Collision and grounding of ships and offshore structures

Jørgen Amdahl

Sören Ehlers

Bernt J. Leira

an informa business

E. Kim

and Offshore Structures

Collision and Grounding of Ships

Collision and Grounding of Ships

and Offshore Structures

Amdahl

Ehlers

Leira

Collision and Grounding of Ships and Offshore Structures contains the latest

research results and innovations presented at the 6th International Conference on

Collision and Grounding of Ships and Offshore Structures (Trondheim, Norway, 17-

19 June 2013). The book comprises contributions made in the fi eld of numerical and

analytical analysis of collision and grounding consequences for ships and offshore

structures in various scenarios, such as narrow passageways and arctic conditions

including accidental ice impact. A wide range of topics is covered:

- Recent large-scale collision experiments

- Innovative concepts and procedures to improve the crashworthiness of ships and

offshore structures

- Ship collisions with offshore renewable energy installations

- Residual strength of damaged ship structures as well as mitigation measures for

the consequences of such accidents

- Statistical analysis of collision and grounding incidents to analyse and predict the

probability of their occurrence

- Developments concerning rational rules for structural design to avoid collisions

- Grounding actions comprising the use of general risk assessment methodologies

Collision and Grounding of Ships and Offshore Structures contributes signifi cantly

to increasing the safety and reliability of seaborne transport and operations, and

will be useful to academics and engineers involved in marine technology-related

research and the marine industry.

COLLISION AND GROUNDING OF SHIPS AND OFFSHORE STRUCTURES

This page intentionally left blank

PROCEEDINGS OF THE 6TH INTERNATIONAL CONFERENCE ON COLLISION AND GROUNDING

OF SHIPS AND OFFSHORE STRUCTURES, ICCGS, TRONDHEIM, NORWAY, 17–19 JUNE 2013

Collision and Grounding of Ships

and Offshore Structures

Editors

Jørgen Amdahl, Sören Ehlers & Bernt J. Leira

Department of Marine Technology, Norwegian University of Science and Technology,

Trondheim, Norway

CRC Press/Balkema is an imprint of the Taylor & Francis Group, an informa business

© 2013 Taylor & Francis Group, London, UK

Typeset by MPS Limited, Chennai, India

Printed and bound in Great Britain by CPI Group (UK) Ltd, Croydon, CR0 4YY.

All rights reserved. No part of this publication or the information contained herein may be

reproduced, stored in a retrieval system, or transmitted in any form or by any means,

electronic, mechanical, by photocopying, recording or otherwise, without written prior

permission from the publishers.

Although all care is taken to ensure integrity and the quality of this publication and the

information herein, no responsibility is assumed by the publishers nor the author for any

damage to the property or persons as a result of operation or use of this publication

and/or the information contained herein.

Published by: CRC Press/Balkema

P.O. Box 11320, 2301 EH, Leiden, The Netherlands

e-mail: [email protected]

www.crcpress.com – www.taylorandfrancis.com

ISBN: 978-1-138-00059-9 (Hbk + CD-ROM)

ISBN: 978-1-315-88489-9 (eBook)

Collision and Grounding of Ships and Offshore Structures – Amdahl, Ehlers & Leira (Eds)

© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00059-9

Table of contents

Foreword VII

Feasibility of collision and grounding data for probabilistic accident modeling 1

M. Hänninen, M. Sladojevic, S. Tirunagari & P. Kujala

Bridge crossings at Sognefjorden – Ship collision risk studies 9

M.G. Hansen, S. Randrup-Thomsen, T. Askeland, M. Ask, L. Skorpa, S.J. Hillestad & J. Veie

VTS a risk reducer: A quantitative study of the effect of VTS Great Belt 19

T. Lehn-Schiøler, M.G. Hansen, K. Melchild, T.K. Jensen, S. Randrup-Thomsen, K.A.K. Glibbery,

F.M. Rasmussen & F. Ennemark

An improvement on a method for estimating number of collision candidates between ships 27

F. Kaneko

Modeling and simulation system for marine accident cause investigation 39

S.G. Lee, S.H. Jun & G.Y. Kong

Development of vessel collision model based on Naturalistic Decision Making model 49

M. Asami & F. Kaneko

Material characterization and implementation of the RTCL, BWH and SHEAR failure

criteria to finite element codes for the simulation of impacts on ship structures 57

J.N. Marinatos & M.S. Samuelides

Prediction of failure strain according to stress triaxiality of a high strength marine structural steel 69

