Lecture Notes in Civil Engineering

Tetsuo Okada
Katsuyuki Suzuki
Yasumi Kawamura Editors

Practical Design
of Ships and Other
Floating Structures
Proceedings of the 14th International
Symposium, PRADS 2019,
September 22–26, 2019,
Yokohama, Japan- Volume III
Lecture Notes in Civil Engineering

Volume 65

Series Editors
Marco di Prisco, Politecnico di Milano, Milano, Italy
Sheng-Hong Chen, School of Water Resources and Hydropower Engineering,
Wuhan University, Wuhan, China
Ioannis Vayas, Institute of Steel Structures, National Technical University of
Athens, Athens, Greece
Sanjay Kumar Shukla, School of Engineering, Edith Cowan University, Joondalup,
WA, Australia
Anuj Sharma, Iowa State University, Ames, IA, USA
Nagesh Kumar, Department of Civil Engineering, Indian Institute of Science
Bangalore, Bengaluru, Karnataka, India
Chien Ming Wang, School of Civil Engineering, The University of Queensland,
Brisbane, QLD, Australia
Lecture Notes in Civil Engineering (LNCE) publishes the latest developments in
Civil Engineering - quickly, informally and in top quality. Though original research
reported in proceedings and post-proceedings represents the core of LNCE, edited
volumes of exceptionally high quality and interest may also be considered for
publication. Volumes published in LNCE embrace all aspects and subfields of, as
well as new challenges in, Civil Engineering. Topics in the series include:
• Construction and Structural Mechanics
• Building Materials
• Concrete, Steel and Timber Structures
• Geotechnical Engineering
• Earthquake Engineering
• Coastal Engineering
• Ocean and Offshore Engineering; Ships and Floating Structures
• Hydraulics, Hydrology and Water Resources Engineering
• Environmental Engineering and Sustainability
• Structural Health and Monitoring
• Surveying and Geographical Information Systems
• Indoor Environments
• Transportation and Traffic
• Risk Analysis
• Safety and Security
To submit a proposal or request further information, please contact the appropriate
Springer Editor:
– Mr. Pierpaolo Riva at [email protected] (Europe and Americas);
– Ms. Swati Meherishi at [email protected] (Asia - except China,
and Australia, New Zealand);
– Dr. Mengchu Huang at [email protected] (China).
All books in the series now indexed by Scopus and EI Compendex database!

More information about this series at http://www.springer.com/series/15087

Tetsuo Okada Katsuyuki Suzuki
• •

Yasumi Kawamura
Editors

Practical Design of Ships

and Other Floating Structures
Proceedings of the 14th International
Symposium, PRADS 2019,
September 22–26, 2019,
Yokohama, Japan- Volume III

123
Editors
Tetsuo Okada Katsuyuki Suzuki
Faculty of Engineering School of Engineering
Yokohama National University The University of Tokyo
Yokohama, Kanagawa, Japan Bunkyo-ku, Tokyo, Japan

Yasumi Kawamura
Faculty of Engineering
Yokohama National University
Yokohama, Japan

ISSN 2366-2557 ISSN 2366-2565 (electronic)

Lecture Notes in Civil Engineering
ISBN 978-981-15-4679-2 ISBN 978-981-15-4680-8 (eBook)
https://doi.org/10.1007/978-981-15-4680-8
© Springer Nature Singapore Pte Ltd. 2021
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface

The first PRADS (International Symposium on Practical Design of Ships and Other
Floating Structures) was held in Tokyo in 1977, in celebration of the 80th
anniversary of the Society of Naval Architects of Japan (later the Japan Society of
Naval Architects and Ocean Engineers), where 245 delegates from 26 countries
attended and 56 papers were presented. In the preface of the first PRADS pro-
ceedings, the chairman advocated two important principles; one is the emphasis on
the practical application to ship design, and the other is the promotion of interna-
tional mutual understanding through discussions and exchange of information on
practical problems in shipbuilding.
Since then, 42 years have passed, our circumstances and environment
experienced tremendous changes, for example, rapid development of information
technology and computer science, expansion of offshore development, and more
and more emphasis on green shipping and renewable energy. Accordingly, the
topics covered by the subsequent PRADS symposia evolved, but we believe that
the first two principles have been kept in mind throughout by all the participants
like a basso continuo.
Respecting the traditions of these past successful PRADS symposia, we are
proud to organize the 14th PRADS Symposium in Yokohama. Yokohama, located
just south of Tokyo, is the second largest city in Japan by population. The city has
been Japan’s major gateway for international transportation and communication.
Yokohama is also a center of excellence for shipbuilding and offshore engineering,
with many leading universities, research institutes, and shipping companies.
Yokohama is also a tourists’ destination with excellent accessibility from all over
the world.
More than 220 abstracts were accepted. The full papers were peer-reviewed by
two reviewers for each paper, and about 170 papers were finally selected for pre-
sentation. The topics cover hydrodynamics, ship dynamics, structures, machinery
and equipment, design and construction, navigation and logistics, and ocean
engineering.

v
vi Preface

On this opportunity, we would like to express our sincere appreciation to the

main sponsor, ClassNK, and other sponsors, Mitsui O.S.K. Lines Ltd., Mitsubishi
Shipbuilding Co., Ltd., Namura Shipbuilding Co., Ltd., Japan Marine United
Corporation, IHI Corporation, and Sumitomo Heavy Industries Marine Engineering
Co., Ltd. In addition, our gratitude is extended to the worldwide reviewers of the
full papers; without whose excellent voluntary works, our high-quality symposium
was not even made possible. Special thanks are also extended to the local orga-
nizing committee members, who devoted themselves to manage everything.
Last but not least, we would like to thank all the delegates for their participation
and invite them to enjoy the symposium as well as this exciting city.

Tetsuo Okada
Katsuyuki Suzuki
Yasumi Kawamura
Organization

Standing Committee

Seizo Motora Professor Emeritus, The University of Tokyo,

(Honorary Chairman) Japan
Yasumi Kawamura Yokohama National University, Japan
(Chairman of Standing
Committee)
Tetsuo Okada Yokohama National University, Japan
(Chairman of PRADS
2019)
Alan J. Murphy Newcastle University, UK
Bas Buchner MARIN (Maritime Research Institute
Netherlands), The Netherlands
Enrico Rizzuto University of Genoa, Italy
Ge (George) Wang Seastel Marine System (USA) LLC, USA
Ilson P. Pasqualino Federal University of Rio de Janeiro, Brazil
Patrick Kaeding University of Rostock, Germany
Quentin Derbanne Bureau Veritas Marine & Offshore, France
Seung-Hee Lee Inha University, Republic of Korea
Sverre Steen Norwegian University of Science
and Technology, Norway
Ulrik D. Nielsen Technical University of Denmark, Denmark
Xiaoming Cheng China Ship Scientific Research Center, China

Local Organizing Committee

Tetsuo Okada (Chairman) Yokohama National University, Japan
Katsuyuki Suzuki The University of Tokyo, Japan
(Vice Chairman)

vii
viii Organization

Yasumi Kawamura Yokohama National University, Japan

(Vice Chairman)
Daisuke Yanagihara Kyushu University, Japan
Kazuhiro Aoyama The University of Tokyo, Japan
Kazuhiro Iijima Osaka University, Japan
Kentaro Kobayashi Japan Society of Naval Architects and Ocean
Engineers, Japan
Koji Gotoh Kyushu University, Japan
Kunihiro Hamada Hiroshima University, Japan
Motohiko Murai Yokohama National University, Japan
Takanori Hino Yokohama National University, Japan
Tsutomu Fukui ClassNK, Japan
Yasuhira Yamada National Maritime Research Institute, Japan
Yasuyuki Toda Osaka University, Japan
Yoshiaki Hirakawa Yokohama National University, Japan
Yukitaka Yasuzawa Kyushu University, Japan

Honorary Committee
Shinjiro Mishima Japan Society of Naval Architects and Ocean
Engineers, Japan
Hiroyuki Yamato National Institute of Maritime, Port and Aviation
Technology, Japan
Takanori Kunihiro IHI Corporation, Japan
Yukito Higaki Imabari Shipbuilding Co., Ltd., Japan
Takashi Nakabe Onomichi Dockyard Co., Ltd., Japan
Yoshinori Mochida Kawasaki Heavy Industries, Ltd., Japan
Takashi Ueda Sanoyas Shipbuilding Corporation, Japan
Kotaro Chiba Japan Marine United Corporation, Japan
Tetsushi Soga Shin Kurushima Dockyard Co., Ltd., Japan
Hideshi Shimamoto Sumitomo Heavy Industries Marine
& Engineering Co., Ltd., Japan
Sachio Okumura Tsuneishi Shipbuilding Co., Ltd., Japan
Kensuke Namura Namura Shipbuilding Co., Ltd., Japan
Tetsuro Koga Mitsui E&S Shipbuilding Co., Ltd., Japan
Koji Okura Mitsubishi Shipbuilding Co., Ltd., Japan
Yasuhiko Katoh The Shipbuilders’ Association of Japan, Japan
Junichiro Ikeda Mitsui O.S.K. Lines, Ltd., Japan
Koichi Fujiwara ClassNK, Japan
Masahiko Fujikubo Osaka University, Japan
Shotaro Uto National Maritime Research Institute, Japan
Tetsuo Okada Yokohama National University, Japan
Katsuyuki Suzuki The University of Tokyo, Japan
Contents

Design and Construction

Design Methodology
A Bayesian Network Approach to Evaluate Shipboard Launch
and Recovery Performance in Early-Stage Ship Design . . . . . . . . . . . . . 3
James A. Coller and David J. Singer
Communication of Design Space Relationships Learned
by Bayesian Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Claire Wincott and Matthew Collette
Optimizing Ships Using the Holistic Accelerated Concept
Design Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Roy de Winter, Jan Furustam, Thomas Bäck, and Thijs Muller
A “Fast Track to Approval” Process for Innovative
Maritime Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Stéphane Paboeuf, Arnold de Bruijn, Franz Evegren, Matthias Krause,
and Marcel Elenbaas
A Modified Design Framework Based on Markov Decision Process
for Operational Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Hao Yuan and David J. Singer
Ship Design and Optimization
IGC Code-Based LNG Filling Limits Computation Using 3D
CAD Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Seok-Ho Byun, Woo-Sung Kil, Myeong-Jo Son, Jeong-Youl Lee,
and Mi-Ho Park
Developments of a Synthesis Model for FLNG Design . . . . . . . . . . . . . . 91
Lariuss Zago, Daniel P. Vieira, Yu K. Kam, Raul Dotta,
Rodrigo M. Amarante, and Claudio M. P. Sampaio

ix
x Contents

The First LNG Fuel Train Ferry for St. Petersburg –

Kaliningrad Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Gennadiy Egorov, Alexander Nilva, and Dmitriy Chernikov
Creation of River-Sea Cruise Passenger Vessels
of New Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
Alexander Egorov, Gennadiy Egorov, and Igor Ilnitskiy
Research on the Accessibility Evaluation of the Cruise Passage
Design Based on Passenger Movement Simulation . . . . . . . . . . . . . . . . . 137
Zhan Chen, Wei Cai, Yazhong Wang, and Yigang Wu
Genetic Algorithm Selection for Ship Concept Design . . . . . . . . . . . . . . 156
Adam Sobey, Przemyslaw Grudniewski, and Thomas Savasta
Multi Objective Design of Ships; A Pareto Procedure . . . . . . . . . . . . . . 173
Sander Calisal
Optimal Arrangement Method of a Ship Considering
the Performance Against Flooding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
Ki-Su Kim and Myung-Il Roh
Optimal Design of a Floating Raft Vibration Isolation System
with Multiple Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
Wenzhang Liu, Wenwei Wu, and Xuewen Yin
A Study of Multi-Objective Optimization for Propulsion
Performance and Cargo Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Yasuo Ichinose, Yusuke Tahara, Tomoki Takami, Azumi Kaneko,
Takayoshi Masui, and Daisuke Arai
Hull Form Optimization to Fulfil Minimum Propulsion Power
by Using Frequency and Time Domain Potential Flow Solvers . . . . . . . 220
Jaekyung Heo, Daehwan Park, Jan Henrik Berg-Jensen, Zhiyuan Pan,
and Torgeir Kirkhorn Vada
Welding and Construction
Controller Design of a Gantry Crane for the Safe Erection of Blocks
in Shipyards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Hye-Won Lee, Myung-Il Roh, Seung-Ho Ham, and Do-Hyun Chun
A Study on Using Hot-Rolled Steel Sheet for Ship Superstructure . . . . . 243
Hisao Narimatsu, Masakazu Kuwada, and Koji Gotoh
Overview of the Joint Industrial Project for Practical Application
of Laser-Arc Hybrid Welding in Construction of General Merchant
Ships in Japan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
Koji Gotoh, Takamori Uemura, Issei Uchino, Hisao Narimatsu,
Toshimitsu Maeda, Takumi Torigoe, and Atsuo Moriyama
Contents xi

Control of Car Decks Welding Deformation in RO-RO Ships . . . . . . . . 273

Junichi Toyoshima, Kiyoharu Hayashi, Mikikazu Murakami,
Masanori Sano, Shogo Imasaki, Yasuo Katayama, Hidekazu Murakawa,
Naoki Osawa, and Sherif Rashed
Big Data in Ship Design and Construction
Development of Basic Planning Support System Using
Marine Logistics Big Data and Its Application to Ship
Basic Planning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
Kunihiro Hamada, Noritaka Hirata, Kai Ihara,
Dimas Angga Fakhri Muzhoffar, and Mohammad Danil Arifin
Evaluation Method for the Maximum Wave Load Based on AIS
and Hindcast Wave Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
Masayoshi Oka, Tomoki Takami, and Chong Ma
“Monitoring Platform” of Monitoring and Visualizing System
for Shipyard: Application to Cutting and Subassembly Processes . . . . . 321
Kazuhiro Aoyama, Takayuki Yotsuzuka, Yoshiki Tanaka,
and Yoshikazu Tanabe
Monitoring Systems in Design of Ships . . . . . . . . . . . . . . . . . . . . . . . . . 338
Serena Lim, Kayvan Pazouki, and Alan J. Murphy
Performance of Ships in Ice
Ice Resistance Calculation in Pack Ice Conditions . . . . . . . . . . . . . . . . . 349
Guiyong Zhang, Biye Yang, Yuyan Jiang, and Zhe Sun
On the Development of the Arc7 LNG Carrier – Hull
Form Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
Ho-Jeong Lim, Ji-In Park, Sahng-Hyon Lee, Sung-Pyo Kim,
and Joong Kyoo Kang
A New Technique for Prediction of Ship-Ice Floes Interaction
Without Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
Donghwa Han, Joonmo Choung, and Kuk-Jin Kang
Study on Estimation of Ice Resistance and Attainable Speed for Ship
of Arbitrary Shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
Hyun-Soo Kim, Donghwa Han, and Ali Erinc Ozden
Estimation of Ice Conditions Along the Northern Sea Route . . . . . . . . . 397
Wei Chai, Bernt J. Leira, Chana Sinsabvarodom, and Wei Shi
Effect of Ship Speed on Bow Impact Load Estimation for Polar
Class Ship Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
Mahmud Sazidy, Claude Daley, Bruce Colbourne, and John MacKay
xii Contents

Navigation and Logistics

Safety vs. Sustainability – How Much Underpowering of Ships
Is Acceptable? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423
Stefan Krüger, Michal Josten, and Leonidas Souflis
Sustainability Assessment of Small Scale Fishing Vessel Operations:
A Case Study in Palabuhanratu, Indonesia . . . . . . . . . . . . . . . . . . . . . . 436
Vita R. Kurniawati, Richard W. Birmingham, and Alan J. Murphy
Assessing Port Congestion Using Ship Movement Data:
A Case Study of Tianjin Port . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
Guanlan Liu, Alessio Tei, and Alan J. Murphy
Numerical Simulation of Hybrid Platform Supply Vessel (PSV)
Fuel Consumption for the Pre-Salt Layer in Brazil . . . . . . . . . . . . . . . . 478
Kazuo Nishimoto, Claudio M. P. Sampaio, Rodrigo J. Vale,
and Felipe Ruggeri
Optimization of River-Sea-Going Ship Type Scheme and Operation
Strategy Under Emission Control Area in China . . . . . . . . . . . . . . . . . . 498
Xinjing Tan, Wei Cai, and Chao Liu
Estimation of the Collision Risk on Planned Route . . . . . . . . . . . . . . . . 516
Makiko Minami and Ruri Shoji
Autonomous Route Planning for Ships in Complex Waterways . . . . . . . 533
Jinsong Xu, Chunxiao Hou, Rongwu Yang, and Ge (George) Wang

Ocean Engineering
Hydrodynamic Analysis of Offshore Aquaculture Platform and Its
Mooring Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 551
Jun Yu, Xiaoming Cheng, and Xiaofeng Wu
Prediction of Extreme Nonlinear Hydrodynamic Responses
and Mooring Line Loads of Floating Offshore Structures . . . . . . . . . . . 564
Dong-Hyun Lim, Yonghwan Kim, and Seung-Hoon Lee
Design Automation of Mooring Systems for Floating Structures . . . . . . 579
Bo Wu, Xiaoming Cheng, Ying Chen, Xinyun Ni, and Kai Zhang
A Study of Riser Oscillations Caused by Slug Flows During Subsea
Petroleum Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 595
Sergio Nascimento Bordalo and Celso Kazuyuki Morooka
A Simplified Nonlinear Analysis Procedure for the Flexible Risers
in Subjected to Combined Loading Effects . . . . . . . . . . . . . . . . . . . . . . . 614
Jeong Du Kim and Beom-Seon Jang
Contents xiii

Modal Parameter Identification of an Electrical Submersible Pump

Installed in a Test Well Using Drop Tests . . . . . . . . . . . . . . . . . . . . . . . 631
Ulisses A. Monteiro, Ricardo S. Minette, Ricardo H. R. Gutiérrez,
and Luiz A. Vaz
Fatigue Analysis of Vortex-Induced Vibration for Marine Risers
with Top-End Platform Motion Excitations . . . . . . . . . . . . . . . . . . . . . . 639
Yuchao Yuan, Hongxiang Xue, Wenyong Tang, and Jun Liu
Real Time Estimation of Local Wave Characteristics from Ship
Motions Using Artificial Neural Networks . . . . . . . . . . . . . . . . . . . . . . . 657
Bulent Duz, Bart Mak, Remco Hageman, and Nicola Grasso
Nonlinear Wave Loads on Deep-Water Semisubmersibles . . . . . . . . . . . 679
Rahul Manohar, Xin Wang, Velizar Prampakov, and Arun Dev
Experimental and Numerical Comparisons of a Floating Absorber
Wave Energy Converter in Regular Waves . . . . . . . . . . . . . . . . . . . . . . 697
A. Novás-Cortés, L. Santiago-Caamaño, M. Miguez-González,
and V. Díaz-Casas
Study on Motion Response of a Floating Offshore Turbines’ Unit
with OWC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707
Motohiko Murai and Ryohei Mochiduki
Development and Design of a Floating Type Ocean Current
Turbine System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 732
Yasushi Dodo, Shigeki Nagaya, Tetsuo Okada, Makoto Toyoda,
and Akio Ito

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 757

Design and Construction
 A Bayesian Network Approach to
Evaluate Shipboard Launch and Recovery
Performance in Early-Stage Ship Design

James A. Coller(B) and David J. Singer

Department of Naval Architecture and Marine Engineering,

University of Michigan, Ann Arbor, MI, USA
[email protected]
https://ancr.engin.umich.edu/

Abstract. Vessel performance in shipboard launch and recovery oper-

ations is a critical element of its overall mission effectiveness. Both com-
mercial and military vessels are required to launch and recover a vari-
ety of manned and unmanned platforms including submersibles, small
boats, aircraft, and amphibious vehicles. A fundamental attribute of
these launch and recovery operations is their dependence on a number
of probabilistic factors such as environmental conditions and design fac-
tors associated with the vessel and the platforms. To facilitate effective
launch and recovery operations, ship designers require a design method
that takes into account these probabilistic factors to gain a better under-
standing the implications of launch and recovery on a variety of early
stage design decisions. Recent advancements in design tools and method-
ologies have pushed critical decisions earlier in the design process which
necessitates evaluating mission critical systems sooner as well.
This paper presents an application of Bayesian networks for evaluating
the performance of shipboard launch and recovery operations. Bayesian
networks provide a framework for understanding system interdependen-
cies by modeling and analyzing conditional probabilities between vari-
able pairs. In previous ship design research, Bayesian networks have been
applied to risk modeling and the examination of principal particulars.
This research utilizes Bayesian networks to model the environmen-
tal and design factors that contribute to the performance of launch and
recovery operations, using a level of information that is consistent with
early-stage ship design. These networks are analyzed and examined to
assess the impact of the overall design attributes on those factors. A ship
design case study is presented to demonstrate the method and elucidate
the driving interdependencies in launch and recovery operations.

Keywords: Bayesian networks · Ship design · Launch and recovery ·

Design tools · Unmanned vehicles · Causal dependency

c Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 3–23, 2021.
https://doi.org/10.1007/978-981-15-4680-8_1
4 J. A. Coller and D. J. Singer

1 Introduction
1.1 Problem and Approach

Advancements in naval architecture continue to push the boundaries of how

much of the design space can be explored for a given ship within time and
budgetary constraints. With methodologies that require large portions of the
design space to be investigated, such as set based design (SBD), [1] the design
space needs to be explored in a computationally efficient manner while providing
novel information that enables decision making with limited information.
Early stage design requires efficient design tools to effectively examine a given
design and evaluate the design performance [2]. These tools are used to evaluate
the potential ship designs and ensure that the key design requirements are met.
While these tools have inherent biases and will have an impact on the decision
making process [3], they remain necessary in modern ship design.
Naval ships are often tasked with the launch and recovery of additional
systems, both manned and unmanned, including boats, helicopters, unmanned
underwater vehicles (UUVs), unmanned aerial vehicles (UAVs), and unmanned
surface vehicles (USVs). The design of the ship can have a large impact on the
effectiveness of these operations, in addition to environmental conditions and
operating procedures. Design features such as well decks, stern ramps, davits,
cranes, and helicopter pads can be complicated to integrate and evaluate.
As the use of unmanned vehicles increases, these operations will continue to
increase in complexity. The US Navy has stated it plans to continue rolling out
unmanned vehicles to integrate with the fleet, across the skies [4], the surface
[5], and subsea [6]. Unmanned and autonomous vehicles can be more difficult
to launch and recover without human operator guidance and with an increased
number of systems that must flawlessly execute.
This paper investigates a novel approach to this problem by using a Bayesian
network to evaluate the environmental and ship design factors for a shipboard
launch and recovery system. The Bayesian Network will allow the designer to
evaluate the interdependencies of the design decision and evaluate the expected
satisfaction with the overall design. The network structure allows the designer
to simulate satisfaction for a wide range of vehicles and complex systems.
The goal of this paper is not to prove the specific network structure is correct,
but rather to show that the use of a Bayesian framework can be beneficial to early
stage designers. The specific network definition is dependent upon an individual
design.
In the remainder of the paper, an introduction to launch and recovery sys-
tems and Bayesian networks will be presented. The methodology is explained,
and a case study is presented to highlight the network structure and examine
variable sensitivity. Additional opportunities for further work and implications
are discussed.
 Bayesian Networks in Ship Design 5

1.2 Launch and Recovery Design Space

In naval ship design, many mission profiles include launching and recovering any
number of other vehicles from the ship platform. A given design may be tasked
with having several launch and recovery areas such as well decks, flight decks,
and davits [7,8]. Multi-mission ships are tasked with flexible spaces that can
serve a wide variety of mission capabilities depending on the call to service and
the changing geo-political climate [9,10].
Within this design space, the possible vehicles to be launched or recovered are
numerous. Categorically, they include aerial, surface, and subsea vehicles. Within
these categories, dozens to thousands of variants exist. Figures 1 and 2 highlight
some of the variety a single ship may see. Additional unique missions may include
recovering non-propulsive vehicles such as a NASA landing module [11].

Fig. 1. Sailors launch the Rigid Hull Fig. 2. A landing craft, air cushion
Inflatable Boat (RHIB) aboard the (LCAC) is launched from the well
guided-missile destroyer USS Russell deck of the USS Boxer (LHD 4).
(DDG 59). US Navy Released Photo, US Navy Released Photo, 190116-N-
060619-N-4166B-009. SH168-1054.

The launch and recovery spaces of a ship can have large impacts on the overall
ship design. Some launch and recovery systems will dictate large portions of the
ship arrangements and will be primary design drivers such as flight decks, well
decks, and stern ramps. Other systems, such as davits and articulating cranes,
will have minimal impact on the overall design. This makes the analysis of the
design alternatives an important aspect of the overall design process.

1.3 Bayesian Networks

Bayesian networks are a type of directed acyclic graph (DAG) encoded with
conditional probability distributions. The network allows the expression of rela-
tionships which can result in simplified knowledge generation and transparency
to relationship dependencies. Bayesian networks allow for both top-down and
bottom-up reasoning, which means that the outcome can be predicted, given
a set of inputs, or given an outcome, the root cause of the outcome can be
identified [12].
6 J. A. Coller and D. J. Singer

Bayesian networks are directed graphs, meaning they show the direction-
ality of relationships. Bayesian networks rely on the use of Baye’s Theorem for
Bayesian inference, shown in Eq. 1. The posterior distribution, P (θ|y), represents
the probability that is the basis for action, given collected knowledge, y. This
utilizes information from a prior distribution, P (θ), and the likelihood function,
P (y|θ).
P (y|θ)P (θ)
P (θ|y) = (1)
P (y)
The marginal distribution, P (y), is a normalizing constant that is often disre-
garded, as it can be difficult to determine [13], and the result is that in Bayesian
inference we may only show the proportionality of the posterior to the likelihood
and prior, shown in Eq. 2.

P (θ|y) ∝ P (y|θ)P (θ) (2)

Bayesian networks rely on joint probability distributions to combine parent

nodes for the child node. In this probabilistic model, each variable, Xi is repre-
sented as a node on the DAG. The probability distribution P (X1 , X2 , ..., Xn ) is
represented as follows:

n
P (X1 , X2 , ..., Xn ) = P (Xi |ΠXi ) (3)
i=1

where ΠXi represents the set of nodes that are the parents to Xi in the network
topology.
Bayesian networks are widely used in engineering and defense research.
Bayesian networks have been used to predict drinking water distribution pipe
breaks [14]. They have also been used for probabilistic assessments for nuclear
waste disposal [15], and predicting air defense threats [16]. In marine structural
engineering, they have been used in structural performance and reliability mod-
els [17]. There have been some limited applications in marine design [18], and in
marine decision making [19].
In some of these cases, the network topology was learned from the data [14],
while in others it was determined by the authors [16,17].

2 Methodology
The Bayesian framework examined in this paper can be adapted to any network
structure. The network structure should be set based on the specific design sce-
nario and structure of a given ship design project by the designers. The network
should be representative of the ship design and factors under consideration.
For the purposes of examining the Bayesian network framework, we will
use an existing sample network. The network structure for this methodology is
adapted from an influence diagram created by Wireman, 2019 [20]. This model
is intended to show the ability of Bayesian Networks to examine a ship design,
 Bayesian Networks in Ship Design 7

and does not necessarily try to perfectly model individual variables such as crane
operating capacities.
In the network, the top level nodes, represented by squares, are the primary
input variables. The middle level nodes, represented by circles, are intermediate
variables that are of importance and are dependent upon the input nodes. Lastly,
the output nodes, represented by diamonds and a hexagon, are calculated metrics
that give probabilistic outputs based on the input and intermediate nodes. In
the structure, the directed edges between nodes represent the dependencies.
The top level and intermediate nodes are established by the design team.
These nodes should represent the high level inputs under consideration and the
functional dependencies that map from them. These nodes will change between
ships, but this current format gives one example of how this can be done.
The previous influence diagram structure was slightly modified. The variables
included are described further in Tables 1 and 2. The full network structure is
shown in Fig. 3, with the four primary subtrees shown separately in Figs. 4, 5, 6
and 7.
All modeling was performed in Python 2.7. All graphing of the networks was
performed using the NetworkX, [21], and Graphviz, [22], packages for Python.

Fig. 3. The full network tree developed. Closer views of the four primary subtrees are
shown in Figs. 4, 5, 6, and 7.
8 J. A. Coller and D. J. Singer

The model takes in 14 primary input variables, listed in Table 1, which are the
first nodes below each of the four variable categories. From here, the input nodes
are passed to the first layer of calculated intermediate nodes. These intermediate
nodes utilize joint conditional probability tables to determine the outcome of the
input node. These joint conditional probability tables are shown in Tables 3, 4,
5, 6 and 7.
In instances where the input variable is a rating, Table 3 is used to determine
the probability that the outcome will be satisfactory. Table 4 is used for vari-
ables where the weight of the launch or recovery vehicle is dependent. Table 5 is
for variables that are dependent on the vehicle volume, Table 6 is for variables
dependent on the ship freeboard, and Table 7 is used for variables dependent on
the crew size.
In this method, P (θ) is the probability of a satisfactory design outcome, given
observations yi . There are several intermediate satisfaction metrics, including
Freeboard Satisfaction, θf , Arrangement Flexibility Satisfaction, θa , Manning
Satisfaction, θm , Performance Satisfaction, θp , and Vehicle Interoperability Sat-
isfaction, θi .
The probability of a satisfactory design outcome was the metric chosen as
the key performance indicator to evaluate the overall design effectiveness from
the network. This metric could have been any number of things including cost
effectiveness or mission effectiveness. This metric was chosen for its simplicity.
There are multiple intermediate variables which contribute to the knowledge
base. In most cases, these are calculated as:


N
P (θchild ) = P (θparent,i |yi ) (4)
i=1

where the child nodes are the dependent intermediate nodes, and the N parent
nodes are the nodes above them in the hierarchy which have edges leading to
the child signifying dependence. The parent probabilities are calculated using
the conditional probability tables previously described.
In the case of the probabilistic estimate of the actual capacity of the launch
and recovery system, Ca , it is calculated as follows:

Ca = Cr − 0.5I (5)

where Cr is the rated capacity, and I is the impact of the other dependent nodes,
which is calculated as:

I = Cr (1 − P (θ|ywind )P (θ|yss )) (6)

where P (θ|ywind ) and P (θ|yss ) are the conditional probabilities from the impact
of the wind and the sea state, respectively.
 Bayesian Networks in Ship Design 9

Table 1. Variables included in the Bayesian network.

Item Description
Host vessel The ship that the secondary vehicle is being
launched from or recovered onto
Arrangement flexibility* Rating of if the ship design’s arrangement is flexible
Crew size* Total number of crew members on the host vessel
Minimum freeboard* Minimum freeboard required for the ship, in meters
Available manning Probability there will be enough crew to man the
launch/recovery
Vehicle The vehicle being launched or recovered
Vehicle volume* The volume of the vehicle, in cubic meters
Vehicle weight* The weight of the vehicle, in metric tons
Variability of vehicle type* Rating of the variability of other vehicles of the same
type
L/R flexibility* Rating of the flexibility the vehicle has for multiple
launch/recovery methods
Vehicle complexity* Rating of how complex the vehicle is to
launch/recover
L/R system The launch or recovery system
System requirements* Rating of the overall requirements of the
launch/recovery system
Arrangement impact Probability that the system does not heavily impact
the ship design’s arrangements
Required manning Probability that the manning requirement of the
system can be satisfied
Freeboard flexibility* Rating of the flexibility of the system to deal with
various freeboard amounts
Relative height* Rating of how close to the water level or top of the
ship the system is
Simplicity of L/R Probability that the system will be simple to execute
Rated capacity* The rated capacity of the system, in metric tons
Actual capacity A probabilistically determined capacity of the system
Environment The environmental conditions
Sea state rating* Rating of the sea state condition and impact on the
launch and recovery
Wind speeds* Rating of the wind conditions and impact on the
launch and recovery
Freeboard available Probability of environmental impacts on the
freeboard
* Denotes input variables to the network
10 J. A. Coller and D. J. Singer

Table 2. The satisfaction variables included in the Bayesian network.

Item Description
Arrangement satisfaction Probability that the arrangement of the ship is
satisfactory
Manning satisfaction Probability that the manning levels are satisfactory
Freeboard satisfaction Probability that the freeboard levels will be
satisfactory
Performance satisfaction Probability that the performance of the
launch/recovery will be satisfactory
Interoperability satisfaction Probability that the interoperability of multiple
vehicles will be satisfactory
Overall satisfaction Overall probability of satisfaction

Table 3. The conditional proba- Table 4. The conditional proba-

bility table for the rated variables. bility table for the mass variables.

Rating P (Satisfactory) Mass (mt) P (Satisfactory)

1 0.99 1 0.99
2 0.98 5 0.98
3 0.95 10 0.95
4 0.80 20 0.90
5 0.50 30 0.80
6 0.20 40 0.70
7 0.10 50 0.60
8 0.05 100 0.40
9 0.01 200 0.30

In the case of the Performance Satisfaction, the weight of the vehicle needs
to be considered compared to the actual capacity of the launch and recovery
system. In this case, the probability of satisfaction is calculated as follows:

0.01, if W > Ca
P (θp ) = (7)
P (θ|ycmplx )((1 − W/Ca ) + 0.25), otherwise

where W is the weight of the vehicle, P (θ|ycmplx ) is the probability of satisfaction

given the complexity of the vehicle, and ((1 − W/Ca ) + 0.25) is capped at 1.0.
The overall satisfaction, P (θ), is computed as the geometric mean of
P (θf,a,m,p,i ):
P (θ) = [P (θf )P (θa )P (θm )P (θp )P (θi )]1/5 (8)
 Bayesian Networks in Ship Design 11

Table 5. The conditional proba- Table 6. The conditional proba-

bility table for the volumetric vari- bility table for the freeboard vari-
ables. ables.

Volume (m3 ) P (Satisfactory) Fbd Min (m) P (Satisfactory)

0.1 0.99 0.5 0.99
1.0 0.95 1.0 0.95
5.0 0.85 2.0 0.85
10.0 0.75 3.0 0.75
20.0 0.55 4.0 0.65
30.0 0.40 5.0 0.55
40.0 0.30 6.0 0.45
50.0 0.22 7.0 0.35
75.0 0.10 8.0 0.25
100.0 0.05 9.0 0.15
150.0 0.01 10.0 0.05

Table 7. The conditional probability table for the crew-based variables.

Crew size P (Satisfactory)

5 0.01
10 0.08
15 0.15
20 0.20
30 0.30
40 0.40
50 0.50
60 0.60
80 0.72
100 0.80
150 0.90
200 0.95
400 0.99
12 J. A. Coller and D. J. Singer

Fig. 4. The Host Vessel portion of the network is shown.

Fig. 5. The Vehicle portion of the network is shown.

 Bayesian Networks in Ship Design 13

Fig. 6. The Launch and Recovery System portion of the network is shown.

Fig. 7. The Environmental portion of the network is shown.

14 J. A. Coller and D. J. Singer

3 Model Sensitivity
To examine the effectiveness of the proposed model, a simple sensitivity analysis
is examined. A sample hose vessel, vehicle, launch and recovery system, and envi-
ronmental conditions are proposed. From this baseline, each variable is changed
independent of the rest between the minimum and maximum values assessed in
the conditional probability tables. The baseline input variables, as well as the
maximum and minimum values, are shown in Table 8.

Table 8. The input variables for the case study. The baseline value is shown along
with the minimum and maximum values used in the simulation.

Input variable Value

Baseline Min Max
Host ship Arrangement flexibility 3 1 9
Crew size 200 5 400
Freeboard minimum 1 1 9
Vehicle Volume [m3 ] 15 0.1 150
Weight [mt] 20 0.5 200
Variability 3 1 9
Flexibility 3 1 9
Complexity 2 1 9
L/R system System requirements 2 1 9
Freeboard flexibility 2 1 9
Relative height 2 1 9
Capacity [mt] 40 0.5 300
Environment Sea state 2 1 9
Wind 2 1 9

The baseline variable selection results in an overall probability of satisfaction

of 0.76. This is reflected in Fig. 8 and Table 9. As shown, under the baseline

Table 9. The results of the baseline variables through the network.

Variable Value
P (θ) 0.76
P (θf ) 0.91
P (θa ) 0.69
P (θm ) 0.63
P (θp ) 0.69
P (θi ) 0.95
 Bayesian Networks in Ship Design 15

conditions the overall satisfaction is being driven positively by the vehicle inter-
operability and is being driven negatively by the manning requirement.

Fig. 8. The baseline performance data from the initial simulation. The overall sat-
isfaction is 0.76, which is largely driven down by the poor probability of manning
satisfaction, arrangement flexibility, and performance.

Using this baseline data as a starting point, 134 test cases are simulated by
varying each input variable independently. From this data, the most useful data
was determined to be the maximum, minimum, and the difference between the
maximum and minimum values for each of the satisfaction probabilities. This
data is shown in Table 10 and Table 11.
We can validate the model by examining the edge cases. Conceptually, if
the sea state is extremely large, the probability of a positive outcome with high
satisfaction is low.
This is accurately depicted in Table 10, where min(P (θ)) = 0.02 corresponds
to the worst sea state condition. Similarly, not having enough crew to perform
the task, having extreme system requirements, having a very complex vehicle,
or having a heavy vehicle with a medium capacity launch/recovery system all
result in low probabilities of satisfaction. This makes sense given our intuition.
Conversely, we can see improvements over the baseline across the board as
we loosen restrictions on the maximum value side of Table 10. Many variables
were already near their best case scenarios in the baseline, so only minimal gains
were observed. However in the cases of the variables where the baseline was of
moderate difficulty (vehicle weight, launch system capacity), larger changes were
observed, as expected. In no case do we expect a perfect situation, given the rest
of the baseline variables, which is why the overall satisfaction is never expected
above 0.89.
16 J. A. Coller and D. J. Singer

Table 10. The minimum and maximum results from the simulation.

Input variable Minimum output Maximum output

P (θ) P (θf ) P (θa ) P (θm ) P (θp ) P (θi ) P (θ) P (θf ) P (θa ) P (θm ) P (θp ) P (θi )
Arr flexibility 0.31 0.91 0.01 0.63 0.69 0.95 0.77 0.91 0.72 0.63 0.69 0.95
Crew size 0.31 0.91 0.69 0.01 0.69 0.95 0.77 0.91 0.69 0.66 0.69 0.95
Fbd minimum 0.53 0.14 0.69 0.63 0.69 0.95 0.76 0.91 0.69 0.63 0.69 0.95
Volume 0.75 0.91 0.61 0.63 0.69 0.95 0.78 0.91 0.76 0.63 0.69 0.95
Weight 0.26 0.91 0.69 0.21 0.01 0.95 0.83 0.91 0.69 0.69 0.92 0.95
Variability 0.56 0.91 0.69 0.63 0.69 0.20 0.77 0.91 0.69 0.63 0.69 0.98
Flexibility 0.56 0.91 0.69 0.63 0.69 0.20 0.77 0.91 0.69 0.63 0.69 0.98
Complexity 0.31 0.91 0.69 0.63 0.01 0.95 0.77 0.91 0.69 0.63 0.70 0.95
Sys requirements 0.12 0.91 0.01 0.01 0.69 0.95 0.85 0.91 0.94 0.78 0.69 0.95
Fbd flexibility 0.34 0.01 0.92 0.77 0.69 0.95 0.85 0.92 0.92 0.77 0.69 0.95
Relative height 0.34 0.91 0.92 0.77 0.01 0.95 0.85 0.91 0.92 0.77 0.70 0.95
Capacity 0.36 0.91 0.92 0.77 0.01 0.95 0.89 0.91 0.92 0.77 0.92 0.95
Sea state 0.02 0.01 0.69 0.00 0.01 0.95 0.77 0.92 0.69 0.64 0.70 0.95
Wind 0.05 0.91 0.69 0.00 0.01 0.95 0.77 0.91 0.69 0.64 0.70 0.95

Table 11. The difference between the minimum and maximum results of the simula-
tion. A zero indicates that the input variable has no influence on the output.

Input variable Max-Min

P (θ) P (θf ) P (θa ) P (θm ) P (θp ) P (θi )
Host ship Arrangement flexibility 0.46 0.00 0.71 0.00 0.00 0.00
Crew size 0.46 0.00 0.00 0.65 0.00 0.00
Freeboard minimum 0.24 0.77 0.00 0.00 0.00 0.00
Vehicle Volume 0.03 0.00 0.15 0.00 0.00 0.00
Weight 0.56 0.00 0.00 0.48 0.91 0.00
Variability 0.21 0.00 0.00 0.00 0.00 0.78
Flexibility 0.21 0.00 0.00 0.00 0.00 0.78
Complexity 0.46 0.00 0.00 0.00 0.69 0.00
L/R system System requirements 0.73 0.00 0.93 0.77 0.00 0.00
Freeboard flexibility 0.51 0.91 0.00 0.00 0.00 0.00
Relative height 0.51 0.00 0.00 0.00 0.69 0.00
Capacity 0.53 0.00 0.00 0.00 0.91 0.00
Environment Sea state 0.75 0.91 0.00 0.64 0.69 0.00
Wind 0.72 0.00 0.00 0.64 0.69 0.00

These data allow the identification of the key variables in the design by
examining the differences between their minimum and maximum values across
the various metrics, shown in Table 11. A result of zero indicates that the variable
has no influence on the predicted satisfaction. A low number indicates it has low
impact, and a high number indicates that it has a large impact. In this case,
the data show that the environment has the largest impact on the overall design
satisfaction outcome.
 Bayesian Networks in Ship Design 17

The sea state influences more aspects of the satisfaction outcome than any
other element. This is due to the sea state’s influence across the board on oper-
ations. As the sea state intensifies, ship motions increase and situations can
potentially become more dangerous. The safe working load for cranes can be
diminished and higher manpower levels can be required for shipboard operations.
While individual ships set manning requirements for given ship operations[23],
these can often be increased in heavier sea conditions.
The volume of the vehicle on the other hand, influences the outcome satis-
faction least, through its impact on the arrangement flexibility. While moving
a small man-portable UUV can be simple and have no impact on the arrange-
ment flexibility, moving a large helicopter will have a larger impact. However,
this larger vehicle will also require a larger launch and recovery system, which is
being accounted for in the system requirements side of the arrangement impacts
as well. The influence paths of these variables are highlighted in Fig. 9.

Fig. 9. The impact of the sea state is (blue path) is high due to its high degree of
connectivity. The vehicle volume (yellow path) has very little impact due to its linear
path.

The largest impact variable, the sea state, is examined in Fig. 10 and Fig. 11.
This shows that the overall output function changes with the sea state’s con-
ditional probability function, as expected. They do not match exactly because
other variables are still influencing the outcome. We can also observe in the
18 J. A. Coller and D. J. Singer

data that the sea state does not influence the vehicle interoperability or the
arrangement flexibility satisfaction levels since they do not share conditional
dependencies, also observable in Fig. 9.

Fig. 10. The sea state’s conditional Fig. 11. The sea state does not effect
probability function impacts the over- the vehicle interoperability or the
all satisfaction due to its heavy influ- arrangement flexibility satisfaction lev-
ence in the design satisfaction across els since they do not share conditional
multiple paths. dependencies.

4 Case Study

In order to examine potential uses of Bayesian networks for early stage ship
design, a simple case study is examined.
In this case study, a ship design has been selected. The ship has an arrange-
ment flexibility score of 3 out of 9, a 100 person crew, and 1 m of minimum free-
board. The design will integrate one of two crane systems shown in Table 12. The
two cranes differ in their requirements and capacities. The design is required to
integrate three possible vehicles for launch and recovery. The parameters of these
three vehicles are listed in Table 13. The three vehicles are a small unmanned
underwater vehicle (UUV), a medium duty rigid hull inflatable boat (RHIB) and
a heavy duty RHIB.
In order to quantify the differences between the two launch and recovery sys-
tems, the network is simulated for both launch and recovery systems across the
three vehicles in nine environmental conditions. In each environmental condition,
the sea state and wind rating match one another.
The results of the two launch systems are shown in Figs. 12 and 13. As
expected from the results of the sensitivity analysis, the overall trend follows
that of the environmental condition conditional probability distribution. Of note,
this analysis allows the designer to observe that despite differences, the small
UUV and medium duty RHIB have an almost identical expected satisfaction
probability. The small UUV only has an approximately 2% advantage over the
 Bayesian Networks in Ship Design 19

Table 12. The variable input values for the two crane systems considered for the case
study.

Variable System 1 System 2

System requirements 2 4
Freeboard flexibility 2 4
Relative height 2 4
Rated capacity [mt] 20 40

Table 13. The variable input values for the three vehicles considered for the case study.

Variable Small UUV Medium RHIB Heavy RHIB

3
Volume [m ] 0.5 10 15
Weight [mt] 0.1 5 10
Variability 2 4 4
Flexibility 2 3 5
Complexity 4 2 3

Fig. 12. The results of the case study Fig. 13. The results of the case study
simulation for crane system 1. The simulation for crane system 2. The
Small UUV and Medium RHIB track Small UUV and Medium RHIB track
almost identically, while the Heavy almost identically, while the Heavy
RHIB has approximately a 10–13% RHIB has approximately a 10–13%
lower predicted performance satisfac- lower predicted performance satisfac-
tion for Sea State Ratings of 1–5. tion for Sea State Ratings of 1–5.
20 J. A. Coller and D. J. Singer

Fig. 14. The predicted performance for crane system 1 and crane system 2 are com-
pared for the medium duty RHIB. The difference in performance is also plotted. System
1 has a higher predicted performance satisfaction, with a 13% difference in lower sea
state ratings.

medium duty RHIB. The heavy duty RHIB’s expected performance satisfaction
is approximately 15% lower than the medium duty RHIB.
In Fig. 14, the medium duty RHIB case for each launch system is compared,
with the difference between them plotted. System 1 has a higher predicted per-
formance satisfaction, with a 13% difference in lower sea state ratings. As the
sea state rating increases, the expected performance satisfaction for both trends
toward zero as the likelihood of a successful launch and recovery goes to zero
due to the heavy seas.
By using the Bayesian network, the designer can quantify the differences
between potential design choices. A key advantage of the Bayesian network is
that the designer can also identify the design drivers of these differences. By
using Table 11 and the differences between the two systems, we can identify the
overall system requirements as the driving factor to differentiate the two systems.
The Bayesian network allows us to understand that amongst the numerous inter-
dependencies, the system requirements vary the most between the systems and
has the potential to have a large impact.
Further, in investigating the differences between the vehicles, we can identify
the vehicle weight as the most likely factor driving the expected probability of
design satisfaction since it impacts a large portion of the tree.
Overall, this method has allowed the designer to quantify the design dif-
ferences and visualize and recognize the design factors that are driving these
differences.
 Bayesian Networks in Ship Design 21

5 Analysis
This model is extremely simplified, but it shows that a Bayesian network struc-
ture can be effective in a ship design case. This model was intended to show
that Bayesian networks can be used in quantifying uncertainty in early stage
design, and it has shown that. The model outcome is heavily dependent upon
both the conditional probability tables as well as the network structure, so it is
no surprise that in this simulated model the outcomes were predicable given our
knowledge of the underlying assumptions.
In this case, no real data was available to train a network structure and create
the conditional probability tables through learning algorithms. Algorithms exist
that will allow the network structure to be built up from provided input data,
based on observed dependencies. Francis, 2014 [14] is an example of this.
One major problem with a ship design Bayesian network model is that the
variables are all very dependant upon one another. Ship design is a complex
problem, sometimes referred to as a “wicked problem” [24]. The ship design,
vehicle, launch and recovery system, and environmental impacts are not readily
decomposed. Because of the lack of decomposition one needs to be cautious that
the structure of the network does not lead to interdependent outcomes that are
heavily impacted on single variable changes.
The advantage of this structure is that it can allow these interdependencies
to be visualized and quantified. While a designer may realize that the choice in
locating a davit on deck 3 versus deck 5 may have an impact on the launch and
recovery operation for a RHIB, by examining those decision trade-offs in a prop-
erly calibrated Bayesian network, the trade space can be quantified. Additional
interdependencies that may not have been immediately realized can be quickly
observed and quantified.

6 Conclusion

Identifying and quantifying the impact of design choices within a ship design
program is critical for the success of the design. Simple early stage design tools
with minimal computational cost are critical in order to evaluate these design
decisions.
In this paper, a simple Bayesian network structure was developed to evaluate
the performance of a shipboard launch and recovery operation. A sensitivity
analysis of the network structure, which is novel within the Bayesian network
literature, and a case study, were developed which provided insight into both
the network structure and design decision making has been presented.
Future work in this area can extend the Bayesian network to other use cases
in ship design such as examining a general arrangement layout, integrating ship
fleets, or modeling casualty response in damage conditions. Other research work
is also being performed to examine the use of Bayesian networks in design opti-
mization. An extension of this work can also incorporate uncertainty in design to
understand the impact of the uncertainty in decision making. Additional factors
22 J. A. Coller and D. J. Singer

can be included in future analysis, including cost, which has a large impact on
ship acquisition.
Additional future work includes validation of the structure. In this frame-
work, the network structure should be validated independently to ensure that it
is representative of the ship design under consideration. The Bayesian framework
has been generalized so that it may be adapted to any network framework and
is independent of the network validation.
The Bayesian network allows for increased transparency and knowledge gen-
eration in the design process. Interdependencies can be visualized and quantified
through probabilistic methods.
While limited, the results of this study justify further research in the area of
modeling uncertainty and the decision making process.

References
1. Singer, D.J., Doerry, N., Buckley, M.E.: What is set-based design? Naval Eng. J.
121(4), 31–43 (2009). https://doi.org/10.1111/j.1559-3584.2009.00226.x
2. Whitcomb, C.A.: Naval ship design philosophy implementation. Naval Eng. J.
110(1), 49–63 (1998). https://doi.org/10.1111/j.1559-3584.1998.tb02385.x
3. Sypniewski, M.J., Singer, D.J.: A Framework for Assessing the Effects of Bias in
Early-Stage Ship Design. Royal Institute of Naval Architects, Singapore (2017)
4. Gertler, J.: U.S. unmanned aerial systems. Technical report R42136, Congressional
Research Service (2012)
5. LaGrone, S.: Navy Wants 10-Ship Unmanned ‘Ghost Fleet’ to Supplement
Manned Force (2019). https://news.usni.org/2019/03/13/navy-wants-ten-ship-
3b-unmanned-experimental-ghost-fleet
6. United States Navy: The Navy Unmanned Undersea Vehicle (UUV) Master Plan
(2004)
7. United States Navy: The US Navy – Fact File: Amphibious Assault
Ships - LHD/LHA(R) (2019). https://www.navy.mil/navydata/fact display.asp?
cid=4200&tid=400&ct=4
8. O’Rourke, R.: Navy LPD-17 Flight II (LX[R]) amphibious ship program: back-
ground and issues for congress. Technical report R43543, Congressional Research
Service (2018)
9. O’Rourke, R.: Coast guard polar security cutter (polar icebreaker) program: back-
ground and issues for congress. Technical report RL34391, Congressional Research
Service (2019)
10. O’Rourke, R.: Navy littoral combat ship (LCS) program: background and issues
for congress. Technical report RL33741, Congressional Research Service (2018)
11. The Maritime Executive: U.S. Navy and NASA Test Space Module Recov-
ery (2018). https://www.maritime-executive.com/features/u-s-navy-and-nasa-
test-space-module-recovery
12. Murphy, K.: A Brief Introduction to Graphical Models and Bayesian Networks
(1998)
13. Gelman, A., Carlin, J.B., Stern, H.S., Dunson, D.B., Vehtari, A., Rubin, D.B.:
Bayesian Data Analysis, 3rd edn. CRC Press, Boca Raton (2013)
14. Francis, R.A., Guikema, S.D., Henneman, L.: Bayesian belief networks for predict-
ing drinking water distribution system pipe breaks. Reliab. Eng. Syst. Saf. 130,
1–11 (2014). https://doi.org/10.1016/j.ress.2014.04.024
 Bayesian Networks in Ship Design 23

15. Lee, C.J., Lee, K.J.: Application of Bayesian network to the probabilistic risk
assessment of nuclear waste disposal. Reliab. Eng. Syst. Saf. 91(5), 515–532 (2006).
https://doi.org/10.1016/j.ress.2005.03.011
16. Johansson, F., Falkman, G.: A Bayesian network approach to threat evaluation
with application to an air defense scenario, p. 7 (2007)
17. Groden, M., Collette, M.: Fusing fleet in-service measurements using Bayesian
networks. Marine Struct. 54, 38–49 (2017). https://doi.org/10.1016/j.marstruc.
2017.03.001
18. Friis-Hansen, A.: Bayesian networks as a decision support tool in marine applica-
tions. Ph.D. thesis, Department of Naval Architecture and Offshore Engineering,
Technical University of Denmark, Lyngby, oCLC: 1032700556 (2000)
19. Eleye-Datubo, A.G., Wall, A., Saajedi, A., Wang, J.: Enabling a powerful marine
and offshore decision-support solution through Bayesian network technique. Risk
Anal. 26(3), 695–721 (2006). https://doi.org/10.1111/j.1539-6924.2006.00775.x
20. Wireman, J., Goodrum, C., Sypniewski, M., Singer, D.: Leading indicators of
design for launch/recovery of naval autonomous underwater vehicles (AUV). In:
18th Conference on Computer and IT Applications in the Maritime Industries,
Tullamore, Ireland, vol. 18, p. 17 (2019)
21. NetworkX: NetworkX (2018). https://networkx.github.io/
22. Graphviz: Graph Visualization Software (2018). https://graphviz.gitlab.io/
23. Gayle, W.: Analysis of operational manning requirements and deployment proce-
dures for unmanned surface vehicles aboard US navy ships. Ph.D. thesis, Naval
Postgraduate School, Monterey, CA (2006)
24. Andrews, D.: A comprehensive methodology for the design of ships (and other
complex systems). Proc. Roy. Soc. Lond. Ser. A: Math. Phys. Eng. Sci. 454(1968),
187–211 (1998). https://doi.org/10.1098/rspa.1998.0154
 Communication of Design Space
Relationships Learned by Bayesian
Networks

Claire Wincott(B) and Matthew Collette

University of Michigan, Ann Arbor, MI 48109, USA

[email protected]

Abstract. Modern ship design often involves the automated creation of

thousands of design alternatives; even when provided a Pareto front of
optimal solutions designers may struggle to understand the differences
in designs and the relationships of design variables. Machine learning
Bayesian networks from automatically developed design data can allow
us to analyze the designs, understand the variable relationships that drive
their differences and optimalities, and lead engineers to better designs.
However, the information about variable relationships in Bayesian net-
work are encoded in difficult to interpret conditional probability tables
(CPTs). Translation of a Bayesian network’s CPTs into simpler edge
weights defining the strength of relationship between nodes allows engi-
neers to more easily interpret and use the complex information encoded
in the network through standard network analysis techniques. Bayesian
networks developed from a multi-objective bulk carrier design problem
developed by Sen are transformed to network adjacency matrices for such
analysis in this work.

Keywords: Bayesian networks · Design space exploration

1 Introduction
Modern ship design practices often involve the creation and evaluation of thou-
sands of design alternatives through automated analysis. However, as most ship
design problems are multi-objective, there is often no single, best solution but
rather a set of Pareto-optimal solutions; designers are often forced to rely on
their own knowledge of best practices to differentiate between solutions. Work-
ing with large complex systems, engineers may struggle to identify the driving
factors of a new design and develop a complete mental model of design space.
Machine learning allows designers to analyze and assess the large design
spaces created through automated design. One such technique is the use of
Bayesian networks to quantify and organize large sets of design data. In [9]
Pelikan pioneered the use of Bayesian networks for design in his development
of the Bayesian Optimization Algorithm (BOA), a estimation of distribution
algorithm similar to a genetic algorithm uses Bayesian principles to identify
c Springer Nature Singapore Pte Ltd. 2021
T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 24–37, 2021.
https://doi.org/10.1007/978-981-15-4680-8_2
 Design Relationships with Bayesian Networks 25

new, plausibly more fit gene sets. Further research uses Bayesian networks to
map sets of promising designs [12], learn continuous design variables without
discretization [4] and act as a surrogate for optimization under uncertainty [7].
Devine proposed using the underlying Bayesian networks and the information
they hold to provide to human engineers some new understanding of the design
space as it develops [2]. While most research utilize Bayesian networks to aid
computer learning and optimization, Devine’s proposal focuses on the transfer of
information in the networks to knowledge for the human designer as well. Unfor-
tunately, Bayesian networks encode information about variable relationships as
conditional probability tables that are not easily interpreted, especially as the
number of connected nodes grows. Translating this coded information into sim-
ple edge weights indicating the intensity of the relationship allows engineers to
easily identify highly coupled variables and perform network analysis to assess
relationships throughout the network of design data.

1.1 Sen Bulker Problem

In [11] Sen presents a non linear preliminary ship design model for bulk carriers
for use in multiobjective ship design comparison. The model uses six indepen-
dent design variables: length (L), draft (T ), depth (D), block coefficient (CB ),
breath (B) and speed (V ). Sen’s model identifies three objectives, transportation
cost minimization, light ship mass minimization and annual cargo maximization,
which are computed from the independent design variables through a set of 27
intermediate variables. The model defines 13 constraints: eight on the indepen-
dent design variables and five that include intermediate design variables. All 33
independent and intermediate design variables as well as the three objectives
are listed in Table 1. Of note, a(CB ), b(CB ) and P (Δ, V, a, b, L) are abstract
intermediate variables with no physical design meaning. The expressions defin-
ing each intermediate variable are excluded from this paper for brevity but can
be found explicitly in [11].

1.2 Bayesian Networks

The design data in this paper is analyzed through the use of Bayesian networks.
Bayesian networks are probabilistic, directed, acyclic graphs which encode vari-
able relationship data as a set of conditional probability tables, one for each
node [5]. The conditional probability tables of a Bayesian network define the
relationship between the likelihood of the child node states and the states of its
parent nodes. When some observation about the state of a node in the network
is made, the beliefs about the likelihood of states in the surrounding nodes of
the network can be updated. Figure 1 shows a basic Bayesian network which
holds information about our belief that it will rain, the sprinkler will be turned
on and that the grass will be wet. This network captures the relationship that
the occurrence of rain or a sprinkler turning on will make the grass wet, however
it takes some effort to understand that the occurrence of rain effects whether the
sprinkler will be turned on but that it is not a direct inverse relationship; the
26 C. Wincott and M. Collette

Table 1. Design variables in Sen bulk carrier model

Independent Design Variables

length, L
draft, T
depth, D
block coefficient, CB
beam, B
speed, V
Intermediate Design Variables
Froude number, F n
displacement, Δ
steel mass
outfit mass
a(CB )
b(CB )
sea days
BM
KG
KB
P (Δ, V, a, b, L)
GM
ship cost
capital charges
fuel carried
deadweight
running costs
port costs
stores/water
voyage cost
cargo weight
port days
round trip per annum, RTPA
annual cost
Objective Variables
light ship mass
annual cargo
transportation cost
 Design Relationships with Bayesian Networks 27

Fig. 1. Basic Bayesian network with relationships between rain, a sprinkler and wet
grass

sprinkler is often not turned on even when it is not raining. The purpose of this
work is to quantify the influence captured by the conditional probability tables
in more succinct way that can be easily communicated. The work presented in
this paper utilizes the commercial software HUGIN 8.7 to learn Bayesian net-
works from developed data sets and propagate observational evidence to evaluate
variable relationships.

1.3 Network Analysis

The focus of this work is to simplify the information contained in the condi-
tional probability tables of Bayesian networks into simple edge weights. Simpli-
fication of the networks allows for the application of standard network analysis
techniques and the straight-forward communication of the design relationships
encoded within. Measures of centrality, degree distribution and path distance
are used to gain greater understanding of the design space relationships [8].

2 Methodology

2.1 Design Space Sampling

To explore the design space of the Sen Bulker problem a full design optimization
was mirrored by creating 30 design datasets, ten sampled sets broken down
into three subsets each: the complete or full set of as-sampled data points, a
feasible set of the data points from the full set that violate no constraints and a
Pareto-optimal set of non-dominated data points from the feasible set. The full
dataset was created from a modified latin hypercube sampling of the Sen Bulker
model. Each of the 10 full datasets was created from an initial normalized latin
hypercube sampling of 100,000 data points of six input variables using python
package pyDOE [6]. The normalized points were then fit to the feasible ranges of
the six independent variables; while some of the feasible upper and lower limits of
the independent variables are concretely defined, others are defined sequentially
as a function of the other independent variables at that point, so the final set of
28 C. Wincott and M. Collette

Table 2. Upper and limits of six independent variables

Variable Lower limit Upper limit

CB 0.63 0.75
V 14 18
L V
( 0.32 )2 /g 362
D L/15 30
T L/19 0.7 ∗ D + 0.7
B 5 L/6

100,000 points is not a true latin hypercube sampling. This was done to reduce
computational expense in creating and identifying a large enough set of feasible
variables. The limits as applied to the normalized latin hypercube data points
are delineated in Table 2. The intermediate variables and objective functions
were evaluated for each of the sampled data points. This set of data points is
referred to as the full dataset for this work. As mentioned, ten full data sets were
developed for analysis.
The feasibility of each design point is then considered by eliminating points
whose intermediate variables violate one or more of the constraints defined in
the Sen Bulker model. (Constraints affecting only design variables are managed
through the sampling process). This smaller subset of design points is referred
to as the feasible set and is obtained for each of the ten full data sets.
Finally, for each feasible set the non-dominated points were extracted to
identify the Pareto front in each of the ten feasible data sets. This subset of
Pareto-optimal points is referred to as the Pareto set and is evaluated for each
feasible data set.
A total of 30 design spaces were developed: a full, feasible and Pareto set for
each of the ten latin hypercube samplings. For all ten datasets, the feasibility
and Pareto restrictions pare down the sample size relatively consistently across
all ten samplings. The number of feasible and Pareto design points identified for
each full set of 100,000 points are listed in Table 3.
While a formal optimization has not been completed, these design data sets
provide a representation of the design space at progressive stages of design.

2.2 Learning Bayesian Networks

A total of 30 networks were machine learned using HUGIN, a software for
Bayesian network research [1]. HUGIN provides a Learning Wizard, which learns
a Bayesian network structure and complies its conditional probability tables from
a provided dataset.
The network structures were learned using a greedy search-and-score algo-
rithm with the Bayesian Information Criterion (BIC) as its score function.
HUGIN’s greedy search-and-score algorithm uses a score function to assess the
merit of possible structures. The algorithm searches through possible network
 Design Relationships with Bayesian Networks 29

Table 3. Number of design data points in each subset

Datset Total points Feasible points Pareto points

1 100,000 11,936 903
2 100,000 11,950 863
3 100,000 11,659 800
4 100,000 11,627 886
5 100,000 11,910 906
6 100,000 11,776 1053
7 100,000 11,846 908
8 100,000 11,792 915
9 100,000 11,773 835
10 100,000 11,670 926

structures by adding, removing and reversing possible edges until the score of
the structure can no longer be improved. The BIC score is measure of the con-
ditional entropy of each node given candidate parent nodes; a higher BIC score
indicates the value of more information about a parent node in learning infor-
mation about the child node. The greedy search-and-score algorithm with a BIC
scoring function in HUGIN was selected to reflect the search algorithm used by
Pelikan in the Bayesian Optimization Algorithm, and further discussion of its
benefits are found in [9].
The conditional probability tables were constructed by binning all data into
ten uniform intervals for design variable, intermediate variable and objective
function value. HUGIN automatically populated the conditional probability dis-
tributions and densities from the provided data sets through batch learning.
HUGIN also stores the evidence counts for each probability table to account for
the number of individual data points fitting each bin. These evidence counts are
used in later edge weight assessment.

2.3 Transforming Conditional Probabilities to Edge Weights

Bayesian networks hold a considerable mass of information in their connections

and conditional probability tables; however, the format of these tables makes
it difficult to interpret and communicate the knowledge the information holds.
To simplify the information into to usable knowledge, this work attempts to
translate the conditional probability tables into simple edge weights defining
relationship importance using two metrics: match distance and K2 score. Match
distance is a measure of difference between two cumulative histograms and was
chosen for its simplicity and clarity. K2 score is a more complex score used in the
learning and scoring of Bayesian structures and, while less intuitive, is already
computed in some optimization algorithms using Bayesian networks, like [2], so
30 C. Wincott and M. Collette

adds little additional computational expense. Each metric and its normalization
to an edge weight between 0..1 is explained in the following subsections.

Match Distance. The match distance between two histograms H, K is

defined as 
dM (H, K) = |ĥi − k̂i | (1)
i

where ĥi = j≤i hi is the cumulative histogram of hi and k̂i is the same [10].
To determine an edge weight between two nodes this distance was computed
between the prior distribution of the child node and the distribution after each
state of the parent node is selected and normalized,
q r 
i=1 j=1 |P (X ≤ xj ) − P (X ≤ xj |Y = y i )|
M Dedge = (2)
q
where q is the number of states yi in parent state Y , r is the number of states xj in
child state X, and P (X ≤ xj ) is the cumulative probability of X in state xj .

K2 Score. The K2 score is traditionally a cumulative score of its values for

each edge of a network summed to evaluate the fitness of a candidate Bayesian
network. This work uses the log gamma K2 of each edge as follows:
⎛ ⎞
q  r
⎝log Γ (r)
loggammaK2 = + log(αij Γ (αij ))⎠ (3)
i=1
Γ (N i + r) j=1

where q is the number of parent states, r is the number of child states, Ni is

the number of observed data points with parent state i and αij is the number of
observed data points with parent state i and child state j. The raw log gamma
K2 score results in scores ranging from O(−105 ) to O(−102 ), where values closer
to zero indicate a stronger correlation. The raw values are also a factor of the
total number of data points assessed and increase in magnitude with a greater
number of data points. To normalize these scores we use the log gamma K2
score between two uniform distributions with no correlation. Since all of our
nodes have 10 states, this normalization factor is constant across all nodes of
the same network but varies between networks trained with different sized data
sets. The normalization factor then equals
⎛ ⎞
q r  
⎝log Γ (r) N N ⎠
loggammaK2normf actor = N
+ log Γ
i=1
Γ ( q + r) j=1
qr qr
(4)
 Design Relationships with Bayesian Networks 31

where q and r are the same as above, and N is the total number of observed
data points [2]. The edge weight for each nonzero network edge is
    
q Γ (r) r N N
i=1 log N
Γ ( q +r)
+ j=1 log qr Γ qr
K2edge = 1 −       . (5)
q Γ (r) r
i=1 log Γ (Ni +r) + j=1 log(α ij Γ (α ij ))

2.4 Network Analysis

The edge weights extracted from the Bayesian networks using match distance
and K2 score are compiled into an adjacency matrix for network analysis. While
the Bayesian networks are learned with directed edges, the adjacency matrices
have been populated representing undirected networks because were are more
interested in the coupling and centrality relationships of nodes than their direc-
tional causality. Figure 2 provides an example of a weighted, undirected network
represented by each adjacency matrix. Nodes, edges and edge weights are the
only network attributes stored by these networks, abandoning the conditional
probability tables used to define relationships in the Bayesian networks.

port days

0.89
displacement transportation costs
cargo weight
0.84 0.70 port costs
0.87
0.74
a deadweight running costs
0.64 T
0.70 0.64 0.72 0.65 0.87
CB Fn crew, stores and water L KB
0.56 0.61
0.64
RTPA 0.71 D
b 0.59 0.87
annual cargo 0.73 annual costs KG
0.72
0.72 0.77
voyage costs
0.56 capital charges outfit mass
fuel cost
0.88 0.62 0.74
0.85
fuel carried ship cost B lightship mass
V
0.77 0.68 0.57 0.81
0.71
P machinery mass GM steel mass
sea days
0.81
0.88
BM
daily consumption

Fig. 2. Example weighted edge Pareto-set network using K2 scored weights Network
with Edge Weights.pdf

Once the adjacency matrices have been compiled for match distance and K2
scores for each of the 30 Bayesian networks, we complete a network analysis to
identify the strongly coupled and central nodes and compare networks at varying
stages of design. The network analysis was completed primarily using the python
software package NetworkX [3].
32 C. Wincott and M. Collette

Three measures of centrality were computed for each of the 30 networks:

degree centrality, closeness centrality and betweenness centrality. Degree central-
ity is the fraction of nodes in the network with which it is directly connected.
This is a simple measure of the number of edges of each node. Closeness central-
ity is a more complex metric, measuring the reciprocal of the average shortest
path distance to all reachable nodes. For node v,


n−1
n−1
CC (v) = , (6)
u=1
d(u, v)

where d(u, v) is the shortest-path distance between nodes u and v for all n
nodes v that can reach u. Betweenness centrality is the sum of the fraction of
all shortest paths that pass through the node. Formally, for node v betweenness
centrality is
 σ(s, t|v)
CB (v) = , (7)
σ(s, t)
s,t∈V

where V is the total set of nodes, σ(s, t) is the number of shortest paths between
node pairs, and σ(s, t|v) is the number of those pairs that pass through v other
than v = s, t [8].

3 Results
The 10 sets of design data produce relatively consistent Bayesian networks from
which an analysis of the design space can be completed. The full sets produced
networks with the most edges and the number of edges decreases consistently
within each data set as the design space is pared down to feasible and Pareto
set networks. The average number of edges in each network, number of edges
found in all ten networks of the same type and the total number of unique edges
found in at least one network of the type are listed in Table 4. As shown by the
decrease in number of connections, the networks learned at each design step are
progressively simpler. This implies that at later design stages fewer relationships
are influential; many of the trivial connections are weeded out by the application
of constraints and identification of Pareto-optimal points. Of the 26 connections
common to all full-set networks, 23 common to all feasible-set networks and 17
common to all Pareto-set networks, fifteen of those edges are in all 30 networks.
The 15 consistently present edges are delineated in Table 5. These edges represent

Table 4. Summary of edges found in the ten full-, feasible- and Pareto-set networks

Average Common Total unique

Full 56.3 26 128
Feasible 36.6 23 63
Pareto 33.0 17 59
 Design Relationships with Bayesian Networks 33

Table 5. Edges present in all 30 networks

Running costs – transportation costs

Annual costs – outfit mass
V – sea days
P (Δ, V, a, b, L) – daily consumption
a(CB ) – CB
b(CB ) – CB
BM – GM
Fuel cost – fuel carried
Cargo deadweight – port days
Displacement – deadweight
Capital charges – ship cost
Crew, stores and water – running costs
KG – D
KB – T
Lightship mass – steel mass

many of the most direct relationships from the Sen Bulker model as well as some
relationships which are logical but not explicit in the equations of the model. For
example, in the model capital charges is a linear function of ship cost, as is KB of
T , KG of D and daily consumption of the intermediate variable P (Δ, V, a, b, L).
Others of these consistent connections are more distantly related in the model’s
computations. For instance, annual cargo and running costs are consistently
connected: annual cargo is a function of RT P A and cargo deadweight, itself a
function of deadweight and two other variables, while running costs is function
of deadweight.

3.1 Analyzing Edge Weights

The Bayesian network edges enumerated in this section are each assigned edge
weight values using normalized match distance and K2 scores. To measure the
consistency between each of the ten full-, feasible- and Pareto-set networks,
the coefficient of variation was measured element wise for each group of ten
adjacency matrices. While it decreases sequentially from the full- to Pareto-set
networks, the average variation of edge weights is substantial when measured
for all unique edges. However, when just the edge weights of the 26, 23 and
17 consistently present edges across full-, feasible- and Pareto-set networks are
measured the coefficients of variation decrease dramatically. This suggests that
the high variance is due to varying edge presence among the non-consistent
edges rather than highly varying values of the match distance and K2 scores for
the same edges in different data sets. In the construction of Bayesian networks,
when redundant connections can be made multiple ways, the learning algorithm
34 C. Wincott and M. Collette

will favor the simplest structure with the fewest edges [5]. This means areas in
the network where multiple similar edge connections are available, the Bayesian
network will only use whichever is strongest for the particular dataset and not
include the other to keep the structure as simple as possible. This may lead to
an apparent inconsistency among subsets though the information passed may be
equivalent (Table 6).

Table 6. Average coefficients of variation for all edges and consistently present edges

All edges Consistent edges

MD score K2 score MD score K2 score
Full 0.723 0.715 0.051 0.041
Feasible 0.469 0.432 0.023 0.017
Pareto 0.511 0.474 0.025 0.018

The average nonzero edge weights and their progression with the stages of
design, found in Table 7, are indicative of the design space relationships. For
both the match distance and K2 scores the standard deviation of the nonzero
edge-wise average weights decreases as the datasets are trimmed of infeasible
and non-optimal solutions. This trends shows that the edges found generally
become more individually impactful, as there is less variation in the influence
of the edges learned by the Bayesian networks. The mean value of the match
distance edge weights increases as the distribution becomes more narrow and
less-impactful connections are lost in later design stages. The K2 edge weights
do experience a drop off in average value, an effect of losing some of the highly-
correlated edges found in the full-set networks with edges values well above 0.9.
Interestingly, the seven edges consistently found in the full-set networks and not
the feasible-set networks all have average K2 edge weights greater than 0.9 in
the full-set networks. The loss of these high scoring edges accounts for much of
the drop in the mean K2 edge weights.

Table 7. Mean and standard deviation of average nonzero edge weights for all edges

MD score K2 score
Mean SD Mean SD
Full 0.562 0.127 0.890 0.160
Feasible 0.575 0.081 0.797 0.144
Pareto 0.580 0.067 0.711 0.095
 Design Relationships with Bayesian Networks 35

3.2 Measuring Centrality

Centrality measures follow a consistent pattern with the trends observed in the
edge weight values. The standard deviations of all three centrality scores, degree
centrality, closeness centrality and betweenness centrality, decrease as the dataset
progress from full to Pareto subsets, shown in Table 8. This indicates that the
networks are learned much more consistently as extraneous design data points
are thrown away. The largest decreases in average standard deviation of the
measured centrality values occur between the full- and feasible-set networks,
indicating that the randomly generated, inviable design points have the biggest
impact on how consistently the networks can be learned and analyzed.

Table 8. Mean standard deviation of centrality measures

Degree centrality Closeness centrality Betweenness centrality

MD score K2 score MD score K2 score
Full 0.024 0.068 0.039 0.019 0.019
Feasible 0.010 0.035 0.027 0.006 0.006
Pareto 0.009 0.031 0.026 0.008 0.008

The mean values of degree centrality, closeness centrality and betweenness

centrality consistently decrease across the networks from full to Pareto datasets,
as seen in Table 9. This appears consistent with the decrease in standard devi-
ation of the average edge weights decreasing because as the individual edges
become more uniform in value, impacts spread more broadly across the network.
While highly coupled nodes create strong centers in the Pareto-set networks and
remaining edge weights become stronger in the feasible and Pareto networks,
the influential centers weaken. Interestingly, the standard deviations of the cen-
trality values show more complex relationships; especially when considered as
a coefficient of variation, the standard deviation as a fraction of the mean, the
increasing trend of these values are significant. For closeness centrality, a measure
of a node’s path-wise closeness to all other nodes, variation increases significantly
as the total values decrease with the progression of the design set; a few nodes
lose almost all centrality while others maintain a significant amount. The same
general trend is observed for both match distance and K2 edge weight networks.
The betweenness centrality distributions show an even more exaggerated trend.
While the full-set network nodes have a multimodal distribution, as the design
points are pared down to feasible and Pareto datasets these distributions become
much narrower, with a majority of variables maintaining very little betweenness
centrality and only a few nodes having significant centrality in later stages of
design. These variables, running cost, annual cost, annual cargo and crew, stores
and water for this model, may be most indicative of the impact of changes to
the more peripheral variables on the overall design solution.
36 C. Wincott and M. Collette

Table 9. Mean and standard deviation of average centrality values

Degree centrality Closeness centrality Betweenness centrality

Mean SD MD score K2 score MD score K2 score
Mean SD Mean SD Mean SD Mean SD
Full 0.089 0.037 0.565 0.082 0.352 0.046 0.023 0.016 0.023 0.015
Feasible 0.058 0.024 0.330 0.099 0.231 0.069 0.010 0.012 0.010 0.012
Pareto 0.052 0.026 0.279 0.106 0.222 0.085 0.009 0.009 0.009 0.009

4 Discussion
This analysis of learned Bayesian networks using traditional weighted edge net-
work analysis provides promising insights into the possibly for its use in large,
more complex design systems. While the networks learned provide little new
information that cannot be identified from the explicit equations of the Sen
Bulker model, this work verifies that the method can be successful in identifying
important relationships in more complex models whose variable relationships
are obscured by computations across multiple disciplines or black box analy-
ses. For the Sen Bulker model, this method is able to identify variables which
were strongly linked, either explicitly in the model’s equations or through more
distant connections. The methods identification of central variables like annual
cost, annual cargo and running cost shows promise for larger applications where
variable relationships are less explicit; however, the identification of crew, stores
and water as the most central node signifies more exploration of the method is
needed to insure intermediate placeholder nodes are given less importance. To
make this method viable for large scale, complex design optimization problems
further exploration is needed of alternate edge weight scoring methods, addi-
tional weighted network analysis methods and its ability to scale to more than
36 variables at three distinct design stages.

5 Conclusion
This work automatically learned 30 Bayesian networks from 10 datasets of full,
feasible and Pareto-optimal design data from Sen’s bulk carrier model. From
these thirty Bayesian networks, edge weights were calculated using match dis-
tance and K2 score to extract influence information from the networks’ condi-
tional probability tables into easily interpreted single values. Network adjacency
matrices populated with these edge weights were used in network analysis to
identify highly coupled and central nodes. The networks and resulting analy-
sis were successful in identifying relationships both explicit in the Sen Bulker
model and logically consistent with designers’ knowledge. Further work in this
area would allow designers to identify strong relationships implicit within large
multidisciplinary optimizations too complex for engineers to form an accurate
mental model.
 Design Relationships with Bayesian Networks 37

Acknowledgements. This material is based upon work supported by the National

Defense Science and Engineering Graduate Fellowship under the Office of Naval
Research ID No. ND-NAV-004-085.

References
1. Hugin Researcher (2018), http://www.hugin.com
2. Devine, T.E.: Using and Interpreting the Bayesian Optimization Algorithm to
Improve Early Stage Design of Marine Structures. Ph.D. thesis, University of
Michigan, Ann Arbor (2016)
3. Hagberg, A., Schult, D., Swart, P.: NetworkX (2018), https://networkx.github.io/
4. Hu, Z., Mahadevan, S.: Bayesian network learning for data-driven design. ASCE-
ASME J. Risk Uncert. in Eng. Syst. Part B: Mech. Eng. 4(4), 041002 (2018)
5. Kjrulff, U.B., Madsen, A.L.: Bayesian networks and influence diagrams: a guide
to construction and analysis. Information Science and Statistics, Springer-Verlag,
New York (2008), https://www.springer.com/us/book/9781441925466
6. Martinez, J.M., Collette, Y., Baudin, M., Christopoulou, M.: pyDOE, https://
pythonhosted.org/pyDOE/#source-code
7. Nannapaneni, S., Liang, C., Mahadevan, S.: Bayesian network approach to mul-
tidisciplinary, multi-objective design optimization under uncertainty. In: 18th
AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference (2017)
8. Newman, M.: Networks. Oxford University Press, Oxford (2018)
9. Pelikan, M.: Bayesian Optimization Algorithm: From Single Level to Hierarchy.
Ph.D. thesis, University of Illinois at Urbana-Champaign, Urbana, Illinois (2002)
10. Rubner, Y., Tomasi, C., Guibas, L.J.: The earth mover’s distance as a metric for
image retrieval. Int. J. Comput. Vis. 40(2), 99–121 (2000)
11. Sen, P., Yang, J.B.: Multiple objective decision making. In: Multiple Criteria Deci-
sion Support in Engineering Design, pp. 113–175. Springer London, London (1998),
https://doi.org/10.1007/978-1-4471-3020-8 4
12. Shahan, D.W., Seepersad, C.C.: bayesian network classifiers for set-based collab-
orative design. J. Mech. Des. 134(7), 071001–071014 (2012). https://doi.org/10.
1115/1.4006323
 Optimizing Ships Using the Holistic
Accelerated Concept Design Methodology

Roy de Winter1,3(B) , Jan Furustam2 , Thomas Bäck3 , and Thijs Muller1

1
C-Job Naval Architects, 2132JN Hoofddorp, The Netherlands
{r.dewinter,t.muller}@c-job.com
2
NAPA Group, FI-00180 Helsinki, Finland
[email protected]
3
Leiden Institute of Advanced Computer Science, 2333CA Leiden, The Netherlands
{r.de.winter,t.h.w.baeck}@liacs.leidenuniv.nl

Abstract. The holistic accelerated concept design optimization method

is a new method to optimize ships in an efficient way. This optimization
method is developed to automatically check feasibility and at the same
time optimize ships by generating different design variations. The naval
architect uses his creativity and imagination to develop a parametric
three-dimensional ship model which can then be checked and optimized
by automated software. This way, the naval architect can focus on the
innovative and creative aspects of ship design while repetitive labor inten-
sive tasks can be executed by the computer. It is shown in this paper
that the holistic design optimization method leads to improved design
variations in a short amount of time.

Keywords: Ship design · Holistic optimization · Multi-objective ·

Automated · Design change · 3d product model

1 Introduction
Conventionally, ships are designed and optimized in an iterative design process
illustrated by the well-known design spiral [1]. In the design spiral several sequen-
tial decisions are made in such a manner that the objectives are optimized and
constraints imposed by physics and regulating authorities are met. However, this
approach is very time consuming and the design process likely leads to a so called
locally optimal solution. The locally optimal solution is found due to the fact
that it is impossible for human experts to consider all the dependencies between
the decision variables, the constraints and the objectives [2].
Ship design optimization problems can be solved much more efficiently.
Therefore, C-Job Naval Architects1 has implemented a holistic design method-
ology. The holistic design methodology, better known as the Accelerated Con-
cept Design (ACD) framework, uses the design circle as presented in Fig. 1 and
deals with all the decision variables, constraints, and objectives simultaneously.
1
C-Job Naval Architects, Ship design and Engineering (2019), https://c-job.com/.
c Springer Nature Singapore Pte Ltd. 2021
T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 38–50, 2021.
https://doi.org/10.1007/978-981-15-4680-8_3
 Accelerated Concept Design 39

Fig. 1. The Accelerated Concept Design Circle illustrates the method where all con-
straints, objectives, and parameters are considered simultaneously. The level of accu-
racy of the design can be picked and mixed by the user. However, high fidelity simu-
lations typically need more computational effort compared to the low fidelity approxi-
mations.

This simultaneous decision-making is done with automated software tools and a

multi-objective optimization algorithm named CEGO [3]. This multi-objective
optimization algorithm considers the whole design space which makes it more
likely that a globally optimal solution will be found.
The software tools that are implemented in the ACD framework are able to
deal with every design aspect. The design aspects represented in the ACD frame-
work are floating position, intact stability, damage stability, strength, weight
& cost, space reservation, resistance and motions. Given a parametric concept
design, every possible design variation can be automatically generated and eval-
uated using the ACD framework. Because every aspect of the ACD framework is
fully automated, the concept design can easily be varied and optimized towards,
for example, the total cost of ownership.
The paper is organized as follows: Section 2 reviews ship design optimization
in practice. Section 3 describes the novel Accelerated Concept Design approach.
Section 4 illustrates a ship design optimization problem. Section 5 describes
the conducted experiments. Section 6 discusses the results on the ship design
optimization problem. Finally, in Sect. 7 conclusions are drawn and an outlook
is given.
40 R. de Winter et al.

2 Ship Design Optimization

When optimizing a ship, all disciplines should work together to come to an opti-
mal solution. This solution, however, is often hard to find since there are many
choices to be made. In practice they are made based on experience and heuristics
while they should be made based on estimated or simulated performance data.

2.1 Ship Design Optimization in Practice

Typically in the first design phase, the main dimensions of a ship are to be
decided on. This brings many complex engineering questions which will have a
strong impact on the final result. However, it are these decisions that need to be
made in the shortest amount of engineering time. The more unique the ship is,
the higher the challenge will be to answer the questions with confidence. As the
design progresses, these early choices are hard to reverse without major delay.
This pushes the designers towards risk mitigation instead of innovation. Innova-
tion is therefore often seen over several designs, each being a small improvement
over the other. In order to allow for more radical design changes in a short time
period and to manage the involved risk, a Holistic Ship Design method is needed
to assist the naval architect in making informed decisions in a short amount of
time.

2.2 Holistic Ship Design Optimization

Holistic optimization has been applied in different fields of engineering. For

example it has been studied before in ship design optimization by Papaniko-
lau et al. [5]. It has also been used to optimize aircraft designs [7] and passenger
cars [6]. Similar techniques have also been used to optimize plant wide produc-
tion processes [4]. All these optimization problems have in common that they
are solved by taking a holistic, integrated perspective while solving the problem
with an optimization algorithm.

3 Accelerated Concept Design

The ACD framework consist of three elements, namely (1) a software package
where the ship is defined, parameterized and evaluated, (2) the CEGO opti-
mization algorithm, and (3) some guidelines on how to set up the optimization
problem.

3.1 Ship Definition and Evaluation Software

The ACD framework at C-Job Naval Architects is implemented in the NAPA

software package2 . In this software it is easy to model, parameterize and evaluate
2
NAPA Oy, Release 2017.3-3 (2019), NAPA software, http://www.NAPA.fi/.
 Accelerated Concept Design 41

any type of vessel. The framework takes as input the parameters that define the
design variation and then automatically computes the corresponding constraint
and objective function scores. Note that this is not the only software package
in which it would be possible to set up the ACD framework. However, the pro-
gramming capabilities in NAPA make the platform modular, easy to adjust, and
give users the possibility to establish a connection to other software packages.

3.2 CEGO

An algorithm specially developed to optimize constrained multi-objective prob-

lems is the Constrained Efficient Global Optimization (CEGO) algorithm [3].
CEGO is a surrogate-assisted optimization algorithm that uses Design and Anal-
ysis of Computer Experiments (DACE) to explore and model the design space.
Moreover, it is designed for optimizing multiple objectives at the same time since
CEGO optimizes the (hyper)volume between a reference point and the Pareto
optimal set as described in [12]. A benefit of using this algorithm is that after an
initial sample of design variations has been evaluated, for any new decision vari-
able combination, the surrogate models can be used to predict the corresponding
objective and constraint scores. This makes the ship design optimization process
efficient and fast since a lot of human effort and simulation time can be saved.
An implementation of the CEGO algorithm can be found on Github [11].

3.3 Optimization Problem Definition

There are some guidelines to follow when setting up a ship design optimization
problem in the ACD framework. The guidelines deal with parameterization of
the vessel, constraint and objective functions.

Parameterization Guidelines. Important to remember when defining a para-

metric model is that it are these decision variables that will define the future
performance of the vessel. Therefore, the ship should be parameterized in such
a manner that the performance of the vessel depends on the decision param-
eters. Defining decision variables that only marginally influence the constraint
and/or the objective values will only have marginal influence so they can better
be neglected or be regarded as a constant throughout the whole process. This is
the case because more parameters means more required computation effort for
CEGO to propose a new solution.
Secondly, it is important that after all the decision variables have been set, the
geometry is formed accordingly. After this, the geometry should not be changed
anymore (for example to meet a constraint) until a new combination of decision
variables is chosen. This way, the input truly corresponds with one unique geom-
etry. After the geometry has been evaluated, it will be these decision parameters
that define the geometry and the corresponding constraint and objective values.
This approach can sometimes result in infeasible solutions since some combina-
tions of decision parameters will result in a geometry that violates one or more
42 R. de Winter et al.

of the constraints. However, this is not a major problem since CEGO directly
links the decision parameters of the infeasible solutions to the corresponding
constraint and objective values and therefore will learn not to try this decision
parameter combination again.

Constraint Function Guidelines. A guideline to take into consideration

when defining constraint functions is to make sure that the output of the con-
straint function is real valued and preferably continuous. Bad practice is a simple
boolean constraint value corresponding to the constraint being violated or met.
The real valued constraint function is important since CEGO takes into account
how much a constrained is violated or met. This information can be learned by
the optimization algorithm and can be crucial since, more often than not, the
optimal design variation is located on the border of the feasible area.

Objective Function Guidelines. When defining an objective function, it is

good practice to write a real valued, and again preferably continuous, function
that only represents one objective at a time. This is preferable to a weighted sum
of all objectives. Similarly, one should not add penalties to the objective score if it
violates any of the constraints. This way, the objective function remains smooth
and the true optimal solution can be found more easily by the surrogate-assisted
optimization algorithm.

4 Ship Design Optimization Problem Definition

In this section a ship design optimization case is presented. In practice, every
ship design optimization problem is different but in theory, it is nothing more
than defining a ship design optimization problem and solve it. Note that this
is the stage where the naval architect can use his imagination and creativity to
come to new innovative designs. This is the case, since this is where the first
design variation is set up and evaluated.
To illustrate this process of defining and optimizing a vessel, a Trailing Suc-
tion Hopper Dredger (TSHD) designed by C-Job Naval Architects is described
and optimized in this paper. This TSHD will be optimized for total cost of owner-
ship by executing a Design and Analysis of Computer Experiments (DACE) [13]
to find the optimal geometry while taking into account the requirements imposed
by physics and the regulating authorities. The original design with the annotated
decision parameters is shown in Fig. 2.

4.1 Parametric Geometry

The first step is to generate a parametric geometry of a vessel. The geometry
depends on decision variables which are numerical quantities for which values
can be varied in the optimization process [9]. The decision variables in this case
are used to rearrange the bulkheads of the vessel and as input of a Free-Form
Deformation technique (FFD) [8] to modify the hull.
 Accelerated Concept Design 43

Fig. 2. Trailing suction hopper dredger with design parameters annotated.

