Professional Documents
Culture Documents
Risk Analysis - 2011 - Li - An Overview of Maritime Waterway Quantitative Risk Assessment Models
Risk Analysis - 2011 - Li - An Overview of Maritime Waterway Quantitative Risk Assessment Models
The safe navigation of ships, especially in narrow shipping waterways, is of the utmost con-
cern to researchers as well as maritime authorities. Many researchers and practitioners have
conducted studies on risk assessment for maritime transportation and have proposed risk
reduction/control measures accordingly. This article provides a detailed review and assess-
ment of various quantitative risk assessment models for maritime waterways. Eighty-seven
academic papers and/or project reports are summarized and discussed. The review then pro-
ceeds to analyze the frequency and consequence estimation models separately. It should be
pointed out that we further summarize the advantages and disadvantages of frequency esti-
mation models and provide recommendations for their application. From the overview, we
find that the quantification of the impact of human error is of great importance and should
be considered in future studies. Possible solutions are also proposed in the discussions.
KEY WORDS: Causation probability; maritime waterway; quantitative risk assessment; ship collision
and grounding
496 0272-4332/12/0100-0496$22.00/1
C 2011 Society for Risk Analysis
Maritime Waterway Quantitative Risk Assessment Models 497
Collision while on
p=8.48E-05
dangerous course
Failure on Failure on
other ship own ship
4.24E-05
0.5 0.5
Fig. 3. Fault tree applied to derive causation probability for collision in good visibility.(12)
For a complex system, the size of the fault tree is Besides fault trees and Bayesian networks, the
likely to become too large. The Bayesian network, expert judgment elicitation method has also been
meanwhile, will remain compact even for large-scale successfully used to estimate causation probabili-
problems. Friis-Hansen and Pedersen(13) devel- ties.(16) In Merrick and van Dorp,(17) the probability
oped a Bayesian network for a navigating officer of a collision is estimated by summing over all the
reacting to the event of being on a collision course possible situations:
(dangerous encounter). Pedersen(10) and Otto
et al.(11) used a similar Bayesian network to de- P(Collision)
rive the causation probabilities of various meeting
k
l
scenarios (e.g., head-on, overtaking, and crossing). = P(Collision|Incidenti Situation j )
j=1 i=1
Friis-Hansen and Simonsen(14) provided a Bayesian
network that considers correlations between two × P(Incidenti |Situation j )P(Situation j ), (2)
approaching vessels to estimate the causation prob-
ability of a ship-to-ship collision. Trucco et al.(15) where P(Situationj ) is the probability that a particu-
applied Bayesian networks to study the risk of a lar situation or combination of values of the factors
maritime transport system. occurs in the system; P(Incidenti |Situationj ) is the
500 Li, Meng, and Qu
probability that a particular triggering incident oc- where D is the average distance between the ships
curs in the given situation; and P(Collision|Incidenti , (in miles), X is the actual length of path to be
Situationj ) is the probability that a collision acci- considered for a single ship (nautical miles), and L
dent occurs in the defined situation once the trig- is the average vessel length (m). In this formula, the
gering incident has occurred. Obviously, the prod- main ship stream is assumed to move with speed V,
uct P(Incidenti |Situationj ) × P(Collision|Incidenti , and there is a single ship approaching the stream with
Situationj ) is more or less equivalent to the cau- the same speed V and at angle θ . This model can eas-
sation probability mentioned above. P(Situationj ) ily be expanded to more general scenarios. However,
can be obtained using Bayesian simulation tech- the geometrical probability is again overestimated by
niques; P(Incidenti |Situationj ) can be assessed us- this formula, especially when the angle is very small.
ing traditional Bayesian estimation techniques; and In addition, the equal speed assumption is not rea-
P(Collision|Incidenti , Situationj ) can be obtained us- sonable and may result in an underestimation of the
ing the expert judgment elicitation method. The probability. On the basis of its proof, the model can
pairwise comparison approach is used to evaluate be extended to more complex conditions.(10)
the uncertainty of expert judgments.(16) Using such
techniques, point estimation and full distribu-
tional information about the collision risk can be 2.1.2.2. Fujii’s model and later developments. Fu-
obtained. jii(18) proposed the following model to estimate the
average number of evasive actions by a ship passing
through an area:
2.1.2. Geometrical Probability Estimation
exit
2.1.2.1. Macduff’s model. Macduff (3) proposed two
(ρ De Vr el /V)dx, (5)
models: one for estimating probabilities of ground- entrance
ings and one for collisions.
