Report, August 2013.
Craiger, J.P., Goodman, D.F., Weiss, R.J., Butler, A., 1996. Modeling organizational
behavior with fuzzy cognitive maps. Int. J. Comput. Intell. Org., 120–123
Darbra, R.M., Casal, J., 2004. Historical analysis of accidents in seaports. Saf. Sci. 42,
85–98.
DNV, 2012. Port state control reports: synopsis of frequent findings and detention
items, DNV Serving the Maritime Industry, January 2012.
Eide, M.S., Endresen, O., Brett, P.O., Ervik, J.L., Røang, K., 2007. Intelligent ship traffic
monitoring for oil spill prevention: risk based decision support building on AIS.
Mar. Pollut. Bull. 54, 145–148.
Gausemeier, J., Fink, A., Schlake, O., 1996. Szenario-Management: Planen und
Führen mit Szenarien, 2. Auflage. Carl Hanser Verlag, München.
Godet, M., 1994. From Anticipation to Action: A Handbook of Strategic Prospective.
UNESCO Publishing, France.
Hänninen, M., Valdez Banda, O., Kujala, P., 2014. Bayesian network model of
maritime safety management. Expert Syst. Appl. 41, 7837–7846.
Hee, D.D., Pickrell, B.D., Bea, R.G., Roberts, K.H., Williamson, R.B., 1999. Safety
management assessment system (SMAS): a process for identifying and
evaluating human and organization factors in marine system operations with
field test results. Reliabil. Eng. Syst. Saf. 65, 125–140.
Hetherington, C., Flin, R., Mearns, K., 2006. Safety in shipping: The human element.
J. Saf. Res. 37, 401–411.
John, A., Paraskevadakis, D., Bury, A., Yang, Z., Riahi, R., Wang, J., 2014. An integrated
fuzzy risk assessment for seaport operations. Saf. Sci. 68, 180–194.
Table 4
Preventive action proposals.
Suggested mechanisms Eliminated root causes
Application priorities
1234567
Safe Ship System Mechanism
(SSSM)
C72 C62 C71
Safe Ship Operation
Mechanism (SSOM)
C74 C66 C48 C65 C67 C26 C49
Safe Ship Execution
Mechanism (SSEM)
C73 C58 C64 C21 C75
40 O. Soner et al. / Safety Science 77 (2015) 25–41
Karahalios, H., 2014. The contribution of risk management in ship management: the
case of ship collision. Saf. Sci. 63, 104–114.
Karahalios, H., Yang, Z.L., Williams, V., Wang, J., 2011. A proposed system of
hierarchical scorecards to assess the implementation of maritime regulations.
Saf. Sci. 49, 450–462.
Kelangath, S., Das, P.K., Quigley, J., Hirdaris, S.E., 2011. Risk analysis of damaged
ships – a data-driven Bayesian approach. Ships Offshore Struct., 1–15
Knapp, S., Franses, P.H., 2009. Does ratification matter and do major conventions
improve safety and decrease pollution in shipping? Marine Policy 33 (5), 826–
846.
Knudsen, O.F., Hassler, B., 2011. IMO legislation and its implementation: accident
risk, vessel deficiencies and national administrative practices. Marine Policy 35
(2), 201–207.
Kosko, B., 1986. Fuzzy cognitive maps. Int. J. Man Mach. Stud. 24, 65–75.
Kosko, B., 1988. Hidden patterns in combined and adaptive knowledge networks.
Int. J. Approximate Reasoning 2, 377–393.
Kosko, B., 1992. Neural Networks and Fuzzy Systems. Prentice Hall, New Jersey.
Kosko, B., 1997. Fuzzy Engineering. Prentice Hall, New Jersey.
Kuo, H.C., Chang, H.K., 2003. A real-time shipboard fire-detection system based on
grey-fuzzy algorithms. Fire Saf. J. 38, 341–363.
Lambrou, M.a., Fjørtoft, K.E., Sykas, E.D., Nikitakos, N., 2008. Ambient intelligence
technologies in support of shipping markets’ operations. Telematics Inform. 25,
72–83.
Lazzerini, B., Mkrtchyan, L., 2011. Analyzing risk impact factors using extended
fuzzy cognitive maps. IEEE Syst. J. 5 (2), 288–297.
Lee, J.-O., Yeo, I.-C., Yang, Y.-S., 2001. A trial application of FSA methodology to the
hatchway watertight integrity of bulk carriers. Mar. Struct. 14, 651–667.
Li, K.X., Zheng, H., 2008. Enforcement of law by the Port State Control (PSC).
Maritime Policy Manage. 35 (1), 61–71.
