Ocean Engineering 156 (2018) 333–346

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Operational risk assessment of offshore transport barges

Nagi Abdussamie a, *, Ashraf Zaghwan b, Mohamed Daboos c, Ismail Elferjani d, Ahmed Mehanna e,
Wenming Su f
a
National Centre for Maritime Engineering and Hydrodynamics, Australian Maritime College, University of Tasmania, Launceston, Tasmania 7250, Australia
b
Entrepreneurship, Commercialisation and Innovation Centre, The University of Adelaide, Adelaide, Australia
c
Marine and Offshore Engineering, Faculty of Engineering, University of Tripoli, Libya
d
RMIT University, Melbourne, Australia
e
Department of Marine Engineering, Arab Academy for Science & Technology & Maritime Transport, Alexandria, Egypt
f
Jiangsu Maritime Institute, Nanjing, China

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

Keywords: Offshore operations such as load-out/float-off, transportation and installation involve a large number of opera-
Transport barges tional hazards which impose high risks on human safety and assets. During such operations, there are many
Load-out situations in which transport barges and Heavy Lift Vessels (HLVs) may expose to significant hazards including
Float-off structural and mechanical failures. In this work, the system of a submersible barge was reviewed based on a
Risk assessment Hazard Identification (HAZID) technique in order to identify the worst-case scenarios during load-out/float-off
Fuzzy sets
operations. In addition, a fuzzy set approach was developed to quantify and assess the risk level during
offshore operations of an offshore structure due to several hazardous scenarios. Rule-based fuzzy logic models
were created and tested using different types of membership functions to calculate risk values, and the potential
hazard impacts on the safety of crew members, the environment, the barge and the offshore structure being
loaded/launched were evaluated. Fuzzy set techniques enabled a further sensitivity analysis to be conducted for
the top-ranked failure modes. Overall, the paper contributes towards the development of the current guidelines
for offshore operations.

1. Introduction
Safe transportation and/or installation of offshore structures such as
steel jackets and jack-up units is one of the greatest challenges in the
offshore industry. These structures are transported offshore either by dry-
or wet-tows. In severe weather conditions, the likelihood of trans-
portation accidents which include grounding, collision and loss of sta-
bility or buoyancy (Vinnem, 2007) can be extremely high which can
impose high risks on human safety and assets. Among several accidents,
the Marathon LeTourneau (Denton, 1989) and the West Gamma (Vin-
nem, 2007) jack-up units capsized when they were being towed. Another
example is the loss of stability due to flooding, which led to the capsizing
of the Bohai 2 jack-up rig and caused the death of 72 people. In 2006, at
the CNOOC (Offshore Oil 298) project, 68 people died in a towing vessel
accident during a typhoon (Fang and Duan, 2014). According to Gunter
et al. (2013), offshore transportation events were the leading cause of
fatalities of workers involved in offshore oil and gas operations in the US
during the period 2003–2010.
As the use of dry-tow techniques has been known to be safer than wet-
tows, most of load-out, transportation and float-off/launching of offshore
structures are performed using barges and Heavy Lift Vessels (HLVs) (Van
Hoorn, 2008). In doing so, the load-out and marine transportations of
offshore structures and topside modules are usually performed in
accordance with GL Noble Denton guidelines (GLND, 2005; GLND,
2009). However, such guidelines do not provide a detailed risk assess-
ment procedure which can be used for hazard identification and risk
mitigations. Nevertheless, during such operations, there are many situ-
ations in which transport barges can expose to significant hazards
including structural and mechanical failures. Therefore, the necessity for
developing new standards based on reliable knowledge is important to-
wards establishing a new milestone to assess and mitigate transport barge
and HLV risks.
The literature shows that most of the recent research efforts have
been focused on risks of conventional ships and offshore structures. On
the other hand, the safety of barges and HLVs has received less attention.
It is, therefore, important to assess the possible consequences of hazards

* Corresponding author.
E-mail address: nagia@utas.edu.au (N. Abdussamie).

https://doi.org/10.1016/j.oceaneng.2018.03.006
Received 30 November 2017; Received in revised form 17 January 2018; Accepted 3 March 2018

0029-8018/© 2018 Elsevier Ltd. All rights reserved.

N. Abdussamie et al. Ocean Engineering 156 (2018) 333–346

Fig. 1. A submersible barge with two stabilised col-

umns (Gerwick, 2002).