On the basis of Buffon’s Needle Problem, the ge- where ρ is the ship density (the number of ships per
ometrical probability of hitting the walls of a channel unit area), De is the diameter of evasion, Vr el is the
(grounding) was estimated to be: relative speed, and V is the speed of the passing ship.
4T The model is based on geometry and laws of motion.
Pg =, (3) De varies from 9.5–16.3 times the ship’s length, ac-
πC
cording to Fujii et al.(4)
where T is the track length of the ship or its stop- The assumptions of Fujii’s model are more rea-
ping distance and C is the breadth of the channel. It sonable than those of Macduff’s model. However,
should be pointed out that T is a function of the size the results show that Fujii’s model also overestimates
and speed of the ship. However, an inappropriate as- the geometrical probability (to an even higher ex-
sumption means that this formula overestimates the tent than Macduff’s model does). This is because the
geometrical probability (the angle of the stopping evasive action is assumed to be 9.5–16.3 times the
track cannot be assumed to be uniformly distributed ship’s length, which is quite conservative. In reality,
between 0 and π/2). the minimum ship distance in some straits (e.g., the
To estimate the geometrical probability of a ship Singapore Strait) is approximately three times the
collision, Macduff(3) applied molecular collision the- ship length.
ory. On the basis of this theory, the geometrical After Fujii’s pioneering work, the concept of a
probability of a ship collision has an approximately ship domain was proposed and widely applied in nav-
linear relationship with the traveling distance, that igational safety studies. Various types of ship do-
is, the quotient of traveling distance and the mean main, with distinct shapes and sizes, were defined
free path (constant) is an approximation of the by Fujii,(18) Goodwin,(19) Coldwell,(20) Davis et al.,(21)
geometrical probability.1 The probability can be and others. Zhao et al.(22) commented on the pre-
calculated: viously proposed ship domains and proposed the
XL sin(θ/2) concept of fuzzy ship domains. Since then, fuzzy
Pa = 2 , (4) ship domains have been developed and applied to
D 925
estimating the frequencies of ship collisions.(23−26)
1 The second-order Taylor expansion of an exponential function is Szlapczynski(23) also proposed a measure of colli-
linear. sion risk on the basis of the concept of a ship’s
Maritime Waterway Quantitative Risk Assessment Models 501
⎧ ⎫1/2
⎨ (2) 2 ⎬
(1) Vj
+ Bi 1 − sin θ ,
⎩ Vi j ⎭
(1)
where Li is the length of a vessel in class i in wa-
(2)
terway 1, Lj is the length of a vessel in class j
(1)
in waterway 2, Bi is the width of a vessel in class i
Fig. 5. Crossing waterways collision.(11) (2)
in waterway 1, Bj is the width of a vessel in class j in
(2)
waterway 2, Vj is the velocity of a vessel in class i in
domain, which is an extension of some earlier (2)
waterway 1, Vj is the velocity of a vessel in class j in
works.(20−22,27) waterway 2, and θ is the angle between the directions
of the two waterways.
The geometrical probability can be estimated by:
2.1.2.3. Pedersen’s model. Pedersen and his re-
search collaborators conducted a series of well-
(1)
Qi Q j
(2)
(1)
Na = fi
known studies and provided a further approach for (1) (2)
(Zi Zj ) V V
i j i j
estimating geometrical probability(10) that was ap-
(9)
plied in several later research works.(11,28,29) (2)
×(zi ) f j (z j )Vij Dij d At,
Pedersen’s model assumes that there are two
crossing waterways, 1 and 2 (as shown in Fig. 5), where A is the considered water area.
and the traffic intensity (number of ships passing per It should be pointed out that, different from
unit of time), velocity, and the lateral distributions of Fujii’s model, which uses ship density, that is, the
the ships’ routes are all known. Pedersen(10) reported number of ships per unit area, Pedersen’s model
that the number of ships belonging to class j on col- uses traffic volume Q (ship flow per unit of time) in
lision course with one ship of class i in waterway 1 Equations (6) and (9).
during the time period t in a segment dz j of water- Pedersen provides a comprehensive approach
way 2 is: for estimating the geometrical probability. On the
basis of a more practical assumption, the vessels
(2)
Qj are categorized on the basis of their distances from
(2)
Pt = (2)
f j (z j ) Dij Vij dz j t, (6) the predetermined navigational course (centerline),
Vj which takes the vessels’ meandering navigation into
account using probability theory. However, deter-
where Q j is the number of movements of a ship in mining the probability distribution and characteris-
class j per unit of time, z is the distance of ships from tics is still challenging, as ship movement data are not
the centerline of the waterway, the lateral distribu- easily obtained.