Repositioning Bulkheads. The rooms inside the hull can easily be parame-
terized. This is very useful since it will be the rooms and their loading that will
define the floating position, intact stability, damage stability, draft, heel, trim
ect. Parameterizing rooms can be done by moving locations of bulkhead along
the x, y and z axis. As an example, the bulkheads that define a hopper of a
TSHD are moved along the x-axis in Fig. 3. This way, we can make the hopper
smaller, bigger, or change the hopper position completely.

Fig. 3. Bulkheads moved forward and backward to define different hopper lengths.

Good practice is to make bulkheads dependent of each other. To illustrate

this the following example is given: if bulkhead 1 is located at frame 0, the room
between bulkhead 1 and 2 is 7 frames long, then bulkhead 2 is located at frame 7.
The room between bulkhead 2 and bulkhead 3 is 10 frames long. Then bulkhead
3 is located at frame 7 + 10 = 17. In case the length of room 1 is increased by x
frames, then move bulkhead 2 to 7 + x and move bulkhead 3 to 7 + x + 10.

Free-Form Deformation. When using automated optimization tools for alter-

ing the hull form of the vessel, a challenge may arise with keeping the fairness
(and other important characteristics) of the hull surface. Using the Free-Form
Deformation technique, the hull form can undergo rather radical changes in
a robust way. The optimization by FFD is done by applying a lattice (box)
around the surface or part of the surface which is to undergo transformation.
The lattice is then deformed by translating its vertices. The control points of
44 R. de Winter et al.

Fig. 4. Free form deformation applied to fore- and aft-hull.

the circumscribed surface will consequently undergo transformation leading to

a new surface geometry. An example of a FFD applied on the fore- and aft-ship
of a hull is illustrated in Fig. 4.

Trailing Suction Hopper Dredger Parameters. The TSHD design we have

parameterized is presented in Fig. 2. The model has the following six decision
variables: Δbreadth ∈ [−1.6, 3.4], Δlength ∈ [−2.8, 9.8], foreship length ∈ [16, 22],
hopper length extension ∈ [5, 9], hopper breadth ∈ [5, 9], hopper height ∈ [12, 16].
The ship length and ship breadth are annotated with a Δ because these variables
are used to apply a FFD on the initial hull. The other parameters are all bulkhead
locations which can be moved and/or extended along the x,y, and z direction.

4.2 Constraints
After defining the parameters and the geometric model, the ship should be evalu-
ated to check if it meets the constraints given by all stakeholders. The constraints
can be seen as hard boundaries which should not be exceeded for the ship to be
feasible. The constraints are typically defined as function inequalities: gi (x) ≤ 0
where x are the decision parameters and gi () is one of the evaluation functions
used to evaluate one of the m constraints [3].
A typical constraint function gi (x) consists of two parts: a continuous func-
tion f (x) which takes as input arguments, the decision variables or (a part of)
the geometry of the model, and a constraint boundary ci that represents the
maximum value for the ith constraint. These two parts taken together form
the inequality constraint fi (x) − ci ≤ 0. This inequality constraint can then be
rewritten to its general form: gi (x) ≤ 0. When constraints have a minimum value
they can be rewritten without loss of generality to a function with a maximum
value by multiplying by −1. Note that the function f (x) does not need to be
a function which can easily be written by mathematical formulas. The function
f (x) can also be a complex function which represents the wave resistance, or
any other assessment function represented by a simulator, for example.

Trailing Suction Hopper Dredger Constraints. For the TSHD some quite
specific constraints have to be set in order to create sensible design variations
which meet the stakeholders demands and which do not violate any of the regu-
lations. These constraints mainly focus on the hopper volume, stability criteria,
space reservation, fuel capacity, water ballast usage, and the dredging area where
 Accelerated Concept Design 45

the vessel will be dredging and dumping. The details on these constraints are
defined below:

Hopper volume: The hopper where the dredged soil is dumped in, should
at least be 8000 cubic meters to meet the client’s demand. The inequality
constraint for the hopper is therefore: 8000 m3 − VOL(hopper) ≤ 0.
Fuel capacity: The design variation has 9200 kW installed power, the fuel type
is Marine GasOil (MGO), and the design variation should at least be able to
sail for 21 days. Therefore a relatively conservative fuel capacity constraint is
defined: 820 m3 − VOL(fueltanks) ≤ 0.
Propulsion Engine: The engine room should at least be large enough to accom-
modate the required space and volume of the propulsion engine. Therefore,
the propulsion engine and the engine room are modelled and intersected. The
volume of the intersection of the two rooms should be at least be equal to
engine volume, if this is not the case, we know the engine does not fit in the
engine room. V OL(Engine) − V OL(EngineRoom ∩ Engine) ≤ 0.
Dredge Pump: The procedure for the dredge pump is the same as for the
Propulsion Engine. Therefore the constraint for the dredge pump is the fol-
lowing: V OL(P ump) − V OL(P umpRoom ∩ P ump) ≤ 0.
Accommodation Space: The accommodation on the design variation should
host 29 crew members. For 29 persons an estimation of the required space is
made, this estimation resulted in a required 1100 cubic meters accommoda-
tion space. Therefore the inequality constraint for the accommodation is the
following: 1100 m3 − VOL(Acc) ≤ 0.
Draft: The design variation should be able to sail in shallow waters, therefore
the draft T should not exceed 7 m when fully loaded. The corresponding
inequality constraint is therefore: T − 7m ≤ 0 such that when the design
variation is fully loaded it is still capable of sailing in shallow waters.
Fore peak bulkhead: The fore peak bulkhead, sometimes referred to as col-
lision bulkhead, should be positioned according to the Maritime SOLAS
rules M IN (0.05L, 10 m) ≤ d ≤ MAX(0.08L, 0.05L + 3), where L is ship
length and d is the collision bulkhead position. Rewritten and separated
into two inequality constraints, this rule then gives the following inequali-
ties: M IN (0.05L, 10 m) − d ≥ 0, and M AX(0.08L, 0.05L + 3) − d ≤ 0.
Stability Criteria: The stability requirements are evaluated by expressing the
applicable regulatory requirements as a meta-centric height value GMc . Giv-
ing any possible loading condition of the ship, the meta-centric height of the
design variation should at least be larger or equal to the prescribed metra-
centric height value to pass all the stability requirements. The stability func-
tion inequality therefore is the following: GMo − GMc ≤ 0.
Trim: Giving different loading conditions with the hopper empty and the hopper
filled, without using water ballast, the trim varies. Because we are dealing
with a ballast-less design, it is necessary to keep the difference in trim within
practical limitations. The trim has therefore been limited to a maximum
difference of 2.1 m. The inequality constraint for trim is therefore described
as follows: (M AX(trimlc ) − M IN (trimlc )) − 2.1 m ≤ 0.
46 R. de Winter et al.

Heel: The same procedure as for trim is followed for heel. The difference between
the maximum and minimum heel is not allowed to be larger than 0.2◦ . The
function inequality is therefore the following: (M AX(heel) − M IN (heel)) −
0.2◦ ≤ 0

Strength is also a critical constraint, but because we are optimizing the steel
weight, we make sure that we only generate design variations which comply with
the strength regulations, therefore a strength constraint is not added.

4.3 Objectives
The objectives are the results which the client wants to achieve, it can be any-
thing as long as it is measurable and dependent on the decision parameters.
Objectives are typically conflicting and non-commensurable. Therefore, often
one objective can not be improved without sacrificing an other objective. As a
consequence, there is not one perfect solution but a set of non-dominated solu-
tions which form the Pareto optimal set. The definition of Pareto optimal set is
used as described in [9].

Trailing Suction Hopper Dredger Objectives. The objective to be min-

imized for this optimization problem is towards the total cost of ownership of
a dredger. The total cost of ownership of a dredger is typically calculated by
the client, the ship owners and operators of dredgers. The calculations are often
based on business cases, which the clients consider their property and therefore
not further explained.
Still we can provide them with the main cost drivers of the design, enabling
the client to evaluate a given Pareto optimal solution and come to a decision on
which one of the Pareto optimal solutions should be used in the next the design
stages.
In this case the total cost of ownership can be separated into building costs,
operational gains, and operational expenses. Building cost are largely driven by
the material cost and consequently can be directly linked to the weight of the
vessel. Operational gains can be defined as ship speed and moved cargo, which
are both fixed for this optimization case and operational cost is largely driven
by fuel consumption, maintenance and crewing cost. Minimizing the total cost
of ownership can thus be achieved by minimizing the steel weight while also
minimizing the resistance. These objectives are a classical example of conflicting
ones, a long and slender design variation will have a smaller resistance factor
and a higher estimated steel weight compared to a wider shorter variation of the
same vessel.
The resistance of the different design variations is estimated with a poten-
tial flow solver [10]. Because of the nature of a potential flow solver, the mesh
describing the hull shape has been idealized. The potential flow code cannot
provide absolute resistance values, but is suitable for comparing the resistance
of different design variations.
 Accelerated Concept Design 47

The steel weight is more complex since the light ship weight, maximum bend-
ing moments, and the loading conditions should be in balance with each other.
Therefore, for every design variation a minimum steel weight needs to be esti-
mated that meets the maximum bending moment requirement while keeping
Eqs. 1 and 2 in balance:

Δactual = L · B · T · Cb · ρ (1)

Δrequired = DW T + LSW (2)

where Δactual is Displacement, L is Length, B is Breadth, T is Draft, Cb is
Block-coefficient, and ρ is the water density, DW T is the Dead Weight, and
LSW is Light Ship Weight.

5 Experiments
As an experiment, the TSHD as described in Sect. 4 has been optimized by
the ACD framework. The original design has a resistance coefficient of 1.08
and a steel weight of 2039 tonnes. This original design is parameterized and
optimized by the ACD framework using only 200 evaluations. We only allow
the framework to do 200 evaluations to simulate a real world case where time is
typically very limited. The reference point is fixed and set to [5000, 2]. By setting
the reference point to [5000, 2] we disregard, and don’t allow the algorithm to find
solutions with a higher steel weight than 5000 tonnes, or with a higher resistance
coefficient than 2. The experiment has been conducted five times independently
with different initial starting points to check for consistency.
To show the severity of the constraints, an experiment was conducted to check
how much of the design space was feasible. To get an indication 200 random
design variations are generated and evaluated. 24% of these design variations
turned out to be feasible.

6 Results
The results of the five independent runs were very similar. The hypervolume
measure as used in multi-objective optimization [12] between the reference point
and the Pareto optimal set was on average 3819 and the standard deviation
of this volume was 3.3. The parameter combinations of a typical run of 200
evaluated design variations are displayed in the parallel coordinate plot in Fig. 5.
The red lines represent the obtained infeasible solutions, the blue lines represent
the feasible solutions, and green lines represent the Pareto optimal solutions.
The Pareto optimal results of a typical run are presented in Fig. 6. During
this run, the ACD framework found a set of 10 non-dominated design variations
where the most interesting solution has a resistance coefficient of 0.87 and a steel
weight of 1748 tonnes. Therefore, compared to the original design, the improved
design has a 19% smaller resistance coefficient and 14% less steel weight.
48 R. de Winter et al.

Fig. 5. Parallel coordinate plot of the 200 different design variations.

Fig. 6. Obtained non-dominated design variations and original design. The marked
solutions are the original solution and the most interesting optimized solution.

In Fig. 7 and Fig. 8 the original and the improved design are shown respec-
tively. From the first observation, there is not a lot of difference, but the opti-
mized result is 9 m longer, and 50 cm less wide. Typically when a ship is longer
more steel is needed to fulfill the strength requirements. But in this case, the
hopper is longer which eases the imposed longitudinal bending moment on the
 Accelerated Concept Design 49

ship. The extra strength results in less thick required steel plates and smaller
profiles required to meet the longitudinal strength.
Furthermore, because the vessel is longer and less wide, the resistance also
significantly decreased. This can also be seen from the wave pattern around the
two design variations in Fig. 7 and Fig. 8.

Fig. 7. Original trailing suction hop- Fig. 8. Trailing suction hopper dredger
per dredger design optimized by human design optimized with the accelerated
experts. concept design.

7 Conclusion and Outlook

In this paper we optimized a Trailing Suction Hopper Dredger with the Acceler-
ated Concept Design framework. The original design had a resistance coefficient
of 1.08 and a steel weight of 2038 tonnes. The most interesting obtained opti-
mized solution showed a 19% smaller resistance coefficient and a 14% smaller
steel weight.
We can conclude that the Accelerated Concept Design framework is capable
of optimizing parameterized vessels in a fully automated manner and in a very
efficient way. On top of this, it is shown that this design process can reduce
time and human effort while significantly improving ship designs. Furthermore,
because of the use of surrogate-assisted models, and therefore objective and
constrained prediction, the whole design space can be explored which would
never have been an option for a human expert alone.
For future work, C-Job Naval Architects aims to automate and connect more
software tools to the Accelerated Concept Design framework to gather more high
fidelity results. On top of this, a more in-depth analysis of ship parameterization
should be done to achieve even better results.
50 R. de Winter et al.

References
1. Evans, J.H.: Basic design concepts. In: 4th Journal of american society for Naval
Engineers, vol. 71, pp. 671–678 Blackwell Publishing Ltd (1959). https://doi.org/
10.1111/j.1559-3584.1959.tb01836.x
2. Papanikolaou, A., Harries, S., Wilken M., Zaraphonitis G.: Integrated design and
multiobjective optimization approach to ship design. In: Proceedings of Interna-
tional Conference on Computer Application in Shipbuilding, At Trieste. vol 3, p.
4. RINA, Royal Institution of Naval Architects. (2011)
3. De Winter, R., Van Stein, B., Dijkman, M.: Designing ships using constrained
multi-objective efficient global optimization. In: Proceedings of International Con-
ference on Machine Learning, Optimization, and Data Science. pp. 191–203. Spring
International Publishing. (2019). ISBN: 978-3-030-13709-0
4. Chai, T.Y.: Challenges of optimal control for plant-wide production processes in
terms control and optimization theories. Acta Automatica Sinica 35, 641–649
(2009)
5. Papanikolaou, A.: Holistic ship design optimization. Comput. Aided Des. 42(11),
1028–1044 (2010)
6. Sieger, D., Stefan, M., Mario, B.: A comprehensive comparison of shape deforma-
tion methods in evolutionary design optimization. In: Proceedings of the Interna-
tional Conference on Engineering Optimization. (2012)
7. Alonso, J.J., LeGresley, P., Pereyra, V.: Aircraft design optimization. Math. Com-
put. Simul. 79(6), 1948–1958 (2009)
8. Coquillart, S.: Extended free-form deformation: a sculpturing tool for 3D geometric
modeling. SIGGRAPH Comput. Graph. 24, 187–196 (1990)
9. Coello, C.A.C., Lamont, G.B., Van Veldhuizen, D.A.: Evolutionary algorithms for
solving multi-objective problems. Springer, New York (2007)
10. Tahara, Y., Stern, F., Himeno, Y.: Computational fluid dynamics-based optimiza-
tion of a surface combatant. J. Ship Res. 48(4), 273–287 (2004)
11. De Winter, R.: CEGO: Constrained Multi-Objective Efficient Global Optimiza-
tion.(2018) https://github.com/RoydeZomer/CEGO/
12. Beume, N., Naujoks, B., Emmerich, M.: Sms-emoa: multiobjective selection based
on dominated hypervolume. Eur. J. Oper. Res. 181(3), 1653–1669 (2007)
13. Simpson, T.W., et al.: Metamodels for computer-based engineering design: survey
and recommendations. Eng. Comput. 17(2), 129–150 (2001)
 A “Fast Track to Approval” Process
for Innovative Maritime Solutions

Stéphane Paboeuf1(&), Arnold de Bruijn2, Franz Evegren3,

Matthias Krause4, and Marcel Elenbaas5
1
Bureau Veritas Marine & Offshore, 8 Boulevard Albert Einstein,
44323 Nantes, France
[email protected]
2
Netherlands Maritime Technology, Boompjes 40,
3011 XB Rotterdam, The Netherlands
3
RISE Research Institutes of Sweden AB, Brinellgatan 4,
504 62 Borås, Sweden
4
Center of Maritime Technology, Bramfelder Straße 164,
22305 Hamburg, Germany
5
Damen Schelde Naval Shipbuilding, De Willem Ruysstraat 99,
4381 NK Vlissingen, The Netherlands

Abstract. RAMSSES (Realisation and Demonstration of Advanced Material

Solutions for Sustainable and Efficient Ships) is an Innovation Action project
partly funded by the European Commission in the framework of the H2020
program. During the 4 years of the project, 13 demonstrators will be developed
and realised to demonstrate the feasibility and efficiency of the introduction of
innovative materials in shipbuilding. Objectives of using new materials such as
composite materials and high tensile strength steel are to reduce the weight of
the vessel and to increase performance compared to when using conventional
material.
The role of the classification society Bureau Veritas Marine & Offshore in the
project RAMSSES is to help and support shipyards, designers, naval architects
and owners to reach the above objectives while achieving existing rules and
international regulations.
The paper gives an overview of current international regulations and espe-
cially the recent IMO (International Maritime Organization) interim guidelines
MSC.1/Circ.1574 concerning the use of Fibre Reinforced Plastic (FRP) ele-
ments within ship structures. The global survey scheme applied by classification
societies is described and the material approval process is detailed in the scope
of innovative material certification. An accelerated approval process called “Fast
Track to Approval” (FTA), developed in collaboration with RAMSSES partners
and leading to the certification of innovative solutions in ship construction is
presented in detail. The goal of the FTA is to have a rapid and efficient
methodology for the approbation of new designs or solutions involving
advanced materials. This process is based on risk analysis, including model and
demonstrator case tests performed during the RAMSSES project. Moreover, a
materials database has been obtained from mechanical and fire test results
carried out on innovative materials, which is utilized for the FTA. The FTA also
makes use of proposed standard risk scenarios, standard tests and standard

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 51–63, 2021.
https://doi.org/10.1007/978-981-15-4680-8_4
52 S. Paboeuf et al.

solutions, in addition to the database to look up reusable data. This will

accelerate the process for approval of innovative maritime solutions for sus-
tainable and efficient ships.

Keywords: Fast Track to Approval
Composite materials
Risk analysis

Certification

1 Introduction

Innovative and lightweight materials applied in ship structures have the potential to
improve life cycle performance and to reduce the environmental footprint significantly.
In recent years, considerable progress has been made in research and development,
leading to the first commercial application, such as the car carrier SIEM Cicero, built
by Uljanik shipyard in Croatia which is equipped with cargo deck panels made of
composite material. However, the use of advanced materials in the maritime sector is
lagging behind the potential. The Innovation Action RAMSSES, co-funded by the
European Union, is aiming to tackle the most relevant barriers that hinder a broader and
quicker technology uptake, thus to obtain recognition and an established role for
advanced materials in the maritime industry. As Fig. 1 illustrates, RAMSSES is acting
on three layers, each of them addressing at a particular strategic objective.

Fig. 1. Layer structure within the RAMSSES project and its main objectives (right).

The development and demonstration represents the bottom layer in the RAMSSES
pyramid and includes 13 demo cases RAMSSES. They cover the entire range of
innovative materials (metallic, non-metallic), ship types (from small work or leisure
boats to big cruise vessels), types of structures (ship hull, components not contributing
to global strength, major outfitting equipment) and volumes (custom made, standard-
ised structures). Furthermore, the project is developing all the production, assembly and
after sales processes which are required for successful technology introduction both in
shipyards and on board. Figure 2 shows a representation of a virtual ship in which all
13 demonstrators are implemented.
 A “Fast Track to Approval” Process for Innovative Maritime Solutions 53

Fig. 2. RAMSSES demonstration cases overview.

A series of test campaigns and life cycle analyses are performed in the assessment
layer, with the aim to show that the suggested solutions are ready to gain regulatory
approval and acceptance by customers. This readiness means that the innovative
material have passed a process defined in RAMSSES for class rules application and
alternative design for fire safety.
The paper at hand is concentrating on the top layer in Fig. 1, i.e. the integration and
communication of project results into the industry and into the regulatory community.
Particularly, this paper focuses on those activities which aim at developing and sug-
gesting measures towards regulatory frameworks that support a quicker exploitation of
new materials’ potentials, the so called “Fast Track to Approval”.

2 Motivations

There is a limited field history of FRP composite structures in commercial application

on SOLAS [1] ships. Approval of the materials from a mechanical perspective requires
a thorough testing procedure, and from a fire safety perspective it is not possible to
refer to the prescriptive regulations in SOLAS [1] since the material is combustible. It
is instead necessary to perform a risk assessment which demonstrates that equivalent
fire safety is achieved. These procedures need to be done case by case and are often
time consuming and costly. From a designer perspective, the general wish is a simpler
and more straight forward procedure, with examples of often acceptable solutions.
A guideline providing such recommendations would speed up the design and approval
process in several ways, making it shorter and more efficient. From the approving
authority perspective, it is also possible to improve the efficiency of the procedure and
there is a need to harmonize in competence and acceptance of different solutions.
Hence, “Fast Track to Approval” guidelines with recommendations on a more stan-
dardized alternative design and approval procedure for mechanical and fire safety,
including examples of promising solutions, would work on different levels to make the
54 S. Paboeuf et al.

design and approval of new material structure solutions less costly, less time con-
suming, more efficient, more harmonized and quality assured.

3 Global Survey Scheme

From class rules perspective on structural aspects, different schemes are generally used
by the classification societies to assess the design of ship constructions with composite
materials. All these schemes are typically based on the following steps of assessment:
1. Raw materials,
2. Structure design,
3. Coupon tests of FRP,
4. Manufacturing and inspection at works,
5. Final tests and inspections.
Each step of the global survey scheme is further described below.

3.1 Raw Materials Assessment

The first step of the global survey scheme is the assessment of raw materials used for
the construction of ships, based on the two following steps according to typical clas-
sification society rules:
• Raw material approval: carried out by tests to check the characteristics of the
material considered (resin, reinforcement, core materials…). These tests may be
performed by the raw manufacturer, or a testing laboratory accepted by the clas-
sification society and validated on the basis of the minimum characteristics defined
by the classification society or the raw manufacturer (these tests are generally based
on ISO standards or equivalent).
As a rule, after satisfactory completion of tests, a certificate is generally issued by
the classification society to the manufacturer for the raw materials tested. Required
mechanical tests for raw material assessment (sometimes called homologation) are
given in Bureau Veritas rules NR546 [2].
• Works approval: based on the quality system (complying with ISO 9000 or
equivalent) of the manufacturer and on recognition of the production and quality
control processes in order to demonstrate the ability to consistently manufacture the
raw materials according to the raw material approved. As a rule, upon satisfactory
completion of the process, a recognition certificate is issued by the classification
society to the manufacturer for the raw materials considered.
The process is more detailed in Bureau Veritas NR320 Certification Scheme of
Materials and Equipment for the classification of Marine Units [3].
 A “Fast Track to Approval” Process for Innovative Maritime Solutions 55

3.2 Structure Design Assessment

The second step of the global survey scheme is the hull structure design approval,
based on the hull structure rules of the classification society. These rules are generally
ordered as follow:
• Definition of structure design principle: including general requirements about the
hull structure general arrangements, definition of structure detail construction…)
• Definition of design loads to be used for the hull structure assessment: Global hull
girder loads and local loads
• Definition of structure calculation methodology: approach of structure computation
similar to those considered for other type of hull structure (steel or aluminum
structure for example)
• Definition of FRP approach for the scantling calculation based on:
– Theoretical methodology to apply to the FRP material analysis
– Breaking values reference characteristics of FRP material
– Scantling criteria: Based on safety coefficients in relation to the breaking stress
or strain values reference of the composite and the actual stresses or strains.
Different general basic approaches are considered in the rules of the classification
societies for the analysis of the composite materials:
• 1st approach: Approach based on the hypothesis that a composite laminate can be
considered as a homogeneous material having mechanical characteristics (mainly
young modulus and breaking stresses) defined taking into account the different
reinforcement layers in the laminate: According to this theory, the hull scantling
may be directly designed by thickness for plates and modulus for stiffeners.
In this case, the reference values of breaking stresses are considered as a global
laminate breaking stress.
• 2nd approach: Approach based on a ply by ply analysis, taking mainly into account
the compression and tensile stresses or strains in the direction of fibres of the
reinforcements.
In this case, the reference values of breaking stresses are compression and tensile
stresses of each layer.
• 3rd approach: Approach based on a ply by ply analysis taking into account the
stresses in the direction and perpendicular to fibres, and the interlaminar stresses
between layers.
In this case, the reference values of breaking stresses are compression and tensile
stresses in the direction of and perpendicular to fibres of layers, and interlaminar shear
stresses between each layer.
As a rule, longitudinal hull girder strength and local strength are examined inde-
pendently in the rules of the classification society. For large ship, it is reasonable to
consider the stress combination between the effects of the global hull structure loads
and the local loads. It should be noted that such combination is not possible with the
first approach defined here above.
56 S. Paboeuf et al.

The scantling check is based on the definition of safety factors equal to the ratio
between the theoretical breaking stress/strain defined by the different methodology and
the actual applied stresses/strains determine by the structure calculation. Even if it is
difficult to directly compare the values of the safety coefficients taken into account in
the structure rules of the different classification society (these values being dependent
of the values of loads considered, the type of stress considered in the laminate, the
theoretical breaking values considered…). Some classification society defined also a
criteria based on maximum deflection values of the structure (plate and stiffeners).

3.3 Coupon Tests of FRP

The third step of the global survey scheme is the mechanical and physicochemical tests
on laminate panel produced by the yard in charge of the hull construction. The aim of
these tests is to show that the mechanical characteristics of the laminates, produced
with raw materials and production process used for the hull construction are at least
equivalent to the theoretical mechanical characteristics considered during the structure
design assessment.
The main mechanical tests, carried out according to ISO standards or equivalent,
required by the classification society are:
• Tensile tests,
• Bending tests,
• Interlaminar shear tests,
• Measurement of density and content in fibre.
As a rule, the test panel may be taken from hull cut-outs or hull extension tabs or
may be produced with the same process of the hull construction (same materials, layup,
same process and cure system).
Specific process such as bond strength for the assembly of laminate or hull part may
be required to be tested, see Sect. 3.6.

3.4 Manufacturing and Inspection at Works

The fourth step of the global survey scheme is the manufacturing and inspection at
works. The assessment of the manufacturer’s arrangement for production aims to verify
that the hull construction is in compliance with the rules of the classification society.
As a rule, this assessment is based on:
• Periodical surveys carried out by a surveyor of the classification society during
construction,
• Quality system used by the yard to ensure the conformity of the composite con-
struction in relation with the hull structural drawings examined by the classification
society.
 A “Fast Track to Approval” Process for Innovative Maritime Solutions 57

3.5 Final Tests and Inspections

The final tests and inspection, within the scope of hull construction, are provided to
confirm:
• The watertightness of hull, tanks and watertight boundaries (carried out by
hydrostatic, hydropneumatic, air, hose testing or equivalent),
• The compliance of these tests and inspections with the rules of the classification
society and the approved conditions.
Non-destructive testing may be used during the final tests (ultra-sonic testing,
stereoscopy, radiography…).
Specimen tests are required on a case by case basis depending on the uncertainty.
This kind of test is especially required for bonding assembly, “non-standard” joining
technics, non-conventional design, new materials, etc.
In general, full scale tests are not required by the class. However, full scale tests can
be useful to validate a structure in the framework of a research project for a scientific
assessment. Full scale test may be required by the Risk Analysis.

3.6 Material Process Approval

The “classical process” for the testing of composite materials structure in shipbuilding
can be represented by a pyramid as indicated on the Fig. 3. The test pyramid is
composed on 4 levels:
1. Raw materials
2. Material coupon tests
3. Specimen tests
4. Full scale tests

Fig. 3. Composite material test pyramid.

58 S. Paboeuf et al.

In general, tests are required by the classification society in the scope of the
classification or certification. However, sometimes tests are also required by a risk
analysis, especially in the case of an alternative design study.

4 Fast Track to Approval

The RAMSSES project is elaborating and testing a new approach to achieve a “Fast
Track to Approval” (FTA) for innovative solutions, including new materials. This will
be fed into the maritime rule making processes in the form of common positions and
project guidelines. RAMSSES partners are also contributing to the development of
cross-industry guidelines e.g. for new materials.
The “Fast Track to Approval” is to be:
• Simple,
• Generic,
• Readable by shipyards, engineering, naval architects, …
• Applicable to all RAMSSES demonstrator cases.
The FTA principle is based on the experience gained during previous research or
commercial projects, by sharing knowledge with other industrial sectors, i.e. aero-
nautics, railway and automotive, by using existing database and by standardizing
results.
The FTA is composed of two layers: one led by the classifications societies and one
led by the shipyards, engineering or design offices, see Fig. 4. The FTA will be a first
step to prepare an Approval In Principle (AIP), leading to the certification or the
classification of a vessel or a part of the vessel. The FTA is developed in the
RAMSSES project to propose an alternative to the prescriptive rules concerning the
introduction of innovative materials in the shipbuilding industry. Figure 4 resumes the
process and gives the main steps of the FTA.

Fig. 4. Fast Track to Approval, process overview.

 A “Fast Track to Approval” Process for Innovative Maritime Solutions 59

The ambition is that the FTA will mainly be used for alternative design solutions
performed in accordance with the circular MSC.1/Circ.1455 [4], see Fig. 5. In the
context of RAMSSES, several risk analyses are carried out to demonstrate the feasi-
bility of new concepts, to validate different design options, to confirm the performance
of innovative materials and to prove equivalent fire safety compared to conventional
structures. All risk analyses will be supported by numerical simulations, such as Finite
Element Analysis, Computational Fluid Dynamics, evacuation simulations, thermo-
mechanics calculations, etc. Model or full scale test for the RAMSSES demonstrator
cases, see Fig. 2, will be used to verify the assumptions in the assessment.

Fig. 5. Design and approval process [4].

60 S. Paboeuf et al.

From the results of demonstrator case risk analyses, the FTA will propose standard
risk scenarios, standard tests, standard solutions and a database to lookup for reusable
data. The quantitative analyses are currently carried out on a case-by-case basis. The
objective aims to develop standard risk scenarios covering a range of similar appli-
cations. These can be referred to in the future without the need to carry out extensive
quantitative analyses or tests. The definition of standard risk scenarios will be based on
the demonstrator cases developed in RAMSSES.
The standard risk scenarios will cover the following aspects:
• Fire safety,
• Stability, including damage stability,
• Materials,
• Structural arrangement.
For example, when evaluating fire safety for FRP composite structures, the
encountered challenges are often similar. Standard risk scenarios developed in the
project cover several applications and include scenarios defining suitable reaction to
fire properties for external and internal surfaces as well as required fire resistance for
internal and loadbearing bulkheads. As an input to the fire risk assessment, significant
testing is often required to prove the functional properties. The second objective of the
FTA is therefore the creation of a materials database, named RAMSSES Knowledge
Repository, see Fig. 6. It is a web browser based platform on invitation access (https://
repository.ramsses.eu/). In addition to descriptions of the maritime materials, it will
define for example fire safety, mechanical and acoustic properties of materials tested
standardized tests. A large number of tests, mechanical and fire tests, will be performed
during the 4 years of the RAMSSES project on a multitude of materials, composites
and metallic. The database of test results and pre-approved solutions, to be developed
in RAMSSES, should avoid the necessity of repetitive tests if a simple qualitative risk
analysis shows that relevant results and solutions are already available. In addition,
numerical or statistical models will be developed during the project that may replace
certain physical testing in the future. The approach will be documented in a project
guideline, which may form the basis for future modified class rules. The feasibility of
this procedure will be demonstrated in a showcase, using a RAMSSES demo case as
reference, as presented in Sect. 5.
The experience and procedures developed in RAMSSES need to be fed into the rule
making process, with the final goal to implement the findings into new rules and
regulations in the medium to long term. On the short term RAMSSES could deliver
input to the evaluation of MSC.1/Circ.1574 [5]. Input to IMO and SOLAS needs to
involve flag states, which are represented in the IMO. This will be achieved through
consortium members working closely with their national authorities, like RISE
(Sweden), Bureau Veritas (France), DSNS and NMTF (Netherlands). In addition to
that, two other channels will be used: class societies, represented by IACS and ship-
yards represented by SEA Europe, have an advisory role as Non-Governmental
Organization at IMO. NMTF as a national shipbuilding association member of SEA
Europe and Bureau Veritas as a member of IACS will serve these channels as well.
 A “Fast Track to Approval” Process for Innovative Maritime Solutions 61

Fig. 6. RAMSSES Knowledge Repository platform (https://repository.ramsses.eu).

5 Shipyard Perspective

Today, as in many industries, there is a drive for more durable solutions, CO2
reduction, reduction of lifecycle costs and reduction of operational costs. In the
transportation industry this is almost always is related to weight reduction (either over-
all weight, or lower center of gravity). Composites offer weight saving as compared to
traditional metallic solutions and reduce operational costs. Also composites could
prolong lifetime as these structures do not corrode and, provided it is designed prop-
erly, can have a prolonged fatigue life.
For these benefits, composites have been a focus point for future ship structures for
a long time. Though already commonly applied in leisure and fast passenger craft, the
applications in large blue water vessels is lacking behind. The benefit from weight
saving only really becomes evident if larger structures are being replaced from metal to
composites. Then weight saving can result in a positive design cycle leading to less
power required, leading to smaller and lighter propulsion further decreasing the weight,
or allows adding for another key functionality set by our customers. In those cases
composite structures are integrated properly in the overall design, offering a compet-
itive advantage to our customers in Capex as well as Opex. Downside however, is that
these larger integrated composite solutions, inevitably effect the overall design and
need to be approved by classification societies. Despite significant improvements in the
last decade, especially for fire safety, the legislation is still unclear, other than that the
overall safety should be demonstrated at least to be equivalent to the original steel
design (Fig. 7).
This is where the importance of a fast track to approval comes in place. Yards can
only offer large one-off capital products into the market for which the design is
approved, or at least clear evident is available that the product, with integrated com-
posite structure, can be approved within a well-defined functionality, cost and lead-time
62 S. Paboeuf et al.

Fig. 7. Damen demonstrator case used for the FTA application.

as set in the contract between yard and customer. The design appraisal should therefore
be set prior to contract, giving time and cost pressure on the approval trajectory. This
illustrates the importance of the “Fast Track to Approval”.
The Damen involvement in various EU research projects (De-Light, BESST,
RAMSSES, and Qualify), and real commercial applications such as the composite
waterbus, gives Damen knowledge and confidence that composites can be applied
sensibly, and ensuring safety, at least equivalent to metallic structures. Recommen-
dations from a shipyard perspective on how to fasten this process, especially prior to
contract, are reflected in this paper. These may lead to recommendations what should
be set prior to contract, and which issues can be dealt with after contract.

6 Conclusion

Any product design, material or production process in the maritime sector is subject to
approval by supervising authorities, mostly based on prescriptive rules and regulations.
This may partly be attributed to a certain conservative attitude in the sector. However,
this approach is duly justified by the extreme operational conditions and long life times
under which maritime products need to survive, the difficulty of human emergency
interaction at sea, and the potentially extreme consequences of the failure of maritime
products on human life, maritime assets and the environment.
The proposed “Fast Track to Approval” respects the need for a thorough evaluation
of safety of the design while avoiding the necessity of repetitive tests if a simplified
qualitative risk analysis shows that relevant results and solutions are already available
in the database. The intention of RAMSSES is to define and apply a smart and efficient
approach while never compromising safety.
 A “Fast Track to Approval” Process for Innovative Maritime Solutions 63

During the next phase of RAMSSES project, this approach will be applied on the
demonstrator case developed by DAMEN in order to demonstrate its feasibility and
efficiency.

References
1. SOLAS - International Convention for the Safety of Life at Sea
2. Bureau Veritas NR546, Hull in Composite Materials and Plywood, Material Approval,
Design Principles, Construction and Survey
3. Bureau Veritas NR320, Certification Scheme of Materials and Equipment for the classifi-
cation of Marine Units
4. MSC.1/Circ.1455 Guidelines for the approval of alternatives and equivalents as provided for
in various IMO instruments
5. MSC.1/Circ.1574 Interim guidelines for use of Fibre Reinforced Plastic (FRP) elements
within ship structures: Fire safety issues
 A Modified Design Framework Based
on Markov Decision Process for Operational
Evaluation

Hao Yuan(&) and David J. Singer

University of Michigan, Ann Arbor, MI 48109, USA

{haoyuanl,djsinger}@umich.edu

Abstract. It is well known that operational evaluation is an indispensable

component of the ship design process. This paper outlines the development of an
extension to the previously developed Ship-Centric Markov Decision Process
(SC-MDP) framework to include the identification, evaluation and impact of
seakeeping performance on operating vessels within early stage design. The
results from previous SC-MDP development efforts at the University of
Michigan validate the success and applicability of this framework towards
furthering the evaluation of ship temporal operational events in early stage
design. As an extension, the SC-MDP framework modification presented in this
paper will provide a mechanism for the evaluation of the performance of a
vessel operating across the ocean through the consideration of physical
response, which is limited when using traditional seakeeping and Computational
Fluid Dynamics methods. Traditional seakeeping analysis in the frequency
domain analyzes the physical response of the vessel from a static perspective,
though this perspective lacks the ability to evaluate the vessel in terms of its
temporal behaviors. Additionally, the analysis in the time domain is limited by
short time frames and high computational expenses. Therefore, the opportunity
to incorporate the implications of the vessel’s physical response from a temporal
perspective can provide great value if the constraints present in early stage
design are mitigated. The formulation presented in this paper extends the SC-
MDP framework to include the seakeeping performance of a vessel in early
stage design context.

Keywords: Ship-Centric Markov Decision Process
Ship preliminary design

Seakeeping response
Operational evaluation

1 Introduction

Seakeeping performance refers to the ability of a ship to successfully operate at various

sea states [1]. The seakeeping performance of a vessel is a critical topic of consider-
ation when evaluating ship operational behaviors. From a design perspective, it is
essential to not only analyze seakeeping responses, but to also investigate the rela-
tionships between seakeeping responses and other operational aspects. Therefore,
establishing a framework that analyzes such interdependencies within the ship system
significantly benefits the designers during preliminary design stages [2].

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 64–79, 2021.
https://doi.org/10.1007/978-981-15-4680-8_5
 A Modified Design Framework Based on MDP for Operational Evaluation 65

A critical component of seakeeping analysis is ship motion prediction. Historically,

a considerable number of seakeeping studies have been focused on this subject. Sayli
et al. [3] took advantage of multiple regression models in order to analyze the rela-
tionships between seakeeping responses and hull parameters of Mediterranean fishing
vessels. This is an example of analyzing statistical methods for a single vessel type. In
terms of the theoretical improvement of seakeeping predictions, research has been
performed in both the time and frequency domains. A JAVA program has been
developed by Augustine et al. [4] to estimate coupled heave and pitch motions in the
time domain. Seakeeping Prediction Program (SPP) [5] implements a version of
SCORES Program [6] for the approximation of heave, pitch and roll motions in the
frequency domain. The advantage of these programs is the reliable response estima-
tions for different circumstances that they provide for designers. However, one dis-
advantage of these estimations is the poor representation of the interconnections
between vessel motions and temporal operational behaviors explicitly, especially when
taking into consideration fickle wave conditions worldwide. As a result, an operational
evaluation framework which can address such disadvantages is necessary in the con-
tinued analysis.
Ship-Centric Markov Decision Process (SC-MDP) framework provides the
opportunity to analyze operational evaluation in the preliminary design stage in a novel
way. The information generated by the SC-MDP framework serves as a unique ref-
erence to identify lifecycle decision path dependencies and relate them back to early
stage of ship design, which provides the design data needed to avoid design lock-in [7].
SC-MDP was originally developed by Niese and Singer [8] to study the lifecycle
decision process for the ballast water treatment of a ship when taking into consideration
potential environmental policy changes. Niese and Singer [9] advanced the process by
defining a set of metrics to evaluate changeability of a design based on their previous
SC-MDP results. Kana et al. [10] applied eigenvalue spectral analysis to the original
SC-MDP for lifecycle compliance of ballast water treatment. The work in these papers
demonstrates the applicability and flexibility of SC-MDP in evaluating ship operational
events. Current research developments of the SC-MDP framework concentrate on the
various impacts of environmental policies, economic benefits, and safety factors [11]
during the ship design process. However, the influence of seakeeping performance on
vessel design has not been well-developed yet. Ship motions will impact and possibly
reduce the reliability of operational behaviors and cause uncertain results, therefore it is
critical to incorporate this impact into the SC-MDP framework.
The purpose of this paper is to present the formulation of a modified SC-MDP
framework which takes into consideration the impact of seakeeping performance on
vessel transit. A commercial ship design for passage over the North Pacific Ocean is
discussed as a case study to illustrate this formulation. The main efforts of this paper
have been divided into two components. First, an appropriate seakeeping calculation
method is required to conduct ship motion predictions. These calculation results will
determine the values of transition probability and reward in the Markov Decision
Process (MDP) structure according to some transformation rules defined by designers.
Second, the MDP framework is defined as a surrogate model to emulate vessel sailing.
The inherent relationship between physical responses and operating decisions is
embedded within the MDP structure. The following sections outline the seakeeping and
66 H. Yuan and D. J. Singer

MDP methodologies, the results of the commercial ship design case study, and future
work to be performed.

2 Method

2.1 Background
Due to the constant increase in international trade, a more advanced design method of
commercial ships is required to maintain relevancy in such a competitive industry. One
of the most basic missions of a commercial ship is to successfully reach its destination.
Without taking into consideration the physical responses of the sea state, it is a
desirable decision to take the shortest trade route to complete the vessel’s mission.
Unfortunately, these responses are inevitable as long as the ship operates in water. As a
consequence, alternate routes that deviate from the shortest trade route must be con-
sidered due to the probable existence of adverse ship motions.
The ultimate value of this SC-MDP framework is to differentiate multiple ship
designs through the evaluation of the trade-offs between seakeeping responses and
operational performance with limited information provided in the preliminary design
stage. Currently, the work in this paper accomplishes the formulation of offering
rational results for a single ship design. In the future, the differentiation of various ship
designs will be realized based on the utilization of those results. The case study of the
single ship design is generated according to such background and the following
methods are selected for the appropriate evaluation of seakeeping responses and
operational performance.

2.2 Seakeeping Prediction Program (SPP)

SPP, a University of Michigan seakeeping software, applies strip theory and long-
crested wave assumptions in the frequency domain [5]. This program offers relatively
reliable estimations of seakeeping responses given only the main hull parameters,
which makes it an effective method during the preliminary stage. It utilizes hull
approximation, ship speed, ship heading relative to the waves, and sea spectrum data as
inputs and generates the response spectrum of heave, pitch, and roll.
The main parameters required to define a hull include waterline length (LWL),
beam (B), and draft (T), in addition to some ship-form coefficients such as prismatic
coefficient (CP), maximum section coefficient (CX), and waterplane coefficient (Cwp).
These coefficients are necessary for approximating the sectional area curve and
waterline curve of the vessel using a mathematical model in the program. Furthermore,
SPP applies Lewis Form to model hull sections at each station. These procedures
produce a rough form based on the provided hull parameters.
Ship speed, in this case study, is a constant value, meaning that variations in speed
throughout the vessel’s operation are not taken into consideration. The variable factor
of the utilized vessel is its direction during the transit, resulting in various values of
ship headings relative to the waves during temporal steps.
 A Modified Design Framework Based on MDP for Operational Evaluation 67

The sea spectrum is used to represent the real irregular waves in which the vessel is
operating. For the purpose of calculating responses during trans-ocean operations, the
ISSC spectrum has been utilized to characterize the fully developed wave conditions.
The ISSC spectrum is a two-parameter spectrum which depends on significant wave
height in meters (Hs) and the mean wave period in seconds (T), whose formula is
shown in Eq. (1) [12]:

173:6Hs2 T 4

Sw ðxÞ ¼ exp 694:4T 4 x4 ð1Þ
x5
where x is the wave frequency in radians per second.
In combination with the selected hull parameters, the sea spectrum Sw ðxÞ can be
converted into the encounter frequency spectrum Sw ðxE Þ which the ship will actually
experience. Furthermore, the energy spectrum Smotion ðxE Þ for any given motion can be
obtained through Eq. (2):

Smotion ðxE Þ ¼ ½RAOmotion ðxE Þ2 Sw ðxE Þ ð2Þ

where xE is the encounter frequency in radians per second and RAOmotion ðxE Þ is the
response amplitude operator for the corresponding motion.
From these equations, numerous values can be obtained. These values are the
statistical parameters that can be found from the heave, pitch, and roll motion energy
spectrum. The values include the root mean square, the average of the 1/3 highest
motion response and the average of the 1/10 highest motion response which can be
calculated for heave, pitch, and roll spectrum. The average of 1/10 highest values
indicate the extreme motion amplitude, which are extracted from SPP for use in MDP
framework in order to learn the maximum trade-offs between physical responses and
deviations away from the shortest route during the operational mission.

2.3 Markov Decision Process (MDP)

MDP focuses on sequential decision-making problems in regard to stochastic condi-
tions. The outcomes of this framework are influenced by both human control and
uncertainties, while including current and future opportunities. A stationary and fully
observable MDP is a 4-tuple <S, A, P, R>, where:
• S, is the set of states that represent the system;
• A, is the set of actions that can be executed in each state;
• Pðs0 js; aÞ; is the transition probability of achieving state s’ from state s through the
execution of action a;
• R(s, a), is the reward obtained due to taking action a in state s, a discount factor c
may be introduced to count for the preference of current reward over future reward.
The MDP outputs an optimal policy which yields the largest cumulative reward
defined as the Bellman equation [13]. In this case study, the optimal policy is solved
using a value iteration algorithm [14].
68 H. Yuan and D. J. Singer

State. Different locations over the ocean are modelled as states in the MDP frame-
work. The wave system in the North Pacific Ocean is selected as an example of the
operating environment. To be more specific, the latitude range is designated from 0 to
60N and the longitude range is designated from 120E to 104W (256E). The overall
range is discretized to distributed points with respect to certain latitude and longitude
intervals. These points at varying latitudes and longitudes are the states for which the
ship is in transit. Figure 1 displays a map including all states when intervals are 5° and
8° along latitude and longitude respectively. For example, locations M and N are two
samples of state representations.

Fig. 1. A map representation of states in the case study

Each state contains relevant wave information including significant wave height,
mean wave period, and mean wave direction. A public dataset from ECMWF [15] has
been utilized to obtain these values. In order to analyze the worst-case scenario, a
month containing large significant wave heights and long mean wave periods is pre-
ferred for use in this case study. Figure 2 summarizes the rough variation of significant
wave height.

Fig. 2. A demonstration of significant wave height distribution from ECMWF dataset

 A Modified Design Framework Based on MDP for Operational Evaluation 69

Figure 3 and Fig. 4 represent the extracted values of mean wave period and mean
wave direction which occurred in the selected month.

Fig. 3. A demonstration of mean wave period distribution from ECMWF dataset

Action. At each state in Fig. 1, a ship is assumed to travel to the surrounding states
through 8 different actions. In other words, the ship is allowed to continue the operation
in 8 azimuths approximately. Therefore, there are 8 elements in the set of action A,
each of which is recorded by an action index in Table 1.
Performing different actions in a certain state requires operation through each wave
environment at varying relative headings, which will ultimately cause diverse ship
motions. Such motions depreciate the original expectation of an action, making it more
challenging to execute the corresponding action in addition to making it less desirable
to select that action than one in a circumstance without seakeeping.
Transition Probability. The transition probabilities are assumed to be affected by the
roll responses of the vessel. Intuitively, even though the roll responses in all 8 direc-
tions may not exceed normative seakeeping criteria, the different degrees of difficulties
that they will impose on a vessel in transit must be taken into consideration. The target
of a certain action is to successfully operate to the next location in its direction.
Though, the roll responses may decrease the probability of success of maintaining the
required angle and reaching the target. These probabilities are written as
Pðtargetjcurrent; aÞ where P is the probability of arriving at the expected target in the
corresponding direction through the execution of action a at the current state. Other-
wise, failure of an action a means that the ship stays in place at the current state, with a
probability written as Pðcurrentjcurrent; aÞ. Therefore, the following transformation
rules demonstrated in Table 2 are assumed in this case to distinguish the impact of
different ranges of roll motions on determining transition probabilities.
Utilizing states M and N in Fig. 1 as an example, if the roll response at state M to go
north equals 3.5°, then PðNjM; northÞ will be 0.7 and PðMjM; northÞ will be 0.3 in the
transition matrix. In order to achieve the goal of obtaining optimal policy without roll
impact using the same MDP structure, the successful probability Pðtargetjcurrent; aÞ is
always assumed to be 0.99 as a dummy value in calculations.
70 H. Yuan and D. J. Singer

Fig. 4. A demonstration of mean wave direction distribution from ECMWF dataset

Table 1. Summary of actions

Action index Action Action index Action
1 North 5 South
2 Northeast 6 Southwest
3 East 7 West
4 Southeast 8 Northwest

Table 2. Transition probability rules

Roll response range Pðtargetjcurrent; aÞ Pðcurrentjcurrent; aÞ
roll < 2° 0.9 0.1
2°  roll < 4° 0.7 0.3
4°  roll < 6° 0.5 0.5
6°  roll < 8° 0.3 0.7
roll  8° 0.1 0.9

Reward. For a voyage across the North Pacific Ocean, the strategy of this operation is
assumed to follow the great-circle azimuth as close as possible at each step. The great
circle of two points on the earth directs the shortest route, which is used as a reference
of operational efficiency in the case study when the ship speed remains constant.
Reward is formulated in Eq. (3), which reflects the deviation from that great circle:

Rðs; aÞ ¼ juðs; destinationÞ  uðaÞj ð3Þ

where uðs; destinationÞ is the great-circle azimuth between the current state s and the
destination; uðaÞ is the azimuth associated with the action.
Logically, an action that makes the ship closer to great circle will generate smaller
absolute deviation and a larger reward in return due to the negative sign. What’s more,
 A Modified Design Framework Based on MDP for Operational Evaluation 71

traveling at certain deviations means experiencing the related pitch motions in the
seaway. It is possible that a large pitch motion will worsen the previous reward cal-
culated by Eq. (3). If the impact of pitch motions on depreciating reward R(s, a) is
added, Eq. (3) will be transformed to Eq. (4) with a depreciation amplifier factor f,

Rðs; aÞ ¼ juðs; destinationÞ  uðaÞj  f ðhÞ ð4Þ

where f ðhÞ is a function of pitch motion h, using f ðhÞ ¼ eh in this case.

Decision Criteria. The objective of the SC-MDP framework is to find an optimal
policy with the least accumulative deviation from the great circle at all steps. The
optimal policy can be solved via the Bellman equation, shown in Eq. (5):
X
UðsÞ ¼ max ½Rðs; aÞ þ s0
cPðs0 js; aÞU ðs0 Þ ð5Þ
a2AðsÞ

where U(s) is the expected utility of state s and c is the discount factor where c = 1.
Equation (6) gives the expression of the optimal policy. Some insights and impli-
cations can be gained through further analysis of these results.
X
pðsÞ ¼ arg max ½Rðs; aÞ þ s0
cPðs0 js; aÞU ðs0 Þ ð6Þ
a2AðsÞ

3 Result

The ship design sample parameters which are imported into the SC-MDP framework
are shown in Table 3. It is possible for this ship to take any action at each state where
the probable seakeeping responses are summarized in Fig. 5. The horizontal axis in
Fig. 5 is the action index, each of which represents one of the actions mentioned in
Table 1. And the vertical axis stands for the responses in degrees. Overall, the range of
roll motions that can be experienced equate up to 12° approximately, and the maximum
value of pitch motions is roughly 3.2°. As mentioned previously, these responses are
only static values and their influence can be seen in the SC-MDP.

Table 3. Hull parameters of the design sample

Parameter Value
LWL (m) 185
B (m) 32.2
T (m) 9.5
Displacement (tonnes) 35624
Speed (knots) 20
72 H. Yuan and D. J. Singer

Fig. 5. Scatter plot of roll and pitch responses over the North Pacific Ocean

There are two scenarios that exemplify the application of the methods, which
provide a new way to identify vessel operational performance in the preliminary stage.
Two parameters are applied to describe the main characteristics of vessel operational
behaviors with respect to physical responses.
On one hand, the introduction of seakeeping responses will change some optimal
policies that are desired in a circumstance without seakeeping. The percentage of such
changes reflects the major range of seakeeping impact over the ocean. In the case study,
this percentage parameter refers to the ratio of the number of states that show different
optimal policies between seakeeping and non-seakeeping situations to the total number
of states located in the sea area, written as P(policy change). A vessel with a smaller P
(policy change) is more capable of handling the predefined wave conditions when
emulating transits in this framework, which indicates a potentially preferable ship
design for the operation.
On the other hand, the overall deterioration of operational performance is
demonstrated in not only the change of optimal policies, but also the increase in
accumulative deviation cost during transits. In this framework, the utility value from
Eq. (5) represents such accumulative deviations at each state, so it is helpful to com-
pare the two sets of utility values between seakeeping circumstances and non-
seakeeping circumstances to learn the differences caused by physical responses. The
second parameter, namely E(utility difference), which is defined as the average of all
the differences between those two sets of utility values, is calculated as another
quantification of the deterioration of operational efficiency. In the early stage evalua-
tion, ship designs with small values of E(utility difference) are anticipated to perform
efficiently with seakeeping impact.
The two parameters above, P(policy change) and E(utility difference), serve as
unique references to reflect the degree of impact from seakeeping responses. If multiple
vessel designs are analyzed in the preliminary stage, vessels that yield smaller results of
these parameters will be distinguished from others for better operational performance.
As a start point, this paper only tests the simulation of a single ship design to explain
the rationality of the results. Moreover, the future utilization and modification of these
 A Modified Design Framework Based on MDP for Operational Evaluation 73

results to realize differentiation of vessel designs are discussed at the end of this
section.

3.1 Scenario 1 (Destination Location: 45N 128W)

In scenario 1, the destination is set as 45N 128W (232E), which can be regarded as a
representation of a generic United States western ports. In order to traverse the North
Pacific Ocean, the vessel mainly operates in the direction from west to east. Compared
with the optimal policy without seakeeping responses (black arrows), the impact of roll
and pitch individually is highlighted in bold red arrows in Fig. 6 and Fig. 7.

Fig. 6. The impact of roll response on optimal policy to destination 1

Table 4. MDP results with roll impact (destination 1)

Parameter Value
Number of states located in the sea area 186
Number of states with different optimal policies 19
P(policy change) 10.22%
E(utility difference) (unit: degrees) 27.99

In order to evaluate the impact of roll responses, values of P(policy change) and E
(utility difference) are summarized in Table 4.
The black arrows in the Fig. 6 denote the optimal policy without seakeeping
consideration. Except for several states that are constrained by land area, the arrows in
the open sea area follow the tendency of great-circle strategy. A bold red arrow denotes
an optimal policy change different from the non-seakeeping result, which indicates the
impact of the related motion. Two main changes are displayed in Fig. 6, which reflect a
logical relationship between seakeeping responses and operational transit.
First, some optimal policies between 35N and 45N are converted to east instead of
northeast. According to the wave environment shown in Fig. 2 and Fig. 4, action
74 H. Yuan and D. J. Singer

Fig. 7. The impact of pitch response on optimal policy to destination 1

northeast meets beam seas in large significant wave heights resulting in extreme roll
angles from 8° to 11°. Such large roll responses are associated with high probabilities
of maintaining station-keeping, while action east means experiencing roll motions at
around 1 or 2° and accumulating the second smallest deviation from the shortest route.
Therefore, this can be regarded as a typical change to avoid large roll motions in bad
wave conditions.
Second, starting near latitude 20N and longitude 208E when operating to the
destination location, the optimal policy changes from northeast to north. The envi-
ronment conditions are quite mild along this diagonal line, and most roll responses
caused by action northeast and north both approximate to 5°. In other words, these
changes are not a result of avoiding large roll responses. In fact, the introduction of
uncertainty depreciates the utilities of states along the northeast direction more than
those along the north direction. As a result, it is worthy to take an action to get away
from the northeast route. In general, these changes derive from roll motions implicitly,
which are difficult to identify without the help of such framework.
Similarly, in order to evaluate the impact of pitch responses, values of P(policy
change) and E(utility difference) are summarized in the following Table 5.

Table 5. MDP results with pitch impact (destination 1)

Parameter Value
Number of states located in the sea area 186
Number of states with different optimal policies 46
P(policy change) 24.73%
E(utility difference) (unit: degrees) 225.64

The changes in Fig. 7 are mainly distributed in the area from 5N to 20N. These new
policies manifest relevant operational behaviors of a ship to adapt to pitch motions. In
this area, a ship is more likely to encounter head seas with the action northeast, which
 A Modified Design Framework Based on MDP for Operational Evaluation 75

may cause the consecutive negative effects. Firstly, an action that ends up in experi-
encing head seas or following seas leads to larger pitch angles than that of the other
actions. Secondly, those large angles are exponentially exaggerated based on the
assumptive amplifier function f(h) in reward calculation. Thirdly, another action is
chosen to avoid either the unsatisfactory reward at the immediate step or the summation
of such rewards at future steps.
Moreover, the impact of both roll and pitch motions can be incorporated at the same
time, which is depicted in Table 6 and Fig. 8. It demonstrates similar changes to Fig. 6
and Fig. 7, but it is not a simple addition of red arrows in those figures. It is a new
round of interaction based on this framework with 36.02% optimal policy change and
359.78° change in utilities on average.

Table 6. MDP results with roll and pitch impact (destination 1)

Parameter Value
Number of states located in the sea area 186
Number of states with different optimal policies 67
P(policy change) 36.02%
E(utility difference) (unit: degrees) 359.78

3.2 Scenario 2 (Destination Location: 30N 128E)

The second destination is assigned at 30N 128E, a representative of ports in China,
which will motivate the transit across the North Pacific Ocean in the opposite direction.
Figure 9 through Fig. 11 represent the corresponding optimal policies of scenario 2
(Fig. 10).
It is apparent that there are few changes under the impact of roll responses but
significant changes due to the pitch responses in scenario 2. The degrees of change are
primarily determined using the initial definitions in transition probabilities and rewards.
The transition probability Pðtargetjcurrent; aÞ decreases gradually with a broad range of
roll angles in this case study, making the optimal policy more sensitive to extreme roll
responses and less sensitive to small ones. The reward R(s, a) is exponentially dete-
riorated by pitch responses in this case study, therefore the optimal policies have been
modified starting at small pitch angles.
The results in scenario 1 and scenario 2 are accurate reflection of model settings in
this specific case study, which are appropriately used as a proof of the formulation. The
operational changes under different circumstances are summarized in total in Table 7
and Table 8 for reference.
These results describe the operational performance of a ship design in the early
stage. They can be used as elementary information to evaluate vessels that operate
through seaway. However, they are not convincing enough to conduct meaningful
comparisons of multiple ship designs. It is essential to expand on these results in the
following two aspects to support further ship differentiations.
76 H. Yuan and D. J. Singer

Fig. 8. The impact of both roll and pitch on optimal policy to destination 1

Fig. 9. The impact of roll response on optimal policy to destination 2

Fig. 10. The impact of pitch response on optimal policy to destination 2

 A Modified Design Framework Based on MDP for Operational Evaluation 77

Fig. 11. The impact of both roll and pitch on optimal policy to destination 2

Table 7. Summary of P(policy change) due to roll, pitch, roll and pitch together
Scenario Destination Roll impact Pitch impact Roll & pitch impact
1 45N 128W 10.22% 24.73% 36.02%
2 30N 128E 4.84% 29.03% 25.81%

Table 8. Summary of E(utility difference) due to roll, pitch, roll and pitch together (in degrees)
Scenario Destination Roll impact Pitch impact Roll & pitch impact
1 45N 128W 27.99 225.64 359.78
2 30N 128E 11.53 235.97 366.98

– Two scenarios have been used as examples in the case study. If this framework is
activated by random destinations over the predefined area, designers will obtain
more general statistical values of the relevant parameters for the evaluation and
differentiation of ship designs.
– This framework takes advantage of designers’ subjective opinions in defining
transformation rules of physical responses. Other than the given definitions in this
case, designers should adjust the definitions based on their own requirements to
compare different ship designs.

4 Conclusion and Future Work

The authors propose a formulation of an extension to the SC-MDP framework to

incorporate the impact of physical responses on transits across the ocean. A case study
has been discussed in this paper to prove the feasibility of generating rational logic and
results from this formulation. The influence of seakeeping responses has been explored
78 H. Yuan and D. J. Singer

over the entirety of the North Pacific Ocean utilizing the combination of SPP and MDP,
which demonstrate a possible method to process the information constraints in the
preliminary stage. However, this framework is simply an elementary formulation in
order to prove the mechanism of introducing appropriate seakeeping tools to the SC-
MDP framework. There are simplifications and assumptions that restrict its application
for more problems. Consequently, significant future work is required to improve this
framework.
Further improvements should be made on the SC-MDP framework in order to
expand its access to other seakeeping parameters, which will broaden the framework’s
overall inclusion of physical information. Additionally, the transformation rules of
adding seakeeping components to MDP are anticipated to enable sensitivity analysis in
the future. New functions will likely be included within this framework to support
operational evaluation for ships in the preliminary stage. In reality, seakeeping analysis
is related to far more than roll and pitch motions, it is a significant component
impacting the overall feasibility of a vessel. Therefore, vessels must be designed to
successfully operate in diverse circumstances.

Acknowledgement. Hao Yuan’s work was supported by the CSC scholarship from China
Scholarship Council affiliated with the Ministry of Education in China. The authors would like to
offer special thanks for their support.

References
1. Molland, A.: Chapter 7 - seakeeping. In: The Maritime Engineering Reference Book,
pp 483–577. Butterworth-Heinemann (2008)
2. Brefort, D., Shields, C., Habben Jansen, A., Duchateau, E., Pawling, R., Droste, K., Jasper,
T., Sypniewski, M., Goodrum, C., Parsons, M.A., Kara, M.Y., Roth, M., Singer, D.J.,
Andrews, D., Hopman, H., Brown, A., Kana, A.A.: An architectural framework for
distributed naval ship systems. Ocean Eng. 147, 375–385 (2018). https://doi.org/10.1016/j.
oceaneng.2017.10.028
3. Sayli, A., Alkan, A.D., Nabergoj, R., Uysal, A.O.: Seakeeping assessment of fishing vessels
in conceptual design stage. Ocean Eng. 34, 724–738 (2007). https://doi.org/10.1016/j.
oceaneng.2006.05.003
4. Augustine, E.A., Emmanuel, D.I., Ezebuchi, A., Ibitoru, D.: Development of preliminary
ship motion prediction tool for coupled heave and pitch. Am. J. Eng. Res. 7, 195–204 (2018)
5. Parsons, M., Li, J., Singer, D.J.: Michigan Conceptual Ship Design Software Environment -
User’s Manuals (2012)
6. Raff, A.I.: Program SCORES - Ship Structural Response in Waves (1972)
7. Niese, N.D., Kana, A.A., Singer, D.J.: Ship design evaluation subject to carbon emission
policymaking using a Markov decision process framework. Ocean Eng. 106, 371–385
(2015). https://doi.org/10.1016/j.oceaneng.2015.06.042
8. Niese, N.D., Singer, D.J.: Strategic life cycle decision-making for the management of
complex Systems subject to uncertain environmental policy. Ocean Eng. 72, 365–374
(2013). https://doi.org/10.1016/j.oceaneng.2013.07.020
9. Niese, N.D., Singer, D.J.: Assessing changeability under uncertain exogenous disturbance.
Res. Eng. Des. 25, 241–258 (2014). https://doi.org/10.1007/s00163-014-0177-5
 A Modified Design Framework Based on MDP for Operational Evaluation 79

10. Kana, A., Singer, D.J., Kana, A.A., Brefort, D.C., Seyffert, H.C., Singer, D.J.: A decision-
making framework for planning lifecycle ballast water treatment compliance. In: PRADS
2016, Copenhagen, Denmark (2016)
11. Kana, A.A., Droste, K.: An early-stage design model for estimating ship evacuation patterns
using the ship-centric Markov decision process. J. Eng. Marit. Environ. 233, 138–149
(2017). https://doi.org/10.1177/1475090217720003
12. Parsons, M.G.: University of Michigan NA470/NA570 Informal Course Notes (2018)
13. Bellman, R.: Dynamic Programming. Princeton University Press, Princeton (1957)
14. Russell, S.J., Norvig, P.: Artificial Intelligence: A Modern Approach, 2nd edn. Prentice Hall,
Upper Saddle River (2003)
15. European Center for Medium-Range Weather Forecasts Homepage. https://www.ecmwf.int/
en/forecasts/datasets/browse-reanalysis-datasets
 IGC Code-Based LNG Filling Limits
Computation Using 3D CAD Model

Seok-Ho Byun1(&), Woo-Sung Kil1, Myeong-Jo Son1,

Jeong-Youl Lee1, and Mi-Ho Park2
1
ICT Solution Team, R&D Division, Korean Register,
Busan, Republic of Korea
[email protected]
2
Environment and Piping Team, Technical Division, Korean Register,
Busan, Republic of Korea

Abstract. In order to increase the safety of reduced volume of LNG (Liquefied

Natural Gas) at the extremely low temperature, the IGC code establishes
international standards. IGC code suggested with the filling limits which is a rate
of maximum LNG by total volume of tank to prevent the accident by LNG at a
sealed tank. The filling limits of LNG storage tank was calculated with the safety
factors. The calculation progress of the filling limits is very complicated and
time consuming because the tank volume and LNG volume for several situations
must be calculated under various factors. The objective of this study is to
develop the algorithm of filling limit calculation based on the 3D CAD model
according to 2018 IGC Code. The developed algorithm carries out applying a
“scenario in which isolated vapor pockets are not generated in cargo tank”
which required by new IGC as well as a “scenario where the pressure relief
valve inlets are maintained as vapor zones” which required by existing IGC. It
supports CAD interface of vessel tank 3d model as well as parametric modeling
of various tank shapes. It not only carries out a numerical calculation on tank
volume under the condition of the Trim & List, but also reviews the various
accident risks by using the viewer of 3D model and VR environment.

Keywords: LNG carrier
Filling limits
VR
IGC code 2018
Isolated vapor
pocket

1 Introduction

1.1 Filling Limits

If LNG in a sealed cargo tank expands and its volume expands, a major accident can
occur. Especially, the risk of LNG is higher because the volume of LNG is increased
about 600 times when LNG is vaporized. To prevent these accidents, the IGC Code
specifies the cargo tank filling limits, which are the ratio of the maximum liquid volume of
the cargo to the total volume of the tanks. Basically the filling limits are defined as 98%,
but if additional safety factors are considered, the filling limits may be increased corre-
spondingly with the approval of the class or the flag State concerned. The conditions to

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 80–90, 2021.
https://doi.org/10.1007/978-981-15-4680-8_6
 IGC Code-Based LNG Filling Limits Computation Using 3D 81

consider are the following, which is for the New IGC Code to be applied to gas carriers
built after July 1, 2016 [2].
• An isolated vapor pocket is not created in the cargo tank.
• The pressure relief valve (PRV) inlet is maintained as a vapor zone.
• Recognition of the allowance for cargo volume expansion at 1.2 times the set
pressure of the safety valve of the cargo tank.
• Recognition of the allowance for operational margins of at least 0.1% of tank
volume.
• Recognition of the tolerance of measuring instruments such as liquid level gauges
and temperature gauges.
“Isolated vapor pocket” is spaces where the PRV can not relieve pressure as the
cargo tank are tilted. Figure 1 shows the state of isolated vapor pockets. which means
that the PRV can not control its pressure. Figure 2 shows the PRV suction port is
submerged in LNG. Both cases are dangerous conditions that can lead to major
accidents since the pressure in a specific space can not be controlled. Therefore, the
charge limit must be calculated to meet the requirements.

Fig. 1. Isolated vapor pocket. Fig. 2. Submerged PRV in LNG.

1.2 Research Background

For verification of the filling limits above 98%, Several variables such as the placement
of the safety valve, the accuracy of the level and temperature measurement, the loading
temperature, etc. must be considered simultaneously. The LNG volumes for “scenario
in which isolated vapor pockets are not generated in cargo tank” and “scenario where
the pressure relief valve inlets are maintained as vapor zones” should be calculated
respectively under the four conditions of trim and list given in the IGC Code. The LNG
volume calculation is largely affected by the shape of the cargo tank, the position of the
safety valve and the dome, so tank 3D modeling is basically necessary. It requires a lot
of work since LNG conditions and various variables according to various situations
must be considered together.
Among of them, “scenario in which isolated vapor pockets are not generated in
cargo tank” recently applied to the New IGC Code has a great influence to the filling
limits. In particular, the tanks of prismatic shape are more likely to generate isolated
82 S. H. Byun et al.

vapor pockets when they are tilted, because the upper space is horizontal. Their
influence is greater since the most widely used membrane LNG cargo tanks in the
world are prismatic shapes.
Since the filling limits directly affect the amount of LNG cargo, careful verification
is required by stakeholders such as shipowners, shippers and class. In particular, the
class is obliged to check that the filling limits higher than 98% are possible accurately
considering the presence of isolated vapor pockets. Thus, this study has developed a
algorithm to calculate the filling limits accurately and time-efficiently. In addition, we
have developed the SeaTrust-FillingLimits software with the development algorithm,
which is intended to provide not only accuracy but also convenience of verification
through additional auxiliary functions.

2 Development

2.1 Algorithm Overview

Fig. 3. Entire algorithm process.

Figure 3 represents the entire development algorithm process. A parametric 3D tank is

created by parameter input corresponding to each initial tank shape. Basically, 3D
models of Cylindrical shape like Type C tank and prismatic shape like No. 96 & Mark
lll tank are supported. Even a special shape that is not supported by default can be
extended to the development algorithm if only its 3D shape is input, because the
development algorithm is fundamentally based on the 3D model.
 IGC Code-Based LNG Filling Limits Computation Using 3D 83

From the generated 3D tank, each tilted tank is created with ± 0.015L Trim, ±15°
List specified by IGC Code. And a 3D LNG corresponding to each tilted tank is
generated. Two scenarios defined by the IGC Code apply to slanted 3D LNG.
This study defines “the state where LNG is the lowest amount in the situation where
the tilted LNG touches the PRV” as “scenario where the pressure relief valve inlets are
maintained as vapor zones”, and defines “the state where is the lowest amount in sit-
uation where the isolated vapor pockets occur at the sloped LNG” as “scenario in
which isolated vapor pockets are not generated in cargo tank”.
Figure 4 shows the tilted tank shapes and the LNG status corresponding to each
scenario. (a), (b), (c) and (d) are states of “scenario in which isolated vapor pockets are
generated in cargo tank”, (e), (f), (g) and (h) show the states of “scenario where the
pressure relief valve inlets are not maintained as vapor zones” at each tilted condition.

Fig. 4. Tilted tank shapes and the LNG status corresponding to each scenario.

The volume of each LNG generated from the scenario definition is calculated.
Among of them, the volume corresponding to the severe case is extracted. The filling
limit of the tank, which reflects the total volume and the expansion coefficient in the
severe volume, is calculated. This process applies to all tanks placed on a LNG carrier,
and the final filling limits are calculated.
84 S. H. Byun et al.

2.2 Parametric 3D CAD Modeling

Tank modeling is basically required for calculation of filling limits since filling limits
are heavily influenced by the shape of the cargo tank and dome, the position of the
safety valve and the dome. In this study, parametric tank 3D CAD modeling technique
is applied to development algorithm to flexibly correspond to variable shape.
The shape of the membrane tank is based on bilateral symmetry. Its shape
parameters include the left half width of the lower horizontal part on the front, the
width and height of the lower left side on the front, the height of the vertical part on the
front, the width and height of the upper left side on the front. The shape of the
cylindrical tank is exactly capsule form. Its shape parameters are the radius of the
cylinder part, the length of the cylinder part on the side, and the length of the front and
rear partial sphere. If the membrane or cylindrical tank is tapered, an additional tapering
height parameter is used. The arrangement parameters of the tank are the distance from
the ship aft perpendicular (A.P.) to the back of the tank, the distance from the ship
centerline (C.L.) to the tank centerline, and the distance from the ship baseline (B.L.) to
the bottom of the tank.
The shape parameters of the hexahedral dome are width, height, and length. The
shape parameter of the cylindrical dome is the radius and height. The arrangement
parameters of the hexahedral and cylindrical dome are the distance from the front of the
tank to the front of the dome, and the distance from ship centerline to tank centerline.
If the inner thickness of the tank and dome is to be taken into account, then
thickness parameters are additionally used. This creates 3D models offset by thickness
in the generated tank and dome 3D model, and is computed as an internal model.
The shape parameters of PRV are radius and height. The arrangement parameters of
the vertical PRV on the hexahedral and cylindrical dome are the distance from the back
of the dome to the center of the PRV, the distance from the dome centerline to the PRV
centerline. The arrangement parameters of the horizontal PRV on the hexahedral dome
are the head direction of the PRV (left to right, right to left, front to back, front to back),
and distance from the bottom of the dome to the center of the PRV. The arrangement
parameters of the horizontal PRV on the cylindrical dome are the angle on the top view
(0 to 360°) and distance from the bottom of the dome to the center of the PRV.
Additional tank 3D models have a rotational arrangement parameter, which allows
the tank to be rotated by ±90 degrees to allow for special cases. A parametric 3D
model with the final shape and placement is created through the entire parameter.
Finally, a parametric 3D model of shape and arrangement is created through the
entire parameters. Typical tanks produced by parametric modeling are shown in Fig. 5,
tapering tank in Fig. 6 and dome and PRV in Fig. 7.
 IGC Code-Based LNG Filling Limits Computation Using 3D 85

Fig. 5. 3D CAD geometry shape of GTT NO96 (top), GTT MARK lll (bottom-left) and Type C
(bottom-right)

Fig. 6. 3D CAD geometry shape of tapered tank

86 S. H. Byun et al.

Fig. 7. 3D CAD geometry shape of hexahedron dome, cylindrical dome, horizontal PRV and
vertical PRV

2.3 Expansion Coefficient Calculation

Since LNG can be expanded by human error and equipment accuracy, filling limits
should be considered for expansion coefficient. The expansion coefficient is a value that
subtracts the scenario calculation volume. The following items in “Acceptance criteria
for cargo tank filling limits higher than 98%” of the International Association of
Classification Societies (IACS) are involved in the expansion coefficient [3].
• Level and temperature gauge accuracy.
• pressure rise above opening pressure.
• an operational margin to account for liquid loading lines drained back to cargo.
• tanks and closing time of loading valves and operators reaction time.
The expansion coefficient (at ) is given by:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
at ¼ a21 þ a22 þ a3 þ a4 ð%Þ ð1Þ

• a1 = relative increase in liquid volume due to tolerance of level gauges.

• a2 = relative increase in liquid volume due to the tolerance of temperature gauges.
 IGC Code-Based LNG Filling Limits Computation Using 3D 87

• a3 = expansion of cargo volume due to pressure rise when pressure relief valves are
relieving at maximum flow rate.
• a4 = operational margin of 0.1%.
dV Dh
a1 ¼  ð% Þ ð2Þ
dh V

• dV
dh = variation of tank volume per meter filling height at the filling height h (m =m)
3

• h = filling height (m) at the filling limit FL to be investigated (FL > 98%)
• V = accepted total tank volume (m3 )
• Dh = max. total tolerance of level gauges (m)

a2 ¼ b  DT ð%Þ ð3Þ

• b = volumetric thermal expansion coefficient at reference temperature (%/°K)

• DT = max. tolerance of temperature gauge (°K)
 
qPRV
a3 ¼ 1  100ð%Þ ð4Þ
qPRV1:2

• qPRV = cargo density (qR Þ at reference conditions, i.e. corresponding to the tem-
perature of the cargo at set opening pressure of the pressure relief valve (PRV)
• qPRV1:2 = cargo density corresponding to the temperature of the cargo at 1.2 times
the set opening pressure of the pressure relief valve (PRV)

3 SeaTrust-FillingLimits
3.1 CAD Interface & Virtual Reality
The development algorithm calculates the filling limits based on parametric 3D CAD
geometry. 3D CAD geometry data of created parametric tanks and domes are expor-
table. The exported data can be imported into the development algorithm and can be
calculated in the imported 3D CAD geometry. The Export & Import CAD Interface
function works separately for tank and dome for development algorithm.
Since this is the manufacturer’s confidential information, only the difference
between the result of the parametric function and the resulting filling limit of the CAD
interface function is compared. The two shapes of GTT MARK lll was compared.
There was an error of 0.00001847% p in the first case and 0.00001430% p in the
second case. It is estimated to be due to the number of polygons. This error seems to be
able to be reduced from the number of export polygons, but it is already very slight.
Additionally, exported CAD data can be extended to other software or virtual
reality. Figure 8 shows the virtual reality simulator for the surveyor under development
by the Korean Register [1]. The Surveyor is looking at the hold of the cargo ship for
learning the structure and being educated about the inspection method. Therefore, the
LNG tank model exported from the development algorithm can be extended to the
virtual reality simulator, so that the training related to the LNG tank can be performed.
88 S. H. Byun et al.

Fig. 8. Virtual reality simulator for the surveyor developed by the Korean Register

3.2 Implementation
This study develops additionally SeaTrust-FillingLimits software that can operate the
development algorithm so that it can be easily utilized by users. SeaTrust-FillingLimits
includes not only the development algorithm, but also additional features such as
contour graphs of volume trend by angle, calculation of loading limits, and report
generation (Figs. 9, 10, 11 and 12).

Fig. 9. SeaTrust-FillingLimits Fig. 10. Contour graphs of volume trend by

angle
 IGC Code-Based LNG Filling Limits Computation Using 3D 89

Fig. 11. Loading limits graph Fig. 12. Auto-generated reports

4 Conclusion

The calculation progress of the filling limits is very complicated and time consuming
because the tank volume and LNG volume for several situations must be calculated
under various factors. “scenario where the pressure relief valve inlets are maintained as
vapor zones” was mainly considered in the case of the existing IGC Code. However,
the revised New IGC Code should take into account “scenario in which isolated vapor
pockets are not generated in cargo tank” as well as existing scenario. So the filling
limits calculation becomes more complicated and time consuming. This research
improves this problem by developing automated and optimized the filling limits cal-
culation algorithm based on parametric 3D CAD geometry. The development algo-
rithm does the following:
• Is applied to the parametric 3D CAD geometry concept. so easily make 3d models
of tanks, dome and PRV.
• Supports basically 3D models of Cylinderical shape like Type C tank and prismatic
shape like No. 96 & Mark lll tank. In addition, it supports cylindrical and hexa-
hedral dome, vertical and horizontal PRV type.
• Even applies to special shape that is not supported by default, if only its 3D shape is
input
• Simultaneously calculates multiple tanks that are placed within a vessel.
• Is based on a parametric arrangement of tanks, so it is usually arranged in 3-axis
coordinates and can arrange specially like a rotated tank.
• Simultaneously calculates all slope conditions to derive the most severe values, and
derives final filling limits reflecting the calculated expansion coefficient.
• Because it is based on 3D CAD geometry, the parametric model created in the
calculation algorithm can be exported to the common CAD format.
• Imports existing geometric models as well as models with parametric modeling.
• The CAD geometry exported from it can be extended to the virtual reality simulator.
In addition, this study develops SeaTrust-FillingLimits software with development
algorithm. This includes not only filling limits calculation but also additional functions
such as tank shape and LNG status 3D viewer, volume trend graph by angle, calcu-
lation of loading limits, and report generation.
90 S. H. Byun et al.

In the future, it will be possible to carry out a study to directly interface the CAD
geometry exported from other software where use ship 3D model data [4].

Acknowledgements. This research was performed as a part of the research project below and
supported by the organizations indicated. We acknowledge and appreciate the support provided.
“Development and commercialization of safety education and training VR contents of sailor
using virtual reality technology” project funded by Ministry of Science and ICT of Korea (NO.
S0602-17-1016)

References
1. Kil, W.S., Son, M.J., Lee, J.Y.: Development of VR ship environment for the educational
training of ship survey. J. Soc. Naval Architects Korea 55(4), 361–369 (2018)
2. International Maritime Organization (IMO): International Code for the Construction and
Equipment of Ships Carrying Liquefied Gases in Bulk (IGC Code). Consolidated Edition
(2018)
3. International Association of Classification Societies (IACS): Acceptance criteria for cargo
tank filling limits higher than 98% (on ships constructed before 1 July 2016), vol. 109, pp. 1–
3 (2017)
4. Son, M.J., Lee, J.Y., Park, H.G., Kim, J.O., Woo, J.J., Lee, J.H.: Development of 3D
CAD/CAE interface in initial structural ship design. Korean J. Comput. Des. Eng. 21(2), 186–
195 (2016)
 Developments of a Synthesis Model
for FLNG Design

Lariuss Zago1, Daniel P. Vieira1(&), Yu K. Kam1, Raul Dotta2,

Rodrigo M. Amarante1, and Claudio M. P. Sampaio1
1
Numerical Offshore Tank, Department of Naval Architecture and Ocean
Engineering, University of Sao Paulo, Sao Paulo, Brazil
[email protected]
2
Technomar Engenharia Oceânica, Sao Paulo, Brazil

Abstract. The increasing demand for natural gas is encouraging the develop-
ment of novel floating units’ designs, capable of processing large quantities of
hydrocarbon. These units called FLNG (Floating Liquefied Natural Gas) are
facilities that produce, process and store liquefied natural gas (LNG) offshore.
Once the topside and tanks of an FLNG are larger and more complex than the
regular FPSO vessels, a design process considering these particularities must be
used. Once just a few FLNG units are under construction and design and not yet
in operation, the information on the design first stages is poor. It is difficult to
obtain a first hull sizing without taking into account the complexity mentioned
above. A previous work (Vieira et al. 2016) presented a set-based approach that
works with sets of possible solutions that are analyzed and compared using a
merit function to select the best and feasible solution. The initial modeling
consists of a synthesis model that evaluates, based in a set of parameters, the
FLNG topside general arrangement and weight estimative, the initial design of
the midship vessel section, undamaged stability and seakeeping. This design
approach is particularly useful to deal with project trade-offs and to optimize
multiple characteristics. The present work shows the most recent developments
of this modeling. New features were implemented as the consideration of three
different loading conditions (30%, 60%, and 90%), evaluation of topside plants
with a capacity between 2 and 4 MTPA, hull motions coupled with the dynamic
of liquids inside the tanks. Furthermore, a more rational way of generating the
set of cases to be analyzed was implemented as well as a user interface capable
of providing more direct visualization of the vessel performance parameters.
This paper aims to achieve a platform design capable of producing, storing, and
offloading liquefied natural gas considering productions of 2, 3, and 4 MTPA.
For motions purposes, environmental conditions of Santos Basin in São Paulo,
Brazil was considered. In the same way, the design should guarantee the shortest
downtime as well as keep costs, of acquisition and operation, as low as possible.
Each of these characteristics must be quantified to allow a ranking of the gen-
erated solutions through an objective function.

Keywords: FLNG
Set-based design
Synthesis model

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 91–105, 2021.
https://doi.org/10.1007/978-981-15-4680-8_7
92 L. Zago et al.

1 Introduction

Natural gas is a fossil fuel and one of the few energy sources used in all sectors of the
economy, including the domestic, commercial, industrial and transportation sectors.
Other advantages of this energy source are that natural gas is the cleanest burning
hydrocarbon and relatively abundant which attracts investments to this sector. Is
expected that the global energy demand of natural gas to grow by 30% between 2015
and 2040, according to the International Energy Agency (IEA) New Policies Scenario.
Many of the reservoirs of hydrocarbon energy sources are in deep water. Brazilian
oil reserves present a high gas-oil ratio (GOR) in pre-salt condition, thus representing a
high possibility for natural gas monetization. And although its attractiveness, there are
technological issues, particularly concerning the storage of the gas and due to
restrictions imposed by the inexistence of export lines, most companies seek to reinject
the associated gas into the reservoirs. Consequently, if the natural gas is monetized, the
use of FLNG platforms with adequate processing and storage capacity seems to be one
of the main feasible solutions to put natural gas on the market.
Due to its unique characteristics, the offshore exploration, the storage system and
the transport of this energy source require a specialized platform named FLNG
(Floating Liquefied Natural Gas) platform. The most suitable technology for storing
large amounts of gas offshore is the liquefaction, which implies the need for a huge
processing plant in the deck of an FLNG unit and large insulated tanks. These new
challenges demand additional evaluation and analysis of the development of an FLNG
platform design. Although several FPSO (Floating Production and Storage Units) have
been designed and built for the last three decades, only recently the first platforms
dedicated to process and store natural gas are being constructed as explained in Vieira
et al. (2016). In addition to the increased difficulty to size and to install a LNG
processing plant onto a floating unit, the FLNG design process also presents several
challenges due to maritime requirements (stability and structural) and due to opera-
tional characteristics such as the loading and offloading process to a LNG carrier ship
(LNGc), especially in a scenario that the LNGc is in side by side position with the
FLNG platform, Zhao et al. (2011) summarized the main aspects in the FLNG side-by-
side application problems in his recent studies. The result is a very complex problem
with multiple parameters to evaluate by the designer in regards the wave performance,
production and costs, Vieira et al. (2018).
Another considered as a major issue is the free surface effect inside the tanks, as
shown in Rocha et al. (2015) and Vieira et al. (2016). Finally, the processing plant
depends on low tilt angles and low acceleration levels to operate, Pettersen et al. 2013,
making a careful seakeeping investigation necessary.
In 2016, extensive research undertaken at the Numerical Offshore Tank
(TPN) laboratory of the University of Sao Paulo evaluated several factors and
parameters of an FLNG floating unit design. The research experimentally and
numerically investigated various inherent physical processes as LNG sloshing, side-by-
side offloading, wind and current loadings, etc. Tests were performed both in the ocean
and towing tanks; for further details, refer to Rocha et al. (2015) and Vieira et al.
(2018).
 Developments of a Synthesis Model for FLNG Design 93

The present work aims to present a computational tool capable of parametrically

evaluate the performance of FLNG floating systems according to a previously defined
merit function with production capacities ranging from 2 MTPA to 4 MTPA (Million
Tonnes per annum). The routine uses the metocean conditions of Santos Basin in the
Brazilian coast. The design should also guarantee the shortest downtime as well as keep
costs, of acquisition and operation, as low as possible. Each of these characteristics
must be quantified to allow a ranking of the generated solutions through an objective
function.
This work presents a method to, given a set of basic parameters, design, evaluate
and optimize a FLNG platform in agreement with the most established international
rules considering topside layout, volume and weight distribution, cargo and ballast
filled conditions, intact stability conditions, longitudinal and transversal structural
specifications and seakeeping analysis in a unified synthesis model. Also, by com-
pressing all the meaningful information in regards to the platform’s performance with
an automatic routine presenting it a graphic interface in an organized way to the client
define which performance characteristic to optimize. Finally, the results are compared
and analyzed for a typical production level and storage capacity, highlighting that the
parametric model approach is usually applied to the conceptual design due to provide
basic performance indicators on feasible solutions for the client’s decision process.

2 Methodology

As mentioned in Vieira et al. (2016), the kernel of a parametric design routine is its
synthesis model. The mathematical models, describing and evaluating each aspect of
the design object, are assembled in a single continuous algorithm. Due to the FLNG
system complexity, this methodology incorporates the integration of the areas of the
process plant, stability, structure, and dynamics of the system in waves.
The amount of information required as input defines the sequence of numerical
routine. Thus, each run of the software generates a topside configuration and evaluates
the structure in three big blocks, respectively: stability, structure, dynamic of the
system in waves, as shown in Fig. 1.

2.1 Stability
A package of numerical routines evaluates the parametrized hulls of FLNG floating
systems. With the definition of the main dimensions, ship configuration and cargo
loading as input parameters, the hydrostatic properties of the FLNG are calculated as
the verification to IMO (International Maritime Organization) criteria. If the criteria are
attended, the geometry is saved in the database; else, it is discarded. Figure 2 shows a
flowchart of the main phases that make up the stability routines.
It is noteworthy that in this design phase, the standards of DNV-GL for maximum
heel and trim were taken into consideration to determine the amount of ballast.
94 L. Zago et al.

Fig. 1. Hull generation numerical routine flowchart

Fig. 2. Stability numerical routine flowchart

2.2 Structure
The structural arrangement evaluation is based on two main hypotheses. It was
assumed that the tanks would be detached modules from the main structure, and they
will be installed after the construction of the hull. This hypothesis is important since the
hull structure can be designed independently of the tank structure. Thus, the primary
DNV-GL (2012) requirements are completely achieved without regard to inertia pro-
vided by tank structure addition. Thus, using both the main dimensions L, B, and D as
the hull geometry obtained from the previous phase, the structural midship section and
the structural elements of the hull are determined. As in the previous phase, the results
are verified with the DNV-GL criteria, and the database is updated.
Figure 3 illustrates the routine flowchart and consists basically of evaluating the
central section arrangement, the cross-sectional arrangement, the steel weight of the
hull and center of gravity using an iterative process.
As explained in Vieira et al. (2016), two main assumptions were made to evaluate
the structure arrangement. First, it was assumed that tanks would be separated in
modules which will be installed after the hull construction. An IMO Type B
(SPB) prismatic tank will be used, which have the ideal characteristics to store LNG
and LPG (IHI Offshore Group, 2019).
 Developments of a Synthesis Model for FLNG Design 95

Fig. 3. Structure numerical routine flowchart

Fig. 4. Structural midship section in detail

This assumption is important once the hull structure can be designed independently
of the tank structure. Thus, the DNV-GL primary requirements are fulfilled without
considering the inertia due to the addition of the tank structure. It was considered a
conservative approach once the tank structure presence will contribute to the increase
of midship section properties.
The second assumption is that the other hull sections are, initially, considered equal
to the midship section. It is also a conservative approach, once the main efforts are in
the hull center. In future steps, a study can be made to evaluate the impact of this
assumption in the hull weight.
Consider that a typical arrangement of the hull structure is composed by primary
(single or double side; single or double bottom; single or double deck) and secondary
(heavy reinforcements and light reinforcements) elements, as shown in Fig. 4.
96 L. Zago et al.