A. Woongshik Nam & J. Choung

Fracture mechanics approach to assess the progressive structural failure of a damaged ship 77

A. Bardetsky

Evaluation of the fendering capabilities of the SPS for an offshore application 85

G. Notaro, K. Brinchmann, E. Steen & N. Oma

Collision tests with rigid and deformable bulbous bows driven against double hull side structures 93

I. Tautz, M. Schöttelndreyer, E. Lehmann & W. Fricke

Side structure filled with multicellular glass hollow spheres in a quasi-static collision test 101

M. Schöttelndreyer, I. Tautz, W. Fricke & E. Lehmann

Response of a tanker side panel punched by a knife edge indenter 109

R. Villavicencio, B. Liu & C. Guedes Soares

A study on positive separating bulbous bow 117

B. Li, L.S. Zhang & L.P. Sun

Calculation of a stranding scenario 127

B. Zipfel & E. Lehmann

Grounding resistance capacity of a bulk carrier considering damage confined to the bow 135

Y. Quéméner & C.H. Huang

Loading on stranded ships 143

C. Souliotis & M.S. Samuelides

Plastic mechanism analysis of structural performances for stiffeners on outer bottom plate

during shoal grounding accident 151

Z. Yu, Z. Hu, G. Wang & Z. Jiang

V

A simplified approach to predict the bottom damage in tanker grounding 161

M. Heinvee, K. Tabri & M. Kõrgesaar

Residual ultimate longitudinal strength – grounding damage index diagram of a corroded

oil tanker hull structure 171

D.K. Kim, H.B. Kim, X.M. Zhang, J.K. Paik & J.K. Seo

Towards an integrated approach to collision and grounding damage assessment 179

E. La Scola & G. Mermiris

Towards more rational design of ship structures against collisions 187

S.R. Cho, J.M. Kim, Y.H. Kim, J.S. Lee & M.I. Roh

Structural safety assessment of ship collision and grounding using FSI analysis technique 197

S.G. Lee, T. Zhao & J.H. Nam

Ship-ice collision analysis to define ice model according to the IACS Polar Rule 205

M.J. Kwak, J.H. Choi, O.J. Hwang & Y.T. Oh

On the plastic and fracture damage of polar class vessel structures subjected to impact loadings 213

D.K. Min, Y.M. Heo, D.W. Shin, S.H. Kim & S.R. Cho

Review of existing methods for the analysis of the accidental limit state due to ice actions 221

E. Kim & J. Amdahl

A particle swarm optimization-based procedure to obtain a crashworthy ice-classed LNG tanker 233

S. Ehlers

Drop tests of ice blocks on stiffened panels with different structural flexibility 241

E. Kim, M. Storheim, J. Amdahl, S. Løset & R. von Bock und Polach

Risk analysis for offloading operations in the Barents, Pechora and Caspian seas 251

N.G. Popov, L.G. Shchemelinin & N.A. Valdman

Safe jacket configurations to resist boat impact 261

A.W. Vredeveldt, J.H.A. Schipperen, Q.H.A. Nassár & C.A. Spaans

Collision between a spar platform and a tanker 267

T. de Jonge & L. Laukeland

Ship collisions against wind turbines, quays and bridge piers 273

P.T. Pedersen

Experimental and numerical investigations on the collision between offshore wind turbine support

structures and service vessels 281

S.R. Cho, B.S. Seo, B.C. Cerik & H.K. Shin

Ultimate strength of an intact and damaged LNG vessel subjected to sub-zero temperature 289

S. Ehlers, S. Benson & K. Misirlis

Ultimate strength of damaged hulls 297

C. Pollalis & M.S. Samuelides

Longitudinal strength assessment of damaged box girders 305

S. Benson, M. Syrigou & R.S. Dow

The analysis and comparison of double side skin crashworthiness 315

A.Y.F. Gong, J.X. Liu, B.S.M. Xiao & N. Wang

A methodology for comparison and assessment of three crashworthy side-shell structures:

The X-core, Y-core and corrugation panel structures 323

J.W. Ringsberg & P. Hogström

Crashworthiness study of LPG ship with type C tanks 331

S. Rudan, B. Ašˇci´c & I. Viši´c

Study on influence of striking bow strength to the side structure during ship collision 339

K. Liu, Y. Zhang & Z. Wang

Author index 345

VI

Collision and Grounding of Ships and Offshore Structures – Amdahl, Ehlers & Leira (Eds)

© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00059-9

Foreword

We are pleased to host the 6th International Conference on Collision and Grounding of Ships and Offshore

Structures in Trondheim this year. This conference has now served for almost two decades as an important

and internationally recognized platform to disseminate the latest research results in the field of collision and

grounding of ships and offshore structures.