Unlike binary systems, the outcome of an operation is expressed as a

Table 1
probability in fuzzy logic systems. For instance, the outcome may mean
Typical dimensions of submersible barges.
“probably true”, “possibly true” or “probably false”. Fuzzy logic based
Overall length, LOA 80–160 m approaches have recently been used for different engineering problems
Breadth 0.20–0.33 LOA such as risk assessment of LNG carriers (Elsayed, 2009; Elsayed et al.,
Depth 0.07–0.08 LOA 2014), risk management of seaports and terminal (Mokhtari et al., 2012)
and expert systems (Samantra et al., 2014).
Elsayed et al. (2009) developed a fuzzy inference system to assess
on the operation and safety of offshore transport barges.
risks of LNG carriers during loading/offloading operations at terminals
Currently, there are numerous methods used for marine and offshore
and found that the use of a fuzzy set approach is particularly suited for
risk assessment, which can be broadly categorised into quantitative and
handling multiple attribute risk problems with imprecise data. Nwaoha
qualitative approaches (ABS, 2000). Quantitative risk assessment (QRA)
et al. (2013) developed a framework for the risks of hazards of LNG
techniques have been regarded in the industry as the only reliable
carrier operations using the combination of a risk matrix approach and a
sources (Wang and Pedersen, 2007). However, due to limitations with
fuzzy evidential reasoning method. Stavrou and Ventikos (2016) used a
accident/incident reports about the likelihood and severity of transport
process mode and effects analysis approach to evaluate different risk
barge and HLV hazards, the application of QRA techniques would be
scenarios related to a ship-to-ship transfer of petroleum cargo operations.
difficult. On the other hand, qualitative risk assessment approaches such
Zhao et al. (2015) applied a Bayesian network method to assess the risks
as the risk matrix technique assess the risk in descriptive terms by using
of accidents in the anchoring system of an LNG carrier. It is worth
experts' opinions. The use of the risk matrix has been adopted with the
mentioning that neither of the aforementioned studies investigated risks
assistance of experts' opinions from the offshore and shipping industries
of offshore operations such as load-outs or float-off/launching.
whose decisions on the degree of hazard are usually based on their
The aim of this paper is to present and discuss risk assessment results
experience in the field (Aronsson, 2012). Experts often use Hazard
of a transport (submersible) barge. In this study, load-out/float-off of an
Identification (HAZID) and Hazard and Operability (HAZOP) techniques
offshore structure, and the operational risks during the load-out/float-off
in order to cover a wide range of possible hazardous scenarios (Spouge,
phase were considered, whilst hazards due to weather conditions are
1999). The likelihood and consequence associated with a failure mode
beyond the scope of this paper. Potential hazards/operability difficulties
are weighted based on the experts' knowledge and combined by the risk
of the system were identified using the HAZID technique. The fuzzy set
matrix to obtain the risk level/index (Spouge, 1999; ABS, 2000). This
approach was developed to synthesise the risk levels of the basic failure
implies that the results of qualitative risk assessment techniques are ex-
modes to enable comparisons among top-ranked failure modes. A rule-
perts’ knowledge dependent “subjective”. Furthermore, the risk results
based fuzzy model using different types of membership functions was
obtained by these techniques are not often reproducible due to an un-
created to calculate and assess operational risks to four consequence
certain risk level. One more limitation with the application of qualitative
components which include on-board barge crew members, the barge, the
risk assessment techniques is that they rely on the use of discrete attri-
structure being loaded/launched and the environment.
butes, which do not account for uncertainty or vagueness of a hazardous
scenario (Elsayed, 2009).
2. Transport barges
Recently, the use of fuzzy logic in artificial intelligence applications
such as expert systems has been increasing. Fuzzy logic is a mathematical
The design of submersible barges used for offshore transportation is
tool for modelling inaccuracy and uncertainty of the real world and
quite similar to that of floating dry-docks, and therefore they can be
human thinking in which variables have degrees of falsehood or truth-
operated in floating and submersible modes (ABS, 2009). Submersible
fulness represented by a range of values between 0 (false) and 1 (true).

Fig. 2. Main steps of offshore transportation and

installation of a jacket structure.

334
N. Abdussamie et al. Ocean Engineering 156 (2018) 333–346

Table 2 Table 4
HAZID analysis of a submersible barge during load-out/float-off operations. Definition of frequency indices.
# Operation Guide Deviation Possible Possible Index Frequency Definition
words causes consequences
8 Very frequent Will occur very frequently
1 Positioning Collision/ Damage to e.g., tug Barge taking 7 Frequent Will occur frequently
Impact hull impact severe heel or 6 Probable May occur several times
trim. Barge 5 Reasonable probable May occur one or two times
sinking. 4 Occasional Likely to occur during lifetime
Drifting Out of e.g., broken Damage to 3 Remote Unlikely to occur during lifetime
position mooring barge and to 2 Very remote So unlikely event
line structure. Injury 1 Extremely remote It may not be experienced during lifetime
to barge crew.
2 Ballasting Flow/ Pump e.g., shut in It would reduce
pressure failure power flow rate. middle of the barge for the passage of pipes and control valves. These
supply Ballasting or de-
barges are also equipped with a pump room at one end of the tunnel to
ballasting may
take long time.
provide the required pumping capacity during ballasting and
Flow/ Valve e.g., control It can prevent de-ballasting operations (ABS, 2009). The depth control during sub-
pressure failure signal fault the ballasting or mergence is performed using draught sensors mounted on the barge at
de-ballasting designated locations from which the vertical height of water column
from continuing
above the sensor is measured and monitored.
as per the
ballast The main steps of the transportation of a structure such as steel jackets
sequence. from an onshore fabrication yard to offshore using submersible barges or
3 Power supply Load Generator e.g., Failure of one of HLVs can be seen in Fig. 2. The submersible barge considered in this work
variation/ failure mechanical two Generators was assumed to perform load-out/float-off operations for an offshore
black-out failure of would limit the
prime number of
structure, and operational risks during these operational phases were
mover equipment that considered, whilst hazards due to weather conditions are beyond the
can be operated scope of this paper. The potential hazard impacts on the safety of on-
simultaneously board barge crew members (denoted as C), the barge (denoted as B),
causing some
the offshore structure being loaded/launched such as steel jackets and
delay.
4 Control Flow/ Hydraulic e.g., seal Prevent valves jack-up units (denoted as S) and the environment (denoted as E) were
pressure leak failure and capstans assessed.
from operating,
or slow or limit 3. Hazard identification (HAZID) study
their operation.
Flow Hydraulic e.g., piping Environmental
fluid spill system impact and/or A HAZID study was undertaken to identify and determine potential
damage toxic effect. hazards that may significantly affect the safety and operability of sub-
5 Miscellaneous Flame/ Fire e.g., human A damage to mersible barges. These included deviations (or failure modes) and their
heat error barge and to
severity degrees to be determined. The significant hazards or operability
structure. Injury
to barge crew. difficulties of a system can be identified using the HAZID technique based
on the review of the system and guidewords (Pitt, 1994). The HAZID
process was carried out on the basis of a team-approach to hazard
analysis in order to identify the hazardous scenarios during
Table 3 load-out/float-off operations. For each credible failure mode, possible
A typical multi-level risk matrix adopted from (Østvik et al., 2005).
causes and consequences were considered, and the associated frequency
Frequency (F) Consequence (C) and consequence indices were obtained from several experts with vary-
Minor Significant Severe Catastrophic Disastrous ing backgrounds and expertise.
The process started by dividing the submersible barge into nodes; a
Very frequent L3 L3 L4 L4 L4
Frequent L3 L3 L3 L4 L4 node might cover ballast system failure such as a failure of a pump or a
Probable L2 L3 L3 L3 L4 valve. Five major operations, which are expected to be conducted at the
Reasonably L2 L2 L3 L3 L3 time of load-out/launching phases, were considered. The review of nodes
probable
was focused on the significance of potential hazards/failure modes in
Little probable L2 L2 L2 L3 L3
Remote L1 L2 L2 L2 L3
each operation. A brief description of each operation and its associated
Very remote L1 L1 L2 L2 L2 failure modes is given below. Table 2 summarises the results of HAZID
Extremely remote L1 L1 L1 L2 L2 study where each hazard/deviation and its possible causes and the po-
tential impacts are presented, whereas the associated frequency and
consequence indices assigned by experts are presented in Section 4.
barges are typically non-propelled vessels and not designed for complete
The incorrect installation of transported structures on-board barges
submergence (Gerwick, 2002). Therefore, they must be neutrally or
can cause several types of structural damage to the barge and/or the
negatively buoyant in order to launch structures. A typical submersible
structure. During launching a structure, the barge is designed to be
barge used for launching and offshore transportation is illustrated in
grounded or beached, thus a sufficient freeboard must be provided to
Fig. 1, which consists of a continuous clear main deck with two stabil-
ensure stability even if one side compartment or end compartment of the
ising columns at one end. Table 1 presents the typical range of the main
barge is flooded (Gerwick, 2002). During submergence, ballast tanks are
particulars of existing submersible barges where the breadth and depth
filled in a specified order and water is made to overflow through the
are given as a function of the overall length (Gerwick, 2002). As can be
outlets/vent pipes.
seen in Fig. 1, the space below the main deck is divided into a number of
Damage to hull. An impact may occur between the barge and the sea
ballast tanks by means of transverse and longitudinal watertight bulk-
floor (during submergence), which may lead to hull damage. Similarly,
heads. A longitudinal pipe tunnel is constructed inside the hull in the
an impact between tugboats and the barge may occur due to the loss of