2.3 Seakeeping
A frequency domain numerical model, performed by WAMIT software for wave
analysis based on potential linear theory (Newman 1977), allowed the study of the
ocean system under wave excitation. The main properties evaluated are wave forces,
hydrodynamic coefficients (added mass, damping potential, and hydrostatic restoration)
as well as the Response Amplitude Operator (RAO).
To evaluate the first order forces, the code standard method was used, as presented
in Newman and Sclavounos (1988). First, the wet surface of each vessel was designed
in a computational routine environment that created a geometrical data file (GDF) for
each loading condition. Figure 5 shows a wet hull surface to 90% percent loaded ship
that was used in the analysis.

Fig. 5. 90% loaded ship wet surface: Hull (grey) and Tanks (blue).

The tanks wet surfaces were also provided to evaluate the influence of liquid cargo
motion on ship motions (Newman 2005). The setup was made using WAMIT (2016)
instructions. This influence has a significant impact on ship motions as larger tanks are
used.
Figure 6 presents the processes and outputs involved in this stage of the numerical
routine. First, the wet surface of each vessel was designed as a hull mesh in a com-
putational routine environment using the database obtained in the previous phases that
created a geometrical data file (GDF) for each loading condition. Each loading con-
dition is then analyzed with WAMIT, and the wave behavior of the floating platform is
characterized by the motions and accelerations of each loading condition being used to
evaluate the downtime by classification societies.
Simulations were performed considering four-wave incidences angles (180°, 165°,
150°, and 135°, under equiprobable occurrence hypothesis as presented in Fig. 7) and
fifty sea conditions (Hs , Tp ) for incidence angle (Table 1).
Restrictions for the FLNG platform accelerations and motions were imposed to
downtime analysis, as described in Table 2.
The downtime analysis was performed considering 30%. 60% and 90% load
conditions, relative to LNG tanks total capacity. Each considered condition is associ-
ated with a fraction of total time of the operation, as shown in Table 3 and used to
estimate the downtime.
 Developments of a Synthesis Model for FLNG Design 97

Table 1. Sea condition occurrence probability

Fig. 6. Seakeeping numerical routine flowchart

Fig. 7. Wave incidences angles considered in simulations

Table 2. FLNG downtime criteria for motions and accelerations

Criteria Description
Motions The maximum angle of inclination of the vessel considering the coupled
(Angle) pitch and roll movement should not exceed 1°
Accelerations The acceleration at the centers of gravity of the liquefaction modules shall
not exceed 0.1 g
98 L. Zago et al.

Table 3. FLNG loading conditions and operation time prevalence

Loading Condition Operation Prevalence
30% 25%
60% 50%
90% 25%

The freeboard of the FLNG platform is a key parameter, used for feasibility system
decision about stability analysis and loading/offloading operational conditions, espe-
cially in side-by-side operations. It is noteworthy that FLNG platforms usually have
high depth dimensions comparing with the LNG ship’s usual depth, which implies in
maximizing the platform draft to minimize its freeboard. Thus, In the simulations, each
platform model was initially considered fully ballasted, and its reduction was used to
correct pitch and roll angles.
It is also noteworthy the importance of the liquid motions inside the LNG tanks to
the analyses, since its direct influence on system stability and tanks configuration,
besides affect the weight of the system and structural arrangement. A more extensive
analysis is presented in Rocha et al. (2015) and Vieira et al. (2018).
At last, it is important to cite that the FLNG design software has a graphical
interface in which it is possible to provide the input data for the merit function, which
will be used to evaluate the generated cases. The interface also allows you to analyze
and graphically observe all data obtained from simulated cases.

3 Application of the Algorithm to a Generic Case: Results

and Analysis

The original numerical code, detailed in Vieira et al. (2016), was improved and is
described in detail in this paper. The main advances regarding the original model are:
• Tank dimensions are no longer a data input, based on load capacity but output for
the merit function;
• Consideration of three loading conditions (30%, 60%, and 90%) for stability and
seakeeping analysis;
• Consideration of the dynamics of the liquid cargo inside the tanks;
• Improvement and optimization of the model and structural assessment processes;
• Update of the process plant database for estimation of weight, area, and volume;
• Implementation of a hierarchical algorithm to increase the sample resolution.
With the objective to implement a merit function that can be flexible to a chosen
metocean condition, the algorithm was developed allowing the client to value each
performance characteristic. To estimate a conceptual design FLNG, this work presented
a merit function prioritizing the structural weight (S), load capacity (L), and downtime
(D) as the following equation shows:
 Developments of a Synthesis Model for FLNG Design 99

f ðD; S; LÞ ¼ 2D þ 2S þ L ð1Þ

Despite the scarce database, as emphasized by Vieira et al. (2016), an analysis of

the similar system was adopted, which presents the main characteristics of the existing
FLNGs in the world, either in operation or in the design phase. Thus, from the
parameters related to production capacity, storage, main dimensions, and others, it was
possible to perform parametric studies for the characteristic estimation of the platform
or its upper and lower bounds.
Using the information obtained, an analysis of a similar system was performed due
to apply the methodology previously explained by a hierarchical algorithm. The
hierarchical algorithm consists of three phases explained below.
The first phase consisted of the definition of the parameters to study the initial
performance of each case in a given environmental condition and respective production
capacity. Since the present work seeks to optimize each design parameter and there are
many variables to be evaluated given the system complexity, it was chosen a method to
group each parameter, classified concerning its relevance to the global performance and
about the remaining parameters. Therefore, three groups of parameters were
established:
1. those related to the main dimensions: length L, beam B, and depth D;
2. three associated with the platform secondary dimensions: DD, DB, and DS (double
deck, double bottom, and double side) width; and
3. the two last ones correlated to platform second-order effects: number of tanks, and
tank arrangement.
Since the parameters L, B, and D have a significantly higher effect on the global
FLNG performance; they were studied first and evaluated separately, with the
remaining geometric features initially obtained from those main dimensions.
The first design case was generated from a study of similar FLNG platforms in CH-
TPN (Sao Paulo, Brazil), by TPN and Frade Japão Petróleo Ltda teams. As described in
Vieira et al. (2016), from this initial geometry, 300 random cases with similar
geometries were developed by the Synthesis model (SM) to ensure an adequate sample
space. The variation range of each dimension is presented in Table 4.
To avoid presenting a large amount of information and easy reader understanding,
the study case of 4 MTPA liquid gas production was selected, and its partial design
data is presented in Table 5.
Each case evaluation considered three main analysis variables: structural mass1, the
seakeeping/downtime2 performance, and the maximum load capacity. After the initial
calculation, a new iteration of the SM is performed to improve the parameters double
deck (DD), double bottom (DB) and double side (DS) previously described

1
The structural mass is a variable directly related to the number of LNG tanks, the FLNG construction
complexity and ultimately its total costs.
2
Downtime is time lapse that the system is not operational, usually represented as a percentage of the
total time available and is related to the ability to operate in adverse environmental conditions
especially in rough seas.
100 L. Zago et al.

Table 4. FLNG main dimensions per production

Production Minimum Maximum Minimum Minimum Minimum Minimum
length (m) length (m) breadth (m) breadth (m) depth (m) depth (m)
2 MTPA 300 400 50 70 24 39
3 MTPA 350 470 60 80 28 43
4 MTPA 420 540 70 90 32 47

Table 5. Structure dimensions and performance characteristics of the best five FLNG systems
evaluated – Phase 1
4 MTPA - Phase 1
1 2 3 4 5
ID flng0004 flng0272 flng0001 flng0188 flng0148
L [m] 521 530 472 525 481
B [m] 74 73 74 73 74
D [m] 46 46 46 44 44
hDD [m] 4 4 4 4 4
wDS [m] 3 3 3 3 3
hDB [m] 3 3 3 3 3
Nº of tanks 6 6 6 6 6
Nº of tank rows 1 1 1 1 1
Structural mass [t] 168578 172996 153336 168603 151635
Downtime [%] 0.4 0.2 0.9 0.4 0.9
Load mass [t] 534755 536869 482849 503237 465938

Table 6. Dimension variation of double deck, double side and double bottom
Parameter variations
hDD [m] 3.0 4.0 2.5 2.0 2.5 2.0 3.0 2.5 2.0 3.5
wDS [m] 2.0 3.0 4.0 3.5 3.0 4.0 2.0 4.0 2.0 3.5
hDB [m] 4.0 3.0 2.5 2.5 2.0 4.0 2.5 2.0 3.5 3.0

(Table 6 presents the considered range of these parameters). This hierarchical algo-
rithm procedure guarantees a better sample resolution as described by Zitzler et al.
(2000) (Fig. 8).
As pointed by Vieira et al. (2019), it is expected that increasing of the double side
dimensions, and consequently filling the space between the LNG tanks and the hull
structure with ballast, contributes significantly to reduce the vessel’s freeboard as a
consequence of displacement increasing. From the freeboard perspective, this assertion
suggests that it is favorable to build narrower and higher tanks as well as the increase in
ballast mass provided by larger double side width contributes to reducing the roll
motion amplitudes of the FLNG, which is associated with the downtime in loading/
offloading operations.
 Developments of a Synthesis Model for FLNG Design 101

hDD
Legend:

ba l last

ca rgo

i ns ulation

empty

hDB

wDS wDS

Fig. 8. Schematic of cargo and ballast tanks arrangement

Table 7. Structure dimensions and performance characteristics of the best five FLNG systems
evaluated – Phase 2
4 MTPA - step 2
1 2 3 4 5
ID flng0001_0009 flng0001_0001 flng0272 flng0004 flng0188
L [m] 472 472 530 521 525
B [m] 65 65 73 74 73
D [m] 40 40 46 46 44
hDD [m] 2 3 4 4 4
wDS [m] 2 2 3 3 3
hDB [m] 3.5 4 3 3 3
Nº of tanks 6 6 6 6 6
Nº of tank rows 1 1 1 1 1
Structural mass [t] 129492 129548 172996 168578 168603
Downtime [%] 0.4 0.4 0.2 0.4 0.4
Load mass [t] 378352 361094 536869 534755 503237

Table 7 presents the five best scenarios cases for the second phase of the algorithm.
The final phase of the synthesis model corresponds to the evaluation of the LNG
tanks arrangement. Vieira et al. (2019) observed that configurations with smaller tanks
have a reduction of sloshing effects, that is, movement of liquid (LNG) inside the tanks,
which is associated with the platform downtime and its stability. Vieira et al. (2019)
state that for smaller tanks, the downtime varies less than for large tanks. Besides that,
tanks with high height/width ratio have less influence of the tank width than tanks with
low height/width ratios. Table 8 presents the best design for all production capacity
evaluated.
102 L. Zago et al.

Table 8. Structure dimensions and performance characteristics of the best FLNG systems
evaluated for each production capacity – Phase 3
BEST FLNG DESIGN
ID flng0058 flng0060_0009 flng0001_0009
Production [ MTPA] 2 3 4
L [m] 414 467 472
B [m] 56 64 65
D [m] 36 40 40
hDD [m] 3 2 2
wDS [m] 2 2 2
hDB [m] 2 3.5 3.5
Nº of tanks 6 6 6
Nº of tank rows 1 1 1
Structural mass [t] 94426 125905 129492
Downtime [%] 2.7 0.4 0.4
Load mass [t] 252813 367280 378352

Fig. 9. FPSO and FLNG around the world

It is noteworthy that the third phase has a minor contribution, in that case, due to the
number of tanks increasing leading to the total load volume decreasing and consequent
negative effect regarding structural mass and corresponding total costs of the system. In
that case, the changes in the configurations of the tanks owing to improve the dynamics
of the liquid cargo resulted in adverse effects in the other parameters of the merit
function.
Figure 9 presents a scatter distribution of existing or in design FPSO and FLNG
around the world, together with the design model obtained by SM running. It is
possible to notice that the three FLNGs designed are close to each other and have the
beam range between 55 m and 70 m and length range about 400 m and 500 m.
 Developments of a Synthesis Model for FLNG Design 103

Table 9. Dimensions and performance parameters comparison of the FLNG design with SM to
the years 2013, 2018 and 2019 – 4 MTPA production capacity
2013 2018 2019 Unit
Length 450.0 462.0 472 m
Beam 81.0 74.0 65 m
Depth 38.0 46.0 40 m
Maximum draft 17.1 28.9 25.9 m
Displacement 609731 953804 767953 t
Structural mass 192100 159459 129492 t
Topside mass 88007 98840 98778 t
Load capacity 240000 500735 378352 t
Number of tanks 20 10 6 units
Tank length 32.8 75.9 63.2 m
Tank beam 37.0 33.0 57 m
Tank height 19.7 40.0 35 m
Double deck 3.5 2.0 2 m
Double bottom 4.5 3.0 2 m
Double side 2.5 3.0 3.5 m

Table 9 presents the main characteristics and the performance parameters for three
FLNG design process obtained with the newest version of SM (2019) and using 2013
and 2018 previous versions. It is important to mention that the new algorithm leads in
an FLNG design with a higher length/beam ratio, as also with a smaller number of
tanks.

4 Conclusion

For the study case, the first phase generates random cases to evaluate the effects of the
main dimensions in the performance of the system. In this phase, it is perceptible the
definition of a region of the best FLNG dimensions with beam between 55 and 70 m
and length between 400 and 500 m.
The second phase evaluates the dimensions of double deck, double side, and double
bottom, which directly affects the dimensions of the tank. The variation of these
parameters is predefined, and the performance of the system is evaluated. The tank
dimension is one of the major characteristics of the system as it can induce changes in
the structural mass, the dynamics of the liquid cargo, and the total volume available to
LNG.
It is noteworthy that the third phase has a modest contribution. Since the parameters
of load capacity, structural mass and downtime are used to ponder the FLNG design
performance, increasing the number of tanks affects negatively the total load capacity
and the structural mass, decreasing the volume of cargo and increasing the structural
mass and costs of the system.
104 L. Zago et al.

It is also important to highlight that downtime has a slight variation in the third
phase. The increase in the number of tanks highly affects the cargo capacity, which
reduces the total mass of the system and turns the FLNG dynamics worst.
The designed vessel obtained in 2013 was a conceptual project using the same
design software, which the main objective was only the improvement of the downtime
factor.

Table 10. Comparison of the rank in the systems developed in the years 2013, 2018 and 2019
flng_2019 flng_2018 flng_2013
8.10 6.50 0.42

As observed in Table 10, this FLNG system has lower performance rank in
comparison with the other solutions due to the storage capacity and the structural
weight factors not being the main objective in the initial project. Even though the 2013
FLNG is still inside the optimum region defined by the third stage of the actual method,
which shows the coherence of the design. As explained above, the 2018 FLNG has the
inclusion of both the storage capacity and structural weight aspects in the merit
function, beyond refinements in wave behavior and project criteria. These improve-
ments ensure a higher rank than the 2013 solution.
The 2019 vessel, as the 2018 model, used the same merit function. But, differently,
from the previous solution, it employed a three-phase evaluation defining suboptimum
regions which the next solution is evaluated related to the previous one.
It is important to note the evolution in the final rank of each system, which shows
significant improvements in the merit function employed. Besides that, all systems
presented consider a theoretical mooring system called soft mooring, which does not
influence the system’s dynamic behavior significantly. A real mooring system should
be employed in software design in future works.

5 Future Works

Future works intend to improve the software and implement new effects as:
• Simulations with different topside layouts;
• In the structural model, the analysis considering the transversal efforts must be
implemented to carry out the classification society rules in the initial phases;
• Once a solution is obtained additional studies such as riser, mooring, and offloading
analysis;
• Implementation of the effects of wind and current efforts;
• A more extensive simulation, considering a larger number of cases.
 Developments of a Synthesis Model for FLNG Design 105

Acknowledgments. The authors acknowledge Frade Japão Petróleo Ltda for the financial
support and the motivation of this work. Partially funded by the INTPART SUBSEA
(NTNU/USP) project, supported by the Research Council of Norway.

References
DNV-GL: (DNV-OSS-103) rules for classification of floating LNG/LPG production, storage and
loading units (2012)
IHI Offshore Group: Self-supporting Prismatic shape IMO type B (SPB) Technology. http://
www.ihi.co.jp/offshore/spbmenu_e.htm. Accessed May 2019
Newman, J.N.: Marine Hydrodynamics. MIT Press, Cambridge (1977)
Newman, J.N., Sclavounos, P.D.: The computation of wave loads on large offshore structures. In:
5th International Conference on the Behavior of Offshore Structures (BOSS). Trondheim,
Norway (1988)
Newman, J.N.: Wave effects on vessels with internal tanks. In: 20th International Workshop on
Water Waves and Floating Bodies (IWWWFB). Spitsbergen, Norway (2005)
Rocha, T.P., Dotta, R., Vieira, D.P., Mello, P.C.D., Malta, E.B., Nishimoto, K.: Experimental
investigation on the influence of liquid cargo in floating vessels motions. In: Offshore
Technology Conference (OTC Brazil). Rio de Janeiro, Brazil (2015)
Vieira, D.P., Lavieri, R.S., Rocha, T.P., Dotta, R., Rampazzo, F., Nishimoto, K.: A synthesis
model for FLNG design. In: 35th International Conference on Ocean, Offshore & Arctic
Engineering (OMAE). Busan, Korea (2016)
Vieira, D.P., de Mello, P.C., Dotta, R. Nishimoto, K.: Experimental investigation on the
influence of the liquid inside the tanks in the wave behavior of FLNG vessels in side-by-side
offloading operations. Appl. Ocean Res. 74, 28–39 (2018)
Vieira, D.P., Kam, Y.K., Sampaio, C.M.P.: Analysis of the performance of FLNG vessels
according to basic parameters. In: XXVI Congreso Panamericano de Ingeniería Naval,
Transporte Marítimo e Ingeniería Portuaria (Copinaval). Cartagena, Colombia (2019)
WAMIT Inc.: WAMIT User manual, Version 7.2 and 7.2PC (2016)
Zitzler, E., Deb, K., Thiele, L.: Comparison of multi-objective evolutionary algorithms: empirical
results. Evol. Comput. 8, 173–195 (2000). https://doi.org/10.1162/10636560056820
Zhao, W., Yang, J.M., Hu, Z.Q., Wei, Y.F.: Recent developments on the hydrodynamics of
floating liquid natural gas (FLNG). Ocean Eng. 38(14–15), 1555–1567. https://doi.org/10.
1016/j.oceaneng.2011.07.018
 The First LNG Fuel Train Ferry
for St. Petersburg – Kaliningrad Line

Gennadiy Egorov(&) , Alexander Nilva ,

and Dmitriy Chernikov

Marine Engineering Bureau, Tenistaya 15 Str.,

Odessa 65009, Odessa Region, Ukraine
{egorov,nilva,chernikov}@meb.com.ua

Abstract. Results of complex of works on optimization of concept of the first

train ferry of LNG with taking into account ice and ecological requirements of
Baltic region are provided. Therewith new ferry is purely cargo without function
of passengers’ transportation. Considering that on Ust-Luga (St. Petersburg
region) – Baltiysk (Kaliningrad region) line passenger function is subsidized and
passenger requirements significantly influence on general arrangement and on
ferry’s design in common, and also on its speed and power, and, hence, con-
siderably reduce railway car capacity and increase cost value of freight trans-
portations, decision to make new ferry in cargo variant has made this project
economically acceptable. Grounding of main characteristics of researched type
of ferries is executed. The new concept of train and automobile ferry for Baltic is
LNG fuel steel self-propelled vessel with Arc4 ice category, with raking stem,
with one cargo deck providing placement of 80 railway cars in one plane
without usage of cargo lifts, aft located ER, with diesel propulsion plant, aft
auxiliary thruster and two bow thrusters.

Keywords: Baltic sea
Sea train ferry
Design
Ecology
Efficiency

Economy
LNG technologies
Innovation

1 Introduction

Majority of UN system organizations as well as IMO included in own activity essential

ecological component oriented on transition to sustainable development of mankind.
Example is step-by-step toughening of requirements to air pollutants contained in
vessels’ exhaust gas including sulphur oxides (SOx), nitrogen oxides (NOx) and par-
ticulate matters according to Annex VI of MARPOL.
The Baltic Sea is a special NOx (NECA) and SOx emission control area (SECA).
Since 1st January 2015 the maximum sulphur content of marine fuels used in SECAs
has been reduced to 0.1%. Standard liquid fuel with such content of sulphur becomes
extremely expensive and significantly increases expenses on sea cargo transportations.
Gas fuel allows completely to exclude SOx emissions and particulate matters, to reduce
NOx emissions by 90% and CO2 emissions by 30%.

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 106–119, 2021.
https://doi.org/10.1007/978-981-15-4680-8_8
 The First LNG Fuel Train Ferry for St. Petersburg 107

Therefore, urgent need of LNG fuel new generation ferries which meet the most
modern requirements to safety has appeared: new train ferries of Marine Engineering
Bureau CNF19M project for Kaliningrad region [1] will be exactly operated in
emission control areas.

2 Aim of the Paper

Search of such technical solutions which will allow to create effective concept of
perspective CNF19M train ferry for special NOx and SOx emission control areas.

3 Main Text

On the Baltic Sea full range of ferry transportations including cruise line transporta-
tions, cargo-passenger, cargo and railway is present. Ferry transportations are carried
out regularly and practically without delays.
In total more than 50 ferry services work at Baltic. Only six of them are railway and
quantity of railway lines is being reduced.
New train ferries for connection of Kaliningrad region with rest part of the Russian
Federation are indispensable as main freight traffic now follows by railway through
Lithuania and Belarus.
Therewith alternative variants of delivery of specific cargoes including military
ones (ferry communication is the only way to deliver cargo without transportation
through neighbouring states) are almost absent.
There are also no active train ferries with maximal possible overall dimensions and
characteristics for Ust-Luga – Baltiysk line. As for March 2019 only two 187 m diesel
train ferries “Ambal” (1990 year built) and “Baltiysk” (1984 year built) of CNF05
modernization project were working on the line. Vessels which can effectively work in
ice conditions of Ust-Luga port are necessary.
Researches of CNF19M concept of cargo ferry for the Baltic Sea are carried out by
Marine Engineering Bureau since 2014 (see Fig. 1 for design stages):
a) CNF19 – 2014 “triple deck” variant;
b) CNF19M – 2016 “single deck” variant without vehicles placement on open deck;
c) CNF19M – 2017 “single deck” variant with vehicles placement on open deck and
fore located superstructure;
d) CNF19M – 2018 “single deck” variant with vehicles placement on open deck, fore
located superstructure, side ramp and optimized fore end.
Therewith new ferry is purely cargo without function of passengers’ transportation.
Considering that on Ust-Luga (St. Petersburg region) – Baltiysk (Kaliningrad region)
line passenger function is subsidized and passenger requirements significantly influ-
ence on general arrangement and on ferry’s design in common, and also on its speed
and power, and, hence, considerably reduce railway car capacity and increase cost
value of freight transportations, decision to make new ferry in cargo variant has made
this project economically acceptable.
108 G. Egorov et al.

Fig. 1. Stages of creation of CNF19M train ferry concept.

3.1 Concept
The new concept of CNF19M train ferry for Baltic (see Fig. 2 and Fig. 3) is 200-m
steel self-propelled LNG fuel automobile and train ferry with Arc4 ice category, raking
stem, closed cargo deck for railway cars and opened upper deck (top of superstructure)
for truck trailers and automobiles, fore arrangement of superstructure and aft located
ER, twin screw propulsion unit, aft auxiliary thruster and two bow thrusters, with gas
tanks located in hold in middle part on safe distance from ER and superstructure, with
double bottom and double sides, with possibility of gas bunkering simultaneously from
four road-train LNG-carriers with a help of booster module, and also LNG bunkering
vessel or with LNG coastal bunkering station.

Fig. 2. General view of CNF19M ferry.

Main characteristics of the ferry is given in Table 1. Principal dimensions and

endurance have been obtained on the basis of the analysis of way conditions of
potential sailing area of CNF19M LNG fuel ferry. New idea of concept is placement of
truck trailers on upper deck in addition to railway cars located on cargo deck. It allows
 The First LNG Fuel Train Ferry for St. Petersburg 109

to provide parallel unloading of trucks from upper deck to berth with loading/unloading
of railway cars to/from main cargo deck (see Fig. 4).

Fig. 3. General arrangement of CNF19M LNG fuel train ferry.

Table 1. Main characteristics of LNG fuel ferry of CNF19M project

Parameters Values
Length overall, m 199.90
Length on summer load waterline (SWL), m 196.62
Breadth overall (with fenders), m 27.40
Breadth, m 27.00
Depth to the Main deck, m 8.60
Depth to the Upper deck, m 14.70
Draught at SWL, m 6.0
Deadweight at SWL (about), t 11062
Range of sailing, nautical miles 3000
Endurance, days:
By oil, fuel, provision and fresh water stores 10
By sewage water 3
(in areas where any discharge are prohibited)
Length of railway lines, m 970
Class notation of Russian Maritime Register of KM Arc4 AUT1-ICS OMBO
Shipping (RMRS) ECO GFS Ro-ro ship
Rudder and propeller complex 2 controllable pitch propellers + 2
semi-balanced rudders
Bow/aft thrusters, kW 2  1000/1  1000
Auxiliary power plant (diesel-generators/shaft- (2  1140/2  2400) + 1  400
generator) + emergency diesel-generator, kW
Crew/berths, pers 24/36
110 G. Egorov et al.

For connection of the ferry with coastal ferry complexes in Baltiysk and Ust-Luga
recess with special supports for the bridge in the aft part of the main deck is foreseen.
The vessel is assigned for transportation of Russian standard railway cars with
1520 mm gauge, and also other ro-ro cargoes, including dangerous cargoes (see Table 2).
It is possible to transport fifteen 40 tons truck trailers on top of superstructure. There are
lashing pockets for securing of railway cars on the main deck. Fastening of railway cars is
carried out with chain lashings and speed lashes and also with railway stoppers.
Design temperature of an outside air is +22 °C at humidity 60% in summer and
−23 °C at humidity 85% in winter, water temperature is from +16 °C down to 0 °C
accordingly.
For accommodation of crew following single cabins are foreseen: 3 single block-
cabins for accommodation of master, chief officer and chief engineer; 21 single cabins
with lavatory and douche for officers and other crew members.

Fig. 4. Parallel loading of railway cars and trucks.

Table 2. 1-T type (Russian standard) railway cars’ cargo carrying capacity
Railway car’s type Length between automatic couplers, mm Quantity, units
Four-axle tank car 12020 80
Open freight car (1st type) 17670 54
Open freight car (2nd type) 13920 70
Freight car 24620 36
Hopper car 14720 64
Autorack car of 11-1291 type 26720 34
Autorack car of 11-287 type 24260 36
Truck trailer 16500 (overall length of truck trailer) 58
(instead of railway cars)

3.2 Hull Form

In process of creation of hull form three main factors were taken into account:
• Ensuring of high-speed characteristics (relatively small block coefficient CB 0.661
while draught at SWL is 6.0 m, sharp angles of waterlines in fore part);
• Hull form has to meet requirements for vessels with Arc4 high ice class (slope
angles of SWL, frames and stem post on SWL level are regulated), see Fig. 5;
 The First LNG Fuel Train Ferry for St. Petersburg 111

• Cargo main deck has to stow the maximal number of railway cars (wide hull
contours in fore part), see Fig. 6.

Fig. 5. CNF19M ferry ice strengthening in accordance with Arc4 requirements.

Fig. 6. Marshalling of railway cars on main deck of CNF19M ferry.

Fig. 7. Optimization of hull form of CNF19M ferry.

112 G. Egorov et al.

Optimization of hull form with the help of CFD modelling and following tests in
experimental tank allowed significantly to improve CNF19M hull form and to reduce
fuel consumption, including due to rational approach to choice of operational speed and
organization of operational line (see Fig. 7, 8 and Fig. 9).
According to results of preliminary propulsion calculation, design speed of
18 knots is obtained with 2 main engines 8100 kW each (on 85% of maximal con-
tinuous rating).

Fig. 8. CFD modelling in flowvision software. Pressure distribution. Fore and aft view.

Fig. 9. CFD modelling in flowvision software. free surface. Fore and aft view.

3.3 Hull Strength Calculation

Overall hull strength meets requirements of Russian Maritime Register of Shipping
according to vessel’s class.
The maximal hogging (70.7% from allowed) is noted in loading case “The vessel
with eighty 94 tons railway cars on main deck, fifteen 40 tons truck trailers on top of
superstructure and 10% of stocks with icing”.
Moment resistance requirements of new hull for the deck are fulfilled with 8.1%
margin, for the bottom – with 10.4% margin.
Thicknesses of side shell plating and plating of longitudinal bulkheads meet
requirements with noticeable margin.
Calculations were carried out by means of finite element method (FEM) in ElBrus
program developed by Marine Engineering Bureau and which is certified by RMRS
(see Fig. 10). Design temperature for materials of hull and superstructures is accepted –
30 °C. Cargo main deck is designed for transportation of railway cars (weight up to
94 t) with load not greater than 23.5 t on axle on all tracks.
 The First LNG Fuel Train Ferry for St. Petersburg 113

There are installed 12 main watertight transverse bulkheads which are sectioning
the hull into 13 impermeable compartments.

Fig. 10. Hull strength calculations in ElBrus program.

3.4 Features of LNG Application as Main Fuel

There are some ways of solution of problem of compliance to requirements of
Annex VI of MARPOL to air pollutants contained in vessels’ exhaust gas including
SOx and NOx as usage of low-sulphur marine diesel oil (MDO), heavy fuel oil
(HFO) with scrubbers installation (purification system of exhaust gases from sulphur
oxides), liquid natural gas.
Usage of low-sulphur MDO is unreasonable because of its high price in comparison
with HFO and LNG. Usage of scrubbers is inexpedient as installation of scrubbers
which are offered for delivery by different companies (complete unit dimensions about
5400  2700  1700 mm) will cardinally affect on railway car capacity and on
economy of the ferry respectively. Therewith, independently of applied fuel type, to
fulfill Tier III NOx standards installation of additional purification systems of exhaust
gases from nitrogen oxides are required.
Therefore usage of LNG as primary ship fuel for CNF19M train ferries for solution
of problem of compliance to requirements for emissions in atmosphere has basic
advantages:
114 G. Egorov et al.

• Allows to exclude completely sulphur emissions, to reduce significantly emissions

of nitrogen oxides (by 85%) and carbon (by 25%) and also particulate matters [2];
• Level of NOx emissions will fulfill Tier III requirements of MARPOL without
installation of additional purification systems of gases [1].
While design and operation of LNG fuel vessels special attention should be paid to
safety. There are following characteristic types of dangers caused by LNG and its
vapours availability on the vessel [3]:
• Volume gas explosion appears from its leak in gaseous state in closed volume with
presence of ignition source;
• Explosion of gas tanks because of overpressure;
• Fire as a result of burning of spread gas or jet fire as a result of burning of gas under
pressure;
• Gas cloud in which fire can be appeared;
• Fast phase transformation when liquefied gas contacts water which is similar to
explosion without ignition;
• “Rollover” – rapid significant increase of pressure in cryogenic tank;
• Suffocation as a result of man’s contact with gas cloud;
• Injuries from low temperatures because of influence of liquefied gas on man’s skin;
• Atmosphere pollution as a result of gas leaking.
Following characteristic dangers connected with usage of LNG in internal com-
bustion engines should be kept in mind:
• Volume gas explosion because of gas leaking in crankcase, exhaust case, air col-
lector and ER;
• Possibility of gas dissolution in circulating oil with subsequent detachment in lube
oil tank and formation of explosive concentration and finally explosion.
At the moment Russian rules for LNG fuel vessels (not gas carriers) are not developed.
Thence DNV GL Rules requirements, IMO recommendations and requirements of
RMRS Rules of classification and building (which includes requirements of Interna-
tional Code of Safety for Ships Using Gases or Other Low-Flashpoint Fuels) have been
taken into account at study of complex of actions which provide increase of safety of
CNF19M project.
Following main constructive features are applied:
• Gas supply system located within ER has to be put into gastight envelope, i.e. has to
be with double-layer walls;
• For each engine near ER separate line of gas supply with possibility of remote
automatic shutdown in case of detection of leaks has to be provided;
• Possibility of instant automatic transition from gas fuel to liquid in case of shut-
down of gas supply;
• Sensors of LNG leakages’ control in places of the most probable leaks (ER,
interpipe space of doubled gas supply pipelines etc.);
• Gas leak signal generation in ship’s warning protection system (20% of explosive
concentration);
 The First LNG Fuel Train Ferry for St. Petersburg 115

• Automatic shutdown of LNG supply to equipment with signal generation in ship’s

warning protection system (60% of explosive concentration);
• Induced ventilation of interpipe space of pipelines with double-layer walls and
internal spaces of gas valve units;
• Ventilation of ER gas outlet circuits while engines are stopped;
• Supply of inert gas, for example nitrogen, for blowing off gas supply pipelines to
engines;
• Preliminary ventilation of boiler furnaces before their ignition.
The main difficulty when using LNG on vessels is rather big required space for
cryogenic tanks. In comparison with fuel oil equal amount of LNG requires approxi-
mately by 1.9 times of bigger volume. With taking into account thermal isolation of
tanks required volume increases approximately by 2.3 times. In case of installation of
LNG storage tanks in vessel’s hull required volume can be increased by 4 times.
Two possible options of applied cryogenic tanks are considered: LNG-Pac cryo-
genic tanks supplied by WARTSILA (see Fig. 11) and MAN companies (see Fig. 12).

Fig. 11. Main elements of Wartsila LNG fuel system which will be applied on the ferry.

LNG-Pac cryogenic tank is complex construction consists of: cryogenic tank with
isolation and external protection cover; fuel gas heater and gas evaporator units with
armature and pipelines; necessary sensors and control-measuring devices; attachments
to external pipelines and exchange of information interfaces with ship integral control
system. Top way of loading and unloading of LNG is applied. Discharge is carried out
by the effect of creation of pressure in cryogenic tank by outside pressurization.
Arrangement of LNG tanks and LNG acceptance station see Fig. 13 and Fig. 14.
Variant of usage of container-cryogenic tanks also was examined. This variant will
allow to make LNG delivery to the vessel by means of railway or motor transport
without classic bunkering operations. Replacement of tanks will be required only.
116 G. Egorov et al.

One of the main suppliers of container-cryogenic tanks in Russia OJSC “URALK-

RIOMASH” offers container-cryogenic tank of KTSM-35/0.6 model in dimensions of
standard FEU which is assigned for reception, storage and LNG delivery and can be
used as container for its transportation by railway, sea, river and motor transport
providing full safety of product under different weather conditions.

Fig. 12. Main elements of MAN LNG fuel system which was proposed for CNF19M ferry.

Fig. 13. Arrangement of LNG tanks on CNF19M ferry.

 The First LNG Fuel Train Ferry for St. Petersburg 117

If to use container-cryogenic tanks following circumstances must be kept in mind:

effective volume of container-cryogenic tanks will require their significant amount, for
example, for providing 5 days autonomy with main engines, boilers and diesel-
generators working on LNG 8 containers will be required, for 10 days autonomy –
16 containers; in addition onboard the vessel stationary units of evaporators and gas
heaters with armature and pipelines will also be required; for pressure monitoring of
gas supply to gas valve units and replacement of containers in ports appropriate load-
lifting means have to be foreseen.

Fig. 14. Arrangement of LNG acceptance station.

3.5 Endurance of the Ferry on St. Petersburg – Kaliningrad Line

The total volume of LNG tanks is 800 cbm (2  400 cbm). Taking into account 10%
dead stock it is available for consumption approximately 720 cbm. As well, reserve
diesel fuel tanks with total volume about 450 cbm are foreseen (see Fig. 15).
Partial bunkering is supposed at each call at the port. At speed of 18 knots round
trip takes 58 h, downtime at mooring berth is 3 h for each port. Overall time of round
trip is 64 h.
Fuel consumption at speed of 18 knots makes: 2030 kg/h and 300 kg/h while cargo
handling at port for MAN main engines; 2135 kg/h and 300 kg/h while cargo handling
at port for Wartsila main engines. Autonomy on LNG makes 2 full round trips – 128 h
(about 5 days).
118 G. Egorov et al.

Fig. 15. LNG round trip endurance at 18 kn operational speed with ballast distribution.

At speed of 15 knots round trip takes 70 h, downtime at mooring berth is 3 h for

each port. Overall time of round trip is 76 h.
Fuel consumption at speed of 15 knots makes: 1128 kg/h and 300 kg/h while cargo
handling at port for MAN main engines; 1172 kg/h and 300 kg/h while cargo handling
at port for Wartsila main engines. Autonomy on LNG makes 3 full round trips – 223 h
(about 10 days).
Anyway, stability of CNF19M ferry in all operational loading cases meets
requirements of RS Rules in accordance with class notation.

3.6 LNG Bunkering

Options of filling stations: from road-train LNG-carriers, LNG bunkering vessel, port
LNG bunkering station, LNG tank-containers.
Examples of all decisions of gas filling stations already exist in the world.
Road-train LNG-carriers which is offered now in the market on the Baltic Sea (for
example, Kriogaz company which bunkers LNG ferries in the port of Tallinn) are not
equipped with pumps for LNG delivery. Hereat for LNG supply on the ferry the
booster module is necessary. This module includes manifold for connection of four
road-train LNG-carriers and pump which will deliver LNG to the ferry. From the
module there will be one hose for connection to the vessel bunker station. The booster
module is supposed to be stored onboard the ferry (see Fig. 14), and for bunkering
operations will be exposed on mooring berth. If bunkering port will not be changed, the
booster module can be stored on mooring berth (agreement of port authorities is
required).
 The First LNG Fuel Train Ferry for St. Petersburg 119

4 Conclusions

Technical feature of developed by Marine Engineering Bureau CNF19M LNG ferry is

compliance to the most modern requirements to safety including requirements to
ecological safety. The ferry of CNF19M project is designed with maximal possible
dimensions and characteristics for Ust-Luga – Baltiysk line and now is under con-
struction. Keel was laid down in November 2018. Delivery is planned on 2021.
This is the modern, safe and economic vessel which will allow to make up to
135 round trips per year, to transport up to 10800 railway cars or 534 600 tons of
cargo per year. Round trip duration is 2.7 days if bunkering will be in time of cargo
operations.
Usage of LNG as primary ship fuel for train ferries has basic advantages in com-
parison with usage of low-sulphur MDO or HFO with scrubbers as allows to exclude
completely sulphur emissions, to reduce significantly emissions of nitrogen oxides,
carbon and also particulate matters. Level of NOx emissions fulfills Tier III require-
ments of MARPOL which came into force since 01.01.16 without installation of
additional purification systems of gases.
Usage of native technical experience and potential of cryogenic equipment will
allow to provide in short term broad appliance of LNG on vessels. Creation of shore
infrastructure for filling vessels with LNG will not take long time and notable capital
expenditures. By specialists’ estimations pilot shore complex can be put into operation
in one and a half year if there will be funds for design and construction. Economic and
ecological results will allow to repay expenditures in 5–7 years term [4, 5].

References
1. Egorov, G.V., Ilnitskiy, I.A., Chernikov, D.V.: Features of design of hull of train ferry for
Ust-Luga – Baltiysk – Ssasnitz – Mukran line (in Russian). In: Proceedings of Ship Structural
Design Science-Technical Conference Named for Prof. P.Ph. Papkovich, pp. 24–25. Krylov
State Research Center, Saint-Petersburg (2015)
2. North European LNG Infrastructure Project: A feasibility study for an LNG filling station
infrastructure and test of recommendations. Information http://www.lngbunkering.org/sites/
default/files/2012%20DMA%20North%20European%20LNG%20Infrastructure%20Project_
0.pdf. Accessed 31 Jan 2019
3. Blinkov, A.N., Vlasov, A.A., Shurpyak, V.K.: Analysis of methods of gas fuel usage in ship
power plants (in Russian). Science-technical collected book of Krylov State Research Center,
vol. 34, pp. 177–197 (2011)
4. Pronin, E.N.: Green light to “Blue corridor” (in Russian). National Gas-Engine Association.
Information http://www.ngvrus.ru/st6_1.shtml. Accessed 31 Jan 2019
5. McArthur, R.: Gas-fuelled mechanical propulsion solutions offer major emissions reductions.
Twentyfour7 1, 66–68 (2011)
 Creation of River-Sea Cruise Passenger Vessels
of New Generation

Alexander Egorov(&) , Gennadiy Egorov , and Igor Ilnitskiy

Marine Engineering Bureau, Tenistaya 15 Str., 65009 Odessa,

Odessa Region, Ukraine
{egorovag,egorov,ilnitskiy}@meb.com.ua

Abstract. The choice is grounded and concepts of river-sea cruise passenger

vessels including for the Caspian, Black and Mediterranean Seas with taking
into account requirements of main operators of the market are developed.
New concepts are not only new vessels, but also new tourist product. For
example, perspective Caspian cruise attracts not only citizens of Caspian
countries (Iran, Azerbaijan, Kazakhstan), but also tourists from the USA,
Australia, Europe and China. In market of sea and river cruises there are
practically no offers on earlier popular cruises between ports of Black,
Mediterranean and Baltic, and also Caspian seas. Thus, absence of offers is
connected, first of all, with lack of river-sea cruise vessels. For the solution of
problem of connection of sea and river routes modern cruise vessels of PV09
and PV300, PV300VD projects developed by Marine Engineering Bureau are
offered. Researches have shown that generally for native conditions traditional
single hull type passenger vessel can be recommended. CFD-modelling tests
were executed to optimize hull form. Module principles of creating inner living
space were used widely. Standard module cabins usage let alternate passenger
capacity by using passenger cabin modules with different area in accordance
with required comfort level.
PV09, PV300, PV300VD concepts are the safe, ecologically “pure” passen-
ger vessels which are completely satisfy to all International Conventions
including SOLAS, MARPOL, COLREGS and also to all national requirements
(Sanitary Rules, Technical Regulations of Safety of Inland Transport Objects).

Keywords: River and River-Sea cruises
River-Sea passenger vessels

Design
Level of comfort
Innovation
Ecology
Safety

1 Introduction

Today river and river-sea cruise vessels (CV) transport about 350–400 thousand
tourists on Russian and Ukrainian rivers, which includes 75 thousand foreigners.
Realities of native tourism are that to favorite river cruises of “entry” tourists from the
USA, Germany, Australia on route Moscow – St. Petersburg, “partly” sea lines
Moscow – Sochi, “Kiev” – Constanta”, “circular” cruise on Caspian Sea were added.
At this moment in Russia, Ukraine and in general in world practice of operation of
CV there is practically no fleet of river-sea cruise vessels.

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 120–136, 2021.
https://doi.org/10.1007/978-981-15-4680-8_9
 Creation of River-Sea Cruise Passenger Vessels of New Generation 121

From 254 river cruise passenger vessels built in Soviet period 98 (38.6%) have
been written off (6.7% – 17 vessels were lost in accidents, 31.9% – 81 vessels were
utilized). There were 32 vessels out of operation (12.6%) with middle age of
57.7 years. In operation – formally 124 vessels (48.8%) with middle age of 45.2 years.
Mean age of utilization – 47.1 year. In the next decade this fleet most likely will be
decommissioned and we will have only about 50 vessels [1] (without vessels which
can be built during these years but which are not ordered yet).
By some expert estimates the market size will make 1.0 million tourists by 2030.
The market assessment including previous years experiment on operation of first-
class vessels PV08 “Alexander Grin” and “Victoria” with native tourists allows to
draw conclusion that, with taking into account the predicted reduction of vessels of old
series, not less than eight new cruise vessels will be required by 2025.
Thus, problem of creation of river and river-sea cruise vessels became actual and
important for native water transport and tourist industries.

2 Aim of the Paper

Aim is grounding on basis of 2010–2017 Marine Engineering Bureau (MEB) re-

searches of concepts of modern river CV which are of interest for native shipping
companies: PV300, PV300VD, PV09 projects. These vessels will be operated on
classical river lines, and also make episodic cruises in coastal zones in accordance with
assigned class.

3 Review of Existing River-Sea Vessels in Researched Area

Operation of river CV on lines with sea parts is widespread in Ukraine. “Chervona

Ruta” and “Viking River Cruises” cruise companies use river CV which earlier have
worked along Black Sea coast [2]. The first cruises in Post-Soviet time on river CV in
Black Sea became possible after executed by MEB work about expansion of sailing
area of 302 project river passenger vessel of “T.G. Shevchenko” type. Work included
grounding of assignment of river-sea “M-PR” class of Russian River Register (RRR).
Since then and till 2014 following vessels had been successfully operated both in
Dnieper and Black Sea:
• two CV of 302 project (“General Vatutin” and “Zvezda Dnepra” of 1988 year built)
of “Chervona Ruta” cruise company;
• two CV of 301 project (“Princessa Dnepra” of 1976 year built and “Mikhail
Lomonosov” of 1979 year built of “Chervona Ruta” and “Viking River Cruises”
cruise companies respectively).
In 2014 all Ukrainian river-sea lines were stopped. As for 2019 only two vessels are
into operation on those routes: “Princessa Dnepra” of “Chervona Ruta” company and
“Viking Sineus” (ex “Mikhail Lomonosov”) of “Viking River Cruises” company of
301 project. “Viking Sineus” was modernized in accordance with MEB PV17 project
in 2013–2014.
122 A. Egorov et al.

After modernization it is four-deck, triple-screw motor vessel with overall length of

125.00 m, breadth of 16.70 m, with four-level superstructure along whole vessel’s
length, with ER located aft, with wheelhouse located fore, with double bottom, with
inclined stem and cruiser aft end (see Fig. 1). Passenger capacity is 196 passengers.

Fig. 1. General arrangement of PV17 river-sea modernized cruise vessel.

98 double-place cabins for location of 196 passengers are foreseen, including:

4 suites with balcony with area 30.1–31.6 m2; 2 “Deluxe” cabins with balconies with
area 24.2 m2; 60 cabins with balconies with area 13.5–16 m2; 32 cabins with balconies
with area 11.1–12.9 m2.
Passenger accommodations are located onto 3 decks: main deck, middle deck and
boat deck.
The restaurant which allows feeding all the passengers in one shift is located onto
the upper deck with area of 350 m2 for 212 places.
Panoramic bar with area of about 100 m2 for 55 places is located in the fore part on
the boat deck.
 Creation of River-Sea Cruise Passenger Vessels of New Generation 123

Music salon with bar with area of 150 m2. for 100 places and large opened area
with beach chairs and tents are located on the shade deck.
Lifesaving gear consists of four life boats and 46 inflatable freefall droppable rafts.
Vessel satisfies an increased ecological safety standard. Closed sewage and
domestic water systems are foreseen. All sewage and domestic waters are forwarded to
collector tanks. There is also sewage water treatment plant.
The next example of river-sea CV is Russian “Rus’ Velikaya” vessel. She was built
in 2013 with usage of elements of existing vessel of 588 project “N.F. Vatutin”.
Vessel’s length – 95.6 m, breadth – 13.5 m, draught – 2.45 m. Passengers’ capacity –
196 persons, crew – 66 persons.
But all these vessels are modernized.

4 Design Features of New Generation River and River-Sea

Cruise Passenger Vessels

New generation of cruise vessels for Russian inland waterways, designed by MEB, is
characterized by the following design features.

4.1 Maximization of Main Dimensions

Restrictions of way conditions of vessel’s operational region are determinative (e.g.
depths, dimensions of vessel’s navigation pass, dimensions of hydraulic engineering
facilities, etc.) – “Volga-Balt Max”, “Volga-Don Max”/”Dnipro Max”, “Belomor-
Baltiysk (BBK) Max”, “Danube Max” classes. Main dimensions of CV concepts are
accepted depending on data, which are shown in Table 1. For further details see [3]. It
is clear that river-sea concept will have restrictions on passengers’ capacity as length
and breadth of the vessel appointed in accordance with way conditions will limit useful
area which can be allocated for cabins and for passenger serving spaces (restaurants,
salons etc.). The number of decks is defined by air draught.

Table 1. General information about possible main particulars of river-sea CV.

Vessel’s class Overall length Overall Draught Air draught
LM, m breadth DM, m d, m HAD, m
“Volga-Balt Max”  185  16.95–20.1 2.90–  13.2
3.80
“Volga-Don Max”/  140  17.0 3.20–  13.2
”Dnipro Max” 3.70
“BBK Max”  132  13.63 2.90–  11.9
3.80
“Danube Danube –  135  13.40 2.00–  6.7
Max” Passau 2.50
Upstream  135  11.45 1.70–  6.03
Passaua 2.00
a
Including pass-through to Northern Sea (Danube-Main-Rhine system)
124 A. Egorov et al.

4.2 Architectural and Constructive Vessel’s Type

Due to architectural and constructive vessel’s type new CV can be as follows [4]:
• so called “floating hotels” with superstructure at all vessel’s breadth without pas-
sages by sides (with full-grown balconies/terraces or with French balconies at
cabins for CV that are oriented to foreign tourists. Upper “sun” deck is equipped
with swimming pool and tents;
• with promenade gallery by sides on all decks; such arrangement provides all-round
view for tourists. This variant is used rarely for new projects because many tourists
don’t like when other people are close to the windows of their cabins;
• with fore saloon that provides both-sides view;
• interim option (for CV oriented on Russian tourists) is applied. This option com-
bines variants mentioned above in different degrees. Practically all new CV (sea and
river) built abroad are foreseen with maximal number of balconies.

4.3 Inner Compartments Design and Level of Comfortableness

New CV are characterized by following features:
• module principles of forming living blocks in total and cabins themselves (standard
cabins) and vertical zoning of living and public compartments were used;
• restaurant, salons and bars are arranged in vertical zones separate from living
compartments;
• noise spaces are not arranged above passenger or crew cabins. Total cabins area is
16–43 m2 including balcony/French balcony (cabins are double usually);
• restaurants with total places equivalent to passengers’ number, conference hall,
bars, internet-salon, baby rooms, beauty salon, souvenir shop, fitness salon, solar-
ium are located on the open sun deck;
• lifts connected all decks, communications and toilets are foreseen for physically
challenged people.
It is unequivocally that in process of design of new vessels one should start from
the comfort level set by the customer as a quantity of conditional stars or comfort factor
which will serve as the regulating factor of the future vessel’s efficiency as function that
determines required areas of decks and compartments, overall vessel’s dimensions,
vessel’s propulsive characteristic, and vessel’s economic efficiency as a result.
As a criterion of an estimation of passenger vessels design comfortableness it is
possible to apply the five-stars scale informally used practically by all tour operators.
The quantity of stars increases according to increase of service quality and comfort. It is
necessary to emphasize especially CV with enhanced comfort (per se, yachts) and
vessels for single-day cruises (cabinless).
Such vessels often don’t relate to CV due to their significant differences in the
approaches to passenger zone and a number of special questions (availability of
additional opportunities for water types of activity, etc.). However, such vessels are
often become platform for development including for classical cruise transportations.
Example of this thesis is shown on Fig. 2 (comparison of side views of PV09 and
 Creation of River-Sea Cruise Passenger Vessels of New Generation 125

PV300, PV300VD projects). Already realized in 2017 PV09 river-sea cruise passenger
vessel became the first in “line-up” of new generation cruise vessels (see Fig. 3).

Fig. 2. Comparison of side views of PV09 and PV300/PV300VD concepts.

4.4 Other Design Features

Choice of classification society class of perspective cruise passenger vessel assumed
following:
• assignment of class based on sailing areas in accordance with planned directions of
transportations;
• definition of ice class in accordance with saved-up experience and tendencies to
extension of the navigation period including winter operation;
• assignment of economically reasonable life-time cycle of the vessel.
For safe and accelerated process of passengers’ evacuation modern evacuation
systems consisting of sleeve and accepting platform (sometimes landing of passengers
can be carried out straight on rescue device) on which passengers during evacuation go
down and with which passengers move on rescue vessel boats and rafts can be used.
Crew has to be formed with taking into account ship power plant automation,
automation of deck works (automatic anchor mooring winches), automation of process
of loading of supplies and mechanization of process of cleaning of decks (washing
machines, etc.) and with taking into account usage of specialties combination.
126 A. Egorov et al.

Fig. 3. General view of river-sea cruise vessel of PV09 project.

4.5 Theoretical “Line-up”

Choice of main particulars of river and river-sea CV is defined by way conditions, tour
operator strategy and his position in the market, adherence to those or other lines and
comfort level in view of necessities, propensities and interests of passengers for which
these vessels are built.
Such external subjective choice should be carried out basing on discrete number of
the alternatives which quite objectively have arisen in the domestic tourist market as a
result of the compromise due to choice of way restrictions and sizes of tourist groups
(400–500 people, 250–300 people, 200–250 people, 100–150 people).
Analysis of such alternatives has allowed to build parametrical line-up of river and
river-sea cruise vessels objectively demanded by native shipowners and on its basis to
develop in MEB projects of new river-sea vessels.
The most attractive are “Volga-Don Max”/”Dnipro Max” PV300, PV300VD
concepts (dimensions are defined by way conditions of Volga-Don Ship canal, Dnipro
River, passengers’ capacity is about 300 people).
Such vessels are planned to be operated on lines connecting St. Petersburg port and
Moscow port, on Volga-Don Ship Canal with exit in Azov and Black Seas, on Volga
with exit in Caspian Sea, on Volga-Baltic waterway with exit in Baltic Sea or on Kiev –
Kherson – Nikolayev – Odessa – Izmail – Constanta line.
 Creation of River-Sea Cruise Passenger Vessels of New Generation 127

4.6 Description of “Volga-Don Max”/“Dnipro Max” PV300VD

and PV300 Concepts
PV300VD concept architectural-structural type is as follows (see Fig. 4): steel, triple-
screw motor ship, with slopping stem and transom stern, with excess freeboard, with
extended forecastle, with four-tiers living superstructures through the whole vessel’s
length located amidship, with wheelhouse located fore and engine-room located aft.
Comfort level on PV300VD vessel will correspond 4–5* hotel standard.

Fig. 4. General view of PV300VD river-sea cruise vessel.

PV300VD “Volga-Don Max” class vessel will work on lines Moscow – St.
Petersburg and also will make voyages Moscow – Sochi, Astrakhan – Caspian ports.
The vessel’s class will allow also to visit ports at Black and Mediterranean seas.
PV300 “Volga-Don Max” class vessel will work on lines Moscow – St. Peters-
burg, and also will make voyages in coastal zones in accordance with assigned RRR
class M-PR 3,0.
PV300 concept architectural-structural type is as follows: steel, self-propelled river-
sea motor ship with two full-rotated rudder propellers, with plumb stem and transom
stern, with excess freeboard, with extended forecastle, with four-tiers living super-
structures through the whole vessel’s length, with wheelhouse located fore and engine-
room located aft (see Fig. 5). Comfort level on PV300 vessel will also correspond 4–
5* hotel standard.

Fig. 5. General view of PV300 river-sea cruise vessel.

PV300 river-sea cruise vessel “Mustay Karim” will be operated on long cruise lines
Moscow – St. Petersburg, Moscow – Astrakhan and Moscow – Rostov-on-Don with
128 A. Egorov et al.

replacing classical cruise vessels of 301 and 302 projects. The vessel was laid down in
2017 and is planned to be built in the beginning of 2020.
PV300VD river-sea cruise vessel “Petr Velikiy” is intended for work on lines
Moscow – St. Petersburg, Moscow – Rostov-on-Don – Sochi, Astrakhan – the ports of
the Caspian Sea, and in Black and Mediterranean Seas. The vessel was laid down in
2016 and is planned to be built in 2021. She was launched in May, 2019 (see Fig. 6).
PV300VD will be the only (as for 2019) river-sea vessel in Russia which meet
requirements of International Conventions.

Fig. 6. Launch of PV300VD river-sea cruise vessel “Petr Velikiy”.

General Arrangement
General arrangement of PV300VD vessel is presented on Fig. 7. Vessel’s main par-
ticulars are as follows: length overall 141 m; length between perpendiculars 135.56 m;
breadth overall 16.82 m; breadth 16.60 m; depth 5.50 m; draught maximal 3.20 m; air
draught from BL 16.25 m; crew and service staff 90 people; autonomy by fuel
15 days; autonomy by other stores 10 days.
General arrangement of PV300 vessel is presented on Fig. 8. Vessel’s main par-
ticulars are as follows: length overall 141 m; length between perpendiculars 135.36 m;
breadth overall 16.80 m; breadth 16.60 m; depth 5.00 m; draught maximal 3.00 m; air
draught from BL 16.25 m; crew and service staff 144 people; autonomy by fuel
15 days; autonomy by other stores 5 days.
 Creation of River-Sea Cruise Passenger Vessels of New Generation 129

Fig. 7. General arrangement of PV300VD river-sea new generation cruise vessel.

Passenger’s Accommodation and Public Rooms

Arrangement of the PV300VD cruise vessel in the variant of the Moscow River
Shipping Company (shipowner) foresees 310 passengers who travel in 155 comfort-
able cabins of different classes: 7 luxury apartments with a balcony of about 36–43 m2
area (33–35 m2 without considering balcony); 3 luxury cabins with a balcony of about
26 m2 area (21 m2 without considering balcony); 2 luxury cabins for disable people
with a balcony of about 33 m2 area (27 m2 without considering balcony); 133 standard
cabins with a balcony of about 19–21 m2 area (16–18 m2 without considering bal-
cony), see Fig. 9; 8 standard cabins of about 16 m2 area; 2 standard cabins for disable
people of about 21 m2 area.
Passenger cabins are located in the middle vertical zone on 4 superstructure decks.
Cabins on three upper decks have own balconies with sliding sections that provide
through deck pass at emergency situations.
Arrangement of the PV300 cruise vessel in the variant of “Vodohod” shipping
company (shipowner) foresees 342 passengers who travel in 171 comfortable cabins of
130 A. Egorov et al.

Fig. 8. General arrangement of PV300 river-sea new generation cruise vessel.

different classes: 1 luxury cabin of about 29 m2 area; 1 luxury cabin of about 26 m2

area; 168 standard cabins of about 19 m2 area (see Fig. 10); 1 standard cabin for
disabled people of about 29 m2 area.
Passenger cabins are located in the middle vertical zone on 3 superstructure decks.
All passenger cabins on both vessels are equipped with double beds, lavatory and
douche, air-conditioner, wardrobe, safe, TV set with satellite and internal vessel’s
channels, hairdryer, sockets 220 V and 110 V, wireless Internet (Wi-Fi), vessel’s radio
translation, external and internal telephone connection. Fore, middle and aft vertical
zones are used for providing rest of passengers on the vessels.
Forward vertical zone on PV300VD vessel includes as follows: 380 m2 panoramic
theatrical music saloon with a bar, gift shop, hairdressing saloon, a spa saloon with
sauna, fitness center, massage room and rest room with jet tub.
Middle vertical zone includes as follows: double level solarium with grill-bar,
530 m2 main restaurant.
 Creation of River-Sea Cruise Passenger Vessels of New Generation 131

Fig. 9. Design view of standard passenger cabin on PV300VD river-sea cruise vessel.

Fig. 10. Design view of standard passenger cabin on PV300 river-sea cruise vessel.
132 A. Egorov et al.

Aft vertical zone includes as follows: 170 m2 24 days bar/disco, restaurant with
110 m2 opened kitchen. One change meal period for passengers is foreseen on the
vessel. Two restaurants are applied for this purpose: main restaurant and restaurant with
opened kitchen. All restaurants have maximal available glass cover that enables to take
pleasure not only in tasty meal but also in window’s view.
6 persons passenger lifts for moving passengers between decks are installed (1 in
fore hall and 1 in aft hall).
2 cargo lifts for moving cargoes between galley and restaurant are foreseen. Cargo
lift from laundry complex on lower deck connecting all decks of the vessel is provided.
Fulfillment of Safety Requirements
PV300VD is not only up-to-date high comfortable and handy vessel but also safe,
ecology “clear” passenger one that fully meets requirements of all International Con-
ventions (IC), including IC for the Safety of Life at Sea (SOLAS), IC for the Prevention
of Pollution from Ships (MARPOL), International Regulations for Preventing Colli-
sions at Sea (COLREGS). Additionally, the vessel meets all Russian requirements such
as Sanitary Regulations and Norms, Fire Safety Regulations, Technical Regulations for
Safety of Inland Water transport Objects.
The vessel fully satisfies SOLAS norms and Class requirements of damage stability,
including probability assessment of hull subdivision, special requirements for pas-
senger vessels of Russian Maritime Register of Shipping, and all normative require-
ments of Russian River Register Rules.
PV300 vessel is planning to be operated on internal waterways without international
voyages, so she only meets Russian requirements.
Fire safety of new concepts is provided by division of vessels into main vertical
zones with constructive and thermal barriers; subdivision of accommodation spaces
from other spaces with thermal and constructive barriers; exception of application of
inflammable materials; detection of any seat of fire in zone of its emergence; restriction
of distribution and fire extinction in places of its emergence; protection of means of
escape and accesses to them for fight against the fire and readiness of fire-protection
equipment to fast application.
From each waterproof compartment located below bulkhead deck two outmost from
each other exits are made.
Special requirements to equipment of ventilation systems of vessels were revealed
on design stage. Spaces which are in different vertical fire zones are not connected
between each other through vent ducts. Vent ducts are arranged in way so that they
could be blocked from the outside of ventilated spaces.
Ventilation control is executed from two posts located on maximal outmost from
each other.
Except meaningful measures of fire protection, complex of actions for fit out of
accommodation, public and service spaces by automatic systems of fire detection and
automatic sprinkler systems in local vertical fire zones is executed.
Hull Structure
Main 7 watertight transverse bulkheads are installed on both vessels. They divide hull
into 8 impermeable compartments. At the aft part of vessels from each side two fast
150 persons’ capacity life-boats and two workboats are installed. For safe and
 Creation of River-Sea Cruise Passenger Vessels of New Generation 133

accelerated process of passengers’ evacuation 2 evacuation systems with closed

inflatable self-righting rafts (4 rafts on 101 persons each) are installed on vessels.
Earlier such systems on river and river sea vessels were not applied.
Middle part of vessel’s hull is constructed by mixed framing system (see Fig. 11).
Main and upper decks are constructed by longitudinal framing system on whole length.
Lower, boat and tent decks are constructed by transverse framing system on whole
length. Bottom and double bottom are constructed by longitudinal framing system in
middle part and transverse framing system in ends and in ER. Outer side is constructed
by transverse framing system on whole length. Constructions of ends are constructed
by transverse framing system.

Fig. 11. Transverse cross-sectional view of PV300VD vessel’s hull.

Construction of sides and bottom is reinforced in accordance with RRR rules and ice
class “Ice 30”.
For ensuring of maximal glazing execution of cuts for windows and doors practi-
cally from deck to deck required solution of number of problems connected with
strength of attachment fittings and material of windows.
Construction of vessel’s decks in area of big cutaways under atriums is reinforced
with thickened sheets and additional connections.
CFD Optimization of Hull Form
CFD-modelling tests were executed to optimize hull form (underwater part of hull is
the same on PV300 and PV300VD projects), see Fig. 12 and Fig. 13. Researches have
shown that traditional single hull type passenger vessel can be recommended.
According to received results of calculation full speed makes 26.3 km/h (14.2 knots).
Choice of Main Engine Type
Main power plant on PV300VD vessel is diesel-electric. It consists of four main diesel-
generators of alternative current 690 V, 50 Hz with electrical power 1140 ekW each,
134 A. Egorov et al.

Fig. 12. CFD modelling in flowvision software. Pressure distribution. Fore and aft view.

Fig. 13. CFD modelling in flowvision software. Free surface. Fore and aft view.

feeding electric propulsion plant and other customers. Classic river vessels in Russia
have diesel power plant.
For cruise passenger vessel, for which needs of passenger part for energy are
commensurable with needs for energy for the movement, use of electric propulsion
provides considerable advantages on building stage and, especially, in operation.
Electric propulsion has allowed to exclude on building stage one diesel-generator on
power and to reduce in general number of diesels –4 main diesel-generators are
installed instead of 3 main engines and 3 diesel-generators. During standard operation
two main diesel-generators are providing power to propulsion motors and other vessel
consumers. The third diesel-generator is in reserve or works when speed is increased or
under bad weather conditions. And it is possible to carry out technical maintenance on
the fourth diesel-generator. Diesel-electric power plant provides higher profitability on
low and half speeds, and also during vessel voyages with frequent stops and maneu-
vers. Electric drive allows to apply screw-propellers with the best performance coef-
ficient and to use primary engines in the most economy mode because between primary
engine and screw practically any gear ratio can be chosen.
Auxiliary power plant consists of: emergency diesel-generator of 214 kW electrical
power; auxiliary boiler plant consisting of two oil fired thermal oil heaters with heat
 Creation of River-Sea Cruise Passenger Vessels of New Generation 135

power 800 kW each, fore thermal oil exhaust gas heaters with heat power 250 kW
each; auxiliary mechanisms and apparatus working for power plant.
On PV300 vessel main power plant is classic diesel one. It consists of two main
diesel engines Wartsila 6L20 of maximal continuous power 1200 kW each. Auxiliary
power plant consists of: three diesel-generators of electrical power of 760 ekW each;
emergency diesel-generator of 214 kW electrical power; auxiliary mechanisms and
apparatus working for power plant. Auxiliary boiler plant consists of: one oil fired
thermal oil heater with heat power 800 kW; three thermal oil exhaust gas heaters with
heat power 170 kW each.
As fuel for main engines IFO380 with viscosity 380 cSt at 50°C and low sulfur
content is used. For emergency diesel-generator MDO with low sulfur content with
flash point over 60°C is used.
Improvement of Maneuverability and Grounding of Propulsion Complex Type
In aft part of PV300VD vessel three full-rotate fixed pitch rudder propellers (FPRP)
with mechanical power on input shaft is of 1000 kW each are installed. Two FPRP
with 1200 kW power each are installed on PV300 vessel. For improvement of con-
trollability at low speeds, at passing narrow waters and at moorings bow thruster is
foreseen. Traditionally three-shaft unit was used on big Russian river CV as propulsion
complex.
As special researches have shown better maneuverability of the vessel is provided
on wide angles at FPRP turn. In process of decrease in motion speed of the vessel
advantage of FPRP increases up to the most low speed <3 kn when the vessel with
rudders becomes almost uncontrolled and FPRP allow to carry out vessel turn in pos.
Usage of FPRP is especially effective while vessel’s operation in river when passing
narrow places and numerous river turns the vessel cannot move with maximum speed
and efficiency of rudders falls down. Thus, the vessel equipped with FPRP can pass the
same way section faster. If to operate vessel on one FPRP there is no loss of con-
trollability and thus vessel speed falls slightly to 7.5–8.0 kn. In economy mode of
operation on one FPRP considerable fuel saving exists.
Purchase value of FPRP is approximately equal to the total cost of speed-reduction
gear, shaft line, stern gear, screw, nozzle, steering gear and rudder.
Therewith FPRP installation is simpler: there is no need for laying of shaft line, thus
less qualified personnel can be used, time of FPRP installation is less than time of
mechanical transmission installation. All this allows to cut down expenses on
mounting. FPRP installation can be carried out afloat after vessel’s launching. Delivery
and installation of equipment of rudder propeller system can be made for some months
later that also lead to economy of funds.
For providing full control over the vessel’s movement while maneuvering on slow
speed, mooring operations and docking installation of vessel’s control system which
unites all propulsion units of the vessel (FPRPs and bow thrusters) in integral module is
provided and thus control of all these propulsion units is executed by means of one
joystick. Therefore, FPRPs have been applied on PV300 and PV300VD concepts.
136 A. Egorov et al.

5 Conclusions

The most serious restraining factor of growth of native cruise tourism industry is lack
of modern river and river-sea cruise vessels. For the solution of problem of updating it
is offered to connect river and some coastal sea routes on the basis of modern cruise
vessels of PV300, PV300VD and PV09 projects. PV09, PV300 and PV300VD vessels
are not only up-to-date high comfortable and handy vessels but also safe, ecology
“clear” passenger carriers that fully meet requirements of all International (PV300VD
vessel) and Russian requirements (PV09, PV300, PV300VD vessels).
PV300 and PV300VD concepts are the first native cruise vessels which were
designed in accordance with maximization of main dimensions methodology and
module principles of forming living blocks and public places. These decisions allowed
to increase comfort level for passengers and to make river-sea passengers’ trans-
portations economically reasonable. For ensuring of maximal glazing execution of cuts
for windows and doors problems connected with strength of attachment fittings and
material of windows were solved.
Application of FPRP allowed to improve maneuverability of PV300 and PV300VD
vessels. On existing native river and river-sea vessels triple-screw propulsion complex
is installed.
Hull form was specially designed and optimized by means of CFD.

References
1. Egorov, A.G., Egorov, G.V.: Analysis of decommissioning trends of river cruise vessels and
prognosis of river fleet structure with definition of most demanded types of vessels for
conversion and modernization (in Russian). J. Morskoy Vestnik 1(69), 21–27 (2019)
2. Egorov, G.V., Ilnitskiy, I.A., Kalugin, Ya.V.: River-sea cruise passenger vessels for
Russian IWW (in Russian). Science-technical collected book of Russian Maritime Register of
Shipping, vol. 36, pp. 255–276 (2013)
3. Egorov, G.V., Ilnitskiy, I.A., Kalugin, Y.V.: “Line-up” of river cruise passenger vessels (in
Russian). J. ONMU Report. 2(38), 20–40 (2013)
4. Egorov, G.V.: Analysis of European river cruise experience and its influence on native cruise
vessels of new generation (in Russian). J Morskoy Vestnik 3(43), 23–28 (2012)
 Research on the Accessibility Evaluation
of the Cruise Passage Design Based
on Passenger Movement Simulation

Zhan Chen1,2, Wei Cai1,2(&), Yazhong Wang1,2, and Yigang Wu1,2

1
School of Transportation, Wuhan University of Technology,
Wuhan, Hubei, China
[email protected]
2
Cruise and Yacht Design Institute, Wuhan University of Technology,
Wuhan, Hubei, China

Abstract. Cruise passage accessibility is defined as a kind of difficulty degree

of getting close to the demanded place within the expected time in a typical
cruise scenario. In consideration of the space utilization of cruise ships and the
efficiency of passenger movement, the concept of cruise passage accessibility is
proposed to improve cruise passage design. This study aims to build the pas-
senger movement model under different scenarios in the complex cruise layout
and the evaluation is carried out with the simulation results. According to the
analysis of passenger movement characteristics, the social force model is
improved. Besides, Analytic Hierarchy Process and Fuzzy Comprehensive
Evaluation Method (AHP-FCA) are combined in the cruise passage accessibility
evaluation. With the improved social force model, the passenger movement
model on the Mediterranean Armonia Cruise are constructed through the
AnyLogic simulation platform. Then, the cruise passage accessibility is evalu-
ated by the simulation results and the evaluation method. Through the acces-
sibility evaluation method, some effective suggestions are proposed to improve
cruise passage accessibility.

Keywords: Cruise passage
Passenger movement
Social force model

Simulation experiment
Accessibility evaluation

1 Introduction

Cruise passage accessibility is an important measurement of cruise passage design

quantity and the passenger movement efficiency. In order to study cruise passage
accessibility evaluation, Anylogic Software Simulation and AHP-FCA evaluation
method are applied.
Currently, research on evaluation of ship layout design has made good progress.
Brown and Salcedo (2003) used multi-objective genetic algorithm to construct the
evaluation system of ship layout design in the preliminary design of cruise [1]. Anil
(2005) used parametric space exploration techniques to reduce ship design dimensions
and simplify ship design evaluation [2]. Chen Miao (2009) deduced relationship for-
mula between the speed and density of personnel movement by the dynamic equation

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 137–155, 2021.
https://doi.org/10.1007/978-981-15-4680-8_10
138 Z. Chen et al.

in a crowded state, and evaluated the ship accessibility by the evacuation time [3].
Gesine Hofinger et al. (2014) analyzed the influence of psychological factors in
evacuation process [4]. Li Feng (2014) established the ship accessibility evaluation
index system with the environment, layout, process, and time indicators as the
framework [5]. Marina Balakhontceva et al. (2016) proposed a multi-agent model and
crowd dynamics modeling method by considering the evacuation process of ship
movement [6]. However, cruise passage accessibility has not been clearly defined in the
previous studies, and the evaluation of cruise passage design is still inadequate.
Therefore, passenger movement simulation is constructed to improve cruise pas-
sage accessibility. Combined with the cruise passage layout characteristics and
movement characteristics, the social force model is improved. Then, AHP - FCA
method is used to support the evaluation method. Finally, the Mediterranean Armonia
cruise ship is taken as an example to simulate three kinds of passenger movement
scenarios. Through the analysis of simulation results, the evaluation score, which is
obtained by AHP-FCA evaluation method, comprehensively reflects the pros and cons
of cruise passage accessibility, and has a good guiding significance for the cruise
passage design.

2 Passenger Movement Characteristics and Improvement

of Social Force Model

2.1 Passenger Movement Characteristics

The distribution and movement efficiency of cruise passengers are mainly subject to the
surrounding environment and psychological quality of passengers [7].
For dining activities, the passenger behaviors are simple with the assistance of
staffs. When passengers enter into or leave off the dining room, the passenger density in
the dining room is highest.
For entertainment activities, the distribution of passengers will be gathered in a
short time, and it is possible to cause the phenomenon of conformity, small groups and
overcrowding.
For emergency evacuation, the behaviors of passenger movement are always dif-
ferent from the usual:
1) Due to panic psychology, passengers expect to move faster than usual.
2) In the narrow areas of cruise passage, passenger movement could produce over-
crowding and even colliding.
3) Some passengers are unfamiliar with the cruise passage layout so that they
sometimes get wrong in evacuation direction.
4) Nervousness would affect the efficiency of passenger movement.
5) The number of people in the different evacuation exits is uneven.
 Research on the Accessibility Evaluation of the Cruise Passage Design 139

2.2 Improvement of Social Force Model

Through the analysis of the movement characteristics of passengers, influence factors
would be considered to improve the social force model and build a model for simu-
lation experiment analysis.
According to the concept of social force, Helbing et al. introduced dynamic
characteristics and constructed a social force model [8]
8 P P
> vo ðtÞeo ðtÞv ðtÞ
>
> mi dv i
¼ mj i i si i þ j6¼i fij þ w fiw
< n dt h i o
r d
fij ¼ Ai exp ij Bi ij þ kgðrij  dij Þ nij þ jgðrij  dij ÞDvtji tij ð1Þ
>
> n h i o
>
: fiw ¼ Ai exp d
ri
Bi
iw
þ kgðri  diw Þ niw  jgðri  diw Þðvi
tiw Þtiw

where, mi is the mass of person i, voi ðtÞ is the expected speed of person i, eoi ðtÞ is the
unit vector of the person moving in the desired direction at time t, vi(t) is the actual
motion velocity vector of person i at time t, si is the relaxation time of i, Ai is the
intensity of interaction, Bi is the range of interaction. The parameters k and j determine
the obstruction effects in case  of physical
 interactions. Besides, rij = ri + rj is the sum
of their radii ri and rj. dij ¼ ri  rj  is the distance between 2 people’s mass centers,
nij is normalized vector pointing from person j to person i, tij is the tangential direction,
Dvtji is the tangential velocity difference, diw is the shortest distance between person
i and barrier j, and niw is the direction perpendicular to it.
In combination with the above passage design characteristics and the movement
characteristics of passengers, the social force model is improved: one is the influence of
introducing panic psychology, conformity psychology and small group behavior on the
expected speed of passengers; the other is the influence of introducing congested areas
on the movement space of passengers.
Panic Psychology. Sharad Sharma et al. indicated that panic could slow movement
during an evacuation [9]. Ppanic is used to express the impact of panic factors on the
actual velocity of personnel [10], and the actual expected velocity, as amended

vi ðtÞ0 2 ðvmin
i ðtÞ  Ppanic  vD ðtÞ; vi
max
ðtÞ þ 2  Ppanic  vD ðtÞÞ ð2Þ

where, vi ðtÞ0 is revised actual speed, vmin

i ðtÞ and vi ðtÞ are the maximum and minimum
max

of the normal movement speed of person i, and vD ðtÞ is a unit of speed (m/s).
Small Group Behavior. In the emergency evacuation, dining activities or large-scale
entertainment activities, cruise passengers have the characteristics of gathering toge-
ther. The small group phenomenon has a more significant impact on the movement
efficiency of passengers. Therefore, when the model is running, there is a certain
probability that passengers will move in a small group, and the passenger group
behavior is expressed by Lakoba’s definition of attraction [11]
140 Z. Chen et al.

 
fiq ¼ Fi exp ðriq  diq Þ=Bi niq ð3Þ

where, q is a companion of person i, riq is the sum of their radii, diq is the distance
between their centers of mass, Bi is the range of the force between them, Fi is negative
constant, and niq is the acting force is opposite to that of the original model.
Guidance of Crowded Areas. Due to the high tonnage and large passenger capacity
of cruise ships, it is possible to cause local regional congestion in the process of
evacuation, which reduces the evacuation efficiency of passengers. Therefore, it is
necessary to arrange staff to guide the evacuation path and direction. The guiding path
of evacuation is determined by specific scenes, while the guiding direction is the
desired direction of passengers

eo ðtÞ þ timpact eleader

eoi ðtÞ0 ¼  io  ð4Þ
ei ðtÞ þ timpact eleader 

where, eleader is the direction guided by cruise crew, timpact is the guiding force of staff,
and the range is [0, 1], eoi ðtÞ0 is the direction of desired velocity after correction.

3 Accessibility Evaluation Method

AHP AHP ̶ FCA FCA

Hierarchy Structure Model

Evaluation Grade Criteria Selection

Judgment Matrix
Membership Matrices Construction
Single-factor Index Weight Calculation

2nd-level Indicator Strata Evaluation

Consistency Matrix
Checking Adjustment
1st-level Indicator Strata Evaluation

Combined Weight Calculation

Weighted Average Comprehensive
Evaluation
Consistency
Checking

Evaluation
Conclusion

Fig. 1. Accessibility evaluation process.

 Research on the Accessibility Evaluation of the Cruise Passage Design 141

According to the definition of cruise passage accessibility, the accessibility eval-

uation of cruise passage needs to be carried out from the perspectives of passage layout
and passenger movement. Besides, accessibility evaluation is a synthesis evaluation
which refers to many factors.
Therefore, based on the simulation results of the social force model, AHP-FCA
comprehensive evaluation method is applied to evaluate cruise passage accessibility
[12]. The specific evaluation process is shown in Fig. 1.

3.1 Index System and Empowerment Analysis

Hierarchical Structure Model. By taking the accessibility of the overall passage

layout of cruise ships as the overall goal, a 3-level evaluation index system is con-
structed (Fig. 2).
In Fig. 2, U is the comprehensive evaluation set, u1, u2, and u3 respectively rep-
resents the evaluation set under the scenarios of emergency evacuation, dining activ-
ities and large-scale entertainment activities; uij (i = 1, 2, 3; j = 1, 2, 3, 4, 5) represents
each indicator in the 2nd-level indicator strata.

Cruise passage
accessibility(T)

Emergency leisure and

Catering activity(B2)
evacuation(B1) entertainment(B3)

u11 u12 u13 u14 u15 u21 u22 u23 u24 u31 u32 u33 u34

u11 u21 u31 Mean value of movement distance

u12 u22 u32 Number of stairways
u13 u23 u33 the maximum of crowd density
u14 u24 u34 movement time
u15 passage width

Fig. 2. Evaluation index system of cruise passage accessibility.

Single-Factor Index Weight. The judgment matrix, weight vector and consistency
ratio of the following levels are obtained by pairwise comparison and characteristic root
method (see Table 1, Table 2, Table 3, and Table 4).
142 Z. Chen et al.

Table 1. 1st-level index Bi in hierarchical single arrangement

A B1 B2 B3 b
B1 1 4 5 0.6833
B2 1/4 1 2 0.1998
B3 1/5 1/2 1 0.1169
kmax 3.0246
CI 0.0123
RI 0.58
CR 0.0212

Table 2. 2nd-level index uij in hierarchical single arrangement (1)

B1 C1 C2 C3 C4 C5 w
C1 1 2 2 1/2 1 0.1956
C2 1/2 1 1 1/4 1/2 0.0978
C3 1/2 1 1 1/4 1/2 0.0978
C4 2 4 4 1 3 0.4274
C5 1 2 2 1/3 1 0.1814
kmax 5.0198
CI 0.0049
RI 1.12
CR 0.0042

Table 3. 2nd-level index uij in hierarchical single arrangement (2)

B1 C6 C7 C8 C9 w
C6 1 2 3 1/3 0.2372
C7 1/2 1 2 1/4 0.1406
C8 1/3 1/2 1 1/4 0.0913
C9 3 4 4 1 0.5309
kmax 4.0875
CI 0.0292
RI 0.89
CR 0.0328
 Research on the Accessibility Evaluation of the Cruise Passage Design 143

Table 4. 2nd-level index uij in hierarchical single arrangement (3)

B1 C10 C11 C12 C13 w
C10 1 2 3 1/3 0.2372
C11 1/2 1 2 1/4 0.1406
C12 1/3 1/2 1 1/4 0.0913
C13 3 4 4 1 0.5309
kmax 4.0875
CI 0.0292
RI 0.89
CR 0.0328

According to the consistency checks of single-level ranking (CR  0.10), the

weight values of 2nd-level index uij are calculated.
Combined Weight. According to the consistency checks of single-level ranking
(CR  0.10), the weight values of 1st-level index Bi are calculated, and the total
hierarchical ranking is calculated to obtain the synthetic weight of 2nd-level index uij
relative to target T (Table 5).

Table 5. Combined weight calculation.

2 -level indicator strata 1st-level indicator strata
nd
Combined weigh (aij)
B1 B2 B3
0.6833 0.1998 0.1169
u11 0.1956 0.1337
u12 0.0978 0.0668
u13 0.0978 0.0668
u14 0.4274 0.2920
u15 0.1814 0.1240
u21 0.2372 0.0474
u22 0.1406 0.0281
u23 0.0913 0.0182
u24 0.5309 0.1061
u31 0.2372 0.0277
u32 0.1406 0.0164
u33 0.0913 0.0107
u34 0.5309 0.0621
144 Z. Chen et al.

According to the result of weight set calculation of AHP method, the combined
weight of the 2nd-level indexes is determined
8
>
> A ¼ fa1 ; a2 ; a3 g
<
A1 ¼ fa11 ; a12 ; a13 ; a14 ; a15 g
ð5Þ
>
> A2 ¼ fa21 ; a22 ; a23 ; a24 g
:
A3 ¼ fa31 ; a32 ; a33 ; a34 g

where, A is combined weight set, A1, A2, and A3 respectively represent the weight set of
evaluation factors under the scenarios of emergency evacuation, dining activities and
entertainment activities, ai (i = 1, 2, 3) is the weight set of 1st-level indicator strata, and
aij (i = 1, 2, 3; j = 1, 2, 3, 4, 5) is the weight set of 2nd-level indicator strata.

3.2 Fuzzy Evaluation and Scores

Based on their own experience and relevant specifications, 10 experts who have
experience in ship general arrangement evaluation were invited to score each index of
the cruise passage accessibility evaluation index system on a 5-point scale. According
to the scoring results, the membership degree rijk of the second-level indicator uij on
each rating grade vk is obtained. Then the fuzzy evaluation matrix Ri of the single-factor
ui is established
8 Nijk
>
> 2 rijk ¼ 3
>
>
N
< ri11 ri12


ri1p
6 ri21 ri22


ri2p 7 ð6Þ
>
> Ri ¼ 6
4






7
>
> rijk


5
:
rin1 rin2


rinp

where, Nijk represents the number of people who have made vk grade evaluations on the
2nd-level indicator uij, N = 10 means total number of evaluators.
Fuzzy evaluation method is used to evaluate 1st-level indicator (u1, u21, and u3), and
single factor evaluation vector Bi is constructed
8
< B1 ¼ A1
R1
B ¼ A2
R2 ð7Þ
: 2
B3 ¼ A3
R3

The 2nd-level fuzzy comprehensive evaluation vector is established by synthesizing

single factor evaluation vector

R ¼ ½B1 ; B2 ; B3 T
ð8Þ
B¼A
R

The evaluation vector B is weighted by score, and the comprehensive evaluation

score of passage accessibility evaluation is obtained
 Research on the Accessibility Evaluation of the Cruise Passage Design 145

P ¼ B
VT ð9Þ

where, V = [5, 4, 3, 2, 1].

If the evaluation score is equal to or more than 3 marks, passage accessibility meets
the requirement. Besides, through the result of 10 experts’ single-factor evaluation,
passage accessibility can be improved with specific purpose.

4 Case Study

Based on the improved social force model and Anylogic 7.3.5 simulation platform [13],
the Mediterranean Armonia Cruise Ship is used as the prototype to construct passenger
movement simulation tests under three scenarios: emergency evacuation in cabin area,
dining in dining area, and large-scale entertainment activity in recreation area. To make
the simulation results are approach to the actual condition, we respectively improve the
simulation of static environment and dynamic environment. For static environment
simulation, the deck model is constructed with the actual size and shape. For dynamic
environment, the social force model is improved by panic psychology, small group
behavior, and guidance of crowded areas. The specific modeling process [14] is shown
in Fig. 3.

Set the relevant Create a logical

Analysis of scene Draw 3-D Layout
physical flowchart of
characteristics of the scene
properties people behavior

function
Java variable Set the feature
code Implement time
attribute of each
Export data Model running behavior attribute
module's matter-
correlation
element

Fig. 3. Anylogic simulation modeling flowchart.

4.1 Scene Simulation

Emergency Evacuation in Cabin Areas. There are 5 main staircases on each of the
three decks (Fig. 4). The deck parameter information is shown in Table 6. The pas-
sengers start from their own living cabin and move to the cabin door, then they select
the nearest stairway and the gathering place for evacuation. When all of passengers
arrive on the lifeboat deck (Deck 7), the evacuation on this floor will be regarded as the
end. Professor Helbing D. [8] pointed out that “under normal conditions, the average
146 Z. Chen et al.

Fig. 4. Cabin area scene on Decks 8–10.

Table 6. Relevant parameters of Decks 8–10.

Deck Cabin quantity Passenger quantity
8 300 968
9 297 709
10 182 376

speed of evacuees in the horizontal passage is 1.25 m/s”. However, due to the unfa-
vorable environment of the cruise ship and the influence of panic psychology, the initial
speed is set to 0.7–1.2 m/s and the movement speed is 1.2–1.4 m/s.
Dining in Marco Polo Main Dining Room. Marco Polo Main Dining Room (located
on Deck 5) has an area of 1050 m2 and a maximum seating capacity of 718 people.
Among them, 352 people enter from Stair 1 to Area 1 and 366 people enter from Stair 2
to Area 2. And there are four exits for passengers (Fig. 5).
Passengers randomly select the entrances (Stair 1 and Stair 2) and enter into the
dining room. Then 3–5 passengers gather as a small group and randomly select the
dining table. The dining time is about 20–35 min. When all of passengers have reached
the exits after meal, the simulation is regarded as an end. As the dining room is divided

Fig. 5. Marco Polo Main Dining Room.

 Research on the Accessibility Evaluation of the Cruise Passage Design 147

into Area 1 and Area 2, there are 2 source points of passenger flow. Due to the average
movement speed of passenger flow is about 1.30 m/s under normal conditions and the
internal layout design is relatively compact, the movement speed of passenger flow is
set to 1.1–1.3 m/s.

Fig. 6. Entertainment area on Deck 5–6.

Large-Scale Entertainment Activity. Teatro La Fenice Theatre with an area of

1135 m2 and a maximum capacity of 557 people, has 2 decks (Decks 5–6). According
to the distribution of cruise passengers, the passenger quantity of Decks 5–6 are set to
385 and 172 respectively. Besides, there are four entrances in the theatre, and there are
eight staircases located on the bow of Decks 5–6 (Fig. 6).
Passengers randomly appear at any staircase of the entertainment deck and guided
by the staff to move towards the theatre in small groups. When all the passengers arrive
at the entrances of the theatre, the simulation is deemed to be over. As the theatre is
divided into 2 parts, there are 8 source points of passenger flow. Under normal con-
ditions, the average speed of young people is 1.32 m/s [15]. According to the complex
environment and the different individual behaviors, the speed of passenger flow is set to
1.2–1.4 m/s.

4.2 Results and Discussion

Results of Emergency Evacuation Simulation. As is shown in Fig. 7, although the

number of evacuees at each exit is approximately equal, the evacuation density of Exit
3 (or Exit 4) are smaller than other exits. In the cabin layout design, there should be
more 3-people rooms and 4-people rooms near Exit 3 and Exit 4.
148 Z. Chen et al.

a Passenger evacuation efficiency. b Number of evacuees at each exit.

c Evacuation speed of each exit. d Evacuation density of each exit.

Fig. 7. Emergency evacuation on Deck 8.

Figure 8 shows that number of evacuees, evacuation speed, and evacuation density
of each exit are relatively close in the whole progress. The results indicates that the
layout of this deck are reasonable.
As is shown in Fig. 9, it takes 167 s from the successful evacuation of first pas-
senger (at 39 s) to the successful evacuation of all passengers. And the number of
evacuees, speed and density at each exit are directly proportional. The numbers of
evacuees at Exit 1 and Exit 2 are respectively 120 and 112. While the numbers of
evacuees at Exit 3–5 are 60, 36 and 48, respectively. The evacuation densities of the
exits are unequal: Exit 1 and Exit 2 are crowded, while there are much fewer people at
other exits.
 Research on the Accessibility Evaluation of the Cruise Passage Design 149

a Passenger evacuation efficiency. b Number of evacuees at each exit.

c Evacuation speed of each exit. d Evacuation density of each exit.