tion of the ships’ routes is denoted by f (which is
often taken to be a normal distribution), Vij is the rel-
ative velocity, calculated by: 2.1.2.4. Kaneko’s model. Kaneko(9) proposed his
models on geometrical probability estimation for two
specific scenarios. One scenario is a circular bound-
(1) (2) (1) (2)
Vi j = (Vi )2 + (Vj )2 − 2Vi Vj cos θ , (7) ary area with random sailing directions and the other
502 Li, Meng, and Qu
situation. For the circular boundary and random sail- meetings within one route segment, where L is the
ing direction scenario, the number of other ships en- length of the route segment, N1 is the number of
countered by a given ship (the main ship) within time times per year ship 1 passes through this route, N2 is
T is: the number of times per year ship 2 passes through
√ the route, V1 is the speed of ship 1, and V2 is the
4ρVr T 2 α
λc = (1 + α)E , (10) speed of ship 2. PG = B1 +B 2
is the geometrical colli-
π 1+α c
sion probability, where B1 is the breadth of ship 1,
where ρ is the average number of sailing ships in the B2 is the breadth of ship 2, and c is the width of the
area, V is the velocity of other ships, α = V0 /V, V0 is segment of waterway. PC is the causation probability,
the velocity of the main ship, and: taken to be 3E-04 in the report. kRR is a risk reduction
√ π factor, taken to be 0.5.
2 V0 V 2 4V0 V Crossing collision frequency (PX ) depends on
E = 1− sin2 θ dθ, (11)
V + V0 0 V + V 0 crossing pattern (X-type or Y-type; see Fig. 8), traf-
fic intensity in each of the two directions, width,
where θ is the angle between the direction of travel length, and speed of the ships, crossing angle, cau-
of the main ship and other ships. sation probability PC , and the probability that the
For the rectangular boundary with fixed sailing traces of the two ships intersect:
direction scenario, the number of ships encountered
by one ship within time T is: PX = PI × PG × PC × kRR, (14)
λc = ρV2r T 1 + α 2 + 2α cos θ . (12) where PI is the probability that the traces of the
two ships intersect: for an X-type crossing PI = 1,
It can also be shown using this model that the for a Y-type crossing with intersection PI = 1, and
random sailing direction scenario results in 10% for a Y-type crossing without intersection PI = 0. If
more dangerous encounters than the fixed direction it is not clear whether there are intersections, for
scenario. As the ship traffic density within a large a Y-type crossing it is assumed that PI = 0.5. PG =
sea area is assumed to be uniform, Kaneko’s model N1 (1 − e−N2 t ) is the geometrical collision probabil-
is more appropriate for use in a wide open sea area ity, which depends on the so-called critical time in-
rather than in a restricted narrow waterway. terval t (see Fig. 9), given by:
1 V2 V1 V1 V2
t = B2 − + B −
tan θ
1
2.1.2.5. COWI models. A practical mathematical |V1 V2 | sin θ sin θ tan θ
model was applied in a report by COWI(30) to calcu-
late the frequency of ship collisions. In parallel wa- + L1 |V2 | + L2 |V1 | . (15)
terways, the collision frequency (PX ) of two ships
(head-on or overtaking; see Figs. 6 and 7) depends The causation probability PC is taken to be
on the length of the route segment, traffic intensity in 3E-04. If one of the two potentially colliding ships
Maritime Waterway Quantitative Risk Assessment Models 503
has a pilot service (or local experience or heightened The frequency of grounding because of a missed
safety standards), the risk reduction factor takes turn is given by:
value of kRR = 0.75. If both ships have such risk re-
PX = PNT × PG × PC × kRR, (18)
duction measures in place, kRR = 0.5.
The model used in the COWI report is on the where
basis of a concept similar to Pedersen’s model. The
PG = F(α1 ) − F(α2 ), (19)
risk reduction factors are considered in this model
as well, which add more flexibility. Nevertheless, the and
risk reduction factors are determined in a somewhat
PNT = e−λ V ,
x
(20)
subjective way.