335
N. Abdussamie et al. Ocean Engineering 156 (2018) 333–346

Table 5
Definition of consequence indices.
Class Minor (1) Significant (2) Severe (3) Catastrophic (4) Disastrous (5)

S Minor damage
B Minor damage and
delay< 24 h
E Negligible
pollution
C Minor injury
Moderate damage Major damage Extensive damage Total loss
Moderate damage and Major damage. Permanent Total loss (of, e.g., a small size of Total loss (of, e.g., a large size of
delay
24 h. Temporary repair is repair is required. barge) barge)
required.
Pollution reported to regulatory Pollution reported to regulatory Pollution reported to regulatory Pollution reported to regulatory
authorities “minor pollution” authorities “major pollution” authorities “severe pollution” authorities “uncontrolled pollution”
Serious injury One fatality Multiple fatalities Large number of fatalities

low.
Table 6 Out of position. During manoeuvrings, the mission of tugboats is to
Example of HAZID results. keep barges in a particular position. Tugboats at fore and aft are usually
FM# Failure mode Expert 1 Expert 2 moored by ropes alongside barges to provide a high level of controlla-
bility during manoeuvring. Meanwhile, other tugboats are to be in a
F index C index F index C index
stand-by position for supporting the operation of station keeping. Oper-
1 Damage to hull 1 3S, 4B, 2E, 2 3S, 4B, 1E,
ational difficulties appear in bad weather conditions, which may result in
3C 4C
2 Out of position 3 3S, 3B, 1E, 3 3S, 3B, 1E, breaking mooring lines. Additionally, when the manoeuvre is required to
3C 4C be done in the vicinity of other barges/ships, this situation may result in
3 Pump failure 7 2S, 3B, 1E, 7 3S, 3B, 1E, an out of position scenario, which increases the risk of collision.
1C 1C The ballast system of submersible barges often consists of a number of
4 Valve failure 6 2S, 3B, 1E, 7 3S, 3B, 1E,
redundant pumps, which are put in use to perform ballasting and de-
1C 1C
5 Generator failure 1 1S, 2B, 1E, 1 1S, 2B, 1E, ballasting operations. These pumps are mainly designed to pump a
1C 1C large quantity of ballast water and often operated by electric motors.
6 Hydraulic leak 1 2S, 2B, 1E, 1 2S, 2B, 1E, Centrifugal pumps are commonly used for ballasting/de-ballasting sys-
1C 1C
tems, as they have a long lifetime and are easy to maintain. Piping system
7 Hydraulic fluid 1 1S, 1B, 4E, 1 1S, 1B, 3E,
spill 1C 1C
of both ballast and de-ballast systems has individual flow lines to each
8 Fire 6 3S, 3B, 1E, 5 4S, 4B, 1E, tank. Each line is equipped with butterfly valves at every individual tank.
4C 4C These valves are operated remotely from the control room. The over-
board discharges are also controlled by remotely-operated valves using
hydraulic system.
Pump failure. The failure of pump(s) is likely due to an electrical fault
Table 7
Parameters of membership functions used for modelling fuzzy sets.
or a mechanical failure and could delay the operation of water pumping.
Valve failure. Failure of a ballasting/de-ballasting valve is a major
Variable Index TMF (a, b, ZMF (a, b, c, GMF (σ,
issue, which may lead to major consequences such as a severe list and
d) d) m)
long delay in the loading/launching process.
Frequency “extremely F1 (0, 1, 2) (0.1, 0.9, 1.1, (0.425, 1) The barge's electric power is supplied by at least two generators such
remote” 1.9)
Frequency “very remote” F2 (1, 2, 3) (1.1, 1.9, 2.1, (0.425, 2)
that the total capacity of which is sufficient to fully operate the ballast/
2.9) de-ballast pumps. In case of one of the generators was out of service,
Frequency “remote” F3 (2, 3, 4) (2.1, 2.9, 3.1, (0.425, 3) the capacity of the remaining generator would be adequate to maintain a
3.9) safe operation (ABS, 2009). These generators are equipped with me-
Frequency “little probable” F4 (3, 4, 5) (3.1, 3.9, 4.1, (0.425, 4)
chanical and electrical protection systems to shut-down the system in
4.9)
Frequency “reasonably F5 (4, 5, 6) (4.1, 4.9, 5.1, (0.425, 5) case of emergency.
probable” 5.9) Generator failure. It is assumed that the barge is equipped with two
Frequency “probable” F6 (5, 6, 7) (5.1, 5.9, 6.1, (0.425, 6) generators, and only one generator may fail (50% power loss) to start due
6.9) to, for example, a mechanical failure (ABS, 2009). Failure of one gener-
Frequency “frequent” F7 (6, 7, 8) (6.1, 6.9, 7.1, (0.425, 7)
7.9)
ator would limit the number of equipment that can be operated simul-
Frequency “very frequent” F8 (7, 8, 9) (7.1, 7.9, 8.1, (0.425, 8) taneously causing some delay.
8.9) A control room is to be located above the full-submergence load-line
Consequence “minor” C1 (0, 1, 2) (0.1, 0.9, 1.1, (0.425, 1) of a transport barge in order to remotely operate the valves of overboard
1.9)
discharges and ballast system. A hydraulic system is often used for the
Consequence “significant” C2 (1, 2, 3) (1.1, 1.9, 2.1, (0.425, 2)
2.9) control of these valves. The hydraulic oil is used to operate valves,
Consequence “severe” C3 (2, 3, 4) (2.1, 2.9, 3.1, (0.425, 3) mooring windlass and capacitance.
3.9) Hydraulic leak and spill. Leakage of hydraulic oil is considered one of
Consequence “catastrophic” C4 (3, 4, 5) (3.1, 3.9, 4.1, (0.425, 4) the failure modes that may occur during normal operations. However,
4.9)
Consequence “disastrous” C5 (4, 5, 6) (4.1, 4.9, 5.1, (0.425, 5)
hydraulic valves are rarely exposing to such failure unless human error
5.9) caused such. The leakage of hydraulic fluid could prevent valves from
operation, cause some delay and affect the entire barge operation.
Overall, the leak of hydraulic oil can easily be detected and dealt with as
position, which may lead to structural damage in the barge side shell. per the on-board operational procedure. Therefore, its impact is expected
Moreover, structural failures due to general and pitting corrosion could to be minor.
increase the severity of the damage to hull. As these barges are required Fire. During load-out/float-off operations and/or offshore trans-
to undergo special survey every five years, where hull thickness mea- portation, fire hazard is expected due to the presence of flammable gases
surements are usually performed and documented by owners. However, and/or chemical substances such as anticorrosive paints, and hence its
the impact of this failure mode on the global integrity is considered to be consequences could be severe on the safety of crew members.