Fig. 8. Emergency evacuation on Deck 9.

Due to the small number of people and the low evacuation density, the evacuation
time on Deck 10 is shortest (Table 7). The evacuation time of passengers in the cabin
areas shall be the largest one of the three, so T is 252 s.
Results of Dining Activity Simulation. As shown in Fig. 10, passengers start leaving
the dining room at 22 min. And it takes 48 min 23 s to serve 718 passengers in the
main dining room.
Although the maximum passenger capacity of Area 1 (352 people) is equal to that of
Area 2 (366 people), the size of Area 1 is about twice as the size of Area 2. Besides, the
width of the exits in Area 1 is wider than the width of the exits in Area 2. Therefore, the
flow densities of Entrance 3 and Entrance 4 are higher than those of Entrance 1 and
Entrance 2 (Figs. 10 and 11).
150 Z. Chen et al.

a Evacuation efficiency. b Number of evacuees at each exit.

c Evacuation speed of each exit. d Evacuation density of each exit.

Fig. 9. Emergency evacuation on Deck 10.

Table 7. Statistics of emergency evacuation simulation results.

Deck Width of Maximum evacuation Maximum Evacuation
passage (m) speed (people/hour) evacuation density time (s)
8 1.38 3072 0.44 220
9 1.36 2100 0.29 252
10 1.28 1776 0.22 167

Results of Large-Scale Entertainment Activity Simulation. Figure 12 shows that it

takes 379 s for 557 passengers to arrive at the entrance of the theatre. Since there is no
fixed form in the recreation areas, the density and speed of passenger flow at each
entrance are greatly different and the passage in the recreation areas is prone to local
overcrowding (Fig. 13).
 Research on the Accessibility Evaluation of the Cruise Passage Design 151

(a) Dining efficiency. (b) Speed of passenger flow at each exit.

(c) Flow density at each exit.

Fig. 10. Passenger movement in the main dining room.

(a) At beginning. (b) In the end.

Fig. 11. Movement density of people in Marco Polo’s main dining room.
152 Z. Chen et al.

(a) Movement efficiency. (b) Speed of passenger flow at each exit.

(c) Passenger flow density at each exit.

Fig. 12. Passenger movement in the recreation areas.

Fig. 13. The movement density of people in the recreation areas.

Evaluation and Analysis of Accessibility. Table 8 shows the layout characteristics of

cruise passage and the simulation results under three scenes. According to the simu-
lation, 10 experts’ evaluation scores is given in Table 9. Combined with the above
results and AHP-FCA comprehensive evaluation method, the Fuzzy score of MSC
Armonia Cruise passage accessibility is 3.4154, which indicates that there are still
problems in some indicators of the accessibility of the cruise passage design:
 Research on the Accessibility Evaluation of the Cruise Passage Design 153

1) In the dining and recreation areas, the accessibility index is good at stairway
quantity and the average distance of passenger movement, but insufficient consid-
eration is given to the maximum evacuation density of the passage. For instance, the
passage in the recreation areas is so narrow to cause possibly the phenomenon of
personnel crowding.
2) Staircase layout of cabin areas is unreasonable. One part of Staircases are crowded,
and another one part of Staircases are not. For example, on Deck 10, the numbers of
people escaping from Exit 1 and Exit 2 are respectively 120 and 112, while the
numbers of people escaping from Exit 3, Exit 4 and Exit 5 are respectively 60, 36
and 48. It indicates that the passenger quantity of Exit 3, Exit 4, or Exit 5 are even
less than half of the passenger quantity of Exit 1 or Exit 2.

Table 8. Accessibility index data of the cruise passage.

Mean value of Stairway Maximum of Passage Movement
movement quantity crowd width time (s)
distance (m) density (m)
Cabin 47 5 0.44 1.32 252
areas
Dining 9.4 3 0.42 / 2903
areas
Recreation 8.9 6 0.84 / 379
areas

Table 9. Statistics of 10 experts’ evaluation scores.

st
1 -level indicator strata 2nd-level indicator strata Number of
evaluation
scores (from
5 to 1)
5 4 3 2 1
Emergency evacuation u11 0 4 6 0 0
u12 5 2 3 0 0
u13 0 0 5 5 0
u14 3 4 1 2 0
u15 1 4 2 3 0
Catering activities u21 2 2 3 3 0
u22 5 3 2 0 0
u23 0 2 4 4 0
u24 0 2 4 3 1
Leisure and entertainment u31 3 4 2 1 0
u32 4 4 2 0 0
u33 0 0 6 3 1
u34 1 3 4 2 0
154 Z. Chen et al.

5 Conclusions

In order to improve cruise passage design, the cruise passage accessibility is proposed.
Based on passenger movement characteristics, social force model is improved, and
passenger movement simulation under different scenarios is constructed. Through the
simulation results, AHP-FCA evaluation method is applied and the fuzzy score of the
cruise passage accessibility is obtained. This study can comprehensively evaluate
cruise passage accessibility and effectively guide cruise passage layout design.
In this study, three concluding remarks could be drawn as listed below:
1) Considering the limitation of passage layout, the minimum passage width should be
noticed to avoid the phenomenon of local overcrowding.
2) As shown in Table 9, the evaluation scores under u13, u23, and u33 is relatively
lower than other factors. So as to improve the maximum of crowded density,
staircase layout should be adjusted by passenger density in the special area, rather
than evenly distributed along length direction.
3) The passenger movement simulation and AHP-FCA evaluation method need to be
continually improved in practice. Through the improvements, the simulation results
and the proposed method can be more approach to the actual condition.

Acknowledgement. This work was funded by Special Project of Large-scale Cruise ShipResearch
and Development ([2017] No. 614), Ministry of Industry and Information Technology,China, and
supported by Cruise and Yacht Design Institute, School of Transportation, WuhanUniversity of
Technology, China. The authors are pleased to acknowledge these supports.

References
1. Brown, A., Salcedo, J.: Multiple-objective optimization in naval ship design. Naval Eng.
J. 115(4), 49–62 (2003)
2. Anil, K.A.: Multi-criteria analysis in naval ship design. Naval Postgraduate School,
Monterey California (2005)
3. Chen, M.: Preliminary study on the simulation of passenger evacuation performance of
passenger ship. Harbin Engineering University (2009)
4. Hofinger, G., Zinke, R., Künzer, L.: Human factors in evacuation simulation, planning, and
guidance. Transp. Res. Procedia 2, 603–611 (2014)
5. Li, F.: Research on accessibility optimization and evaluation method of large ships. Ph.D.
thesis, Department of Shipbuilding Engineering, Harbin Engineering University (2014)
6. Balakhontceva, M., Karbovskii, V., Sutulo, S., Boukhanovsky, A.: Multi- agent simulation
of passenger evacuation from a damaged ship under storm conditions. Procedia Comput. Sci.
80, 2455–2464 (2016)
7. Park, J.-H., Lee, D., Kim, H.: Development of evacuation model for human safety in
maritime casualty. Ocean Eng. 31(11), 1537–1547 (2004)
8. Helbing, D., Farkas, I., Vicsek, T.: Simulating dynamical features of escape panic. Nature
407(6803), 487–490 (2000)
9. Sharma, S., Singh, H., Prakash, A.: Multi-agent modeling and simulation of human behavior
in aircraft evacuations. IEEE Trans. Aerosp. Electron. Syst. 44(4), 1477–1488 (2008)
 Research on the Accessibility Evaluation of the Cruise Passage Design 155

10. Deng, Y., Feng, Y., Hu, L.: The impact of psychology and randomness on the evacuation of
pedestrians: a modification of social force model. Syst. Eng. 11, 144–148 (2004)
11. Lakoba, T.I., Kaup, D.J., Finkelstein, N.M.: Modifications of the Helbing Molnar Farkas
Vicsek social force model for pedestrian evolution. Simul. Trans. Soc. Model. Simul. Int. 81
(5), 339–352 (2005)
12. Li, G., Huang, Z., Wei, S.: Support effectiveness evaluation based on AHP-fuzzy of ship
spare parts. Ship Electron. Eng. 34(09), 108–111, 118 (2014)
13. Dan, Z., Peng, X.: Research on evacuation of public buildings based on anylogic. Comput.
Appl. Softw. 35(06), 107–109, 128 (2018)
14. Cao, N., Chen, Y., Qu, Z., Zhao, L., Bai, Q., Yang, Q.: Pedestrian route choice model based
on social force model. J. Zhejiang Univ. (Eng. Sci.) 52(02), 352–357 (2018)
15. Zhao, S.: Research on the crowd evacuation simulation based on the social force model.
Dalian University of Technology (2016)
 Genetic Algorithm Selection for Ship Concept
Design

Adam Sobey(&) , Przemyslaw Grudniewski, and Thomas Savasta

Fluid Structure Interactions Group, University of Southampton, Southampton

SO16 7QF, UK
[email protected]

Abstract. Engineering designers are increasingly looking to automate the

design process. By utilizing tools to support decision making engineers can
explore a wider range of options and reduce the time to an initial design, which
can then be iterated over a number of cycles. One area which is prominent in the
literature is the design of boat layouts as this is a complex part of the design
which interacts with a range of different subsystems. The layout is often opti-
mized using Genetic Algorithms and the selection of these algorithms has been
shown to be important. In addition, the representation for this problem is also
vital, with a balance between the continuous nature of the design space and the
total size of the space. This paper therefore explores one representation, the
vector-based approach, to determine how reducing the number of constraints
might improve the performance. The new representation is compared over 3
problems of increasing complexity using a number of current state-of-the-art
Genetic Algorithms. The results show that the reduction in constraints increases
the size of the space more than it reduces the complexity, showing poorer results
than previously. cMLSGA is always the top performing Genetic Algorithm and
it can be concluded that the diversity focused general solvers should be pro-
moted on similar kind of problems.

Keywords: Layout design
Evolutionary computation
Automated design

1 Marine Layout Design

Ships are increasingly complex to design. The number of objectives that designers want
to consider is growing, regulation is becoming more complex, customers are more
demanding due to the current economic climate and accounting for the environment is
an ever increasing concern. For a practical, systems-level, approach to design this
means that the designers job is becoming increasingly difficult due to the number of
variables that need to be considered. This is both the manner in which the variables
from their own subsystem interacts with variables in other subsystems, for example
stiffener height impacting on cabin volume in a yacht, but the increase in complexity in
their own subsystem, for example many vessels are now looking at optimising a hull
over an operating profile rather than a maximum speed. Many other industries with
similar issues are now looking to software solutions to help simplify the design space.
This can be in the use of data to support decision making, generation of databases and

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 156–172, 2021.
https://doi.org/10.1007/978-981-15-4680-8_11
 Genetic Algorithm Selection for Ship Concept Design 157

linked software to allow a freer interaction between engineers in different subsystems

or automated design tools.
Automated design tools provide a way to search large design spaces, visualise
results to reduce complexity and incorporate areas of expertise outside of the designer’s
domain knowledge. These tools are becoming increasingly prevalent, such as the use of
feature selection tools to reduce the number of variables down to only the key set or
optimisation tools to find the best set of designs in a given search space. Genetic
Algorithms are a popular tool for the optimization process as they are flexible to a range
of problems and powerful in finding solutions across large search spaces. However, a
key problem in optimisation is the ‘no free lunch’ theory: in essence, that problems
have different characteristics and that different algorithms are suited to solving these
problems. If the problem is well understood then a specialist solver can be used which
will have excellent performance, but if the problem is not known then a general solver
must be used that inevitably has poorer performance. For many of the very difficult
problems in engineering it is likely that the general solvers will be incapable of solving
the problems proposed or take many more computationally expensive function calls to
do so. It is therefore vital that problems are categorized and the corresponding specialist
algorithms are used or developed.
This paper therefore investigates the performance of a number of ‘state-of-the-art’
Genetic Algorithms on the optimisation of a general arrangement, categorising the
problem and highlighting the desired characteristics required from a Genetic Algo-
rithm. A range of top performing algorithms will be used: U-NSGA-III, MOEA/D-
MSF, HEIA, BCE and cMLSGA; to determine which algorithms perform most
strongly. This will be performed on a simple yacht design problem, allowing a rela-
tively complex general arrangement but reducing computational time. In addition the
importance of constraints, population size and number of objectives will be investi-
gated. Recommendations will be given on how algorithm design needs to be developed
further to solve the General Arrangement problems and how extending the problem to
larger and more complex problems can be achieved.

2 Review of Boat Layout Optimization

The automated optimization of a boat layout requires a method to represent the layout
and then a method to optimize the layout suited to this representation. There are a
number of approaches available in the literature with most of them utilizing Genetic
Algorithms for the optimisation. There are four main methods in the literature for
representing the layout of the vessel, each relying on different numbers of constraints,
variables and objectives: intelligent ship arrangement, packing approach, grid based
and vector representation.
The intelligent ship arrangement is originally developed in 2008 by Daniels and
Parsons [1]. This method utilises the Agent-Genetic Algorithm to optimise the layout
which is a hybrid between the agent-based approach and a Genetic Algorithm. They
compare the method to an agent-based approach, which is uncommon in the literature,
and a pure Genetic Algorithm [2] which is based on the original Genetic Algorithm.
The population sizes used are small, 100, and the mutation rate is high, 30%. It is
158 A. Sobey et al.

expected that the initial diversity of points is too low for the problem type and that a
wider population would be more successful. In addition the algorithm developed
appears to be based on Goldberg [3], developed in 1989, or before, meaning that the
solvers used to resolve the optimisation are dated. The representation relies on a high
number of constraints to reduce the design space, with a number of different repre-
sentations compared, this ranges up to 521 constraints in the database. The represen-
tation relies on the use of a binary representation where the inside of the vessel is split
into rectangles, with each block assigned an available compartment type. This is similar
to scheduling type problems seen in the literature on which a number of benchmarking
exercises of optimisation algorithms have been performed. The outputs from the
optimisation are judged to be valid and to provide high quality designs, even on large
problems.
van Oers et al. [4] use a packing approach which is optimised using NSGA-II
implemented in Matlab 2007b, which is an early version of NSGA-II which has since
be updated in 2011 [5]. The code uses a population size of 500 and a 500 generations.
This is the most commonly used Genetic Algorithm and shows strong diversity of
search providing excellent performance as a general solver. However, the algorithm is
not benchmarked against other candidate solutions. The layout is represented using a
block approach which has a relatively low number of variables for each compartment,
making it scalable for larger applications. The aim is to reduce the quantity of overlap
detection and removal as far as possible to make the algorithm more efficient. This
means a solid definition is used where the boundaries of a room are used alongside
information about whether point lie in or outside an object. The available space is then
searched for after the design is generated. The implementation allows 6.35 * 104
evaluations in 10 min on a standard desktop computer. The designs contain a large
number of voids, but the designs are judged to be feasible for the application.
Nam, Kim and Le [6] use a grid based system to represent the layout which is
extended in Nam and Lee [7], both approaches use a Genetic Algorithm to determine
the layout of the vessel with a four deck optimization of the superyacht shown in
Fig. 1. The Genetic Algorithm used is a Constraint-Based Genetic Algorithm (CBGA).
In the development and documentation of the Genetic Algorithm no other algorithms
are compared against and there is no benchmarking on other problems outside of boat
layouts. This makes it difficult to determine how effective the algorithm is in com-
parison to the state-of-the-art. However, the approach is compared favourably to a
branch and leaf hybridised with simulated annealing. The problem is treated to be
similar to scheduling problems seen in the optimization literature, where simulated
annealing has been shown to perform poorly. The algorithm seeks to determine a layout
with cabins while also looking at the position of the stairs, and therefore goes beyond
just the optimization of the layout. The results of the algorithm are simple layouts, with
no unused space, that provide short distances for escape via the staircases.
Finally, Sobey et al. [8] use a vector based approach to the optimisation based on
Riley [9]. This approach is in 3D and utilises a large number of variables to represent
each cabin, making it less scalable than the grid based method. The problem uses 49
variables to define a problem for a 24 m yacht, with 2 objectives of space utilisation
and 3 constraints. They compare 10 different state-of-the-art Genetic Algorithms where
the results show that cMLSGA [10] performs the best on these problems. It raises the
 Genetic Algorithm Selection for Ship Concept Design 159

Fig. 1. Multi deck optimisation considering area requirements [7]

importance of benchmarking recent solvers and to the importance of modern Genetic

Algorithms. This is because there is a separation between the performances of different
algorithms, with the modern Genetic Algorithms substantially outperforming the older
methods.
The four different main methods are summarized in Table 1 showing a range of
representations that are available for the layout optimization. The literature shows that
the methods rely on a high number of constraints, which results in design spaces that
are traditionally harder for Genetic Algorithms to solve. In addition some of the
methods show an escalation of variables with the number of compartments, also
increasing the difficulty. More recent research shows the importance of algorithm
selection and utilization of the current state-of-the-art from evolutionary computation.

Table 1. Marine Layout Optimisation

Heading level Intelligent Packing Mesh Vector representation
ship approach representation
arrangement
Genetic Agent-GA NSGA- CBDA 10 benchmarked with
algorithm hybrid II cMLSGA best
performing
Variables 100 60 30 49
Objectives 3 3 2 2
Constraints 521 3 6 3
Compartments 205 30 40 7
160 A. Sobey et al.

3 Boat Layout Definition

The manner in which the boat layout is represented is important for the optimisation
and so the representation presented in Sobey et al. [8] is updated to make it more
efficient. Genetic Algorithms are generally designed with a poor ability to cope with
discontinuities or constraints and therefore the new representation is focused on pro-
viding a continuous, unconstrained, layout representation. This representation is then
compared across three different yachts to show how the increase in complexity can be
handled by the Genetic Algorithm. In this case a motor yacht is optimised to maximise
the space utilisation of the cabins in the hull. The motor yacht particulars for the 3
different case studies are described in Table 2.

Table 2. Motor yacht particulars.

24 m 50 m 140 m
Length overall (m) 24.8 50 140
Max beam (m) 6.9 11.12 23.47
Draught (m) 1.955 3.50 5.0
Displacement (Tonnes) 152.8 418.6 2658.3
LCB (from transom) 8.673 17.737 48.975
Decks 1 1 2
Compartments 7 10 57
Variables 49 70 399
Objectives 4 4 8

3.1 Cabin Geometry

The new representation is reduced to a 2D definition, as previous studies have shown
that there are limited benefits to a 3D representation if structures, piping and electrical
subsystems are not considered. In this case each cabin is defined using 7 variables
which are shown in Fig. 2:
1. x position from the forward perpendicular (FP);
2. y position from the centreline (CL);
3. length of the compartment (L);
4. distance from the centreline to the forward most point to starboard (W11);
5. distance from the centreline to the forward most point to port (W11);
6. distance from the centreline to the aft most point to starboard (W21);
7. distance from the centreline to the aft most point to port (W22).
In addition, the length of the cabin can’t be longer than the distance between two
bulkheads. These variables will be optimized to minimize 4 objectives for each deck,
with 4 objectives overall for the small vessels, 24 m and 50 m, and 8 objectives for the
larger 140 m vessel. The 4 objectives seek to reduce the unused area for the whole
deck, minimize the distance between certain cabins, minimize the distance between the
 Genetic Algorithm Selection for Ship Concept Design 161

W11 W21

L
y – from CL

W12 W22
x – from FP

CL

Fig. 2. Cabin geometry definition

Centre of Buoyancy and Centre of Gravity and ensure that the area of each cabin is
similar to those seen for this type of vessel in the past.
The space fitness objective minimises the unused area of the deck compared to the
whole deck area and is defined in Eq. (1):
P
Deck area  Cabins area
Fitnessspace ¼ ð1Þ
Deck area

To optimise the flow of people within the yacht the distance between cabins is
minimized where 2 cabins, i and j, are defined as next to each other if one of the
conditions in Eqs. (2)–(5) are satisfied:

xi þ li ¼ xj ; ð2Þ

xi þ lj ¼ xj ; ð3Þ

yi þ maxðwi11; wi21Þ ¼ yj  maxðwi12; wi22 Þ; ð4Þ

yj þ maxðwj11 ; wj21 Þ ¼ yi  maxðwi21; wi22 Þ: ð5Þ

If none of these conditions is satisfied, the condition showing the minimum value is
returned.
To determine the trim of the vessel the centre of gravity and centre of buoyancy
need to be determined. The longitudinal centre of area is found for each cabin using
standard weights per unit area from data available in the literature. This is then mul-
tiplied by the actual area of the cabin to find its weights. The longitudinal centre of area
is assumed to be positioned in the middle of the cabin length and the vertical centre of
area is assumed to be the average of the 2 faces, shown in Eq. (6):
162 A. Sobey et al.

w11 þ w12 þ w21 þ w22

VCA ¼ y þ : ð6Þ
4
The centre of mass of the cabins is assumed to be at the centre of area. The
moments are taken in the transverse and longitudinal direction for each cabin and added
together and divided by the total weight of cabins to find the position of the centre of
gravity (CG) of the deck. The centre of buoyancy (CB) of the yacht is calculated using
the sectional areas found from the yacht lines plan. The aim of this objective is to
minimise the distance between CB and CG, shown in Eq. (7):

CB  CG
Fitnesstrim ¼ : ð7Þ
CB

Cabin are be required to be of a certain size depending on their type. Standards

areas are determine from data available in the literature and all of the cabins are
required to respect these allocated areas to within a margin of 20%. The aim of this
objective is to minimise the difference between the actual area and the standards area of
each cabin, shown in Eq. (8):

1 X Standard area  actual area

Fitnesssize ¼ : : ð8Þ
nb of cabins Standard area

4 Genetic Algorithm Selection

The existing studies shows a limited range of Genetic Algorithms from the Evolu-
tionary Computation literature are used within layout optimisation, all of these are
applied without benchmarking to determine their performance or the characteristics
required from the optimisation algorithm. However, recent research by Sobey et al. [8]
shows that the layout optimisation of a boat hull is a complex optimisation problem for
the Genetic Algorithms to solve, as some of the current state-of-the-art algorithms
struggle to solve even simple representations of the layout. Therefore, it indicates that,
the selection of the correct Genetic Algorithm is vital to ensure that the best Pareto
Optimal Front is found.
A brief review of the current state-of-the-art in evolutionary computation is per-
formed by splitting the algorithms into 4 broad categories: niching, decomposition, co-
evolutionary and multi-level selection algorithms. Niching is exemplified by the
crowding mechanism based niching technique found in the most popular Genetic
Algorithm NSGA-II [11] which uses non-domination to sort the fittest solutions. It is a
robust general solver with few hyper-parameters and retains a high diversity, making it
the most popular algorithm in real applications. This methodology has been upgraded
to a variety of versions, such as U-NSGA-III [12] that unifies NSGA-II and NSGA-III
to be suitable for mono-objective, multi-objective and many-objective problems, the
problem type on which it is currently the top performing algorithm.
 Genetic Algorithm Selection for Ship Concept Design 163

Decomposition methods are a relatively new family where the population is divided
into sub-groups that search different sub-regions of the search space. These algorithms
implement a number of additional setting parameters, such as the weight vectors in
MOEA/D, which greatly influence the solutions and must be optimised. MOEA/D [13]
is the most popular decomposition method with a number of different variants spe-
cialised for different problem types. These methods require a priori knowledge of the
objective space or they can result in extremely poor performance. However, it is hard to
obtain the required knowledge of the objective space for practical optimisation prob-
lems before solving them. When solving discontinuous problems or constrained
problems these algorithms struggle with the large gaps where there are no feasible
solutions as the weight vectors point straight through the gaps and the individuals
struggle to go around these spaces, resulting in a waste of computational power.
However, these algorithms exhibit excellent convergence characteristics, dominating
the benchmarking for dynamic and unconstrained problems.
The term coevolution is first introduced to describe the coexistence of plants and
butterflies [14]. In the co-evolutionary approaches, multiple populations of species of
individuals coexist and evolve in parallel, usually utilising distinct reproduction
mechanisms. There are two currently top performing methods that utilise this approach,
Bi-Criterion Evolution algorithm (BCE) and Hybrid Evolutionary Immune Algorithm
(HEIA). In Bi-Criterion Evolution algorithm (BCE) [15] sub-populations operate on
the same search spaces and individuals for each group are selected at each generation
based on two distinct fitness indicators: the Pareto-based criterion (PC) and the Non-
Pareto-based (NPC). In the Pareto-based criterion, standard Pareto dominance is uti-
lised which rewards convergence whereas in the Non-Pareto based selection an addi-
tional indicator is introduced, based on Hypervolumes (HV) which rewards diversity of
solutions. This leads to an overall improvement in diversity for the entire population,
especially on many-objective cases and problems with irregular search spaces and
variable linkages. However, it is still convergence dominated. A similar approach has
been utilised in Hybrid Evolutionary Immune Algorithm (HEIA) [16], but in this case
two distinct evolutionary computation methods are used, Immune Algorithm and
Genetic Algorithm, instead of separate quality indicators. This method shows excellent
performance on quite a wide range of problems, but is more convergence orientated
and the performance of this method has not been evaluated on highly discontinuous
problems and constrained problems where the performance is expected to be low, as it
only utilises crowding distance for diversity.
Multi-Level Selection Genetic Algorithm was developed to take advantage of the
recent evolutionary theories of Wilson and Sober [17]. Wilson and Sober propose that
evolutionary fitness is not just dependent on the fitness of the individual but can also be
dependent on the collective of individuals that it is associated with, an example might
be that the survival of a wolf is not just dependent on its own fitness but the fitness of
its pack. Multi-level selection Genetic Algorithm (MLSGA) was first introduced by
Sobey and Grudniewski [18] and [19], in this algorithm a collective level reproduction
mechanism is introduced, in addition to the individual level used in standard Genetic
Algorithm, and the fitness function is split between these levels, with this algorithm
being unique in this respect. The algorithm works by randomly generating an initial
population which is classified into collectives according to the design variables. On the
164 A. Sobey et al.

individual level, each individual is evaluated through the individual objective function
and genetic operators are utilised to perform individual reproduction, and these can be
selected from any of the current state-of-the-art methods. Simultaneously, each col-
lective is evaluated using the collective objective function. There is a competition
among the collectives and the worst collective(s) is eliminated. This collective is
replaced by generating a copy of the best individuals from each of the remaining
collectives. The process is stopped when the termination condition is satisfied. There
are two main fitness evaluation methods, MLS1 and MLS2 [18]. MLS1 uses the
aggregate of the individuals in the population to calculate the fitness of a collective.
MLS2 calculates different objectives using a fitness defined for the collective, with
MLS2R defined as being reverse. Therefore, MLS1 focuses on solutions at the middle
of the Pareto optimal front and MLS2 and MLS2R enhance the search ability at the two
sides of the real Pareto optimal front separately. Based on the two main methods, MLS-
U, combining MLS1, MLS2 and MLS2R, can be utilised in MLSGA to maintain the
diversity of the search shown in Fig. 4. This method has recently been combined with
the co-evolutionary approach, co-evolutionary Multi-Level Selection Genetic Algo-
rithm (cMLSGA), to increase its generality. So far MLSGA, and its variants, have
shown to have top performance across a range of state-of-the-art multi-objective
problems. The algorithm seems to thrive in environments with discontinuous fronts and
constrained problems where diversity of the mechanisms is important. Multi-level
selection theory is unique in that it is the only diversity first search Genetic Algorithm.
In this study a total of 5 genetic algorithms are used to solve the presented cases:
MOEA/D-MSF [20] as an improved variant of MOEA/D for imbalanced and uncon-
strained problems with a similar performance to MOEA/D-PSF on this problem type;
HEIA [16] as an algorithm that shows high proficiency across a diverse set of problems
and is a general solver with a bias towards convergence; BCE [15] which is another
more recent algorithm designed as a general solver but with a stronger bias towards
convergence than HEIA; U-NSGA-III [12] as the many-objective universal variant of
NSGA-II [11], which is the current state-of-the-art seen in the marine literature;
cMLSGA as the best general solver [10] which also provides a diversity first search and
an algorithm developed to be similar to the original Genetic Algorithm, representing a
solver which is still common in the marine literature.
The tests are performed over 15 separate runs, and the termination criterion is set at
300,000 function evaluations for each run on the 24 m and 50 m case studies and 50,000
for the 140 m simulations as some solvers perform the analysis slowly with the high
number of objectives. The results are compared using the Hyper Volume (HV) and
Inverted Generational Distance (IGD) indicators, as between them they provide com-
prehensive information on the convergence, accuracy, and diversity of the obtained
solutions. HV is the measure of volume of the objective space between a predefined
reference point and the obtained solutions which has a stronger focus on the diversity
and edge points and can be calculated according to [21], where higher values indicates
the better results. IGD is the measurement of the average Euclidean distance between the
points in a true Pareto Optimal Fronts and the closest solution in the obtained set of
solutions, and is to be minimised, where 0 indicates the perfect convergence. This metric
has stronger emphasis on the convergence and uniformity of the points and for practical
problems can be calculated according to [22]. Different population sizes have been
 Genetic Algorithm Selection for Ship Concept Design 165

evaluated and 1800 is selected as the best value for cMLSGA and 1000 for other
algorithms. The crossover and mutation rates are set as 1 and 0.08 respectively, and the
rest of algorithm-specific operation parameters are set as in the original publications. For
all cases the objective normalization strategy taken from [13] is used.

5 Automated Design of Yacht Layouts

Three case studies are used to test the performance of the algorithms and investigate the
potential complexity that Genetic Algorithms are capable of solving. A 24 m, 50 m and
140 m yacht are compared, each with increasing complexity, and the top Genetic
Algorithms are benchmarked and the resulting layouts compared.

5.1 Superyacht 1 - 24 m
The 24 m superyacht forms the simplest problem with 4 objectives, 7 compartments
and 49 variables. A leaderboard of the algorithms is shown in Table 3. In this case the
cMLSGA is the highest performing of the algorithms showing the best performance on
the IGD and HV metrics; the lowest performer is MOEA/D-MSF using both metrics for
comparison. In this case the strongest convergence improving algorithm of the co-
evolutionary approaches, BCE, shows a stronger performance, with HEIA performing
only 4th for the IGD metric and 3rd for the HV metric. The strongest many-objective
algorithm from the Evolutionary Computation literature, U-NSGA-III, performs quite
poorly, despite the problem having 4 objectives. MOEA/D-MSF, which has the
strongest convergence of the different algorithms, performs the worst with a similar
IGD to HEIA but with a much lower HV score, reflecting its poor diversity retention.

Table 3. Performance metrics for the 24 m superyacht, values indicate the mean and standard
deviation for each metric.
Ranking IGD HV
1 cMLSGA cMLSGA
0.2055 1.3157
0.0303 0.1621
2 BCE BCE
0.3193 1.0797
0.0822 0.2418
3 U-NSGA-III HEIA
0.4012 0.8344
0.1742 0.3194
4 HEIA U-NSGA-III
0.4291 0.8050
0.1507 0.3258
5 MOEA/D-MSF MOEA/D-MSF
0.4749 0.6247
0.1465 0.1770
166 A. Sobey et al.

Fig. 3. Automated layout design for the 24 m yacht, black lines demonstrate the position of the
bulkheads.

A figure representing the layout for the best point on the Pareto Front is shown
below, Fig. 3. The vessel shows a rough layout with the positions of most of the rooms
providing a design that is reasonable for this type of yacht. The machinery is at the
back and there is a large master room. However, the crew is sandwiched between the
masters room and the main rooms, which is unrealistic as the crew is likely to be away
from the customers. In addition the rooms themselves are not making best utilization of
the space nor are they sensible shapes, especially in the case of the double room 1. This
provides a worse layout than the previous research in Sobey et al. [8] for the same
number of function calls and the same boat size.

5.2 Superyacht 2 - 50 m
The 50 m superyacht provides a more complex design space with 4 objectives, 10
compartments and 70 variables. A leaderboard of the algorithms is shown in Table 4.
In this case the cMLSGA is again the highest performing of the algorithms showing the
best performance on the IGD and HV metrics, while MOEA/D-MSF provides the worst
performance of the algorithms. The separation between cMLSGA and the other
algorithms is greater on this problem, with a larger difference to BCE on the IGD
metric and to HEIA on the HV metric. Since HEIA has better diversity retention than
BCE it illustrates that this is becoming a factor in the performance of the Genetic
Algorithm, with the larger search space requiring the successful algorithms to search
more widely to find the optimal solutions. In this case U-NSGA-III is performing
almost as badly as MOEA/D-MSF, despite there being 4 objectives, with a wide gap to
the co-evolutionary and multi-level approaches. MOEA/D-MSF, which has the
strongest convergence of the different algorithms, performs the worst again and in this
case the HV metric shows a large standard deviation of 30.299, reflecting the fact that
on some runs the performance was extremely poor.
 Genetic Algorithm Selection for Ship Concept Design 167

Table 4. Performance metrics for the 50 m superyacht

Ranking IGD HV
1 cMLSGA cMLSGA
0.2674 1.5176
0.0626 0.1669
2 BCE HEIA
0.5205 0.9392
0.0473 0.1830
3 HEIA BCE
0.5309 0.8929
0.0627 0.2385
4 U-NSGA-III U-NSGA-III
0.6102 0.7197
0.0708 0.2480
5 MOEA/D-MSF MOEA/D-MSF
0.6202 0.7112
0.1062 30.299

Fig. 4. Automated layout design for the 50 m yacht, black lines demonstrate the position of the
bulkheads.

In this case the optimisation provides a reasonable design with most of the area
covered and less available space than the previous design, shown in Fig. 4. However,
the galley and dining room are small and the double room 2 is a point, hidden next to
the double room 3 at the front of the vessel. In addition the size of the gym is large,
taking up more space than is perhaps realistic. The galley, double room 4 and double
room 1 also have unrealistic diagonal walls to the cabins. The overall design is further
from a reasonable starting point for a concept design than the 24 m yacht due to the
additional complexity provided by the additional rooms. Further function calls would
be required to solve this problem satisfactorily.
168 A. Sobey et al.

5.3 Megayacht - 140 m

The final study is for a Megayacht design of 140 m, which is large for the leisure boat
industry. However, this design also starts to represent the size and complexity expected
from many commercial vessels. The optimization requires 8 objectives and 399 vari-
ables to optimize the 57 different compartments. In this case many of the solvers had
started to run inefficiently, with noticeably faster performance from cMLSGA than the
others. Due to this inefficiency only 50,000 function calls were made to allow the
simulations to be made in a feasible time frame. The results for the different algorithms
are shown in Table 5. These results show that cMLSGA provides the best convergence
and uniformity with leading performance on the IGD metric. However, in this case it is
only the second best performer on the HV metric. BCE is a stronger convergence based
algorithm than cMLSGA and it appears that with less function call that the separation
between these two algorithms is lower. In addition U-NSGA-III performs more poorly,
with the worst results for IGD and second to worst on HV. This means that MOEA/D-
MSF is no longer the worst performer and this might be that the convergence first nature
of this algorithm is preferred in scenarios where there are less function calls available.

Table 5. Performance metrics for the 140 m megayacht

Ranking IGD HV
1 cMLSGA BCE
0.7424 1.3228
0.3180 0.0773
2 HEIA cMLSGA
0.8531 1.2618
0.2174 0.2328
3 BCE MOEA/D-MSF
1.4735 1.2225
0.1586 0.2918
4 MOEA/D-MSF U-NSGA-III
1.5701 0.9545
0.0682 0.2232
5 U-NSGA-III HEIA
1.6429 0.8629
0.0892 0.3986

In this case the design still provides a relatively good packing of the boat layout,
with a small quantity of unutilized space. However, it results in an extreme situation
with some of the rooms filling out the area between the bulkheads and providing large
cabins, Fig. 5. These cabins are often much larger than is realistic for their task, while
other rooms are points with virtually no volume. This design is some distance from a
feasible concept design, which is likely to be from the reduced function calls and
increased complexity of the design. For the bottom deck, Fig. 6 the results are very
poor demonstrating that a number of additional function calls are required.
 Genetic Algorithm Selection for Ship Concept Design 169

Fig. 5. Automated layout design for the 140 m yacht top deck, black lines demonstrate the
position of the bulkheads.

Fig. 6. Automated layout design for the 140 m yacht bottom deck, black lines demonstrate the
position of the bulkheads.

6 Discussion and Limitations

The results show that the selection of the Genetic Algorithm is important to resolving the
layout for the optimization problem. The state-of-the-art algorithms are already strug-
gling to resolve the Pareto Front by the time the problem reaches 140 m, this is both to do
with the representation and the difficult in solving problems of this size. cMLSGA
provides the best results for all of the problems with the next best performing algorithm,
BCE, also using the co-evolutionary approach which provides a good general perfor-
mance. The worst performing algorithm is MOEA/D-MSF which is a convergence based
algorithm and a specialist solver, which is defined as a solver with a strong performance
on some problems and weak on others. This algorithm performed better on the final,
170 A. Sobey et al.

140 m, design where the function calls are reduced, potentially because of its strong
convergence first mechanisms that quickly drive it to the front. Interestingly the U-
NSGA-III approach performs poorly on the 140 m vessel as this is the highest performing
many-objective solver showing high performance on problems with 4+ objectives in the
evolutionary computation literature. It also performs poorly on the easier problems, 24 m
and 50 m with 4th performance on both, indicating that modern attempts at layout opti-
mization should move to more advanced algorithms than NSGA-II.
The problem representation is adapted to reduce the number of constraints from the
previous research [8]. Most Genetic Algorithms perform poorly with a large number of
constraints and discontinuous spaces and so the reduction in constraints should allow for
a more continuous landscape. However, due to the nature of the layout problem, it
appears that the Pareto front itself is discontinuous, despite the lack of additional con-
straints. The poor performance compared to the previous representation of the problems
shows that in some cases, a limited set of constraints may help to reduce the search
space, which is highly beneficial. Fully merging the constraints into objective functions,
as penalties, may lead to significant increase in complexity of the problem and outweigh
the potential advantages of unconstrained optimisation. Therefore the balance between
constraints is not trivial and more iterations are required to reach an optimal represen-
tation for this type of problem. In addition the designs for the 3 layouts show that the
trimming and space utilization constraints are easy to solve. It is a more complex
problem to ensure that the rooms are in sensible places and of a reasonable shape and
size. This needs further exploration with some adaptions to the representation to provide
these benefits either through constraints or through the cabin definition.

7 Conclusion

The automated design of vessels is of increasing interest to ship designers. However,

the complexity of these problems is high and the representation of the problem is key to
providing a good solution. The use of Genetic Algorithms to solve these problems is
common but the algorithms that are used are often not compared to each other or
outdated methodologies are utilised. In the results shown here, the best performing
Genetic Algorithm is cMLSGA, which is a general solver promoting diversity over
convergence. This means that, the presented layout problem, predominantly requires
high diversity of the search due to its constrained and discontinuous nature. The
representation used for the layout attempts to provide fewer constraints and converts
these to objectives as Genetic Algorithms are generally better at solving problems with
continuous surfaces. However, combining the constraints with objective functions
leads to the increase of the complexity of the problem, which grows to such an extent
that the results for even the 24 m problem are poor, and lower in comparison to the 3
constraints representation that has been used previously. This indicates that a small
number of constraints may be preferred as they reduces the search space to one that is
feasible for the algorithms to investigate and this is necessary for good performance.
The recommendation is that cMLSGA should be used for boat layout optimization and
that other general Genetic Algorithms with diversity first searches should be promoted
on similar problems.
 Genetic Algorithm Selection for Ship Concept Design 171

References
1. Zhang, Q., Li, H.: MOEA/D: a multiobjective evolutionary algorithm based on decompo-
sition. IEEE Trans. Evol. Comput. 11, 712–731 (2007)
2. Jiang, S., Yang, S., Wang, Y., Liu, X.: Scalarizing functions in decomposition-based
multiobjective evolutionary algorithms. IEEE Trans. Evol. Comput. 22, 296–313 (2018)
3. Lin, Q., Chen, J., Zhan, Z.-H., Chen, W., Coello Coello, C., Yin, Y., Lin, C.-M., Zhang, J.:
A hybrid evolutionary immune algorithm for multiobjective optimization problems. IEEE
Trans. Evol. Comput. 20, 711–729 (2016)
4. Li, M., Yang, S., Member, S., Liu, X.: Pareto or non-pareto: bi-criterion evolution in
multiobjective optimization. IEEE Trans. Evol. Comput. 20, 645–665 (2016)
5. Seada, H., Deb, K.: U-NSGA-III: a unified evolutionary optimization procedure for single,
multiple, and many objectives. In: International Conference Evolution Multi-Criterion
Optimisation, Guimarães (2015)
6. Deb, K., Pratap, A., Agarwal, S., Meyarivan, T.: A fast and elitist multiobjective genetic
algorithm: NSGA-II. IEEE Trans. Evol. Comput. 6, 182–197 (2002)
7. Grudniewski, P.A., Sobey, A.J.: cMLSGA: co-evolutionary multi-level selection genetic
algorithm. IEEE Evol. Comput., Under Review
8. While, L., Bradstreet, L., Barone, L.: A fast way of calculating exact hypervolumes. IEEE
Trans. Evol. Comput. 16, 86–95 (2012)
9. Wang, Z., Bai, J., Sobey, A.J., Xiong, J., Shenoi, R.A.: Optimal design of triaxial weave
fabric composites under tension. Compos. Struct. 201, 616–624 (2018)
10. Nam, J., Le, T.: Automatic interior space arrangement of mid-sized superyachts using a
constraint-based genetic algorithm. J. Mar. Sci. Technol. 17, 481–492 (2012)
11. Nam, J., Kim, D., Lee, H.: An approach to computerized preliminary design procedure of
mid-size superyachts from hull modeling to interior space arrangement. Int. J. Naval Arch.
Ocean Eng. 2, 96–103 (2010)
12. van Oers, B., Stapersma, D., Hopman, H.: An optimisation-based space allocation routine
for the generation of feasible ship designs. Ship Technol. Res. 56, 31–48 (2009)
13. Deb, K.: Multi-objective NSGA-II code in C Revision 1.1.6. Softw Dev by KanGAL (2011).
https://www.egr.msu.edu/*kdeb/codes.shtml. Accessed 9 Feb 2019
14. Daniels, A., Parsons, M.G.: An agent based approach to space allocation in general. In: 9th
International Marine Design Conference, Ann Arbor, MI (2006)
15. Goldberg, D.E.: Genetic Algorithms in Search, Optimisation and Machine Learning.
Addison-Wesley, Reading (1989)
16. Sobey, A., Blanchard, J., Grudniewski, P., Savasta, T.: There’s no free lunch: a study of
genetic algorithm use in maritime applications. In: Conference on Computer Applications
and Information Technology in the Maritime Industries, Tullamore, Ireland (2019)
17. Riley, C.: Concurrent engineering for motor yacht design. University of Southampton,
Southampton (2015)
18. Daniels, A., Parsons, M.: Development of a hybrid agent-genetic algorithm approach to
general arrangements. Ship Technol. Res. 55, 78–86 (2008)
19. Ehrlich, P.R., Raven, P.H.: Butterflies and plants: a study in coevolution. Evolution 18, 586–
608 (1964)
20. Wilson, D.S., Sober, E.: Reintroducing group selection to the human behavioral sciences.
Behav. Brain Sci. 136, 337–356 (1989)
172 A. Sobey et al.

21. Sobey, A.J., Grudniewski, P.A.: Re-inspiring the genetic algorithm with multi-level selection
theory: multi-level selection genetic algorithm. Bioinspir. Biomim. 13, 852–862 (2018)
22. Grudniewski, P.A., Sobey, A.J.: Behaviour of multi-level selection genetic algorithm
(MLSGA) using different individual level selection mechanisms. Swarm Evol. Comput. 44,
852–862 (2018)
 Multi Objective Design of Ships; A Pareto
Procedure

Sander Calisal1,2(&)
1
Piri Reis Üniversitesi, İstanbul, Turkey
[email protected]
2
Mechanical Engineering Department, University of British Columbia,
Vancouver, Canada

Abstract. The design calculations for a displacement type vessel and for a
planing hull are done for the main dimensions, starting with a fuel and accel-
eration criteria. A large number of vessels are simulated with the help of
available algorithms in the open literature such as the UBC series for the dis-
placement types. Ship design is continuously changing and new design tech-
nologies developed during the last two decades. The method used in this study is
to calculate ship resistance and ship motion based on suitable algorithms and on
the calculation of dynamic effects such as acceleration, slamming, and added
resistance and observe the influence of the principal dimensions on these
quantities in a simulation procedure. The design requirements are then used to
select a suitable candidate ship. Pareto fronts are obtained for two parameters
such as the effective horse power and the acceleration levels. Pareto procedure
for design is illustrated for a two parameter design procedure. Possible com-
bination of ship principal dimensions is obtained for the relative importance of
the design parameter, as seen by the owner or the designer. Application is then
expanded to planing hulls. Many solutions are first obtained in a range of
desirable speeds, beam, and dead rise angle. The solutions in the range of
planing hull speed ratio are considered. The satisfactory solutions with com-
bined criteria in the Pareto front are tested and studied in some more details.
A ‘better’ solution is recommended to the designer. Additional criteria are
considered in the selection such as the critical trim angle as constraints etc.

Keywords: Ship design
Planing hull design
Multi objective
Pareto front

Pareto method

1 Introduction

This study is on the design of displacement hulls and planing hulls. After a discussion
on the design of a displacement hull, the procedure developed for displacement hulls is
modified for a planing hull design, by selecting an average beam value, a dead rise
angle and a trim angle for a hull with a given amount of mass. The procedure for
planing hull later checks the dynamic trim angle with respect to the critical trim angle,
where the porpoising is likely to start. In contrast to the old ship design procedures
based on the duplication of the features of surviving ships, we used algorithms based
on experimental and statistical data. The concept of a design cycle used for ship design
© Springer Nature Singapore Pte Ltd. 2021
T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 173–189, 2021.
https://doi.org/10.1007/978-981-15-4680-8_12
174 S. Calisal

was based on a perturbation procedure and this design procedure is described in

modern text books such as PNA. The design cycle in fact suggests that the ship design
is a nonlinear process and, in general, it may converge to a solution satisfying the
‘owner requirements’. An old- fashion ship design procedure could be described as
making sure that the ship will float and will have sufficient stability and speed. In that
procedure an equilibrium condition: that is ‘the sum of weight groups as a function of
length, is set equal to the buoyancy of the ship’, is used to obtain the ship length.
Following these calculations, a ship stability criteria for GM is used to obtain the beam
of the ship for a known weight group distribution. The underlying assumption in this
process is that, the new ship will have similar weight groups and similar weight
distributions to those of existing ships. No unknown weight groups and/or their
locations are foreseen at this stage. In fact, that is the way, the author learned, long time
ago, the ship design procedure. After this stage, various available computer programs,
were used to calculate the ‘curves of form’ and floodable length calculations etc. While
one can still use such calculations in a computer, more effective procedures are possible
now, for the preliminary design and possibly new design methods will be forthcoming,
benefiting from the powerful graphical features of new IT systems and easy access to
large data bases. A good example of how the new developments in computers provide
can be seen in the use of expert systems. Akintürk [1] used an expert system to design a
fishing boat with low levels of accelerations for crew comfort and safety. He developed
a ship design methodology suitable for a computer environment, including crew
comfort and safety in the preliminary design. He implemented the procedure in an
expert system called Echidna and applied it for the design of a SWATH and fishing
vessels. The procedure used was based on heuristic rules to complete the design and to
find a solution for a given sea state and satisfying the owner requirements. He was able
to quantify the additional cost of crew comfort and safety. More recently with the
availability of faster computers, powerful optimization procedures were developed. Wu
Chauncey [2] used a Taguchi and Pareto procedure and developed an ‘optimal method’
for the preliminary design of a truss. He used a low order polynomial curve fit to
approximate Pareto optimal designs. He then predicted the frequency of vibration and
mass of the truss. Hong Yoon Kim and Nickolas Vlahopoulos [3] used a multi- level
optimization algorithm to construct a solution satisfying all constraints. With the
application of this algorithm to a case where a non-convex Pareto front existed, a ship
design optimization analysis was demonstrated. These three design methods referred to
above, deviate from the old design procedures and use extensively the computational
and decision-making potential of the algorithms for computer applications. A relatively
recent application of multi-level optimization is by [4] Diez M. et al. that relies on the
calculations of ship resistance and ship motion by well-established computer programs
and usage of a Pareto algorithm, to locate an optimal solution. This present study is
essentially inspired by that study, however, instead of dedicated software, the present
procedure uses algorithms for resistance and motion calculations. This study also
shows that some derived dynamic characteristics of the ship can be foreseen well in
advance and that more objectives could be added in the design process such as
slamming per hour, or ship displacement in the optimization. One can also see that by
reducing the acceleration levels, the added wave resistance in waves may also be
reduced.
 Multi Objective Design of Ships; A Pareto Procedure 175

2 Simulations

Simulations are necessary for a Pareto study and in this study, this is done by using
published algorithms available for resistance and sea keeping. For the displacement
type hull the resistance and motion spectra in head seas of UBC series [5, 6] and the
procedure suggested by Jensen et al. [7] were used as the measure of the fuel con-
sumption and the acceleration levels. A ship vector was defined as: (Ship length, Ship
Beam, Ship Draft, Ship Block coefficient). The length to beam ratio is randomly
selected in the interval 2.6 to 3.89 for the relative motion studies. For planing hulls the
procedure outlined by Faltinsen [8] and the algorithms available for accelerations are
used. The first step of the study was done to establish the existence of a Pareto front.
For this, a relatively large set of hulls with randomly selected dimensions are defined
within the domain of definition of the algorithms.

2.1 Simulation of Still Water Resistance

For displacement hull design, the wave resistance is estimated first and a viscous
resistance is estimated using the ITTC 57 algorithm. No other common resistance
components are added to the resistance, as at this stage of the design, EHP (Effective
Horse Power) is seen as a proper design parameter. Similarly, the length to beam and
the beam to draft ratios are selected in their proper intervals for the algorithms. The ship
wave resistance simulation can be done effectively, using a ship design vector as (1,
length to beam ratio, length to beam ratio squared, beam to draft ratio and beam to draft
ratio squared). The total resistance calculated was then converted into an easily visu-
alized effective horse power value. EHP is considered as one of the performance
parameter in the definition and search of a Pareto front.

2.2 Motion and Its Spectral Estimation

For the displacement type hull design, the data and the algorithm for heave and pitch
estimation for UBC series were obtained from Calisal et al. [6] for head seas. For the
calculation of statistical dynamic motion values, a sea state is assumed. Important
dynamic values are then calculated only for head seas and for the Froude number
assigned earlier. The assumptions in Jensen et al [7] are used to calculate the combined
motion at the bow of the ship and the relative motion with respect to wave elevation. The
main assumption in Jensen et al [7] is that the interactions between different modes of
motions can be neglected. Additional simplifications such as a barge-like geometry and
the estimation of the damping coefficient by a relatively simple equation of Yamamoto
et al. [9] are used as given in Jensen et al The ship added mass is set equal to the ship
displacement as an assumption and for locating the ‘better hull’ in the design space.
The data, provided by Goudey C.A and Heliotis A.D (1985) [10] for a specific
fishing boat, are used to validate the design code described above. The values of the
pitch response as given by the UBC algorithm was between the experimental and
numerical values provided in this report.
176 S. Calisal

Fig. 1. Bow-acceleration validation with independent tests Goudey C.A and Heliotis A.D
(1985).

The experimental bow acceleration and the numerical values given in this report
and the added resistance values were higher but they were still comparable, as given in
Fig. 1.
Jensen’s assumptions provide the estimation of relative motions in a closed form
solution. This makes the calculation or the estimation of the design related to the ship
dynamic parameters rather simple and fast. However in this study, the response
functions available from the experimental study of the UBC series were used in the
relative motion simulations for the bow motion and the estimation of green water on
deck and slamming per hour. The ITTC wave spectrum and the statistical- calculation
procedures, as given in Bhattacharyya [11] and Ochi [12] are used for the calculations
of significant values for heave, pitch and combined motions and derived values such as
the number of slamming per hour, and green water on deck etc.
For the planing hull studies Faltinsen’s [8] procedure is used for the calculation of
the resistance for randomly selected speeds, beams, and dead rise values. An estimate
of resistance and a dynamic trim value are calculated. The acceleration level at the
center of mass or at the bow are simulated using published algorithms by Savitsky
[13, 14], Taunton [15, 16] and Ermina Bagovic [17], F. De Luca, C. Pensa [18].

2.3 The Pareto Procedure Used

The background idea is to simulate the performance of a hull by various available
algorithms and/or software, and then find the principle parameters of the hull that
provides the best possible multi-objective performance. The design procedure is seen as
 Multi Objective Design of Ships; A Pareto Procedure 177

a sequence of simulations and specifically: Performance = Function of (Ship principal

parameters) or in the case of displacement hulls: Objectives = Function of (L, B, T,
Cb) at a constant Froude number and at an assigned sea state; For planning hulls
Objectives = Function (Mass, L, B, U, b).
Objectives are seen as a vector and, in this study, the elements of this vector con-
sidered were acceleration levels at a specific location on the hull and the fuel consumption
in calm water (EHP), and they are estimated by the specific simulation methods outlined
above. The principal parameters corresponding to the ‘best design’ are then retrieved by a
Pareto procedure. In general, certain weighting values could be assigned to the design
objectives, to provide different degrees of preference to the objectives.
1- The Figs. 2 and 3 definitely show the contradiction between the choice of high
values for length and beam and the high-level energy (EHP) and the relationship
between acceleration levels and the EHP required for a displacement-type hull.
While the procedure was tested for a two objectives search, we extended it for a
multi-objective criteria at a later stage of the study. The Pareto procedure used in this
study is a combination of the Pareto-design procedures as described by Grieson [19]
and Kasprazak and by Lewis [20]. The method selected and used can be summa-
rized as follows: A suitable simulation is used for the calculations of the values for
the design objectives. The design-acceleration location is arbitrarily chosen as at the
bow of the ship. No duration of exposure to high acceleration levels are considered.
The acceleration levels such as that of IMO or US Navy requirements may be used
or included to limit the design space for a particular design.

Fig. 2. Variation of EHP with length and beam

178 S. Calisal

Fig. 3. A typical variation of bow acceleration versus EHP

Fig. 4. Plot of nondimensioanal EHP and nondimensioanal acceleration level.

2- The next step is the search for the existence of a specific convex Pareto front as in
Fig. 4. This is done graphically and by visual observation. The EHP and significant
acceleration levels obtained by simulations are plotted and the existence of a Pareto
front confirmed. It was observed that such a front indeed exists for the UBC series
and most of the planing type hulls tested.
As the measurement units of the design parameter such as the acceleration and the
EHP are not the same, the design objective values are nondimensionalized by
dividing the objective values, by the maximum objective value calculated, in the
design intervals by the simulation procedure. This then gives a distribution on a
unit design space, on each axis.
 Multi Objective Design of Ships; A Pareto Procedure 179

3- A utopic point is then selected to calculate the best suitable design in the design
space. Zero acceleration and zero EHP, that is the origin of the frame is used as the
utopic point and the design at the shortest distance to the utopic point is taken as
the location of the ‘better hull’.
4- Finally the ship vectors for the discrete candidate ships are obtained by finding the
candidate ship at the smallest distance from the utopic point. The above procedure
suggest that the importance of the 2 design parameters studied are of equal impor-
tance. The procedure suggested here is that if one parameter is more important than
the other one by a weight factor wi, the coordinate of the candidate designs could be
increased in that direction by that factor reflecting the increased importance of that
objective. This procedure is a recommendation of Kasprzak and Lewis [20]. They
argue successfully that ‘The Pareto points are independent of the relative distance of
the design objectives’. They call this method as ‘the scaling method’ (Fig. 5).

Fig. 5. A simplified linear Pareto front AB and the location of the ‘best point’ T, utopic point O.

One can visualize the scaling operation by assuming that the Pareto line is locally
representable by a straight line AB, and the point T is the tangency point of the line AB
and the circle with a centre on O; T representing the ‘better’ solution. One can show
that as OT is orthogonal to AB and OT = OA cos b and the coordinates of the best
design T is (OT sin b, OT cos b).
As OT = OA sin a; The coordinate of T can be expressed in terms of OA as:

ðOA 0:5 sinð2aÞ; OA sin2 ðaÞÞ:

So by giving more weight to say objective 1, or by reducing the angle b and

increasing the angle a the objective 1 may be increased or reduced and the objective 2
will increase for 0\a\p=2.
180 S. Calisal

3 Application of the Procedure

3.1 Displacement Hulls
The first attempt of this study was to see if a “Pareto front” exists for some ship design
objectives. Figure 4 suggests that for a given EHP or fuel consumption level, a min-
imum bow acceleration exists for this type hull.
The set of minimum acceleration levels at a constant EHP can be seen as a point on
the Pareto Front. For a constant EHP value, the designer can then select the hull with
minimum acceleration level. This observation is taken as the existence of a Pareto front.
The calculations for the motion, the velocity, the acceleration at the bow in head
seas at constant Froude number are done using the UBC motion algorithm. The
acceleration level is seen as a measure of crew safety and comfort of the passenger and
is taken as a separate design parameter. The Utopic point in this study is taken as the
origin of the design space. The Fig. 4 shows the performance space and the “better”
design shown with a green star. The assumption used in the selection of the ‘better
point’ was that the importance of the two design parameters had equal weight or
importance. The set of hulls with a minimum acceleration level and EHP values are
identified next. For this, the abscissa and ordinate axes are divided into n segments and
the hulls with minimum values in each segment are identified. A 4th order polynomial
is then used to represent continuously the Pareto front. The search and identification of
best hull are done using this continuous curve.
The identification of ship vector is then done by searching and finding hulls in close
proximity to the ‘better’ ship. Hulls in the ninety five percent radius, in the vicinity of
the better point, are then identified as suitable hulls. Best hulls do not have the

Fig. 6. Principle dimensions and the mean values of ‘Best’ designs Best design is for Ship
number 1.
 Multi Objective Design of Ships; A Pareto Procedure 181

maximum length allocated in the design. Best length for equal weight for the two
design performances is about 25 m and the best beam is about 8 m and the draft about
3 m. These numbers could provide a converging design process or could narrow down
of the search space for the designer. Figure 6 below shows the main parameters of the
suitable hulls. The first hull is called the ‘Best’ hull.
The results for the derived performance values such as the acceleration levels and
the number of slams per hour can be seen in Figs. 7 and 8 below. The application of
weights assigned to objectives is also searched to see the suitability of this method. The
design parameters correspond to different weights assigned to the objective functions.
W1 refers to the weight assigned to nondimensional EHP, w2 refers to the weight of
nondimensional Acceleration. In Table 1 XO YO are the solution points in the
nondimensional design space. The application of weights assigned to objectives is also
searched to see the suitability of this method. The design parameters correspond to
different weights assigned to the objective functions.

Fig. 7. Significant bow acceleration versus length and beam of ships in design population

Column 2 in Table 1 shows the results for equal weights assigned to the two
nondimensional objectives. The third column represents the results when more weight
is assigned to the EHP and relatively less weight to the acceleration. We can see that we
have larger acceleration levels compared to equal weight solution. In the fourth column
we give more weight to the acceleration. The acceleration level is reduced at the
expense of the increased EHP, and the hull has higher displacement or cost. More
sophisticated design criteria may be possible including added wave resistance, number
of slamming, number of water on deck or combinations of these values that the
designer intends to study. Application of the above procedure with three objectives is
182 S. Calisal

also done. The three objectives version tried were the ship EHP, Bow acceleration and
Displacement of the ship. The ship displacement was considered as a measure of the
construction cost of the ship. The values simulated for the three objectives were non-
dimensionalized by their own maximum values. The best design is searched in the non-
dimensional space and using the origin of this space as the utopic point. The following
values listed in Table 2 were obtained for the ‘better’ hull: This result is not that
different from the results obtained for the two-objective designs; however, the hull
displacement is much smaller. It is almost 60% of the displacement suggested by the
two objective study.

Fig. 8. Number of slams per hour verses the beam and ship length

Table 1. Influence of weight on the ‘Best’ design values non-dimensional at sea state 3 and
Fr = 0.265
Design objectives and parameters W1 = 1, w2 = 1 W1 = 2, = 1 W1 = 1, w2 = 2
X0 (in Unit space) 0.3236 0.354 0.3031
Yo (in Unit space) 0.3084 0.363 0.2992
EHP 92.2 94.16 107
Acceleration (m/s2) 1.97 2.07 1.89
Length (m) 27 24.4 23.8
Beam (m) 7.11 9.22 9.04
Draft (m) 3.35 3.41 3.8
Displacement (m3) 397 419 438
 Multi Objective Design of Ships; A Pareto Procedure 183

Table 2. Best ship for 3 objectives (EHP, Bow Acceleration, Displacement)

Length m. Beam m. Draft m. Displacement m3
22.2 7.36 2.66 231

All computer-based study and programing was done by MATLAB and it takes
about 20 s to run the program on a modest PC including the graphical work (Fig. 9).

3.2 Planing Hulls

To validate the computer program written for planing hulls, a sample example provided
by Savitsky and Brown [13] is numerically tested. The results are shown below in
graphical forms.
The input to the program can be summarized as follows. All input as in the example
by the above reference are given here for completeness. The mass of the hull was taken
as 186000 lb, Length overall = 80 ft, the location of longitudinal center of the mass
was at 34 ft from stern. Beam interval used for the search for the ‘better hull’ was 24 to
25 ft. and the speed interval 24 to 27 knots and deadrise interval was 10° to 20°. Some
of the intervals were rather narrow just to compare the results obtained by the design
program for a ‘better’ hull, with the existing example and for the software calibration
purposes. The following results are provided in graphical forms.

Fig. 9. EHP versus parameters length beam for displacement hull design.
184 S. Calisal

Fig. 10. Plot of the nondimensional acceleration at the centroid of mass versus

The Fig. 10 gives the nondimensional acceleration versus non-dimensional calm

water EHP. The figure suggests that for planing hulls the acceleration increases in
general with the power installed. The better hull does not correspond, however, to the
smallest EHP within the design space defined. At lower EHP Fig. 10 suggests that the
acceleration at LCG is expected to be higher. Red circle is the chosen ‘better’ hull. The
red point is at about 81% of maximum power and at 66% of maximum acceleration
level calculated for the population of the candidate designs.
Figure 11 gives the non-dimensional power, nondimensional acceleration and the
Froude number based on the beam for the better hull and 4 other hulls near the ‘better’
hull. Design parameters for the ‘better’ hull corresponding to the number one design
and four other designs near the better hull on the Pareto front. Figure 12 gives the
dynamic trim, deadrise angle and the average beam selected by the program for the set
of 5 hulls near and at the better hull. Better design number 1 corresponds to the design
nearest to the utopic point. Trim angle is in degrees, Dead Rise in degrees and Beam
value is in meter for the Better design and four near designs. We now compare the
design values by the procedure outlines and the values provided by Savitsky at three
speeds. Figure 13 provides the values of the EHP calculated by Savitsky at three speeds
and the corresponding values for the best speed and EHP obtained by the design
software for the ‘better’ hull. The design effective horse power value for the better hull
at the better speed remains reasonable and possibly somewhat less compared the values
provided by Savitsky [13]. The EHP values of the Savitsky example at three speeds
show an almost linearly changing EHP values with the speed. The EHP value calcu-
lated by the design software seems to be 1.5% higher than the values given by Savitsky
example values.
 Multi Objective Design of Ships; A Pareto Procedure 185

Fig. 11. Non-dimensional power, non-dimensional acceleration and Froude number based on
the beam for the set of ‘better hulls’.

Fig. 12. Trim angle, dead rise angle, and beam values for the ‘better hull’
186 S. Calisal

Fig. 13. Effective horse power vs speed. Red circle is the selected speed by the program

Fig. 14. EHP required for the added resistance at significant waves of 4.6 ft and at the selected
speed.
 Multi Objective Design of Ships; A Pareto Procedure 187

Figure 14 provides the comparison of the EHP value given for the added resistance
values in the Savitsky example and the same value obtained by the design software. At
a relatively less forward speed in head seas, a comparable added resistance value is
simulated by the design software.

Fig. 15. Dynamic stability check for the ‘better’ hull parameters. Trim and the critical trim of
the design population and the critical tau values versus Cv the speed coefficient.

Finally the Fig. 15 compares the dynamic trim calculated for the better design with
the critical trim calculated by the algorithm provided by Celano T. III [21]. The
expected dynamic trim of this hull at the design speed is well below the critical trim
suggested by the algorithm. That means that the hull will not experience porpoising
under the expected design conditions.

4 Conclusion

The study on the possibility of finding a hybrid method for the preliminary design of
displacement-type hulls and planing-type hulls with multiple objectives, suggests that a
computerized method for ship design based on the concepts of Pareto front can be
successfully used by the designers. The key elements necessary for this, are the specific
algorithms based on the Froude number or type of flow around the ship. Once the
188 S. Calisal

design space is narrowed down to a much smaller design space, one can possibly then
use other computational methods such as CFD for resistance, sea keeping, etc.
The major emphasis in ship design is now for environmental protection and for
good reasons on the fuel consumption. This study is aimed to see if an ‘optimum’ hull
could be designed by including more design objectives and still save fuel. The fuel
consumption level and motion parameters are chosen as the design performance
objectives. The study shows that a compromised solution could be obtained. Additional
‘weight’ concepts may be used by the designers and other performance measures could
be added in the design criteria. Acceleration and fuel consumption levels seem to have
a convex Pareto front and allow a quick systematic identification of the ‘better’ hull
parameters.
The procedure outlined for two objectives could be extended to more objectives
and could be expanded easily to multiple performance objectives in a systematic way.
It is observed that as the acceleration level is reduced and so was the added resistance
which also means that the fuel requirement is reduced as well.
Our conclusion is that computerised methods based on Pereto concepts are effec-
tive, and they can be used successfully for the preliminary design of ships.

Acknowledgements. The author extends his thanks to Prof. O Goren for providing information
on a more sophisticated CFD based design optimization procedure that he was involved in, that
in turn inspired this algorithmic study. Thanks are also extended to Professors Ziya Saydam,
Professor Serhan Gökçay and Onur Yurdakul for the improvement of the procedure and the data
provided for validation.
Thanks are also extended to NSERC and Piri Reis University for providing the necessary
support for this study. The programming is done with an academic version of MATLAB at
University of British Columbia and at the Piri Reis University.

References
1. Akintürk, A.: Inclusion of a crew safety node into the preliminary design of fishing vessels.
Doctoral thesis, Mechanical Engineering Department, University of British Columbia, May
1997
2. Wu, C.: Multi criterion preliminary design of a tetrahedral platform. In: 36th AIAA/
ASME/ASCE/ASC Structures Structural Dynamics and Material Conference, New Orleans,
LA, 10–13 April 1995 (1995)
3. Kim, H.Y., Vlahogoulos, N.: A multi-level optimization algorithm and a ship design
application. In: 12th AIAA Aviation Technology, Integration and Operation Conference and
14th AIAA/ISSM, Indianapolis, Indiana, 17–19 September 2012 (2012)
4. Diez, M., Serani, A., Campana, E., Goren, O., Sarioz, K., Danisman, D.B., Grigoropoulos,
G., Aloniati, E., Visonneau, M., Queutey, P., Stern, F.: Multi-objective hydrodynamic
optimization of the DTMB 5415 for resistance and seakeeping. In: 13th International
Conference on FAST 2015, Washington D.C., USA, 1–4 September 2015 (2015)
5. Calisal, S.M., McGreer, D.: A resistance study on a systematic series of low L/B vessels.
SNAME Mar. Technol. 30(4), 286–297 (1993)
6. Calisal, S.M., Howard, D., Mikkelsen, J.: A seakeeping study of the UBC series. SNAME
Mar. Technol. 34(1), 10–24 (1997)
 Multi Objective Design of Ships; A Pareto Procedure 189

7. Jensen, J.J., Mansour, A.E., Olsen, A.S.: Estimation of ship motions using closed form
expressions. J. Ocean Eng. 31, 61–85 (2004)
8. Faltinsen, O.M.: Hydrodynamics of High Speed Marine Vechicles, 454 p. Cambridge
University Press, Cambridge (2005)
9. Yamamoto: See 7 for the equation Original paper not found (1986)
10. Heliotis, A.D., Goudey, C.A.: Tow tank results of bulbous bow retrofits on New England
trawler hulls. Center for Fisheries Engineering Research, MIT, September 1985
11. Bhattacharyya, R.: Dynamics of Marine Vehicles, p. 498. Wiley, New York (1978)
12. Ochi, M.K.: Extreme behaviour of a ship in rough seas. Trans. SNAME 69, 159–170 (1964)
13. Savitsky, D., Brown, P.W.: Procedures for hydrodynamic evaluation of planing hull in
smooth and rough water. Mer. Technol. 13(4), 381–400 (1976)
14. Savitsky, D.: On the subject of high-speed monohulls. Presented to the Greek Section f the
Society of Naval Architects and Marine Engineers Athens, Greece, 2 October 2003
15. Taunton, D.J., Hudson, D.A., Shenoi, R.A.: Characteristics of a series of high speed hard
chine planing hulls part II: performance in waves. Int. J. Small Craft Technol. 153, B1–B22
(2011)
16. Taunton, D.J., Hudson, D.A., Shenoi, R.A.: Characteristics of a series of high speed hard
chine planing hulls part 1: performance in calm water. Int. J. Small Craft Technol.
152(Part B2), 55–75 (2010)
17. De Luca, F., Pensa, C.: The Naples warped hard chine hulls systematic series. J. Ocean Eng.
139, 205–236 (2017)
18. Bagovic, E., Berorello, C., Pennini, S.: Planning hull sea keeping in irregular head seas.
Trans. FAMENA XXXVIII(3), 1–2 (2014)
19. Grierson, D.E.: Pareto analysis of pareto design. In: Joint International Conference on
Computing and Decision Making in Civil and Building Engineering, Montreal, Canada, 14–
16 June 2006 (2006)
20. Kasprzak, E.M., Lewis, K.E.: Pareto in multi objective optimization using the collinearity
theorem and scaling method. Struct. Multi. Optim. 22(3), 208–218 (2001)
21. Celano III, T.: The prediction of porpoising inception for modern planing craft. Trident
report No. 254, US Naval Academy (1998)
 Optimal Arrangement Method of a Ship
Considering the Performance Against Flooding

Ki-Su Kim and Myung-Il Roh(&)

Seoul National University, Seoul, Republic of Korea

[email protected]

Abstract. Most of the ship’s accidents are accompanied by flooding. Also,

water inflow due to the flooding propagates along with complex compartments
of the ship, so the flooding process is very complicated. For the robust design of
the ship against flooding, it is necessary to consider such a process at the stage of
the arrangement design. However, at present, the arrangement design is carried
out considering only the final state of the ship due to damage that generally
causes the flooding. At the intermediate state, dynamic effects and various
factors can lead to a more severe condition of the ship than the final state. On the
other hand, the flooding process can be vary depending on the arrangement
design. To improve the arrangement design in the aspect of safety, we propose
an optimal arrangement method of a ship considering the flooding. For this, an
optimization problem for the arrangement design is mathematically formulated.
The formulation includes conventional objective functions for the arrangement
design such as the operability. In terms of flooding, we add the additional
objective functions such as the total time to flooding and the amount of motion
during the flooding. The optimization problem includes constraints such as the
requirements of damage stability. At this time, the flooding analysis, the most
important part of the problem, is performed using the position based dynamics
(PBD), which is a kind of mesh-free particle methods. Finally, the proposed
method is applied to an example of a naval ship that the survivability is one of
the important issues. The result shows the applicability of the proposed method
at the arrangement design of the ship.

Keywords: Arrangement design
Ship flooding, optimization, position based

dynamics
Naval ship

1 Introduction

1.1 Research Background

The arrangement design of ships is normally being made based on data of the past
design and experts’ experiences. When performing the arrangement design, experts
refer the past design and add additional requirements such as the owner’s requirements
and regulations and additional information to the data for a new design based on their
experiences. When performing the arrangement design, delay in design can occur due
to the missing data or the absence of experts. Moreover, most of the ships are operated
in the ocean so that spaces that can be filled with components for them are restricted.

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 190–197, 2021.
https://doi.org/10.1007/978-981-15-4680-8_13
 Optimal Arrangement Method of a Ship 191

For these reasons, the demand for the optimization method for the arrangement design
for ships is proposed in this study.
During the arrangement design process, there are several things to consider. Firstly,
ships should satisfy the international code and regulations because they have to be
operated in the ocean under the permission of related organizations or governments.
Next, the ship should be designed by the ship owner’s requirements. And the sur-
vivability is the most important factor in the operation of the ship. Therefore, in this
study, we tried to consider the intermediate flooding during the arrangement design.
Traditionally, the rules for ships were judged whether they are satisfied with sta-
bility only by the final state of ships after accidents, so that, there was a lack of
understanding of the flooding process and time to reached the final state. The most
accidents in the ocean accompany flooding. And, the process progresses for quite a
long time. Therefore, the accurate estimation of the flooding process for ships is very
important no less than appropriate after action for accidents [1]. Moreover, the
arrangement design of ships is closely related to the speed of the flooding process, so
that, we tried to consider this as one of the objectives of evaluation of arrangement
design.

1.2 Related Works

There are several pieces of research about the arrangement design and the flooding
analysis. However, research that considers both of them at the same time is hard to find.
Therefore, in this section, the researches about arrangement design and flooding
analysis were reviewed separately.
Several researchers have proposed a method to find an optimal arrangement design
of ships. Helvacioglu and Insel [2] proposed a multi-stage expert system to arrange the
compartments on a container ship. Nick [3] proposed a method to generate, evaluate,
and optimize the arrangement design of a naval surface ship. Roh et al. [4] proposed a
method to layout bulkheads in the bull of a naval surface ship. And Jung et al. [5]
consider the stability, operability, and survivability to optimize the arrangement design
of the naval surface ship.
There are many studies that analyze the flooding process of the ships. Ruponen [6]
used the quasi-static method based on a hydraulic orifice equation to perform analysis
for intermediate flooding. Dankowski et al. [7] analyzed the intermediate state of
flooding of a ship using the quasi-static method and considered the air compression in
the compartments. Kim et al. [8] performed the intermediate flooding simulation for the
barge-type damaged ship. Lee [9] has improved the hydraulic orifice equation to
suggest the dynamic orifice equation and performing quasi-static flooding simulation
for the passenger ship. As mentioned above, there are many studies about arrangement
design and flooding analysis for ships. However, they did not consider the survivability
on the side of flooding. Therefore, in this study, we proposed an optimal arrangement
method of a ship considering the flooding.
192 K.-S. Kim and M.-I. Roh

2 Optimal Arrangement Method

In ships, there are various things to optimize. Thus, it is challenging to solve the
problems considering all the variables at the same time. In this manner, multi-stage
optimization problems and optimization algorithms for solving the problems was
proposed in this study, as shown in Fig. 1.

Fig. 1. The configuration of the optimal arrangement method

As shown in Fig. 1. In the first stage, the main compartments of the ship are
optimized, considering the strength, and performances. In the second stage, the sub-
compartments in the compartment are optimized, considering the operability and the
effectiveness for crews and passengers. Finally, the alternatives derived from the two
stages are evaluated using flooding analysis method.

2.1 Multi-stage Optimization

This section describes an optimization method for the arrangement of ships. As
mentioned earlier, the arrangement method for the ships was formulated in two stages
in this study.
First Stage-Optimization of Compartment Arrangement
The first stage is the compartment arrangement where the positions of the bulkheads
and decks in the hull are determined by minimizing the maximum bending moment at
the intact state, the space for liquid tanks. The bending moment can be calculated by
the weight curve and the buoyancy curve of a ship. By minimizing the bending
moment, the bulkhead arrangement can be safer in terms of structural safety. Also, the
more cargos or passengers can be loaded by minimizing the space for liquid tanks.
These two considerations were set to objective functions of the first stage as shown in
(1), and (2).
 Optimal Arrangement Method of a Ship 193

Minimize F1 ¼ maxðBending momentÞ ð1Þ

Minimize F2 ¼ VFOT þ VFWT þ VWBT þ VLOT ð2Þ

When arranging the bulkheads and decks in the hull, we should obey the minimum
requirements for liquid tanks and the space for engine rooms. In general, the space for
the liquid tanks may have required a given volume for operation. Therefore, in this
study, the upper and lower limits for each liquid tank were constrained, as shown in
(3)–(10). In the equations, FOT is a fuel oil tank, FWT is a freshwater tank, WBT is a
water ballast tank, and LOT is liquid oil tank.

g1 ¼ VFOT
min
 VFOT  0 ð3Þ

g2 ¼ VFOT  VFOT
max
0 ð4Þ

g3 ¼ VFWT
min
 VFWT  0 ð5Þ

g4 ¼ VFWT  VFWT
max
0 ð6Þ

g5 ¼ VWBT
min
 VWBT  0 ð7Þ

g6 ¼ VWBT  VWBT
max
0 ð8Þ

g7 ¼ VLOT
min
 VLOT  0 ð9Þ

g8 ¼ VLOT  VLOT
max
0 ð10Þ

If the length of each of the engine rooms should have a minimum length and width,
such requirements can be formulated as constraints as shown in (11) and (12).

ER  jx5  x6 j  0
g9 ¼ lmin ð11Þ

ER  jx7  x8 j  0
g10 ¼ lmin ð12Þ

Second Stage-Optimization of Sub-compartment Arrangement

After the compartment arrangement which was done in the first stage. Regarding the
operability and effectiveness of a ship, the sequence of rooms and the positions of the
passages in the compartments are important during the sub-compartment arrangement.
To enhance the operability, the adjacency index that represents the effectiveness of the
flow of crews, passengers, and supplies in the ship should be minimized. As following
these reasons, the process for the sub-compartment arrangement was formulated as an
optimization problem. The objective function of this stage minimize the adjacency
index (AI) in (13). In the equation, Ri is sub-compartment to be allocated at the
compartment, qij is adjacency coefficients between the room i and j, and dij is the
distance between the rooms.
194 K.-S. Kim and M.-I. Roh

XNR 1 XNR  
Minimize F1 ¼ AI ¼ i¼1 j¼i þ 1
qij
dij ð13Þ

When arranging the sub-compartment and passages in the compartment, the area for
sub-compartment and required locations of the specific sub-compartment should be
satisfied. These two constraints can be formulated as (14), and (15). In the equations, ai
is an area of rooms, and Ri is sub-compartment to be allocated at the region i.

g1 ¼ amin
i  ai  0 ð14Þ

g2 ¼ Ri  Rreq
i 0 ð15Þ

2.2 Flooding Analysis in the Time Domain

After the optimization, compartments, sub-compartments, and passages are determined,
which means the path of the water inside the hull is determined when flooding is
occurring. To evaluate the survivability in terms of flooding, flooding analysis was
performed after the optimization. In this study, dynamic analysis for flooding was
performed using (16).

Mx ¼ FGravitational þ FBuotancy þ FHydrodynamic ð16Þ

By solving (16) in each time steps, the acceleration of the ship can be obtained.
Using the obtained acceleration, the velocity and position vector can be calculated by
integrating the acceleration. When flooding is occurring, M in (16) is time variant value
about ships. Depending on the inflow and outflow of water and air, the weight and
center of weight can be changed from time to time. To calculate the flow of fluid in
each time, we use the Position based dynamics (PBD), which is one of the Mesh-free
particle method (MPM) [10].
After the first and second stage of optimization, the flooding analysis was per-
formed for ships considering loading conditions and damage cases. By doing this, we
can check accurate stability after the accident if the stability is not satisfied even in one
case. The optimization is re-performed. We are changing the bulkheads or rooms
considering the vulnerable compartments.

3 Application

To check the applicability of the proposed method, the prototype program was
developed. Using the prototype program, optimization of arrangement was performed,
and the solution was evaluated using flooding analysis of the Navy destroyer. The
results of first-stage were shown in Fig. 2. As shown in Fig. 2, the bulkheads were
adjusted to minimize the maximum bending moment. The objectives were to minimize
approximately 20%.
 Optimal Arrangement Method of a Ship 195

Fig. 2. Optimization results of first-stage

Using the results of first-stage, second-stage was performed. The results of the
second-stage were shown in Fig. 3. As shown in Fig. 3, the room arrangement of the
superstructure was optimized. By performing second-stage, the adjacency index for
rooms was enhanced by about 15%.

Fig. 3. Optimization results of second-stage

After two stages of optimization, intact and damage stability was checked using the
prototype program. And it was confirmed that the eight criteria for stability (3 for intact
and 5 for damage) were meet the criteria as shown in Fig. 4.
Moreover, we check the possibility of enhancing the survivability in an accident by
optimizing the arrangement of the compartment and the sub-compartment.
196 K.-S. Kim and M.-I. Roh

Fig. 4. The result of the flooding analysis for oil tanker

4 Conclusions

In this study, a method of optimization of arrangement considering the performance

against flooding. For many variables, a multi-stage optimization was proposed. After
the optimization of compartments and sub-compartments in two stages, the surviv-
ability in terms of flooding was evaluated. Dynamic flooding analysis was performed
using the equations of motion, and PBD was used for calculating the amount of flow at
each time steps. By applying the proposed method to the oil tanker, the applicability of
the proposed method was evaluated.

Acknowledgments. This work was partially supported by (a) The National Research Founda-
tion of Korea (NRF) grant funded by the Ministry of Education, Science and Technology
(No.2016R1A2B4016253) and (b) Research Institute of Marine Systems Engineering of Seoul
National University, Republic of Korea.

References
1. Kim, K.S., Roh, M.I., Kim, J.S.: A method of dynamic flooding analysis in time domain for
the compartment arrangement design of a ship. In: Proceedings of the Twenty-seventh
International Ocean and Polar Engineering Conference 2017, San Francisco, USA, pp. 862–
866 (2017)
2. Helvacioglu, S., Insel, M.: A reasoning method for a ship design expert system. Expert Syst.
22(2), 72–77 (2005)
3. Nick, E.K.: Fuzzy optimal allocation and arrangement of spaces in naval surface ship design.
Ph.D. thesis, The University of Michigan, USA (2008)
4. Roh, M.I., Lee, S.U., Lee, K.Y.: Development of an optimal compartment design system of
naval ships using compartment modeling and ship calculation modules. J. Soc. Naval Archit.
Korea 46(4), 424–434 (2009)
 Optimal Arrangement Method of a Ship 197

5. Jung, S.K., Roh, M.I., Kim, K.S.: Arrangement method of a naval surface ship considering
stability, operability, and survivability. Ocean Eng. 152, 316–333 (2018)
6. Ruponen, P.: Progressive flooding of a damaged passenger ship, Ph.D. theses, Helsinki
University of Technology, Finland (2007)
7. Dankowski, H., Russel, P., Kruger, S.: New insight into the flooding sequence of the costa
Concordia accident. In: Proceedings of the 33rd International Conference on Ocean,
Offshore and Arctic Engineering 2013, Nantes, France, pp. 1–10 (2013)
8. Kim, K.S., Roh, M.I., Lee, S.M.: Flooding analysis based on PBD (Position Based
Dynamics) for ships and offshore structures. In: Proceedings of the Twenty-eighth
International Ocean and Polar Engineering Conference 2018, Sapporo, Japan, pp. 955–
957 (2018)
9. Lee, G.J.: Dynamic orifice flow model and compartment models for flooding simulation of a
damaged ship. Ocean Eng. 109, 635–653 (2015)
10. Bender, J., Muller, M., Otaduy, M.A., Teschner, M., Macklin, M.: A survey on position-
based simulation method in computer graphics. Comput. Graph. Forum 33(6), 228–251
(2014)
 Optimal Design of a Floating Raft Vibration
Isolation System with Multiple Constraints

Wenzhang Liu(&), Wenwei Wu, and Xuewen Yin

National Key Laboratory on Ship Vibration and Noise,

China Ship Scientific Research Center,
222 Shanshui East Road, Wuxi 214082, China
[email protected]

Abstract. This paper presents an optimal design strategy for a floating raft
vibration isolation system which is extensively deployed for mounting engines
in ships and underwater vehicles. Multiple design variables of the floating raft
including total mass, static stress, resonance frequencies, and vibration isolation
effects within a prescribed frequency range are taken into account so as to
determine the optimization objective, which are the critical concerns in early
design stage, i.e., vibration attenuation and the comfort of the crews. In addition
to the conventional constraints like geometric sizes and mass, the static stress
and vibration level of the floating raft vibration isolation system under specific
load cases are also considered. According to the above requirements and prin-
ciples, an optimal model is developed. Through comprehensive optimization
analysis, an optimized design of a floating raft system is provided, which can
ensure lower vibration level of the foundation in a prescribed frequency range
without at expenses of unnecessary weight and excessive equipment motions.
This design strategy would lend itself to bridge structural and acoustic design in
a unified way during early design stage.

Keywords: Floating raft isolation system
Acoustic optimization design

Multiple constraints

1 Introduction

Motors and engines are the major mechanical sources that induce inevitable vibration
and noise within ships and underwater vehicles. To effectively control such mechanical
vibrations plays an important role in improving the acoustic stealth performance and
the comfort of the crews of ships and other vehicles. Mechanical vibration isolation
technology is one of the most effective means to reduce the vibration transmission from
the mechanical equipment to the mounting base. As far as its development history is
concerned, it has experienced three distinct stages, namely, single-layered vibration
isolation, double-layered vibration isolation and floating raft vibration isolation, among
which floating raft vibration isolation device is the most effective in attenuating
vibration transmission, and, as a result, the vibration of the hull structure and under-
water radiation noise can be effectively reduced. The general floating vibration isola-
tion system includes several vibration isolators attached to the top side as well as the

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 198–208, 2021.
https://doi.org/10.1007/978-981-15-4680-8_14
 Optimal Design of a Floating Raft Vibration Isolation System 199

bottom side of the common raft frame and its can roughly reduce radiation noise by
about 20–40 dB.
The traditional design method of the floating raft vibration isolation system is
stemming from a preliminary structural form based on engineering experience or
theoretical knowledge, and after an exhausting multidisciplinary analysis and verifi-
cation, specific design requirements can be secured. Since the design objective and
constraint variables are highly correlated and mutually coupled between each other, the
traditional serial design method is passive and time-consuming. Recently, a novel
design strategy, namely, optimal design process is well accepted instead of the tradi-
tional design method, in which multiple optimization objectives more than floating
vibration isolation effect are proposed by accounting for multiple constraints such as
total mass, static strength, natural frequencies of the isolation system, vibration level of
the hull foundation and vibration intensity of the equipment. At present, exhausting
research works associated with floating raft vibration isolation system are dedicated to
the performance analysis of the vibration isolation system or the specific design with
specific constraints, few attentions are paid on multi-objective comprehensive system
design [1, 2].
This paper proposes an optimal design strategy for a floating raft vibration isolation
system, which considers the mutual coupling relationship between the various objec-
tives and constraints of the floating raft vibration isolation system from the global level.
By improving the design variables such as geometric dimensions, under the estab-
lishment of an optimization model, each objective is calculated in parallel and the most
main design variables are determined based on sensitivity analysis. The rapid collab-
orative optimization design with multi-objective and multi-constraints is achieved, and
the design results meet the performance requirements of each target discipline. This
optimization design method plays a bridge role in the unified design of structure and
acoustics at various design stages of the ship.

2 Establishment of an Optimized Design Model

In this paper, a simplified model of flat frame floating raft is developed so as to study
the whole optimization design process and to achieve the optimized raft frame scheme
[3, 4].

2.1 Geometric Models and Design Variables

The geometric model of a floating raft vibration isolation system for the flat frame type
consists of the excitation machine - four vibration isolators – raft frame - six raft
isolators - hull foundation, as shown in Fig. 1.
The design variables for the whole model are the thickness of each panel of the raft
frame, which are divided into four parts: top, side, inner and bottom. In order to ensure
that the projection coordinates of the center of gravity of the raft in the horizontal plane
are consistent with that of the machine, and because the vibration isolators are uni-
formly symmetrically arranged, the thickness of each panel that is symmetrical about
the center of the raft shall be coincident. Therefore, the thickness of the centrally
200 W. Liu et al.

Fig. 1. Simplified geometric model of floating raft isolation system

symmetrical panel is indicated by the same design variable and the labels are marked
counterclockwise (Fig. 2).

Fig. 2. Label and location of design variables in the model

The thickness of each panel in the initial model is 0.01 m, and the upper and lower
range of the design variables is listed in Table 1.