The COWI report(30) also presented a similar where PNT is the annual frequency of a ship missing a
grounding/contact model. Two patterns are consid- turn and not correcting its error at a sufficiently early
ered: one called grounding because of imprecise nav- point to avoid grounding, λ is the checking frequency
igation, and the other grounding because of a missed (0.5–1 minute), x is the distance between the missed
turn (Fig. 10). The frequency of grounding because turning point and the ground, and V is the vessel’s
of imprecise navigation is given by: speed. The causation factor PC is taken to be 3E-04,
and the risk reduction factor kRR is 0.5.
PX = PG × PC × kRR × kDC , (16) In COWI models, the geometrical probability
where is further divided into several independent prob-
abilities. Compared with Pedersen’s models, they
PG = F(α1 ) − F(α2 ), (17) are more friendly to use in practical projects be-
cause of the better data availability of their input
where α1 and α2 are the upper and lower bounds of
parameters.
the critical angle, respectively, and F is the lateral
distribution of the ship’s course. The causation fac-
tor PC is again taken to be 3E-04, the distance fac- 2.1.2.6. Other models. In addition to the above-
10 nm
tor kDC is distance , and the risk reduction factor kRR mentioned models, several other models have also
is 0.5. been developed by researchers. A simple random
504 Li, Meng, and Qu
Turning point
walk model was built by Tan and Otay(31) to esti- Rensselaer Polytechnic Institute, and Virginia Com-
mate geometrical probability. Roeleven et al.(32) cal- monwealth University (GWU-RPI-VCU group for
culated accident probability by fitting generalized lin- short)—developed a counting model that considers
ear models. Geng et al.(33) fitted the distributions of each vessel of interest against each other vessel in the
vessel courses using field observations, and then cal- simulation in turn.(36) For each pairing, the closest
culated vessel-bridge collision probabilities. Debnath point of approach (CPA) is calculated, along with
and Chin(34) and Chin and Debnath(35) studied per- the angle of interaction and the time to the CPA
ceived collision risks in port waters. (also see Section 2.1.1). The model has been widely
applied by the group to assess and/or manage risks
(e.g., for Washington State ferries) and has been
2.1.3. Computer Simulation developed and continuously improved by the group
Although causation-probability-based mathe- over a span of more than 10 years.(17,36−44) It is of
matical models for estimating accident frequencies note that the methodology allows for the effective
have been well studied and successfully applied in evaluation of a simultaneous implementation of the
real-world projects, they have some inherent limita- likelihoods and consequences of various accidental
tions. These models assume that a ship will appear in scenarios. The framework for risk management is
a certain location with a certain probability and, be- summarized as a six-step procedure:
cause of the large number of ships in a typical study
1. Define a quantitative measure of risk.
area, it is often assumed that the presence of any ship
2. Identify potential risk interventions and de-
can follow only one or a few given probability distri-
termine their impact on the accidental events.
butions in any given location and at any time (e.g.,
3. Develop a comprehensive quantitative model
uniform, normal distributions), which is unrealistic
for comparing the risk interventions in a
for restricted waters. These assumptions are also in-
meaningful manner.
appropriate for waters with navigational regulations.
4. Establish a baseline level of risk by defining
For example, ferry services often operate regularly
a baseline scenario and using the developed
during daytime and cease operation at night, whereas
model to quantify the risk of that scenario.
very large crude carriers can pass through narrow wa-
5. Model the effect of all the risk interventions
terways only during high tide. Causation-based math-
in terms of changes to the model parameters.
ematical models cannot capture time-dependent pat-
6. Implement these changes to the model and
terns well. Computer simulation could be a better
evaluate the risk interventions relative to the
alternative for such cases. The maritime simulation
established baseline level of risk.
model is based on an interaction counting model,
similar in approach to the geometrical probability es- Another team from Turkey took a different ap-
timations. proach to looking at narrower waterways that re-
A joint research group involving three quires less computational time. This approach simply
universities—the George Washington University, partitions the waterway into slices (or cells), then
Maritime Waterway Quantitative Risk Assessment Models 505
takes the vessel of interest and counts its interac- Table III. Fire and Explosion Frequencies (Number of
tions with all other vessels in the same slice.(45) The Accidents per Ship-Hour)(12)
model has been applied to assess risks in the Strait of
Ship Type Frequency
Istanbul. Several attributes—vessel type and length,
vessel reliability (age and flag), pilot request, tug re- Tankers 4.08E-07
quest, vessel proximity, visibility conditions, current, Bulk ships 3.43E-07
geographical difficulty, and local traffic density—are General cargo ships 2.65E-07
Ferries 3.51E-07
taken into account in the Turkish model.