The preparation of this conference and proceedings would not have been possible without the excellent support

from Frank Klæbo, Martin Storheim and Ekaterina Kim and we would like to express our thankfulness to them.

In addition, we would like to thank Leila Dashtizadeh and Rouzbeh Siavashi for their efforts in formatting

the manuscripts where needed. Furthermore, we would like to thank the steering committee for promoting and

supporting the conference as well as the reviewers for their valuable contributions to this event. The financial

support of DNV, MARINTEK and DYNAmore Nordic is also greatly acknowledged Finally, we are wishing all

participants a fruitful, stimulating and professionally rewarding stay at NTNU’s Marine Technology Centre.

the editors

VII

This page intentionally left blank

Collision and Grounding of Ships and Offshore Structures – Amdahl, Ehlers & Leira (Eds)

© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00059-9

Feasibility of collision and grounding data for probabilistic

accident modeling

M. Hänninen, M. Sladojevic, S. Tirunagari & P. Kujala

Aalto University, Department of Applied Mechanics, Espoo, Finland

ABSTRACT: There exist various sources of data related to marine traffic safety, and the amount of data seems

to be further growing in the future. However, the data sets have different formats, scopes, and initial purposes.

The paper discusses the feasibility of maritime traffic accident and incident data to probabilistic modeling of

collision and grounding accidents, especially their causal factors. In addition, a case study is conducted for

examining the data feasibility. First, categorical Finnish accident causal data is utilized in learning a Bayesian

network model from the data. The data feasibility is then evaluated based on the how well the model matches

to unseen accident cases and how it performs in classification of the accidents. The results indicate that the

dataset does not contain enough information for the applied of modeling approach. Finally, recommendations

to improving the data or ways to cope with the uncertainty are given.

1 INTRODUCTION

The purpose of accident modeling is to learn more

about accidents in order to prevent them in the

future. Probabilistic accident models, depending on

the underlying theoretical accident model type used

(see e.g. Hollnagel 2004), quantitatively describe

accident causes, mechanisms, event chains, or sys￾tem variability. Such a model could be utilized within

a cost-benefit analysis, risk management or safety￾related decision making. However, a ship, and further

the marine traffic system as a whole, can be considered

as a complex socio-technical system. In such a system

an accident is hardly ever a result of a single cause or a

chain of events (Hollnagel, 2004). On the other hand,

accidents are low probability events and thus relatively

little data about accidents exists. Therefore, the lack of

data combined to the complexity of the problem might

result in unreliable or invalid probabilistic models.

This paper discusses the feasibility of ship acci￾dent data for probabilistic collision and/or grounding

modeling purposes. In addition, as incidents or near￾misses occur more frequently than accidents but might

be partly governed by the same underlying mecha￾nisms and thus could provide additional information

about marine traffic accidents (Harrald et al. 1998),

also incident data is considered. The study is based

on examining the data itself when available, review￾ing relevant literature, and a case study of evaluating

accident data feasibility to learning a Bayesian net￾work model of the dependencies between the reported

accident causes. The examination is limited to acci￾dent databases providing categorical information on

the accidents, accident investigation reports, a near￾miss reporting database, and Vessel Traffic Service

(VTS) violation and incident reports. Other potential

data sources such as Port State Control inspection data,

occupational safety data, data from insurance compa￾nies or classification societies are not addressed. The

systems and practices of accident or incident report￾ing or the corresponding data formats might differ

from country to country. Here the emphasis is on data

describing the marine traffic in Finland.

The rest of the paper is organized as follows.

Chapters 2–4 describe the features of the aforemen￾tioned accident and incident data sources and discusses

their feasibility to probabilistic collision and ground￾ing modeling. Chapter 5 presents the data, methods,

results and discussion of the case study, learning

a Bayesian network of reported accident causes in

Finnish collisions and groundings. Finally, conclu￾sions from the data, the literature review and the case

study results are drawn in Chapter 6.