2.2 Optimization Design Principles – Objectives and Constraints [5, 6]

For the global optimization of the floating raft vibration isolation system, it is necessary
to consider the optimization design principles, such as natural vibration characteristics,
static strength, frequency response, vibration isolation effect, total mass etc., which are
summarized respectively as follows:
(1) Natural vibration characteristics: The natural frequencies of the system should be
designated to avoid excitation frequencies of the machine and avoid structural
resonance. In this paper, the working frequency of the mechanical equipment is
85 Hz. Calculation and analysis of the initial model shows that the first-order
bending frequency of the raft is 85 Hz, and so the initial model is at risk of
resonance. Therefore, in the optimized design, to the first-order bending frequency
(raft_fre1) and the second-order resonant frequency (raft_fre2) of the raft, the
constraint variables are all greater than 85 Hz, and the optimization direction of the
objectives(frequencies) is set to the maximum (Fig. 3).
(2) Static strength: It is required that the structure cannot fail when subjected to given
excitation forces. The maximum stress of the raft in the initial model under the
excitation and gravity is 4.385 MPa. In order to ensure that the bearing
 Optimal Design of a Floating Raft Vibration Isolation System 201

Table 1. Design variables

Name Lower bound/m Initial value/m Upper bound/m
thickness_bottom1 0.005 0.01 0.02
thickness_bottom2 0.005 0.01 0.02
thickness_bottom3 0.005 0.01 0.02
thickness_bottom4 0.005 0.01 0.02
thickness_inner1 0.005 0.01 0.02
thickness_inner2 0.005 0.01 0.02
thickness_inner3 0.005 0.01 0.02
thickness_inner4 0.005 0.01 0.02
thickness_inner5 0.005 0.01 0.02
thickness_inner6 0.005 0.01 0.02
thickness_inner7 0.005 0.01 0.02
thickness_inner8 0.005 0.01 0.02
thickness_side1 0.005 0.01 0.02
thickness_side2 0.005 0.01 0.02
thickness_side3 0.005 0.01 0.02
thickness_side4 0.005 0.01 0.02
thickness_top1 0.005 0.01 0.02
thickness_top3 0.005 0.01 0.02
thickness_top2 0.005 0.01 0.02
thickness_top4 0.005 0.01 0.02

Fig. 3. Natural frequencies and mode shapes of initial model (left 85 Hz, right 90 Hz)

performance of the optimized model is not reduced, the maximum stress

(max_stress) of the raft in the optimized design is not more than 4380000 Pa, and
the optimization direction of the objectives (stress) is set to the minimum (Fig. 4).
(3) Frequency response: Under the frequency response vibration, the vibration inten-
sity of the mechanical equipment near the working frequency is not exceeded to
standard requirements, and the vibration level of the ship foundation is kept as
small as possible. The vibration levels at the frequencies 80 Hz, 85 Hz and 90 Hz
are calculated and the vibration level at the mechanical feet (vibration_M) is
guaranteed to be no greater than that of the initial model. The optimization
direction of the vibration level of F1 and F2 at the foundation (vibration_F1,
vibration_F2) is set to the minimum (Fig. 5).
202 W. Liu et al.

Fig. 4. Stress contours of initial model under machine excitation force and gravity

Fig. 5. Frequency response of initial model under machine excitation force (85 Hz)

(4) Vibration isolation effect: The vibration isolation effect of the floating raft vibration
isolation system is evaluated by the vibration acceleration drop from machine to
foundation DLT (vibration_level_drop). The calculated frequency points are set to
80 Hz, 90 Hz and 95 Hz, which are identified by 1, 2 and 3 respectively in the
variable suffixes. In the optimization design, the constraint (vibration level drop) is
not lower than that of the initial model, and the optimization direction of the
objectives (vibration_level_drop) is set to the maximum.

DLT ¼ LU  LD ð1Þ

Where: LU -Vibration acceleration level at the machine foot,

dB;LD -Vibration acceleration level at the ship foundation, dB.
(5) Total mass: Lightweight design principles in the green ship design concept are
emphasized, which include eliminating redundant materials without sacrificing
performance. In the optimization process, under the premise of ensuring the per-
formances based on the above design principles, the total mass caused by the
variation of the design variables does not exceed to 12000 kg, and the optimization
direction of the objectives (total mass) is set to the minimum.

2.3 Global Optimization Model

According to the above five optimization design principles, a global optimization
model is established. The design variables are panel thicknesses of the raft, and the
objectives and constraints meet the above design principles and requirements, and the
objectives are coupled to each other (see Table 2 and 3 respectively).
 Optimal Design of a Floating Raft Vibration Isolation System 203

Table 2. Constraints variables

Name Lower bound Upper bound
max_stress 4380000
raft_fre1 85
raft_fre2 85
total_mass 12000
vibration_level_drop1 15
vibration_level_drop2 8
vibration_level_drop3 3.5

Table 3. Objectives direction

Name Direction Scale factor Weight factor
max_stress minimize 1 1
raft_fre1 maximize 1 1
raft_fre2 maximize 1 1
total_mass minimize 1 1
vibration1_F1 minimize 1 1
vibration1_F2 minimize 1 1
vibration2_F1 minimize 1 1
vibration2_F2 minimize 1 1
vibration3_F1 minimize 1 1
vibration3_F2 minimize 1 1
vibration_level_drop1 maximize 1 1
vibration_level_drop2 maximize 1 1
vibration_level_drop3 maximize 1 1

3 Optimization Design Method and Process

3.1 General Optimization Process
Firstly, an integrated optimization design platform is developed. The above objectives
and constraints are calculated by the parameterized finite element method, respectively,
through three modules of frequencies analysis, static analysis, and frequencies
response. And the above modules are integrated into the optimization platform to be
called to calculate the objectives and constraints. Through the sensitivity analysis, the
main variables are determined in order to improve the efficiency of optimized design.
After the results collection and objectives calculation each step, all variables are sub-
mitted to the optimization platform. The genetic evolution algorithm (NCGA) is written
and the design process is completed. (see Fig. 6)
204 W. Liu et al.

Fig. 6. General flow of optimization design process

3.2 Sensitivity Analysis [7, 8]

Considering the structural characteristics of the floating raft vibration isolation system,
the sensitivity of each objective to the structural parameters is calculated. During the
sensitivity analysis process, through the optimal Latin hypercube method, the design
matrix of design variables is obtained. Then the value of each single design variable is
changed, and the average value of the effect of the structural parameters on the
objectives is calculated with all possible combinations of this changed design variable
and other variables of the whole design matrix. This above process is performed once
for each design variable, and Pareto diagrams are got (see below Fig. 7), which reflect
the percentage of contribution of every design variable to each objective. Conse-
quently, the main design variables are determined. As shown in Fig. 7, sensitivity
results for frequencies, static stress and total mass are listed. The main variables which
have the greater weight factor are shown in Table 4.

3.3 Optimization Procedure [9, 10]

Multi-objective genetic evolutionary algorithm is applied in optimization process. The
whole iterative process has a total of 1200 steps, as shown in Fig. 8, which lists the
changes of the values of the frequency and static stress in the iterative calculation
process. After the comprehensive evaluation, the global optimal solution appears at
step 927.
 Optimal Design of a Floating Raft Vibration Isolation System 205

Fig. 7. Sensitivity analysis plot

Table 4. Sensitivity analysis results

Objectives The main design variables
max_stress thickness_side1, thickness_side3, thickness_top4
raft_fre1 thickness_bottom4, thickness_top4, thickness_inner2
raft_fre2 thickness_side4, thickness_inner7, thickness_top2
total_mass thickness_top1 * 4, total thickness_bottom1 * 4
vibration_F thickness_bottom4, thickness_inner2, thickness_top4
vibration_level_drop thickness_side4, thickness_inner7, thickness_top2

Fig. 8. Change history of objectives in optimization procedure

206 W. Liu et al.

4 Comparative Analysis of Results Before and After

Optimization

The optimal design variables obtained after optimization are shown in Table 5.

Table 5. Optimal design variables

Name Optimal value/m Initial value/m
thickness_bottom1 0.014 0.01
thickness_bottom2 0.005 0.01
thickness_bottom3 0.0063 0.01
thickness_bottom4 0.0076 0.01
thickness_inner1 0.0081 0.01
thickness_inner2 0.016 0.01
thickness_inner3 0.019 0.01
thickness_inner4 0.014 0.01
thickness_inner5 0.0055 0.01
thickness_inner6 0.005 0.01
thickness_inner7 0.011 0.01
thickness_inner8 0.0058 0.01
thickness_side1 0.015 0.01
thickness_side2 0.010 0.01
thickness_side3 0.018 0.01
thickness_side4 0.0076 0.01
thickness_top1 0.0056 0.01
thickness_top3 0.0074 0.01
thickness_top2 0.012 0.01
thickness_top4 0.0064 0.01

The objectives values before and after optimization are shown in Table 6.

Table 6. Optimal objectives

Name Optimal value Initial value Change proportion
max_stress 4384530 Pa 2728440 Pa −37.8%
raft_fre1 85 Hz 88 Hz 3.5%
raft_fre2 90 Hz 93 Hz 3.3%
total_mass 9816 kg 10000 kg −1.9%
vibration2_F1 (85 Hz) 2.6 * 10−4m/s2 1.6 * 10−4 m/s2 −38.5%
vibration2_F2 (85 Hz) 4.1 * 10−4 m/s2 3.8 * 10−4 m/s2 −7.3%
vibration_level_drop1 15 dB 23 dB 53.3%
vibration_level_drop2 8 dB 14 dB 75.0%
vibration_level_drop3 3 dB 21 dB 600%
 Optimal Design of a Floating Raft Vibration Isolation System 207

5 Conclusion

Through the above optimization process and comparison of optimization results, the
following conclusions can be drawn:
(1) With the total mass increase by 1.9%, the maximum stress is reduced by 37.8%,
which increases the structural safety factor. The natural frequency is increased by
about 3.5%, which is offset from the operating frequency of the machine at 85 Hz
to avoid resonance. The vibration level at the support points of ship foundation is
reduced by 38.5% and 7.3% respectively, and the vibration level drop (vibration
isolation effect) of the floating raft isolation system is increased by more than 50%,
effectively reducing the foundation vibration level. The radiation noise passing
through the ship hull to the underwater is reduced, and the acoustic stealth per-
formance is enhanced.
(2) This method breaks through the drawbacks of the traditional iterative design
method based on experience. Considering the multi-objectives and multi-
constraints in the optimization design of the floating vibration isolation system,
and analyzing the high coupling relationship between the objectives and con-
straints, the collaborative optimization design between the objectives is realized.
(3) The optimization method is applicable to the design of the vibration isolation
system at each design stage of the ship. The optimization design principles con-
sidered in this paper are basically the same in different design stages. Through this
method the structural parameters that meet the design requirements can be quickly
and efficiently obtained.

Acknowledgment. The authors wish to thank High-Tech Ship Fund from the Ministry of
Industry and Information Technology (MIIT): Deepwater Semi-submersible Support Platform
(Grant No.: 2016 [546]), High Quality Brand Ship Board Machinery (Grant No.: 2016 [547]).
This work was also partially supported National Science Foundation of Jiangsu Province-Youth
Fund (Grant No.: BK20170217).

References
1. Li, T.Y., et al.: Structural power flow analysis for a floating raft isolation system consisting
of constrained damped beams. J. Sound Vib. 202(1), 47–54 (1997)
2. Wuhan: Simulating calculation of onboard floating raft system based on finite element
method. J. Wuhan Univ. Tech. 16(20):2023–2029 (2006)
3. Zhang, W.Y., Mesbahi, E.: Optimization design on floating raft system for advancing
vibration isolation characteristics. In: International Conference on Consumer Electronics
(2012)
4. Liang-Bi, L.I., Wang, G.Z., Wen, H.B.: Finite element analysis and experimental research on
the dynamical properties of the ship model with a floating raft installation. J. East China
Shipbuilding Inst. (2001)
5. Allaire, G., Jouve, F.: A level-set method for vibration and multiple loads structural
optimization. Comput. Methods Appl. Mech. Eng. 194(30), 3269–3290 (2005)
208 W. Liu et al.

6. Nigam, N.C.: Structural optimization in random vibration environment. AAIA J. 10(4), 551–
553 (2012)
7. Allaire, G., Jouve, F., Toader, A.M.: Structural optimization using sensitivity analysis and a
level-set method. J. Comput. Phys. 194(1), 363–393 (2004)
8. Choi, K.K., Haug, E.J., Komkov, V.: Design sensitivity analysis of structural systems.
Academic Press Inc, Orlando Fl vol. 19, no. 8, pp. 1087–1088 (1986)
9. Chen, X., Xiaoli, Y.U., Binwei, J.I.: Study of crankshaft strength based on iSIGHT platform
and DOE methods. In: International Conference on Measuring Technology & Mechatronics
Automation (2010)
10. Padula, S.L., et al.: Multidisciplinary Optimization Branch Experience Using iSIGHT
Software. An Overview of An Experimental Demonstration Aerotow Program, vol. 20, no.
2–4, pp. 216–224 (1999)
 A Study of Multi-Objective Optimization
for Propulsion Performance
and Cargo Capacity

Yasuo Ichinose1(&) , Yusuke Tahara1 , Tomoki Takami1,

Azumi Kaneko1, Takayoshi Masui2, and Daisuke Arai3
1
National Maritime Research Institute, Mitaka, Japan
[email protected]
2
NAPA Japan, Kobe, Japan
3
Sanoyas Shipbuilding Corporation, Kurashiki, Japan

Abstract. The purpose of this paper is development of hull form design system
which can optimize both propulsive performance and cargo capacity, utilizing
automatic design tool of engine room and cargo hold arrangement, and high-
fidelity evaluation tool for propulsion performance. The present system is
developed based on 3D CAD system, NAPA, coupling with RaNS equation
based CFD solver, Neptune. In this system, engine room arrangement is auto-
matically designed considering hull form shape and structural members to keep
the clearance for workers and deck space for installation of auxiliary engines.
After deciding the position of the bulkhead from the engine room arrangement,
the cargo hold is also automatically designed to maximize cargo capacity. Hull
form transformation is performed using FFD technique that facilitates main-
taining the smoothness of the deformed hull form. Demonstration of the present
design system with 62,000-deadweight class bulk carrier, shows very promising
results for the holistic design of hull form. Comparing to the convectional
manual design study, the proposed system enables to present dozens of optimal
design results in wide range like as an optimal surface in design range. Fur-
thermore, this optimal surface helps the designer to understand effective limits of
design parameters. As a result, the present system can help designer’s high-level
decision making.

Keywords: Holistic ship design
Hull form
3D CAD
Automatic design

EEDI

1 Introduction

Demands for advanced hull from design is growing continuously. In hull form design,
computational fluid dynamics (CFD) based hull form optimization, which is called
Simulation Based Design (SBD), is widely accepted. In the past decades, SBD has
been widely studied in hull form optimization research for propulsive performance.
Those works were successfully demonstrated SBD ability and effectiveness for hull
form optimization in propulsive performance (for example [1], [2]).

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 209–219, 2021.
https://doi.org/10.1007/978-981-15-4680-8_15
210 Y. Ichinose et al.

Meanwhile, from January 2013, regulations by Energy Efficiency Design Index

(EEDI) began to be implemented. EEDI is one of the indicators to evaluate trans-
portation efficiency. Hence, it is becoming increasingly important that the hull form
design focusing on transportation efficiency. Transport efficiency is an index calculated
from propulsion performance and cargo capacity (deadweight). Maximizing propulsion
performance and maximizing cargo capacity at the same time are generally in a trade-
off relationship. Therefore, we need to apply multi-objective optimization technique for
hull form design aiming at maximizing EEDI.
Those interdisciplinary multi-objective optimizations are called total (holistic) ship
design. The focus of conventional studies is to determine the principal dimension of
ships (for example [3]). These studies use simplified formulas in evaluation of hull
form. For example, in the propulsion performance evaluation, the propulsive perfor-
mance is evaluated using the Holtrop’s method [4] or regression formula of original
database of towing-tank results (for example [5]), which is defined propulsion per-
formance by principal dimension of hull form such as ship length or block coefficient
(CB). In addition, cargo capacity is also evaluated by empirical formula defined by
principal dimension. Although these results are effective for determining the principal
dimension, it is unsuitable for detailed design of hull from. The hull form optimization
that determines the principal dimension is called global concept design and is distin-
guished from local geometry optimization [6]. In recent years, since 3D CAD
(Computer-Aided Design) system has become popular at the design site, local geom-
etry optimization using these CAD data is more important.
In recent years, the HOLISHIP Project [7], which focus on holistic (total opti-
mization) design of ships, has been underway as a European project since 2013. In this
project, the 3D CAD system is highly coordinated with various performance evaluation
tools. We are aiming for the holistic design of the ship, and the focus of our research is
exactly there. In this project, several hull design demonstrations are being conducted,
and permissive results are obtained [8–10]. These demonstrations focus more on global
concept optimization. However, local geometry optimization is also important, such as
stern hull form design considering the clearance between the main engine and structural
member for workers. In hull form optimization aimed at EEDI, the increase in cargo
capacity (deadweight) is important. In bulk carriers and tanker, the shapes of the main
machine room, especially the length in the longitudinal direction, has a particularly
strong influence on the cargo capacity. Therefore, in stern-shaped local geometry
optimization, multi-objective optimization of propulsion performance and Cargo
capability for changes in main machine room shape is important. Nevertheless, little
studies has been done to local geometry optimization with holistic design.
The purpose of this paper is development of hull form design system which can
optimize both propulsive performance and cargo capacity utilizing high fidelity CFD
and 3D CAD model to consider cargo hold and main engine arrangement. The present
system is developed based on 3D CAD system, NAPA [11], coupling with Reynolds-
averaged Navier Stokes (RaNS) equation based CFD solver, Neptune. Utilizing NAPA
system, Engine room arrangement is automatically designed considering space for
pedestrian passage between main web frame and main engine 3D form imported as iges
format in advance. After deciding the position of the bulkhead from the main engine
room arrangement, the cargo hold is also automatically designed to maximize cargo
 A Study of Multi-Objective Optimization for Propulsion 211

capacity. Hull form transformation is performed using Free Form Deformation

(FFD) technique that facilitates maintaining the smoothness of the deformed hull form.
Furthermore, demonstration with 62,000 deadweight class bulk carrier, called NMRI-
N-BC is also performed to discuss the effectiveness of the present system.

2 Overview of the Design System

The present system is characterized by automatically designing engine room arrange-

ment and evaluating their propulsion performance and cargo capacity with high
accuracy. Figure 1 shows a flow chart of present optimization system. Firstly, hull form
is transformed by FFD technique. Based on the deformed hull form, engine room and
engine room bulkhead (BHD) is automatically designed with constraints. Finally,
propulsive performance and cargo capacity are evaluated and optimized.

Fig. 1. A flow chart of present optimization system and calculation time.

One of the advantages of the present system is increase of the total value of the
ship, such as EEDI. In addition, reducing of design time due to reduction of redesign
effort can be also emphasized (Fig. 2). Those advantages of the present system can be
achieved by using high-fidelity evaluation method, comparing to the empirical formula.

2.1 Hull Form Transformation (Geometry Manipulation)

The hull form transformation (geometry manipulation) is taken by Free Form Defor-
mation (FFD) in the present study. Figure 3 shows an example of a bulbus bow
transform by FFD. In the FFD, a box is defined in the deformation space, and then
some vertices of the box are moved to realize desired transformation of the hull form.
212 Y. Ichinose et al.

Fig. 2. One of the advantages of total optimization design.

Along with this deformation of the box, the inner hull form is also deformed. The
feature of FFD is its high degree of freedom of deformation and smoothness of the
deformed hull surface. By proper definitions of the boxes, it is possible to achieve
various transformations such as deformation of frame line, cross sectional area curve,
position of the buoyancy and so on.

Fig. 3. An FFD example for bulbus bow transformation.

2.2 Automatic Main Engine Room Arrangements

The present system can design the main machine room with some constraints automat-
ically. Figure 4 shows an overview of automatic main engine room design. The position
of the main engine is optimized considering hull form shape and structural members to
keep the clearance for workers and installation space for auxiliary engines and pipes. The
main engine shape is handled as 3D CAD data, and basic shapes of web frames which is
structural members can be defined by designer. The 2nd and 3rd deck’s area is automat-
ically kept in order to ensure the installation space for auxiliary engines and pipes. To
achieve those automatic design, iteration of design is necessary in this step.
 A Study of Multi-Objective Optimization for Propulsion 213

Setup of web frame in Distance between main engine and engine room bulkhead
engine room

Clearance between main

engine and web frame
3D main engine data was imported as iges data

Fig. 4. Overview of automatic main engine room design.

2.3 CFD Method and Propulsive Performance Evaluation

Propulsion performance of deformed hull form is evaluated by RaNS equation based
CFD solver Neptune developed at NMRI [12], for which validation and verification
with towing tank test and full-scale measurements [13] has been carried out for many
years. Calculation conditions in the present study is design ship speed in full load
condition at model scale Reynolds number (Re = 5.0  106). The turbulence model
used in the present work is modified Spalart-Allmaras (MSA) one-equation model [14]
without wall function. An arbitrary constant parameter, Cvor in MSA set as 10 in all
cases of the present work. The presence of rotating propeller is represented by sim-
plified body-force model based on an infinitely bladed-propeller theory [15]. The
calculation grids of basic hull forms without any appendages at full-load and even-keel
were generated with HO topology, 1.8 million cells (i  j  k ¼ 169  129  81: both
sides), 1.5 million cells (i  j  k ¼ 169  113  81: both sides) for free surface and
double model calculation cases, respectively. A double model assumption is applied
only to self-propulsion simulations. The empirical correlations, analyzed from towing
tank results, are adopted to CFD result to ensure powering results. Evaluation of
propulsion performance in full-scale is based on the normal powering method from the
towing tank test results in NMRI.

2.4 Automatic Cargo Holds Design and Capacity Evaluation

The objective of the cargo holds design is maximize the cargo capacity. The present
system automatically design cargo hold 3D shapes considering hull form and based on
bulk head position defined above mentioned steps. By those 3-dimensional design
approach, the accurate evaluation of cargo capacity is achieved in this system.
214 Y. Ichinose et al.

3 Demonstration Case

The target ship for demonstration of the present system is 62,000 deadweight class bulk
carrier, called NMRI-N-BC. The ship has principal dimensions on Table 1, where Lpp:
ship length, B: ship breadth, D: ship depth, d: design draft and Vs: design speed. The
objectives are propulsion performance and cargo capacity.

Table 1. Principal dimensions of the target ship.

62,000 DW
NMRI-N-BC
Lpp 180.0 [m]
B 32.0 [m]
D 18.0 [m]
d 12.0 [m]
Vs 14.0 [knot]

Fig. 5. FFD in LCB (top: LCB to aft, BF −2.0; bottom: LCT to fore, BF 3.0)
 A Study of Multi-Objective Optimization for Propulsion 215

Fig. 6. FFD in AFT CP curve (top: aft heavy, BF −0.5; bottom: midship heavy, BF 2.0).

3.1 Parametric Hull Form Deformation

The hull form deformation is parametrized by Blending Factor (BF) of FFD. In the present
demonstration, the original hull form is deformed by following parametrized FFD.
1. Sectional area curve (CP curve)
a. LCB: position of longitudinal center of buoyancy, Fig. 5.
b. AFT CP curve: CP curve trend in aft-part (Whether the weight is placed on the
midship side or on the aft side), Fig. 6.
2. Sectional frame line shape (generally called U-type and V-type), Fig. 7.
Constrains of the present demonstrational optimization is displacement, clearance
between main engine and web frame and area of 2nd and 3rd deck.

3.2 Result and Discussion

Response surfaces of the objectives of cargo capacity and propulsive performance
(Delivered horse power at design speed: DHP) are shown in Fig. 8 and Fig. 9, respec-
tively. The represented design parameters are LCB and AFT CP curve. From the results, to
increase hold volume of the target ship, the LCB is placed to more aft side, and the trends
of CP curve is weight more on stern side. This is because the main engine can be arranged
more aft side by placing the displacement on the stern side. Thus, by moving the engine
bulk head to stern side, the cargo space increased. On the other hand, to improve the
propulsion performance, the LCB must be arranged to the forward, which is opposite to
the direction mentioned above. Hence, there is a trade-off relationship between propulsive
performance and hold volume in this demonstration case of design.
Multi-objective optimization results are shown in Fig. 10. A Pareto solution
maximizing propulsion performance and cargo capacity is found with the present
system. The multi-objective optimization in the present demonstration has succeeded
because a hull form better than original is found in terms of the cargo capacity and
propulsive performance.
216 Y. Ichinose et al.

Fig. 7. FFD in sectional frame line shape (left: U-type, BF 2.0; right: V-type, BF −2.0).

Fig. 8. Response surface of cargo capacity.

In this research, we discuss the ship type of red and blue ▽ hull form among Pareto
solutions in Fig. 10. Figure 11 shows the lines of red and blue hull forms compared
with the original hull form. The blue hull form is superior in propulsive performance.
And this blue hull form has displacement on the bow side, and V-type frame lines. On
the other hand, the red hull form has both better propulsive performance and larger hold
volume. This hull form has U-shaped frame lines.in this respect, the present system
enables multi-objective optimization of hull form interdisciplinary.
 A Study of Multi-Objective Optimization for Propulsion 217

Fig. 9. Response surface of propulsive performance.

In addition, in Fig. 10, we found interesting findings that could not be obtained by
conventional methods. That means that the practical upper design value of hold volume
of the target ship type is about 625,000 m3. As shown in Fig. 11, when trying to obtain
the hold volume at 625,000 m3 or more, DHP becomes unacceptably large. Comparing
to conventional and manual design approach, the present method has the advantage that
many studies enables designer to obtain new findings or knowledge.

Fig. 10. Multi-objective optimization results

218 Y. Ichinose et al.

Fig. 11. Comparison of hull forms on the pareto front comparing to original hull form.

4 Concluding Remarks

In this paper, a hull form design system which can optimize both propulsion perfor-
mance and cargo capacity has been developed. The system utilizes high fidelity CFD
and 3D CAD model, considering cargo hold and main engine placement. In hull shape
optimization aimed at EEDI, it is important to increase the cargo capacity (deadweight).
For bulk carriers and tankers, the shape of the main machine room, especially the
length in the longitudinal direction, has a particularly strong influence on the load
capacity. Therefore, in the optimization of the local shape of the stern shape, the multi-
objective optimization of the propulsive performance and the cargo capacity to the
change of the shape of the main machine room are important.
Demonstration of this design system has been carried out in this paper. The target
ship is a 62,000-deadweight class bulk carrier called NMRI-N-BC. Multi-objective
optimization of propulsion performance and load capacity has been performed by
deforming hull forms based on rear frame lines and sectional area curve. The present
system can provide a Pareto solution, maximizing propulsion performance and cargo
capacity. Compared to conventional manual case study, the proposed system can
present dozens of optimized design results in a wide range like the optimum design
surface. In addition, this new design knowledge can be obtained from the results, that is
the upper limit value of the cargo capacity of the target class ship. From these dis-
cussions, the effectiveness of the proposed multi-objective ship design system has been
confirmed.

Acknowledgments. This work was partially supported by JSPS KAKENHI Grant Number
17K06975. The authors would like to express their appreciation to those who concern for the
support and encouragement.
 A Study of Multi-Objective Optimization for Propulsion 219

References
1. Campana, E., Peri, D., Tahara, Y., Stern, F.: Shape optimization in ship hydrodynamics
using computational fluid dynamics. Comput. Methods Appl. Mech. Eng. 196, 634–651
(2006)
2. Tahara, Y., Peri, D., Campana, E.F., Stern, F.: Single- and multi-objective design
optimization of a fast multihull ship: numerical and experimental results. J. Mar. Sci.
Technol. 16(4), 412–433 (2011)
3. Nowacki, H., Brusis, F., Swift, P.M.: Tanker preliminary design—an optimization problem
with constraints. Trans. SNAME, 78 (1970)
4. Holtrop, J., Mennen, G.G.J.: An approximate power prediction method. International
Shipbuilding Progress (1982)
5. Ichinose, Y., Kume, K.: A Program named “HOPE Light” for optimizing hull-form
dimensions. Papers of National Maritime Research Institute, vol. 15, no. 4, pp. 407–419
(2015). https://www.nmri.go.jp/en/_src/26755/PNM23150403-00.pdf. Accessed 6 Mar 2019
6. Tahara, Y., Ichinose, Y., Takami, T., Kaneko, A.: Simulation based global concept and local
geometry optimization for ship design considering propulsive performance in actual seas and
fatigue damage evaluation by using global oceanographic model and onboard monitored
data. In: 32nd Symposium on Naval Hydrodynamics (2018)
7. EU HOISHIP Project. http://www.holiship.eu. Accessed 6 Mar 2019
8. Papanikolaou, A.: Holistic ship design optimisation. In: Papanikolaou, A. (ed.) A Holistic
Approach to Ship Design, pp. 9–42. Springer, Cham (2019)
9. Marzi, J., Papanikolaou, A., Brunswig, J., Corrignan, P., Lecointre, L., Aubert, A.,
Zaraphonitis, G., Harries, S.: HOLISTIC ship design optimization. In: Kujala, P., Lu, L.
(eds.) IMDC 2018, Marine Design XIII. Taylor and Francis group, Routledge (2018)
10. De Jongh, M., Olsen, K.E., Berg, B., Jansen, J.E., Torben, S., Abt, C., Dimopoulos, G.,
Zymaris, A., Hassani, V.: High-level demonstration of holistic design and optimization
process of offshore support vessel. In: Kujala, P., Lu, L. (eds.) IMDC 2018,Marine
Design XIII. Taylor and Francis group, Routledge (2018)
11. NAPA. https://www.napa.fi. Accessed 6 Mar 2019
12. Hirata, N., Hino, T.: An efficient algorithm for simulating free-surface turbulent flows
around an advancing ship. J. Soc. Naval Architect. Japan 185, 1–8 (2000)
13. Hirata, N., Hino, T.: Numerical computations of ship flows in full-scale. In: Computational
Methods in marine Engineering, Barcelona, Spain (2007)
14. Hirata, N., Hino, T.: A comparative study of zero- and one-equation turbulence models for
ship flows. J. Kansai Soc. Naval Architects 234, 17–24 (2000)
15. Ohashi, K., Hirata, N., Hino, T.: A comparative study of body force models representing
effects of contrarotating propellers. Trans. West-Japan Soc. Naval Archit. 105, 55–64 (2003)
 Hull Form Optimization to Fulfil Minimum
Propulsion Power by Using Frequency
and Time Domain Potential Flow Solvers

Jaekyung Heo1(&), Daehwan Park2, Jan Henrik Berg-Jensen3,

Zhiyuan Pan3, and Torgeir Kirkhorn Vada3
1
Digital Solutions, DNV GL, Busan, Korea
[email protected]
2
Friendship Systems, Potsdam, Germany
3
Digital Solutions, DNV GL, Høvik, Norway

Abstract. This paper presents hull form optimization processes which can be
practically applied at hull form design stage. Traditionally hull form develop-
ment is carried out in calm sea condition to reach target design speed with
specified engine power. However, the design condition has been changing since
the evolution of EEDI. It is required to satisfy not only the design speed in calm
sea, but also minimum propulsion power in wave condition for ships complying
with regulations on energy efficiency of IMO.
Hull form optimization processes with single and multi-objective functions
are described. The single-objective optimization seeks for optimized hull forms
in waves from a hull form initially optimized in calm sea condition. The multi-
objective optimization evolves hull forms by improving resistance performance
in both waves and calm water. For the optimization processes, a practical
approach using two potential flow solvers, Wasim and Wadam, which can
handle forward speed problems, is presented.
The approach consists of two stages: The first stage is to optimize and find
best candidates by running Wadam inside CAESES, a parametric hull form
optimization program. Added resistance at a required advance speed is com-
puted and evaluated for several sea states specified by the simplified assessment
of IMO MEPC.1/Circ.850. KVLCC2 hull form is modified and evaluated in the
first optimization stage. The second stage is a verification stage using Wasim.
Wasim has no limit in forward speed and is more accurate than Wadam in
computing added resistance, while it requires more computational time. Based
on the verification by Wasim, the wave added resistance of the optimal hull form
is confirmed.
The proposed approach employing the two solvers will help hull form
designers save design time and investigate many more alternatives. It will also
provide practical design information by showing a direction of how hull forms
are varied to minimize the added resistance in waves.

Keywords: Minimum propulsion power
Potential flow solver
Hull form

Optimization

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 220–236, 2021.
https://doi.org/10.1007/978-981-15-4680-8_16
 Hull Form Optimization to Fulfil Minimum Propulsion Power 221

1 Introduction

Hull form has been traditionally developed and optimized to minimize resistance in
calm sea condition. Experience of engineers was the main resource to draw out hull
forms to satisfy the design speed with given engine power. Emergence of computer
aided engineering opened a new era of hull form design, enabling engineers to test
candidates prior to model tests. Potential flow solvers based on Dawson’s approach [1]
drew attention and has been regarded as a common design tool in the industry. Due to
the characteristics of potential flows, those programs are employed for developing fore
body such as bulbous bow by examining bow waves. CFD based on RANSE has been
increasingly utilized as performance of computers improves. It becomes common to
solve whole viscous flow fields including free surface to obtain total resistance and
further to carry out self-propulsion tests with a propeller. In addition to the progress of
the numerical methods and hardware, parametric hull form modelling which eventually
aims for optimizing hull forms, has drawn attention from the shipbuilding industry. The
parametric modelling with optimization techniques improves efficiency of iterative
design process by coupling numerical solvers with optimization process [2]. It has been
applied not only for hull forms but also for propellers and energy saving devices.
On top of the classical design target for achieving design speed in calm sea,
additional requirement has been added by environment regulations. EEDI is one of
such regulations in shipping industry, requiring a minimum energy efficiency level for
new ships to reduce greenhouse gas. Several options are available to comply with
EEDI, and reducing speed can be an attractive measure. Lower speed with less engine
power, on the contrary, can endanger ships in harsh weather conditions. IMO has set
requirements for tankers, bulk carriers and combination carriers which comply with
EEDI requirements to install engines of which propulsion power is sufficient to
maintain the maneuverability in adverse conditions. The simplified assessment pro-
cedure introduced in the IMO guideline needs calculation of total resistance including
added resistance in waves to predict minimum propulsion power [3].
The additional requirement on the minimum propulsion power led hull form
designers, who have been acquainted with the calm sea condition, to wave conditions.
They now pay more attention than before on added resistance in waves at design stage
and make experimental and numerical attempts to estimate powering performance in
waves to comply with the requirement. This has potential effects on the hull form
design procedure based on the typical design spiral, which now should account for the
minimum propulsion power that needs to be investigated inside the design spiral.
The present study introduces the use of two 3-D panel method programs, Wasim
and Wadam, for parametric hull form optimization with aim of how iterative opti-
mization processes for wave added resistance can become effective by using the sol-
vers. Optimization is carried out with a parametric modelling program, CAESES, for
KVLCC2 as baseline hull form. Accuracy of the solvers is compared in terms of
motions and added resistance in waves. Discussion on optimization processes is
addressed in Sect. 4. Optimized hull forms from the parametric optimization are
selected in Sect. 5, and conclusions are drawn.
222 J. Heo et al.

2 KVLCC2 Hull Form

Parametric optimization process is applied to the hull form of KVLCC2 with

assumption that it has been originally optimized at the design speed in calm sea
condition. KVLCC2 has been selected for many benchmarking studies, and the authors
used information provided by SIMMAN 2014 workshop [4]. KVLCC2 has length of
320.0 m, breadth of 58.0 m, draft of 20.8 m, and Cb of 0.8098.
In addition to its design speed of 15.5 knots, advance speed should be defined to
estimate minimum propulsion power. The required ship advance speed is to be defined
with various design data, e.g. rudder area, windage area, etc., according to the interim
guidelines [3]. Due to lack of design information, the required speed of KVLCC2 is
assumed to be 4 knots.

3 Potential Flow Solvers

3.1 Wasim-Time Domain Solver

Wasim is a 3-D panel method program, distributing Rankine sources on free surface as
well as body, and solves equations of motion in time domain. Wasim is applicable for
ships of arbitrary shapes with forward speed and has been used for linear and nonlinear
motion and wave load analyses.
Pan et al. [5] implemented a method to evaluate wave added resistance by
employing a control surface close to body (see Fig. 1). Added resistance is obtained
from the integral over the control surface by applying the momentum conservation.
They showed several comparisons in good agreement with results of model tests and
other computations.
DNV GL recently had a joint project with DSME to estimate minimum propulsion
power of real ships and compared thoroughly with model tests. It was observed that
Wasim can exactly capture distinguishing results affected by characteristics of each
different hull form. One of the promising results is that Wasim computed the closest
results to the experimental results in added resistance as well as heave and pitch
motions [6].

Fig. 1. Distribution of panels on KVLCC2, free surface and control surface

 Hull Form Optimization to Fulfil Minimum Propulsion Power 223

3.2 Wadam-Frequency Domain Solver

A typical linear frequency domain solver, Wadam, which was originally limited for
zero speed problems, is used. Pan et al. [7] slightly increased the limitation of forward
speed by exploiting the wave drift damping technique [8]. The forward speed technique
implemented in Wadam is based on a perturbation method in which wave amplitude
and forward speed are taken as a small parameter in the approximate solution. Pan et al.
[7] presented results of benchmark tests for a ship and a semi-submersible and showed
a validity of the range of forward speed of Wadam.
Wadam, like other similar programs, has limitation in use when s ¼ xe U=g
approaches 0.25, where xe , U, g are encounter frequency, forward speed, gravitational
acceleration, respectively. As the nondimensional frequency s is dependent on the
group velocity of the radiating waves and the forward speed of the body, Wadam
should be used up to a certain range where the group velocity of the radiation waves is
higher, i.e. s \ 0:25, so that the waves can be radiated away.

3.3 Comparative Study of Wasim and Wadam

To properly set up parametric optimization process, sets of analyses were performed to
compare results by the two solvers. The same hull mesh of 1680 panels was used for
the both programs. Wasim required additional 6,650 and 600 panels for free surface
and control surface. Wadam also needed a free surface mesh to compute wave drift
damping forces and 1,100 panels were distributed on the free surface.
The time domain solution started from a calm sea analysis to find a steady condition
where basis and local potentials should be computed. Double body model was selected
for the computations of the basis potentials. To obtain wave added resistance transfer
function, regular wave analyses for each wave period were executed. Time step was set
to 0.01 s, and the time marching lasted until 6 cycles of harmonic results were stored
for FFT analysis.
Figure 2 compares quadratic transfer functions of added resistance in waves. The
abscissa (L0 ) represents wave lengths divided by the length of KVLCC2, and the
ordinate (R0 ) is added resistance nondimensionalized by dividing with qgB2 n2 =L,
where q is the density of water, B is the breadth and L is the length of the vessel and n
is the wave amplitude. At zero speed, the two solvers gave exactly the identical results.
The results start to deviate little by little as the forward speed increases. At the required
ship advance speed, V = 4 knots, Wadam shows larger values in short waves, and
smaller values in long waves.
In motion transfer functions shown in Fig. 3 and Fig. 4, discrepancy is observed at
higher speeds over wave lengths shorter than the length of the vessel.
According to the interim guidance, wave added resistance was calculated for
JONSWAP wave spectrums with significant wave height of 5.5 m, and peak periods
from 7.0 to 15.0 s (see Fig. 5). The trend of the transfer functions of added resistance is
reflected on the wave added resistance. The two curves show similar pattern, and the
highest peaks are located at the same sea state.
224 J. Heo et al.

Fig. 2. Wave added resistance RAOs

Fig. 3. Heave RAOs

 Hull Form Optimization to Fulfil Minimum Propulsion Power 225

Fig. 4. Pitch RAOs

Fig. 5. Wave added resistance in sea states defined by JONSWAP wave spectrums

Overall, the results of Wasim is more accurate than those of Wadam at large
forward speeds, which was found in our previous study [7] as well. However, the
results of Wadam cannot be regarded unreasonable, since the deviations are not sig-
nificant and, above all, Wadam produced similar trends as those of Wasim.
From a usage perspective, the solvers took different time in preparation and
computation. Wadam needed about 12 min for 19 wave frequencies, while Wasim took
more than 3.5 hours per each wave frequency on a laptop with i7-6820HQ CPU. More
preparation time should be consumed for Wasim to generate fine free surface mesh, and
to obtain the steady solution.
226 J. Heo et al.

4 Parametric Hull Form Optimization Process

Parametric optimization process for hull form design consists of definition of design
parameters, coupling of solvers and evaluation. All these setting is made inside
CAESES program that have been used for hull form optimization with CFD programs.
In hull form design, a serial process with a single solver for a single-objective
function is commonly applied. For example, it modifies a hull form with design
parameters in specified ranges, solves for the modified hull form and returns results,
then the optimizer evaluates and makes judgement for the next variation. The process is
iterated until the maximum number of iterations is reached. Minimizing resistance in
calm sea condition with CFD or equivalent tools can be a typical job of this kind that is
frequently carried out by hull form designers. In case of two or more objective func-
tions, e.g. resistance and seakeeping performances, more relevant results should be
evaluated in the optimization cycle [9, 10]. With multiple solvers, the multi-objective
optimization works in parallel and collects results and evaluates them for decision
making.
Minimum propulsion power as an objective function can be treated by either serial
or parallel process. In the serial process, an optimized hull form in calm sea condition
can be a proper starting point (see the left of Fig. 6). Hull form variations and added
resistance computations are iterated for many candidates generated by CAESES. It
should be mentioned that the serial process including only the added resistance com-
putation might deteriorate the performance in calm sea condition. Nonetheless, it can
give a straightforward indication to reduce added resistance, that will be useful
information for hull form designers.

Fig. 6. Parametric optimization processes: single objective (left) and multi objectives (right)
 Hull Form Optimization to Fulfil Minimum Propulsion Power 227

Another solver for calm water resistance can be included as shown on the right of
Fig. 6. Two solvers compute resistance in waves and calm sea conditions, respectively,
and the results are sent to the optimizer of CAESES to evaluate and develop hull forms.
The parallel process can start from scratch or an optimized hull form in calm sea
condition. The parallel process looks more ideal in seeking for a solution taking the
multi-objective functions into consideration, which can end up with optimized hull
forms satisfying the both requirements.
As a main solver to collaborate inside the optimization loop for minimum
propulsion power, Wadam can be a proper choice from a perspective of computational
efficiency and accuracy. Although Wadam provides less accurate results than Wasim,
results of Wadam are always stable and consistent. The deviation from those of Wasim
is small for conditions with small s. These are important factors to be selected as
numerical solvers for optimization. On top of that, CAESES can automatically generate
mesh for Wadam, which makes the process seamless. The less computational time of
Wadam will make the iterative optimization task fast and it can be an overnight task.
However, despite the merits of Wadam, Wasim, the more accurate solver, should not be
underestimated nor excluded from the optimization process. Wasim needs to be exe-
cuted outside the loop for optimized hull forms to verify the results of Wadam.
The results of Wadam, quadratic transfer functions of added resistance, are auto-
matically forwarded to a statistical postprocessing program, Postresp. It computes added
resistance for the specific sea states and returns results to the optimizer of CAESES.
For the optimization, Dakota optimization engine embedded in CAESES is used.
Response surface method is selected for minimum value search mechanism. Firstly,
initial computations are performed with the selected design parameters. Dakota dis-
tributes the design parameters in the design space as evenly as possible in order to avoid
computing unnecessarily closed design points. In the second step, a few more compu-
tations are performed on the response surface to find out the optimum more precisely.

5 Results and Discussion

5.1 Variation of Bow Shapes

It is known that wave added resistance consists mainly of two components, diffraction
in the bow area and radiation by ship motions. Various bow shapes to control the
reflected waves from the bow have been introduced [11], and bulbless bow forms, e.g.
LEADGE bow, have been designed in the industry.
Two sets of design parameters were defined to study two kinds of fore body forms,
bulbless bow and LEADGE bow (see Fig. 7). The bulbless bow has no trace of the
bulbous bow of the original hull form. In the bulbless bow, the entrance fullness
controls the radius at the stem line, and the shoulder fullness compensates the changed
volume to keep the original volume the same as much as possible. The bulbless bow
might have more blunt shape than the LEADGE bow. The entrance fullness defined for
the LEADGE bow can be regarded as entrance angle. Figure 8 shows smooth transition
of LEADGE bows generated by CAESES. The changes of volume and wet surface area
were found small enough to be ignored.
228 J. Heo et al.

Fig. 7. Design parameters for bulbless bow (top) and LEADGE bow (bottom)

Fig. 8. LEADGE bows varied from the baseline hull form

The hull form variation was applied up to the station 18, and 24 hull forms were
generated for each kind of the bows by the Dakota method. Figure 9 shows the effects
of the design parameters on wave added resistance. The reduction of wave added
resistance is proportional mainly to the increase of the bow length, however the other
parameters do not show remarkable correlations with wave added resistance. The
maximum reduction reached about 20% in the both bow shapes. The best bulbless bow
needed about 4 m longer bow than the original hull form, and the LEADGE bow was
lengthened more by about 8 m. It is found that all the bulbless bows reduced the wave
added resistance, while the LEADGE bows needed to be lengthened at least by 2 m to
reduce it. It can be explained by the used variation techniques. The LEADGE bows
were varied through the transition as shown in Fig. 8. However, the bulbless bow
removed the original bulbous bow without transition for its first variation. The bulbless
bows filled up volume in the bow area depending on the entrance fullness, and com-
pensated volumes to the shoulder area accordingly.
 Hull Form Optimization to Fulfil Minimum Propulsion Power 229

Fig. 9. Variation of wave added resistance with design parameters of bulbless bows (left) and
LEADGE bows (right)

Fig. 10. Bulbless and LEADGE bows

230 J. Heo et al.

The best bow forms of the two types generated in the comparative study are shown
in Fig. 10. The bulbless bow has vertical stem line and very sharp entrance like typical
LEADGE bows. The LEADGE bow also has sharp entrance and protruding part of the
original bulbous bow. Their added resistance RAOs are compared in Fig. 11. They two
are very similar each other, and show large decrease in short waves compared to the
baseline hull form.

Fig. 11. Wave added resistance RAOs of the best bulbless and LEADGE bows

5.2 Bow Optimization with Single-Objective Function

Further optimization for the LEADGE bow was carried out using the Dakota optimizer.
Initially 9 bows were generated to create response surface, and 4 additional iterations
were made to search for the minimum value from the response surface.
Figure 12 shows wave added resistance varied with the design parameters. The
trend lines were drawn based on the initial 9 hull forms marked with the blue circles.
Similar to the results in the previous section, the length variation shows the strongest
correlation with the wave added resistance. The four red circles represent the final hull
forms generated based on the information of the response surface. It is obviously
shown that the optimizer located the final hull forms in the right direction.

Fig. 12. Variation of wave added resistance

 Hull Form Optimization to Fulfil Minimum Propulsion Power 231

Figure 13 shows the wave added resistance RAOs of the 13 hull forms computed by
Wadam. All the developed LEADGE bows show smaller RAOs than the original bul-
bous bow in short waves. In long waves, however, some of the LEADGE bows
increased the RAOs. The results agree well with that the LEADGE bows are superior in
short waves governed by wave reflection. It is also interesting to note that the RAOs of
13 bows are spread quite much as depicted by the envelope. The variation in short waves
is almost twice the minimum wave added resistance. The large envelope of the wave
added resistance RAOs means that wave added resistance depends heavily on bow
shapes, in other words, it can be controlled by design of bow shapes. Also, numerical
tools should be able to take account of bow shapes in computing wave added resistance.

Fig. 13. Envelope of wave added resistance RAOs

The finally optimized bow is shown in Fig. 14. The reduction of the wave added
resistance is about 25% of that of the baseline hull form by increasing 8 m in length.
The wave added resistance RAOs computed by Wadam inside the optimization loop
should be confirmed by additional computation by Wasim especially for the final hull
form. The wave added resistance by Wadam and Wasim are compared in Fig. 14. The
trend appeared in the Wadam results for the baseline and optimized hull forms are
similarly found in the Wasim results as well.

Fig. 14. Optimized bow shape and wave added resistance RAOs by Wasim and Wadam
232 J. Heo et al.

Fig. 15. Wave added resistance in irregular waves computed by Wadam and Wasim (lines) and
reduction ratios (bars)

The resulting wave added resistance in the specified sea states is shown in Fig. 15.
The optimized hull form shows improvement from the baseline hull form in the both
computations by Wasim and Wadam. The largest resistance of the baseline and opti-
mized hull forms occurred at the same peak period of 13 seconds in the Wasim results.
The most noticeable finding is that the reduction ratios of the added resistance by the
two solvers are agreed very well especially near the peak period of the largest wave
added resistance.

5.3 Bow Optimization with Multi-objective Functions

Resistance in calm sea condition as well as wave added resistance were specified as
objective functions in the optimization process as shown in Fig. 6. The multi-objective
optimization started from the original KVLCC2 hull form using the Dakota optimizer,
and 78 different hull forms were generated.
Calm sea resistance at the design speed was computed by Shipflow. As only bow
parts were modified, steady potential flows were computed for wave making resistance
and boundary layer theory was used to estimate friction resistance. Total resistance was
evaluated in the optimization process.
Figure 16 distributes all the created hull forms on the plane of the design param-
eters. The added resistance (Raw) was decreased up to 20%, and the reduction of total
resistance in calm sea (RT) reached about 3%. The variation in the wave added resis-
tance is larger than that in the calm sea resistance, i.e. the bow shapes affected more on
the wave added resistance. The conflicting characteristics of the two objective functions
can be clearly identified by the Pareto front marked in red and green symbols.
 Hull Form Optimization to Fulfil Minimum Propulsion Power 233

Fig. 16. Distribution of design parameters (top) and wave added resistance and calm sea
resistance (bottom)

A final hull form can be selected from the Pareto front depending on the required
amount of the reduction in wave added resistance on top of the allowable margin in
calm sea resistance. In this study, a hull form marked by the green symbol in Fig. 16
was selected as the optimal hull form. The shape of the selected hull form is shown in
Fig. 17. The LWL of the optimized hull form was increased by 2.7 m, and the bulbous
bow remained.

Fig. 17. Bow shape of the optimized hull form (left) and comparison of body plans (right):
baseline hull form in black lines and optimized hull form in red lines
234 J. Heo et al.

Fig. 18. Wave added resistance RAOs of the optimized hull form by Wasim and Wadam

Wasim analysis was executed for the optimized hull form, and RAOs of wave
added resistance are compared in Fig. 18. Although the maximum peaks are found at
the different wave lengths between the Wadam and Wasim results, the trends are agreed
very closely. The reduction amount is almost similar in short waves and at the peak
periods. It is interesting that the significant improvement found in Fig. 14 from the
single-object optimization was not reproduced, most probably due to the existence of
the bulbous bow.
Figure 19 compares wave added resistance for the given sea states. Both Wasim
and Wadam show very good agreement in tendency of the wave added resistance.
Moreover, the reduction ratio of the two programs are very close each other, especially
over the peak periods where the maximum wave added resistance occurred.
The optimization process can be practically utilized in hull form design process to
reduce wave added resistance. Wadam inside the optimization loop makes the process
practical and Wasim outside the loop confirms the Wadam results.

Fig. 19. Wave added resistance in irregular waves computed by Wadam and Wasim (lines) and
reduction ratios (bars)
 Hull Form Optimization to Fulfil Minimum Propulsion Power 235

6 Conclusions

The parametric hull form optimizations were performed for KVLCC2 to reduce wave
added resistance from the perspective of minimum propulsion power. Comparisons of
the two potential flow solvers, Wasim and Wadam, were made at low forward speeds.
The solvers quantitatively revealed small discrepancy in wave added resistance, on the
other hand, showed very similar trends which is necessary to be included in the
optimization processes.
The single and multi-objective optimization processes for the wave added resis-
tance were set up with Wadam as the main solver inside the optimization controlled by
CAESES. Wasim analysis for the optimized hull form followed outside the opti-
mization system to confirm the Wadam results and convert them into actual design
values.
It was found that the reduction ratios of wave added resistance between the two
solvers are almost identical. It convinces that the use of Wadam in the optimization
loop can efficiently provide optimized hull forms that can reduce wave added resis-
tance. Additional confirmation by Wasim outside the optimization cycle must be car-
ried out.
The multi-objective optimization process recognized the Pareto front and generated
optimized hull forms in waves and calm sea conditions. This optimization process can
be practically utilized in hull form design stage to fulfill minimum propulsion power.
In addition, CAESES showed that it can generate the two bow types, bulbless and
LEADGE, with the defined design parameters. The LEADGE bows showed improved
results in short waves than the baseline bulbous bow hull form. It was found that the
lengthening effect was dominant in reducing wave added resistance.

References
1. Dawson, C.W.: A practical computer method for solving ship-wave problems. In: 2nd
International Conference on Numerical Ship Hydrodynamics (1977)
2. Abt, C., Harries, S.: A new approach to integration of CAD and CFD for naval architects. In:
6th International Conference on Computer Applications and Information Technology in the
Maritime Industries (COMPIT 2007), Cortona, Italy (2007)
3. IMO MEPC: 1/Circ.850/Rev.2.: 2013 Interim guidelines for determining minimum
propulsion power to maintain the manoeuvrability of ships in adverse conditions (2017)
4. SIMMAN 2014 workshop: https://simman2014.dk/. Accessed 21 Mar 2019
5. Pan, Z., Vada, T., Han, K.: Computation of wave added resistance by control surface
integration, OMAE 2016-54353. In: Proceedings OMAE 2016, Busan, Korea (2016)
6. Jung, D., Choi, J., Lee, Y., Kim, Ch., Kim, S., Heo, J.: Investigation of Wave added
resistance for minimum propulsion power of an Aframax-class tanker with numerical
simulations and model tests. In: Proceedings ISOPE 2019, Honolulu, USA (2019)
7. Pan, Z., Vada, T. Finne, S., Nestegård, A., Hoff, J.R., Hermundstad, E.M., Stansberg, C.T.:
Benchmark study of numerical approaches for wave-current interaction problem of offshore
floaters, OMAE 2016-54411. In: Proceedings OMAE 2016, Busan, Korea (2016)
8. Finne, S., Grue, J., Nestegard, A.: Prediction of complete second-order wave drift damping
force for offshore structures. In: Proceedings ISOPE 2000, Seattle, USA (2000)
236 J. Heo et al.

9. Papanikolaou, A., Harries, S., Wilken, M., Zaraphonitis, G.: Integrated design and
multiobjective optimization approach to ship design. In: Proceedings 15th International
Conference on Computer Applications in Shipbuilding, Trieste (2011)
10. Yu, J., Lee, Ch., Lee, I., Choi, J.: Bow hull-form optimization in waves of a 66,000 DWT
bulk carrier. Int. J. Naval Archit. Ocean Eng. 9, 499–508 (2017)
11. Hirota, K., Matsumoto, K., Takagishi, K., Yamasaki, K., Orihara, H., Yoshida, H.:
Development of bow shape to reduce the added resistance due to waves and verification of
full scale measurement. In: Proceedings of the First International Conference on Marine
Research and Transportation, Ischia, Italy (2005)
 Controller Design of a Gantry Crane
for the Safe Erection of Blocks in Shipyards

Hye-Won Lee1 , Myung-Il Roh1(&) , Seung-Ho Ham2 ,

and Do-Hyun Chun1
1
Seoul National University, Seoul, Republic of Korea
[email protected]
2
Changwon National University, Changwon, Republic of Korea

Abstract. In shipyards, ships and offshore structures are constructed by joining

several blocks on the dock. For this, a gantry crane generally conduct block
erection operation including the lifting, turn-over, and lowering procedure.
During the erection operation, the operator controls the trolleys, crane girder,
and wire ropes to control the motion of the block to the certain trajectory and the
orientation. The gantry crane lifting a block is an underactuated mechanical
system, which has more degrees of freedom than control inputs. Therefore, the
traditional control theories are not applicable to the operations of block erection.
In this study, a method for the controller design of the gantry crane is proposed
for the trajectory tracking control of the block. Then, the simulation based on the
multibody dynamics is performed to analyze the motions of the crane and the
block. The result has shown that the motion of the block is successfully con-
trolled during the operation.

Keywords: Block erection
Simulation
Controller design
Multibody

dynamics

1 Introduction

1.1 Research Background

The ship and offshore structures are generally constructed by composing several
blocks. For this, the block should be lifted, transported, turned-over if required, and
lowered to the other structures using cranes. These processes are called block erection.
During the block erection procedure, the position and orientation of the block should be
precisely controlled to avoid any potential risks such as collision or overweight. In
shipyards, the block erection operations are mostly conducted by using gantry cranes.
The gantry crane is consists of crane girder, upper and lower trolleys, hooks, and wire
ropes. During the operation, the operator controls the translational motion of the
trolleys and crane girder, and the length of the wire ropes connected to the hooks to
control the block. In this case, the block has more degrees of freedom than control
inputs, which makes it an underactuated mechanical system. Therefore, the traditional
control theories are not directly applicable to the gantry crane system.

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 237–242, 2021.
https://doi.org/10.1007/978-981-15-4680-8_17
238 H.-W. Lee et al.

In the actual operation, the control is performed manually by operator. The control
of the block is therefore depends on the decision of the operator or the direction of the
outside workers. In such cases, the block cannot be precisely controlled, and the
operation is conducted very slowly to avoid any risks such as collision. Meanwhile,
physics-based simulation is required before the actual operation to check potential
risks. However, the simulations based on the traditional control methods have some
limitations. To solve these problems, the trajectory tracking control of a block is
presented in this study for block transporting and turn-over operation simulation using
gantry crane.

1.2 Related Works

There are several researches about the control of the gantry crane. Liu and Guo [1]
performed trajectory tracking control of a point mass payload using gantry crane by
sliding mode control method. The gantry crane is considered as a two-dimensional with
three degrees of freedom. Mehra et al. [2] also performed point mass payload control
using four DOF (Degrees Of Freedom) gantry crane. The set-point regulation which
maintains certain position or orientation is carried out using passivity based control.
Bakhtiari-Nejad et al. [3] performed step value tracking of a point mass payload using 2
DOF gantry crane. The optimal feedback control is performed for the tracking to the
target value. In this study, the block transporting and turn-over operation is performed
using 6 DOF gantry crane. For the trajectory tracking control, the feedforward control
using index reduction method is applied.

2 Theoretical Background

2.1 Underactuated Mechanical System

A system with more degrees of freedom than the number of the independent control
inputs is called underactuated mechanical system [4]. In underactuated mechanical
systems, the inverse dynamic control problem is more challenging, and many tradi-
tional non-linear control methods are not directly applicable [5]. The configuration of
the gantry crane is as shown in Fig. 1.
The gantry crane including upper and lower trolleys, girder, three hooks and wire
ropes has 6-DOF. As the block has translational and rotational motions about three
axes, the system has nine total redundant coordinates. In the gantry crane, three wire
ropes connected to the hooks controls the motion of the block. Therefore, three are
three length constraints. To sum up, the gantry crane with a block has nine degrees of
freedom, while it has six control inputs. It implies that the block is kinematically
undetermined, and it is movable while the length of the wire ropes are kept constant.
The block may perform sway motions with three degrees of freedom while lifting [6].
 Controller Design of a Gantry Crane for the Safe Erection 239

Fig. 1. Configuration of a gantry crane

2.2 Motion Planning

A possible task of the underactuated mechanical system can be divided into set-point
regulation and trajectory tracking. The set-point regulation is to control the output
variables to certain value. While set-point regulation is a well-defined control issue, the
trajectory tracking control is a challenge topic [5]. The desired system outputs can be
modeled as servo constraints which can be imposed on the system as additional
constraints.
The required trajectory of the block can be prescribed by start orientation and target
orientation as shown in Fig. 2. Following Rodrigues’ rotation formula, the target
rotation axis er , and angle sre required to rotate to the target orientation can be cal-
culated. It can be obtained as follows (1).
a
Re ¼ cosðsre ÞI3 þ sinðsre Þ~er þ ð1  cosðsre ÞÞer eTr ð1Þ

a
Re is the rotation matrix from initial orientation to target orientation. In the same
way, the translational motion can be described using target position. Finally, the pre-
scribed trajectory that block has to follow can be specified by using servo constraints.

Fig. 2. Motion planning of the block

240 H.-W. Lee et al.

2.3 Dynamic Equations

The wire rope constraints can be derived as the distance between the hook and the
trolley should maintain same with the length of the wire rope. Three constraints can be
obtained using following equations where ci is ith wire rope vector start from the lug
point of the block to the trolley, and pci is the length of ith wire rope.

1 T

U1 ðp; xÞ ¼ ci ðp; xÞci ðp; xÞ  p2ci ð2Þ
2

Then, the dynamic equations of the gantry crane and the block can be obtained
 T 
using crane coordinate p ¼ pg0 ; pg1ð2Þ ; pg3 ; pc1 ; pc2 ; pc3 , control force u ¼ fg0 ; fg1ð2Þ ;
fg3 ; mc1 ; mc2 ; mc3 T and constraint Jacobian Gp and Gs as shown in (2) and (3).

€ ¼ f 1 ðp; p_ Þ þ B1 u  GTp ðp; xÞk

M1 p ð3Þ

M2 €x ¼ f 2  GTs ðp; xÞk ð4Þ

Finally, the equations of motion of the system composed of the dynamic equations
of the crane and the block, constraint equations, and the servo constraint can be
formulated. In underactuated mechanical system, the use of servo constraints leads to
high index of DAEs (Differential Algebraic Equations) which can cause difficulties in
the numerical integration. To solve the final equations, we adopted index reduction by
minimal extension method [7] to reduce the high index value of the DAEs. Then, the
backward Euler discretization is adopted for the numerical stability. The final form of
the equation is then obtained as following. The equations can be solved using Gauss-
Newton method.

pn þ 1  pn  Dtvn þ 1 ¼ 0


M1 ðvn þ 1  vn Þ  Dt Bun þ 1  G1 ðpn þ 1 ; cn þ 1 Þkn þ 1 ¼ 0
ð5Þ
M2€cn þ 1  f 2 þ G2 ðpn þ 1 ; cn þ 1 Þkn þ 1 ¼ 0
Uðpn þ 1 ; cn þ 1 Þ ¼ 0

3 Application

For the application, the block transporting and turn-over operation is modeled as
presented in Fig. 3. The transporting and turn-over operations are conducted simulta-
neously with constant speed as shown in Fig. 3. The upper and lower trolleys and the
wire rope is controlled automatically to rotate the block 90° along x-axis and move 40
m along y-axis.
 Controller Design of a Gantry Crane for the Safe Erection 241

Fig. 3. Block transporting and turn-over simulation

For the comparison, the PID (Proportional-Integral-Differential) controller is

adopted in case 1. The transporting velocity and angular velocity for turn-over oper-
ation is kept constant during the operation. Therefore, the velocity set-point control
using PID control is performed. In case 2, the trajectory tracking control explained in
this study is applied. The result is shown in Fig. 4.

Fig. 4. Result of the block transporting and turn-over control simulation

In case 1, the trajectory of the block is difficult to control due to the simultaneous
movement of the trolleys and the wire ropes. As the block turned-over by the hoisting
of the wire rope, it affects the motion of the lower trolley. In case 2, the trajectory of the
position and the orientation of the block is included in the dynamic equations as servo
constraints, which makes it easier to control.
242 H.-W. Lee et al.

4 Conclusions

In this study, a method for the controller design of the gantry crane is presented to keep
the position and the orientation of the block to certain trajectory during the operation.
The complex equipment called gantry crane is modeled as a multibody system. For the
control of the gantry crane, which is underactuated mechanical system, the servo
constraints are formulated using the initial and target position. Then, the equations of
motion are solved using index reduction by minimal extension method. Finally, the
controller is applied to the block transportation and turn-over operation simulation.
Two cases are compared using PID controller and proposed method. Then, the result
has shown that the motion of the block is successfully controlled to target trajectory
using suggested method.

Acknowledgments. This work was partially supported by (a) Brain Korea 21 Plus Program,
(b) Research Institute of Marine Systems Engineering of Seoul National University, Republic of
Korea.

References
1. Liu, D., Guo, W.: Tracking control for an underactuated two-dimensional overhead crane.
J. Appl. Res. 10, 597–606 (2012)
2. Mehra, R., Satpute, S., Kazi, F., Singh, N.: Geometric-PBC based control of 4-DOF
underactuated overhead crane system. In: 21st International Symposium on Mathematical
Theory of Networks and Systems, Groningen, Netherlands, pp. 1232–1237 (2014)
3. Bakhtiari-Nejad, F., Nazemizadeh, M., Arjmand, H.: Tracking control of an underactuated
gantry crane using an optimal feedback controller. Int. J. Autom. Mech. Eng. 7, 830–839
(2013)
4. Krafes, S., Chalh, Z., Saka, A.: A review on the control of second order underactuated
mechanical systems. Complexity 2018, 1–17 (2018)
5. Liu, Y., Yu, H.: A survey of underactuated mechanical systems. IET Control Theory Appl.
7(7), 921–935 (2013)
6. Heyden, T., Woernle, C.: Dynamics and flatness-based control of a kinematically
undetermined cable suspension manipulator. Multibody Syst. Dyn. 16, 155–177 (2006)
7. Yang, Y.: Numerical Methods for the Inverse Dynamics Simulation Of Underactuated
Mechanical Systems. Kit Scientific Publishing, Karlsruhe (2017)
 A Study on Using Hot-Rolled Steel Sheet
for Ship Superstructure

Hisao Narimatsu1, Masakazu Kuwada1, and Koji Gotoh2(&)

1
Tsuneishi Shipbuilding Co., Ltd,
Numakuma-cho 1083, Fukuyama City, Hiroshima, Japan
[email protected]
2
Kyushu University,
Motooka 744, Nishi-ku 744, Fukuoka City, Fukuoka, Japan
[email protected]

Abstract. Application of the hot-rolled steel sheet for the members of super-
structure, where the strength requirement of these parts in ships is not severe, is
expected as a solution to reduce the hull construction cost. Because the hot-
rolled steel sheet is applied after recoiling by cold working to return it to a flat
plate, the material properties is deteriorated due to excessive plastic working and
the residual stress induced by the plastic work might affect for the thermal
cutting and welding. There is a possibility that the hot-rolled steel sheet has poor
properties comparing with the flat rolled steels applied for hull parts where the
structural integrities are required by the classification societies’ rules. The dif-
ference of properties between the hot-rolled steel sheet and the flat rolled steel
are investigated in this research. Investigated items are mechanical properties,
microstructures, initial residual stress distributions, hardness distributions, dis-
tortion after thermal cutting and laser-arc hybrid welding.

Keywords: Hot-rolled steel sheet (Hot-coiled sheet)
Residual stress

Deflection
Superstructure
Welding deformation

1 Introduction

As one of the cost reductions in hull construction, there is an idea to use a hot-rolled steel
sheet (as stated in the below, simply referred as hot-coiled sheet) which is relatively easy
to procure in the production of a portion where the strength requirement is not severe
such as accommodation blocks of ship superstructure. The hot-coiled sheet is shipped
after rolled in a coil shaped form in steel mill manufacturers. After delivery to the
factory, the hot-coil material is recoiled by cold machining and returned to flat plate
shape for application of members in structures. Since the hot-coil material is processed
into flat plate by plastic deformation, there is a concern that the material will deteriorate
and that the material properties will be different from general flat rolled plate material (as
stated in the below, simply referred as flat plate) at the time of thermal cutting and
welding by the residual stress due to plastic work. The differences in the residual stress
distributions and the difference in welding difficulty with the aim of application of the
hot-coiled sheet to the hull superstructure are investigated in this study.

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 243–257, 2021.
https://doi.org/10.1007/978-981-15-4680-8_18
244 H. Narimatsu et al.

2 Comparison of Hot-Coiled Sheet and Flat Plate

2.1 Mechanical Properties, Chemical Composition and Microstructure
Specimen shall be equivalent to Grade KA specified by ClassNK [1] were prepared in
hot-coiled sheet and flat plate. According to consideration of variations in the material
properties by steel mill manufacturers, materials were prepared from foreign manu-
facturers and from Japanese manufacturers which have enough experience of hull
material approved by classification societies. Applied plate thickness in this study is
5.5 mm for all materials.
Mechanical properties, chemical composition and microstructure observation were
carried to confirm the differences. The tensile test carried out with verified to Class NK
rules [1]. Specimen shape U1 was applied. For chemical composition analysis, nitrogen
was measured by the inert gas melting-TCD method. Oxygen was measured by the
inert gas fusion and infrared absorption method. Other components was measured by
the QV method from material surface. Microstructure observation was performed 100
and 400 times at the depth position of a quarter from the plate surface. Results of the
tensile test, chemical composition measurements are shown in Tables 1 and 2,
respectively. Photographs of the microstructural observation is shown in Fig. 1.

Table 1. Mechanical properties of applied materials.

Materials Yield stress (0.2% Tensile Elongation Yield
Proof stress) (N/mm2) stress (%) ratio
(N/mm2) (%)
Hot-coiled sheet (308) 430 22 72
(made Overseas)
Flat plate (made (344) 468 24 74
Overseas)
Hot-coiled sheet 366 480 24 76
(made in Japan)
Flat plate (made in 356 445 29 80
Japan)

On the steel plates made overseas, 0.2% proof stress was adopted because a clear
yield point did not identified. Although the value of yield stress and tensile stress of
hot-coiled sheet made overseas is lower than those made in Japan, all measured values
of both materials satisfied the specification of Grade KA steel. Both microstructures are
ferrite and pearlite. There is a slight difference in grain size and average grain diameter
of hot-coil material made in japan is the finest. In the chemical composition, although
the addition of a small amount of Al, Ti, etc. is observed in materials made overseas,
 A Study on Using Hot-Rolled Steel Sheet for Ship Superstructure 245

Table 2. Chemical composition of applied materials (unit in wt%).

Component Made overseas Made in Japan
Hot-coiled sheet Flat plate Hot-coiled sheet Flat plate
C 0.094 0.142 0.177 0.133
Si 0.15 0.13 0.02 0.13
Mn 0.86 0.82 0.62 0.64
P 0.016 0.007 0.015 0.013
S 0.007 0.003 0.006 0.005
Cu 0.01 0.01 0.02 0.01
Al 0.034 0.036 0.017 0.015
Ni 0.01 0.01 0.01 0.01
Cr 0.02 0.01 0.03 0.02
Mo <0.01 <0.01 <0.01 <0.01
V 0.002 0.001 0.001 0.002
Nb 0.002 0.001 <0.001 <0.001
Ti <0.001 0.013 <0.001 <0.001
B 0.0001 0.0001 <0.0001 <0.0001
N 0.0023 0.0039 0.0025 0.0028
O 0.0012 0.0013 0.0020 0.0013

Hot-coiled sheet Flat plate Hot-coiled sheet Flat plate

(made overseas) (made overseas) (made in Japan) (made in Japan)

Fig. 1. Microscope observation.

it does not seem to have a significant influence on the difficulty for the thermal pro-
cessing. The amount of C of hot-coil material made overseas is low. This difference
might cause the difference in tensile strength between the hot-coil made in Japan and
overseas.
246 H. Narimatsu et al.

2.2 Residual Stress and Distribution of Hardness

Residual stress and Vickers hardness of each test materials are measured and compared
by each processing method (hot-coil material and sheet material) and by manufacturers.
The residual stress was measured using an X-ray diffraction method with cos a method
[2]. There is a concern that the properties of the hot-coil material might differ
depending on the curvature in the coil winding condition (outer radius 980 mm).
Therefore, the steel sheets with shearing cut four sides in way that as much as refrained
from thermal processing after rolling were prepared. Sample steel sheets with
1,800 mm wide and 3,500 mm long were picked up from the winding center side
(outer radius 380 mm) and the winding outer circumference side (outer radius
830 mm). Figure 2 shows the shape of the sample steel plate and the residual stress
measurement position. X-axis in Fig. 2 corresponds to the rolling direction of the hot-
coiled sheet and the flat plate. In these measurements, we ignored the front and back
side of the plate due to thin specimen thickness.

Fig. 2. Measured points of residual stress.

Figure 3 shows measured compressive residual stress distributions on the front

surface of the steel plate made overseas and Fig. 4 shows the result on the compressive
residual stress on the front surface of the steel plate made in Japan.
From the measured result of the residual stresses, it is confirmed that the com-
pression residual stress of hot-coiled sheet made in Japan is relatively smaller than one
made overseas. No significant difference of residual stress distributions between hot-
coiled sheet and flat plate in case of made overseas. In case of hot-coiled sheet made in
Japan, the compressive residual stress of the sheet picked up outside the coil tends to be
slightly higher than that picked up inner position.
 A Study on Using Hot-Rolled Steel Sheet for Ship Superstructure 247

700
Inside of the coil steel
Outside of the coil steel
600

Compressive residual stress[MPa]

Flat rolled steel

500

400

300

200

100

0
−80 −60 −40 −20 0 20 40 60 80
Y[mm]

(a) Residual stress distribution of x direction component along y axis

700
Inside of the coil steel
Compressive residual stress[MPa]

Outside of the coil steel

600
Flat rolled steel

500

400

300

200

100

0
−80 −60 −40 −20 0 20 40 60 80
X[mm]
(b) Residual stress distribution of x direction component along x axis

Fig. 3. Comparison of residual stress on the front surface of steel for picking position. (Made
overseas)
248 H. Narimatsu et al.

700
Inside of the coil steel
Outside of the coil steel

Compressive residual stress[MPa]

600 Flat rolled steel

500

400

300

200

100

0
−80 −60 −40 −20 0 20 40 60 80
Y[mm]
(c) Residual stress distribution of y direction component along y axis

700
Inside of the coil steel
Outside of the coil steel
Compressive residual stress[MPa]

600
Flat rolled steel

500

400

300

200

100

0
−80 −60 −40 −20 0 20 40 60 80
X[mm]
(d) Residual stress distribution of y direction component along x axis

Fig. 3. (continued)

Subsequently, Vickers hardness test using the test specimen used for residual stress
measurement were performed. Table 3 shows the average values measured 4 points on
each surface of the test specimen (measured position is red circle shown in Fig. 2).
Materials made in Japan have smaller hardness value for flat plate than hot-coiled sheet,
but there is no significant difference in materials made overseas.
 A Study on Using Hot-Rolled Steel Sheet for Ship Superstructure 249

600
Inside of the coil steel

Compressive residual stress[MPa]

Outside of the coil steel
500
Flat rolled steel

400

300

200

100

−100
−80 −60 −40 −20 0 20 40 60 80
Y[mm]

(a) Residual stress distribution of x direction component along y axis

500
Inside of the coil steel
Outside of the coil steel
Compressive residual stress[MPa]

400 Flat rolled steel

300

200

100

−100
−80 −60 −40 −20 0 20 40 60 80
X[mm]
(b) Residual stress distribution of x direction component along x axis

Fig. 4. Comparison of residual stress on the front surface of steel for picking position. (Made in
Japan)
250 H. Narimatsu et al.

600
Inside of the coil steel
Outside of the coil steel

Compressive residual stress[MPa]

500
Flat rolled steel

400

300

200

100

−100
−80 −60 −40 −20 0 20 40 60 80
Y[mm]
(c) Residual stress distribution of y direction component along y axis

600
Inside of the coil steel
Compressive residual stress[MPa]

Outside of the coil steel

500
Flat rolled steel

400

300

200

100

−100
−80 −60 −40 −20 0 20 40 60 80
X[mm]
(d) Residual stress distribution of y direction component along x axis

Fig. 4. (continued)

3 Deformation by Thermal Cutting

3.1 Verification of Distortion by Cutting
Distortion caused by thermal fabrications might affect the appearance of the product,
especially in thin sheet products. During the thin plate is thermally cut, the member is
deformed due to the heat input effect, and in the case where the member is elongated,
 A Study on Using Hot-Rolled Steel Sheet for Ship Superstructure 251

Fig. 5. Cutting plan.

Table 3. Comparison of Vickers hardness test results.

Type of plate Front side Back side
Made in Japan Flat plate 153 158
Hot-coiled sheet (inside) 198 181
Hot-coiled sheet (outside) 194 190
Made overseas Flat plate 195 203
Hot-coiled sheet (inside) 203 192
Hot-coiled sheet (outside) 194 191

longitudinal deformation is often a serious problem due to cutting heat input. Since the
hot-coiled sheet is affected by multiple plastic working for their recoiling process, there
is concern that the appearance of deformation of the member due to thermal cutting
might be different from that of the sheet material.
To compare and evaluate the deformation of each of hot-coiled sheet and flat plate
by the cutting process, thermal cutting test was performed using plasma and laser
cuttings to investigate the amount of deflection caused by the longitudinal deformation.
Each specimen (dimensions: 1,800 wide and 3,500 mm long) was cut into a size of
280 mm wide and 880 mm long as shown in Fig. 5. The deflections were measured at
the position shown in Fig. 6.
252 H. Narimatsu et al.

Fig. 6. Measured points of deflection.

Table 4. Measured deflection after cutting (unit in mm).

Type of plate Laser cutting Plasma cutting
Made in Japan Flat plate 0.0 2.6
Hot-coiled sheet (inside) 3.0 6.2
Hot-coiled sheet (outside) 2.2 4.8
Made overseas Flat plate 9.1 5.1
Hot-coiled sheet (inside) 3.9 9.1
Hot-coiled sheet (outside) 5.8 9.5

Table 5. Welding conditions of submerged arc welding.

Travel speed (mm/min): 600
Arc current (A): 320
Arc voltage (V): 30
Nominal heat input (J/mm): 960
 A Study on Using Hot-Rolled Steel Sheet for Ship Superstructure 253

Table 6. Measured deflection after welding (unit in mm).

Type of plate Laser cutting Plasma cutting
Made in Japan Flat plate 16.0 (16.0)* 18.0 (15.5)*
Hot-coiled sheet (inside) 19.8 (16.8)* 28.3 (22.1)*
Hot-coiled sheet (outside) 22.0 (19.9)* 29.5 (24.7)*
Made overseas Flat plate 20.3 (11.2)* 24.0 (18.9)*
Hot-coiled sheet (inside) 23.8 (19.9)* 27.3 (18.2)*
Hot-coiled sheet (outside) 18.3 (12.5)* 26.5 (17.0)*
*Values of inside of bracket means difference from ‘after cutting’ to ‘after
welding’.

Fig. 7. Measured points of deflection.

3.2 Measurement Results and Consideration

Specimens were cut by plasma and laser cutting to 20 specimens (total 240 specimens)
of 280 mm wide and 880 mm long per each material. Table 4 shows the average value
of the longitudinal deflections at measured position shown in Fig. 6.
It is confirmed except for the hot-coiled sheet made overseas that the longitudinal
deflection in case of laser cutting is limited to about half of case of plasma cutting.
254 H. Narimatsu et al.

Table 7. Welding conditions of laser-arc hybrid welding.

Travel speed (mm/min): 1600
Arc current (A): 290
Arc voltage (V): 26
Laser power (kW): 4
Nominal heat input (J/mm): 433
Distance between laser spot and center of arc (mm): 5
Defocus distance (mm): −5
Drag angle of arc torch (degree): 40
Push angle of laser head (degree): 10
Wire extension of arc welding (mm): 20
Primer coating near weld line Removed

(a)Froont surface (b)Bacck surface

Macroscopic obsservation
(c)M

Fig. 8. Bead appearance and macroscopic observation.

Generally, it is the same tendency as the magnitude relation of general cutting defor-
mation between laser and plasma cuttings. Moreover, it is confirmed that the amount of
deflection is generally smaller for Japanese materials than for overseas materials.
Especially when laser cutting was done on the hot-coiled sheet made in Japan, the
deflection was almost at a negligible level. On the other hand, no significant difference
was confirmed in comparison between inside and outside of the hot-coiled sheet.
Referring these results, Considering these results, it is recommended to apply laser
cutting to the cutting of thin plates used in superstructures.
 A Study on Using Hot-Rolled Steel Sheet for Ship Superstructure 255

Table 8. Measured deflection after welding (unit in mm).

Type of plate Laser-arc hybrid Submerged arc
welding welding
Made in Flat plate (null) 16.1
Japan Hot-coiled sheet 5.7 16.8
(inside)
Hot-coiled sheet 6.2 19.9
(outside)
Made Flat plate (null) 11.2
overseas Hot-coiled sheet 7.6 19.9
(inside)
Hot-coiled sheet (null) 12.5
(outside)

Fig. 9. Measured points for the angular distortion.

Table 9. Measured angular distortion.

Type of plate Angular distortion
(degree)
Made in Japan Flat plate 4.06
Hot-coiled sheet (inside) 5.16
Hot-coiled sheet (outside) 5.56
Made overseas Flat plate 2.93
Hot-coiled sheet (inside) 5.93
Hot-coiled sheet (outside) 3.39
256 H. Narimatsu et al.

4 Evaluation of Thermal Deformation by Welding

4.1 Verification of Welding Deformation by Different Cutting Method
To verify the difference in welding deformation between hot-coiled sheet and flat plate,
the cut specimens by the plasma and laser cuttings obtained in the previous section was
joined by the butt both-side submerged arc welding. Each the deflection before welding
and after welding were measured. Welding conditions are shown in Table 5 and
measurement positions of deflection are shown in Fig. 7. The joint was manufactured
by welding the longitudinal direction (except for the outer peripheral part) of two
adjacent test pieces in Fig. 5.
Table 6 shows the measured value of the deflection of each specimen after welding.
The values enclosed in parentheses in the table represent the difference between the
amount of deflection before and after the welding by subtracting the deflection amount
immediately after tack welding for the fitting.
The total amount of deflection is smaller for the laser cut material than for the
plasma cut material, since the amount of deflection before welding affect total
deflection. Regarding the deformation due to welding, no significant difference was
found in the relationship between the sheet and plate making process and the manu-
facturing method or the cutting method.

4.2 Verification of Welding Deformation Due to Difference in Welding

Heat Input
In the previous section, we confirmed that similar deflection is added as weld defor-
mation regardless of steel type, when welding is applied to the steel sheet once sub-
jected to the thermal cutting process. Therefore, we investigated the deflection before
and after welding in the case of using a laser-arc hybrid welding method, which is a
relatively low heat input welding method, to verify the possibility of reducing the
amount of deformation. The measurement positions of the deflection are the same as in
Fig. 7. Table 7 shows the welding conditions for laser-arc hybrid welding, Fig. 8
shows the photographs of the welded specimens. Measured results are shown in
Table 8.
Although the number of samples might be insufficient, measured results shown in
Table 8 indicates that the welding deformation by hybrid welding is suppressed to
about 1/3 of submerged arc welding and it is expected that welding deformation
occurring in the current shipbuilding work can be drastically reduced.
The measurement positions of angular distortion are shown in Fig. 9 Angular
distortion was calculated by the displacement in the thickness direction between the
two reference points and the two measuring points of deflection for the specimen
immediately after temporary fitting by laser welding alone and the specimen after the
main welding. The measurement results are shown in Table 9. The value of angular
distortion of the flat plate tended to be smaller than that of the hot-coiled sheet from
measured results. However, there was no significant difference in the degree of influ-
ence on the angular distortion in the hot-coiled sheet by the manufacturing country or
the sampling position.
 A Study on Using Hot-Rolled Steel Sheet for Ship Superstructure 257

5 Conclusion

From the viewpoint of the practical application of hot-coiled sheet to construct thin
wall structures, the difference of the properties, initial residual stress distributions and
the welding deformation of hot-coiled sheet and flat plate were investigated in this
study.
It was confirmed that there was no serious difference regardless of hot-coiled sheet
or flat plate, sampling location (inside or outside of the coil diameter) and orientation in
the hot-coiled sheet, manufacturing country, etc. in case of using conventional ship-
building methods such as plasma cutting and submerged arc welding. Therefore, the
difference in material strength and weldability between the hot-coiled sheet and the flat
plate is negligible.
On the other hand, it is confirmed that the application of low heat input construction
methods such as laser cutting and laser-arc hybrid welding can decrease the defor-
mation during the construction.
With regard to welding deformation reduction, it is important to carefully consider
the balance between human costs and new equipment investment costs in the con-
ventional manner to reduce the total construction cost.

References
1. Nippon Kaiji Kyokai (ClassNK): Rules for the survey and construction of steel ships (Part K
Materials) (2018)
2. Pulstec Industrial Co., Ltd.: Portable X-ray Residual Stress Analyzer l-X360n. https://www.
pulstec.co.jp/en/product/x-ray/ Accessed 15 Mar 2019
 Overview of the Joint Industrial Project
for Practical Application of Laser-Arc Hybrid
Welding in Construction of General Merchant
Ships in Japan

Koji Gotoh1(&), Takamori Uemura1, Issei Uchino2,

Hisao Narimatsu3, Toshimitsu Maeda4, Takumi Torigoe4,
and Atsuo Moriyama5
1
Kyushu University, Motooka 744, Nishi-ku, Fukuoka 819-0395, Japan
[email protected]
2
Namura Shipbuilding Co., Ltd., Shioya 5-1, Kurokawa, Imari 848-0121, Japan
3
Tsuneishi Shipbuilding Co., Ltd., Tsuneishi 1083, Numakumao,
Fukuyama 720-0393, Japan
4
Maeda Kogyo Co., Ltd., 22 Yon-No-Shimo, Nawa, Tokai 476-0002, Japan
5
Japan Ship Technology Research Association, Akasaka 2-10-9,
Minato City 107-0052, Tokyo, Japan

Abstract. The application of advanced welding technologies is important for

improving the efficiency of hull construction processes and reducing production
costs. Laser-arc hybrid welding technology, which combines the advantages of
laser and normal arc welding, is of high quality and is expected to improve
productivity. The Japan Ship Technology Research Association has organized
the joint industry project in collaboration with a university, a classification
society and some shipyards to apply this technology in the construction of
general merchant ships, in which the plate thickness of the main structural
member is thicker than passenger ships and high speed craft. The overview of
the first phase JIP results were introduced in PRADS 2016. In this presentation,
an overview of main research results of the second and third phase JIP, which
are listed below, are introduced.

Keywords: Laser-arc hybrid welding
Hull construction
Full-penetration tee

joint by one-side single-pass welding
Weld groove processing
Weld seam
tracking

1 Introduction

Advanced welding technologies can contribute to improving the efficiency of hull

construction and to reducing production costs. In addition, the accuracy of hull con-
struction is directly related to the structural integrity of the hull structures.
Laser-arc hybrid welding technology, which has the advantages of laser and con-
ventional arc welding, provides high-quality welds and is expected to improve pro-
ductivity and accuracy [1, 2]. Sufficient reductions in residual welding deformation in

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 258–272, 2021.
https://doi.org/10.1007/978-981-15-4680-8_19
 Overview of the Joint Industrial Project for Practical Application 259

hull structures are expected by applying laser-arc hybrid welding because a major
advantage of laser-arc hybrid welding is its low heat input compared with conventional
arc welding. Such improvements will increase the productivity of hull construction and
permit structural styles that reduce the net weight of the hull. However, the following
difficulties arise when applying hybrid welding to hull construction.
1. The plate of a stiffener is expected to have thickness of at least 10–15 mm. Such
plates are often used for general cargo ships. To date, the applicable plate thick-
nesses of laser-arc hybrid welds are no more than about 10 mm.
2. Although grooves are usually mechanically cut to avoid the occurrence of weld
defects, it is necessary to thermally cut grooves during hull construction.
3. A pure-CO2 shield gas needs to be applied in the place of a CO2 and argon mixed
gas to reduce the construction cost.
4. Laser power sources are expensive compared with other construction tools.
5. It is necessary to check the effect of primer paint, which prevents corrosion of
welded joints, on the structural integrity of the welded joint.
A joint industrial research project focusing on the applicability of laser-arc hybrid
welding in the construction of general merchant ships was organized by the Japan Ship
Technology Research Association and supported by the Nippon Foundation and the
Shipbuilding Research Centre of Japan. The research was conducted from 2012 to
2014. The achievement of this JIP was presented in PRADS 2016 [3] and the guide-
lines of laser-arc hybrid welding for hull construction was published by Class NK [4]
according to the JIP results. However, several research subjects that could not be
addressed in this research project were pointed out. Therefore, new members joined the
JIP and a new JIP worked to solve these problems. In this presentation, an overview of
the main research results of the second phase JIP are introduced.

2 Full-Penetration Tee Joint by One-Side Single-Pass

Welding

2.1 Investigation of Appropriate Welding Conditions

Achievement of our first phase JIP indicates the possibility of the fabrication of the full-
penetration tee joint with stiffener thickness is about 15 mm by one-side single-pass
welding is expected by applying the laser-arc hybrid welding. We conducted experi-
ments to establish the welding procedure for fabrication of this type joint. The applied
material, welding consumable and shielded gas are shown in Table 1. Specifications of
power sources applied in this JIP are shown in Table 2. These materials were applied
by considering the situation currently used in hull construction in Japan. The same
materials were also applied in the investigations introduced the following sections.
The specimen configuration is shown in Fig. 1. Residual oxide film generated by
laser cutting and anti-rust primer was removed to prevent the generation of blowhole
and other weld defects.
260 K. Gotoh et al.

Table 1. Materials of the plate and weld wire used.

Material High-tensile steel, grade KD36 approved by Nippon Kaiji Kyokai
(Class NK)
Plate thickness Thickness 14 mm for stiffener and 21 mm for base plate
Plate cutting process Laser cutting
Groove preparation I-shaped groove (no groove preparation)
Weld consumable Solid wire MG-50 (diameter, 1.6 mm) approved by Nippon Kaiji
(Wire) Kyokai (Class NK)
Shielded gas for arc Pure CO2
welding

Table 2. Specifications of the applied welding power sources.

Laser 20 kW fiber laser (IPG Photonics, YLS-20000-S2T)
Shielded gas for arc Digital inverter type CO2 arc welding machines (Daihen Welbee
welding Inverter M500G)

t =21

10 0
200

x z
t =14

y y
300

Un it:m m
x

Fig. 1. Specimen configuration of the tee shaped joint for detecting the suitable welding
condition.

To investigate suitable welding conditions, various welding parameters as control

factors and noise factors were set, as shown in Table 3, and assigned control factors to
Taguchi experimental design–L18 orthogonal array tables. Other welding conditions
were set as constant values shown in Table 4.
In these investigative experiments, we set the laser head angle with horizontal hL at
a low angle so that the boundary surface between the base plate and the stiffener
connecting region lay within the laser irradiation area to increase the laser irradiation
area to enlarge the melting region.
 Overview of the Joint Industrial Project for Practical Application 261

Table 3. Control factors and noise factors for detecting the suitable welding condition.

Control factors Level

Laser power 2
Arc current 3
Arc voltage 3
Defocusing distance, Df 3
Arc torch angle, θA 3
Laser head angle, θL 3
Arc push angle, θp 3
Laser drag angle, θd 3
Noise factors Level
Wire extension, d 3
Laser radiation point, HL 3
Distance between web
surface and arc torch 3
aiming point, DAW

Table 4. Fixed welding conditions applied in this study.