336
N. Abdussamie et al.
Fig. 3. The intersection of two fuzzy sets for frequency indices F2 and F3.
4. Qualitative risk matrix approach
In order to qualitatively evaluate the risk level of a hazardous sce-
nario, the likelihood and severity are combined in a risk matrix (ABS,
2000). The risk matrix is a commonly used technique in formal safety
assessments (ABS, 2000; Elsayed et al., 2011; Vanem et al., 2008; Veritas,
2010; Vinnem, 2007), which are rational and systematic processes for
assessing the risks associated with shipping activity, and hence the results
obtained by this technique can be served as a guideline to validate fuzzy
risk values. The risk is commonly divided into three or four (pictured)
categories which should be the same for safety, environmental and
economic aspects (DNV, 2013). In order to achieve a sufficient resolution
in the obtained risk levels, an 8  5 (frequency x consequence) matrix
was used in this study for the qualitative risk assessment as given in
Table 3 which is similar to the one used for LNG hazards (Østvik et al.,
2005). According to DNV (2013), the output risk levels can be cat-
egorised into three or four levels (denoted by L). In this work, four risk
levels from low to high including “L1”, “L2”, “L3” and “L4” were adopted.
L4 is high “unacceptable” risk, and actions shall be taken to reduce the
risk level. The medium risk was further divided into tolerable (L3) and
broadly acceptable (L2) regions to focus on efforts for risk control (ABS,
2009; DNV, 2013; Vinnem, 2007). L1 is negligible risk, which does not
require actions to be taken.
Based on the results of HAZID study and using technical question-
naires, opinions and judgements from 22 experts with 10–15 years of
experience in the field of offshore operations were used to define the
frequency and consequence indices of each failure mode. The selection
criteria of these experts were based on their design and operational
background and expertise. In order to analyse the input from experts,
they were divided into two expert groups such that group 1 (denoted as
expert 1) consisted of 11 senior naval architects and ocean engineers,
whilst group 2 (denoted as expert 2) consisted of 11 senior seafarers
including chief engineers and captains. Such a combination of design and
operational experiences provided a confidence in the collected data.
During the consultation, each expert has been provided with the defini-
tion of frequency and consequence indices as presented in Tables 4 and 5,
337
Ocean Engineering 156 (2018) 333–346
respectively, along with the risk matrix adopted for this study.
Table 6 summarises a sample of the results of frequency and conse-
quence assessment obtained based on the two groups of experts’ input for
each failure mode (denoted by FM) where the numbers 1–7 denote the
weight/index of each frequency/consequence. For instance, FM 1
“damage to hull” during the operation of positioning the barge using
tugboats was rated by expert group 1 as extremely remote (Fi ¼ 1) with a
moderate damage for the structure being loaded (Ci ¼ 3 → 3S). In order to
account for the differences in the assigned indices by expert groups, the
mean value of each index was used as input for the fuzzy logic models
presented hereafter.
5. Fuzzy logic model
The fuzzy risk model works based on if-then rules, which represent the
relationship between the frequency and consequence (inputs) and the
outcome risk (output). A fuzzy logic system consists fundamentally of
three steps including fuzzification, inference and defuzzification.
In the fuzzification step, each crisp/fuzzy input into a fuzzy model
may have multiple sets assigned into it. In this work, eight fuzzy sets were
used to model the frequency indices and five fuzzy sets were used to
model the consequence indices. An input membership function can be
generated by specifying a number; the degree of membership [0–1], for
each possible input value for a given label (e.g., frequent). A membership
degree is a real number on [0, 1]. The element does not belong to the set
if the degree is 0; whilst the element belongs completely to the set if the
degree is 1. The fuzzification takes a real time input value and compares
it with the stored membership function information so that the fuzzy
input values can be calculated.
In order to create fuzzy sets, different types of membership functions
were tested in this study including triangular membership function
(denoted as TMF), trapezoidal membership function (denoted as ZMF)
and Gaussian membership function (denoted as GMF). Triangular
membership functions were selected due to their simplicity, convenience
of mathematical operations and effectiveness in representing judgment
distributions of multiple experts (Elsayed et al., 2014; Mentes and
N. Abdussamie et al. Ocean Engineering 156 (2018) 333–346

Fig. 4. Frequency fuzzy sets using TMF (top),

ZMF (middle) and GMF (bottom).