2 ACCIDENT DATABASES

2.1 EMCIP

All Member States of the European Union are obli￾gated to report any marine casualty or accident occur￾rence involving merchant ships, recreational crafts

and inland waterway vessels to the European Marine

Casualty Information Platform (EMCIP) operated by

European Maritime Safety Agency EMSA (Correia

2010). In EMCIP, the casualty events are classified

into 25 event types. Collisions and groundings can

be categorized as a collision with another ship, a col￾lision with multiple ships, a collision when the ship

is not underway, contact with floating cargo, contact

1

with ice, contact with other floating object, contact

with unknown floating object, contact with a fixed

object, contact with a flying object, drift ground￾ing/stranding, or powered grounding/stranding. The

collected information is divided into factual data and

casualty analysis data. To describe the sequence of

the events related to a casualty, the results obtained in

the Casualty analysis methodology for maritime oper￾ations (CASMET) project (Caridis 1999) are used.

Special focus has been paid to verifying the quality

of the reporting and accomplishing the application of

the same principles in the investigations of casualties

and data analyses across the EU (EMSA 2010).

EMCIP database had operated on a voluntary basis

for two years until June 2011 when it became manda￾tory. Therefore, the data it currently contains might

still be too scarce for probabilistic modeling purposes.

Further, all accidents stored in the system are available

only to EMSA. A particular Member State has access

only to her own data, and the access is only granted to

authorities. Nevertheless, despite the low number of

records and the limited access to the system, EMCIP

manages to establish a common taxonomy. This could

facilitate different comparison studies in the future.

So far the Finnish EMCIP data has only been uti￾lized in reporting marine traffic accident statistics for

the years 2009–2010 (Trafi 2011) and the authors were

unable to find studies of further accident modeling

based on EMCIP data. Due to researchers not having

access to the data, further examination of its feasibility

is impossible.

2.2 DAMA

Before EMCIP, from the year 1990 to 2010, the marine

accidents of Finnish vessels and accidents to foreign

vessels within Finnish territorial waters were stored

in accident database DAMA (Laiho 2007, Kallberg

2011). In 2001–2005, the average number of acci￾dent cases stored per year was 50, of which 15 were

groundings and 5 collisions (Laiho 2007).

DAMA included 20 accident type categories,

including ship-ship collision, collision with an off￾shore platform, collision with a floating object, col￾lision with a bridge or quay, and grounding/stranding.

Besides the accident type, DAMA entries included

fields listed inTable 1. However, not all fields had been

filled in all accident cases. DAMA had 78 alternatives

for the accident causes and a possibility to report up

to four causes per accident. These causes had been

categorized under the following seven cause groups:

external factors; ship structure and layout; technical

faults in ship equipment; factors related to equipment

usage and placement; cargo, cargo and fuel handling

and related safety equipment; communication, orga￾nizing, instructions and routines; and people, situation

assessment, actions.

Based on the DAMA data, statistical analyses of

accident characteristics such as ship types, circum￾stances and causes have been conducted (Heiskanen

2001, Laiho 2007, Kujala et al. 2009). However, more

Table 1. Data fields in the DAMA accident database.

Field Format Field Format

Case number number Country text

Ship name text Waters cat

Home port text Voyage phase cat

Nationality text Working ac. cat

Type of ship cat Wind direction cat

Constuction year number Wind force cat

Renovation year number Sea cat

Material cat Visibility cat

GRT number Light cat

DWT number Cargo cat

Length number Pilot onboard y/n

Classification soc. text 2. ship name text

Year number 2. ship nation text

Month number Loss/dam. severity cat

Day number Evacuated y/n

Time of event number Hull damage y/n

Day of the week number Hull dam. severity cat

Event #1 cat Damage length Number

Event #2 cat Damage width Number

Event #3 cat Damage depth Number

Cause #1 cat Hull dam. locat. y cat

Cause #2 cat Hull dam. locat. z cat

Cause #3 cat Hull dam. locat. x cat

Cause #4 cat Death people Number

Departure port text Injured people Number

Destination port text Oil pollution Number

Latitude number Bridge manning Free text

Longitude number Damages Free text

in-depth analyses of the marine accidents in Finland,

such as an analysis of the correlations between the

different factors, or studies for finding subgroups or

clusters within the accidents, could not be found.

2.3 HELCOM

Baltic Marine Environment Protection Commission

HELCOM (Helsinki Commission) gathers data on

Baltic Sea accidents (HELCOM 2012a) covering all

accidents of tankers over 150 GT and/or other ships

over 400 GT within the states’ territorial waters or

EEZs. Due to a change in the reporting format, the

data before 2004 and the subsequent years are not fully

comparable. In 2005–2009, the average annual num￾ber of accidents in HELCOM database was 125. The

accident dataset, from 1989 on, can be accessed online

with a map based web tool (HELCOM 2012b) and is

also available on request.