Item Value
Joint gap (mm) less than 0.05
Travel speed (mm/min) 900
Distance between laser radiation point and arc torch aiming point, DLA (mm) 3

Table 5. Detected suitable welding conditions.

Travel speed (mm/min) 900
Laser power (kW) 20
Arc current (A) 425
Arc voltage (V) 27.3
Wire extension, d (mm) 3
Vertical distance between laser radiation point and arc torch aiming point, DLA (mm) 3

As a result of the above experiments, welding conditions were derived to obtain a

joint with good bead appearance and no internal defects. Part of the welding condition
is shown in Table 5 and the photograph of a cross section macroscopic observation is
shown in Fig. 2. It is confirmed that a sound fitting can be manufactured up to a gap of
1.5 mm by finely adjusting the welding current and voltage.
262 K. Gotoh et al.

(a) HL = 1.65 mm (b) HL = 2.0 mm

Fig. 2. Example of macroscopic observations.

2.2 Mechanical Test Required Guidelines

The mechanical performance of the welded joint was investigated according to the
Guidelines on Laser-Arc Hybrid Welding (ver.2) [4]. The test joint configuration to
extract the specimens is shown in Fig. 3. In addition, the nondestructive tests (radiation
inspection test, ultrasonic inspection test and magnetic particle inspection test) were
performed to judge the integrity of the welded joint according to ISO 5817 [5] and
acceptable results were obtained. Figure 4 shows the appearance of weld bead. Fig-
ure 5 shows the macroscopic observation of a cross section of the weld region. The
Vickers hardness test was also performed and the measured maximum value was
258 HV, which was measured at the weld toe on the base plate of specimen M1. This
value is less than the maximum allowable value of 380 HV.

Fig. 3. Specimen configuration of the tee shaped joint (weld length, 1000 mm).

Front side Back side

Fig. 4. Appearance of a weld bead near the center of the welded joint.
 Overview of the Joint Industrial Project for Practical Application 263

⟵ Back side ⟵ Back side ⟵ Back side

Front sid
de Front sid
de Front side
e

( Specimen M1
(a) (b
b) Specimen M
M2 (c) specimen M33

Fig. 5. Macroscopic observations.

2.3 Fatigue Strength

Although a fatigue test is required to approve the welding procedure, it is important to
ensure the fatigue performance of welded joints in hull structures. Fatigue tests were
performed to investigate the fatigue performance of the welded joints. Fatigue strength
was investigated by conducting three-point bending with a zero stress ratio according to
ISO TR14345-2002 [6]. A similar test was also conducted using a non-load carrying
type cruciform shaped joint fabricated by the same welding procedure mentioned
above.
Test results are shown in Fig. 6. Three fatigue design curves (IIW FAT100 for a
transverse non-load-carrying tee joint with an as welded fillet [7], IIW FAT80 [7] and
JSSC Class-E [8] for a transverse non-load-carrying tee joint with two-sided toe
grounded fillets) are drawn in this figure. In addition, fatigue test results for the twin-
pass tee welded joint fabricated by the hybrid welding presented in PRADS 2016 [3]
are plotted.

600 600
0
Single-pass T joint
Nominal stress range [MPa]

Nominal stress range [MPa]

400 400
0

200 200
0

100 100
0
80 IIW FAT100 80
0 IIW FAT100

60 Tw
win-pass T joint IIW FAT80
J
IIW FAT80
JSSC Class E
60
0 JSSC Class E
Specimens includin ng no weld defect
40 40
0 3 point bending)
: Tee joint (3
Specimens containing embedded flaws
(max. size: less tha
an 1 mm) : Cruciform joint
j
Specimens taken frrom tack weld zone (Non load ccarrying type, axial loading)
20 20
0 5 6 7
5 6 7
10
1 10
0 10 10 10 10
Number of cycles to Failure [Cyccle] Number of cycles to Failure [Cycle]
(a) Tee joint
j (bb) Non-load carrying type cruciform
joint

Fig. 6. S-N curves.

All joints fabricated by one side single pass welding had high fatigue performance
compared with the IIW FAT 80 fatigue design curve and twin-pass welded joints.
Remarkably, the slope of the S-N curve of the tee joint fabricated by single pass welding
264 K. Gotoh et al.

is shallow compared with the others. However, the slope of the S-N curve of the
cruciform joint is similar to the design curves. The reason is that the joint might contain
misalignment. By considering such high-fatigue performance, it is expected that most
fillet welded joints in a hull structure can be replaced with laser-arc hybrid welded joints.

3 Influence of the Weld Groove Processing Method

3.1 Influence of Groove Shape Cutting Methods on Joint Quality
Because the accuracy of groove preparation generally affects the joint quality of hybrid
welds, European shipyards, which are leaders in the introduction of hybrid welding for
passenger ships, apply mechanical cutting to groove preparation. In general merchant
ship construction, groove preparation by thermal cutting is suitable in terms of cost. For
this reason, previous studies [3] have been restricted to laser cutting that has excellent
cutting accuracy in thermal cutting for the groove preparation method. However,
plasma and gas cutting are often used in Japanese shipyards, and there are grooves (e.g.
Y-shaped grooves) that can be processed only by only gas cutting. Therefore, in the
case of groove processing by plasma or gas cutting, welding conditions for manu-
facturing a good quality joint by hybrid welding were examined. The applied materials
are the same as in Table 1.
Suitable welding conditions were detected by conducting a similar procedure
described in the previous section. However, only visual inspection of bead appearance,
cross section macroscopic observation and Vickers hardness measurement (measured
maximum value, 297 HV, which is less than the allowable limit of 380 HV) was
conducted to judge the validity. The obtained results are shown in Table 6. This
condition is applicable to the maximum gap of 0.5 mm. Thermal cutting conditions for
plasma and gas cutting are followed as the standard cutting conditions in shipyards.
The arithmetically average groove surface roughness of a specimen fabricated by
plasma and gas cutting were about 3 lm and 10.5 lm. The value by plasma cutting is
almost the same as laser cutting.

Table 6. Detected suitable welding conditions.

Travel speed (mm/min) 1200
Laser power (kW) 12
Arc current (A) 400
Arc voltage (V) 26.0

Figure 7 shows the bead appearance and macroscopic observation of the joints for
which groove preparation was conducted by plasma and gas cutting. In both cases,
spatter was observed on the front bead-side surface, but the bead shape itself was good.
In addition, no weld defects were observed on the cross section of the weld bead. By
considering these results, it is expected that plasma and gas cutting can be applied to
groove preparation for laser-arc hybrid welding.
 Overview of the Joint Industrial Project for Practical Application 265

(1) Front side appear- (1) Front side appear-

ance ance

(2) Back-side appear- (3) Cross section (2) Back-side appear- (3) Cross section
ance macroscopic observa- ance macroscopic observa-
tion tion
(a) Groove preparation by plasma cutting (b) Groove preparation by gas cutting.

Fig. 7. Comparison of bead appearances and cross sections by different groove preparation
methods.

3.2 Influence of Retained Primer and Oxide Film on Joint Quality

In general, it is useful to remove the antirust primer near connected regions prior to
construction to prevent the formation of weld defects. In the case of groove preparation
by thermal cutting, welding defects sometimes occur if an oxide film remains on the
groove surface, as in the case of the primer. However, the conventional arc welding
used in the standard shipbuilding process enables fabrication of a good quality welded
joint without removing the primer and the oxide film in advance. Therefore, it is
necessary to enable fabrication of good quality joints, even if the primer and oxide film
remain. The influence on joint quality when these are retained near the connection
region was investigated. The test condition of the primer and oxide film is shown in
Table 7. The primer and oxide film were removed by a disk grinder. Bead appearances
and macroscopic observations of each specimen are shown in Fig. 8.

Table 7. Test conditions.

Specimen Joint Gap Primer Oxide film on the groove
ID type (mm) surface
B1 Butt 0.5 Retained on the front Removed
surface only
B2 Butt 0.5 Retained on the back Removed
surface only
B3 Butt 0.5 Removed Retained
T1 Tee 0 Removed Removed
T2 Tee 0 Retained Retained
266 K. Gotoh et al.

(1) Front-side appear- (1) Front-side appear-

ance ance

(2) Back-side appear- (3) Cross section (2) Back-side appear- (3) Cross section
ance macroscopic observa- ance macroscopic observa-
tion tion
(a) Specimen B1 (b) Specimen B2

(1) Front-side appear-

ance

(2) Back-side appear- (3) Cross section

ance macroscopic observa-
tion
(c) Specimen B3

(1) Front-side appear- (1) Front-side appear-

ance ance

(2) Back-side appear- (3) Cross section (2) Back-side appear- (3) Cross section
ance macroscopic observa- ance macroscopic observa-
tion tion
(d) Specimen T1 (e) Specimen T2

Fig. 8. Weld bead appearance and macroscopic observation.

 Overview of the Joint Industrial Project for Practical Application 267

In specimen B1, although a slight decrease in surface bead height was confirmed, a
good bead appearance and cross section were obtained. However, underfill occurs on
both the front and back beads in specimen B2 and occurs on the front-side bead in
specimen B3. These results indicate that the primer on the back surface of the test
specimen and the oxide film can contribute to the formation of defects when producing
a welded joint using hybrid welding. As a cause of the generation of the underfill, it is
considered that the primer and the oxide film are thermally decomposed by welding to
generate carbon monoxide (CO) gas, and the gas stirs the molten pool and scatters from
the back surface of the material. The gas generated when the oxide film is thermally
decomposed by laser welding and the gas generated when the primer is thermally
decomposed by arc welding contains a large amount of CO. In the case of the primer on
the front surface, it is presumed that the primer effect does not occur because the gas
generated at the time of welding is released into the atmosphere and is not mixed in the
molten pool. However, a good bead appearance and no defects were observed in the
case of the tee joint, see Fig. 8(d) and (e). The generated CO gas is released and not
mixed in the molten pool because the preceding laser beam drills a tunnel and the gas
can be released into the atmosphere.

4 Application of Weld Seam Tracking Technology

To construct long welded joints with allowable quality, it is necessary to track the weld
line and gap variation, and to appropriately change the welding conditions when
necessary. This JIP focused on the seam tracking device for welding in automobile
manufacturing, as reported by some authors [9–11], and examined the possibility of
stable long-distance welding by introducing this device. The approximate installation
location of the sensor is shown in Fig. 9.

Laser head

Arc torch

Tracking sensor

Fig. 9. Tracking sensor.

268 K. Gotoh et al.

As shown in Fig. 10, the welding line was aimed at the start point, and the spec-
imen was set up by lifting the flange so that the welding target positions were each
shifted 15 mm away from the end point. To perform welding with the seam tracking
function enabled, the specimen was deformed freely during welding by clamping the
two ends of the specimen only. From the information obtained by tracking, linear
interpolation teaching of only the start and end points was performed. In the evaluation,
the imaging window was checked visually to determine whether the sensor was con-
taminated with fumes or spatters during the welding. The welding line was also
checked to determine whether missed or abnormal values could be removed. Verifi-
cation was performed at a welding length of 4800 mm, a laser power of 20 kW, a
welding current of 300 A, and a speed of 1000 mm/min.

Top View
Start
point Tracking distance 4800 mm End point
Weld line shift:
Web 15 mm
Flange

Side View Initial setting of

trajectory of
welding robot
Start Web
point End point
Weld line shift:
Flange 15 mm

Fig. 10. Specimen setting condition.

It was confirmed that, in the case of dry operation (with only a robot moving
without laser and arc welding), sufficient tracking accuracy was obtained. However, at
the time of construction of a full-penetration tee joint by one-side single-pass welding,
the lens of the weld line detection sensor became cloudy because of fumes, so detection
accuracy decreased as the welding length became longer, and the sensor overheated,
confirming that the durability was insufficient. Therefore, we have taken measures such
as installing an air purge mechanism for fume removal and a cooling mechanism near
the sensor. As a result, it was confirmed that tracking was possible within ±0.2 mm for
both aiming and height deviation.
In the hybrid welding condition, examination of a full-penetration tee joint by one-
side single-pass welding with a welding length of 2000 mm, even in a situation where
the target position changes significantly because the tacking was removed after the
welding length of 1000 mm, confirms that the tracking mechanism remains opera-
tional. Figure 11 shows the appearance of the specimen and the amount of horizontal
displacement of the upright plate corresponding to this examination.
 Overview of the Joint Industrial Project for Practical Application 269

Distance from original center

20

line of stiffener [mm]

15

10
Welding direction
5

0
0 500 1000 1500 2000
Location of weld line [mm]
(a) Front bead appearance; from 1800 mm to(b) Horizontal deflection of the stiffener along
2000 mm the weld length.

Fig. 11. Performance check result of the tracking system.

5 Prototyping of a 5-m-Long Welded Joint

Using the above-mentioned seam tracking system, a full-penetration tee joint was
manufactured by one-side single-pass welding with a total length of 5000 mm. As a
limitation of the laser oscillator performance, the laser power decreases as the con-
tinuous construction time increases. Therefore, the construction was temporarily
stopped at a welding length of 2000 mm, and then 3000 mm was constructed. In
addition, the laser power was automatically changed according to the gap width
obtained by seam tracking while finely adjusting the part of the welding conditions
derived by this JIP. The initial gap was set to 1.5 mm from the viewpoint of the
tracking detection.
Examples of the bead appearance for the range of the first welding length of
2000 mm are shown in Fig. 12. The front bead was separated from the web plate in
some sections and a bead formed on the base plate. The back bead was poured into the
back-side of the web through the through hole created by the leading laser irradiation,
and it is estimated that the target position of the subsequent arc is improperly shifted
because of this effect. Figure 13(a) shows the change of the laser power during
welding. It was confirmed that gap width detection by tracking and laser power control
according to the laser power changes were appropriate. A 1.5-mm-thick filler gauge
was inserted between the web and base plates at intervals of 500 mm for setting the
initial gap, so it was judged that the gap was 0.0 mm and the laser power increased
instantaneously.
Examples of the appearance of a bead with a welding length of 3000 mm are
shown in Fig. 14, and the laser power history during welding is shown in Fig. 13(b). It
is confirmed from Fig. 14 that the back bead becomes wavy from around 3700 mm.
The laser power history at this point (about 113 s in Fig. 13(b)) was about 17 kW.
However, stable back beads were formed with the same level of output at a welding
length of 2750 mm (about 50 s in Fig. 13(b)). The reason the back bead shape is
different even when the laser power is the same might be that the power decrease from
the laser focal shift due to the thermal expansion of lens occurred. Another possibility
is that fume and plume stagnation around the processing point occurred. However, the
reason for this phenomenon has not yet been identified. The front bead was very short
270 K. Gotoh et al.

(f-1) Front bead from 0 mm to 300 mm (b-1) Back bead from 0 mm to 300 mm

(f-2) Front bead from 600 mm to 900 mm (b-2) Back bead from 600 mm to 900 mm

(f-3) Front bead from 1500 mm to 1800 mm (b-3) Back bead from 1500 mm to 1800 mm

Fig. 12. Weld bead appearance on a full-penetration tee joint fabricated by one-side single pass
welding. (Welding section from 0 mm to 2000 mm).

19 19

18.5 18.5

18 18
Laser power [kW]

Laser power [kW]

17.5 17.5

17 17

16.5 16.5

16 16

15.5 15.5

15 15
0 40 80 120 0 20 40 60 80 100 120 140 160 180 200
Laser irradiation time [s] Laser irradiation time [s]
(a) Welding section: 0 mm to 2000 mm (b) Welding section: 2000 mm to 5000 mm

Fig. 13. History of laser power during the welding.

in leg length and formed a bead along the web plate. Considering the size of the back
bead, it is believed that the total amount of filler metal is insufficient at the current
welding wire feed amount, and it is necessary to make adjustments such as reducing the
welding speed. The laser power fluctuates in six sections divided by filler gauge
insertion, which is consistent with the number of sections between filler gauges
installed at a welding length of 3000 mm (installation interval, 500 mm). It can be
confirmed that each function of the tracking device was fully exhibited.
 Overview of the Joint Industrial Project for Practical Application 271

(f-1) Front bead from 2000 mm to 2100 mm (b-1) Back bead from 2000 mm to 2100 mm

(f-3) Front bead from 2400 mm to 2700 mm (b-3) Back bead from 2400 mm to 2700 mm

(f-5) Front bead from 3000 mm to 3300 mm (b-5) Back bead from 3000 mm to 3300 mm

(f-7) Front bead from 3600 mm to 3900 mm (b-7) Back bead from 3600 mm to 3900 mm

(f-9) Front bead from 4200 mm to 4500 mm (b-9) Back bead from 4200 mm to 4500 mm

(f-11) Front bead from 4800 mm to 5000 mm (b-11) Back bead from 4800mm to 5000mm

Fig. 14. Weld bead appearance on a full-penetration tee joint fabricated by one-side single pass
welding. (Welding section from 2000 mm to 5000 mm).

6 Concluding Remarks and Future Challenges

The second phase research project for introducing laser-arc hybrid welding in the
construction of general merchant ships has been conducted and the following results
were obtained.
1. Welding conditions to fabricate a good quality full-penetration tee joint by one-side
single-pass welding were examined.
2. The influence of groove shape cutting methods was investigated and the possibility
of application of other thermal cutting methods (plasma and gas cutting) was
confirmed.
3. The influence of retained primer near weld line and oxide film on the groove surface
on the joint quality was investigated, and the necessity of removal of both con-
taminants during butt welding was confirmed.
4. It was confirmed that stable weld seam line tracking was possible during long-
distance welding by using a seam tracking system.
272 K. Gotoh et al.

Further Challenges related to this research are as follows.

5. Expansion of welding conditions, e.g. applicable plate thickness and gap width.
6. Application to other welding positions (horizontal, vertical and overhead positions).
7. Examination of fatigue performance of load carrying type cruciform joints.
8. Examination of fracture toughness in each notch position.

Acknowledgement. This research was conducted as a joint research project with the Ship-
building Research Center of Japan and other shipyards, with the organizing committee set up at
the Japan Ship Technology Research Association (JSTRA). The authors express their appreci-
ation to the National Maritime Research Institute for the fatigue test of cruciform joints and to the
JFE Steel Corporation for providing steel plates.

References
1. Gerriten, C., Howarth, D.: A review of the development and application of laser and laser-
arc hybrid welding in European shipbuilding. In: Proceedings of 11th CF/DRDC
International Meeting on Naval Applications of Materials Technology, Halifax, NS, Canada,
pp. 7–9 (2005)
2. Koga, H., Goda, H., Terada, S., Hirota, K., Nakayama, S., Tsubota, S.: First application of
hybrid laser-arc welding to commercial ships. Mitsubishi Heavy Ind. Tech. Rev. 47(3), 59–
64 (2010)
3. Gotoh, K., Fukui, T., Moriyama, A., Tsumura, S.. Sawato, H.: Overview of a joint industrial
project for introducing laser-arc hybrid welding to the construction of general merchant ships
in Japan. In: Proceedings of 14th Practical Design of Ships and Other Floating Structures
(PRADS 2016), ID 014 (2016)
4. Nippon Kaiji Kyokai (ClassNK): Guidelines on Laser-Arc Hybrid Welding (Ver.2) (2015)
5. ISO 5817–2014: Welding–Fusion-welded joints in steel, nickel, titanium and their alloys
(beam welding excluded) – Quality levels for imperfections (2014)
6. ISO TR14345-2002: Fatigue–Fatigue testing of welded component–Guidance (2002)
7. Hobbacher, A.: Fatigue Design of Welded Joints and Components–Recommendations of
IIW Joint Working Group XIII–XV, Woodhead (1996)
8. Mori, T., Minami, K., Kabuto, Y.: Review of fatigue strength of welded joints specified in
the JSSC fatigue design recommendations, Kou Kouzou Rombunshuu 18(69), 71–81
(2011). https://doi.org/10.11273/jssc.18.69_71
9. Maeda, T., Torigoe, I., Kutsuna, M.: Study on seam tracking for laser welding. In:
Proceedings of Japan Welding Society Annual Meeting, vol. 99, pp. 64–65 (2016). (in
Japanese)
10. Maeda, T., Torigoe, I., Kutsuna, M.: Study on seam tracking and adaptive control for laser-
arc hybrid welding. In: Proceedings of Japan Welding Society Annual Meeting, vol. 100,
pp. 58–59 (2017). (in Japanese)
11. Maeda, T., Torigoe, I., Kasuga, S., Kutsuna, M.: Study on seam tracking system for laser
welding using tack welds identification function. In: Proceedings of Japan Welding Society
Annual Meeting, vol. 101, pp. 358–359 (2017). (in Japanese)
 Control of Car Decks Welding Deformation
in RO-RO Ships

Junichi Toyoshima1, Kiyoharu Hayashi1, Mikikazu Murakami1,

Masanori Sano1, Shogo Imasaki1, Yasuo Katayama2,
Hidekazu Murakawa2, Naoki Osawa3, and Sherif Rashed2(&)
1
Naikai Zosen Corporation, Onomichi, Japan
2
Joining and Welding Research Institute, Osaka University, Osaka, Japan
[email protected]
3
Department of Naval Architecture and Ocean Engineering, Osaka University,
Osaka, Japan

Abstract. Lashing rails are used in RO-RO ships to lash cars to car decks
during transportation. These rails are firmly fillet welded to decks to avoid
damage by car tires during loading and unloading. These rails are welded to
decks after the deck is completely assembled and welded. Due to their small
height, the effect of lashing rails welding tendon farce is additional compressive
stress and shrinkage in deck plating, and additional sagging of deck structure.
Local plate deflection between longitudinal stiffeners also considerably increa-
ses. Correction of this deformation is difficult and consumes work hours and
calendar time.
In this work, Alternative welding sequences are numerically investigated.
Back heating, treatment of gaps and misalignments occurring during welding
and alternative supporting methods during welding are considered. The best
sequence is selected based on comparison of welding deformation caused by
each sequence.
To reduce compressive stress and local deformation in deck plate, mechanical
straitening of deck platting after welding the lasing rails is numerically con-
sidered. This technique can only be applied when lashing rails are welded first.
Reduction of tendon force and transverse bending is numerically predicted and
checked against physical testing of mechanical straitening of a deck plate with
lashing rails welded to it. Such reduction is incorporated in the prediction of
deck structure welding deformation. Its effect on overall as well as local
deformation of the deck is assessed.

Keywords: Lashing rail
Welding deformation
Welding sequence

Mechanical straitening

1 Introduction

Lashing rails are used in RO-RO ships to lash cars to car decks during transportation.
A lashing rail is a short length (1500 mm to 2000 mm) of a shallow flat bar (about
25 mm high) with closely spaced openings to mount lashing wires. Such rails are laid
along car decks on the upper side, closely spaced in the longitudinal direction. In the

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 273–286, 2021.
https://doi.org/10.1007/978-981-15-4680-8_20
274 J. Toyoshima et al.

transverse direction, they are spaced at about 700 or 800 mm, thus covering the whole
area of the deck. These rails are firmly fillet welded to decks to avoid damage which
may be caused by car tires during loading and unloading. These rails are traditionally
assembled and welded to decks after the decks are completely assembled and welded.
Due to their small height, bending neutral axis of the flat bar and the effective plating
attached to it is at or below the weld. Therefore, the effect of welding tendon farce
(acting at the weld) is additional compressive stress in deck plate, additional deck plate
shrinkage and additional deck structure sagging. Such sagging reduces the clear height
of the deck below. Local plate deflection between longitudinal stiffeners also consid-
erably increases. Correction of this deformation is difficult and consumes work hours
and calendar time.
In this work, measures to reduce welding deformation are considered. Other
welding sequences are numerically investigated using the inherent deformation
method. In addition to the traditional sequence, in which the lashing rails are welded
last, sequences in which the lashing rails are assembled and welded early in the
assembly process are considered. Back heating (line heating on the plate surface
opposite to stiffeners weld) to reduce local plate deflection is included in the prediction
of welding deformation. Treatment of gaps and misalignments occurring during
welding is also taken into account. Alternative panel supporting methods during
welding are considered. The best sequence is selected based on comparison of welding
deformation caused by each sequence.
To reduce compressive stress in deck plate and plate deflection between stiffeners,
mechanical straitening of decks after welding the lasing rails is numerically considered.
This mechanical straitening is carried out by enforcing the plate with the lashing rails
welded to it through rollers to straighten lashing rails welding deformation. This
operation greatly reduces the welding tendon force and transverse bending. However,
this technique can only be applied in the sequence where lashing rails are the first
members to be welded to deck plate. Reduction of tendon force and transverse bending
is numerically investigated. Results of the investigation are incorporated in the pre-
diction of welding deformation and the effect of mechanical straightening on overall as
well as local deformation of the deck is assessed. Physical testing of mechanical
straitening of a plate with lashing rails welded to it is carried out to confirm the findings
of the numerical investigation.

2 Car Deck Structure

The considered car deck section is 11.230 m long  21.000 m in breadth. It is a

longitudinally stiffed structure as shown in Fig. 1. Deck plating is 7 mm thick and is
assembled of 7 flat plate panels, each 11.230  3.000 m. These blanks are welded
together over longitudinal I beams 100  5 + 75  7 mm, marked by small arrows in
Fig. 1, as shown in Fig. 2, except one butt welded joint at one side. Deck plate panels
are supported by longitudinal inverted angles 100  75  7 mm. Longitudinals are
spaced at 750 mm. Deck girders and deck beams are as shown in Fig. 1. Transverse
beams are continuous and spaced at 3200 mm. Lashing rails are 1600 mm lengths of
25  8 mm flat bars closely spaced in the longitudinal direction with spaces just to
 Control of Car Decks Welding Deformation in RO-RO Ships 275

round the weld. In the transverse direction they are spaced at 800 mm. An end angle is
fitted at one end of the deck section as shown in Fig. 2 to facilitate assembly of the
section to the next section.

Fig. 1. Deck structure

Fig. 2. I Beams joining deck plate panels and end angle used in joining sections

3 Assembly Sequences

3.1 Traditional Assembly Sequence

In the traditional assembly sequence, assembly is carried out as follows, see the left
hand side of Fig. 3.
276 J. Toyoshima et al.

1. It starts by automatic welding of longitudinals to the 7 flat plate panels. Meanwhile,

webs of girders and deck beams are welded to their flanges.
2. Next, the 7 panels are laid upside down on supporting pins. Joining I beams,
girders, deck beams and end angle are fitted and tack welded then welded. Note that
welding of joining I beams and end angle on the upper side of the deck structure
(now on the lower side) is not carried out at this stage, see Fig. 2. The deck plate but
joint at side is also tack welded and welded.

Weld longitudinals to Fit and Weld lashing rail Weld longitudinals to

deck plate deck plate
Weld beams and girders Weld beams and girders
webs to their flanges Weld longitudinals to
webs to their flanges
deck plate
Weld beams and girders
Fit and Weld all other webs to their flanges Fit and Weld lashing rail
stiffening members to
deck plate
Fit and Weld all other Fit and Weld all other
stiffening members to stiffening members to
Fit and Weld all stiffening deck plate deck plate
members to each other
Fit and Weld all stiffening Fit and Weld all stiffening
members to each other members to each other
Turn over

Complete welding I Turn over Turn over

beams and end angle to
deck plate
Complete welding I Complete welding I
beams and end angle to beams and end angle to
Back heating deck plate deck plate

Fit and Weld lashing rail Back heating Back heating

Sequence 1 (original) Sequence 2 Sequence 3

Fig. 3. Alterative deck section assembly sequences

3. Stiffening members are fitted, tack welded and welded to each other.
4. Panel is turned over to the upright position and supported on 4 supporting blocks at
the outmost 4 intersections of girders and deck beams. One more supporting block
at the center of the deck structure is sometimes provided.
5. Now the joining I beams’ and end angle’s welding on the upper side of the deck
plate are completed.
6. Back heating of all stiffeners welding is carried out.
7. Finally lashing rails are fitted tack welded then welded.
 Control of Car Decks Welding Deformation in RO-RO Ships 277

During fitting and tack welding, any gaps and misalignments are rectified as far as
practicable. Remaining small gaps and misalignments are accepted.

3.2 Alternative Assembly Sequences

If lashing rails are welded to deck plate before fitting and welding stiffeners, deck plate
would shrink due to the tendon force and transverse shrinkage, but will remain almost
flat. Stiffening members can then be fitted and welded. This may not reduce the
compressive stress in deck plate, but would reduce overall deck structure welding
deformation and deck sagging. Sequence 2 shown in the middle of Fig. 3 incorporates
this logic. It is similar to the traditional sequence except that lashing rails are fitted, tack
welded and welded to the 7 plate panels first, before fitting and welding longitudinals.
A shop floor concern was that welding the lashing rails first would cause small
deformations. These deformations, though small, may hamper fitting and automatic
welding of longitudinals. Therefore a third sequence, in which lashing rails are welded
after welding longitudinals to the 7 deck plate panels, is also considered. This sequence
is shown on the right hand side of Fig. 3.

4 Welding Deformation Analysis

Deck structure welding deformation when assembled according to each of these 3

sequences is numerically predicted. Methods and finite element model used in the
prediction are outlined in the following. Obtained results are also summarized and
compared to each other.

4.1 Welding Deformation Prediction Methods

In this study, the inherent deformation method is applied to evaluate welding defor-
mation using a shell finite element elastic analysis. In welding of plate structures, the
inherent deformation can be divided into four components; longitudinal shrinkage,
transverse shrinkage, longitudinal bending, and transverse bending. The inherent
deformation method can also be used to analyze deformation caused by straightening
when straightening heating lines are at the weld lines [3].
Interface elements are inserted between the parts (members) to be welded together
to model the joint state, free (no joint), fitted (positioned and tack welded), or welded.
The interface element also allows consideration of gap and misalignment treatment.
Details of the inherent deformation method and the interface element may be found in
references [1–3].
Evaluation of Inherent Deformation. Inherent deformation of each welding line and
each straitening heating line is evaluated in this study using empirical equations created
from results of thermal Elastic Plastic analysis and physical testing. Welding condi-
tions, i.e. electric potential, current and welding torch velocity are as actually used on
shop floor. In straightening [3], straightening heating lines are applied after welding is
completed and the joint has cooled down to room temperature. Applying line heating
278 J. Toyoshima et al.

on the plate side opposite to the weld line releases some of the residual stresses in and
around the weld. This releases a considerable part of the tendon force. Then when the
joint cools down, inherent deformation corresponding to the straightening heating line
is produced and added to the remaining tendon force, welding transverse shrinkage,
welding transverse and longitudinal bending. Release of the tendon force is estimated
based on work by Vega [4]. Inherent deformation of straitening heating lines is
according to Blandon [5]. The total straitening inherent deformation is evaluated taking
account of the partial release of tendon force.

4.2 Finite Element Model

Finite Element Mesh. The shell Finite Element model of the deck structure is shown
in Fig. 4. Nominal elements size is 130  160 mm in the breadth and length directions
respectively. This allows 6 elements between longitudinals allowing accurate predic-
tion of the deflection between longitudinals. The model has 31426 nodes, 25308 shell
elements and 5366 interface elements.

Fig. 4. Deck structure Finite Element model

Boundary Conditions and External Forces. In Z direction, when the model is upside
down, the model is supported on closely spaced pins. These pins prevent the deck from
falling down under gravity, but allow separation if the deck is to show upward dis-
placements. When the model is upright, two different boundary conditions are con-
sidered. The first is having 4 supporting blocks at the outmost 4 intersections of girders
and deck beams. The second is having one more supporting block at the center of the
deck structure. Supporting blocks, similar to supporting pins, prevent the deck from
falling down under gravity, but allow separation if the deck is to show upward
 Control of Car Decks Welding Deformation in RO-RO Ships 279

displacements. To prevent rigid body motion, points A and B in Fig. 4 are fixed only in
X direction, and points C and D are fixed only in Y direction. Gravity is applied.

4.3 Analysis and Results

Large deflection analysis of welding deformation is carried out in 6 cases, the 3
sequences shown in Sect. 3 above, each sequence with 2 different boundary conditions,
the first using 4 supporting blocks after turn over, and the second using also the
additional center supporting block as outlined in Sect. 4.2 above. The final deformed
shape (deformation scale 5:1) with contour plots of the Z displacement in the 6 cases
are shown in Figs. 5.

(1)

(a) (b)

(2)

(c) (d)

(3)

(e) (f)

Fig. 5. (1) Z-Displacement in sequence 1, with (a) and without center support (b) (2) Z-
Displacement in sequence 2, with (c), and without center support (d) (3) Z-Displacement in
sequence 3, with (e), and without center support (f)
280 J. Toyoshima et al.

Final deformed shapes in the cases with center support are plotted after removing of
the center support to have correct comparison with the cases without center support.
Minimum displacement (maximum displacement downward) and maximum displace-
ment (upward) at the outmost longitudinals are shown in Fig. 6. Twist of the deck
section is defined as follows;

Twist ¼ ðd2  d1 Þ þ ðd4  d3 Þ: ð1Þ

where, d1 ; d2 ; d3 and d4 are Z-displacements at the outmost longitudinals at points 1,

2, 3 and 4 shown in Fig. 6.

Maximum Displacement At Minimum Displacement

Outmost Longitudinals
W/O W
100 0
Displacement (mm)

80 -20
Displacement (mm)

1
60 1 -40
2
40 2 -60
3
3 -80
20
-100
0
W/O W -120

Twist
W/O W
20
A
Displacement (mm)

0
1
-20 B
2
-40 3

-60
Locations of corners 1-4 in Eqn. 1
-80

Fig. 6. Minimum and maximum displacements, and twist of the deck section when assembled
using sequences 1, 2 and 3

These results suggest that sequence 2 or 3 may be selected for assembly.

Employing a center support prevents excessive twist.
Carefully examining the plots in Fig. 5, wavy Z-displacement can be seen along
lines between longitudinals. As an example, Z displacement along line AB shown in
 Control of Car Decks Welding Deformation in RO-RO Ships 281

Fig. 6 in sequences 1, 2 and 3 with center support are plotted in Fig. 7 against the
longitudinal position X. Excessive wavy displacement may be observed. In Fig. 8, the
same displacements in the 3 sequences without center support are plotted. Here we can
see that the traditional sequence (sequence 1), although displaying the largest overall
displacement, the wavy displacement component is absent. This wavy displacement
exists with sequences 2 and 3.

Fig. 7. Z displacement along line AB in cases with center support

Fig. 8. Z displacement along line AB in cases without center support

The reason of this wavy displacement is the additional compressive stress imposed
on deck plating by the tendon force resulting from welding the lashing rails before
assembly of the deck section. Measures to reduce such tendon force are necessary.
282 J. Toyoshima et al.

5 Measures to Reduce Welding Tendon Force

5.1 Effect of Welding Procedure and Mechanical Straitening on Tendon
Force
Overall deflection of deck sections is mainly caused by large longitudinal shrinkage of
the welded joints joining stiffeners to plate panels. This shrinkage is also responsible
for increasing local deflection of plate panels between stiffeners and sometimes their
buckling. An inappropriate assembly sequence can magnify such action.
It is well known that longitudinal shrinkage of welded joints is primarily deter-
mined by welding net heat input per unit length of the weld line, Qnet. Therefore, to
reduce longitudinal shrinkage, reduction of welding heat input is an effective measure.
This can be achieved by replacing both sides continuous fillet welding by one side
continuous welding or both sides intermittent welding. As a post welding treatment to
reduce the tendon force, mechanical straightening may be effective.
In this section, the effectiveness of adopting alternative welding procedures and
mechanical straightening is examined using FE analysis. The model considered is a
plate on which a lashing rail is welded. The thickness of the plate is 7 mm and its width
and length are 600 mm and 1000 mm. The thickness of the lashing rail is 8 mm and its
height is 22 mm. Welding current, voltage and speed are taken according to the yard
practice as 285 A, 25.5 V and 7.5 mm/s. Heat input efficiency is assumed to be 80%.
Figure 9 shows the investigated welding procedures, “both sides continuous”, “one
side continuous”, “intermittent parallel” and “intermittent staggered” welding.

Fig. 9. Investigated welding procedures

The mechanical straightening process is models as shown in Fig. 10. The plate is
made to travel in the longitudinal direction through a pair of rollers whose radius is
200 mm, placed on both sides of the lashing rail while the front and rear of the plate is
hold in the flat position. The plate with the lashing rail bends down then up then back to
the flat position.
Welding deformation and deformation after mechanical straitening are computed
for the four welding procedures. To quantitatively clarify the influence of welding
procedure and mechanical straightening on the tendon force, the total longitudinal
 Control of Car Decks Welding Deformation in RO-RO Ships 283

plastic Strain is evaluated by integrating the plastic strain in the welding direction over
the cross-section normal to the weld line [1–3].

Fig. 10. Modelling the mechanical straightening process.

Distribution of total plastic strain in the weld direction along the weld line after
welding and after mechanical straightening are shown for both sides continuous
welding and intermittent parallel welding in Fig. 11. Here, total plastic strain is the
integration of plastic strain in the welding direction over the cross-section normal to the
weld line. It is clearly observed that the total plastic strain in the welding direction is
significantly reduced by mechanical straightening.

Fig. 11. Distribution of total plastic strain along the weld line before and after mechanical
straightening, (a) Both sides continuous, (b) Intermittent parallel.
284 J. Toyoshima et al.

Table 1 summarizes the computed results. By replacing both sides continuous

welding with one side continuous, intermittent parallel or intermittent staggered
welding, the longitudinal shrinkage can be reduced by nearly 30%. Mechanical
straightening reduces the longitudinal shrinkage to about 50% of its original value,
regardless of the welding procedure.

Table 1. Average plastic strain integrated over cross-section

Weld procedure Both side One side Parallel Staggered
Average heat input ratio 1.0 0.5 0.48 0.42
LID* as welded 3.18 2.18 2.12 2.04
LID* after straightening 1.69 1.22 1.30 1.17
Ratio 0.533 0.560 0.614 0.574
*LID: Longitudinal inherent deformation

For strength reasons, lashing rails must be welded using continuous welding on
both sides. Therefore, to reduce the tendon force and deck compressive stress,
mechanical straitening of the deck plating after welding the lashing rails is selected.
This excludes sequence 3, since it is not possible to carry out mechanical straitening
after welding the longitudinals. Further, for the rest of welded joints, it is recommended
to use one side continuous welding or intermittent welding where ever it is possible.

5.2 Effect of Mechanical Straitening on Deck Structure Welding

Deformation

With center support

Without center support

Fig. 12. Z-Displacement in sequence 2 with mechanical straitening.

 Control of Car Decks Welding Deformation in RO-RO Ships 285

Here only the reduction of the tendon force of welding the lashing rails is considered.
Table 1 indicates a reduction of 46.7% of the longitudinal inherent deformation (lon-
gitudinal shrinkage) when both sides continuous welding is used. Welding deformation
analysis is carried out again following sequence 2 and incorporating a reduction of only
40% of the said tendon force (longitudinal inherent deformation) due to mechanical
straining carried out immediately after welding the lashing rails. Deformed shapes
(Deformation scale 5:1) with the Z-displacement contours plotted on them in the cases
with and without center support are shown in Fig. 12. The deflection along line AB is
shown in Fig. 7 (sequence 2 ms) for the case with center support, and in Fig. 8 for the
case without center support. It may be seen that mechanical straitening did not have an
appreciable effect on overall deflection. However the wavy Z displacement has been
considerably reduced.

6 Conclusions

To control welding deformation of a car deck of a RO-RO ship, different assembly

sequences are considered. Also mechanical straitening of deck plating after welding the
lash rails is considered. Obtained results show that
1. Overall welding deformation can be greatly reduced by correctly designing the
assembly sequence. In the present case, fitting and welding the lasing rails first
reduced overall deflection of the deck section by almost 50%.
2. Method of supporting the deck section during welding may have a considerable
effect on deformation pattern. In the present case adding a central supporting block
after turnover of the deck section to the upright position almost eliminated deck
section twist.
3. Mechanical straitening of deck plating after welding the lashing rails first consid-
erably reduces the wavy deflection between longitudinals.
The yard has constructed a second vessel following sequence 2 applying
mechanical straitening just after welding the lashing rails. A third vessel is under
construction. Comparison with measurements will be presented in a following paper.

References
1. Murakawa, H., Luo, Y., Ueda, Y.: Prediction of welding deformation and residual stress by
elastic FEM based on inherent strain (first report) – Mechanism of inherent strain production.
J. Japan Soc. Naval Architects 180, 739–751 (1996)
2. Murakawa, H., Luo, Y., Ueda, Y.: Prediction of welding deformation and residual stress by
elastic FEM based on inherent strain (second report) - deformation and residual stress under
multiple thermal cycles. J. Japan Soc. Naval Architects 182, 783–793 (1997)
286 J. Toyoshima et al.

3. Hama, Y., Sano, M., Kawaji, M., Katayama, Y., Murakawa, H., Y Okumoto, Y., Rashed, S.:
Prediction and control of car ferry superstructure blocks deformation induced by welding and
straitening. In: International Conference on Computer Applications in Shipbuilding, Busan,
Korea (2013)
4. Vega, A., Escobar, E., Fong, E., Ma, N., Murakawa, H.: Analysis and prediction of parallel
effect on inherent deformation during the line heating process. Comput. Model. Eng. Sci.
90(3), 197–210 (2013). Blandon
5. Blandon, J., Osawa, N., Masanori, S., Rashed, S., Murakawa, H.: Numerical study on heat
straitening process for welding distortion of a stiffened panel. In: 25th International Ocean and
Polar Engineering Conference, Hawaii, USA (2015)
 Development of Basic Planning Support
System Using Marine Logistics Big Data
and Its Application to Ship Basic Planning

Kunihiro Hamada(&), Noritaka Hirata, Kai Ihara,

Dimas Angga Fakhri Muzhoffar, and Mohammad Danil Arifin

Graduate School of Engineering, Hiroshima University, Hiroshima, Japan

[email protected]

Abstract. Recently, the utilization of big data has been conducted in various
sectors. Big data is a large collection of datasets that is effective for under-
standing current situations and future development. In marine industries, big
data in the form of automatic identification system (AIS) data, ship principal
dimension, port facilities, shipping routes, and international trade data has
increased significantly. It is possible to acquire the insights of shipping industry
activities with the effective utilization of big data. Similarly, the dynamic situ-
ation of the global market requires aggressive ship development. Accordingly, it
is critical to realize certain ship specifications to satisfy market requirements.
Considering this background, the authors examined the utilization of marine
logistics big data to support ship basic planning, and developed a ship basic
planning support system that can predict the demand of ships using marine
logistic big data. However, in a previous study, a methodology to generate
effective information using big data was the focus, and the applicability to
practical ship design was not examined. Herein, the applicability of our ship
basic planning support system to practical ship design is examined. Hence, the
reliability of marine logistics big data (such as AIS data and ship movement
data) according to region are examined. Moreover, a ship allocation model is
developed in conformity with a marine logistic database. Applying a ship
allocation model, we can reproduce the actual ship allocation of targeted ship-
ping conditions and predict ships with high future demand. Therefore, effective
ship size, ship performance, and ship demand are examined by assuming some
future scenarios using the developed system, and some examples are presented.

Keywords: Big data
Ship basic planning
Deep learning
Marine logistics

Demand forecast

1 Introduction

In recent years, big data has created new opportunities for various industries. The term
“big data” was originally used to characterize large and complex datasets that humans
and conventional software cannot handle [1]. Big data is associated with three
dimensions: volume, variety, and velocity [2]. Volume represents the amount of data,
variety describes the diversity of data type, and velocity illustrates the rate of data

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 287–307, 2021.
https://doi.org/10.1007/978-981-15-4680-8_21
288 K. Hamada et al.

generated in a certain time period. Correspondingly, big data requires sufficient

methods and tools to gain the maximum value contained within [2, 3]. Further, the
effectiveness of big data has been recognized, and forecast accuracy reinforcement is
the most typical big data utilization. In logistics and transportation, big-data-based
analysis can benefit companies in traffic congestion avoidance, route enhancement, and
operation cost diminution [4, 5].
In the maritime industry, studies on big data utilization to improve ship mainte-
nance, ship operations, and ship construction have been executed [6–8]. Moreover,
marine logistics big data, such as port data, ship data, route data, international trade
data, and data provided by an automatic identification system (AIS), can currently be
used. However, studies regarding the utilization of such marine logistics big data for
ship basic planning or design are limited; they have been identified as important future
tasks [9].
Therefore, the authors examined the utilization of marine logistics big data to
support ship basic planning, and developed a basic planning support system using
marine logistics big data from a previous study [10–12]. Hence, ship basic planning
reflects various meanings. The authors focused on the decision of ship specifications
considering vessel operations in the latest market conditions. To realize this, our system
can generate ship allocation that matches to a real ship allocation. Therefore, through
executing the simulation by changing the ship specifications or future scenarios (such
as fuel price and trade volume between ports) for example, we can examine the
effective ship principal particulars and performance. However, in the previous study,
the methodology to generate effective information using big data was emphasized, and
the applicability to practical ship design was not examined.
Herein, the applicability of our ship basic planning support system to practical ship
design is examined. Hence, the reliability of marine logistics big data (such as AIS data
and port calling data according to region) are examined. Moreover, effective ship
performance and ship demand are examined by assuming some future scenarios.

2 Overview of Previous Study

In the previous study, we developed a marine logistics database (MLDB) and ship
allocation model [11, 12]. It was focused on an iron-ore bulk-carrier ship type and a
crude oil tanker as a foundation to construct the ship basic planning using marine
logistics big data. The outline of the MLDB, the ship allocation model, and demand
forecast simulations are discussed in this section (see Fig. 1).
 Development of Basic Planning Support System 289

STEP 1. Marine Logistics Database (MLDB) Establishment

INPUT DATA GENERATING DATABASE

MLDB
AIS Data SQL Server Express
Ships Data (Error Detection and Correction) Ship Operation Data
Ports Data Cargo Amount
Route Data Port Restriction
Trade Data

STEP 2. Ship Allocation Model Formation

INFLUENTIAL FACTORS OF MODELED SHIP ALLOCATION

MLDB CONFORMED TO
WORLD TRADE
ACTUAL SHIP ALLOCATION
Canal and Port Information
Bunker Price
International Trade Volume
Ship Specification Trend

SHIP ALLOCATION MODEL

Shipper Model Operator Model Shipowner Model

Transportation Estimation of Estimation
Demand Demand Result

Ship
Allocation

STEP 3. Simulations

PRESENT CONDITION

Refer to extracted data of NEW SHIP ALLOCATION

MLDB AND
Simulations using
NEW SHIP DEMAND
Ship Allocation Model
FUTURE SCENARIO

Bunker Price
Fuel Consumption of Ship
Canal and Port Expansion

Fig. 1. Overview of basic planning support system [9]

2.1 Marine Logistics Database

The MLDB is relational database that manages data pertaining to the AIS, ship
operation, ships, ports, routes, and trades [10–12]. All data are incorporated system-
atically to acquire beneficial insights for identifying a compelling ship specification.
The time span of ship operation data is from 2009 until 2019.
290 K. Hamada et al.

SHIP PORTS DATA

OPERATION DATA 1
Port Name
N IMO Number
Country Name
SHIPS DATA Departure / Arrival N
1 Latitude / Longitude
Port Name
IMO Number N
1 Annual Tonnage
MMSI Departure Date
Max. Deadweight
Ship Name Departure Country
AIS DATA (DWT)
Ship Type Departure Draught
IMO Number Max. Ship Length (m)
N Arrival Date
Movement Date Year Built Max. Ship Draught (m)
Length (m) Arrival Country
Latitude / Longitude Max. Ship Breadth (m)
Breadth (m) Arrival Draught
Indicated Speed N Cargo Type
Max. Draught (m) Route Number
Indicated Draught
Service Speed (kn) Operation Number

Main Engine Max. TRADE DATA

Power (kW)
ROUTE DATA Country Code
Deadweight 1 1
Route Number Country Name
(DWT)
Route Period
Gross Tonnage
(GT) Departure Country Report
Shipbuilder Departure Port Name Partner
Ship Operator Arrival Country Commodity Code

1 Ship Owner Arrival Port Name Trade Volume (ton)

One-to-Many
N Relationship Ship Manager Distance (nm) Cargo Type

Fig. 2. Data structure of marine logistics database

However, the AIS data are only available for 2017. For the liquefied natural gas
(LNG) carrier, the ship operation data contain 4818 operations. The data structure of
the MLDB is shown in Fig. 2.
AIS Data. AIS data comprise the ship name, International Maritime Organization
(IMO) number, movement timestamp, indicated sailing speed, indicated draught, and
latitude/longitude coordinates during 2017 obtained from IHS Markit. In this study, we
used the AIS data hourly.
Ship Operation Data. Ship operation data comprise the ship name, IMO number,
indicated draught, arrival/departure dates, and arrival/departure ports for 2009 until
2018 extracted from the Market Intelligence Network by IHS Markit/TheTradeNet [13].
Ships Data. Ship data comprise the ship name, IMO number, principal dimension,
operator, shipbuilder, age, and status extracted from Sea-web Ships [14].
Ports Data. Port data include the port name, principal dimension, longitude/latitude
information, and handled cargo types extracted from Sea-web Ports [15].
Route Data. Route data formed from arrival/departure port information within the AIS
data include distances between those ports collected from Sea-web Ports and IHS
Fairplay Ports & Terminal Guide [15, 16].
 Development of Basic Planning Support System 291

Trade Data. Trade data include the exporter/importer country, trade period, com-
modity code, trade quantity, and commodity/cargo name collected from the United
Nations Comtrade [17, 18].

2.2 Ship Allocation Model

According to [11, 12], the ship allocation model is developed to represent the actual
ship allocation. Three distinctive models are composed based on the MLDB: shipper
model, shipowner model, and operator model.
Shipper Model. The shipper model issues a request for cargo transportation between
two or more ports. It is defined for each type of cargo, i.e., iron ore, coal, crude oil, etc.
This model represents a viable method to reflect the international demand of cargo
transportation.
Shipowner Model. This model provides the estimation of sailing day, cargo volume,
and operation cost in associates to the demand of cargo transportation from the shipper
model. The ship owner is defined for each ship.
Operator Model. This model is designed to determine the most competent ship
allocation to fulfill cargo transportation demand performed based on the estimation
results of the shipowner model.

2.3 Limitation of Previous Study

The MLDB and ship allocation model have been developed in a previous study.
However, the following limitations were present in the previous studies:
• Developed MLDB did not cover worldwide ship operation data.
• Developed MLDB did not validate cargo volume comparison between ship oper-
ation data cargo estimation and statistics trade data.
• Evaluation of developed ship allocation model was insufficient.
Therefore, the above mentioned points are discussed herein.

3 Evaluations of Marine Logistics Big Data

In the MLDB, various data shown in Sect. 2.1 are integrated to construct models and
simulations [11, 12]. Although data cleaning was conducted, a thorough validation of
ship operation data has not been executed to justify its feasibility. To evaluate the
reliability of the MLDB, modification of draught data, and evaluations of draught data,
port calling data, and AIS data were executed in this study.

3.1 Evaluations of Draught Data

Draught changes are described in the AIS data and ship operation data. Owing to
loading/unloading activities, the draught value in the ship operation data tend to be
changed after a ship enters a certain port. However, draught data are input manually;
292 K. Hamada et al.

the changes might not be reflected in the draught data. To evaluate the draught changes,
the value of changed draught percentage can be obtained using Eq. 1. Further, Fig. 3
illustrates the average value of changed draught percentage of ports in each country
intended for crude oil tanker operations in 2014 and 2018.
nchanged
Changed Draught ð%Þ ¼
100% ð1Þ
ntotal:ops

where nchanged is number of ship operation data that contained changed draught and
ntotalops is the number of all ship operation data; Changed Draught(%) is represented in
percentage.

Fig. 3. Changed draught percentage in crude oil tanker operation

As shown in Fig. 3, the changed draught percentage of crude oil tanker operation
indicated improvement throughout 2014 to 2018. Although draught changes percent-
age in certain countries (such as Japan, China, Peru, Argentina, and Spain) are less than
85% in 2014, those values improved in 2018.
 Development of Basic Planning Support System 293

Moreover, the changed draught percentage is different according to country.

Draught data are considered effective for estimating cargo volume [11, 12]. However,
we should consider the characteristic of draught data, i.e., improvement in draught
information based on year, possibility of draught change that may vary depending on
country, and draught change dependency to ship size and/or ship operators.

3.2 Evaluations of Port Calling Data

Port calling data are managed in ship operation data, and are effective for understanding
ship operation. Port calling data are generated using AIS data and reports from agents
in ports; however, all port calling might not be included in the port calling data.
Therefore, the reliability of port calling data is evaluated using the statistic trade
data based on operations of LNG tankers, crude oil tankers, and iron-ore bulk carriers.
Meanwhile, some errors of draught data can be found in the ship operation data.
Draught data that are empty or larger than the ship specifications had been replaced
with the average draught data of the same-size type in the same route. The comparison
results of trade volume between the estimated and trade data intended for an LNG
tanker are shown in Fig. 4. Diagonal pattern bars indicate the estimation result in which
the cargo volume is estimated using Eqs. 2 and 3 [11].

dsail ðmÞ
di ¼ ð2Þ
dmax:i ðmÞ

where dsaili(m) is the sailing draught, dmaxi(m) is the ship maximum draught, and di is
the draught rate.

ðdi  0:2Þ
Vi ðtonÞ ¼ Cargoi ðtonÞ
ð3Þ
ð1  0:2Þ

where Cargoi(ton) is the ship maximum loaded cargo, di is the draught rate, and
Vi (ton) is the cargo volume.
Moreover, the error of estimation is shown in Table 1. From Fig. 4 and Table 1,
following characteristics are exhibited:
• In case of Qatar to Japan, cargo volume estimation using port calling data results in
an identical number as the statistics trade data, with an average error less than 10%
after 2011.
• In case of Qatar to China, cargo volume estimation using port calling data results in
an identical number as the statistics trade data, with an average error less than 10%
after 2014, except in 2016.
• The average errors of the estimation results for Malaysia to Japan and Malaysia to
China are larger than those for Qatar to Japan and Qatar to China. However, the
average error decreases in consecutive years.
294 K. Hamada et al.

Fig. 4. Comparison of cargo volume between port calling cargo estimation and statistic trade
data

Table 1. Average error of cargo volume estimation

Year
2009 2010 2011 2012 2013 2014 2015 2016 2017
Route
Qatar Japan 93.8% 64.4% 7.9% 2.3% 5.8% 8.7% 6.4% 1.0% 0.3%
Qatar China 100.0% 79.3% 38.8% 39.8% 22.7% 2.9% 3.2% 15.6% 7.3%
Malaysia Japan 81.3% 74.1% 64.5% 53.9% 63.2% 55.8% 62.2% 59.5% 57.1%
Malaysia China 76.9% 96.7% 63.6% 35.3% 67.2% 47.1% 56.9% 42.9% 63.7%

From the results above, the reliability of port calling data is considered as different
according to country.

3.3 Evaluations of AIS Data

Other than ship operation data, AIS data contain more detailed ship position data. AIS
data serve the ship position record continuously alongside the movement/activity of the
ship itself. The existence of AIS data must be elaborated to strengthen the interpretation
of ship operation data. The percentage of AIS data existence are collected using Eq. 4
in each country. Arranged in each country visited, the AIS data existence for a crude oil
tanker is visualized in Fig. 5.
 Development of Basic Planning Support System 295

Fig. 5. AIS data existence for crude oil tanker in 2017

nport
AIS Existenceð%Þ ¼
100% ð4Þ
timeðhÞ

where nport is the number of AIS data recorded in the port area, time(h) is the port
staying time in hour, and AIS Existence(%) is the percentage of AIS existence.
As shown in Fig. 5, the AIS data existence varies depending on country. However,
reliable 2017 data can be obtained in many developed countries.

4 Procedure to Develop Ship Allocation Model

4.1 Shipper Model Development

The shipper model issues a request for cargo transportation between two or more ports.
Cluster analysis is performed to generate the shipper model [11, 12]. Cluster analysis is a
method of defining the similarity in datasets, collectively group similar data, and
organize those groups into several classes (clusters). To construct the shipper model, the
following steps were executed. The developed shipper models are discussed in Sect. 5.
Extraction of Ship Operation Data from MLDB. Ship operation data are extracted
from the MLDB. The data extracted are managed in the form of a matrix between the
ports (P1, P2, …, Pn) and ships (S1, S2, …, Sn). An example is presented in Fig. 6(1).
Calculation of Euclidean Distance Between Ports. Figure 6(2) shows the stan-
dardized value of the example data. Subsequently, the Euclidean distance is calculated
using Eq. 5 for each listed port. The example result of the Euclidian distance is pre-
sented in Fig. 6(3).
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
d ðx; yÞ ¼ ð x1  y1 Þ 2 þ ð x2  y2 Þ 2 þ . . . þ ð xn  yn Þ 2 ð5Þ

where xn and yn are the standardized operation number ship n at ports x and y,
respectively, and d(x, y) is the Euclidean distance between x and y.
296 K. Hamada et al.

Clustering Using Hierarchical Clustering Method. The average linkage method is

adopted to measure the distance between two clusters. Figure 6(4) shows a represen-
tation of the average linkage method. This step is completed by utilizing Eq. 6.

1 X X
d ðC1 ; C2 Þ ¼ x1  C1 x2  C2 d ðx1 ; x2 Þ ð6Þ
jC1 jjC2 j

where C1 and C2 represent clusters 1 and 2, respectively; x1 and x2 represent ports 1 and
2, respectively; d(x1, x2) is the hierarchical distance between ports 1 and 2; d(C1, C2) is
the hierarchical distance between clusters C1 and C2.

Ship
Ship Arrival Data S1 S2 S3 S4 S5 S6
Port
extracted from MLDB
P1 0 0 1 0 2 0
(1) P2 0 0 1 0 3 2
P3 0 0 0 1 0 0
Manage Ship Arrival P4 5 1 0 0 0 0
Data in matrix form P5 2 0 0 0 0 0

Ship
S1 S2 S3 S4 S5 S6
Port
P1 -0.7 -0.7 0.65 -0.7 1.96 -0.7
(2) P2 -0.9 -0.9 0 -0.9 1.73 0.87
Calculate
P3 -0.4 -0.4 -0.4 2.24 -0.4 -0.4
Standardization
of Ship Arrival Data P4 2.19 0 -0.5 -0.5 -0.5 -0.5
P5 2.24 -0.4 -0.4 -0.4 -0.4 -0.4

Port
P1 P2 P3 P4 P5
Port
Calculate P1 1.71 3.94 4.04 3.94
Euclidean Distance (3) P2 1.71 4.08 4.21 4.08
of Ship Arrival Data P3 3.94 4.08 3.87 3.79
P4 4.04 4.21 3.87 0.49
P5 3.94 4.08 3.79 0.49

Define Port Cluster

of Ship Arrival Data P2 P3
using Average P1
Linkage Method (4)
P5
P4

Define Dendrogram
using Port Cluster of
Ship Arrival Data

(5)

P2 P1 P3 P5 P4

Fig. 6. Ship arrival data clustering process

 Development of Basic Planning Support System 297

Present Clustering Result Using Dendrogram. The average linkage method defines
the distance between ports. Ports that are located near each other are grouped as
clusters. Further, they are represented in a dendrogram, as shown in Fig. 6(5).

4.2 Shipowner Model Development

The shipowner model is defined for each ship. It estimates the shipment days, cargo
volume, and operation cost in response to demand for cargo transportation from
operators. Deep learning is used to estimate the draught rate, average sailing speed, and
port staying time in loading and ballast conditions.
Training Data. The operation data on each ship type that are extracted from the
MLDB are used as training data, as shown in Fig. 7(1). From the datasets, 75% data
were used for training and the remaining 25% were used for evaluation.

Fig. 7. Deep learning estimation process

298 K. Hamada et al.

Learning Model Generation. As shown in Figs. 7(2) and 7(3), the contents in the
input and output layers are set as follows:
• Input layer: DWT (ton), length (m), breadth (m), depth (m), draught (m), service
speed (knot), main engine power (HP), years of build, operator, shipbuilder, etc.
• Output layer: draught rate, average sailing speed, and port staying time for laden
and ballast conditions.

Shipment Days, Cargo Volume and Operation Cost Calculation. Results obtained
from deep learning were used to determine the number of shipment days, cargo vol-
ume, and operation cost, as shown in Fig. 7(4). Shipment days were determined by
considering the route distance, indicated speed, and port staying time. The cargo
volume was calculated based on the method developed by [19], and the operation cost
was calculated using the method from [20].

4.3 Operator Model Development

The operator model selects the finest ship allocation based on the estimation results of
shipowner model. The following procedures were performed to determine ship allo-
cation [11, 12].
Total Cost and Cargo Volume Calculation. As shown in Fig. 8(1), shipowners bid
for all shipment requests (Ships A–D) based on each selected route (Routes A–B) from
Shippers A and B. The cost per unit transport volume was calculated by considering the
total operation cost and total cargo volume in ton.
Standard Deviation Calculation. The standard deviations of the listed ships were
calculated for each route based on the cost per unit transport volume. The deviation
value was used as a benchmark to justify which ship was competent for transporting a
given cargo type through the designated route. An example of the calculated deviation
values are shown in Fig. 8(2).
Ship Assignment. Decides which ship to be chartered regularly built upon standard
deviation values. Ships with the highest standard deviation value are assigned to the
selected route shipping. Ship B is assigned to Route A1, as presented in Fig. 8(3).
Recalculate the Amount of Cargo Shipment Request. The remaining cargo to be
transported was determined by subtracting the demand for cargo transportation with the
ship capacity of the assigned shipping. Therefore, the cargo volume to ship was
updated and all steps in Table 2 are repeated until all cargo are successfully
transported.
 Development of Basic Planning Support System 299

Fig. 8. Ship allocation process

5 Case Studies

5.1 Shipper Model

In this section, an example of a shipper model intended for iron-ore transportation
between Australia and Japan is discussed. Data used in the clustering are iron-ore bulk
carrier operations in 2017. Clustering results of Australian and Japanese ports are
shown in Figs. 9 and 10, respectively, and the specifications are presented in Tables 2
and 3, respectively.
300 K. Hamada et al.

Distance

Port Hedland A1
Dampier
A2
Port Walcott
Esperance
A3
Parker Point

Fig. 9. Clustering result of iron ore ports in Australia

Distance

Chiba
Mizushima
J1
Fukuyama
Kawasaki
Kashima
Oita
J2
Kisarazu
Kure
J3
Higashi-Harima
Tobata
Wakayama J4
Nagoya

Fig. 10. Clustering result of iron ore ports in Japan

Table 2. Iron ore ports limitation in Australia

Cluster Port name Port limitation
DWT L (m) B (m) d (m)
A1 Port Hedland 340,000 335 60 19.5
A2 Dampier 260,000 330 55 19.2
Port Walcott 250,000 340 55 19.5
A3 Esperance 220,000 300 50 18.5
Parker Point 220,000 300 50 18.0

From the figures of both exporter ports and importer ports, several clusters were
defined based on the company ownership of the port as well as the port limitations.
Therefore, the clustering result matches the actual condition. Moreover, as discussed in
Sect. 3.2, the cargo volume of each operation is estimated using the MLDB. Using the
estimation result, the trade volume between those clusters can be calculated as pre-
sented in Table 4.
 Development of Basic Planning Support System 301

Table 3. Iron ore ports limitation in Japan

Cluster Port name Port limitation
DWT L (m) B (m) d (m)
J1 Chiba 220,000 300 50 18.0
Fukuyama 220,000 300 50 18.0
Kawasaki 260,000 340 50 18.0
Mizushima 220,000 340 50 18.0
J2 Kashima 300,000 340 60 19.0
Oita 400,000 450 60 25.0
Kisarazu 300,000 330 60 19.0
J3 Tobata 160,000 327 43 16.0
Wakayama 160,000 300 43 14.0
Nagoya 110,000 300 43 16.0
J4 Kure 276,000 360 45 18.0
Higashi-Harima 180,000 330 47 17.0

Table 4 shows that JFE steel (cluster J1) exhibits a closer relationship with BHP
Bili-ton (Cluster A1), although NSSMT (cluster J2 and J3) exhibits a closer relationship
with Rio Tinto. Those conditions match the actual partnerships stated in [22] and [23].
From these results and using the port calling data and clustering analysis, we can define
the shipper model that represents the cargo transportation between two or more ports.

Table 4. Iron ore cargo distribution between Australia and Japan clusters

Country Australia
Cluster A1 A2 A3
Company BHP Biliton Rio Tinto
Port Name Port Hedland Dampier Port Walcott Esperance Parker Point
Chiba
Fukuyama 15.6 MT 11.6 MT 3.3 MT
J1 JFE Steel
Kawasaki (19.4%) (14.4%) (4.1%)
Mizushima
Kashima
Nippon Steel 8.6 MT 26.2 MT 3.5 MT
J2
Japan

Oita
& Sumitomo (10.7%) (32.5%) (4.3%)
Kisarazu
Metal
Tobata
Terminal 0 MT 2.9 MT 0 MT
J3 Wakayama
(NSSMT) (0.0%) (3.6%) (0.0%)
Nagoya
Nisshin Steel Kure 3.6 MT 3.6% 1.6 MT
J4
KOBELCO Higashi-Harima (4.5%) (4.5%) (2.0%)
302 K. Hamada et al.

5.2 Shipowner Model

In this section, an example of a shipowner model intended for a crude oil tanker is
presented. As discussed in Sect. 4.2, the shipowner model estimates the draught rate,
average sailing speed, and port staying time using the deep-learning method. Hence,
the accuracy of deep learning is affected by the training data. Meanwhile, ship oper-
ation is affected by the shipping fare and oil price that change intensely (see Fig. 8). To
develop a robust estimation method, we examine the relation between the duration of
training data and the estimation error. The results are shown in Fig. 12 and Table 5. As
shown in Fig. 12, the correlation coefficient (R) between the actual average sailing
speed and deep learning estimation is 0.96. Therefore, the deep-learning estimation is
reliable.

35 600
30 500
25
400
20
300
15
200
10
5 100

0 0
2009 2010 2011 2012 2013 2014 2015 2016 2017
Shipping Rate (US$/ton) HFO Price (US$/ton)

Fig. 11. Changes in the shipping fare and HFO price [24, 25]

30
R = 0.96
Deeplearning Estimation (knots)

25

20

15

10

0
0 5 10 15 20 25 30
Actual Average Sailing Speed (knots)

Fig. 12. Average sailing speed (knots) estimation result

 Development of Basic Planning Support System 303

Table 5. Error of average sailing speed (knots) estimation

No. Training Test
2013 (High fuel price) 2014 (Middle fuel price) 2016 (Low fuel price)
1 2014–2015 0.91 0.74 0.9
2 2014–2016 0.95 0.82 0.86
3 2013–2017 0.79 0.83 0.89

As shown in Fig. 11, year 2013 is considered as the higher limit of fuel price, while
year 2016 is the lower limit. Further, it can be concluded from Table 5 that training
dataset 1 is fair for the middle fuel price, but poor in covering the high and low fuel
price periods. Meanwhile, training dataset 2 is fair for the middle and low fuel price
periods. Meanwhile, training dataset 3 is fair for all fuel price periods. Therefore,
training datasets 3 is selected owing to its characteristics.

5.3 Simulation to Validate Ship Allocation

To confirm the effectiveness of the ship allocation model, we simulated the iron-ore bulk-
carrier ship allocation between Australia and Japan in 2014. In this simulation, ports in
Japan are divided into four clusters, i.e., Shipper A (Chiba, Fukuyama, Kawasaki,
Mizushima) matches ports of JFE Steel in Japan, Shipper B (Kashima, Kisarazu, Oita)
matches deep draught (more than 19 m) ports of Nippon Steel Sumikin (NSSMT) in Japan,
Shipper C (Nagoya, Tobata, Wakayama) matches shallow draught (less than 19 m) ports
of Nippon Steel Sumikin (NSSMT) in Japan, and Shipper D (Others) matches ports owned
by other companies in Japan, i.e., KOBELCO, Nissin Steel, etc.
Subsequently, iron-ore bulk carriers that were used in these routes in 2014 were
allocated using the developed model. The simulation result is compared with the actual
ship allocation in Fig. 13 and Table 6.

Shipper A Shipper B
(Chiba, Fukuyama, Kawasaki, Mizushima) (Kashima, Kisarazu, Oita)
150 150
Actual Simulation Actual Simulation
Operations

Operations

100 100

50 50

0 0
10 17 21 23 25 30 10 17 21 23 25 30
Ship Size (104 DWT) Ship Size (104 DWT)

Shipper C Shipper D
(Nagoya, Tobata, Wakayama) (Others)
150 150
Actual Simulation Actual Simulation
Operations

Operations

100 100

50 50

0 0
10 17 21 23 25 30 10 17 21 23 25 30
Ship Size (104 DWT) Ship Size (104 DWT)

Fig. 13. Comparison of operation number between simulation result and actual ship allocation
304 K. Hamada et al.

Table 6. Error of operation number in simulation

Ship Size (10 4 DWT)

Shipper Operation
10 17 21 23 25 30
Actual 7 107
A Simulation 2 112
Error 71.4% 4.7%
Actual 28 40 45 9 4
B Simulation 33 35 45 9 4
Error 17.9% 12.5% 0.0% 0.0% 0.0%
Actual 27
C Simulation 27
Error 0.0%
Actual 28
D Simulation 28
Error 0.0%

The following characteristics are reflected from Fig. 13 and Table 6:

• For Shipper A, the ship size of 210,000 DWT was primarily utilized. This tendency
is shown in both simulation and actual scenario.
• For Shipper B, various ship sizes were utilized, but the majority were 170,000
DWT, 210,000 DWT, and 230,000 DWT.
• For Shipper C, the ship size of 100,000 DWT was utilized. Simulation results
matched those of actual ship allocation.
• For Shipper D, the ship size of 170,000 DWT was utilized. Simulation results
matched those of actual ship allocation.
From the results above, the simulations are considered capable of reproducing the
actual ship allocation.

5.4 Simulation to Examine Effect of Fuel Cost Improvement

In this section, we pay attention to the crude oil tanker which operates from Saudi
Arabia and United Arab Emirates to Japan and Korea in 2016. The typical principal
particular for this ship is shown in Table 7.
Because marine fuel is the ship primary operation cost [21], fuel efficiency is
defined as a performance parameter to execute further simulations. To inspect the
domination of fuel efficiency on ship demand and describe future development fore-
casts, we simulated increasing fuel efficiencies by 5%, 10%, and 15% from changing
the fuel price. A ship with no fuel efficiency change is defined as S0. Ships with fuel
efficiency increases of 5%, 10%, and 15% are denoted by S5, S10, and S15, respectively.
Table 8 shows the simulation result of ship allocation on the intended route (Saudi
Arabia and United Arab Emirates to Japan and Korea) after modifying the fuel effi-
ciency. Cargo volume between the countries is set based on the result of 2016.
 Development of Basic Planning Support System 305

Ships generated by simulations based on the lowest fuel cost in 2016 would not be
able to replace the existing ships. When the fuel cost is approximately US$ 400/t, the
demand for vessels with a 10% improvement in fuel efficiency increases, even con-
sidering the depreciation cost.

Table 7. Principal particulars Table 8. Fuel efficiency simulation result

of the target ship HFO price Ship Number of Number of
Ship S0 (US$/t) allocated route operation
DWT (t) 311,061 293.9 (2016) S5 0 0
S10 0 0
L (m) 333.0
S15 1 18
B (m) 60.0
347.5 (2017) S5 0 0
d (m) 21.0
S10 0 0
v (kn) 15.5 S15 2 24
408.5 (2018) S5 0 0
S10 2 18
525.6 (2014) S5 2 18
S10 3 36
569.6 (2013) S5 2 24
S10 5 71

Furthermore, in a high-fuel-cost condition of more than US$ 500/t, the simulation

shows that the demand for ship increases even in the case of 5% fuel efficiency
enhancement.
Meanwhile, the cargo transportation demand of new ships with improved fuel
efficiency by 10% has significantly increased, as shown from the simulation based on
the highest fuel price condition in 2013. This proved that it is possible to study the
effect of fuel cost fluctuation on the demand for fuel efficiency improvement in vessels.
Further, it can be concluded that fuel price fluctuation affected the ship specification
demand.

6 Conclusion

Herein, a support system for basic ship planning using marine logistics big data was
discussed. The following conclusions can be drawn from this study:
• The reliability of some marine logistics big data, i.e., draught data, port calling data,
and AIS data were evaluated. The results indicated that the reliability and coverage
of the draught data, port calling data, and AIS data were different depending on the
country and so on. For developed countries, the estimated cargo volume obtained
from draught data and port calling data was the same as the statistics trade data after
2014.
306 K. Hamada et al.

• Using the data extracted from marine logistics big data, the ship allocation model
composed from the following three specified models was developed: shipper model,
shipowner model, and operator model. The ship allocation using the developed
model generally agreed with the actual ship allocations.
• By inputting the future scenario into the simulation, an expected ship in future
demand could be acquired from the categories of ship size, supply–demand balance,
ship fuel efficiency, and fuel price correlation of ship allocation proposed in this
study.

Acknowledgment. This work was supported by JSPS KAKENHI Grant Number 16H04602.
Authors would like to express their gratitude for the support.

References
1. Snijders, C., Matzat, U., Reips, U.: ‘Big data’: big gaps of knowledge in the field of internet
science. Int. J. Internet Sci. 7(1), 1–5 (2012)
2. Press, G.: A very short history of big data. https://www.forbes.com/sites/gilpress/2013/05/
09/a-very-short-history-of-big-data/. Accessed 04 Mar 2019
3. Mirović, M., Miličević, M., Obradović, I.: Big data in the maritime industry. Naše more
65(1), 56–62 (2018)
4. Hossein, H., Emmanuel, S.S.: Forecasting with big data; A review, pp. 5–19. Springer
(2015)
5. Ando, H.: Optimizing fleet performance with smart ship management. The IoT Show 2015
Asia, Monohakobi Technology Institute (MTI) NYK Group (2015)
6. Yoshida, T.: Utilizing big data and the internet of things in shipping. In: Sea Japan 2016
MTI, pp. 1–32 (2016)
7. Aoyama, K., Liu, J., Hiro, Y.: Monitoring system for advanced shipbuilding construction
management: extracting and utilizing monitoring data by considering the reliability of
monitored data. In: ICCAS, pp. 143–154 (2015)
8. Perera, L.P.: Statistical filter-based sensor and DAQ fault detection for onboard ship
performance and navigation monitoring systems. In: IFAC, pp. 323–328 (2016)
9. Wada, Y., Hamada, K., Hirata, N., Yui, I., Arifin, M.D., Shohei, M., Kazutaka, S.: A study
on system dynamics model for the demand forecasting of shipbuilding. In 12th IMDC(2),
pp. 167–182 (2015)
10. Arifin, M.D., Hamada, K., Hirata, N., Ouchi, K.: Development of marine logistics database
using automatic identification system (AIS) and statistical data. In: Proceedings of TEAM
2016 (2016)
11. Arifin, M.D., Hamada, K., Hirata, N., Ihara, K., Koide, Y.: A study on the support system of
ship basic planning using marine logistics big data. In: ICCAS 2017 (2017)
12. Ihara K., Hamada, K., Hirata, N., Arifin, M.D.: Development of the basic planning support
system for tanker using marine logistics big data. In: Proceedings of TEAM 2018 (2018)
13. Market Intelligence Network, MINT. https://www.marketintelligencenetwork.com/. start
using from 2015-present
14. IHS Maritime & Trade, Sea-web Ships. http://maritime.ihs.com/. start using from 2015-
present
15. IHS Maritime & Trade, Sea-web Ports. http://maritime.ihs.com/. start using from 2015-
present
 Development of Basic Planning Support System 307

16. IHS Fairplay, Ports & Terminals Guide version 3.1, CD-ROM (2013)
17. United Nations (UN) commodity trade statistics database. https://comtrade.un.org/data/. start
using from 2015-present
18. International Gas Union (IGU): 2018 World LNG Report. In: 27th World Gas Conference
(2018)
19. Koide, Y., Hamada, K., Hirata, N., Arifin, M.D., Ouchi, K.: Research on basic planning
support of ships using maritime logistics big data. In: Proceedings of JASNAOE(23) (2016,
in Japanese)
20. Oizumi, U., Aoyama, K.: Assessment method of CO2 reduction effect of new technology
and international regulation on maritime logistics. In: 6th EcoDesign, CD-ROM (2009)
21. Journée, J., Meijers, J.: Ship routing for optimum performance. Transactions IME, Rapport
0529-P (1980)
22. Rio Tinto, Fourth quarter operations review. https://www.riotinto.com/documents/RT_
Japan_factsheet_EN.pdf. Accessed 7 Mar 2019
23. Australian Competition & Consumer Commission, Draft Determination. https://www.accc.
gov.au/system/files/public-registers/documents/D05%2B76270.pdf. Accessed 7 Mar 2019
24. Japan Long Course Ferry Service Association (JLCFSA), Fuel oil price transition chart.
http://www.jlc-ferry.jp/nenryouyu.html. Accessed 7 Mar 2019. (in Japanese)
25. BP: Approximate conversion factors: BP statistical review of world energy (2017)
 Evaluation Method for the Maximum Wave
Load Based on AIS and Hindcast Wave Data

Masayoshi Oka(&), Tomoki Takami, and Chong Ma

National Maritime Research Institute, Mitaka-shi, Tokyo, Japan

[email protected]

Abstract. Improvement of the estimation accuracy of the wave load is nec-

essary to realize a rational ship structure design. However, a rational method has
not been established to determine the operation limit which influences the wave
loads. In this study, the ship encounter sea condition including the operation in
wave such as wave direction and ship speed, and the wave loads acted on actual
ships were investigated using AIS data and wave hindcast data, and following
were confirmed.
– The maximum significant wave height in North Atlantic sea by wave
hindcast is not significantly differ from IACS Recommendation No. 34.
– Significant wave height that ships encountered is reduced by about 10–20%
owing to the seamanship of storm avoidance.
– The maximum wave loads of individual ships were estimated for container
ships, tankers and bulk carriers using the strip method. It is clarified that the
vertical bending moment in waves, which is estimated to have acted in the
actual sea conditions, is almost below the design load.

Keywords: AIS data
Hindcast wave data
Ship encounter sea condition

Long-term prediction
Vertical bending moment

1 Introduction

Improvement of the estimation accuracy of the wave load is necessary to realize a

rational ship structure design. However, a rational method has not been established to
determine the operational limit which influences the wave loads.
In the current ship design, the wave scatter diagram for the North Atlantic (Inter-
national Association of Classification Societies, Recommendation No. 34 (IACS Rec.
34) [1]) has been used as standard. Because IACS Rec. 34 was created based on Global
Wave Statistics (GWS) in 1986 [2] and it is old, and recent climate change are not
reflected, a project team was set up at IACS, and a review of the standards is under
consideration [3]. On the other hand, compared to the 1980s, the accuracy of weather
and sea condition prediction has been improved, and storm avoidance has been per-
formed generically, and the wave ship encountered may be more relaxed than at the

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 308–320, 2021.
https://doi.org/10.1007/978-981-15-4680-8_22
 Evaluation Method for the Maximum Wave Load 309

time of 1986. In the previous researches, because it could not clear how often storm
weather avoidance has been conducted and how much the encounter waves have been
changed, the design and regulation that considered the storm avoidance could not come
true. However, in recent years, the mandatory installation of the Automatic Identifi-
cation System (AIS) specified in the “SOLAS Convention Chapter V, Regulation 19”
has come to transmit the data including location information from ships of 300 GT or
more engaged in international voyages. In addition, because several company who
collects ship location information by satellite [4] appeared and data collection pro-
gressed, ship location information came to be clarified gradually. Moreover, the
technology of wave hindcasting has also been developed, and it has become possible to
obtain the value of sea state parameter (e.g. significant wave height) with high density
and high accuracy. By using the AIS data and the wave data, the sea condition which
ships encountered has been clarified. Furthermore, it is expected to realize a safer and
more economically rational ship design.
Concerning about ship encounter sea condition, Soares [5] proposed a probabilistic
model for ship navigation in heavy sea based on the interviews with ship operators and
carried out a Monte Carlo simulation. And he showed that the effect of the storm
avoidance on wave load is up to 25%. Vettor et al. [6] analyzed wave data based on the
ship report and show that a 20% wave height reduction effect can be obtained by the
storm avoidance. From these previous studies, it has been found that the navigation for
storm avoidance contributes to the reduction of encounter wave height and wave load.
However, these studies could not able to be verified because the voyage data required
for quantitative evaluation is insufficient. Authors [7] analyzed the sea state which ship
encountered using AIS (Automatic Identification System) data and the hindcast wave
data. The effects of the weather routing and/or storm avoidance in North Atlantic and
the maximum wave load acted on actual container ships was evaluated.
In this study, the effect of ship type and ship size on encounter wave height was
analyzed. And the maximum wave load acted on actual ships for various types of ships,
container ship, oil tanker and bulk carrier were estimated. Moreover, the maximum
wave load of individual ships were predicted, and compared with the design load
required by IACS [8].

2 Probability of Sea-State

2.1 Global Ocean Wave Hindcast Database

In this research, Global Ocean Wave Hindcast Database provided by the Japan Weather
Association [9] based on wave hindcasting using the WAMcycle4 (JWA revision)
third-generation ocean wave model were utilized.
Data periods and areas are shown in Table 1 and Fig. 1. The period is one year, and
the area is the North Atlantic area specified in IACS Recommendation No. 34 [1].
310 M. Oka et al.

Table 1. Subjected period and area

Period 2016/9/30/01:00–2017/10/1/0:00
Time interval 1h
Area (ref. Fig. 1) (1) N40.00–60.00, W10.00–60.00
(2) N40.00–45.00, W60.00–70.00
Grid interval 0.5° (50 km)
Number of grid 4,201 points

Fig. 1. Subjected area (corresponding to IACS recommendation No. 34/Area 8, 9, 15, 16)

2.2 Exceedance Probability of Significant Wave Height

The exceedance probability of significant wave height (Hs) in the North Atlantic was
calculated based on the hindcast wave data. The results are shown in Fig. 2. The
significant wave height of the probability P = 10−5, which corresponds to the maxi-
mum value during 25 years, is Hs = 16 m. It is confirmed that remarkable difference
does not occur.

1.0E+00
IACS/Rec.34
1.0E-01 Hindcast (JWA)
Probability of exceedance

1.0E-02

1.0E-03

1.0E-04

1.0E-05
0 2 4 6 8 10 12 14 16 18
Significant wave height [m]

Fig. 2. Significant wave height (Hs) in North Atlantic area

 Evaluation Method for the Maximum Wave Load 311

3 Ship Encountered Sea State Based on the Locations

of Ships

In this chapter, the ship encountered sea state based on the locations of ships were
created. And, effects of ship type and ship size on encountered sea state were analyzed.

3.1 AIS Data

AIS data are utilized in order to get the information of ship locations. In this study,
NAPA Fleet Intelligence [4] provided by NAPA was utilized. The outline of AIS data
is shown in Table 2 the types of ships included in the AIS data are shown in Fig. 3.
Although the period of AIS data is one year, it is one month apart from the global
database, so 11 months from 2016/11 to 2017/9 are considered as the evaluation
period. 13,496 ships are included in the NAPA Fleet Intelligence. There are 9,050 ships
that had passed through the subjected area in Fig. 1 during the subjected period. Since
the AIS data is obtained at any time, the average value for one hour is used for analysis.

Table 2. Summary of AIS data provided by Napa Ltd.

Period 2016/11/1 0:00–2017/10/31 23:59
Number of ships 13,496 (Among them, 9,050 ships have navigated through the subjected)
Ship profile data IMO Number, Ship Name, Ship type, Length, DWT, Service speed, etc.
Voyage data Time stamp, Latitude, Longitude, Course over ground, speed etc.

Ship_numbers
Ro-Ro Ship
Ferry
Passenger Ship ¥N

LNG Tanker
LPG Tanker
Bulk Carrier
General Cargo
Ship

Vehicle Carrier

Chemical
Tanker Container Ship

Oil Tanker
Mul purpose
Ship

Bulk Carrier Container Ship Mul purpose Ship

Oil Tanker Chemical Tanker Vehicle Carrier
General Cargo Ship LPG Tanker LNG Tanker
Passenger Ship Ferry Ro-Ro Ship
\N

Fig. 3. Detail of the ship kind included in the AIS data

312 M. Oka et al.

3.2 Ship Encountered Significant Wave Height

Analyzing the AIS data and wave data, the exceedance probability of the encounter
significant wave height was determined. The value which located the nearest from ship
position among the grid points in Fig. 1 was applied. The results are shown in Fig. 4.
The significant wave height at P = 10−5 is Hs = 16 m for wave data (Hindcast) and
Hs = 13 m for ship encountered seas (AIS + Hindcast). In terms of ship type, con-
tainer ships (CS) have the highest reduction rate, followed by bulk carriers (BC) and oil
tankers (OT). Oil tankers are thought to be affected by the a little number of ships
which encountered heavy sea, because the encountered significant wave height sud-
denly rises in the low probability.

1.0E+00 IACS/Rec.34
Hindcast
AIS+Hindcast (All)
1.0E-01 AIS+Hindcast(CS)
Probability of exceedance

AIS+Hindcast(OT)
AIS+Hindcast(BC)
1.0E-02

1.0E-03

1.0E-04

1.0E-05
0 2 4 6 8 10 12 14 16 18
Significant wave height [m]

Fig. 4. Probability of exceedance of significant wave height

Fig. 5. Spatial distribution of the maximum significant wave height

 Evaluation Method for the Maximum Wave Load 313

NEWFOUNDLAND

(a) Container ships

NEWFOUNDLAND

(b) Oil tanker

NEWFOUNDLAND

(c) Bulk carrier

Fig. 6. Positions of the ships obtained from the AIS data

314 M. Oka et al.

It is assumed that the cause of the ship encounter sea state being calmer than the
wave data is the effect of the rough weather avoidance by ships. The spatial distribution
of the maximum significant wave height for one year is shown in Fig. 5, and the
positions of the ships obtained from the AIS data are shown in Fig. 6(a) to (c).
Comparing the Fig. 5 and Fig. 6, it can be seen that the number of ships (especially
container ships) passing near W40-N55 (Point A in Fig. 5) and W25-N45 (Point B in
Fig. 5), where heavy seas are likely to occur, are small. One of the reasons for the
reduction of the significant wave height is considered that the amount of traffic thor-
ough such heavy sea area is small.

3.3 Effect of Ship Size

The exceedance probability of significant wave heights which ship encountered were
compared by dividing into three category of L = 200 m or less, 200 to 300 m, and
300 m or more. The results are shown in Fig. 7(a) to (d). Bulk carriers is classified into
the following two classes, because only one ship categorized into L = 300 m or more.
In all types of ships (Fig. 7(d)), it is understood that the significant wave height is low
when L = 300 m or more. Comparing by ship type, container ship is most affected by
length, the significant wave height is large for L = 200 m or less. Oil tankers are less
affected by length. Bulk carrier tends to have a high significant wave height as it be
longer, contrary to the tendency of other ships.

CS OT
1.0E+00 1.0E+00
L < 200m L < 200m
1.0E-01 1.0E-01
Probability of exceedance

Probability of exceedance

200m < L < 300m 200m < L < 300m

300m < L 300m < L

1.0E-02 1.0E-02

1.0E-03 1.0E-03

1.0E-04 1.0E-04

1.0E-05 1.0E-05
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18
Significant wave height [m] Significant wave height [m]

(a) Container ship (b) Oil tanker

BC All
1.0E+00 1.0E+00
L < 200m L < 200m
1.0E-01 1.0E-01
Probability of exceedance

Probability of exceedance

200m < L < 300m

200m < L < 300m
300m < L
1.0E-02 1.0E-02

1.0E-03 1.0E-03

1.0E-04 1.0E-04

1.0E-05 1.0E-05
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18
Significant wave height [m] Significant wave height [m]

(c) Bulk carrier (d) All kind of ships

Fig. 7. Ship encountered wave height comparing to each ship size

 Evaluation Method for the Maximum Wave Load 315

4 Wave Load Considering with Actual Encounter Sea State

Wave load estimation was performed using the actual encounter sea state. Performing
long-term predictions for vertical bending moment (VBM), the effect of encounter sea
states on wave load were confirmed. Subsequently, the wave load estimations for the
individual ships were performed using the actual encounter sea conditions which
includes steering and speed of each ship.

4.1 Long-Term Prediction

4.1.1 Response Amplitude Operator (RAO)
Using the linear strip method, the response amplitude operator (RAO) of the VBM at
the mid-ship were determined. Three types of ships, container ship (CS), tanker
(OT) and bulk carrier (BC), were subjected, and loading were assumed to be full load.
Example of the RAOs are shown in Fig. 8(a), (b). The tanker and bulk carrier used the
same RAO. The vertical axis means a non-dimension value of VBM, and the horizontal
axis means the ratio of the wavelength k to the ship length L. As the encounter wave
direction v, 0° means following seas and 180° means head seas. The Froude number Fn
is a non-dimension number of ship speed V.

RAO RAO
0.025 0.025
χ= 0deg
χ= 0deg
0.02 χ= 30deg 0.02
χ= 30deg
χ= 60deg
0.015 0.015 χ= 60deg
Amp. /ζ

Amp. /ζ

χ= 90deg
χ= 90deg
0.01 0.01
χ= 120deg
χ= 120deg

0.005 χ= 150deg 0.005 χ= 150deg

χ= 180deg χ= 180deg
0 0
0.00 0.50 1.00 1.50 2.00 2.50 0.00 0.50 1.00 1.50 2.00 2.50
λ/L λ/L

(a) Container ships (b) Oil tanker and bulk carrier

Fig. 8. Response amplitude operator (RAO) of VBM (Fn = 0.06)

4.1.2 Short Term Response

The short-term response function was determined by superimposing the RAO and the
wave spectrum [10]. IACS Recommendation No. 34 was applied for wave spectrum
and wave directional dispersion. An example of a short-term response function is
shown in Fig. 9(a), (b). The vertical axis means the non-dimension value of the
standard deviation (R) of the VBM, and the horizontal axis means the zero-crossing
wave period (Tz).
316 M. Oka et al.