The Hong Kong Marine Department(46) pre-
sented a brief risk assessment project looking at the
Hong Kong port waters. First, a comprehensive anal- Table IV. Estimated Frequency of Initiating Events for Crude
Oil Ships(2)
ysis of marine traffic activities within Hong Kong wa-
ters was conducted by capturing the latest traffic ac- Frequency of Accidents
tivity through field surveys. Second, the study built a Initial Event (per ship-year)
marine traffic risk assessment simulation model, and
Collision 1.03E-02
the risks of future scenarios were compared with that
Contact 3.72E-03
of the baseline scenario. Grounding 7.49E-03
Qu and Meng(47) proposed a simulation ap- Fire 3.65E-03
proach based on a cellular automata model, focusing Explosion 1.90E-03
on the Singapore Strait. Discrete event models are Double hull (DH) ships: 1.93E-03,
NASF All ships: 5.74E-03
applied to generate vessels of different categories
and velocities from four portals of the straits. Ten
ship-following and five ship-crossing rules are used
to simulate mariners’ responses to various naviga- Table V. Estimated Frequency of Initiating Events for
Container Ships(48)
tional scenarios. This is the first attempt to apply
the relatively mature simulation models of roadway Accidents Frequency
vehicle movement to maritime transportation by Accident Scenario (per ship-year)
taking into account ship interactions. The model
can be used to generate the lateral distributions of Collision 1.61E-02
Contact 3.65E-03
a ship’s trajectory (needed for Pedersen’s model) Grounding 6.84E-03
and to estimate a ship’s domain violation (for Fujii’s Fire/explosion 3.55E-03
model). Heavy weather 2.64E-03
Table VI. Estimated Frequency of Initiating Events for Various Ship Types(49)
Table VII. Historical Accident Statistics for Various Initiating Events: Cruise Ships(50)
Table VIII. Estimated Frequency of Initiating Events for further research. However, because of their com-
Ro-Pax Ships(51) plexity and the large amount of expert knowledge
Frequency (per ship year)
and other resources incorporated into developing the
event trees, it is not possible for other researchers to
Initiating Event Total Serious validate the results.
Collision 1.25E-02 1.29E-03
Contact 1.25E-02 1.36E-03 3.2. Mechanical Model and Simulation for
Fire/explosion 8.28E-03 3.23E-03
Wrecked/stranded 9.57E-03 3.04E-03
Consequence Estimation
Hull damage 2.26E-03 4.53E-04 For particular scenarios in an event tree, me-
Foundered 1.29E-04 1.29E-04
Machinery damage 1.87E-02 2.00E-03
chanical models and simulation methods have been
Miscellaneous 4.07E-03 1.94E-04 developed to estimate the expected damage to ships.
The damage to the ship’s body as a result of an acci-
dent can be estimated using these models.
Minorsky (54) proposed a method for calculating
distinct consequences. Ronza et al.(52) also con- the kinetic energy when two ships collide, on the ba-
structed event trees for consequence estimation us- sis of the following assumptions:
ing historical data analysis.
Some conceptual event tree structures from the 1. The collision is totally inelastic;
IMO’s FSA reports(2,48−51) are shown in Figs. 11–15.2 2. The system kinetic energy along the struck
Note that the event tree structures are different for ship’s longitudinal direction is negligible; and
distinct vessel types and accident categories. Major 3. The rotations of the struck and striking ships
factors affecting the consequences of an accident are are small and can be neglected.
the vessel speed (operational state), the extent of On the basis of the conservation of momentum,
damage, survivability (sink or stay afloat), the evacu- the final velocity of the two ships is given by:
ation model, and the dangerous cargo leakage model
(related to vessel type and cargo type). MBv B
v= , (21)
The series FSA reports from the IMO laid solid MA + MB + dmA
foundations for a vessel’s risk assessment. The event where MA is the mass of the struck ship, MB is the
trees developed in the reports provide guidelines for mass of the striking ship, dmA is the added mass of
the struck ship in the sway direction (it is assumed
2 The high-level risk model style was originally proposed by that dmA = 0.4MA), v is the final velocity in the ver-
Vanem et al.(53) tical direction (toward the struck ship’s centerline),
Maritime Waterway Quantitative Risk Assessment Models 507
Probability of DG Evacuation
and fuel oil release Probability of timely evacuation
model
Operational
Probability of being in port, restricted waters, or at sea
state
Fire protection
Probability of fire detection / fire fighting system failing
model
Probability of DG Evacuation
and fuel oil release Probability of timely evacuation
model
Fig. 12. High-level risk model for a fire/explosion scenario: container ship.(48)
and vB is the initial velocity of the striking ship in the Pedersen and Zhang(28) proposed an analytical
vertical direction. The total kinetic energy absorbed model to quantify ship damage when the characteris-
in the collision is: tics of both the striking and struck ships are known.