HELCOM database accidents are divided into colli￾sions, fire, groundings, machinery damages, physical

damages, pollutions, sinkings, technical failures and

other accidents. Collisions can be further classified as

collisions with another vessel, with an object, or as the

ones with another vessel and an object. The HELCOM

data fields and the numbers of times the field has been

filled in the 1989–2009 data can be seen in Table 2.

Only one cause per involved ship is reported.The cause

categorization into a human factor, a technical factor,

2

Table 2. Data fields in HELCOM accident database. The

number of times reported describes the number of cases

where the corresponding field has not been left blank or

reported as “n.i.”, “unknown” etc. in 1989–2009. Ship2 size

values were found to be identical to the reported Ship1 size in

all but one collision with another vessel, so its correctness can

be questioned and the reporting percentage is not presented

in the table.

# of times Reporting

Data field Entry format reported (%)

Date dd.mm.yyyy 1251 100,0%

Ship1 name text 1251 100,0%

Ship2 name text 145 100,0%*

Year numeric 1251 100,0%

Latitude numeric 1250 99,9%

Longitude numeric 1250 99,9%

Accident type cat. 1249 99,8%

Ship1 category cat. 1230 98,3%

Pollution no/yes/n.i. 1166 93,2%

Type of pollution text 133 93,0%***

Amount of poll. numeric 1021 81,6%

Collision type cat. 273 78,0%**

Ship1 type text 964 77,1%

Ship2 category cat. 108 75,0%*

Ship2 type text 87 65,9%*

Country text 756 60,4%

Ship1 size (gt) numeric 725 58,0%

Time hh.mm 646 51,6%

Ship2 size (gt) numeric 68 51,5%*

Cause, ship1 cat 616 49,2%

Ship1 draught (m) numeric/ 590 47,2%

interval

Pilot, ship1 cat 572 45,7%

Cargo type text 535 42,8%

Ice conditions no/yes/n.i. 507 40,5%

Damage text 478 38,2%

Cause, ship2 cat 46 34,8%

Accident details text 423 33,8%

Ship1 size (dwt) numeric 395 31,6%

Offence text 277 22,1%

Cause details text 274 21,9%

Assistance need text 209 16,7%

Ship1 hull single/ 170 13,6%

double/n.i.

Pilot, ship2 cat 15 11,4%*

Ship2 hull single/ 13 9,8%*

double/n.i.

Ship2 draught (m) numeric/ 55 4,4%*

interval

Additional info text 38 3,0%

Consequences/ text 36 2,9%

response actions

Amount of poll. (tons) numeric 15 1,2%

Crew trained in ice no/yes/n.i. 14 1,1%

navigation

Ship2 size (dwt) numeric 157 –

*of the collisions with another vessel.

**of collisions.

***of accidents with pollution.

an external factor, or another factor is coarser than the

one in DAMA. It is supplemented with a text field for

describing the cause in more detail. However, as can

be seen from Table 2, it has been filled in only 22% of

the cases.

From the data, HELCOM publishes annual acci￾dent statistics (HELCOM 2012c). A combination of

DAMA data and HELCOM data from the years 1997-

1999 and 2001–2006 were also used in evaluating

accident statistics for the Gulf of Finland (Kujala et al.

2009). Mazaheri et al. (in prep.) have studied corre￾lations between the ship traffic and the location of

the grounding accidents within the HELCOM data.

HELCOM data was also used by Hänninen & Kujala

(2013) when modeling the dependencies of the Gulf

of Finland Port State Control inspection findings and

accident involvement.

Compared with DAMA, HELCOM contains fewer

accidents from Finnish waters: as an example, in

DAMA there are 46 accidents from Finnish waters in

2004, whereas in HELCOM database the number is 8.

On the other hand, some of the accidents present in the

HELCOM data are missing from DAMA. Neverthe￾less, although not complete and even containing some

errors (Salmi 2010), at the moment HELCOM data is

the largest database with a uniform data format of the

Baltic Sea accidents.

3 ACCIDENT INVESTIGATION REPORTS

In Finland, Safety InvestigationAuthority (SIA) inves￾tigates and reports “all major accidents regardless of

their nature as well as all aviation, marine and rail

accidents and their incidents” (SIA 2012a). Marine

accidents are investigated if they have occurred within

Finnish waters, or if a Finnish vessel has been involved

in the accident. SIA investigates and reports how the

accident occurred, what were the circumstances, the

causes, the consequences and the rescue operations.