338
N. Abdussamie et al. Ocean Engineering 156 (2018) 333–346

Fig. 5. Consequence fuzzy sets using TMF (top),

ZMF (middle) and GMF (bottom).

339
N. Abdussamie et al. Ocean Engineering 156 (2018) 333–346

Table 8 “significant” then risk is “level 3”.

Sugeno's consequence parameters for risk levels. Rule 3: If frequency is “very frequent” and consequence is “severe”
Risk level Range Zeroth-order (r) First-order (p, q, r) then risk is “level 4”.
00
Rule 4: If frequency is “very frequent” and consequence is “cata-
Level 1 “L1 2–4 3 (1, 1, 0)
Level 2 “L200 5–7 6 (1, 1, 0) strophic” then risk is “level 4”.
Level 3 “L300 8–10 9 (1, 1, 0) Rule 5: If frequency is “very frequent” and consequence is “disas-
Level 4 “L400 11–13 12 (1, 1, 0) trous” then risk is “level 4”.
Rule 6: If frequency is “frequent” and consequence is “minor” then
risk is “level 3”.
Helvacioglu, 2012). The Gaussian membership functions were used by
Rule 7: If frequency is “frequent” and consequence is “significant”
other researchers (Elsayed, 2009; Elsayed et al., 2011) and found to be
then risk is “level 3”.
suitable for risk assessment problems. The ZMF is defined by four ele-
Rule 8: If frequency is “frequent” and consequence is “severe” then
ments: a, b, c and d. For a trapezoidal fuzzy set x ¼ (a, b, c, d; w), the
risk is “level 3”.
membership function is given by (Sivanandam et al., 2007):
Rule 9: If frequency is “frequent” and consequence is “catastrophic”
8 then risk is “level 4”.
>
> wðx  aÞ
>
> ; axb Rule 10: If frequency is “frequent” and consequence is “disastrous”
>
> ba
>
> then risk is “level 4”.
< w; bxc
μA ðxÞ ¼ (1) The computation of fuzzy rules is called a fuzzy inference. The Sugeno
>
> wðx  dÞ
>
> ; cxd fuzzy inference method (Takagi and Sugeno, 1985) implemented in
>
> cd
>
> MATLAB was used. The Sugeno method is a more compact and compu-
:
0; otherwise tationally efficient than the Mamdani method (Mamdani and Assilian,
1975). A typical rule in a first-order Sugeno fuzzy model has the
where μA(x) denotes the membership function of a variable x. If w ¼ 1,
following form (Sivanandam et al., 2007):
the set x (a, b, c, d; 1) is called normalised trapezoidal. If b ¼ c, the
If F (frequency) is x and C (consequence) is y then z
trapezoidal fuzzy set reduces to a triangular fuzzy set as:
(risk) ¼ px þ qy þ rwhere x and y are fuzzy sets, p, q and r are the
8 consequence parameters.
> 0; xa
>
> As the risk level is linearly proportional to both frequency and
>
> x  a
>
>
< b  a; a<xb consequence R ¼ f(F, C) (Vinnem, 2007), in this study, the zeroth- and
μA ðxÞ ¼ (2) first-order models were adopted for all calculations. For the zeroth-order,
>
> dx
>
> ; b<x<d r was defined as the mean value of the associated range of risk levels
>
> d b
>
: given in Table 8. The consequence parameters of the first-order model
0; x
d were obtained based on the given premise variables in the risk matrix
The Gaussian function is given by: using the formula R ¼ F þ C (Østvik et al., 2005) which yields p ¼ q ¼ 1
and r ¼ 0 for all risk levels using linear regression. 3D surfaces of the
ðxmÞ2
relationship between the inputs (frequency and consequence) and the
μA ðxÞ ¼ e 2σ 2 (3)
output (risk) are shown in Fig. 6. The GMF seems to produce a smoother
where m and σ are the mean and standard deviation, respectively. The surface than that was obtained by the TMF and ZMF, which can be
parameters of each membership function tested are summarised in attributed to the mathematical model of a Gaussian function being more
Table 7 for the frequency and the consequence (denoted by Fi and Ci). It effective in handling the range of Fi and Ci. The output level z of each rule
should be noted that the central element of the triangular membership is weighted by the firing strength wi of the rule. For example, for an AND
functions, b, of each set was defined to be equal to the index associated rule, the firing strength is obtained by:
with the linguistic variable given in the risk matrix. Other membership
wi ¼ AndMethodðμA ðxÞ; μB ðyÞÞ (4)
functions (ZMF and GMF) were then generated on the basis of the pre-
defined TMF. For instance, the set of F1 “extremely remote” was assigned where μA(x) and μB(y) are the membership functions for inputs x and y,
as (a, b, d) ¼ (0, 1, 2) such that the central element of the triangular MF respectively.
“b” was defined to be equal to the index associated with the linguistic The final output of the system is the weighted average of all rule
variable given in the risk matrix. On the other hand, the Gaussian MFs outputs and computed as (Sivanandam et al., 2007):
were then generated on the basis of the predefined triangular MFs in
PN
which m ¼ b for all fuzzy sets. In order to achieve a similar intersection wi zi
between any two consecutive fuzzy sets for all types of MFs, a standard R ¼ Pi¼1
N (5)
i¼1 wi
deviation σ of 0.425 was estimated and used for all fuzzy sets of the
Gaussian functions, as seen in Table 7. As an example, this was illustrated where N is the number of rules.
in Fig. 3 for F2 and F3 such that at x ¼ 2.5 a degree of membership In order to allow for the risk matrix results to be compared with those
function μF(x ¼ 2.5) ¼ 0.5 was obtained for triangular, trapezoidal and predicted by the fuzzy set techniques, a fuzzy risk index (FRI) was used to
Gaussian MFs which yields σ ~ 0.425. Figs. 4–5 show fuzzy sets repre- combine the various consequence attribute risks into a unified risk
sented by the different membership functions for both frequency and measure:
consequence parameters, respectively.
PM
A fuzzy model was created based on 40 if-then rules which represent 1 ci Ri
FRI ¼  Pi¼1 (6)
the size of the qualitative risk matrix (8  5). The precondition can M M
i¼1 ci
consist of multiple conditions linked together with AND conjunction. The
first 10 if-and-then rules which represent the first two rows of the risk where M is the number of consequences (M ¼ 4 in this case), ci is a weight
matrix can be written as: factor, i.e., severity level for each consequence, and Ri is the resulting risk
Rule 1: If frequency is “very frequent” and consequence is “minor” value for each consequence attribute calculated from Eqn. (5), whereas
then risk is “level 3”. R ¼ F þ C for the qualitative risk matrix technique.
Rule 2: If frequency is “very frequent” and consequence is