External and internal dynamics are considered. The
1 1 method is applied to a specific RoRo passenger ship.
KE = MBv 2B − (MA + MB + dmA)v 2 . (22)
2 2 Chen(56) developed a simplified collision model
(SIMCOL) on the basis of a time domain simulta-
On the basis of an investigation of 26 ship-to-ship neous solution of external dynamics and internal de-
collisions, the volume of damaged structural steel as formation mechanics. The model was built into soft-
a result of the energy absorbed during the collision ware and has been used in a tanker design study.(57)
can be derived. The SIMCOL model is also described in Brown and
Servis and Samuelides(55) analyzed the damage Chen(58) and in Brown.(59)
to the struck ship by using the finite element tech- Van de Wiel and van Dorp(60) developed an
nique. Their method can be used to assess the col- oil outflow model for collision and grounding acci-
lision behavior of a ship under a specific collision dents involving tankers. Various factors, such as the
scenario and thus to compare the survivability of dif- tankers’ hull design (single or double), the struck
ferent structural arrangements. ship’s displacement and speed, the striking ship’s
508 Li, Meng, and Qu
Collision
Collision frequency
frequency model
Loading
Probability of being loaded/in ballast
condition model
Fig. 13. High-level risk model for a collision scenario: LNG carrier.(53)
Grounding
Grounding frequency
frequency model
Loading
Probability of being loaded/in ballast
condition model
Fig. 14. High-level risk model for a fire/explosion scenario: LNG carrier.(53)
displacement and speed, and the interaction angle, Some relevant studies focus on the energy
are considered in the model. The extent of the lon- calculation,(61−67) some models analyze the impact on
gitudinal and transversal damage to the tanker can a ship’s structure,(68−79) and others evaluate the ship’s
then be calculated. On the basis of the estimates of performance and oil spill scales after accidents.(80−85)
the damage to the ship, the volume of oil outflow can Recently, Pedersen(6) presented a review paper
be estimated. on ship collision and grounding analysis tools.
Maritime Waterway Quantitative Risk Assessment Models 509
Number of fatalities,
Consequence environmental pollution,
property
Fig. 15. High-level risk model for a collision scenario: oil tanker.(2)
Pedersen’s paper introduces a mathematical model situation, among other factors. It is universally ac-
for the estimation of collision probabilities and some knowledged that uncertainty is an unavoidable com-
mechanical methods for estimating ship damage. ponent in any risk assessment procedure. However,
the lack of data, combined with this parameter un-
certainty, exacerbates the difficulties. The issue of
4. DISCUSSIONS AND RECOMMENDATIONS propagating uncertainty in maritime transportation
Although the framework of maritime risk assess- systems has been addressed by the GWU-RPI-VCU
ment has been well developed, two issues are still team.(17,43,87) They apply the Bayesian simulation
of concern. One major concern is how to quantify technique to propagate uncertainty throughout their
the effect of human errors. Statistics show that about analysis, and their case study (the Washington State
80% of collisions occur because of human error,(86) ferries and ferry service expansions in San Francisco
and it is the primary cause of most transportation Bay) shows that the results are robust to the inherent
accidents.(40) The IMO has provided some guidance uncertainties. It would be of great significance to con-
on the use of human reliability analysis,(1) incorpo- duct further studies incorporating the Bayesian simu-
rating some generic tools, such as the technique for lation technique into other risk assessment models.
human error rate prediction and the human error
assessment and reduction technique. However, very
5. CONCLUDING REMARKS
few research studies have followed along these lines.
Modeling human error is still challenging in a mar- In this study, we have reviewed the literature
itime risk assessment context, and future research on QRA models for shipping waterways, comprising
would be highly desirable. It is also extremely neces- frequency and consequence analysis. The frequency
sary that maritime authorities collect near-miss and estimation models are illustrated and discussed in
human-error data similar to that collected in the avi- Section 2. Event tree analysis, a mechanical model,
ation domain. and simulation models for consequence analysis are
The other issue is how to address the parameter presented in Section 3. On the basis of our review,
uncertainties. On the basis of the above mentioned two issues are proposed as requiring further study:
approaches, risk assessment is determined by a vari- parameter uncertainty and the quantification of hu-
ety of input parameters, such as the hydrographical man factors. Possible solutions are proposed in the
condition, ship flow characteristics, and navigational discussion and recommendations section.