0.004 0.004

0.0035 0.0035

0.003 0.003
χ=0deg χ=0deg
R/rho*g*L^2*B*H

R/rho*g*L^2*B*H
0.0025 0.0025
χ=30deg χ=30deg
0.002 χ=60deg 0.002 χ=60deg
χ=90deg χ=90deg
0.0015 0.0015
χ=120deg χ=120deg
0.001 χ=150deg 0.001 χ=150deg

0.0005 χ=180deg 0.0005 χ=180deg

0 0
0 5 10 15 20 0 5 10 15 20
Tz(s) Tz(s)

(a) Container ships (b) Oil tanker and bulk carrier

Fig. 9. Short term response of VBM (L = 150 m, Fn = 0.06)

4.1.3 Long Term Prediction

Long-term prediction was performed using the short-term response and the probability
distributions of wave (wave scatter diagrams), and the maximum value of the excee-
dance probability Q = 10−8 (corresponding to 25 years) was obtained. Long-term
prediction were performed for each ship types and ship sizes. The ship speed and
encounter wave direction were assumed to be V = 0.75 Vs (Vs: navigation speed),
uniform wave direction (All Headings) according to the guidelines of the classification
society [11]. Vs used the same value for each type of ship. The average value of the Vs
of the subjected ships was adopted. Container ship was set to Vs = 12.0 m/s (about 23
knot), and the oil tanker and bulk carrier was set to Vs = 7.6 m/s, (about 14 knot). The
results of long-term predictions are shown in Fig. 10. It can be seen that the maximum
load is reduced by 10–20% due to considering the actual encounter seas.

VBM of container ship in North Atlan c VBM of OT & BC in North Atran c

0.0016 0.0016

0.0014 0.0014

0.0012 0.0012
VBM / rho*g*L^3*B

VBM / rho*g*L^3*B

0.001 0.001

0.0008 0.0008

0.0006 0.0006

0.0004 0.0004

0.0002 0.0002

0 0
0 50 100 150 200 250 300 350 400 450 0 50 100 150 200 250 300 350 400 450
L (m) L (m)

Longterm pred. 75%Vs, Allheadings, Hindcast Longterm pred. 75%Vs, Allheadings, Hindcast
Longterm pred. 75%Vs, Allheadings, AIS+Hindcast Longterm pred. 75%Vs, Allheadings, AIS+Hindcast

(a) Container ship (b) Oil tanker and bulk carrier

Fig. 10. Maximum wave load (VBM) correspond to Q = 10−8

 Evaluation Method for the Maximum Wave Load 317

4.2 Maximum Wave Load of Individual Ships

The maximum wave load of individual ships was estimated using AIS data and wave
data. In this study, it was assumed that the VBMs are proportional to L2B. An NMRI
post-Panamax container ship with L = 283.8 m and an NMRI bulk carrier with
L = 278.0 m are used as a standard model. The oil tanker and bulk carrier used the
same model. In order to estimate the wave load depending on the actual ship speed,
response functions were prepared at intervals of Froude number DFn = 0.02 from
V = 0 to Vs. A linear load was assumed for the wave height.
The procedure for estimating the maximum wave load of individual vessels is
shown below.
(1) Obtaining the time histories of actual sea conditions that the individual ships en
countered from AIS data and wave data
(2) Finding the standard deviation (R) of the VBM correspond to the actual sea con-
ditions (Hs, Tz, v, V) using short term response function shown in Fig. 9,
(3) Finding the worst sea state (Worst Sea State) where R is the largest in the time
history of each ship.
(4) Determine the 1/1000 maximum expected value (VBMMAX) in the worst sea state.
VBMMAX can be calculated by the following equation using Rwss, assuming that
the short-term response is a Rayleigh distribution.

VBMMAX ¼ 3:87Rwss ð1Þ

The maximum wave load of each individual ships for about one year was estimated
according to the above procedure. The results are shown in Fig. 11. And, the histogram
and probability distribution of the ratio of the estimated load to the design load required
by structural rule [8, 12] are shown in Fig. 12. The maximum load of the individual
ship estimated based on the AIS data and the wave data is approximately below the
design load is confirmed.
318 M. Oka et al.

VBM of container ship in North Atlan c

0.0016

0.0014

0.0012

VBM / rho*g*L^3*B
0.001

0.0008

0.0006

0.0004

0.0002

0
0 50 100 150 200 250 300 350 400 450
L (m)

Maximum value based on AIS and HINDCAST.

IACS/URS11 (Hogging moment)
IACS/URS11 (Sagging moment)

(a) Container ship; N=1,184

VBM of Oil tanker in North Atran c
0.0016

0.0014

0.0012
VBM / rho*g*L^3*B

0.001

0.0008

0.0006

0.0004

0.0002

0
0 50 100 150 200 250 300 350 400
L (m)
Maximum value based on AIS and HINDCAST
IACS/HCSR (HOG)
IACS/HCSR (SAG)

(b) Oil tanker; N=1,098

VBM of Bulk carrier in North Atran c

0.0016

0.0014

0.0012
VBM / rho*g*L^3*B

0.001

0.0008

0.0006

0.0004

0.0002

0
0 50 100 150 200 250 300 350 400
L (m)
Maximum value based on AIS and HINDCAST
IACS/HCSR (HOG)
IACS/HCSR (SAG)

(c) Bulk carrier; N=2,410

Fig. 11. The maximum wave load (VBM) correspond to Q = 10−8 estimated by using actual sea
condition each ships encountered.
 Evaluation Method for the Maximum Wave Load 319

180 1
160
140 0.8

120
Histogram 0.6
100

P
80 Probability distribu on
0.4
60
40 0.2
20
0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
VBM_actual / VBM_rule

(a) Container ship; N=1,184

160 1

140
0.8
120

100
Histogram 0.6
80
N

P
Probability distribu on
60 0.4

40
0.2
20

0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
VBM_actual / VBM_rule

(b) Oil tanker; N=1,098

350 1

300
0.8
250

200 Histogram 0.6

N

Probability distribu on
150
0.4
100
0.2
50

0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
VBM_actual / VBM_rule

(c) Bulk carrier; N=2,410

Fig. 12. Histogram and probability distribution of the maximum wave load (VBM)

5 Conclusions

The sea conditions which ships encountered and the wave loads acted on actual ships
have been investigated using AIS data and wave hindcast data, and following con-
clusions were obtained.
– The maximum significant wave height by wave hindcast do not differ significantly
from IACS/Rec.34.
320 M. Oka et al.

– The maximum significant wave height in North Atlantic sea by wave hindcast is not
significantly differ from IACS Recommendation No. 34.
– Significant wave height that ships encountered is reduced by about 10–20% owing
to the seamanship of storm avoidance.
– The maximum wave loads of individual ships were estimated for container ships,
tankers and bulk carriers using the strip method. It is clarified that the vertical
bending moment in waves, which is estimated to have acted in the actual sea
conditions, is almost below the design load.

Acknowledgment. This research includes the contents of the research conducted by the Japan
Ship Technology Research Association in FY2018 (global strategy project), which is a support
project of the Nippon Foundation. We express our gratitude to the parties concerned.

References
1. International Association of Classification Societies (IACS) Recommendation No. 34,
Standard Wave Data (2001)
2. Hogben, N., et al.: British Maritime Technology “Global Wave Statistics”. Unwin Brothers
Limited, London (1986)
3. IMO/MSC 99/INF. 19, 13 March 2018
4. NAPA FREET INTERIGENCE (2019). https://fleetintelligence.napa.fi/
5. Soares, C.G.: Effect of heavy weather maneuvering on the wave induced vertical bending
moments in ship structures. J. Ship Res. 34(1), 60–68 (1990)
6. Vettor, R., Soares, C.G.: Assessment of the storm avoidance effect on the wave climate along
the main North Atlantic routes. J. Navigation 69(1), 127–144 (2016)
7. Oka, M., et al.: Estimation of wave loads acted on ships in service based on AIS Data. -
evaluation of operational effects on the maximum loads. J. Jpn. Soc. Naval Archit. Ocean
Eng. 28, 89–98 (2018). (in Japanese)
8. International Association of Classification Societies (IACS) Unified Requirement S11,
Longitudinal Strength Standard (2010)
9. Matsuura, K., et al.: Estimation and accuracy of the marine weather. Bull. Jpn. Soc. Naval
Archit. Ocean Eng. 77, 6–10 (2018). (in Japanese)
10. Fukuda, J.: Statistical prediction for ship response. In: 1st Symposium about Seaworthiness,
pp. 99–119. The Society of Naval Architects of Japan (1969)
11. Class, N.K.: Guidelines for Direct Load Analysis and Strength Assessment March 2018
12. International Association of Classification Societies (IACS) Common Structural Rules for
Bulk Carriers and Oil Tankers, January 2019
 “Monitoring Platform” of Monitoring
and Visualizing System for Shipyard:
Application to Cutting and Subassembly
Processes

Kazuhiro Aoyama1(&), Takayuki Yotsuzuka2, Yoshiki Tanaka3,

and Yoshikazu Tanabe4
1
Research into Artifacts, Center for Engineering, School of Engineering,
The University of Tokyo, Tokyo, Japan
[email protected]
2
Shin Kurushima Dockyard Co., Ltd., Onishi, Ehime, Japan
3
Namura Shipbuilding Co., Ltd., Imari, Saga, Japan
4
Koike Sanso Kogyo Co., Ltd., Toke, Chiba, Japan

Abstract. Owing to the rapid progress of information and communication

technology (ICT), a new manufacturing environment that utilizes the ICT
infrastructure effectively is expected, and various activities are performed
proactively worldwide. This activity is positioned as the Fourth Industrial
Revolution (Industrie 4.0) by the information network and is attracting world-
wide attention. Even in the shipbuilding industry, the realization of Industrie 4.0
by IoT (Internet of Things) is important. The cyber–physical system (CPS) is the
core concept to realize Industrie 4.0; further, it is important to realize the cyber
system (virtual world) that constitutes this CPS and the “factory visualization
system” technology.
In this application, we will develop a foundation necessary for developing the
monitoring/visualizing system for shipyards based on the factory visualization
system and enhance the quality of data obtained by the visualization system. In
this study, this foundation is used for developing the “monitoring platform” and
will be applied to the actual process of a shipbuilding factory to evaluate its
effectiveness. In the factory visualization system that is the base of this research
and development, monitoring information is acquired primarily on the image
analysis technology of video images.
In this research, we add functions to acquire more monitoring information
from the video image and develop an additional function of the foundation for
developing the monitoring platform. To acquire the level of monitoring infor-
mation available at the actual factory, we aim to improve both the accuracy and
processing speed of image analysis and expand the target area of image analysis.
A monitoring platform to develop a visualization system is necessary to
realize Industrie 4.0 for shipbuilding. In this study, we organized and imple-
mented additional functions for developing a visualization system of the cutting
and subassembly processes and verified the practicality of the monitoring
platform.

Keywords: Monitoring system
Industrie 4.0
Shipbuilding management

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 321–337, 2021.
https://doi.org/10.1007/978-981-15-4680-8_23
322 K. Aoyama et al.

1 Introduction

Owing to the rapid progress of information and communication technology (ICT), a

new manufacturing environment that effectively utilizes the ICT infrastructure is
expected, and various activities worldwide are being developed proactively. This
activity is positioned as the Fourth Industrial Revolution (Industrie 4.0) by the infor-
mation network and is attracting worldwide attention. In the shipbuilding industry, the
realization of Industrie 4.0 at shipyards is emphasized. In realizing Industry 4.0, the
cyber–physical system (CPS) is fundamental. The CPS is a concept of a system in
which the real world (physical world) and cyber world (cyber world) are united by
information and are closely coupled (Fig. 1). In the CPS, a digital copy of the real
world (transfer to digital data) is built in the cyber world. This digital copy depends on
methods such as simulation or elucidates the cause of the challenges that the real world
is facing, or deepens the discovery and insights; thus, an optimal solution can be
obtained. The results obtained are fed back to the real world, thereby solving problems
and optimizing the real world. The CPS possesses such a mechanism and hence many
manufacturing industries are interested in it.
From 2015 to 2016, one of the authors had studied the practical use of a ship-
building visualization system to improve construction productivity with the competi-
tive process in the Japanese shipbuilding industry, in cooperation with the Japan Ship
Technology Research Association supported by the Nippon Foundation. As a result of
our research, we developed a system that can automatically generate a function that can
collect the information of “worker” movement and work content using information
such as a video camera’s moving image in the small assembly process of a shipyard to
generate the operator’s movement map and work Gantt chart. The developed system is
a system that primarily performs the visualization of workers (people); however, its
ultimate purpose is a system that realizes the digital twin conversion of a shipyard in
the CPS. By visualizing various “products and services” at the shipbuilding factory, the
physical system, which is a real shipbuilding factory, is expected to be converted to a
cyber system and used as a basis for cooperation. In this study, we develop the

Fig. 1. CPSS: cyber physical systems for shipbuilding

 “Monitoring Platform” of Monitoring and Visualizing System for Shipyard 323

necessary development base for developing a visualization system/monitoring system

of a shipyard foundation based on the “factory visualization system” to enhance the
quality of data obtained by a visualization system. In this study, this is known as the
foundation for developing the “monitoring platform.” We will introduce its outline and
discuss the functions added to realize the visualization system of the cutting and
assembly processes. Moreover, we organize the expectations and issues for Industrie
4.0 by the IoT of shipbuilding.

2 Background of the Research and Related Study

2.1 Outline of Innovative Shipbuilding Technology Research

and Development (i-Shipping)
The authors’ research and development were adopted for an innovative shipbuilding
technology research and development (i-shipping), which is a subsidized project by the
Ministry of Land, Infrastructure, and Transport. In this research and development,
based on the factory visualization system, the database obtained by the visualization
system is improved by developing the foundation of development for constructing the
visualization/monitoring system of the shipyard. We aim to develop an environment in
which each company can build a visualization system based on its purpose and
strategy.
– Goal 1: We reaffirm the function of the factory visualization system and develop a
foundation for the monitoring platform to build the visualization system to realize
Industrie 4.0 in a shipbuilding factory.
– Goal 2: Utilizing the foundation developed for the monitoring platform, we attempt
to organize and implement the required functionality to build a visualization system
of the cutting and subassembly processes to verify the practicality of the monitoring
platform.

2.2 Overview of Monitoring Platform

Generally, in developing the visualization system of a factory, it is necessary to develop
a system according to the characteristics of the factory and realize the visualization
system separately in cooperation with a system-developing company. Here, a platform
that can be used easily to develop a visualization system that is required by many
shipyards is expected. The factory visualization system based on a monitoring plat-
form inherited the major basic functions from the monitoring system for shipbuilding
factory that is developed by the Manufacturing Systems Engineering Laboratory
(Aoyama Laboratory) in the Department of Systems Innovation of the Graduate School
of Engineering at the University of Tokyo. This system was redeveloped as a Web
application with support by the Japan Ship Technology Research Association while
receiving a grant from the Japan Foundation in 2016 (Fig. 2). The monitoring target of
this system is a subassembly process (job shop for panel construction) with a good
video recording environment. The outline of the monitoring platform is summarized
below (Fig. 3).
324 K. Aoyama et al.

Jobshop countermeasures

Opera
on
on Gan

Fig. 2. Overview of PDCA by CPSS Plan: Create construction schedule. Do: Generate
operation data by recording video, im-age processing, and sensor data collection. Check: Browse
current progress of construc-tion and find problems. Action: Change construction schedule and
simulate construction to obtain optimized construction plan.

Data Collection System

In labor-intensive shipyards where welding workers weld based on their experience for
large welded structures, the volume and quality of shipbuilding data are poor. Eco-
nomically monitoring a wide range of large factories is important. Therefore, equip-
ment such as a fixed-installation-type video camera and smartphone are introduced, and
data related to the video image of the factory and the position and motion of the worker
are collected.
Data Analysis System
We have been developing an image analysis system that acquires the position data of
tasks, workers, etc. with time-stamped data using image analysis technology. Using the
feature amount of the image, the several types of work are determined. The welding,
gouging of the weld, grinding using a grinder, heating and cutting by a gas burner are
collected as digital data. In addition, by analyzing the acceleration data obtained from
the smartphone, the worker’s activity state is conjectured, extracted as data, and
collected.
Information Monitoring Processing System
We integrate various types of collected data, extract the work information to be
monitored, and generate a construction system model (digital copy of shipbuilding), in
which the relation between construction resource and worker are described. For
example, the position and movement of the worker are detected from the result of the
image analysis system, and if the position is close to the detected work, it is presumed
that it was involved in the work. Subsequently, the extracted work data are linked and
extracted as process information. Hence, meaningful work data are generated such as
what, how, when, where, and by whom work is processed.
 “Monitoring Platform” of Monitoring and Visualizing System for Shipyard 325

Fig. 3. Configuration of the monitoring platform

Information Monitoring Display System (Monitoring Browser)

We have developed a monitoring browser that realizes the visualization of shipbuilding
by processing and displaying data obtained as a construction system model. The work
performed by each worker is displayed for each color, and the process is shown in a
Gantt chart. Consequently, process information that is the workflow of the worker is
extracted and managed. Similar to the process information of the worker, the con-
struction state of the work for each constructing block is output as a Gantt chart, and the
number of workers for each block and total working time are output as well.

2.3 Information Defining Work Expected in Monitoring: 5W1H

The purpose of the monitoring factory is diverse; however, the purpose of the system
to be developed based on the monitoring platform is to grasp the construction state of
the whole factory. To satisfy this purpose, it is necessary to grasp the condition of the
worker, state of the work subject (product), and state of the production facility, all of
which constitute the operation state, and to obtain information on the work situation. It
is indicated that 5W1H is important as it conveys information that clearly speci-
fies work. 5W1H contains information such as when (time), who (worker), where
(place), what (object), why (reason), and how (method). Although it is desirable to
construct a monitoring technology that collects such information at one time, the
currently available monitoring technologies only have limited collected information
owing to the characteristics of the method.
For example, from the image of a video camera, we can collect information well on
when (time) and where (location) for a certain task; however, to obtain information
such as who (worker) or what (article), it is necessary to improve the accuracy of the
analysis. Therefore, it is necessary to introduce and build a method to effectively
acquire information that is difficult to collect. In this research and development, mul-
tiple monitoring systems are introduced, and the data obtained from those systems are
326 K. Aoyama et al.

integrated to develop a system that can monitor the construction status of the factory.
Therefore, we herein introduce a monitoring system that can cooperate with the
monitoring platform, organize the information obtained by that system, and organize
the information obtained by integration and cooperation between that system and other
systems. During information integration, the correspondence relationship between
information is highly important; furthermore, as a key issue in monitoring, we arrange
and discuss the method of constructing the correspondence relationship between
information.

2.4 Coordination Function Between IoT Equipment and Monitoring

Platform
The monitoring platform is built as a platform that can collect, integrate, and visualize
various data regarding shipyard operations. To use this platform at actual shipbuilding
sites, it is necessary to extend the function to link data obtained from various devices
and equipment. To discuss and realize concrete collaborations, we will consider the
development and introduction of IoT devices, devices and systems for collecting data,
while targeting cutting and assembly processes.
An IoT device that can be used in the cutting process is an NC cutting machine.
The NC cutting machine cuts the steel plate according to the NC data created in the
design, where both the cutting plan and log data of the operation panel are recorded.
Joining the trend toward IoT in recent years, IoT has been applied to NC cutting
machines, and cutting data managed locally inside the cutting machine has become
information that can be managed in the cloud by an Internet connection.
In the assembly process, the digitization and application of the IoT of a welding
power source are being promoted. In a digital welding power source, the data of the
welding voltage and welding current are monitored in real time to control the welding
process digitally. These data are provided through a network on the cloud. However, in
the current situation of shipyards in Japan, many analog welding power sources exist,
and it is difficult to acquire data with the high accuracy of current and voltage in detail.
As the IoT of the analog-type welding power source, a welding arc management system
that monitors the ON/OFF status of a welding power source and manages data on the
cloud has been developed and sold, and welding process management is expected.
It is useful to understand the construction status of facilities and equipment, even
with data obtained from the facilities and equipment of a shipyard. However, the
acquired data are limited and incomplete for understanding the construction status of
the shipyard as a whole. Therefore, to effectively use the data obtained from the
facilities and equipment of the shipyard, cooperation with the monitoring platform is
expected (Fig. 4).
 “Monitoring Platform” of Monitoring and Visualizing System for Shipyard 327

Fig. 4. System cooperation by monitoring platform

3 Visualization of Cutting Process

3.1 Overview and Monitoring of the Cutting Process

The cutting process supplies many parts in ship construction, and managing this pro-
cess is important for improving shipbuilding productivity. The cutting operation is
primarily performed by the NC cutting machine; however, in an actual operation, the
operator and assistant must handles many operations. It is important to analyze the
operating status of NC cutting machines and cranes as well as the activity status of the
workers such that the reasons for the trouble and delay occurring in the cutting process
can be understood. Furthermore, necessary and sufficient data should be collected to
create an optimal plan for the cutting process according to situation.

3.2 Visualization of Cutting by Data Collection from NC Cutting

Machine
To visualize the operating status of the NC cutting machine, we developed a tool to
manage the operating status data of the NC cutting machine, as shown in Fig. 5. This
tool allows one to verify if the cutting process is being performed on schedule. If it is
not implemented as scheduled, the cause is analyzed from the captured video. By
utilizing data such as worker movements and crane movements collected in coopera-
tion with the monitoring platform, the cause of failure to perform cutting according to
the process can be ascertained. For example, shortage of cranes for transporting steel
plates or parts to be cut, problems with the flow lines of workers, and insufficient
operators. These causes are difficult to obtain only with data collected from the NC
cutting machine.
The tool was researched and developed by customizing Konnection to apply the
IoT in the cutting process and the effective use of the monitoring platform. In addition
to the management of the cutting process, the standard function involves collecting the
cutting amounts, operating rates, maintenance information, and cost information of
each cutting machine, and displaying a list of information managed by a web browser
(Fig. 6). As information on the operation status of the cutting machine, the planned
328 K. Aoyama et al.

Cutting Results
Operation History of Cutting Workshop: Carry In, Cutting and Carry Off Steel Plates

Cutting Results Management

Cutting Plans

Fig. 5. Monitoring the operation status of NC cutting machine

KAP WEB Dashboard Company-wide Machine status Equipment status by floor

WEB application

Machine alarm history Working time Operating status analysis

Maintenance and consumables Operating status monitoring Cutting process

Fig. 6. Functions of konnection (KOIKE)

cutting amount of the day and the actual cutting amount are compared and displayed in
a pie chart, and the operating rates of the previous and current weeks are compared and
displayed.
 “Monitoring Platform” of Monitoring and Visualizing System for Shipyard 329

Data logs required for maintenance such as travel distance, operating time (cutting,
printing, marking, piercing, etc.), grease up, and cooling water exchange are collected
and displayed.

3.3 Integration with Monitoring Platform and Visualization of Cutting

Process
Although IoT NC cutting machines can provide information for understanding the
operating status of cutting machines, many issues must be solved to understand the
operating status of the whole cutting process. Therefore, we consider a method of
understanding the operation status of the cutting process by integrating the data
obtained from the NC cutting machine and that obtained by the monitoring platform
(Fig. 7).
In the cutting process, the worker operates the NC cutting machine and handles
various works before and after cutting. Therefore, we studied a method to extract the
work to be processed by the worker for the NC cutting machine. Specifically, we
attempted to acquire various works by integrating data obtained from the image
analysis of the video image provided by the monitoring platform and acceleration data
according to the worker’s operation obtained from the acceleration sensor built into the
smartphone. As the worker moves his entire body in the cutting process, we attached an
acceleration sensor to the chest pocket and acquired the x-, y-, and z-axis three-axis
acceleration data. We calculated 27 types of features for the obtained time series data of
acceleration, and we conjectured the work from those features.
To conjecture the work, we used the relationship between “where” and “type of
work.” Specifically, the information on three types of “where (places)” such as “control
panel,” “on the cutting workspace,” and “outside the cutting workspace” is determined.
The work related to the cutting is processed on the cutting workspace; therefore, when

Cutting workshop NC operation

1,2,3,4 log chart

Cutting
workshop 1
Video when cutting machine is down

Fig. 7. Integration of monitoring platform and konnection

330 K. Aoyama et al.

the worker is on the cutting workspace, four types of work, i.e., “grinder,” “burner,”
“marking,” and “walking/confirmation” are conjectured. The flow of the work method
estimation is shown below (Fig. 8).
1) Combining the classification by calculating the feature quantity and the classifica-
tion by image analysis of the video data, the operator’s workplace is classified into
“NC control panel”, in the cutting workspace, and out of cutting workspace (Fig. 8
④).
– Conjectures of working area: The worker area determined from the feature data
of the acceleration data and the worker area obtained from the image analysis
system of the video data are compared. If they match, the working area is
decided. If they do not match, it is considered as an “uncertain area” (Fig. 9).
– The uncertain area is decided by considering the area before and after it.
2) The works considered to be on the cutting workspace are classified into four types
of work: “manual cutting,” “grinder,” “marking,” and “walking/confirmation”
(Fig. 8 ⑤).
– Classify features using the random forest, which is a machine-learning method.
Assign labels of work to the classification results. Those that cannot be labeled
definitely will be labeled as uncertain work.

Processing of Manual Cutting

Operation
acceleration Stand
data
Grinder
Classification of
Estimation of In the
work by machine
worker's location Workspace
learning
Marking

Image Out of
Analyzing Workspace
Working

Fig. 8. Flow of conjectures of works on the cutting workspace

Fig. 9. Conjectures of the working area

 “Monitoring Platform” of Monitoring and Visualizing System for Shipyard 331

Fig. 10. Conjectures of worker’s activity on cutting workspace

– The work processed on the cutting workspace in the cutting process of the
shipyard accounts for the feature that the same work condition continues for a
certain duration or more. The uncertain work is decided in consideration of the
pre-work and post-work (Fig. 10).

4 Visualization of Assembly Process

4.1 Overview of the Assembly Process and Points of Visualization
The assembly process connects members with members, members with blocks, blocks
with blocks, etc. by the welding operation. Therefore, in monitoring the assembly
process, it is important to understand the situation of the welding work and that of the
work associated to welding. Arc time measurement and task extraction are important in
monitoring the welding operation, which is the core of the assembly process. In arc
welding, arc time implies the time during which an arc is generated, that is, the welding
time. This welding time is important index to control the manufacturing process. The
work efficiency can be observed by calculating, evaluating, and analyzing the arc time
rate that indexes the arc time in the working time.
Generally, the arc time rate is −15% to 50%; however, it is affected significantly by
the preparation setup, transportation handling, waiting time, etc. before and after
welding. Additionally, the arc-time rate changes depending on the skill of the welding
operator, and efficiency can be improved by appropriate education and training. Fur-
thermore, based on the results of calculating and analyzing the arc time rate for each
block or region, man-hour reduction can be improved by feeding back to the exami-
nation of the number of weld joints and joint positions in the structural design stage.

4.2 Collection of Arc Time Data

As a monitoring system that can measure the arc time quantitatively, we introduced a
system called the “arc time watcher” to the assembly process. The arc time watcher
contains a function to determine whether an arc occurs if the read current value exceeds
332 K. Aoyama et al.

a certain value. It is configured to accumulate data, and its associated procedures are as
follows: Attach a small clamp-type current sensor to each welding power supply, read
the current value flowing through the power cable on the primary side, receive data
using wireless communication with the receiver, and receive the data management
software through the company network.

4.3 Output Function of Arc Timer

The data collected by the system are visualized by various graphs and tables to
understand and analyze the basic arc times. In addition, cooperation with external data
other than arc time is considered, and the output data can be analyzed in various forms
using spreadsheet software, etc. In the operation status report, the time zone in which
an arbitrary arc of each welding machine has occurred can be determined. A blank time
zone represents a time zone where no arc occurs.
Furthermore, the real-time arc occurrence status called the signage function can be
monitored. For example, by installing a monitor in the factory, it can be used for
understanding and disseminating goals and achievements.

4.4 Application to Actual Process of Assembly Process

In the experiment, 120 sets of arc time watchers were introduced to the assembly
process of the parallel part of the hull for the welding power source primarily used by
welding workers. This process exhibits a relatively high welding efficiency because the
blocks in the parallel part of the hull flow on the conveyor with the welding machine
are fixed. The obtained data are organized and analyzed to use for various cost
reductions.
Arc Time Rate: By identifying workers who use individual welding machines and
aggregating the data of arc time and working time jointly, the arc time rates for each
worker are as shown in Fig. 11 (a). In this graph, blue is arc time and red is additional
work; furthermore, the efficiency of each worker can be determined quantitatively. This
tabulation/analysis is not limited to individuals but can be tabulated by workgroup or
work segment, and improvement points can be extracted by comparing similar facilities
and construction environments in the workplace.
Block Work Efficiency: Figure 11 (b) shows a graph in which the arc time for each
block constructed in the survey period and its arc time rate are summarized. By
investigating the block that each worker had installed and at what time, the arc time rate
for each block can be determined, that is, the work efficiency in the welding operation.
The results of this tabulation can be used as feedback to future improvements in
construction procedures and block shapes.
Operation Rate of Welding Machine: As shown in Fig. 11 (c), the arcing situation
of each welding machine can be obtained. Using this data, the layout of the welding
machines and the required number of welding machines can be optimized.
 “Monitoring Platform” of Monitoring and Visualizing System for Shipyard 333

Fig. 11. Collection of welding data

Arc Generation Status: If the time when each welding machine is generating arcs can
be determined, and when many welding machines in a certain area do not generate arcs,
the efficiency can be improved by investigating and identifying the cause.
Current Waveform: The waveform of the current value can be obtained using a
sensor. Figure 11 (d) is a waveform of an individual welding worker; however, the
stability of the current value clearly indicates the movement of the torch and the
stability of the welding position. The stability of the current value is directly linked to
the stability of the welding quality, which facilitates the understanding of the quality.

4.5 Visualization of the Assembly Process in Collaboration with Factory

Visualization System
Until the previous section, we have explained the case of introducing an arc time
watcher in the actual process of a shipbuilding factory and the data and analysis
obtained by the system. Although this analysis alone can be used to examine factors
that inhibit productivity from the aspect of welding work, it is difficult to judge why the
arc time rate was bad and what was the situation when it was bad. Meanwhile, the
factory visualization system enables the collection and analysis of data pertaining to
welding operations, worker activities, block loading and unloading, etc. (Fig. 12). By
combining this system with the arc time watcher, it can assume the operation status of
the actual assembly process.
In the example presented in the previous section, the assembly line introduced the
conveyor line, and it is easy to obtain the total arc time for every pallet of the conveyor,
that is, for every construction block. However, in general, a block is placed on the
workspace such as a job shop, and the work starts. At this point, it is important to
identify the blocks on the workspace, identify the work area, and collect the welding
work related to the construction blocks. With the arc time watchers alone, it is difficult
334 K. Aoyama et al.

Fig. 12. Correspondence of two types of welding data

b. Work Time

a. Browsing Monitored Data

Fig. 13. Browsing monitored data in a visualization system

to associate welding work with construction blocks. At this time, the correspondence
between the welding work and construction block can be defined clearly using the
visualization system.
Figure 13 shows the browsing of the monitored data of construction. The data
related to welding work etc. are combined, and the work status in the job shop can be
confirmed with a heat map, or the amount of work is calculated.

5 Future Tasks and Prospects

5.1 Visualization of Logistics in Factory

When we consider shipyards from the viewpoint of intra-factory logistics, shipyards
carry various materials, from the transportation of steel plates and small members to the
 “Monitoring Platform” of Monitoring and Visualizing System for Shipyard 335

transport of large construction blocks. The construction blocks are transported from the
cutting process to the subassembly process, from the subassembly process to the
assembly process, and further to the erection process before reaching the construction
dock. If problems occur in the distribution that links these processes, the entire process
of the shipbuilding shop will become haywire. For example, if the assembly at the
assembly shop falls into a waiting state, the work will be stopped, and the delay will
propagate to the block erection. Therefore, it is important to understand the information
on how parts and blocks, which are essential targets for monitoring, are transported.
The facility and equipment that are fundamental in the transport of parts and blocks in
the shipyard are cranes and moving carriages. Furthermore, it is important to monitor
the movement of these facilities and equipment.
The monitoring platform can store and manage the data of the crane movement.
Figure 14 (a) shows the history of the crane movement; Fig. 14 (b) shows the allo-
cation history of a block on the assembly shop. If we can obtain the data of a movement
from an IoT crane and integrate it with the allocation data of blocks on a job shop, we
can use them to verify the allocation history of a block. As mentioned above, the
movement of a block is necessary for monitoring logistics in a shipyard. We will
discuss the possibility and effectiveness of the visualization of logistics.

Crane
10:30

Crane moves
Crane
11:00

The block was carried in.

Crane
11:30

The block was carried off.

12:00

Carry In Carry Off

Time Crane
a. Movement of a Crane b. Allocation History of blocks on Assembly Workshop

Fig. 14. Movement of a crane and block in the visualization system

5.2 Applying IoT of Welding Carriage/Cutting Carriage

It is important to collect data pertaining to the NC cutting machine, portable cutting
machine, and welding cart. The collected data can provide a more detailed under-
standing of the equipment’s operating status. The IoT mobile cutting machine equipped
with Bluetooth communication function and the IoT welding carriage has been put on
the market after the research and development. Using this built-in communication
function, the operation log is sent to the monitoring platform and visualized as an NC
cutting machine. By linking the log of the workers operating multiple trucks with the
worker’s work status by image analysis, visualizations such as the confirmation of the
appropriate personnel placement, and the comparison between design welding amount
and actual welding amount are possible.
336 K. Aoyama et al.

6 Summary

The configuration of the construction system differed significantly depending on the

conditions of the shipbuilder, shipyard, etc. From the viewpoint of visualizing the
production system, the purpose of the visualization itself varied significantly. There-
fore, the visualization systems to be developed were different. To efficiently develop a
visualization system for construction, it was desirable to develop an environment that
could comprehensively use various monitoring element technologies and build appli-
cations suitable for the purpose.
The development of a visualization platform, or the monitoring platform was to
streamline the system for a shipyard to process monitoring information easily. In
addition, using Konnection, we built a visualization system for cutting and sub-
assembly processes, which were the actual processes of the shipyard. Consequently,
various data could be collected from each process, and information on unreasonable
(Muri), waste (Muda), and unevenness (Mura) in the process could be obtained. In
addition, analysis and improvement measures could be performed for factors that
impeded productivity. This is expected to improve productivity by at least 10%.
By linking the welding machine arc time rate and monitoring platform acquired
from arc time data, quantitative data such as “welding worker efficiency and skill” and
“block and work area working efficiency” were acquired. This data could be visualized
quantitatively and instantaneously. Therefore, it is expected to be used as a new control
index in the future; however, this is not foreseeable with the existing data collection
methods.
By operating this system, for example, in addition to the man-hours used in the
previous indicators, an index called “arc time rate” could be used that could be eval-
uated more clearly. Using this indicator, it is possible to meet the goal and to carry out
analysis and improvement. To lead to improve productivity, it can be used how to use
acquired data, and each manager can use this data to improve workers and the work
environment. To integrate the monitoring system, it is necessary to organize the
obtained data before integrate them. Finally, it extremely important to carefully study
the procedure for the improvement of the workers and their working environments
from the viewpoint of each manager.

Acknowledgement. This research was promoted with the support from the Maritime Produc-
tivity Revolution (i-Shipping) of the Ministry of Land, Infrastructure, Transport, and Tourism.
We are deeply grateful for the support. In addition, the base of the monitoring platform developed
in this research utilizes the factory visualization system developed by the Japan Institute of Ship
Technology Research with the assistance from the Japan Foundation. We are grateful the
assistance.
 “Monitoring Platform” of Monitoring and Visualizing System for Shipyard 337

References
1. Acatech (National Academy of Science and Engineering), Recommendations for implement-
ing the strategic initiative INDUSTRIE 4.0 (2013)
2. Liu, J., Aoyama, K.: Monitoring system for sub-assembling-shop to manage effective
construction. In: Proceedings of the 15th International Conference on Computer Applications
in Shipbuilding (ICCAS 2011), Trieste, Italy, September 2011
3. Liu, J., Hirota, J., Wu, C.X., Aoyama, K.: Monitoring system for advanced construction
management (integration methods of monitoring data to generate virtual job shop. In:
Proceedings of the 16th International Conference on Computer Applications in Shipbuilding
(ICCAS 2013), Busan, Korea, September 2013
4. Hiro, Y., Aoyama, K., Liu, J.: Monitoring system for advanced shipbuilding construction
management: extracting and utilizing monitoring data by considering the reliability of
monitored data. In: 17th International Conference on Computer Applications in Shipbuilding
2015 (ICCAS 2015), Bremen, Germany, September 2015
5. Aoyama, K., Mizushima, T., Hiro, Y., Oizumi, K.: Advanced management of shipbuilding
construction using monitoring system and job shop simulation. In: Proceedings of PRADS
2016, Copenhagen, Denmark, September 2016
6. Aoyama, K., Hoshi, M., Hiro, Y., Kitamura, R., Oizumi, K.: Cyber physical system in
shipbuilding toward realize smart shipyard: integration of monitoring system and shipyard
simulation system. In: 17th International Conference on Computer Applications in
Shipbuilding 2017 (ICCAS 2017), Singapore, September 2017
 Monitoring Systems in Design of Ships

Serena Lim(&), Kayvan Pazouki, and Alan J. Murphy

Offshore and Subsea Technology Research Group, School of Engineering,

Newcastle University, Newcastle upon Tyne, UK
[email protected]

Abstract. Recent advances in sensor technology and data communications

have given rise to a market for creating and retrofitting monitoring systems on-
board vessels. Systems on-board the ships are commonly monitored indepen-
dently, however, an integrated monitoring system that can communicate
between systems will allow a wider array of applications. Typically, most of the
monitoring systems and data acquisition units are retrofitted into ships. This
presents a layer of complication during sensor selection and installation. The
inclusion of an integrated monitoring system at the design stage is therefore
proposed to enable the optimal selection of sensor types and installation loca-
tions to achieve accuracy in the quality of data. The monitoring system
addressed during the design stage should include information that is necessary
and important to the operators and owners. This approach will allow collection
of a structured, accurate and continuous time series of data, specific to the vessel
type and operation, for further analysis. A monitoring system that is designed as
part of the ship during the commissioning stage will also be able to log data for
analysis throughout the entire life cycle of the ship, including the initial sea
trials. This paper showcases the benefits of and experience gained from an on-
board monitoring system and addresses the lessons learned that will highlight
the added benefit of embedding the monitoring system within the design stage of
the hull and machinery arrangement. This paper considers aspects relating to
monitoring systems and highlights how these processes can be integrated into
the design processes of a ship.

Keywords: Integrated monitoring system
Sensor selection and installation

Data collection platform

1 Introduction

Shipping is the most energy efficient way to move large volumes of cargo. However,
there are still environmental concerns related to shipping operations and the contri-
bution this sector makes to climate change [1]. Recent advances in sensor technology
and data communications has given rise to a market for creating and retrofitting
monitoring systems on-board vessels to allow for better safety and monitor and
improve energy efficiency of ships.
Using sensors to monitor ship systems is common practice. Most monitoring
systems focus on a single goal, such as evaluation of hull conditions [2], engine
efficiency monitoring and dynamic positioning monitoring system [3]. Each of these

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 338–348, 2021.
https://doi.org/10.1007/978-981-15-4680-8_24
 Monitoring Systems in Design of Ships 339

monitoring systems is important in achieving specific goals. By collecting data from all
vessel systems onto one uniform platform a holistic analysis of vessel performance is
possible [4, 5]. A complete monitoring system inevitably should interact with all
systems. However, retrofitting monitoring systems into existing ships presents a layer
of complication during sensor selection and installation. Existing vessels are not
designed for sensing and complications arise when fitting sensors into existing systems,
where the sensors might be intruding or might interfere with other systems.
Efforts to incorporate data monitoring systems during the design stage of ships is in
line with the efforts of the International Maritime Organization (IMO) in developing
rules for autonomous ships that operates mainly based on sensors and data from the
ship system. The scoping exercise for Maritime Autonomous Surface Ships (MASS)
will end in 2020 [6]. Moreover, the monitoring systems in design of ships will con-
tribute significantly towards the shipping industry’s outlook as this sector moves
increasingly towards digitalisation and big data [7].
This paper discusses typical approaches used in retrofitting sensors on-board
existing vessels and the challenges in doing so. The added benefit of embedding the
monitoring system at the early design stage of the hull and machinery arrangement is
discussed. A suitable approach in embedding a data monitoring system at the design
stage of ships is presented.

2 Challenges in Data Monitoring Systems

Recent technological advances in sensor technology, ability to digitize information

from sensors and store data in a structured database for data processing has encouraged
the fitting of data monitoring systems on ships. The key goal of fitting a monitoring
system on-board is for assessment and management of ship performance and condition.

2.1 Common Practices in Developing Ship Monitoring Systems

The holistic data monitoring system can be implemented in a systematic manner, made
up of five steps as illustrated in Fig. 1. These five steps comprise of the vessel survey
step, sensor selection step, sensor installation step, sensor communication step and data
analysis step. Each of these steps will be further clarified to highlight the importance of
each step in designing the data monitoring system. The standardized method can be
effectively applied to an individual vessel as well as an entire fleet. This top down
approach is adopted to obtain data from systems on-board the vessel to a single
platform for structured data storage that eases data processing procedures, resulting in a
holistic vessel monitoring system [8]. These raw data can be converted to useful
information using appropriate processing methods, turning raw data into knowledge
when analysed, thus underpinning rational decision making for improved vessel
performance.
Vessel Survey. The vessel is surveyed to gather information from each system on-
board and to identify specific equipment to be monitored. During the survey infor-
mation is gathered from the Master and Chief Engineer and other vessel crew, to
340 S. Lim et al.

Fig. 1. Methodology for fitting data monitoring systems on ships

identify the systems that need to be monitored. The survey step is time consuming but
is of absolute importance due to the fact that for many older vessels, drawings on-board
the vessel might not be up to date, due to rewiring or repairs that were carried out
throughout the lifetime of the vessel. In order to carry out a more time efficient and
productive survey, the system drawings and the vessel log are taken into consideration
prior to the vessel survey to identify the main vessel operations. Existing sensors that
are reliable are identified during the vessel survey and additional sensors that are
needed are identified.
Sensor Selection. Following the survey, the procurement of new sensors is made, by
selecting the most appropriate sensor and conducting factory acceptance tests for all
hardware. Most vessels have sensors on-board, these sensors are considered and their
accuracy is also taken into account. If the existing sensors are not reliable, new sensors
are proposed for installation. For cases where sensors are not available, new sensors
could be installed. In the current market, there are many sensors available with different
methods of measurement and different levels of accuracy. For example, the measure-
ment of the torque of the shaft can be done using the strain gauge method or the optical
method. The fuel flow on the other hand could be measured using the positive dis-
placement method or the Coriolis method. During the process of selecting sensors,
accuracy is important, as small inaccuracies of individual sensors connected in series to
monitor the key performance of a system can lead to large aggregated errors.
Sensor Installation. Once the type and number of sensors required for energy mon-
itoring is established, location of the sensors and how they can be installed is
addressed. Complete sensor installation also includes building hardware for accurate
data acquisition and setting up reliable monitoring systems that are able to unify all
sensors under a common hub.
Sensor Communication. Data from the sensors has to be transformed into useful
information in order to understand the data from different systems on-board the vessel.
With existing sensors and new sensors installed on the vessel, a data collection plat-
form is needed to unify the data from the sensors to a server or a computer. The
platform must be able to communicate with different devices with different commu-
nication protocols. Typical communication protocols that are currently found on sen-
sors on-board the vessel are Controller Area Network (CAN bus), Modbus and
 Monitoring Systems in Design of Ships 341

National Marine Electronics Association (NMEA) systems. All these different types of
physical sensors need to be connected to the same network via an Ethernet.
Analysis. The raw data is processed using mathematical formulae and statistics to
generate useful information. More-over, quality parameters such as energy production,
hull condition, propeller condition, engine efficiencies are calculated by further analysis
of this raw data. After the energy breakdown structure of the overall energy flow on-
board the vessel is produced, data collection and data processing, and an analysis and
interpretation of the data is undertaken. This step will transform the information into
useful knowledge, which can be used to improve current operation and best practice for
the future design. The processed data is monitored and plotted for different modes of
operation of the vessel. An automated recommendation for each operation would be the
ultimate goal of the vessel monitoring system. At different operating speeds and
operating modes, the energy distribution would vary significantly. The result of this
analysis would be recommendations for the optimised use of active equipment for a
particular operation.
The top down approach has been the typical methodology adopted to fit monitoring
systems on ships and has proven to be a smooth and unified system for gathering data
from various systems on-board the ship for data analysis.

2.2 Challenges of Each Step (Lessons Learned)

The top down approach has been applied on different types of ships and has proven to
be effective in gathering data from different systems on a common data collection
platform and perform energy management for ship. However, there are many chal-
lenges and obstacles that need to be overcome. The challenges in fitting a data mon-
itoring system on ships are highlighted in Fig. 2. Each of these challenges and the
lessons learned overcoming the challenges are discussed.

Fig. 2. Challenges in fitting monitoring systems on ships

342 S. Lim et al.

During the vessel survey, the piping on the vessel has to be traced as the drawings
received prior to the survey could be inconsistent with the actual piping on the vessel
due to retrofitting and repair works carried out on the vessel. Miscommunication and
human error could also happen when a survey is carried out to understand operation of
the ship.
When it comes to sensor selection, there are restrictions on the types of sensors that
can be selected due to compatibility of sensors with systems, inability to install built-in
sensors that might breach the warranty of certain system and also space restrictions.
During sensor installation, cable runs are needed to connect the sensor to the main
communication. The cable runs have to comply to fire safety, including watertightness
and gastightness depending on ship type and often has to be done when the ship is
alongside. When carrying out installation of current and voltage sensors on busbars, the
complete busbar panel or switchboard has to be turned off or disconnected from
electrical supply for safety purposes. These installation procedures causes ship down
time and obstruct daily operations. Due to the pipeline configurations on ships, some of
the sensors cannot to be installed in an optimum location to acquire correct sensor data,
for example, when measuring fuel flow in a pipe, only sections of the pipe with single-
phase flow can be fitted with fuel flow sensor to be able to quantify the fuel flow
accurately.
Certain sensors are readily available on-board the ship, however, the sensors often
have limited transmission interfaces that do not have, for example, a spare port for data
transmission to the data monitoring system.
Data analysis of a non-structured database with different frequency and quality of
data will lead to unreliable conclusions. Gathering systems from individual systems
only provide surface analysis for their individual goals and not with respect to a
broader collective goal. Due to restrictions on the installation of sensors at certain
locations, certain assumptions and approximations have to be made in order to analyze
data.
After the installation of the monitoring system, it is essential to improve training of
crew to ensure proper maintenance for optimum performance of the sensors. The crew
should also be trained to understand on-board visualization of data, so that the
advantages of having real time display of data can be realised.

3 Design of Monitoring Systems Using the Bottom Up

Approach

By implementing a bottom up approach and including the planning of monitoring

systems at the design stage, the challenges discussed in Sect. 2.2 can be mitigated.
During the start of a ship design, the operation profile is determined; the main key
performance indicators for ships of such operation can be determined. The design of a
structured database is also important for optimal data collection from different systems
into one uniform platform for ease of data processing and data quality control.
 Monitoring Systems in Design of Ships 343

3.1 Key Performance Indicators

Data from sensors can also be used to detect degradation of machinery over time,
where the new trends show deviation from ideal conditions. A set of key performance
indicators (KPIs) are used to provide a baseline curve to mark the performance of the
ship for analysis and determine the type of data analytics that need to be performed.
This will highlight and determine which components need further attention. In order to
understand the exact performance of the vessel during each operation, real data from
the vessel is needed. To obtain this, sensors are needed to detect different parameters
form different systems on-board the vessel. By specifying the KPIs, the key parameters
that are needed can be specified and appropriate sensors can be installed to obtain the
required parameters. The flow chart of an bottom up approach is shown in Fig. 3.

Fig. 3. Flow chart of fitting data monitoring systems on ships

Vessel KPIs can be further divided into three different categories, monitoring of the
individual system, as a collective system and for fleet management. Performance
monitoring of individual system are such as the monitoring of ballast water, emissions
monitoring, engine efficiencies, pump efficiencies, structural integrity and cargo han-
dling time. The KPIs for collective system monitoring include hull and propeller
performances, vessel energy management, load distribution management, operational
optimisation and environmental performances. The KPIs that can be achieved through
monitoring systems for fleet management can include ship availability and dry dock
scheduling.
For each KPI, a set of parameters is required. For example, to measure hull and
propeller performance, parameters that should be measured include the ship speed,
engine brake power, thrust on shaft, environmental conditions and fuel consumption.
And in order to measure KPI for energy efficiency of vessel, fuel consumption, energy
convertors such as main engines, and energy consumed by each consumer can be
measured.
By determining the KPI and identifying the parameters needed to perform the
necessary analysis, data needed and frequency of data needed can be determined. The
appropriate data logging system and structured database can be designed. Appropriate
sensors can be fitted to ensure all required parameters are captured unto a unified
database. Typical sensors normally found on-board the vessels are fuel flow meter,
torque meter, electrical power metre, GPS, pressure and temperature sensors.
344 S. Lim et al.

3.2 Structured Database

A standardised data mining and data monitoring system on-board ships has been
developed and improved through experience of fitting such systems on different types
of ships using the top down approach shown in Fig. 1. There are many sensors on-
board the ships that are under-utilised, such as the fuel flow metres, and the shaft torque
and thrust metres. The main reason is that these sensors feed into separate systems with
individual goals. The sensor standardisation methodology developed, by introducing a
common hub, allows sensors to communicate with each other, maintaining the features
to provide information on the individual goals as well as aggregating data from dif-
ferent sensors to provide collective goals. Following a series of sensor installation and
communication on different types of vessel, the methodology shown in Fig. 4. was
developed to standardise data mining and data monitoring on-board a ship.

Fig. 4. Standardise data mining and data monitoring on-board a ship [9]

In between the physical sensors and visualisation of data is the common commu-
nication platform with software interfaces. Software targeting different sensors can be
programmed to translate the output signal into digital data. These data gatherers and
translators software are also known as the sensor containers. The sensor container also
contains the ability to filter stray data, remove inconsistent data, and is calibrated to log
input of a certain pre-defined frequency and time stamp. Data from all sensors are fed
into a data aggregator, where the data can be processed using mathematical algorithms.
The added benefit of having a data aggregator is to log all data from sensors at a
standard time stamp for easy processing of data. It also consolidate data over a period
of pre-defined intervals and avoid storing every individual raw data point that does not
add value to the database, avoid data overload incident and preventing program failure.
The output of the data aggregator are displayed to show the real time data as well as
sending to an on-board database by transmitting the data via a cloud service.
 Monitoring Systems in Design of Ships 345

The common hub developed is a ready system that can operate in harsh conditions
and hardy towards humidity and temperature change, equipped with large data storage
capabilities and ability to communicate data in a concise manner. The system is also
able to communicate to different types of sensor output, allowing a wider selection of
suitable sensors. The system can also collect data continuously, process data and be
programmed to incorporated pre-determined KPIs. Now that the system and the
structure of the common platform is established, a bottom up approach can be applied
to include data monitoring system at the design stage of the ship.

4 Benefits of Data Monitoring System at Design Stage

Inclusion of an integrated monitoring system early on in the design stage is proposed to

enable the optimal selection of sensor types and their installation to achieve accuracy in
the quality of data. The monitoring system addressed during the design stage should
include information that is necessary and important to the operators and owners. This
approach will allow a structured, accurate and continuous time series of database,
specific to vessel type and operation, for further analysis. A monitoring system that is
designed as an integral part of the ship during the commissioning stage will be able to
log data for analysis throughout the entire life cycle of the ship, including the sea trials.
A concept design of monitoring system that suit the ship can be developed. The
design can be presented to the client for evaluation. Additional data required by the
clients to suit specific requirements can be added at this stage. When the monitoring
system is presented together with the ship and system design, better planning of space
required by the sensors, selection of sensor types and installation procedures can be
planned. The technical designs together with all drawings, documentations and cal-
culations can be submitted together, to be approved by a class society, national
authority and shipowner.
The installation of data monitoring systems at the design stage of ships is beneficial
for system efficiency and operational efficiency. In terms of system efficiencies, the
monitoring system fitted when the ship is being built will be able to log data for
analysis throughout the entire life cycle of the ship, including the sea trials. The sea
trials data that is measured with the same sensors will be comparable to the bench-
marking of ship activities. Figure 5 shows a comparison of data gathered from a newly
fitted monitoring system with the original sea trials data carried out in 1999 when the
ship was newly built. A difference in the curve is observed, however, it was not
possible to identify the methodology used when the ship was tested in 1999 to ensure
compatibility of data gathering process. The data from the newly fitted monitoring
system shows that the output power is similar for both sides of the engines, giving
confidence in the installed sensors.
A data monitoring system can benefit from planning of sensor installations during
the design stage of the vessel to allow a holistic approach in managing data to provide
useful information. For example, current methods to quantify and manage vessel
energy flow is obtained from a combination of data from sensors and also from
technical specifications and theoretical methods. Figure 6 is an example of energy flow
of a tug presented as a Sankey diagram. The main engines are fitted with fuel flow
346 S. Lim et al.

Fig. 5. Data from new monitoring system as compared to sea trials data

meters and shaft torque meters, which allows main engine efficiency to be determined.
However, due to space restrictions on board the vessel, fuel meters on the auxiliary
engines and the electrical sensors are not able to be fitted. Figure 6 shows the main
engines fuel consumption and brake power when operating at 70% Maximum Con-
tinuous Rating (MCR), derived from sensor data. Reverse calculations have to be
carried out to quantify the approximated fuel consumption of the auxiliary engine based
on the electrical load balance sheet. This solution is proposed for a retrofitted system on
the vessel. Had the data monitoring system and sensors installation been planned
during the design stage, a fully data-centric analysis could have been carried out instead
of using rough estimations.
Furthermore, a monitoring system fitted during the design stage will give confi-
dence in the measurements of quantifying savings, and a difference to the performance
after hull cleaning activities and ship retrofitting activities. Moreover, irregularity or
deviation from a baseline condition can lead to system degradation identification or
fault detection. Data from the entire lifecycle of the ship can also be used as feedback to
improve new generations of ship design.
In terms of operational efficiencies of monitoring systems, on duty engineers will
be able to undergo appropriate training for sensor maintenance and also training to
understand the monitoring systems on-board. A structured database with quality data
will give a larger dataset to perform further analysis such as using machine learning
techniques.
This platform can be used for moving towards full automation regulatory. This
method of obtaining quality and structured data for the entire lifecycle of the ship will
evolve and feed into improvements in new designs and operations.
 Monitoring Systems in Design of Ships 347

Fig. 6. Sankey diagram of energy flow on a tug

5 Conclusions

The goal of including an advanced monitoring systems in ship design is to provide the
design team with such information throughout the design process that an efficient
monitoring system be included during the initial design. A bottom up approach was
proposed for integration of the monitoring system at the design stage, where a stan-
dardised data format for ships will create a network where ships can communicate and
share data with each other automatically. Ship design is a process that blends complex
models and systems to create a reliable and functional product, the novelty of including
data monitoring systems in the early design stages allows all necessary condition of the
vessel to be logged throughout the lifecycle of the vessel and allows smart systems and
analytics to be used in vessel performance analysis. The challenges of fitting moni-
toring systems on-board vessels were discussed and the benefits of having a data
monitoring system installed from the beginning of the lifecycle of the vessel were
presented. With the adoption of such design improvements, the shipping industry will
be able to move forward into the age of digitalization with more confidence.
348 S. Lim et al.

References
1. Bows-Larkin, A.: All adrift: aviation, shipping, and climate change policy. Clim. Policy
15(6), 681–702 (2015)
2. Carchen, A., Pazouki, K., Atlar, M.: Development of an online ship performance monitoring
system dedicated for biofouling and anti-fouling coating analysis. In: 2nd Hull Performance &
Insight Conference (HullPIC 2017), Newcastle University, United Kindgom (2017)
3. Bhatti, J., Humphreys, T.E.: Hostile control of ships via false GPS signals: demonstration and
detection. Navigation: J. Inst. Navig. 64(1), 51–66 (2017)
4. Armstrong, V.N., Banks, C.: Integrated approach to vessel energy efficiency. Ocean Eng. 110,
39–48 (2015)
5. Lim, S., Pazouki, K., Murphy, A.J.: Holistic energy mapping methodology for reduced fuel
consumption and emissions. In: ASME 2017 36th International Conference on Ocean,
Offshore and Arctic Engineering (V07BT06A030-V07BT06A030), Norway (2017)
6. Maritime Safety Committee (MSC). 100th session, 2018. International Maritime Organisation.
http://www.imo.org/en/MediaCentre/MeetingSummaries/MSC/Pages/MSC-100th-session.
aspx. Accessed 06 Mar 2010
7. DNV GL. Technology Outlook 2025: Shipping (2015)
8. Lim, S., Pazouki, K., Murphy, A.J., Younessi, S., Graham, N.: Systematic rationale for
assessing energy flow across the entire vessel operation. In: Energy Efficient Ships 2016,
London, United Kingdom (2016)
9. Lim, S., Pazouki, K., Coraddu, A., Murphy, A.J.: Standardisation of data monitoring on-board
vessels. In: 12th Conference on Sustainable Development of Energy, Water and Environment
Systems, Croatia (2017)
 Ice Resistance Calculation in Pack Ice
Conditions

Guiyong Zhang1,2,3(&), Biye Yang1, Yuyan Jiang1, and Zhe Sun1

1
Liaoning Engineering Laboratory for Deep-Sea Floating Structures,
School of Naval Architecture, Dalian University of Technology,
Dalian 116024, People’s Republic of China
[email protected]
2
State Key Laboratory of Structural Analysis for Industrial Equipment,
Dalian University of Technology, Dalian 116024, People’s Republic of China
3
Collaborative Innovation Center for Advanced Ship and Deep-Sea
Exploration, Shanghai 200240, People’s Republic of China

Abstract. With the decline of the Arctic sea ice cover, more and more marine
transportation and oil production are moving into ice-covered waters. Pack ice
conditions may be the most common scenario for the ice-going ships in air-
worthiness seasons, except the icebreaker. During the ship advancing in pack ice
conditions, the ice pieces will contact with the hull, interact with the water and
even break or crush due to the bend stress, which is a complex process for the
numerical simulation. In this study, the ice resistance of ship under the pack ice
condition has been estimated using Multi-Material Arbitrary Lagrangian Eulerian
formulation and penalty-based fluid structure interaction technique in
LS-DYNA. Emphasis in present research has been placed on the hydrodynamic
loads occurring during the ship-ice interaction in the numerical model without
significant simplification. Validation of the employed fluid structure interaction
model was given, especially focusing on the buoyancy and hydrodynamic
behaviors of an ice cube. The ice pieces and the ship hull were assumed to be
rigid. In addition, an ice filed generation algorithm to generate randomly dis-
tributed pack ices was adopted. Finally, the ice resistance of an oil tanker has
been calculated and validated by comparing with the empirical formula. More-
over, the necessity to consider water effect in the interaction between ship and ice
was discussed. When the hydrodynamic influence of ship is neglected, the ice
resistance will be large obviously. In order to simulate the fracture behavior of
ice, the material model such as cohesive element model was also considered.

Keywords: Ice resistance
Pack ice condition
Cohesive element model

1 Introduction

Oil production and marine transportation in the Arctic have to face the extreme envi-
ronment conditions, such as the cold temperature and the harsh sea state. The appearance
of the ice floe increases the risk of the operation and navigation in the polar region.
Although there has been a long time since the first scientific publication of the icebreakers
[2], the prediction of the ship’s performance in ice is still imperfect. With the deepening

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 349–362, 2021.
https://doi.org/10.1007/978-981-15-4680-8_25
350 G. Zhang et al.

global warming, the solid level ice and iceberg melt significantly. Large area of broken ice
or pack ice gradually diffuses in the ship route. Compared with the interaction with the
level ice or iceberg, most ice-going ships’ structures are able to withstand the small-size
broken ice’s load. However, navigation in the pack ice field with the same propulsion and
speed as in the open water is still dangerous. The ship will be fully embedded into ice
when a large number of ice pieces accumulates around the hull. Therefore, it is necessary
to investigate the ship’s performance in pack ice conditions.
Numerous studies have been performed to calculate the resistance in level ice
conditions. But there are only a few investigations about the pack ice or broken ice.
Kim conducted synthetic and refrigerated ice model tests and the corresponding
numerical simulation about the ship’s resistance in pack ice conditions [3, 4]. The
feasibility of the synthetic ice was validated and the influence of the hull form was also
discussed. The simulation results are in good agreement with the experiment qualita-
tively. A series of model-scale and full-scale tests was performed by the Institute for
Ocean Technology in Canada for over thirty years [5–7]. The test and analysis methods
were summarized and the correlation formula between model-scale and full-scale was
also derived. The discrete element method has been widely applied in the interaction
between the ship and the broken ice [8–10]. Recently, considering the efficiency and
real-time simulation, Metrikin [11], Yulmetov [12] and van den Berg [13] has predicted
the ship’s performance in broken ice conditions with the non-smoothed discrete ele-
ment, which has been widely used in computer games. However, the hydrodynamic
model in the DEM is oversimplified and the material failure model is generally from
the perspective of experience.
In this study, the resistance of an oil tanker in the pack ice field was simulated with
LS-DYNA. The hydrodynamic effect was considered and the calculated results was
compared with the empirical formulas. In order to simulate the ice failure’s behavior,
the cohesive element model was introduced as a preliminary attempt.

2 Numerical Simulation of an Oil Tanker in the Pack Ice

Conditions

2.1 The Fluid Structure Interaction Method

The Multi-Material Arbitrary Lagrangian-Eulerian description has been widely applied
to the computation of the fluid structure interaction problem in LS-DYNA. The
equations of mass, momentum and energy conservation can be described as:

@q ! !
þ qr
!v þ ð!v ! w Þ
ðr qÞ ¼ 0
@t
@! ! !
þ q½ð!
v ! w Þ
r !
v ¼ r r þ q!
v
q g ð1Þ
@t
@e ! !
q þ q½ð!
v !w Þ
r e ¼ ðr : r Þ!v þ q!
g
!
v
@t
 Ice Resistance Calculation in Pack Ice Conditions 351

* *
where q is the density of the fluid, r is the Cauchy stress, e is the energy, v and w
represent the material velocity and mesh velocity respectively. The couple force is
calculated using the penalty method. During a calculation timestep, once the structure
node penetrates the interface of the interaction, there will be a couple force pushing the
node back to the structure’s surface. The spring-like force can be calculated by:

F ¼ kd ð2Þ

where d is the penetration distance and the numerical stiffness k can be described as:

KA2
k ¼ pf ð3Þ
V
where K is the bulk modulus of the fluid element, V is the fluid element’s volume and
A is structure element’s area locating at the coupling region. pf is the scale factor which
is recommended as 0  pf  1. More details can be found in Aquelet’s research [14].

2.2 Description of the Numerical Model

In this study, we calculated an oil tanker’s resistance in the broken ice field. The oil
tanker’s principal dimensions and parameters are summarized in Table 1. A snapshot
of the hull is given in Fig. 1.

Fig. 1. The oil tanker model

Table 1. Principle dimension of the oil tanker

Length (m) Beam (m) Draft (m) Block coefficient
180 32.9 11.4 0.807

As the deformation of the ship’s structure is the secondary consideration, the hull
was assumed to be rigid in this simulation. The influence of the interaction between the
ship and ice floes was also assumed to be ignored. The resistance in the surge direction
is the main objective in this study. When the small size ice floes contact with the ship or
the structure, the failure or deformation of the ice floe is small. Most of the floes
generally slide or flow around the hull. Due to the complex of the nature sea ice, the ice
352 G. Zhang et al.

material model is still unreliable, although some significant developments have been
made recently. Furthermore, using the rigid material model can to some extent save the
time cost. Therefore, the ice material was also assumed to be rigid in this simulation.
The ice material’s property is shown in Table 2.

Table 2. Ice material properties

3
Density (kg/m ) Young’s modulus (Gpa) Poisson’s ratio Size (m) Thickness (m)
870 9 0.3 4–7 0.6

Th calculation region was set as shown in Fig. 2. The length of the total calculation
region was set 600 m and the length of the ice field was set about 350 m. There was a
distance from the ship and the ice filed at the beginning of the simulation, in order to
guarantee that the ship accelerates to a constant velocity when the ship contacts with
the ice floes. The width of the channel was set about 3 times of the ship’s breadth. An
additional fence was set at the boundary of the calculation region to keep the ice floes
moving in the channel. The influence of the channel width is still need further research.
Moreover, at the boundary of the fluid calculated domain, a boundary condition called
reservoir boundary was applied to avoid the wave reflection.

90m

600m

Reservoir
Boundary

Fence

Fig. 2. The numerical domain of the simulation

The ice floes were distributed randomly in the broken ice field. The cutting convex
polygon method which was proposed by Metrikin [11] was introduced in this study.
The convex polygon with arbitrary edge number can be produced randomly. The ice
 Ice Resistance Calculation in Pack Ice Conditions 353

floes were packed in the calculation region randomly. Meanwhile, the detection of
intersection was applied to avoid the overlap of floe-floe and floe-boundary. The ice
concentration was set as the termination criterion of the packing algorithm. In this
simulation, the ice floe shape was assumed as quadrilateral and the size was about 6 m
to 8 m. The effect of the ice floe shape and size were not considered in this study.
The Multi-Material Arbitrary Lagrangian-Eulerian formulation was used to solve
the fluid-structure interaction (FSI) in this simulation. The Volume of Fluid
(VOF) model was applied to simulate the nonlinear free surface and the FSI force was
calculated by the penalty coupling method. For the contact calculation, the contact
option between ice and ice was set as AUTOMATIC_SINGLE_SURFACE and the
interaction between ship and ice was set as AUTOMATIC_SURFACE_TO_
SURFACE.

2.3 Results and Comparison with Empirical Formula

The simulation was conducted at the ship velocity range from 2 m/s to 6 m/s and the
initial ice concentration was set 50%. Figure 3 shows the snapshot of the interaction
between the ship and the ice floes. The ice floes collided with the ship and some of
them rotated and accumulated near the hull as shown in Fig. 4. The scenario occurred
in the simulation is similar to the experiment conducted by Zhou [15]. In order to valid
the accuracy of the simulation, we compared the mean value of the ice resistance with
the empirical formulas.

Fig. 3. The ship moves through the pack ice field with a constant speed of 5 m/s
354 G. Zhang et al.

Fig. 4. The interaction between ice floes and ship

Kashtelian’s Empirical Formula

Kashtelian et al. proposed a well-known semi-empirical formula to calculate the ice
resistance when the ship moves through a broken ice channel [16].
pffiffiffiffiffiffiffiffi B L
Rice ¼ 10½qi bf hi ð Þk1 ð1 þ 2laW Þ
2 B ð4Þ
þ k2 qi bf hi Bðl þ aW tan a0 ÞFr þ k3 qi bf hi L tan2 a0 Fr 2 

where:
Rice is the only ice resistance (without the open water resistance);
qi is the density of the ice floe;
bf is the ice floe size;
hi is the ice floe thickness;
B is the beam of ship;
L is the length of ship;
l is the coefficient of friction between ice floe and ship hull;
aw is the waterline area coefficient;
a0 is half of the entering angle of the bow part of the ship;
Fr is the Froude number;
k1, k2 and k3 are empirical coefficients;
 Ice Resistance Calculation in Pack Ice Conditions 355

DuBrovin’s Empirical Formula

Dubrovin performed a series of ice model test including the icebreakers and cargo ships
[17]. A semi-empirical formula was established and verified with the full-scale test.

Rice ¼ p1 A þ p2 UFrn ð5Þ

where:
p1 and p2 are empirical coefficients which are influenced by the ice concentration
and the channel width;
n is the powering coefficient depending on the ship form;
A and U are defined as:

1 pffiffiffiffiffiffiffiffi L
A ¼ B2 bf hi qi ð1 þ 2 laH Þ ð6Þ
4 B
L
U ¼ hi bf qi B½l þ tan a0 ðaH þ tan a0 Þ ð7Þ
B

aH is the prismatic fore-body coefficient;

Aboulazm’s Analytic Formula
Aboulazm proposed an analytic formula based on the conservation of linear momentum
between the ship and the ice floe [18].
pffiffiffiffiffiffiffiffi
rice ¼ qw V 2 sin2 hðCCs y cos h=ð1  CÞ þC=khw Cf Þdy
pffiffiffiffiffiffiffiffi
þ qw V l sin hðCCs y cos h=ð1  CÞ þ C=khw Cf Þdx
2 2 ð8Þ
þ qi V hi ðC=ð1  CÞÞ sin h cos hydy
2 2

where:
rice is the ice resistance of the unit segment;
qw is the density of the sea water;
Cs is the skin friction coefficient;
Cf is the form drag coefficient;
hw is the draft of ice floe;
C is the ice concentration;
V is the ship velocity;
h is the slope of tangent to waterline entrance curve with the x axis;
k is the ice floe form factor;
356 G. Zhang et al.

Fig. 5. Comparison of ice resistance between numerical and empirical

Figure 5 shows the comparison of the mean ice resistance varying with the ship
velocity between the numerical results and the empirical and analytic formulas. As the
curves show, the difference between the empirical formulas is huge. Kashtelian and
Aboulazm’s formulas are mainly considered in two dimensions. The submergence
force is not involved in these formulas. Therefore, the calculated values are relatively
small. The present results are close to the Dubrovin’s quantitatively, which was totally
based on the model test’s regression curves. From the qualitative perspective, the
tendency of the ice resistance increasing with the ship velocity is close to the
Dubrovin’s and Kashtelian’s, but different from the Aboulazm’s. The Aboulazm’s
formula is totally via the analytical derivation in 2D and the influence of the channel
width is not considered in this formula. It also should be noted that the ice resistance in
the numerical simulation did not always increase with the ship velocity, which has also
been found in the experiment [19]. The reasons causing the phenomenon still need to
be investigated.

3 Preliminary Simulation of the Cohesive Element Model

3.1 Cohesive Element Model

The material model is always the secondary during the collision between ship and the
broken ice in the numerical simulation. However, when the size of the ice floe is large
enough, the deformation and the failure of the ice floe have to be considered.
Cohesive element model (CEM) which was firstly proposed by Barenblatt [20, 21]
and Dugdale [22] in order to simulate the fracture process, has been widely applied to
simulate study ice sheet-structure interaction process and got more and more attention.
 Ice Resistance Calculation in Pack Ice Conditions 357

As a preliminary attempt, we simulated the interaction between the level ice and the
structure.
The ice sheet is modelled with two kinds of elements, bulk element and cohesive
element. The bulk element is generally the eight-node brick element and the cohesive
element whose thickness is minimal but limited [23] is inserted between the bulk
elements. These two elements are connected together by sharing the same nodes.
Figure 6 illustrates the mesh configuration of the cohesive element model.

(a) bulk elements (b) cohesive elements (c) connections of elements[1]

Fig. 6. Ice sheet model

Due to the different modeling purpose, these two elements have different material
properties. The bulk elements are defined by the curve of yield stress-effective plastic
strain and the cohesive elements follow the traction-separation law (TSL) curve. These
mechanical property curves are shown in Fig. 7 respectively. The curve in Fig. 7(b)
can expressed by:
8 h i
>
> T0 2ðdd1 Þ  ðdd1 Þ2 for d\d1
<
T¼ h T0 for d \d\d
1 i 2 ð9Þ
>
>
: T0 2ð dd2 Þ3  3ð dd2 Þ2 for d2 \d\d0
d1 d2 d0 d2

where T is the traction force, d is crack-opening displacement, T0 is the maximum

traction force, d0 is the maximum crack-opening displacement, d1 and d2 are both
special values of crack opening accordingly. In the first stage of loading, T increases
with d until the crack-opening displacement reaches d1, which means that the ice sheet
starts to damage. Then T maintains the same value T0 while the crack-opening dis-
placement increases from d1 to d2 and finally T goes down to zero with d developing.
Meanwhile, the cohesive elements fail completely and the nodes of bulk elements
which are attached to the cohesive elements originally are free.
358 G. Zhang et al.

(a) stress-strain curve (b) TSL curve

Fig. 7. Mechanical property curves

3.2 Numerical Simulation Analysis

Simulation of Cylinder-Level Ice Collision

A simple case about the collision between ice and cylinder structure using the CEM is
presented in this section. The configurations are suggested by Feng [24]. For ice sheet,
bulk element’s material property curves used in the simulation can be found in Fig. 7.
*PIECEWISE_LINEAR_PLASTICITY was set as the bulk element’s material and
*COHESIVE_GENERAL was set for the cohesive element. The main parameters of
the ice material are summarized in Table 3.

Table 3. Main parameters of the ice material

Bulk element Cohesive element
Items Value Items Value
Density (kg/m3) 910 Maximum traction force (Mpa) 1.04
Elastic modulus (Gpa) 5.0 d1 =d0 0.1
Poisson’s ratio 0.3 d2 =d0 0.8
Yeild stress (Mpa) 2.0 Fracture energy (N/m) 5200

*CONTACT_ERODING_SINGLE_SURFACE was set for the interaction between

the ice bulks and *CONTACT_ERODING _SURFACE_TO_SURFACE was used for
the contact between ice and the structure. The column moved at a constant velocity of
0.15 m/s. A plate was placed under the ice sheet to simplify the buoyance effect. As the
Fig. 8 shows, during the interaction between the column and level ice, the accumu-
lation of ice fragments existed around the front of the structure and the collision force
fluctuated varying with time. After reaching the peak value at around 4 s, the collision
force entered a relative steady state.
 Ice Resistance Calculation in Pack Ice Conditions 359

(a) Ice floes accumulate (b) Collision force

Fig. 8. Column-level ice collision results

Simulation of Ship-Ice Sheet Interaction

In this section, the interaction between an oil tanker and the level ice was investigated
using CEM. The tanker was set to move along the surge direction at a constant velocity
of 1.0 m/s. The ice sheet was set 100 m  100 m  1.5 m with the bulk elements of
0.5 m  0.5 m  0.5 m. The fixed and non-reflecting boundary was applied to the ice
sheet, as shown in Fig. 9. Figure 10 shows the behavior of the bulk elements and the
vertical cohesive elements during the collision process, separately. The cohesive ele-
ments generally failed and were deleted from the model at the contact region. The bulk
elements accumulated at the bow and continued to contact with the hull. Besides the
cracking, the splashing and crushing phenomenon of fragments could also be observed
in the numerical simulation.

Fig. 9. Numerical configuration of the ship-level ice interaction

360 G. Zhang et al.

(a) Bulk element (b) Vertical cohesive element

Fig. 10. Elements’ behavior during the collision

The contact force between the ship and the ice sheet was recorded during the
tanker’s advancing, as shown in Fig. 11. The curve fluctuated with a roughly growing
trend until the bow was fully embedded into the ice sheet. The validation of the
numerical results still needs further research and a more complicated hydrodynamic
model will be introduced in the further model.

Fig. 11. Contact force-duration curve

4 Conclusion

The pack ice resistance was calculated considering the hydrodynamic effect in
LS-DYNA. The behavior of the ice floes during the collision in the simulation is
similar to the experiment qualitatively. Some empirical and analytical formulas were
listed and estimated the pack ice resistance varying with the ship velocity. The mean
value of the surge resistance in the simulation fits the Dubrovin’s formula well. The
trend of the numerical results is close to the empirical formulas but there are some
differences from the analytic formula. A preliminary simulation of the interaction
 Ice Resistance Calculation in Pack Ice Conditions 361

between the ice sheet and the ship was also conducted. The ice sheet cracked at the
contact area and the broken ice pieces accumulated at the bow. From the phe-
nomenological perspective, the results seemed plausible. In the further study, the CEM
would be extended to the broken ice and more detailed hydrodynamic model should be
considered.

References
1. Pang, S.D., et al. The modelling of ice-structure interaction with cohesive element method:
limitations and challenges. In: International Conference on Port & Ocean Engineering Under
Arctic Conditions (2015)
2. Runeberg, R.: On steamers for winter navigation and ice-breaking. In: Proceedings of
Institution of Civil Engineers, pp. 277–301 (1888/1889)
3. Kim, M.C., et al.: Numerical and experimental investigation of the resistance performance of
an icebreaking cargo vessel in pack ice conditions. Int. J. Naval Arch. Ocean Eng. 5(1), 116–
131 (2013)
4. Kim, M.C., Lee, W.J., Shin, Y.J.: Comparative study on the resistance performance of an
icebreaking cargo vessel according to the variation of waterline angles in pack ice
conditions. Int. J. Naval Arch. Ocean Eng. 6(4), 876–893 (2014)
5. Molyneux, W.D., Kim, H.S.: Model experiments to support the design of large icebreaking
tankers. In: Design and Construction of Vessels Operating in Low Temperature Environ-
ments, London, UK, pp. 45–53 (2007)
6. Wang, J., Jones, S.J.: Resistance and propulsion of CCGS terry fox in ice from model tests to
full scale correlation. Trans.-Soc. Naval Arch. Mar. Eng 116 (2008)
7. Derradji-Aouat, A., Thiel, A.: Terry fox resistance tests – phase III (PMM) testing ITTC
experimental uncertainty analysis initiative. Institute for Ocean Technology, National
Research Council (2004)
8. Ji, S.: Discrete element modeling of ice loads on ship and offshore structures. In:
Proceedings of the 7th International Conference on Discrete Element Methods, Dalian,
pp. 45–54 (2017)
9. Liu, L., Zhan, D., Spencer, D., Molyneux, D.: Pack ice forces on floating offshore oil and gas
exploration systems. In: International Conference and Exhibition on Performance of Ships
and Structures in Ice, pp. 264–274 (2010)
10. Lau, M., Lawrence, K.P., Rothenburg, L.: Discrete element analysis of ice loads on ships
and structures. Ships Offshore Struct. 6(3), 211–221 (2011)
11. Metrikin, I.: A software framework for simulating stationkeeping of a vessel in
discontinuous ice. Model. Ident. Control 35(4), 211–246 (2014)
12. Yulmetov, R., Lubbad, R., Løset, S.: Planar multi-body model of iceberg free drift and
towing in broken ice. Cold Reg. Sci. Technol. 121, 154–166 (2016)
13. van den Berg, M., Lubbad, R., Løset, S.: An implicit time-stepping scheme and an improved
contact model for ice-structure interaction simulations. Cold Reg. Sci. Technol. 155, 193–
213 (2018)
14. Aquelet, N., Souli, M., Olovsson, L.: Euler-Lagrange coupling with damping effects:
Application to slamming problems. Comput. Method Appl. M. 195(1–3), 110–132 (2006)
15. Zhou, L., Ling, H., Chen, L.: Model tests of an icebreaking tanker in broken ice. Int. J. Naval
Arch. Ocean Eng. 11(1), 422–434 (2019)
16. Kashtelian, V.I., Poznyak, I.I., Ryvlin, A.Ia.: Ice resistance for ship motion. Sudostroenie,
Leningrad (1968)
362 G. Zhang et al.

17. Dubrovin, O.V., Aleksandrov, M., Moor, R.: Calculation of broken ice resistance based on
model testing. Translation Department of Navel Architecture and Marine Engineering,
University of Michigan (1970)
18. Aboulazm, A.F.: Ship resistance in ice floe covered waters. Memorial University of
Newfoundland (1989)
19. Luo, W.-Z., Guo, C.-Y., Wu, T.-C., Su, Y.-M.: Experimental research on resistance and
motion attitude variation of ship-wave-ice interaction in marginal ice zones. Mar. Struct. 58,
399–415 (2018)
20. Barenblatt, G.I.: The formation of equilibrium cracks during brittle fracture. General ideas
and hypotheses. Axially-symmetric cracks. J. Appl. Math. Mech. 23(3), 622–636 (1959)
21. Barenblatt, G.I.: The mathematical theory of equilibrium cracks in brittle fracture. Adv.
Appl. Mech. 7, 55–129 (1962)
22. Dugdale, D.S.: Solids, yielding of steel sheets containing slits. J. Mech. Phys. Solids 8(2),
100–104 (1960)
23. Gürtner, A.: Experimental and numerical investigations of ice-structure interaction (2009)
24. Feng, D., Pang, S.D., Zhang, J.: Parameter sensitivity in numerical modelling of ice-structure
interaction with cohesive element method. In: ASME International Conference on Ocean
(2016)
 On the Development of the Arc7 LNG Carrier
– Hull Form Development

Ho-Jeong Lim(&), Ji-In Park, Sahng-Hyon Lee, Sung-Pyo Kim,

and Joong Kyoo Kang

Department of Ship & Ocean R&D Institute, Daewoo Shipbuilding & Marine
Engineering Co., LTD., 96, Baegot 2-ro, Siheung-si, Gyeonggi-do 15011, Korea
[email protected]

Abstract. This paper focuses on the brief history and some fundamental
technical issues which are raised during the hydrodynamic design development
of DSME Arctic LNG carrier for Yamal area.
In 2010, DSME investigates ship design concept and economic aspects for
shuttle LNG transportation services from Yamal peninsula to Murmansk. In this
study, Arc7 class 85,000 m3 shuttle LNG carrier which is driven by twin pod
propulsion system has been developed. The economic evaluation for round trip
with three bow design alternatives (bulbous bow, moderate ice bow and extreme
ice bow) have been made and found that the extreme type of bow variant can
transport the highest annual volume of cargo.
The design development for 170,000 m3 class Arctic LNG carrier was
undertaken early in 2011. Various types of bow including bulbous bow type
were compared based on the model tests and, the performance comparison
between different propulsion concepts. After the newbuilding contract for
Yamal LNG in 2013, a set of challenging target for open water speed perfor-
mance and ice capabilities are given. In order to comply with the target, a hull
form optimization using utmost advanced CAE technologies were carried out
considering multi-operation profiles, i.e. laden and ballast draft conditions
including speed variations. In addition to the optimization for conventional open
water, bow and specially designed stern shapes are also optimized in view point
of ice breaking abilities by applying the constraints of hull form characteristics
to maintain the target ice capabilities. And those full scale open water and ice
breaking performances are finally well confirmed in official sea trials and ice
trial.

Keywords: Arc7
LNG
Shuttle
Hull form

1 Transport Economic Analysis of 85,000 m3 Shuttle LNG

Carrier

Since 2010, with the good example of 85k Shuttle LNGC development shown in
Fig. 1, special technical issues (STI) were deeply investigated by technical cooperation
with Aker Arctic Technology (AAT 2010).

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 363–372, 2021.
https://doi.org/10.1007/978-981-15-4680-8_26
364 H.-J. Lim et al.

Fig. 1. DSME 85,000 m3 shuttle LNG carrier

In view point of dimension of the vessel, major dimension parameter such as beam,
L/B and block coefficient were investigated for the influence on ice performance
including turning ability. For the route A and B in Fig. 1, three different bow shapes
(bulbous bow, moderate ice bow and extreme ice bow) shown as in Fig. 2 and five
kinds of propulsion systems in Table 1 were evaluated in consideration of summer and
winter season operation as shown in Fig. 3.
From the economic analysis results, the most favorable configurations such as
extreme ice bow and POD propulsion system in consideration of the Kara sea operation
were finally selected.

Fig. 2. Bow type

 On the Development of the Arc7 LNG Carrier – Hull Form Development 365

Table 1. Comparison of machinery and propeller alternatives

Machinery Response Power Fuel Fuel Emission Layout Machinery Investment Manoeuvring Existence
alternatives to ice flexibility consumption flexibility SOx flexibility weight cost flexibility
loads
Steam 3 2 1 5 4 1 2 4 3 1
turbine
Slow speed 2 1 5 2 1 1 2 3 3 2
diesel with
CPP
Geared 2 3 4 2 3 1 3 3 3 2
medium
speed with
CPP
DF D-E 4 4 4 4 4 3 2 2 3 3
with
shaftlines
DF D-E 5 5 4 4 4 4 2 2 5 4
with podded
drives
Scale: 5-excellent; 4-good; 3-medium; 2-tolerable; 1-poor

Fig. 3. Economic evaluation for different bow types in Kara Sea

2 Hull Form Optimization

As the pre-development study for 170,000 m3 ARC7 classed LNG Carrier, bow shape
and propulsion type study were undertaken since 2011 referring to the previous
research outcomes.
After the contract award for 15 vessels in 2012, main tasks were completed to fully
meet the specified requirements such as the typical design speed of 19.5 knots in open
water and up to 5 knots of ice navigation. Especially performance balancing between
open water and ice performance was mainly focused and successfully completed via
assistance of the most advanced optimization tool and model tests (Fig. 4).
366 H.-J. Lim et al.

Fig. 4. Optimization strategy in Yamal project

2.1 Bow Shape Study

A comparative model testing with two different bow designs was carried out for open
water performance and ice navigation performance (SSPA 2014; VTT 2014; AARC
2014).
The bow F1 is a conventional bulbous bow type. The F1 is designed to have higher
ridge of the bulbous bow nose than normal one to have an upward ice breaking pattern
as shown in Fig. 5. The bow F2 is designed to have an extreme ice bow to have a
downward ice breaking pattern.

Fig. 5. Fore body configuration

At the model tests in open water condition, the bulbous bow F1 shows much better
wave pattern with approximately 15% lower resistance than F2. During the level ice
tests in ahead moving condition we found the bulbous bow creates a high ice resistance
due to complicated turning over of broken ice pieces flow along the ship side as shown
in Fig. 6.
 On the Development of the Arc7 LNG Carrier – Hull Form Development 367

Fig. 6. Ice breaking pattern of bulb type bow

The Ice bow F2 was finally confirmed as more suitable concept for this type of high
Arctic class vessels via performance comparison based on the test results shown in
Fig. 7.
For the comparison tests, the hybrid propulsion having two PODs and one center
fixed pitch propeller were equipped as the propulsion configurations.

Fig. 7. Bow shape study (bulbous bow vs. ice bow)

2.2 Stern Configuration with Propulsion System

When we design the stern shape including POD arrangement for DAS concept vessel,
the ice-hull-POD interactions in astern operation mode have to be carefully considered
keeping good resistance in normal ahead operation mode.
The POD mounting height is one of the important design parameters. Sometimes
the head box of POD unit become too large to avoid direct ice contact to the body of
POD units.
368 H.-J. Lim et al.

In view point of astern ice breaking and turning performance, stern configuration
variations were studied in combination with different propulsion systems as shown in
Fig. 8. A1 model is designed to have hybrid propulsion having twin POD units with
conventional driven center propeller. The aftbody A2 is three POD propulsion concept.
From this study, A2 concept was confirmed as the excellent design considering
open water and ice performance especially for ice turning ability of the vessel, and
therefore it was selected as the basic stern configurations in this project.

Fig. 8. Stern configuration study (2PODs + center FPP vs. 3PODs)

2.3 Performance Balancing Between Open Water and Ice Covered Water
Since the 19.5 knots of open water ship speed is very tough target for this kind of high
arctic class vessel, the Bow concept was optimized to meet the target performance in
step wise design process. The Bow shape is finally reached to a moderate type of ice
bow F3 as shown in Fig. 9.
Considering the following major design parameters for good ice going performance
in astern mode are listed below, A2 hull form was designed.
– Stern profile angle to the level ice
– Ice-Hull-POD interaction
Finally F3 combined with A2 hull form (F3A2) have been confirmed that both
performances had been achieved simultaneously.
 On the Development of the Arc7 LNG Carrier – Hull Form Development 369

Fig. 9. Performance balancing between open water and ice navigation performance (F3A2)

In order to make a further improvement based on the overall required operational

profile by well balancing both performances, i.e. improving ahead open water per-
formance and astern ice breaking performance, especially level ice 1.5 m thickness was
tried with F4A3 hull form shown as in Fig. 10 that was selected as the best alternative
among the several candidates based on the CFD results and its performance finally
confirmed at model tests.

Fig. 10. Performance balancing between open water and ice navigation performance (F4A3)
370 H.-J. Lim et al.

2.4 Sensitivity Check of Increased Displacement

Lastly but not least item for volume adjusting in the final stage, further optimization of
minimization of defects on the performance from the increased volume was intended
and turned out at fore body hull form (F5) using by the advanced optimization tech-
nique, Friendship system and VOF-RANS as a performance evaluation tool. Hull form
variations were carried out with fully parametric model by combination of 30 kinds
form parameters as the major control variables such as FOS, FOB, CPC etc. shown as
in Fig. 11. For the minimum resistance and maximum displacement as the objective
function with design speed at design draft as the considered operational profile, several
promising candidates were picked up among 40,000 variants via pareto plot shown as
in Fig. 11 (DNV-GL 2013). The chosen alternatives had been checked the detail
performance and chosen the best one to be confirmed through model testing.
Figure 11 also shows design constraints which were controlled two categories; the
one (yellow) is hard condition for reserving the cargo hold arrangement with minimum
clearance and the other one (red) is soft condition for reserving optimized ice perfor-
mance. As a result of the optimization, it was confirmed as the relatively successful one
that just 1.6% performance reduction by 1.1% increased displacement.