340
N. Abdussamie et al. Ocean Engineering 156 (2018) 333–346

Fig. 6. 3D surfaces for the output risk using

TMF (top), ZMF (middle) and GMF (bottom).

341
N. Abdussamie et al. Ocean Engineering 156 (2018) 333–346

Table 9
Summary of fuzzy risk values.
FM# Failure mode TMF/ZMF GMF-0

S B E C S B E C

1 Damage to hull 4.50 6.00 3.00 5.25 4.50 5.92 3.01 5.25
2 Out of position 6.00 6.00 3.00 6.00 6.00 6.00 3.32 6.09
3 Pump failure 9.00 9.00 9.00 9.00 9.09 9.31 8.84 8.84
4 Valve failure 9.00 9.00 7.50 7.50 9.00 9.09 7.59 7.59
5 Generator failure 3.00 3.00 3.00 3.00 3.00 3.01 3.00 3.00
6 Hydraulic leak 3.00 3.00 3.00 3.00 3.01 3.01 3.00 3.00
7 Hydraulic fluid spill 3.00 3.00 4.50 3.00 3.00 3.00 4.59 3.00
8 Fire 9.00 9.00 6.00 9.00 9.00 9.00 6.09 9.09

Fig. 7. FRI values versus qualitative risk matrix results.

6. Results and discussion
The results of risk assessment obtained by the qualitative risk matrix
and fuzzy set techniques are discussed in this section.
6.1. Risk results
Table 9 summarises the calculated fuzzy risk values for the 8 failure
modes and for all consequence attributes using the zeroth-order fuzzy
Sugeno model. The TMF and ZMF type functions were found to provide
identical risk values for all failure modes, while the GMF type function
provided different values for some consequences. For instance, the cor-
responding risk values to FM 1 “damage to hull” for the environment (B)
predicted by the GMF and TMF/ZMF were 5.92 and 6.00, respectively. By
investigating the obtained results, the GMF was found to be more sen-
sitive to a small change in Fi or Ci than the other functions, which makes it
advantageous over the TMF and ZMF in risk ranking/prioritising. For
example, FM 3 “pump failure” in Table 9, same risk value R of 9.0 was
obtained by the TMF/ZMF for all consequences despite different Ci values
(2.5S, 3B, 1E, 1C), which indicates that neither of these functions
detected the change in Ci. In contrast, the GMF predicted different risk
values of FM 3 for the four consequence attributes such that higher risk
values were calculated for larger Ci values, R ¼ 9.31 for B and R ¼ 9.09
for S which support the linear relationship between the risk level and the
severity. Based on the risk values obtained by the GMF, the safety of
barge and structure seems to be affected significantly by three failure
modes including FM 3 “pump failure”, FM 4 “valve failure” and FM 8
“fire”. On the other hand, as expected FM 8 “fire” caused the highest risk
level for the safety of crew members. The risk results of these failure
modes are discussed in more details below.
Fig. 7 shows FRI values for the 8 failure modes using the three
different membership functions tested and the risk matrix technique
(denoted as QRM). Likewise, both the TMF and ZMF type functions
provided identical risk values for all failure modes and different from
those obtained by the Gaussian-type functions. By referring to Fig. 7,
fuzzy risk values exactly matched those obtained by the QRM when the
first-order fuzzy Sugeno model (denoted as GMF-1) was used, which
indicates that the RI values obtained by the fuzzy set approaches are
consistent with the results of the QRM technique. This confirms that the
relationship between risk parameters was correctly modelled in the fuzzy
logic models using if-then rules. It should be noted that all the subsequent
results discussed hereafter were obtained using the GMF-0 and GMF-1
models.
6.2. Consequence analysis
The safety of each consequence attribute was assessed by presenting
the risk profile due to different failure modes. For instance, the safety of