The reports based on the investigations also provide

recommendations of actions for preventing similar

accidents.

The marine traffic accident investigation reports of

accidents from 1997 on and 10 older reports canbe

downloaded from SIA web pages (SIA 2012b). In

October 2012, 187 reports of accidents, serious inci￾dents, incidents, damages, minor accidents and other

incidents were available.

Accident reports are in text format and their usage

typically requires human effort in extracting informa￾tion of interest from the text. The task can become

tedious while humans may not always be capable of

extracting the information objectively. Text mining is

an emerging technology that can be used to augment

existing data in electronic textual databases by making

unstructured text data available for analysis (Francis &

Flynn 2010).

Zheng & Jin (2010) used accident reports and a

text data mining technique called attribute reduction

for extracting the most frequent human factors which

they considered as reasons leading to human errors in

marine traffic accidents. Artana et al. (2005) devel￾oped and evaluated software utilizing text-mining for

encountering maritimmarinee hazards as well as a

risk management system covering organizations and

human resources. Tirunagari et al. (2012) applied NLP

methods text mining to cluster the marine accident

3

reports. However, the utilization of text mining is a

complex task as it involves addressing text data which

is very unstructured and fuzzy (Tan 1999). Moreover,

there are quite many challenges when accident reports

are concerned as the reports are written in natural lan￾guage with no standard template and often contain

misspellings and abbreviations. Also, the detection

of multi words such as “safety culture” is difficult

because it is not known which word is of greater

importance and the words “safety” and “culture”

have a different meaning when appearing separately

compared with when considered as a single word.

4 NEAR-MISS REPORTING

4.1 Insjö and ForeSea

ForeSea is an anonymous and voluntary experience

database initiated by Finnish and Swedish organi￾zations and government agencies. The aim of the

database is “to capture the conditions that are nor￾mally not reported to authorities” including accidents,

near misses and non-conformities (ForeSea 2012).

The database is a refined version of the Swedish Insjö

system which was launched in 2002 and the plan

is to replace Insjö with ForeSea.

In September 2011, twelve companies were report￾ing to ForeSea and 76 to Insjö (Bråfelt, pers. comm).

Approximately one report per year per ship has been

obtained to Insjö. On the 7th of December 2011, Insjö

contained 1282 accident reports, 841 near misses and

532 non conformity records. 1268 of these reports had

been transferred to ForeSea. After ForeSea becomes

fully operational in July 2013, every individual mem￾ber company will be required to provide reports to the

database every year.

The philosophy behind the ForeSea taxonomy is

“what can be got into”, compared with EMCIP’s phi￾losophy of “what the collector wants to get in” (Bråfelt,

pers. comm). The database administrator is responsi￾ble for classifying the event into 27 categories based on

his interpretation. Data can be separated into five main

categories: prerequisite data, the course of events, the

causes, the consequences, and the measures. Each of

these is further divided into subcategories. The causes

are divided into human/manning, working environ￾ment, marine environment, technical ship and cargo

and management causes.

Data stored in the Insjö database is available to four

categories of users with different rights and accesses

to features. Researchers have access to the most of the

features, including also a right to export data to Excel

format.

Insjö and ForeSea contain only a short description

of the event in narrative textual form, with very lit￾tle factual data available (the ship type, type of event,

the activity of the ship, the location) and its quality

depends on the reporter’s skills (Bråfelt, pers. comm).

As with accident investigation reports, the utilization

of the data would require information extraction from

the text, conducted either manually or possibly with

text mining. So far, the data has not been utilized for

even establishing trends (Erdogan 2011).

4.2 VTS violation and incident reporting

VesselTraffic Service (VTS) provides information and

navigational guidance to the vessels and can orga￾nize the traffic within a VTS monitoring area (FTA

2011). In the Gulf of Finland, areas not included in

the VTS areas are covered by Mandatory Ship Report￾ing System GOFREP.Within Finnish territorial waters,

vessels with a GT of at least 300 are obliged by law

to participate in the VTS monitoring and report their

arrival to the GOFREP area or when they are leaving

a port in the Gulf (FTA 2012).

VTS operators should report all violations they

observe within the Finnish VTS areas and the

GOFREP area. Also, incidents or near misses within

Finnish waters are reported. However, differences in

the numbers of reported violations between VTS oper￾ators have been detected (Talja, pers. comm.). In 2010,

a total number of 125 incident and violation reports

were made at the Gulf of Finland VTS center.