510 Li, Meng, and Qu
model the impact of human error in a maritime system. Safety 60. Van de Wiel G, Van Dorp JR. An oil outflow model for tanker
Science, 1998; 30(1–2):235–247. collisions and groundings. Annals of Operations Research,
41. Merrick JRW, van Dorp JR, Harrald JR, Mazzuchi TA, 2011; 187(1):279–304.
Spahn JE, Grabowski M. A systems approach to managing 61. Pedersen PT, Zhang S. On impact mechanics in ship collisions.
oil transportation risk in Prince William Sound. Systems En- Marine Structures, 1998; 11(10):429–449.
gineering, 2000; 3(3):128–142. 62. Pedersen PT, Zhang S. Absorbed energy in ship collisions and
42. Merrick JRW, van Dorp JR, Mazzuchi TA, Harrald JR. Mod- grounding-revising Minorsky’s empirical method. Journal of
eling risk in the dynamic environment of maritime transporta- Ship Research, 2000; 44(2):140–154.
tion. Pp. 1090–1098 in Winter Simulation Conference Pro- 63. Glykas A, Das PK. Energy conservation during ground-
ceedings, Arlington, VA, 2001. ing with rigid slopes. Ocean Engineering, 2001; 28(4):
43. Merrick JRW, van Dorp JR, Dinesh V. Assessing uncertainty 397–415.
in simulation-based maritime risk assessment. Risk Analysis, 64. Glykas A, Das PK. Energy conservation during a tanker colli-
2005; 25(3):731–743. sion. Ocean Engineering, 2001; 28(4):361–374.
44. van Dorp JR, Merrick JRW. On a risk management analy- 65. Glykas A, Das PK, Barltrop N. Application of failure and frac-
sis of oil spill risk using maritime transportation system sim- ture criteria during a tanker head-on collision. Ocean Engi-
ulation. Annals of Operations Research, 2011; 187(1):249– neering, 2001; 28(4):375–395.
377. 66. Lehmann E, Peschmann J. Energy absorption by the steel
45. Ulusçu ÖS, Özbaş B, Altiok T, Or I. Risk analysis of the structure of ships in the event of collisions. Marine Structures,
vessel traffic in the Strait of Istanbul. Risk Analysis, 2009; 2002; 15(4–5):429–441.
29(10):1454–1472. 67. Tabri K, Broekhuijsen J, Matusiak J, Varsta P. Analytical
46. Hong Kong Marine Department. Marine Traffic Risk Assess- modelling of ship collision based on full-scale experiments.
ment for Hong Kong Waters (MARA Study). Hong Kong: Marine Structures, 2009; 22(1):42–61.
Hong Kong Marine Department, Report Ref: R8152/20 Issue 68. Asadi GV, Vaughan H. Dynamics and structural damage of
1, March, 2004. tanker ships running aground. Journal of Hazardous Materi-
47. Qu X, Meng Q. A Simulation Model for Ship Movements als, 1990; 25(1–2):61–73.
in the Singapore Strait and its Applications. Presented at 69. Ohtsubo H, Kawamoto Y, Kuroiwa T. Experimental and
the 90th Annual Meeting of Transportation Research Board. numerical research on ship collision and grounding of oil
Washington, DC, January 23–27, 2011. tankers. Nuclear Engineering and Design, 1994; 150(2–3):
48. IMO. Formal Safety Assessment FSA—Container Vessels. 385–396.
London: International Maritime Organization, Document 70. Paik JK, Pedersen PT. Modelling of the internal mechanics in
No.: MSC83/INF.8, July 3, 2007. ship collisions. Ocean Engineering, 1996; 23(2):107–142.
49. IMO. Formal Safety Assessment FSA—Liquefied Natural 71. Simonsen BC, Wierzbicki T. Grounding bottom damage and
Gas(LNG) Carriers. London: International Maritime Organi- ship motion over a rock. International Journal of Offshore and
zation, Document No.: MSC83/INF.3, July 3, 2007. Polar Engineering, 1996; 6(3):195–202.
50. IMO. Formal Safety Assessment FSA—Cruise Ships. London: 72. Simonsen BC. Ship grounding on rock—I. Theory. Marine
International Maritime Organization, Document No.: Structures, 1997; 10(7):519–562.