342
N. Abdussamie et al.
Fig. 8. Failure mode risk contributions for the safety of structure (a), barge (b), the environment (c) and crew members (d).
structure is shown in Fig. 8a where pump failure contributed 20% to-
wards the overall risk imposed on the structure. Failure of generator
caused the lowest risk contribution with 6%. For the impact on the barge
safety and operation, refer to Fig. 8b where pump failure, valve failure
and fire had the same risk contribution of 19% towards the overall risk
imposed on the barge. Failure of generator, hydraulic leak and hydraulic
fluid spill caused the lowest risk contribution of 6%. For the environ-
mental impacts (Fig. 8c), pump failure had the largest risk portion of 22%
followed by valve failure (19%), fire (15%) and hydraulic fluid spill
(12%). On the other hand, failure of generator and hydraulic leak
contributed with 8%. For the safety of crew (Fig. 8d) where fire hazard
had the largest risk portion of 20% followed by pump failure with 19%
and valve failure with 17%, whilst failure of generator contributed with
only 6%.
6.3. Sensitivity analysis
In order to evaluate how uncertainty in model inputs (e.g., frequency
index) affects the model output (risk value), sensitivity analyses were
343
Ocean Engineering 156 (2018) 333–346
conducted for the three top-ranked failure modes including FM 3 “pump
failure”, FM 4 “valve failure” and FM 8 “fire”. One-way sensitivity
analysis and tornado diagram techniques were employed in the analyses.
In the one-way sensitivity analysis, the effect of the frequency index Fi
was specified as multiple point estimates and varied from Fi ¼ 1–8 with
0.25 increments as shown in Figs. 9 and 10 in which the values of Fi
assigned by the two groups of experts were denoted by vertical lines (Exp.
1 and 2). On the other hand, the tornado diagram is a set of one-way
sensitivity analysis for multiple input variables as shown in Fig. 11 in
which each variable (F, B, S, B and C) is represented by a horizontal bar.
In order to test the effect of the zeroth- and first-order Sugeno models
on the results of sensitivity analysis, the output of both models was
compared for FM 3 risk to the safety of barge (Ci ¼ 3). As seen in Fig. 9,
the first-order model (GMF-1) resulted in a linear relationship between Fi
and R, whereas a different trend was obtained using the GMF-0 model. As
already mentioned, based on the resulting risk levels obtained from the
risk matrix technique, the risk level is indeed linearly proportional to the
frequency index. It is worth noting that the GMF-0 model produced
smaller risk values at Fi ¼ 1 and larger risk values at Fi ¼ 8 than those
N. Abdussamie et al. Ocean Engineering 156 (2018) 333–346

Fig. 9. Results of sensitivity analysis of “FM 3: pump failure” risk to barge.

Fig. 10. Results of sensitivity analysis for “FM 3: pump failure” (top), “FM 4: valve failure” (middle) and “FM 8: fire” (bottom).

obtained by the GMF-1 model. It was concluded that the first-order
Sugeno model produced more accurate risk values than the zeroth-
order in comparison with the risk matrix technique, and hence it was
employed for the further analyses discussed hereafter.
By referring to Fig. 10, the risk level associated with pump and valve
failures increased for all consequence classes with increasing Fi with a
maximum R value ¼ 11 and 10.5 for B and S, respectively. For fire haz-
ard, the R value largely increased within the Fi range between 6 and 8,
which is higher than the values assigned by the expert groups. The
maximum risk value for C (R ¼ 12) is categorised as a high-risk level “L4”

344
N. Abdussamie et al.
Fig. 11. Tornado diagrams showing multiple sensitivity analyses for FM 3 (top),
FM 4 (middle) and FM 8 (bottom).
(Table 3) indicating safety measures need to be provided for this scenario
to be controlled. Such findings demonstrate the effectiveness of using
fuzzy sets to model risk parameters and assist safety assessors in risk
evaluation. Further action is appropriate where cost-effective, or where
needed to ensure risks do not increase (DNV, 2013). It should be noted
that risks are tolerable once all reasonably practicable actions have been
taken to reduce them. Further reduction action is needed, unless the costs
are grossly disproportionate to the benefits.
For the tornado diagrams shown in Fig. 11, each horizontal bar rep-
resents the range of expected FRI values obtained by varying the asso-
ciated variable. The most sensitive variable (or most critical) is at the top
and the least sensitive is at the bottom of the diagram. A change of 1 in
the index was applied for the frequency and consequence parameters, as
this variation could be the case in the results of individual expert
judgements used in risk analysis. It should be noted that the base values
of FRI (prior to sensitivity analyses) were 2.33, 2.20 and 2.24 for FM 3,
345
Ocean Engineering 156 (2018) 333–346
FM 4 and FM 8, respectively. The expected value of FRI was most affected
by the change in F-index (a wider bar at the top of each diagram) for the
three failure modes tested (for example, for FM 3, FRI ¼ 2.08–2.58). The
consequence index of B and S were the most second and third critical
parameters for both FM 3 and FM 4, whilst the consequence index of C
was the second most critical variable for FM 8 (FRI ¼ 2.16–2.35).
7. Conclusions
A fuzzy set approach was developed to quantify and assess the risk
due to several hazardous scenarios during load-out/float-off operations
of an offshore transport barge. The potential hazard impacts on the safety
of crew members involved in the operations, the barge and an offshore
structure being loaded, and the environmental impacts were assessed
using the developed techniques. Based on the results presented in this
paper, the following conclusions were drawn:
1. Based on the HAZID study performed, the worst-case operational
scenarios were identified including eight failure modes in different
five operations. A qualitative risk assessment was conducted with the
assistance of judgements from 22 senior experts with 10–15 years of
experience in the field of offshore operations. Even though the
number of experts provided insights into the expected frequency and
consequence of each failure mode, it is recommended that a large
number of experts should be consulted for future studies in order to
thoroughly investigate the operational hazards and risks of offshore
barges.
2. The rule-based fuzzy models used different types of membership
functions including the triangular, trapezoidal and Gaussian func-
tions. The triangular type functions were found to provide identical
results to those obtained using the trapezoidal functions but slightly
different from those obtained by the Gaussian functions. This can be
attributed to the mathematical model of a Gaussian function being
more effective in handling the range of frequency and consequence
indices than the triangular and trapezoidal functions as illustrated in
the surface plots. It was concluded that the Gaussian-type functions
were found to be more effective in risk ranking/prioritising than
other membership functions tested.
3. Based on the consequence analyses conducted for the safety of each
consequence attribute, the safety of the structure could be highly
affected by the pump failure, which contributed 20% towards the
overall risk imposed on the structure. For the impact on the barge
safety and operation, the pump and valve failures as well as the fire
hazard had the same risk contribution of 19%. For the safety of crew
members, fire hazard had the largest risk portion of 20%. Other
failure modes such as the failure of generator, hydraulic leak and
hydraulic fluid spill caused the lowest risk contribution with 6%. It is
therefore recommended that the scenarios of “pump failure”, “valve
failure” and “fire hazard” should be prioritised in the risk analysis for
offshore transport barges.
4. The results of sensitivity analyses for top-ranked failure modes using
both one-way sensitivity analysis and tornado diagram techniques
indicated that fuzzy set approaches are useful to test the effect of risk
parameters on the output risk values. This demonstrates the effec-
tiveness of using fuzzy sets to model risk parameters to assist safety
assessors in risk evaluation.
5. Overall, the present fuzzy sets approach could be a useful tool in
addition to the risk matrix technique to assess risks of barges and
Heavy Lift Vessels used in offshore transportation. For further work,
other risk measures such as downtime and the reputation of com-
pany/shipyard should be studied using fuzzy models.
Acknowledgements
The authors are thankful to all experts for providing their opinions
and useful discussions about the HAZID study.
N. Abdussamie et al. Ocean Engineering 156 (2018) 333–346