The format of the violation and especially the

incident reporting forms has slightly varied over the

years but the basic structure, a narrative text field

for describing the event and a few check box-type

options for the location or circumstances has remained

unchanged. The information the reports covered in the

first half of the year 2009 and the fill-up percentages

is presented in Tables 3–4. At the beginning of 2012,

the reporting system was reformed and all reporting is

to be done into an electrical system.

The work of the VTS was described both verbally

and statistically based on two two-week periods of

VTS operators reporting all situations requiring VTS

intervention (Westerlund 2011). Salmi (2010) used

violation reports for identifying accident-prone ves￾sels by comparing the vessels present at the violation

reports to HELCOM accident statistics. Unfortunately,

the categorized data the reports contain does not

provide much input to probabilistic models and the

information about the situation, the vessel(s) and the

circumstances must be transformed into categorical

data, which may introduce some uncertainty. On the

other hand, as with accident investigation and near

miss reports, finding the truth behind the textual infor￾mation may also be challenging. Nevertheless, the

advantage of VTS violation and incident reports is that

violations and incidents occur more frequently than

accidents and thus there is more data available.

5 CASE STUDY: FEASIBILITY OF

CATEGORICAL ACCIDENT CAUSE

DATA FOR LEARNING A BAYESIAN

NETWORK MODEL

5.1 Purpose of the case study

Although textual descriptions provide the rich infor￾mation on accidents, the terms or expressions when

4

Table 3. Information fields of the Finnish VTS violation

reports from the year 2009. In addition, a capture of the situ￾ation on ECDIS is attached to the report which may include

additional AIS information about the speed, course and head￾ing of the vessel. The filling percentages are calculated from

37 VTS violation reports from January–July 2009.

Type of Type of field

information Field and filling %

Vessel Name Text (100%)

identification Flag Text (100%)

Port of registry Text (65%)

Callsign Text (100%)

Type Text (100%)

IMO Number Text (100%)

MMSI Text (100%)

GT Text (76%)

Time Date and time Text (92%)

Position, speed Latitude & longitude Text (100%)

and course

Location Territorial waters of Check box

Finland/international (100%)

waters

Outside scheme/ Check box

Traffic Separation Check box/

Scheme/ Text (name)

Lane/ Check box/

Text (desc.)

Separation zone/ Check box

Other location Check box/

Text (desc.)

(76%)

Identification Plotted by Radar/ Check box

Plotted by AIS (89%)

Identified by Text (GOFREP

or VTS) (0%)

Weather Wind direction Text (68%)

Wind force (m/s) Text (68%)

Sea state (douglas) Text (22%)

Visibility (m) Text (8%)

Contravened Rule 10 (b) I Check box

regulations Rule 10 (b) ii Check box

Rule 10 (b) iii, joining Check box

Rule 10 (b) iii, leaving Check box

Rule 10 (c) Check box

Rule 10 (d) Check box

Rule 10 (e) Check box

Rule 10 (f) Check box

Rule 10 (g) Check box

Rule 10 (h) Check box

Rule 10 (i) Check box

Rule 10 (j) Check box

IMO Resolution Check box

MSC.139(76) Annex 1 Check box

Other rules Check box/

Text (95%)

Additional Details of the incident Text (97%)

information

referring to a similar factor or cause might vary, which

might complicate any probabilistic modeling based

on the data. Categorical accident information requires

less effort on preparing the data for probabilistic

analyses and removes the problem of unambiguity.

Table 4. Information fields of the Finnish VTS incident

reports from the year 2009. In addition, a capture of the situ￾ation on ECDIS is attached to the report which may include

additional AIS information about the vessel’s speed, course

and heading. The filling percentages are calculated from 21

incident reports from January–July 2009.

Type of Type of field

information Field and filling%

Vessel Name Text (95%)

identification Callsign Text (90%)

IMO Number Text (76%)

Pilot Text (38%)

Master Text (0%)

Time Date and time Text (100%)

Position, speed Position Text (86%)

and course Destination Text (81%)

Location Hanko VTS Check box

Helsinki VTS Check box

Kotka VTS Check box

GOFREP Check box

(95%)

Weather Weather Text (visib. 67%,

wind dir. 95%,

wind force 95%)

Type of Near miss Check box

non-conformity Accident Check box

AIS Check box

Environment Check box

Pilot Check box

Equipment Check box

Personal injuries Check box

Emergency Check box

Other Check box

(100% )

Additional Description of

information incident Text

Actions taken by Text (Descr. and/or

VTS Operator actions 100%)

Operator Text (100%)

Supervisor Text (95%)

However, fixed categorization might result informa￾tion loss and thus introduce uncertainty in the model.