LiPing, C., GuoJun, P., XingGu, Z., 2011. The application and research of navigationaids inspection and maintenance based on video surveillance. Proc. Eng. 15,
3088–3092.
Lois, P., Wang, J., Wall, a., Ruxton, T., 2004. Formal safety assessment of cruise ships.
Tourism Manage. 25, 93–109.
Lun, Y.H.V., Wong, C.W.Y., Lai, K., Cheng, T.C.E., 2008. Institutional perspective on
the adoption of technology for the security enhancement of container transport.
Transport Rev. 28, 21–33.
Nasserzadeh, S.M.R., Jafarzadeh, M.H., Mansouri, T., Sohrabi, B., 2008. Customer
satisfaction fuzzy cognitive map in banking industry. Commun. IBIMA 2, 151–
162.
Nozicka, G.J., Bonham, G.M., Shapiro, M.J., 1976. Simulation techniques. In: Axelrod,
R., (Ed.), Structure of Decision, Princeton University Press, Princeton, NJ.
O’Neil, W.A., 2003. The human element in shipping. World Maritime Univ. J.
Maritime Affairs 2, 95–97.
Özesmi, U., Özesmi, S.L., 2004. Ecological models based on people’s knowledge: a
multi-step fuzzy cognitive mapping approach. Ecol. Model. 176, 43–64.
Pam, E.D., Li, K.X., Wall, a., Yang, Z., Wang, J., 2013. A subjective approach for ballast
water risk estimation. Ocean Eng. 61, 66–76.
Papageorgiou, E.I., 2011. A new methodology for decisions in medical informatics
using fuzzy cognitive maps based on fuzzy rule-extraction techniques. Appl.
Soft Comput. 11, 500–513.
Papageorgiou, E., Kontogianni, A., 2012. Using fuzzy cognitive mapping in
environmental decision making and management: a methodological primer
and an application. In: Young, S. (Ed.), International Perspectives on Global
Environmental Change. InTech, Croatia, pp. 427–450.
Papageorgiou, E.I., Stylios, C.D., 2008. Fuzzy cognitive maps. In: Pedrycz, W.,
Skowron, A., Kreinovich, V. (Eds.), Handbook of Granular Computing. John Wiley
& Sons Ltd., West Sussex.
Papageorgiou, E.I., Markinos, A., Gemptos, T., 2009. Application of fuzzy cognitive
maps for cotton yield management in precision farming. Expert Syst. Appl. 36,
12399–12413.
Papakostas, G.A., Boutalis, Y.S., Koulouriotis, D.E., Mertzios, B.G., 2008. Fuzzy
cognitive maps for pattern recognition applications. Int. J. Pattern Recognit.
Artif. Intell. 22, 1461–1486.
Paris MoU, 2012. Taking port state control to the next level, Port State Control
Annual Report.
Patterson, J.M., Shappell, S.A., 2010. Operator error and system deficiencies: analysis
of 508 mining incidents and accidents from Queensland, Australia, using HFACS.
Accid. Anal. Prev. 42, 1379–1385.
Reason, J., 1990. Human Error. Cambridge University Press, New York.
Reinach, S., Viale, A., 2006. Application of a human error framework to conduct train
accident/incident investigations. Accid. Anal. Prev. 38, 396–406.
Ross, T.J., 2004. Fuzzy Logic with Engineering Applications, second ed. McGrawHill, New York.
Rothblum, A.R., 2000. Human Error and Marine Safety. National Safety Council
Congress and Expo, Orlando.
Rothblum, A., Wheal, D., Withington, S., Shappell, S.A., Wiegmann, D.A., 2002.
Improving incident investigation through inclusion of human factors. United
States Department of Transportation-Publications and Papers (Paper 32).
Saaty, T.L., 2004. Decision making – the analytic hierarchy and network processes
(AHP/ANP). J. Syst. Sci. Syst. Eng. 13 (1), 1–34.
Schinas, O., Stefanakos, C.N., 2012. Cost assessment of environmental regulation
and options for marine operators. Transport. Res. Part C: Emerg. Technol. 25,
81–99.
Schröder-Hinrichs, J.U., Baldauf, M., Ghirxi, K.T., 2011. Accident investigation
reporting deficiencies related to organizational factors in machinery space
fires and explosions. Accid. Anal. Prev. 43 (3), 1187–1196.
Serdar Asan, S., Asan, U., 2007. Qualitative cross-impact analysis with time
consideration. Technol. Forecast. Soc. Chang. 74, 627–644.
Shappell, S.A., Wiegmann, D.A., 2000. The human factors analysis and classification
system – HFACS. Federal Aviation Administration Technical Report No. DOT/
FAA/AM-00/7. National Technical Information Service, N Springfield.