References Mokhtari, K., Ren, J., Roberts, C., Wang, J., 2012. Decision support framework for risk
management on sea ports and terminals using fuzzy set theory and evidential
reasoning approach. Expert Syst. Appl. 39, 5087–5103.
ABS, 2000. Guidance Notes on Risk Assessment Applications for the Marine and Offshore
Nwaoha, T.C., Yang, Z., Wang, J., Bonsall, S., 2013. Adoption of new advanced
Oil and Gas Industries. American Bureau of Shipping, Houston, USA.
computational techniques to hazards ranking in LNG Carrier operations. Ocean Eng.
ABS, 2009. Rules for Building and Classing, Steel Floating Dry Docks. American Bureau of
72, 31–44.
Shipping, Houston, USA.
Østvik, I., Vanem, E., Castello, F., 2005. HAZID for LNG Tankers. SAFEDOR Report
Aronsson, E., 2012. FLNG Compared to LNG Carriers-Requirements and
D.4.3.1.
Recommendations for LNG Production Facilities and Re-gas Units. Masters Thesis.
Pitt, M.J., 1994. Hazard and operability studies: a tool for management analysis. Facilities
Chalmers University of Technology.
12, 5–8.
Denton, A., 1989. The loss of a jack-up under tow. Mar. Struct. 2, 213–231.
Samantra, C., Datta, S., Mahapatra, S.S., 2014. Risk assessment in IT outsourcing using
DNV, 2013. DNV-OS-J201 Offshore Substations for Wind Farms. Det Norsok Veritas,
fuzzy decision-making approach: an Indian perspective. Expert Syst. Appl. 41,
Norway.
4010–4022.
Elsayed, T., 2009. Fuzzy inference system for the risk assessment of liquefied natural gas
Sivanandam, S., Sumathi, S., Deepa, S., 2007. Introduction to Fuzzy Logic Using MATLAB.
carriers during loading/offloading at terminals. Appl. Ocean Res. 31, 179–185.
Springer.
Elsayed, T., Leheta, H., Belhaj, I., 2011. Fuzzy inference system for fire and explosion risk
Spouge, J., 1999. A Guide to Quantitative Risk Assessment for Offshore Installations.
assessment of floating storage and offloading vessels. Proc. IME M J. Eng. Marit.
CMPT.
Environ. 225, 171–180.
Stavrou, D.I., Ventikos, N.P., 2016. A novel approach in risk evaluation for ship-to-ship
Elsayed, T., Leheta, H., Shehadeh, M., 2009. Multi-attribute risk assessment of LNG
(STS) transfer of cargo using process failure mode and effects analysis (PFMEA).
carriers during loading/offloading at terminals. Ships Offshore Struct. 4, 127–131.
J. Risk Res. 19, 913–933.
Elsayed, T., Marghany, K., Abdulkader, S., 2014. Risk assessment of liquefied natural gas
Takagi, T., Sugeno, M., 1985. Fuzzy identification of systems and its applications to
carriers using fuzzy TOPSIS. Ships Offshore Struct. 9, 355–364.
modeling and control. IEEE Trans. Syst., Man, Cybernetics 116–132.
Fang, H., Duan, M., 2014. Offshore Operation Facilities: Equipment and Procedures. Gulf
Van Hoorn, F., 2008. Heavy-lift transport ships-overview of existing fleet and future
Professional Publishing.
developments. In: Proceedings of the Marine Operations Speciality Symposium
Gerwick, C., 2002. Construction of Marine and Offshore Structures. CRC press.
(MOSS).
GLND, 2005. General Guidelines for Marine Transportations, 0030/NDI. GL Noble
Vanem, E., Ant~ao, P., Østvik, I., De Comas, F.D.C., 2008. Analysing the risk of LNG Carrier
Denton.
operations. Reliab. Eng. Syst. Saf. 93, 1328–1344.
GLND, 2009. Guidelines for Loadouts, 0013/ND. GL Noble Denton.
Veritas, D.N., 2010. DNV-RP-F107 Risk Assessment of Pipeline Protection. Norway.
Gunter, M.M., Hill, R., Retzer, K.D., Lincoln, J.M., 2013. Fatal injuries in offshore oil and
Vinnem, J.E., 2007. Offshore Risk Assessment Principles, Modeling and Applications of
gas operations-United States, 2003-2010. MMWR. Morbidity and mortality weekly
QRA Studies. Springer, London.
report 62, 301.
Wang, G., Pedersen, P.T., 2007. A Literature Review of Risk Assessment of Ship-FPSO
Mamdani, E.H., Assilian, S., 1975. An experiment in linguistic synthesis with a fuzzy logic
Collisions. OMAE2007-29760, San Diego, California.
controller. Int. J. Man Mach. Stud. 7, 1–13.
Zhao, S., Soares, C.G., Zhu, H., 2015. A Bayesian network modelling and risk analysis on
Mentes, A., Helvacioglu, I.H., 2012. Fuzzy decision support system for spread mooring
LNG carrier anchoring system. In: Transportation Information and Safety (ICTIS),
system selection. Expert Syst. Appl. 39, 3283–3297.
2015 International Conference on. IEEE, pp. 432–436.

346