MSC85/INF.2, July 21, 2008. 73. Simonsen BC. Ship grounding on rock—II. Validation and ap-
51. IMO. Formal Safety Assessment FSA—RoPax Ships. Lon- plication. Marine Structures, 1997; 10(7):563–584.
don: International Maritime Organization, Document No.: 74. Pedersen PT, Zhang S. Effect of ship structure and size on
MSC85/INF.3, July 21, 2008. grounding and collision damage distributions. Ocean Engi-
52. Ronza A, Félez S, Darbra RM, Carol S, Vı́lchez JA, neering, 2000; 27(11):1161–1179.
Casal J. Predicting the frequency of accidents in port ar- 75. Kitamura O. FEM approach to the simulation of collision
eas by developing event trees from historical analysis. Jour- and grounding damage. Marine Structures, 2002; 15(4–5):
nal of Loss Prevention in the Process Industries, 2003; 16(6): 403–428.
551–560. 76. Wang G, Chen Y, Zhang H, Peng H. Longitudinal strength
53. Vanem E, Antão P, de Comas FDC, Skjong R. Formal safety of ships with accidental damages. Marine Structures, 2002;
assessment of LNG tankers. In 10th International Symposium 15(2):119–138.
on Practical Design of Ships and Other Floating Structures, 77. Ozguc O, Das PK, Barltrop N. A comparative study on the
PRADS, Houston, 2007. structural integrity of single and double side skin bulk carri-
54. Minorsky VU. An analysis of ship collisions with reference to ers under collision damage. Marine Structures, 2005; 18(7–8):
protection of nuclear power plants. Journal of Ship Research, 511–547.
1959; 3(2):1–4. 78. Ehlers S, Broekhuijsen J, Alsos HS, Biehl F, Tabri K. Simu-
55. Servis DP, Samuelides M. Ship collision analysis using finite el- lating the collision response of ship side structures: A failure
ements. In SAFER EURORO Spring Meeting, Nantes, 1999. criteria benchmark study. International Shipbuilding Progress,
56. Chen D. Simplified Ship Collision Model, Master Thesis,: 2008; 55(1–2):127–144.
Virginia Polytechnic Institute and State University, Blacks- 79. Luı́s RM, Teixeira AP, Guedes Soares C. Longitudinal
burg, Virginia; January, 2000. strength reliability of a tanker hull accidentally grounded.
57. Transportation Research Board. Environmental Performance Structural Safety, 2009; 31(3):224–233.
of Tanker Designs in Collision and Grounding, Methods for 80. Cuesta I, Grau FX, Giralt F. Numerical simulation of oil spills
Comparison. Washington, DC: Marine Board, Transportation in a generalized domain. Oil and Chemical Pollution, 1990;
Research Board, The National Academies, National Academy 7(2):143–159.
Press, Special report 259, 2001. 81. Sirkar J, Ameer P, Brown AJ, Goss P, Michel K, Nicastro F,
58. Brown A, Chen D. Probabilistic method for predicting ship Willis W. A framework for assessing the environmental per-
collision damage. Ocean Engineering International Journal, formance of tankers in accidental grounding and collision.
2002; 6(1):54–65. SNAME Transactions, 1997; 105:253–295.
59. Brown A. Collision scenarios and probabilistic colli- 82. Rawson C, Crake K, Brown AJ. Assessing the environmental
sion damage. Marine Structures, 2002; 15(4–5):335– performance of tankers in accidental grounding and collision.
364. SNAME Transactions, 1998; 106:41–58.
512 Li, Meng, and Qu
83. Skjong R, Vanem E. Damage stability evaluation in collision 85. Fang C, Das PK. Survivability and reliability of damaged
of bulk carriers. In Proceedings of the 3rd International Con- ships after collision and grounding. Ocean Engineering, 2005;
ference on Collision and Grounding of Ships (ICCGS2004), 32(3–4):293–307.
Japan, 2004. 86. Macrae C. Human factors at sea: Common patterns of error in
84. Vanem E, Skjong R. Collision and grounding of pas- groundings and collisions. Maritime Policy and Management,
senger ships—Risk assessment and emergency evacuations. 2009; 36(1):21–38.
In Proceedings of the 3rd International Conference on 87. Merrick JRW, van Dorp JR, Singh A. Analysis of correlated
Collision and Grounding of Ships (ICCGS2004), Japan, expert judgments from extended pairwise comparisons. Deci-
2004. sion Analysis, 2005; 2(1):17–29.