Shappell, S.A., Wiegmann, D., 2001. Applying reason: the human factors analysis
and classification system. Human Fact. Aerospace Saf. 1, 59–86.
Shappell, S.A., Detwiler, C., Holcomb, K., Hackworth, C., Boquet, A., Wiegmann, D.A.,
2007. Human error and commercial aviation accidents: an analysis using the
human factors analysis and classification system. Hum. Factors 49, 227–242.
Sii, H.S., Ruxton, T., Wang, J., 2001. A fuzzy-logic-based approach to qualitative
safety modelling for marine systems. Reliabil. Eng. Syst. Saf. 73, 19–34.
Stach, W., Kurgan, L., Pedrycz, W., Reformat, M., 2005. Genetic learning of fuzzy
cognitive maps. Fuzzy Sets Syst. 153, 371–401.
Stylios, C.D., Groumpos, P.P., 2004. Modeling complex systems using fuzzy cognitive
maps. IEEE Trans. Syst., Man, Cybernet.—Part A: Syst. Humans 34 (1), 155–162.
Stylios, C.D., Groumpos, P.P., 2000. Fuzzy cognitive maps in modeling supervisory
control systems. J. Intell. Fuzzy Syst. 8, 83–98.
Stylios, C.D., Georgopoulos, V.C., Groumpos, P.P., 1997. Introducing the theory of
fuzzy cognitive maps in distributed systems. In: Proceedings of 12th IEEE
International Symposium on Intelligent Control, Istanbul, Turkey, 1997, pp. 55–
60.
Toffoli, A., Lefevre, J.M., Bitner-Gregersen, E., Monbaliu, J., 2005. Towards the
identification of warning criteria: analysis of a ship accident database. Appl.
Ocean Res. 27, 281–291.
Tokyo MoU, 2013. Annual Report on Port State Control in the Asia-Pacific Region.
Trucco, P., Cagno, E., Ruggeri, F., Grande, O., 2008. A Bayesian belief network
modelling of organisational factors in risk analysis: a case study in maritime
transportation. Reliabil. Eng. Syst. Saf. 93, 845–856.
Tsadiras, A.K., 2008. Comparing the inference capabilities of binary. Trivalent and
Sigmoid Fuzzy Cognitive Maps, Information Sciences 178, 3880–3894.
Tsadiras, A.K., Kouskouvelis, I., Margaritis, K.G., 2003. Using fuzzy cognitive maps as
a decision support system for political decisions. Adv. Inform. Lect. Notes
Comput. Sci. 2563, 172–182.
Tzannatos, E., 2005. Technical reliability of the Greek coastal passenger fleet. Mar.
Policy 29, 85–92.
Tzannatos, E., Kokotos, D., 2009. Analysis of accidents in Greek shipping during the
pre- and post-ISM period. Mar. Policy 33, 679–684.
Vanem, E., Ellis, J., 2010. Evaluating the cost-effectiveness of a monitoring system
for improved evacuation from passenger ships. Saf. Sci. 48, 788–802.
Vanem, E., Skjong, R., 2006. Designing for safety in passenger ships utilizing
advanced evacuation analyses—a risk based approach. Saf. Sci. 44, 111–135.
Vasantha Kandasamy, W.B., Smarandache, F., 2003. Fuzzy Cognitive Maps and
Neutrosophic Cognitive Maps, Xiquan, Phoenix.
Wang, J., 2002. Offshore safety case approach and formal safety assessment of ships.
J. Saf. Res. 33, 81–115.
Wang, J., Foinikis, P., 2001. Formal safety assessment of containerships. Mar. Policy
25 (2), 143–157.
Wang, H., Jiang, H., Yin, L., 2013. Cause mechanism study to human factors in
maritime accidents: towards a complex system brittleness analysis approach.
Proc. – Soc. Behav. Sci. 96, 723–727.
Wang, J., Li, M., Liu, Y., Zhang, H., Zou, W., Cheng, L., 2014. Safety assessment of
shipping routes in the South China Sea based on the fuzzy analytic hierarchy
process. Saf. Sci. 62, 46–57.
Wieslaw, T., 2012. Origins of ship safety requirements formulated by international
maritime organization. Proc. Eng. 45, 847–856.
Yaman, D., Polat, S., 2009. A Fuzzy Cognitive Map Approach for Effect-Based
Operations: An Illustrative Case. Inf. Sci. 179, 382–403.
Zhang, D., Yan, X.P., Yang, Z.L., Wall, a., Wang, J., 2013. Incorporation of formal safety
assessment and Bayesian network in navigational risk estimation of the Yangtze
River. Reliabil. Eng. Syst. Saf. 118, 93–105.
O. Soner et al. / Safety Science 77 (2015) 25–41 41

Published by

Medical

View all posts