Further, given the complexity of the problem, the cate￾gorization of accident causes should be rather detailed

while the dataset would need to be large in order to

have enough data within each category.

The aim of the case study is to evaluate if cate￾gorical accident-cause data is a feasible information

source for a probabilistic model of collisions and

groundings and their reported causes. The model is

constructed directly from the data. The feasibility is

evaluated based on how well the model matches to

unseen accident cases.

5.2 Data and methods

In order to avoid problems from taxonomy differences,

a single accident database is used as an input. As

EMCIP data is not available, DAMA accident database

is chosen as the input data due to featuring the most

5

Table 5. The number of cases in the dataset and the number

of cases with at least one, two or three reported causes.

Collisions Groundings Total

No. of cases 55 160 215

>0 reported causes 55 157 212

>1 reported causes 11 21 32

>2 reported causes 4 10 14

Table 6. The number of cause types within cause categories

of the dataset and the number of cases with the reported cause

category.

Collisions Groundings Total

People, situation 32 117 149

assessment,

actions (13)

External conditions (7) 33 29 62

Technical failure (5) 0 30 30

Communication 8 7 15

organization

procedures etc. (8)

Ship structure and 0 2 2

layout (2)

Equipment and 0 2 2

layout (1)

Other (1) 0 2 0

Total (37) 73 189 262

detailed cause categorization and a possibility to report

more than one cause per accident. Table 5 summa￾rizes the data consististing of 55 ship-ship collisions

and 160 grounding cases. The accidents have occurred

within 1997–1999 and January 2001–June 2006. From

the accidents, accident type (collision/grounding) and

the reported primary, secondary and third causes are

considered. 37 different cause types are present in the

dataset. These causes can be grouped into seven cate￾gories. The frequencies of these categories within the

data are shown in Table 6.

A Bayesian network (BN) model (Pearl 1988) con￾sisting of 38 variables in total is learned from the

data. In brief, BN is a graphical representation of

the joint probability distribution of a set of variables

describing a certain problem (Darwiche 2009). The

case study model variables describe the accident type

(collision/grounding) and whether each cause type had

been reported in an accident (yes/no). The graph struc￾ture is learned using NPC algorithm (Steck & Tresp

1999) whereas Expectation-maximization method

(Dempster et al. 1977) is applied for determining

the network probability parameters. Hugin Expert

software (Mädsen et al. 2005) is utilized in the

construction.

For evaluating the quality of the resulting model,

the dataset is divided into a training set (143 cases)

which is used for learning the model and a test set

(72 cases) for evaluating how well the resulted model

Figure 1. A part of the BN model structure learned from

the data with the NPC algorithm. The rest of the variables

are unconnected or pairwise connected variables and are not

shown.

Table 7. The performance of the Bayesian network model

given the test set compared to an empty graph (for the scores,

higher values are better).

BN model Empty graph

Log-likelihood score −599.8 −603.2

AIC score −659.8 −640.2

BIC score −728.1 −682.3

Classification error 16.7% 19.4%

Precision (collision) 0.625 0.000

Recall (collision) 0.357 0.000

F-measure 0.455 0.000

AUC 0.87 0.80

performs with unseen data. Log-likelihood score is

calculated for comparing the model fit to the test

set. However, as log-likelihood favors densely con￾nected networks, the Akaike Information Criterion

(AIC) (Akaike 1974) and the Bayesian Information

Criterion (BIC) (Schwarz 1978) scores, which addi￾tionally penalize a model based on its complexity, are

also determined. The scores are then compared with

the ones of an empty graph, i.e., a model with no depen￾dencies between the variables. In addition, the model’s

ability to correctly classify test set cases as collisions

is evaluated by calculating the collision misclassifica￾tion rate, precision and recall and the area under the

ROC-curve (AUC) (e.g. Bradley 1997).

5.3 Results and discussion

From the data, NPC algorithm learns a Bayesian

network of ten connected variables (including the

event type and the presences of nine cause types), 17

unconnected cause type variables, and five pairs of

dependent causes. Figure 1 presents the ten connected

variables.

The data itself (Table 5) already suggests that it

cannot produce a very informative model on the con￾nections between different causes, as in less than 15%

6