Fig. 11. Sensitivity check of increased displacement (F5A3)

2.5 Full Scale Performance Confirmation

All of the optimized results were confirmed in full scale via open water and ice
performance trial. Open water sea-trial had been conducted on 2016 at southern part of
Korea as Fig. 12. The sea state was around Beauport 3–4 and speed trial performed at
least a double run for the each measurement load point to have an accurate trial results.
As a result, we were satisfied the 19.5 knots successfully which results was proved
again by sister vessels speed-trial. The major tested items described as Table 2 and the
all of the results were satisfied design target.
 On the Development of the Arc7 LNG Carrier – Hull Form Development 371

Fig. 12. Open water speed trial (top) and trial area (bottom)

Table 2. Full scale open water performance evaluation

No Test item Target Result
1 Speed (Ahead) 19.5 knots Satisfied
2 Turning Circle (port/stbd) 5.0 L Satisfied
3 [10/10] Z-Maneuvering (port/stbd) 20° Satisfied
4 [20/20] Z-Maneuvering (port/stbd) 25° Satisfied
5 Crash Stop 15 L Satisfied

In 2017, the ice performance trial had been carried out at Russian arctic area. Ahead
and astern ice going performance, ridge penetration, and ice turning performance was
evaluated during the trial as Fig. 11 and all of the ice performance was successfully
satisfied design target described as Table 3. In 2018, ice performance was proved again
completely by 2nd vessel’s ice trial (Fig. 13).

Table 3. Full scale ice performance evaluation

No Test item Target Result
1 Level Ice Ahead 2 kts at 1.5 m Level ICE Satisfied
2 Level Ice Astern 5 kts at 1.5 m Level ICE Satisfied
3 Astern Level Ice Turning 3 km (Dia.) at 1.7 m Level ICE Satisfied
4 Ridge Penetration Keel depth 15 m Satisfied
372 H.-J. Lim et al.

Fig. 13. Ice performance trial

3 Conclusions

For Yamal project, DSME pre-investigated transport economic analysis for 85k
Shuttle LNGC with the several bow shapes and propulsion systems in Kara Sea and
then successfully completed the optimization work for 170k ARC7 LNGC with
moderate ice bow and three PODs system as the basic configurations in order to fully
meet simultaneously both open water performance and ice navigation performance as
required in contracted specifications.
During optimization process, the most advanced optimization technology was
applied to improve both performances.
In 2016 and early of 2017, open water and ice performance trial had been suc-
cessfully completed and confirmed vessel’s performance well match the design target.

References
Aker Arctic Report K-129: Study on special technical issues related to the design of vessels for
service in the Artic regions, Helsinki (2010)
Aker Arctic Report A501: Development of 3-podded Yamal Arctic 170,000 cbm LNG carrier for
DSME, Helsinki (2014)
DNVGL Report 131216: ECO-Lines for an arctic LNG-carrier, Hamburg (2013)
SSPA Report RE20136785-01-00-A: DSME/Aker Arctic Arc7 LNGC towing tank tests,
Gothenburg (2014)
VTT Report VTT-CR-08647-13: Performance tests for DSME/Aker Arctic Arc7 LNG-carrier,
Helsinki (2014)
 A New Technique for Prediction of Ship-Ice
Floes Interaction Without Fluids

Donghwa Han1, Joonmo Choung1(&), and Kuk-Jin Kang2

1
Department of Naval Architecture and Ocean Engineering,
Inha University, Incheon, Korea
[email protected]
2
Korea Research Institute of Ships and Ocean Engineering (KRISO),
Daejeon, Korea

Abstract. As global warming reduces Arctic sea ice level, Russia’s Arctic
resource development is taking place in earnest. For this reason, it is important
to determine the structural safety at the initial stage of the ship design, because
the ship operating at the Arctic route has large loads to the ship hull. For this
purpose, many researches have been conducted in various ways including
analytical methods and model tests. Analytical methods are used in two ways.
One is ice resistance of ship, and the other is structural crashworthiness. The
approaches of ice resistance prediction of a ship are is including empirical
formulas, multi-body dynamics (MDB), and discrete element method (NDEM).
Structural crashworthiness can be estimated using empirical formulas and FEA
simulation.
In this study, a new approach is given considering both ice resistances of ship
and structural crashworthiness without modeling fluids, simultaneously. There-
fore, analysis cost has reduced due to eliminating fluid consideration. The
hydrostatic and hydrodynamic forces acting on ice floes are realized by directly
controlling the real time hydro-forces during the ship-to-ice collision analysis.
The procedure is achieved through a user defined-subroutine.
The developed user-subroutine was applied to the towing test simulations of
Korean icebreaking research vessel ARAON. As a result, ice-induced resistance
was calculated and compared with the experimental results.

Keywords: Pack ice
Ice resistances
User-subroutine
Drag-force

Buoyancy

1 Introduction

As global warming reduces Arctic sea ice, Russia’s Arctic resource development is
taking place in earnest. In recent years, Russia has successfully constructed
Yamal LNG vessels, interest in the Arctic route has been increasing as a result of the
use of the Arctic Ocean as a means of shipping and transportation, which saves time
and money compared to the existing Suez Canal. Ships operating on the Arctic route
are exposed to various ice conditions such as collision with ice and friction. For this

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 373–382, 2021.
https://doi.org/10.1007/978-981-15-4680-8_27
374 D. Han et al.

reason, it is important to determine the ship safety at the initial stage of the ship design,
because the ship operating at the Arctic route has a large resistance to the hull. For this
purpose, research is being conducted in various ways including model tests, empirical
equation and analytical methods.
In the case of model tests, accurate results can be obtained, but there are many
disadvantages considering time, space, and economic constraints. The empirical
equations such as Shimansky (1938), Enkvist (1972) and Poznyak and Ionov (1981)
have the advantage of estimating the ice resistance in a short time, but there is a
disadvantage that the modeling uncertainty is large, and the result analysis is difficult
depending on the type of ship.
Among analytical methods, there are methods such as multi-body dynamics
(MBD), discrete element modeling methods (DEM), and non-smooth discrete element
modeling (NDEM). MBD is a useful method for analyzing the dynamic behavior of
rigid bodies with contact mechanisms and CFD which is to interpret the interaction
between fluid and rigid body (Kim et al. 2018), DEM can broadly be divided into two
main categories: smooth discrete element modelling (SDEM) and non-smooth discrete
element modelling (NDEM). The difference between the two can be seen as the dif-
ference between explicit and implicit time integration, allowing much larger time steps,
while maintaining stable simulations, when using NDEM (Raed et al. 2018).
These methods have the advantage of obtaining relatively accurate results and easy
analysis of results. However, these methods have also the disadvantage such that it
takes much time to define the shapes and characteristics of ships and ice, and thus takes
a long time to analyze. Therefore, in this study, both ice resistance and structural
collision were considered simultaneously, and the time was shortened by the analysis
method which does not model fluid.

2 Subroutine Verification

2.1 Development of the Subroutine

In this study, we developed a user subroutine supported by Abaqus/Explicit to apply
the modeling technique without explicitly considering fluid, and considered buoyancy
and drag force in real time.
If the buoyancy center of gravity-driven ice is below free surface (z = 0) as shown
in Fig. 1, buoyancy is calculated as shown in Eq. (1) using the z-coordinate infor-
mation of ice elements at each simulation time interval. At this time, buoyancy acts on
the pressure of each element, and the density of seawater is 1.025E−9 ton/mm3.

Pb ¼ qsea gz ð1Þ
 A New Technique for Prediction of Ship-Ice Floes Interaction Without Fluids 375

Fig. 1. Concept of buoyancy

When ice rotates or moves in Fig. 2, user subroutine retrieves normal vector (~ n) and
velocity vector (~ v) information from the center of each element of ice. The velocity
vector (v!n ) acting normal to the element is calculated through the inner product of the
normal vector (~ n) and velocity vector (~
v), and the drag force is calculated as given in
Eq. (2). In this case, when the angle between normal vector (~ n) and velocity vector (~
v)
exceeds −90° to 90°, the drag force does not act, and drag coefficient (Cd ) is 2.0, which
is the square drag coefficient suggested by DNV.

1
Pd ¼ Cd qsea v2n ð2Þ
2

Fig. 2. Concept of drag force

376 D. Han et al.

2.2 Assessment of Single Ice Motions

In order to verify the developed user subroutine, numerical analysis of heave motion
and roll motion of single ice motion performance was carried out. All numerical
analyzes were performed using commercial finite element analysis program
Abaqus/Explicit. The information of single ice is the same as Table 1 and modeled
with 480 rigid elements.

Table 1. Shape information of a single ice

Unit Value
Density of ice kg/m3 870
Density of sea kg/m3 1025
Breadth m 0.28
Length m 0.28
Depth m 0.057
Draft m 0.048

As shown in Fig. 3 (a), the heave motion moves in the z direction for numerical
analysis, and the heave motion converges to 0 as shown in Fig. 4 (a).
In order to confirm the effect of the roll damping, Fig. 3 (b), the numerical analysis
was performed by inclining 3° from the x axis. As shown in Fig. 4 (b), the roll motion
damping effect of a single ice was confirmed, and the performance of the single ice was
verified.

(a) Heave

(b) Roll

Fig. 3. Modeling of ice motions

 A New Technique for Prediction of Ship-Ice Floes Interaction Without Fluids 377

4.00 3.00
Heave motion Roll motion
3.00
2.00
2.00
Displacement (mm)

1.00 1.00

Degree (˚)
0.00
0.00
-1.00

-2.00 -1.00

-3.00
-2.00
-4.00

-5.00 -3.00
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Time (sec) Time (sec)

(a) Heave (b) Roll

Fig. 4. Results of ice motion

2.3 User Subroutine Verification

CEL (Coupled Eulerian Lagrangian) simulations were performed on a single ice to
verify user subroutines. The simulation proceeds with cores mesh and fine mesh as
shown in Fig. 5 and Table 2. The simulation was carried out for a total of 3.3 s as
shown in Fig. 6, and the effect of the hydro force was verified from 0.3 s after applying
the initial force for 0.12 s.

Fig. 5. Modeling of ice

378 D. Han et al.

Fig. 6. Simulation condition

Table 2. Information of a single ice

Dimension Value
Length L 1000.0 mm
Breadth B 500.0 mm
Depth D 500.0 mm
Draft T 250.0 mm
Shell thickness t 20.0 mm
Shell density qsteel 2.5625  10−9 ton/mm3
Sea density qsea 1.025  10−9 ton/mm3
Displacement D 128.125 kg
Drag coefficient Cd 2.0 & 3.0

Simulation results show that there is no mesh size effect as shown in Fig. 7 (a). In
Fig. 7 (b), it can be seen that CEL and the subroutine have many errors. In order to
check the error with CEL, the theoretical period and the comparison test were per-
formed. The theoretical result is 1.003 s, the CEL result is 1.710 s, and the subroutine
result is 1.020 s, confirming that the subroutine results are close to the theoretical
results. Also, the smaller element size, the closer to subroutine result.

Fig. 7. Simulation result

 A New Technique for Prediction of Ship-Ice Floes Interaction Without Fluids 379

3 Simulation of ARAON Model Test

3.1 Model Test Condition
ARAON model test was carried out at Korea Research Institute of Ships & Ocean
engineering (KRISO). The model test was conducted using the 18.667 scale, and the
experiments were performed as shown in Table 3. Table 4 and Table 5 show the
experimental conditions of model test and ARAON information. Figure 8 is the model
test performed at concentration 80%. Experimental data were collected in 1/100 s and
information such as the friction coefficient used in the experiment can be found in
(Jeong et al. 2017).

Table 3. Model test case of pack ice

Ice thickness (m) Concentration (%) Ship speed (knots)
0.057 60 1
3
5
80 1
3
5
90 1
3
5

Table 4. Model test conditions

Item Value
Channel width (m) 4.010
Density of ice (kg/m3) 875.000
Friction coefficient (Ice and Ship) 0.050
Friction coefficient (Ice and Ice) 0.081
Flexural strength (kPa) 62.000

Table 5. Principle dimensions of ARAON

Item Value
Length, O.A. (m) 111.0
Length, B.P. (m) 95.0
Breadth (m) 19.0
Design draft (m) 6.8
380 D. Han et al.

Fig. 8. Model test of ARAON

3.2 Simulation Condition

The simulations of the ARAON model tests were performed using Abaqus/Explicit.
The numerical model for the simulation is shown in Fig. 9, and the total length of pack
ice was modeled by 2.5 times of ARAON. ARAON and the pack ice were modeled
with 174,855 rigid elements. The drag coefficient used in the simulation is 2.0.

Fig. 9. Modeling of ARAON towing test in pack ice

3.3 Simulation Result

Table 6, Fig. 10 show the results of the ARAON model test and simulation results. At
the concentration 60%, the error rate was 0.79% at 3 knots. At the concentration 80%,
the error rate was −2.08% at 3 knots, and the error rate was 18.93% at 3 knots in
concentration 90%. The maximum error rate for each concentration occurred at 3 knots
of line speed, and the error rate tends to increase with increasing concentration. It is
judged to be an error caused by the same channel width depending on the concen-
tration. Also, the results are better than the empirical formulas.
 A New Technique for Prediction of Ship-Ice Floes Interaction Without Fluids 381

Table 6. Comparison of model test and simulation

Concentration Ship Ice resistance (N) Accuracy (%)
(%) speed Model Simulation Empirical Simulation Empirical
(knots) test
60 1 16.33 16.31 16.39 0.10 51.00
3 22.49 22.34 22.39 0.68 22.00
5 31.67 31.57 31.54 0.31 −11.00
80 1 36.92 36.25 36.17 1.81 −21.00
3 43.26 44.16 44.12 −2.08 −10.00
5 49.99 49.03 48.99 1.92 −9.00
90 1 42.36 35.84 35.80 15.39 −11.00
3 51.04 41.387 41.36 18.93 −6.00
5 61.06 69.66 69.60 −14.08 −8.00

80.0 80.0
Model test Model test
Simulation Simulation

60.0 60.0
Ice risistance (N)
Ice risistance (N)

40.0 40.0

20.0 20.0

0.0 0.0
0 2 4 6 0 2 4 6
Ship speed (kts) Ship speed (kts)

(a) Concentration 60% (b) Concentration 80%

80.0
Model test
Simulation

60.0
Ice risistance (N)

40.0

20.0

0.0
0 2 4 6
Ship speed (kts)

(c) Concentration 90%

Fig. 10. Comparison of model test and simulation (concentration 60%)

4 Conclusion

In this study, a new ice resistance estimation simulation method using a user subroutine
of a commercial finite element analysis code is presented. By not modeling the fluid
explicitly, the analysis time can be drastically shortened, and the results can be easily
analyzed by using commercial finite element analysis codes.
382 D. Han et al.

In order to realize the hydrostatic and dynamic forces, buoyancy and drag force are
considered in real time, and the damping effect is verified by heave motion and roll
motion simulation of a single ice. For further verification, the simulation of ARAON
water tank experiment was carried out and the simulation was performed according to
concentration of ice and speed of ship.
Simulation results are similar to the model test as a whole, but the maximum error
rate is 18.93% at a concentration of 90% and 3 knots. It is considered that this is due to
the influence of the channel width of towing tank and the modeling of regular pack ice.
In order to compensate for this error, channel width analysis and irregular pack ice
simulation are required.

Acknowledgements. This work was supported by Korea Environment Industry & Technology
Institute (KEITI) through Plant Research Project, funded by Korea Ministry of Environment
(MOE) (RE201901128) and “Development of hull form of year-round floating type offshore
structure based on the Arctic Ocean in ARC7 condition with dynamic positioning and mooring
system” through the Ministry of Trade, Industry, and Energy (MOTIE) (10063405, 2018).

References
Kim, J., Park, J.C., Kim, S.P., Kim, H.S., Cho, Y.G.: Multibody dynamics simulation on ice
resistance of ship in pack ice environment. In: Proceedings of the ASME 2018 37th
International Conference on Ocean, Offshore and Arctic Engineering, Madrid, Spain (2018)
Poznyak, I., Ionov, B.: The division of icebreaking resistance into components. In: Proceedings
of the 6th STAR Symposium, New York, US, pp. 249–252 (1981)
Read, L., Sveinung, L., Wenjun, L., Andrei, T., Marnix, N.: Simulator for arctic marine
structures (SAMS). In: Proceedings of the ASME 2018 37th International Conference on
Ocean, Offshore and Arctic Engineering, Madrid, Spain (2018)
Shimansky, Y.: Conditional Standards of Ice Qualities of a Ship, vol. 130. Northern Sea Route
Administration Publishing House (1938)
Spencer, D.: A standard method for the conduct and analysis of ice resistance model tests. In:
Proceeding of the 23rd American Towing Tank Conference, pp. 301–307. NRCC, Institute
for Marine Dynamics, New Orleans (1992)
 Study on Estimation of Ice Resistance
and Attainable Speed for Ship
of Arbitrary Shape

Hyun-Soo Kim1 , Donghwa Han2(B) , and Ali Erinc Ozden2

1
Department of Naval Architecture and Ocean Engineering, Inha Technical College,
Incheon, Korea
2
Department of Naval Architecture and Ocean Engineering, Inha University,
Incheon, Korea
[email protected]

Abstract. While operating on the Arctic route, ships may encounter

various issues. The ice environment, such as level ice, pre-sawn, pack
ice, ice ridge and brash ice, is one of the sources of those issues. There
are three different approaches to estimate the ice resistance: analytical
methods, numerical simulations, and using empirical formulas. In this
paper, empirical formulas are used for the estimation. The main aim
of the study is to develop a computer program (I-RES) that predicts
an attainable speed for ship of arbitrary shape and estimates the ice
resistance. To achieve this, first, various ice environment characteristics
such as the level ice, pre-swan, pack ice, brash ice and ice ridge were
analyzed. The results of I-RES were compared with Araon model test
results. Secondly, to obtain the information of the ship geometry, shape
information of arbitrary shape, igs file scaling work of the ship with
bulbous bow was performed, and ice resistance and attainable speed
were estimated.

Keywords: Ice resistance estimation · Empirical formula · Attainable

speed

1 Introduction
Since the sea ice in the Arctic has been decreasing due to global warming,
resource development in this area is getting attention from many countries
around the world. Recently, the Arctic coastal states (such as Russia, USA, and
Canada) are pursuing infrastructure construction projects for resource develop-
ment in the Arctic region. According to the US Geological Survey (USGS), the
arctic region is estimated to account for about 20% of the world’s undiscovered
oil and gas exploration resources. Interest in the Arctic route has been increasing
as a result of the use of the Arctic Sea as a means of shipping and transporta-
tion, which saves time and money compared to the existing routes. Ice properties,
such as ice concentration, ice thickness, strength of ice, and speed-power relation
under various ice conditions are highly influential for determining the optimal
route in terms of low operational costs. Also, ships operating on the Arctic
c Springer Nature Singapore Pte Ltd. 2021
T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 383–396, 2021.
https://doi.org/10.1007/978-981-15-4680-8_28
384 H.-S. Kim et al.

route are exposed to various ice environment such as level ice, which is frozen
flat in a large area, pack ice, where ice pieces of different sizes float, ice ridge,
which is formed by overlapping ice and flat ice, and brash ice, which is formed
by overlapping small ice. So, the determination of ice resistance and the engine
power becomes important since the ship operating at the Arctic has a larger hull
resistance. To determine the ice resistance for different ice environment, several
estimation methods such as analytical methods, empirical formulas, model tests
and numerical simulations are performed by several researchers. In terms of ease
and cost of application and time consumption, empirical formulas should be
preferred in comparison to other methods for estimation especially in the early
design stage. However, empirical formulas may show discrepancies in resistance
estimation because several parameters in these formulas are based on the results
of full scale experiments or model tests. Even if these parameters were adjusted
by using full scale tests, they mostly depend on the dimensions and geometry
of the vessels which were the subjects of the performed experiments. A detailed
comparison of some empirical formulas for level ice resistance estimation can
be found in [2]. Also, the effect of bow design on level ice resistance estimation
results can be found in [8]. Both of these studies show that the level ice resistance
estimation by using empirical formulas is strongly depend on the ship geometry
and ice characteristics and they are sensitive to change of these properties. So,
the empirical ice resistance estimation methods should be modified to include
the effect of changes in geometry and ice characteristics. An example of this type
of modifications, applied to Lindqvist [7] estimation formula, is presented in [9]
and [10]. Also, the modified versions of the empirical formulas of Shimansky [12],
Enkvist [1] and Ionov [11] are given in [6]. As a result of the modifications in
[6], the hull form information and the parameters that define the geometry of
the ship can be given as an input to the estimation formulas. So, the change in
the geometric form and the change in ship dimensions can be reflected into the
estimation process to obtain a better accuracy in the prediction process. In this
study, above mentioned modified empirical formulas are used to estimate the ice
resistance and the estimation results are obtained by use of an in-house soft-
ware called I-RES. First, the validation of the I-RES program, which was first
developed for level ice, for various ice environments is presented by comparing
the estimation results with model test results. Also, the maximum engine power
of the ship is determined by using the I-RES software. Then, for the purpose of
using in the early design process, ice resistance of two ships, a container with
bulb and a bulk carrier, is estimated for pack ice of different concentrations such
as 60%, 80% and 90%. Finally, in the estimation process, the dimensions of each
ship are increased by 11.56 m in length, 2.29 m in width and 2.22 m in depth
to obtain the effect of the change of size in ice resistance and engine power for
different geometries.
 Study on Estimation of Ice Resistance 385

2 Ice Resistance and Attainable Speed Estimation

In this section, the estimation process of I-RES program is introduced first, then
the empirical formulas for ice resistance and the method of attainable speed are
presented.

2.1 I-RES Estimation Process

The I-RES program estimates the ice resistance and determines the attainable
speed of the ship by employing various empirical formulas that are presented
in the following section. The I-RES program algorithm consists of four steps
which can be seen in Fig. 1, to estimate ice resistance and attainable speed. The
first step is to define the shape of the ship and the properties of ice. The ship
properties, such as principal dimensions, design speed, nodal coordinates on the
hull surface, a vector set for the hull surface at that coordinates, is extracted from
the IGS file or the DA file. Properties of ice, such as ice thickness, ice density, and
ice strength are to be given as input to the program. Then in the second step,
by using the empirical formulas, the estimation of the level ice resistance based
on the inputs from the first step is performed. In the third step, selection of the
ice condition such as level ice, pre-sawn ice, pack ice, ice ridge, and brash ice,
is made by using the GUI provided by the I-RES program and an ice condition
file is created. Finally, the attainable speed, that is, one of special speeds that is
used for operation in ice covered sea with a designated power level, is estimated
in the last step by calculating the available net thrust at a given engine power
for the ice environment based on the generated ice condition file.

Fig. 1. Ice resistance and attainable speed estimation process

386 H.-S. Kim et al.

2.2 Ice Resistance Estimation Method

In this part, first, the coordinate system which is adopted in this study for the
estimation process is introduced. As it can be seen from Fig. 2, the conventional
coordinate system x, y, z in naval architecture is employed, i.e., the x-axis is
longitudinal from A.P to F.P, the y-axis is athwartship from starboard to port
and the z-axis is vertical from bottom to deck. Angle α is defined as an angle
between the x-axis and the tangent line of water line at an arbitrary point or
node in the x–y plane, angle β is an angle between the z-axis and the tangent
line of a section at a node in the y–z plane. Angle γ is an angle between the
z-axis and the tangent line of a buttock at a node in the x–z plane. Vector n is
the normal vector at a node point of the hull surface. Angle φ is the bow angle
in profile view.

Fig. 2. Coordinate system

Ice resistance estimation method for level ice is based on the estimation
theory of Spencer [13], and it is calculated as the sum of four independent ice
resistance components which can be given as

RT = RBR + RCL + RBU + ROW (1)

where RBR is the ice breaking resistance of level ice, RCL is the clearing resis-
tance that occurs when ship hull pushes pieces of ice, RBU is the buoyancy resis-
tance caused by the buoyancy of the ice under the bottom of the ship, and ROW
is the open water resistance. For the estimation of the icebreaking resistance,
Shimansky [12] icebreaking resistance estimation method is employed without
any modification and RBR can be given as

λσf h2i Fx
RBR = (2)
1.93 Fz
 Study on Estimation of Ice Resistance 387

where

4
3ρw g
λ= ,
Eh3i
 LE
√
tan2 α 1 + tan2 α
Fx = dx,
0 1 + tan2 α + tan2 β
 LE
√
tan α tan β 1 + tan2 α
Fz = dx. (3)
0 1 + tan2 α + tan2 β
Here, ρw is the water density, g is the gravitational acceleration, E is Young’s
modulus of sea ice, hi is the ice thickness, LE is the length from fore end to
maximum beam of a ship, angle α is defined before, and β is the angle in the
x-z plane as shown in Fig. 2.

Fig. 3. Schematic diagram of buoyancy force acting on hull surface

For the estimation of buoyancy and clearing resistances given in Eq. 1, the
formula of Enkvist [1] and the formula of Poznyak and Ionov [11] are used,
respectively. To calculate the ice buoyancy resistance, geometric information on
the estimated object is extracted and it is applied to the empirical equation of
[1]. According to this equation, buoyancy resistance is divided into the pres-
sure component Rsp induced by ice submergence into the water and the friction
component Rsf caused by broken ice piece. Figure 3 shows the diagrams of the
ship information for calculating the ice buoyancy resistance and the buoyancy
resistance is defined as follows
RBU = Rsp + Rsf . (4)
Here, the resistance components Rsp and Rsf can be given as
L/2 B
Li 0 ρΔ ghs̄bi
Rsp = 0 ,
L/2

L/2

B
Rsf = Li fg ρΔ ghbi sin β, (5)
0 0
388 H.-S. Kim et al.

where ρΔ represents the density difference between water and ice, Li is the length
between each section, bi is the beam of piece of ice in each section, s̄ represents
the average depth of ice cubes immersed in water, h is the ice thickness, g is the
gravitational acceleration, B is the breadth of the hull, and fg is the coefficient
of friction.
For the calculation of ice clearing resistance, the empirical formula of Poznyak
and Ionov [11] is employed. Clearing resistance is a function of ship speed in ice
and it is described by the relations including ship speed, ice properties and
geometry of ship as follows
⎡ ⎤

L/2
 2 
L/2

⎢ (y (x)) y (x) ⎥
RCL = 2ρi ghBF rB ⎣K3 dx+K3 fg dx⎦ (6)
cos α cos α
0 0

where y is the length function representing the line breadth, ρi is the density
of ice, h is the ice thickness, fg is the friction coefficient, g is the gravitational
 
acceleration, K3 and K3 are constants to be taken as 1.0 and 0.5, respectively.
F rB is the Froude number for the hull breadth which is defined as
Vs
F rB = √ (7)
gB

where Vs is the ship speed and B is the breadth of the hull. Also, there exists
some modifications to these formulas such as the application of each immersion
depth of ice pieces instead of the mean depth for the buoyancy resistance and
the definition a new clearing plane according to the hull form geometry using
the normal vector of the node points for clearing resistance. The details of these
modifications can be found in [6].
Attainable speed estimation is based on the specifications and procedures
which are determined by FMA (Finnish & Swedish Maritime Association). The
process of estimation can be briefly summarized as follows. First, a model test
is conducted to measure the ice resistance in a brash ice channel under the
FMA specifications (1A super, 1A, 1B, 1C). The speed according to the thick-
ness, specified in the regulation, is fixed as 5 knots. In open sea condition, wake
(W ) and thrust deduction factor (t), which depends on the ship resistance and
propulsion, are determined. Then, the thrust, torque curves and the charac-
teristic curves of the engine mounted on the ship at random rpm or advance
ratios are obtained using the Propeller Open Water (POW) test results. Sub-
sequently, available net thrust that can overcome ice resistance is calculated by
using POW, propulsion test results, engine characteristic curve, W , t, and pro-
peller information. Figure 4a shows the results of bollard pull experiment, that
gives the thrust according to the number of revolution of the propeller wing
of the Korean icebreaker RV Araon whose particulars can be found in [4]. To
calculate the attainable speed, the resistance for a given ice condition is consid-
ered to be equal to the thrust, that is obtained from the bollard pull test. Then
the resistance under given ice conditions is derived for the corresponding Rpm
 Study on Estimation of Ice Resistance 389

(a) Thrust-Rpm Curve (b) Power-Rpm Curve

(c) Power-Thrust Curve

Fig. 4. Attainable speed estimation

values from the bollard pull test. Also a power value (DHP) that corresponds to
Rpm is obtained which is shown in Fig. 4b. Then the relation between the ice
resistance (I-RES estimation) and the power (DHP) is represented in Fig. 4c.

3 I-RES Program Validation

In this section the validity of the estimation results is shown by comparing the
estimated results with the model test results on level ice, pre-sawn ice, pack ice
and brash ice.

3.1 Level and Pre-sawn Ice

The model test parameters for level ice and pre-sawn ice are given as follows:
the ice strength is 746 kPa, the ice thickness is 0.56 m, and the ship speed are
1, 3 and 5 knots. The model test results and the estimation results are shown in
Table 1. Level ice is represented by the total ice resistance RT and the pre-sawn
390 H.-S. Kim et al.

resistance can be determined as the value excluding the ice-breaking resistance

RBR from the total ice resistance in Eq. 1. Level ice estimation gives the best
result for ship speed of 3 knots. The pre-sawn estimation gives the best result
for ship speed of 5 knots. Overall, the maximum error rate of the I-RES program
was about 12% when the speed of the ship was 5 knots in level ice.

Table 1. Level ice & Pre-sawn ice resistance comparison of model test results and
I-RES estimation

Ship speed (knots) Level ice resistance (kN) Pre-sawn ice resistance (kN)
Model test 1 454.02 156.95
3 570.61 218.56
5 761.46 339.51
I-RES 1 490.30 153.60
3 580.10 243.40
5 669.90 333.20
Error 1 −7.99% 2.14%
3 −1.66% −11.37%
5 12.02% 1.86%

3.2 Pack Ice

Fig. 5. Pack ice correlation coefficient

The model test parameters for pack ice are given as follows: the ice strength is
1157 kPa, the ice thickness is 1.06 m, the ship speed are 1, 3 and 5 knots and
the ice concentration are 60%, 80% and 90%. The model test results and the
estimation results are shown in Table 2. Pack ice resistance can be determined as
the value excluding the ice-breaking resistance RBR from the total ice resistance
 Study on Estimation of Ice Resistance 391

in Eq. 1 as in the case of the pre-sawn resistance. Based on the relationship

between ice concentration and ship speed, the correlation coefficient is derived
as shown in Fig. 5. Some details of pack ice estimation process can be found
in [3]. The maximum error between the model test and I-RES estimation for
pack ice is about 51% when the ship speed is 1 knot and the ice concentration
is 60%. This error is the result of the test environment and it is excluded from
the verification results. When the ice concentration is 80% and 90%, the I-RES
estimation shows a smaller resistance value than the model test results, and the
maximum error is about 20%. It can be deduced that the error occurs because
the correlation coefficient is derived by expanding ice concentration and ship
speed under limited experimental conditions.

Table 2. Pack ice resistance comparison of model test results and I-RES estimation

Ship speed (knots) Pack ice 60% (kN) Pack ice 80% (kN) Pack ice 90% (kN)
Model test 1 114.00 384.11 414.22
3 171.02 453.50 534.74
5 292.12 534.09 651.37
I-RES 1 165.39 298.66 365.29
3 209.49 397.36 491.29
5 232.35 457.58 570.20
Error 1 51% −21% −11%
3 32% −10% −6%
5 -11% −9% −8%

3.3 Brash Ice

The model test for brash ice is carried out at ship speed of 1, 3, 5 knots, brash ice
thickness of 0.86 m, and 1B environment, so that the real ship speed is extended
to 17 knots based on the model test. The details of the model test of brash ice
is given in [5]. The model test results and the estimation results are shown in
Table 3. Brash ice resistance can be determined as the value excluding the ice-
breaking resistance RBR from the total ice resistance in Eq. 1 as in the case of the
pack ice resistance. Based on the extended brash ice resistance, the correlation
coefficient, that depends on the ship speed and brash ice thickness, is derived
with comparison to the I-RES estimation results and it is shown in Fig. 6.
392 H.-S. Kim et al.

Fig. 6. Brash ice correlation coefficient

Table 3. Brash ice resistance comparison of model test results and I-RES estimation

Ship speed (knots) Brash ice resistance (kN)

Model test 1 332.91
3 371.20
5 434.87
I-RES 1 314.01
3 390.55
5 438.30
Error 1 −5.68%
3 5.21%
5 0.79%

4 Ship of Arbitrary Shape

After the validation of the estimation results of I-RES in the previous section,
the resistance estimation of ship of arbitrary shape is presented in this section.
In the early design process, the effect of the ship geometry on the ice resistance is
an important information and it also important to know the influence of the size
change on the ice resistance in terms of operational planning. For this purpose,
ice resistance of two ships, a container with bulb and a bulk carrier, is estimated
for pack ice of different concentrations such as 60%, 80% and 90%. Then, the
dimensions of each ship are increased by 11.56 m in length, 2.29 m in width and
2.22 m in depth in the estimation process. The geometries, main dimensions and
the increment in the dimensions of the two ships are shown in Fig. 7 and Table 4.
The results of the pack ice resistance estimation of the container ship and the
bulk carrier is shown in Fig. 8 and Fig. 9, respectively. As it can be seen from the
results, the estimated pack ice resistance for the container ship is higher than
 Study on Estimation of Ice Resistance 393

(a) Container with bulb (b) Bulk Carrier

Fig. 7. Ship geometries

Table 4. Dimensions of ship before and after the size change

(a) Container Dimensions (b) Bulk Carrier Dimensions

Container Before After Bulk Carrier Before After
Length (x) 230 m 241.56 m Length (x) 85 m 96.56 m
Width (y) 32.2 m 34.49 m Width (y) 20 m 22.29 m
Depth (z) 19 m 21.22 m Depth (z) 18 m 10.72 m

the resistance of the bulk carrier in all concentrations of pack ice due to larger
dimensions of the container ship. After the same amount of change in dimensions
of both ships, estimated resistance becomes higher in all concentrations. Also,
the increment of the resistance is getting higher with the increasing speed. If
we compare the increment of the resistances of both ships for before and after
conditions, the increment in resistance of the container ship is larger than of the
bulk carrier ship with increasing speed.

(a) 60% concentration (b) 80% concentration (c) 90% concentration

Fig. 8. Pack ice resistance estimation results of the container ship before and after the
increment in dimensions

Also the power and ship speed relations are obtained from the estimation
results and these results can be seen in Fig. 10 and Fig. 11 for the container ship
and the bulk carrier, respectively. The power needed for obtaining the required
speed is increased between before and after conditions due to the increment of
394 H.-S. Kim et al.

(a) 60% concentration (b) 80% concentration (c) 90% concentration

Fig. 9. Pack ice resistance estimation results of the bulk carrier before and after the
increment in dimension

(a) 60% concentration (b) 80% concentration (c) 90% concentration

Fig. 10. Power-speed estimation results of the container ship before and after the
increment in dimensions

(a) 60% concentration (b) 80% concentration (c) 90% concentration

Fig. 11. Power-speed estimation results of the bulk carrier before and after the incre-
ment in dimensions

the size in ship dimensions. By using this relations, the attainable speed can be
calculated if the operator decides the certain power level and ice properties and
the voyage route planing in ice field can be performed more efficiently.
 Study on Estimation of Ice Resistance 395

5 Conclusion
In this study, I-RES program was used to estimate the resistance for pack ice of
60%, 80% and 90% concentrations by employing the described empirical formu-
las. Then, the comparison of the I-RES program estimation results with model
test results was presented for various ice environments. For the level and pre-sawn
ice environment, the minimum and maximum error rate of the I-RES estimation
was found to be about 1.7% and 11% in absolute values, respectively. For the
pack ice estimation of 60% concentration, the verification results, for which the
error was found to be 51%, were excluded because of the limited experimental
conditions. For the 80% and 90% concentration of pack ice, the minimum and
maximum errors were about 6% and 21% in absolute value, respectively. Finally,
for the brash ice, the minimum and maximum errors were about 0.79% and
5.68% in absolute value, respectively. After validation of the estimation results,
the estimation for pack ice resistance of 60%, 80% and 90% was performed for
two different shaped ships, a container with bulb and a bulk carrier. Then, in
the estimation process, the dimensions of each ship were increased by 11.56 m in
length, 2.29 m in width and 2.22 m in depth to obtain the effect of the increment
in dimensions on ice resistance and engine power. The estimation results showed
that the increment in dimensions resulted in a higher value of ice resistance for
both ships, as expected. But the amount of increment in ice resistance of the
container ship is larger than of the bulk carrier. Also, the maximum engine power
of both ships was determined by using the I-RES software. The power needed
for obtaining the required speed was increased between before and after condi-
tions due to the increment of the size in ship dimensions. The attainable speed
can be calculated by the help of these results if the operator decides the certain
power level and ice properties. So, the voyage route planning in ice field can
be performed more efficiently. Since the employed formulas contain coefficients
based on empirical data, they should be interpreted carefully. Even if the esti-
mations, that employ these formulas, were validated for RV Araon ice breaker
in various ice conditions in this study and, they were validated for a large ice
breaking tanker of 267 m length in level ice condition in [6]. Both studies show
that there exist some discrepancies between estimation and model test results.
Also, the issues, that are related with the sensitivity of empirical formulas to the
ship geometry and ice characteristics, mentioned in [2] and [8] should be noted.
So, the results, obtained for two different shaped vessels in this study, should be
considered for qualitative assessment. As a consequence of the validation results
and modified estimation formulas, the results of ice resistance estimation can be
used for the early stage of vessel design.

Acknowledgment. This research is supported by “Development of hull form of year-

round floating-type offshore structure based on the Arctic Ocean in ARC7 condition
with dynamic positioning and mooring system” through the Ministry of Trade, Indus-
try, and Energy (MOTIE) (10063405, 2018).
396 H.-S. Kim et al.

References
1. Enkvist, E.: On the ice resistance encountered by ships operating in the continuous
mode of icebreaking. Technical report 24, The Swedish Academy of Engineering
Sciences in Finland (1972)
2. Erceg, S., Ehlers, S.: Semi-empirical level ice resistance prediction methods. Ship
Technol. Res. 64(1), 1–14 (2017)
3. Kim, H.S.: Investigation of the extension of an ice-resistance system for a various
pack ice condition. In: Proceedings of the 23nd IAHR International Symposium on
Ice Conference, Ann Arbor, Michigan, USA (2016)
4. Kim, H.S., Lee, C.J., Choi, K.S., Kim, M.C.: Study on icebreaking performance
of the Korea icebreaker ARAON in the arctic sea. Int. J. Naval Architect. Ocean
Eng. 3, 208–215 (2011)
5. Kim, H.S., Lee, C.J., Kang, K.J., Jeong, S.Y., Yum, J.G., Jeong, U.C.: Study on
the equivalent model test and analysis method of brash ice at Kriso ice tank. In:
Proceedings of the 23rd International Conference on Port and Ocean Engineering
under Arctic Conditions (POAC 2015), Trondheim, Norway (2015)
6. Kim, H.S., Ryu, C.H., Park, K.D., Lee, J.: Development of estimation system of
ice resistance with surface information of hull form. Ocean Eng. 92, 12–19 (2014)
7. Lindqvist, G.: A straightforward method for calculation of ice resistance of ships.
In: Proceedings of 10th International Conference on Port and Ocean Engineering
under Arctic Conditions (POAC), p. 722 (1989)
8. Myland, D., Ehlers, S.: Influence of bow design on ice breaking resistance. Ocean
Eng. 119, 217–232 (2016)
9. Myland, D., Ehlers, S.: Investigation on semi-empirical coefficients and exponents
of a resistance prediction method for ships sailing ahead in level ice. Ships Offshore
Struct., 1–10 (2019)
10. Myland, D., Ehlers, S.: Model scale investigation of aspects influencing the ice resis-
tance of ships sailing ahead in level ice. Ship Technol. Res., 1–11 (2019). https://
doi.org/10.1080/09377255.2019.1576390
11. Poznyak, I., Ionov, B.: The division of icebreaking resistance into components. In:
Proceedings of the 6th STAR Symposium, New York, US, pp. 249–252 (1981)
12. Shimansky, Y.: Conditional Standards of Ice Qualities of a Ship, vol. 130. Northern
Sea Route Administration Publishing House (1938)
13. Spencer, D.: A standard method for the conduct and analysis of ice resistance
model tests. In: Proceeding of the 23rd American Towing Tank Conference, pp.
301–307. NRCC, Institute for Marine Dynamics, New Orleans, US (1992)
 Estimation of Ice Conditions Along
the Northern Sea Route

Wei Chai1,2,3(&), Bernt J. Leira2, Chana Sinsabvarodom2,

and Wei Shi3
1
School of Transportation, Wuhan University of Technology,
Wuhan 430063, China
[email protected]
2
Department of Marine Technology,
Norwegian University of Science and Technology, 7491 Trondheim, Norway
[email protected]
3
State Key Laboratory of Coast and Offshore Engineering,
Dalian University of Technology, Dalian 116024, China

Abstract. The reduction of ice cover and increased activities related to maritime
transport and exploiting natural resources in the arctic regions in recent decades
have resulted in increased interest and demands for developing ice-capable
vessels to navigate along Arctic shipping routes. Among these routes, the
Northern Sea Route (NSR), which tails along the Russian coast, presents a
potential shortcut for connecting the Europe and East Asia. Ice conditions are
considered as the key factors for design of Arctic ships since they are closely
related to the ship ice loads and structural response. In this work, four repre-
sentative sites along the NSR are selected to study the ice conditions along the
NSR. Specifically, ice thickness is estimated by applying the accumulated
freezing degree days (FDD) method based on the collected air temperature
records. A scheme which consists of the ice thickness, ice temperature, ice
salinity and porosity is introduced in this work in order to estimate the ice flexural
strength and crushing strength. Finally, a comparison is made to evaluate the
severity of ice conditions at the four representative sites selected along the NSR.

Keywords: Northern Sea Route
Ice thickness
Flexural strength
Crushing

strength

1 Introduction

The Northern Sea Route (NSR) is a shipping lane trails along the Russian coast and
crossing several Arctic Seas: the Barents Sea, the Pechora Sea, the Kara Sea, the Laptev
Sea, the East Siberian Sea and the Chukchi Sea. In recent years, the NSR has received
increasing attention as a potential shortcut for connecting the Europe and East Asia due
to the reduction of ice extent and increased activities related to maritime transport and
exploiting natural resources in the arctic regions [1].
For Arctic ships, ice loads caused by the ship and ice interaction process present the
dominant load and main structural challenge. Generally, ice loads depend on the ice
condition (e.g. ice thickness, ice strength and ice age), the ship hull geometry as well as
© Springer Nature Singapore Pte Ltd. 2021
T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 397–408, 2021.
https://doi.org/10.1007/978-981-15-4680-8_29
398 W. Chai et al.

on the relative velocity between the ship and ice features [2]. Although a full under-
standing the complex ship-ice interaction process is far from archived, a good
knowledge of the ice conditions along the NSR would promote the development of
structural design and operational safety of Arctic ships along the route.
In this work, first-year level ice along the NSR is considered. For one thing, the ice
conditions along the commercial Arctic shipping routes, such as the NSR, are mostly
first-year and few ice appears in summer seasons. For another, the physical and
mechanical properties of the first-year level ice have been studied extensively in many
decades and reliable relationships have been established between the physical prop-
erties and mechanical properties of the first-year level ice. Based on the study of ship
and ice interaction process, ice thickness, flexural strength and crushing strength of the
level ice are considered as the most important parameters for determining the ice
actions on the Arctic ships [3].
There are mainly two methods to obtain the ice condition data: by measurements
(e.g. in-situ measurements, remote sensing) or by theoretical estimations [4, 5]. For a
long period, obtaining the ice conditions, especially the strength of ice, at a given site
along the NSR, would be difficult and expensive. In contrast, theoretical estimation
based on the well-established empirical formula is an effective and efficient way to
obtain the ice conditions data along the NSR.
Based on the collected air temperature records at a given site, ice thickness is
estimated by applying the accumulated freezing degree days (FDD) method. Subse-
quently, a theoretical scheme with considerations of the air temperature, ice thickness,
ice temperature, ice salinity and porosity is established in order to estimate the cor-
responding flexural strength and crushing strength of the level ice. Furthermore, four
representative sites along the NSR are selected in order to evaluate the severities of ice
conditions in different sea areas.

2 Theoretical Background

Ice thickness is regarded as the most representative parameter for the severity of ice
condition. At a given site, if the history of daily average air temperature, Ta, is
obtained, the ice thickness, h, can be estimated by applying the Stefan equation [6],
which is given as:

h ¼ að2ki =qi LÞ0:5 ½RðTb  Ts Þt 0:5 ð1Þ

where ki is the thermal conductivity of ice, qi denotes the ice density, L represents latent
heat of fusion of ice and t is the time. The approximate values of ki, qi and L are given
in Høyland [7]. Tb and Ts are the temperatures at the bottom and surface of the ice
sheet, respectively. In order to consider the effect of snow cover, an empirical factor, a,
with a value of 0.75, is introduced in the Stefan equation [8]. Moreover, h is given in
meters and t is 86400 s/day.
The bottom temperature of ice sheet, Tb, is assumed to be the same as the freezing
point of sea water, Tf, about −2 °C. The surface temperature of ice sheet is related to
the air temperature, Ta, by the following relationship, which is given as [9]:
 Estimation of Ice Conditions Along the Northern Sea Route 399

Ts ¼ Ta for  10 o C  Ta   2 o C
ð2Þ
Ts ¼ 0:6Ta  4 for Ta \10 o C

For the temperature of the ice sheet throughout the ice growth season, one can
usually assume that there is a linear temperature gradient through the ice sheet, see
Fig. 1.

Air temperature Ta
Air

Ice surface temperature Ts T

h
Ice Ice thickness, h

Ice surface temperature Tb

Sea water

Fig. 1. Illustration of the temperature gradient through the ice sheet, horizontal axis: temperature
T, vertical axis: ice thickness h.

In addition to model based on the Stefan equation, there are another two empirical
models to describe the growth of sea ice thickness with consideration of the snow cover
effects. The first model is the empirical equation proposed in Zubov [10] as:

h2 þ 50h ¼ 8FDD ð3Þ

where FDD denotes freezing degree days for a winter season (i.e. ice growth season).
FDD is the sum over the difference between the daily average air temperature, Ta and
the freezing point of sea water, Tf:
X
FDD ¼ ðTf  Ta Þ ð4Þ
days

The last model for estimation of the sea ice thickness was proposed in Lebedev [11]
and given as:

h ¼ 1:33FDD0:58 ð5Þ

where the ice thickness estimated by the Zubov model and the Lebedev model are
given in centimeters. The above three models can be categorized as the FDD method.
Furthermore, an example of verification for the FDD method with field measurements
has been reported in Ref. [11].
The ice salinity, S, is usually expressed as the fraction by mass of the salts con-
tained in a unit mass for the first-year level ice. The average salinity of an ice sheet,
400 W. Chai et al.

expressed as parts per thousand (ppt), in the ice growth season, can be related to the
thickness of the ice, h, by the following relationship [9]:
(
13:4  17:4h for h  0:34 m
S¼ ð6Þ
8:0  1:62h for h [ 0:34 m

where h is expressed in meters.

In the sea ice, the brine, the pure ice, the air and the solid salts are assumed to exists
in a thermal equilibrium. The brine volume represents the amount of liquid brine within
the ice and it is usually expressed in terms of the volume in parts per thousand or
expressed as a volume fraction. In some cases, such as brine drainage has occurred, the
air volume can be significant and it is useful to know the amount of air volume in the
ice.
The total porosity, vT, of the ice is expressed as:

vT ¼ vb þ va ð7Þ

where vb and va are the relative brine volume and relative air volume, respectively.
Sea ice porosity of growing ice can be estimated by the values of ice temperature,
T, sea ice density qi and salinity of the sea ice, S as [9, 12]:

vb ¼ qi S=F1 ðTÞ ð8Þ

and the function, F1(T) is given as:

4:732  22:45T  0:6397T 2  0:01074T 3 ;  2  T   22:9
F1 ðTÞ ¼ ð9Þ
9899 þ 1309T þ 55:27T 2 þ 0:716T 3 ;  22:9  T   30

where the variable T is given in °C.

The relative air volume can be expressed as:

va ¼ 1  qi qp ðTÞ þ qi S
F2 ðTÞ=F1 ðTÞ ð10Þ

where qp represents the pure ice density (in Mg/m3) and it is described as:

qp ðTÞ ¼ 0:917  1:403
104 T ð11Þ

and the function, F2(T) is given as:

8:903
102  1:763
102 T  5:33
102 T 2  8:801
106 T 3 ;  2  T   22:9
F2 ðTÞ ¼
8:547 þ 1:089T þ 4:518
102 T 2 þ 5:819
104 T 3 ;  22:9  T   30
ð12Þ

The strength of ice features depends on a large number of parameters, such as the
ice temperature, the ice salinity and porosity, the loading rate and direction and so on.
 Estimation of Ice Conditions Along the Northern Sea Route 401

Based on a large number of small-scale tests, the flexural strength, rf, of the first-year
sea ice in the growth season can be expressed by the following relationship [4, 13]:
pffiffiffiffiffi
rf ¼ 1:76 expð5:88 vb Þ ð13Þ

where rf is given in Mpa and the average ice temperature Tav = (Ts+ Tb)/2 is applied in
Eqs. 8 and 9 in order to estimate the average brine volume of the ice sheet.
For the crushing strength of level ice in the growth season, measurements of the
horizontally loaded samples have shown that the uniaxial crushing strength, rc,
expressed in MPa, is given by Eq. 14 [4]:
 rffiffiffiffiffiffiffiffiffi
0:22 vT
rc ¼ 37ð_eÞ 1 ð14Þ
0:27

where e_ is the strain rate which depends on the relative speed of the interaction process
and the width of the structure. The range of strain rates for Eq. 14 is valid from 10−7/s
to 10−3/s and the crushing strength reaches its maximum when the strain rate is equal to
10−3/s.
Moreover, in order to consider the effect of temperature variation through the ice
sheet (see Fig. 1) on the crushing strength, Timco and Frederking [9] developed a
model to predict the full thickness crushing strength. In this model, the ice sheet is
divided into nine separate layers and the crushing strength of each layer is estimated by
applying Eq. 14 with information of average ice temperature, salinity and density of
each layer. The average value of these calculated crushing strength can be used to
estimate the large-scale crushing strength for the full ice sheet.
Finally, based on the equations proposed in this Section, a theoretical scheme,
which is used to provide estimations of the ice thickness and ice properties, is estab-
lished and shown in Fig. 2. Based on the records of daily average air temperature, the
ice thickness, ice salinity, ice temperature, ice porosity and ice strength in the growth
season can be estimated by applying the algorithms and formula included in this
scheme.

3 Numerical Results

3.1 Ice Thickness

In this work, the records of air temperature were obtained from Carbon Dioxide
Information Analysis Center (CDIAC) [14] where the temperature datasets from 518
Russians weather stations are available. Four representative sites along the NSR are
selected and the locations are illustrated in Fig. 3. The name, location and period of
records for the air temperature for each station are given in Table 1. Furthermore, it
should be mentioned that although there could be a tendency that the climate is getting
warmer, the historical data of air temperature is treated without including this tendency
in this work.
402 W. Chai et al.

Fig. 2. Illustration of the scheme used to estimate the ice thickness and ice strength

Chukchi Sea
D East Siberian Sea

C
Laptev Sea

B
Kara sea

A
Pechora sea

Fig. 3. Illustration of the NSR and four representative sites (A, B, C and D) along the NSR.

Table 1. Locations of the weather station and period of air temperature records.
Sites Station name Location Period of records (year)
A Im E.K.Federova 70˚27’N, 59˚05’E 1959–2008
B Russkij 77˚10’N, 96˚26’E 1936–1993
C Shalaurova 73˚11’N, 143˚14’E 1936–1998
D Ostrov Vrangelja 70˚59’N, 181˚31’E 1926–2008
 Estimation of Ice Conditions Along the Northern Sea Route 403

As mentioned above, Ta represents the daily average air temperature, i.e. the
average value during one day and night [15]. The mean daily average temperature
(MDAT) for a given site during a N-year period is defined as [16]:

1 X
MDATk ¼ Ta;jk ð15Þ
N j¼1;N

where Ta,jk represents the daily average air temperature for the kth day in the year with
index number j.
The daily average air temperature collected at site A for the specific period 1959–
2008 is plotted as blue dots in Fig. 4(a) and the corresponding MDAT during this
period is plotted as a black line. Similarly, the MDAT values at four representative sites
for the corresponding record periods (see Table 1), are presented in Fig. 4(b).

Fig. 4. (a) Daily average air temperature (blue dots) and the MDAT (black line) at site A; (b) the
MDAT values at four representative sites.

Based on the records of daily average air temperature, three empirical models, i.e.
the Stefan equation model, the Zubov model and the Lebedev model, are applied to
estimate the (daily) ice thickness in the ice growth season. Similar to the method for
calculating the MDAT in Eq. 15, the mean ice thickness for a day with index k is
calculated as the mean value of the samples, which is a collection of the (daily) ice
thickness corresponds to the same index k during the specific N-year period. The mean
(daily) ice thickness at the four selected sites, estimated by the empirical models for the
specific record periods given in Table 1 are plotted in Fig. 5.
It is seen in Fig. 4(b) that the MDAT values at four representative gradually
decrease from October to February and then increase from March to June. The lowest
values of the MDAT are observed from the beginning of January to the end of
February. Site A has the highest values for the MDAT, while the lowest values of the
MDAT are observed at site C.
404 W. Chai et al.

Fig. 5. Mean (daily) ice thickness estimated by three empirical models at four representative
sites.

For the mean ice thickness estimated by three empirical models, it is shown in
Fig. 5 that the three empirical models provide very close values. In the period with
decrement of the MDAT values, i.e., from October to February, the Stefan equation
model provides the highest values of the mean ice thickness. However, in the period
with increment of the MDAT values, it is found that the mean ice thickness provided
by the Zubov model and the Lebedev model could be higher than the mean ice
thickness calculated by the Stefan equation model.
The overall differences for the mean ice thickness estimated by different models are
not significant. The mean ice thickness based on the Stefan equation model at four
representative sites are plotted in Fig. 6. The mean ice thickness reaches its maxim
value at the end of May and the maximum values are in the range from 1.2 m to 1.7 m.
It is seen in Figs. 4 and 6 that the highest values of the MDAT correspond to the
minimum values of the mean ice thickness. Moreover, it is found that the mean ice
thickness at sites B and C are very close due to the similar climate conditions at these
two sites (see Fig. 4).

3.2 Ice Strength

Based on the MDAT values at four selected sites and the theoretical scheme proposed
in Sect. 2, the flexural strength and crushing strength of the sea ice at these sites is
obtained. It is seen in Fig. 2 that the ice strength depends on the ice thickness which
can be estimated by three empirical models. Therefore, the influence of these ice
thickness models on the ice strength is investigated in this Section.
 Estimation of Ice Conditions Along the Northern Sea Route 405

Fig. 6. Mean (daily) ice thickness estimated by the Stefan equation model at four representative
sites.

The method used to calculate the mean (daily) ice thickness is applied to provide
the mean (daily) values of the ice flexural strength and crushing strength at four
representative sites, which are shown in Figs. 7 and 8, respectively. For the calculation
of the ice crushing strength, the strain rate e_ is equal to 10−3/s.

Fig. 7. Mean (daily) ice flexural strength at different sites, ice thickness is estimated by three
empirical models
406 W. Chai et al.

Fig. 8. Mean (daily) ice crushing strength at different sites, ice thickness is estimated by three
empirical models

It is seen in Figs. 7 and 8 that the influence of different ice thickness estimation
models on the mean values of ice strength is very limited. Subsequently, the mean
values of the ice flexural strength and ice crushing strength based on the same ice
thickness estimation model, the Stefan equation model, at different sites are presented
in Fig. 9(a) and 9(b), respectively.

Fig. 9. (a) Mean (daily) ice flexural strength; (b) Mean (daily) ice crushing strength, the ice
thickness is estimated the Stefan equation model.
 Estimation of Ice Conditions Along the Northern Sea Route 407

For the mean (daily) ice flexural strength, shown in Fig. 9(a), the maximum values
can be observed in February and range from 0.5 MPa to 0.65 MPa. Similarly, it is
observed in Fig. 9(b) that the maximum values of the mean values for the ice crushing
strength range from 3.8 MPa to 4.5 MPa in February.
More importantly, the climate condition has a close relationship with the ice
strength. For one thing, the weakest strength of the sea ice can be observed at site A,
which has the highest values of the MDAT and the lowest values of the mean ice
thickness for the four selected sites. For another, it is seen in Fig. 4(b) and Fig. 9, from
October to February, the ice strength becomes stronger as the MDAT values increase
and then the ice strength decrease with the increment of the MDAT values from March
to June.

4 Conclusions

In this work, a theoretical scheme is developed to study the ice conditions of four
representative sites along the NSR. Ice thickness is estimated by the FDD method with
the records of daily air temperature. Ice flexural strength and crushing strength in the
ice growth season is estimated by the algorithms included in the scheme.
For the ice conditions along the NSR, the mean ice thickness reaches its maxim
value at the end of May and the maximum values of the mean ice strength can be
obtained in February. The maximum values of the mean ice thickness, the mean ice
flexural strength and the mean crushing strength are in the ranges from 1.2 m to 1.7 m,
0.5 MPa to 0.65 MPa and 3.8 MPa to 4.5 MPa, respectively.
It is found that the climate condition has a close relationship with the ice thickness
and ice strength. For the four representative sites, the weakest strength of the sea ice
and the lowest values of the mean ice thickness can be observed at site A, which has the
highest values of the MDAT. Moreover, it is observed that the mean values of ice
strength vary with the variation of air temperature in the ice growth season from
October to the end of May. The correlations between the air temperature, ice thickness
and ice strength will be investigated in future study.

Acknowledgments. This work is supported by the Joint Center of Excellence for Arctic
Shipping and Operations which is funded by the Lloyd’s Register Foundation. The Lloyd’s
Register Foundation supports the advancement of engineering-related education, and funds
research and development that enhances the safety of life at sea, on land, and in the air. Financial
support from the Research Council of Norway (RCN project number: 249272/O80) and the State
Key Laboratory of Coast and Offshore Engineering, Dalian University of Technology (project
number: LP1801) are also acknowledged.
408 W. Chai et al.

References
1. Hahn, M., Dankowski, H., Ehlers, S., et al.: Numerical prediction of ship-ice interaction: a
project presentation. In: 36th International Conference on Ocean, Offshore and Arctic
Engineering Proceedings, ASME 2017, pp. V008T07A2–VT07A2. American Society of
Mechanical Engineers (2017)
2. Chai, W., Leira, B.J., Naess, A.: Probabilistic methods for estimation of the extreme value
statistics of ship ice loads. Cold Reg. Sci. Technol. 146, 87–97 (2017)
3. Zhou, L., Riska, K., Ji, C.: Simulating transverse icebreaking process considering both
crushing and bending failures. Marine Struct. 54, 167–187 (2017)
4. ISO 19906: Petroleum and Natural Gas Industries–Arctic offshore structures. ISO, Geneva
(2010)
5. Chai, W., Leira, B.J.: Environmental contours based on inverse SORM. Marine Struct. 60,
34–51 (2018)
6. Ashton, G.D.: River and Lake Ice Engineering. Water Resources Publication, Colorado
(1986)
7. Høyland, K.V.: Ice thickness, growth and salinity in Van Mijenfjorden, Svalbard, Norway.
Polar Res. 28, 339–352 (2009)
8. Timco, G., Weeks, W.: A review of the engineering properties of sea ice. Cold Reg. Sci.
Technol. 60, 107–129 (2010)
9. Timco, G., Frederking, R.: Compressive strength of sea ice sheets. Cold Reg. Sci. Technol.
17, 227–240 (1990)
10. Zubov, N.: Ice of the Arctic. Izd-vo Glavsevmorputi, Moscow (1945)
11. Lebedev, V.: Rost l’da v arkticheskikh rekakh i moriakh v zavisimosti ot otritsatel’nykh
temperatur vozdukha. Problemy Arktiki. 5, 9–25 (1938)
12. Cox, G.F., Weeks, W.F.: Equations for determining the gas and brine volumes in sea-ice
samples. J. Glaciol. 29, 306–316 (1983)
13. Timco, G., O’Brien, S.: Flexural strength equation for sea ice. Cold Reg. Sci. Technol. 22,
285–298 (1994)
14. Bulygina, O., Razuvaev, V.: Daily temperature and precipitation data for 518 Russian
meteorological stations. Oak Ridge National Laboratory, US Department of Energy, Oak
Ridge, Tennessee (2012)
15. DNV GL: Rules for classification: Ships. Ships for Navigation in Ice. 726 (2016)
16. Bridges, R., Riska, K., Lu, L., et al.: A study on the specification of minimum design air
temperature for ships and offshore structures. Ocean Eng. 160, 478–489 (2018)
 Effect of Ship Speed on Bow Impact Load
Estimation for Polar Class Ship Design

Mahmud Sazidy1(&), Claude Daley2, Bruce Colbourne2,

and John MacKay1
1
Defence Research and Development Canada (DRDC)-Atlantic,
Dartmouth, NS, Canada
[email protected]
2
Memorial University of Newfoundland, St. John’s, NL, Canada

Abstract. The bow impact load is important for design ice load parameter
estimation, and hence for the scantling calculation in Polar ship design and safe
speed methodology. In the Polar Rules, the effective bow impact load is the
minimum ice failure load from the crushing failure model and flexural failure
model. The Polar Rules ice crushing failure model is reasonably good for
estimating the ice crushing load and to account for the ship speed effect.
However, the flexural failure load in the Polar model is insensitive to the ship
speed. In this study, a speed dependent ice flexural failure model developed by
Sazidy et al. [1] is adopted to investigate the ship speed effect on the bow impact
load. The study indicates that the Polar Rules based bow impact load estimation
method is reasonable for thicker ice operation, slow speed interaction or for
ships with lower displacement, as ice crushing represents their limiting ice
failure condition. For thinner ice at medium to higher interaction speeds or the
ship with higher displacements, the flexural failure model provides the limiting
bow impact load, and hence the Polar Rules based method may not be appro-
priate for these interaction scenarios as their flexural failure model is insensitive
to ship speed.

Keywords: Speed effect
Ice flexural failure
Polar ship design
Bow impact
load

1 Introduction

The Arctic Polar Regions are believed to have vast amount of natural resources. Safe
transportation in the region and exploitation of these resources in the Arctic is still a
significant challenge. The heavy multi-year ice poses a great risk to the ships operating
in these regions. Ships are regularly damaged due to impact with ice. These impact
events commonly result in denting, deformation or fracture to the ships. Kubat and
Timco [2] reviewed historical data of ship damage in the Canadian Arctic from 1978 to
2003 and presented a histogram (Fig. 1) indicating the damage events and damage
severity due to impact with the first year ice and multi-year ice. Likhomanov et al. [3]
also reported high levels of hull structural damage during operation in the Northern Sea
Route. These damage events cause operational delays and lead to economic loss. In

© Crown 2021
T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 409–419, 2021.
https://doi.org/10.1007/978-981-15-4680-8_30
410 M. Sazidy et al.

addition, such damage events can be a major threat to the Arctic environment.
Therefore, it is important that design of ice capable ships be well informed to ensure
safe operations in Polar waters.

Fig. 1. Damage events and damage severity in Canadian Arctic [2]

Ice capable ship design and operation require ice strengthening of the hull structure
to resist ice impact forces. Ship classification societies have developed ice strength-
ening requirements based on ice conditions, ship structural particulars and ship oper-
ating conditions [4–6]. Each of the classification societies has its own ice strengthening
requirements [7]. The Polar Rules (also known as Unified Requirements for Polar Ships
or the Polar UR) represent the latest industry standards, and have been in force since
January 1, 2017 [8]. The Polar Rules were developed by the International Association
of Classification Societies Ltd. (IACS) under the guidance of the International Mar-
itime Organization (IMO) in order to replace the individual member societies’ rule sets
and provide a harmonized set of ice classes for ships intended to operate in the Polar
waters [4, 5, 9].
The Polar Rules ship design and ice strengthening process requires determination of
design ice load parameters. The design ice load parameters are rationally linked with
impact forces at the bow which are calculated from a design scenario of a glancing
collision between ship and ice [4, 6]. The bow impact forces from this glancing
collision are limited by the ice edge crushing or flexural failure [9]. Appropriate ice
crushing and flexure failure models are necessary for prediction of the bow impact
force and hence the design ice load parameters.
The Polar Rules adopted an ice crushing model which is based on energy balance
principles, Popov collision mechanics and a pressure-area relationship [9, 10]. The
model has been widely used in the rules based design ice load parameter formulation as
well as ship-ice interaction studies [5, 6, 11]. The model accounts for the ship speed
effect and provides a good estimate of impact load, particularly at thicker ice and slow
ship speed operation [5, 11]. On the other hand, the Polar Rules based ice flexural
failure model is a simple function of hull shape and ice conditions [5, 6, 11], and
 Effect of Ship Speed on Bow Impact Load Estimation 411

ignores the ship speed effects. This makes it difficult to extrapolate the Polar Rules
design method to cases of thinner ice and higher ship speed. Several studies indicated
that the ice inertia and the water foundation make the ice flexural failure process
strongly speed dependent [11–13]. A detailed numerical investigation of ship-ice edge
breaking by Sazidy et al. [1] also suggested that the speed effect should be included in
the ice flexural failure model.
In this study, a speed dependent ice flexural failure model previously developed by
Sazidy et al. [1] has been utilized to estimate the bow impact loads for Polar class ships.
The study considers a higher ice class ship (PC 2) and a lower ice class ship (PC 6) to
demonstrate how the speed effect in the ice flexural failure model influences the bow
impact loads estimated at constant ice thickness, different ice thicknesses and different
ship displacements. The estimated bow impact loads are compared with Polar Rules
based estimates.

2 Polar Design Ice Load Scenario

The Polar Rules design scenario considers a glancing collision of the ship bow shoulder
with a 150° ice wedge in order to derive the bow impact forces and design ice load
parameters [9]. During the glancing collision, the impact forces are limited by the ice
edge crushing or flexural failure [9]. Therefore, the Polar Rules adopted two ice failure
load models for the ice crushing and ice flexure to estimate these bow impact forces
and design ice load parameters. The scenario indicating ice edge crushing and ice
flexure is shown in Fig. 2.

Fig. 2. Polar Rules design scenario indicating ice edge crushing and ice flexure [9]

In general, the Polar bow impact load is calculated at four different bow locations.
The impact load with the maximum magnitude is applied in the form of a load patch to
evaluate the structural strength [5, 6, 9, 11]. This load patch is calculated over a
triangular contact surface but is idealized to a rectangular load patch using specified
formulas for application to the ship structure. This idealized rectangular load patch is
characterized by several ice load parameters such as bow impact load (Fi), line load
(Q), pressure (p), patch height (b) and patch width (w). Details of the design ice load or
412 M. Sazidy et al.

load patch formulation methodology can be found in Rahman [6] and Daley [9].
Figure 3 illustrates the methodology to estimate the design ice load parameters in the
Polar Rules.

Fig. 3. Design ice load formulation methodology in Polar Rules [14]

The magnitude of the bow impact force depends on the crushing failure and flexural
failure modes of the ice wedge. Therefore, the calculation of bow impact load requires
both ice crushing and flexural failure models. The IACS Polar Rules specify an
additional limiting value of ice failure load. According to the Polar Rules, the bow
impact load cannot exceed this limiting value. The minimum of the ice crushing load
(Fc), flexural failure load (Ff) and limiting failure load (Fl) is considered as the bow
impact load at a particular bow region. The Polar Rules based bow impact load formula
can be expressed as [6, 11, 14]:
8 9
>
< Fc ¼ f a CFc D2 >
0:64
=
fh
Fi PC ¼ min Ff ¼ 1:2r 0 ð1Þ
>
: sin b >
;
Fl ¼ 0:6CFc D0:64

The shape coefficient (fa) in Eq. (1) is defined as:

 x 2  a
fa ¼ 0:097  0:68  0:15 pffiffiffiffi0 ð2Þ
L b

Equation (1) indicates that the bow impact load is a function of ice flexural strength
(rf), ice thickness (h), crushing class factor (CFc), ship displacement (D), impact
location (x) from the forward perpendicular, ship length (L) measured on the upper ice
waterline and different hull angles such as water line angles (a) and normal ship frame
angles (b′). In this study, all the ice load parameters are in MN, all the angles are in
radians, ship displacement is in kt, ice flexural strength is in MPa and all other
parameters are in SI units unless otherwise specified. The water line angles and normal
ship angles at impact locations can be calculated from the hull angle definition of Polar
Rules (Fig. 4).
 Effect of Ship Speed on Bow Impact Load Estimation 413

Fig. 4. Hull angle definition in Polar Rules [9]

The crushing class factor (CFc) for a particular Polar class ship can be selected from
Table 1 or estimated directly from following equation:

CFc ¼ P0:36
o Vs
1:28
ð3Þ

Table 1. Polar class dependent parameters for bow load calculation [9, 14]
Polar classes Vs m/s Po MPa hm rf MPa CFc
PC 1 5.70 6.00 7.0 1.40 17.69
PC 2 4.00 4.20 6.0 1.30 9.89
PC 3 3.00 3.00 4.3 1.20 6.06
PC 4 2.50 2.45 3.5 1.10 4.50
PC 5 2.00 2.00 3.0 1.00 3.10
PC 6 1.75 1.50 2.8 0.70 2.40
PC 7 1.50 1.30 2.5 0.65 1.80

The Eqs. (2) and (3) indicate that the crushing class factor (CFc) and hence the
crushing impact force (Fc) is a function of nominal ice crushing strength (Po) in MPa
and ship speed (Vs). On the other hand, the flexural impact force (Ff) is a simple
function of ice flexural strength (rf), ice thickness (h) and normal ship frame angles
(b’), and insensitive to ship speed. In the next section, this flexural failure model is
replaced with a speed dependent flexural failure model to investigate the speed effect
on the bow impact load. Table 1 lists the Polar class dependent parameters for the bow
load estimations. In the table, PC 1 and PC 7 represent highest and lowest ice capability
of ships, respectively. The Polar class dependent parameters will be used for the bow
impact load estimation in Sect. 4.
414 M. Sazidy et al.

3 Speed Dependent Ice Flexural Failure Model

A speed dependent ice flexural failure model was developed by Sazidy et al. [1]
through a numerical model of ship ice wedge breaking process. The numerical model
was based on the local contact mechanism and accounted for the ice edge crushing, ice
flexural failure and the water foundation effect, all of which have important influences
on the speed dependencies of ship icebreaking process. The model investigated the
effect of ice thickness (h), ice wedge angle (h), normal ship frame angle (b’) and ship
speed (Vs) on the icebreaking force and ice breaking pattern. The results from the
investigation were then used to formulate the speed dependent ice flexural failure
model through curve fittings and following one factor at a time approach (OFAT). The
formulated model was validated by Sazidy [15] against full scale ice impact test data
and a non-linear finite element model. The model assumed that the flexural strength
(rf) is linearly related to the impact force similar to the Polar ice flexural failure model
of Eq. (1). The model is well suited to any vessels including low and non-ice class
ships for predicting the flexure dominated icebreaking force for any practical range of
ship, ice and operating parameters. The details of the model can be found in Sazidy
[15]. The vertical component of the impact force is:

Ffvsd ¼ 0:29kv rf h2 h ð4Þ

The dynamic factor (kv) in Eq. (4) is defined as:

kv ¼ 1 þ 2:57sina cos b0 h F0:26

0:2
r ð5Þ

where Froude Number (Fr) is defined as:

Vs sina cos b0
Fr ¼ pffiffiffiffiffi ð6Þ
gh

where g is the acceleration due to gravity.

The normal impact force can be written as:

Ffvsd
Ff ¼ ð7Þ
sinb0
nsd

Now, the Polar ice flexure failure model in Eq. (1) is replaced with the speed
dependent flexural failure model to estimate the bow impact load. Therefore, the
modified model for the bow impact load calculation at a particular position can be
expressed as:
8   2  9
>
> ¼ ð0:097  x  0:15 p affiffiffi 0:64 >
>
> F
< c
0:68 L b0
CF c D >
=
 
Fi m ¼ min 0 0:2
0:29 1 þ 2:57 sin a cos b h Fr
0:26 2
rf h h ð8Þ
>
> Ff nsd ¼ >
>
>
: sin b0 >
;
Fl ¼ 0:6CFc D0:64
 Effect of Ship Speed on Bow Impact Load Estimation 415

4 Bow Impact Load Estimation

The bow impact loads are estimated using Eq. (8) for the PC 2 and PC 6 ice class ships
at constant ice thickness (1 m), different ice thicknesses (0.5–5.0 m) and different ship
displacements (25–200 kT for PC 2 and 10–150 kT for PC 6). At a constant ice
thickness, the bow impact loads are compared with the Polar Rules based estimation
from Eq. (1). The principal particulars of the ships and ice wedges are listed in Table 2.

Table 2. Principal particulars of ships and ice wedge

Parameters PC 2 PC 6
Description Year-round operation in Summer/autumn operation in medium
moderate multi-year ice first-year ice which may include old ice
conditions inclusions
Displacement, D 25,000 tonnes 10,000 tonnes
Block 0.77 0.77
coefficient, Cb
Water plane 0.75 0.75
coefficient, Cpw
Mid-ship 0.94 0.94
coefficient, Cm
Waterline 45° 30°
angle, a
Normal ship 65° 40°
frame angle, b’
Ice wedge 150° 150°
angle, h
Flexural strength 1.3 MPa 0.7 MPa
of ice, rf

For a constant ice thickness of 1.0 m, the bow impact load vs ship speed curves are
plotted in Fig. 5 and Fig. 6 for the PC 2 and PC 6 ships, respectively. These figures
indicate that the crushing load represents the limiting impact load at slow ship speeds.
Within this slow ship speed range, the impact load increases very quickly with the ship
speed. For medium to higher ship speeds, the flexural failure load is lower than the ice
crushing load, and hence the flexural failure dominates the bow impact load. For this
speed range, a gradual change with ship speed is observed in the flexural dominated
bow impact load.
The bow impact load results for the PC 2 and PC 6 ships at different ice thicknesses
(0.5 m to 5.0 m) are plotted in Fig. 7 and Fig. 8, respectively. The figures indicate that
the crushing failure dominates the resulting impact load for a wide range of ship speed
in the thicker ice. In thinner ice, crushing is also a dominating criterion for the slow
speed interaction. The figure also indicates that the flexure is the primary failure
criterion for the thinner ice at medium to higher ship speeds.
416 M. Sazidy et al.

Fig. 5. Speed effect on bow impact load for an ice thickness of 1.0 m (PC 2)

Fig. 6. Speed effect on bow impact load for an ice thickness of 1.0 m (PC 6)

Figure 9 and Fig. 10 show the bow impact load vs ship speed curves for the PC 2
and PC 6 cases, respectively, for different ship displacements. The bow impact loads
for the PC 2 ship at different displacements (25 kT to 200 kT) are estimated using the
class dependent parameters (Table 1) such as the ice thickness (6 m) and the flexural
strength (1.3 MPa). Similarly for the PC 6 ship, the bow impact loads are calculated for
an ice thickness of 2.8 m, flexural strength of 0.7 MPa and displacement range of 10
kT to 150 kT. The PC design speed lines in Fig. 9 and Fig. 10 indicate that the
crushing is the only failure criterion at lower ship displacements. For higher ship
displacements, both the crushing and flexural failure processes contribute to the bow
impact loads.
 Effect of Ship Speed on Bow Impact Load Estimation 417

Fig. 7. Speed effect on bow impact load at different ice thicknesses (PC 2)

Fig. 8. Speed effect on bow impact load at different ice thicknesses (PC 6)

Fig. 9. Speed effect on bow impact load at different ship displacements (PC 2)
418 M. Sazidy et al.

Fig. 10. Speed effect on bow impact load at different ship displacements (PC 6)

5 Conclusions

The ship speed effect on the bow impact loads has been investigated using a speed
dependent ice flexure failure model. The investigation for a PC 2 and a PC 6 ship at
constant ice thickness, different ice thicknesses and different ship displacements found
significant speed effects in both the crushing and flexure dominated bow impact loads.
The Polar Rules based bow impact load estimation method is reasonable for the
crushing dominated thicker ice operation, slow speed interaction or for ships with
lower displacement. For thinner ice at medium to higher interaction speeds or ships
with higher displacements, the Polar Rules based method may not be appropriate as the
flexural failure model is insensitive to interaction speed. Therefore, the speed depen-
dent flexural failure model represents a useful improvement over the current Polar
Rules based model and can be used in the Polar class ship design process for the
formulation of design ice load parameters. Further investigation should aim to explore
the effect of ship waterline angles, ice flexural strength and ice wedge angles consid-
ering different impact scenarios such as head-on impact, impact with pressure ridges
etc.

Acknowledgements. The authors would like to acknowledge the financial support of Atlantic
Canada Opportunities Agency, Research and Development Corporation (RDC) of Newfoundland
and Labrador, American Bureau of Shipping (ABS), BMT Fleet Technology, Husky Energy,
Rolls-Royce Marine, Samsung Heavy Industries and the National Research Council of Canada.
 Effect of Ship Speed on Bow Impact Load Estimation 419

References
1. Sazidy, M., Daley, C., Colbourne, B.: A mathematical model of icebreaking for safe speed
assessment. In: 11th International Conference and Exhibition on Performance of Ships and
Structures in Ice (ICETECH) Banff, Alberta, Canada (2014)
2. Kubat, I., Timco, G.: Vessel damage in the Canadian Arctic. In: Proceedings of the
International Conference on Port and Ocean Engineering Under Arctic Conditions (2003)
3. Likhomanov, V., Timofeev, I.S., Stepanov, I., Kastelyan, V., Iltchuk, A., Krupina, N.,
Klenov, A.: Scientific basis and methodology of the development of an ice passport. In.
Report Prepared for Arctic and Antarctic Research Institute Department of Ship Performance
in Ice, St. Petersburg, Russia (1997)
4. Wallin, F., Åkerström, C.-J.: Ice class requirements on side shell structures-a comparison of
local strength class requirements regarding plastic design of ice-reinforced side shell
structures. Master’s thesis, Chalmers University of Technology (2012)
5. Dolny, J., Yu, H.-C., Daley, C., Kendrick, A.: Developing a technical methodology for the
evaluation of safe operating speeds in various ice conditions. In: Proceedings of the
International Conference on Port and Ocean Engineering Under Arctic Conditions (2013)
6. Rahman, M.: Structural resistance of polar ships and FPSO’s to ice loading. Master’s thesis,
Norwegian University of Science and Technology (2012)
7. Zong, R.: Finite element analysis of ship-ice collision using LS-DYNA. Master’s thesis,
Memorial University of Newfoundland (2012)
8. Brigham, L.: Polar code heralds a new era of safer navigation in Arctic waters (2017). https://
www.adn.com/opinions/2017/01/10/polar-code-heralds-a-new-era-of-safer-navigation-in-
arctic-waters/
9. Daley, C.: IACS unified requirements for polar ships-background notes to design ice loads.
In. Report Prepared for Transport Canada, St. John’s, NL, Canada (2000)
10. Popov, Y., Faddeev, O., Kheisin, D., Yakovlev, A.: Strength of ships navigating in ice. In:
Sudostroenie Publishing House [Technical Translation by U.S. Army Foreign Science and
Technology Center, FSTC-HT-23-96-68], Leningrad, Russian (1967)
11. Daley, C., Kendrick, A., Quinton, B.: Safe Speed in Ice. Draft Final Report Prepared for
BMT Fleet Technology, Reference: 6931DFR. Rev00, Ottawa, ON, Canada (2011)
12. Lubbad, R., Løset, S.: A numerical model for real-time simulation of ship–ice interaction.
Cold Reg. Sci. Technol. 65(2), 111–127 (2011)
13. Su, B., Riska, K., Moan, T.: A numerical method for the prediction of ship performance in
level ice. Cold Reg. Sci. Technol. 60(3), 177–188 (2010)
14. Daley, C.: Polar Class Rules Overview (2014). http://www.engr.mun.ca/*cdaley/8074/
Polar%20Rules%20Overview_CD.pdf
15. Sazidy, M.: Development of velocity dependent ice flexural failure model and application to
safe speed methodology for polar ships. Doctoral dissertation, Memorial University of
Newfoundland (2015)
Navigation and Logistics
 Safety vs. Sustainability – How Much
Underpowering of Ships Is Acceptable?

Stefan Krüger(&), Michal Josten, and Leonidas Souflis

Institute of Ship Design and Ship Safety,

Hamburg University of Technology, Hamburg, Germany
[email protected]

Abstract. The present EEDI regulations have a strong impact on ship design.
Especially for tankers and bulk carriers, the regulation results in the reduction of
the installable main engine power. In 2020, the next EEDI reduction phase is put
into force with the consequence that the main engine power of these ships will
be reduced even further. The development helps to reduce emissions, especially
greenhouse gases, but it puts challenges on ship safety, as the ship must still be
maneuverable even in bad weather conditions. Recently, a grounding incident of
a bulk carrier sea took place in the German Bight where the ship was dragging
its anchor although the main engine was running at full speed. Numerical
investigations of this accident showed that the environmental forces acting on
the ship were larger than the sum of propeller and anchor forces. This incident
was an excellent opportunity to investigate whether our existing numerical
calculation methods are able to cope with such kind of incidents and whether it
was possible to investigate the root cause of that particular grounding event. The
paper gives an overview about the numerical investigations of the grounding
event which address course keeping in heavy weather including anchor forces.
Additionally, operational aspects of the main engine operation and the propul-
sion control systems are considered. The ship did not need to comply with the
existing EEDI regulations. Therefore, the paper addresses also the effect of
reduced main engine power on this particular incident, because it is obvious that
reduced main engine power will make such incidents more probable, as limiting
environmental conditions for a ship with less power will be less severe and
therefore more probable.

Keywords: Maneuvering in adverse weather conditions
EEDI

1 Introduction

Since the Energy Efficiency Design Index (EEDI) has been adopted, the installed
power of ships is subject to a continuous reduction. In fact, the EEDI regime supports
those ships which have a large capacity and a slow ship speed. The fulfilment of the
EEDI is hardly possible without a substantial reduction of the installed engine power.
Therefore, the question has been put forward whether ships with reduced installed
power can still be safely maneuvered in adverse weather conditions. The European
research project SHOPERA [9] has intensively studied this matter. Currently it is under
discussion at IMO whether further regulations for minimum installed power shall be

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 423–435, 2021.
https://doi.org/10.1007/978-981-15-4680-8_31
424 S. Krüger et al.

developed or not. In this context it must be mentioned that already two such rule sets
exist: The minimum required power demands for safe navigation in ice and the min-
imum power requirements for safe return to port. From the ship design point of view it
clearly leads to severe design problems if one regulation prescribes a maximum
installable power and a second regulation prescribes a minimum installable power.
The EEDI formulates a maximum installable power, and the two above mentioned rule
sets clearly describe a minimum power requirement. As it is presently not intended that
a minimum power requirement (from safety considerations) will overrule the EEDI
(environmental view point), ship design is confronted with two conflicting require-
ments concerning the selection of the main engine power. This problem will become
really severe when the next EEDI reduction phase will enter into force in 2020. For the
purpose of investigating maneuvering in adverse weather conditions, it was found
useful to apply a first principle based direct computational procedure to analyze
maneuvering in adverse weather conditions, which is in our analysis defined as course
keeping against wind and waves. The computational procedure must at the same time
be fast enough for application during the basic design phase. It must also ensure that all
maneuvering devices can be designed for a prescribed course keeping procedure. The
method is firstly used to investigate a grounding incident which took place in 2017 in
the German bight, where the ship grounded due to insufficient ability to keep its
position against wind and waves. It is then analyzed if the ship is also endangered in
other situations, and finally, the effect of reduced power due to the EEDI- regulations
after 2020 is numerically studied.

2 Course Keeping Problem

Fig. 1. Principal definitions of the course keeping problem

The principal definitions of the course keeping problem are shown in Fig. 1. The ship
travels with the ship speed vS and it is exposed to environmental forces, which may be
a combination of wind, waves and current. These external forces result at first in a
longitudinal force, which is called added resistance. This added resistance directly leads
 Safety vs. Sustainability 425

to an increase of the propulsion power. The external forces do also result in a hull cross
force and a hull yawing moment. These are generated by a combination of drift motion
of the ship (drift angle b) and a rudder yawing moment (rudder angle d). Both drift
motion and the rudder action result in an additional longitudinal force which also
increases the resistance. The propulsion efficiency of the ship is affected by the fact that
the propulsor(s) now work(s) in oblique flow due to the drifting motion. From the view
point of maneuverability in heavy weather, the most important question is the ability of
the rudder(s) to produce sufficient yaw moment to enable the ship to keep its course. As
the inflow to the rudder(s) is strongly influenced by the propeller(s) (especially under
the heavy loading), it is of interest whether the installed power is sufficient to guarantee
the required inflow to the rudder(s). The calculation is therefore divided into two steps:
At first, two non-linear equations need to be solved to obtain the rudder angle and the
drift angle. If the ship cannot keep its course, there exists no solution for the problem.
During the second step, the added resistance, the propeller and rudder inflow need to be
calculated. This results in propeller rpm and pitch setting and the delivered power of
the drive train(s), provided that the torque/rpm combination is still located in the engine
load diagram. The aim of the procedure is to identify limiting situations for course
keeping. As course keeping problem is a maneuvering problem, we use our maneu-
vering model for the computational procedures. This model is briefly described in the
following section.

3 Maneuvering Model
3.1 General
For the present investigations, we use the body fore maneuvering model which is
implemented in the ship design system E4. The model was developed by Söding [10]
and it is continuously developed by Krüger et al. [6] on the basis of full scale mea-
surements of ships during sea trials and during their operation. The model is used
during the initial design phase for the layout of the hull and the maneuvering devices as
well as during the delivery of the ship for the generation of the maneuvering booklet
and the wheel house poster. The typical accuracy for full scale ships is about ship’s
breath for distances such as tactical diameters or advances and about 2° for overshoot
angles. The overall accuracy of the maneuvering model is demonstrated by Fig. 2.
Figure 2 shows a hardcopy from the sea trial measurements of two nautical maneuvers.
The curves show the calculated maneuver before the ship went on sea trial. The
measurements for the overshoot angles and the turning circle characteristics were taken
from the ship’s instrumentation and they were immediately plotted into the pre cal-
culated maneuvers. It is obvious that the agreement is quite good. This proven
maneuvering model is now adapted to the course keeping problem.
426 S. Krüger et al.

Fig. 2. Two examples for the prediction of nautical maneuvers and for their first visual
validation during the sea trial. Left: 20/20 Zig Zag, Right: 44° rudder turning circle. The
measurements from the ship`s bridge instrumentation are compared against the values of the pre
calculated maneuvers (curves) during the sea trial (hand written values).

3.2 Body Forces

Body Forces are computed according to the slender body theory developed by Söding
[11], see Eq. 1, where X2 denotes the longitudinal force, Y2 the cross force and N2 the
yawing moment:

X2 ¼ RT þ Xvv v2S þ Xvr vs r þ Xrr r 2

_ Z L
RF ðvS þ xT wÞ 
Y2 ¼  0:5q ðvS þ xwÞ_ vð þ xwÞc
_ D d dx
uS S ð1Þ
Z L 0
RF L2 w_ 
_ vð þ xwÞc
_ D d dx
N2 ¼  0:5q xðvS þ xwÞ S
6uS 0

The cross force Y2 and the yawing moment N2 consist of two major contributions:
The first part is due to the acceleration of the section added mass and the second part is
due to the viscous cross resistance of a frame. Both contributions have to be integrated
over ship length. The drag coefficient CD can be identified form full scale measure-
ments or from viscous calculations [8] and it lies in the range of 0.3 (for frames with
large bilge radii) to 1 (for V- shaped frames). But for the course keeping problem at
small drift angles, the viscous contribution is small against the added mass part. It
should be noted that the course keeping of a ship strongly depends on the longitudinal
distribution of section added mass, where the section added mass depends on the hull
form and on the floating condition. Figure 3 shows the longitudinal distribution of the
section added mass for 0 and 15° heel floating condition for the sea trial design floating
condition of the investigated bulk carrier during grounding condition. Our body force
model includes the roll degree of freedom indirectly: From hydrostatic considerations,
the heel angle is found, and section added mass as well as viscous forces are corrected
for the heel [10]. The coefficient Xvv which gives the longitudinal force per drift angle
can be identified from full scale measurements, such as shown in Fig. 2.
 Safety vs. Sustainability 427

Fig. 3. Longitudinal distribution of section added mass for 0 and 15° heel of the investigated
bulk carrier.

3.3 Propeller and Rudder Forces

Propeller forces are obtained from lifting line calculations. This includes computations
in oblique inflow to include the propeller cross forces and the alterations of thrust and
torque. These calculations include alterations of the propeller blade angle if the pro-
peller is a controllable pitch propeller. Rudder forces are computed by a direct panel
method by Söding [11] where the rudder inflow comes from a lifting line theory [4].
Figure 4 shows the propeller/rudder model of the investigated bulk carrier for zero
degree rudder angle and design speed, trial conditions.

Fig. 4. Panel model for rudder forces in the propeller slipstream for the investigated bulk carrier.
428 S. Krüger et al.

3.4 Interaction Effects

Most important interaction effects are the alteration of the effective wake and the cross
flow to the pro-peller. The cross flow vc to the propeller due to the drifting and turning
motion can be modelled according to the Kose approach [10, 12] as:

Vc ¼ k1 vy þ k2 rot xP

Where vy and rot denote the rigid body lateral speed and turning rate of the ship. xP
denotes the distance between propeller and pivoting point. k1 and k2 are factors which
are below 1, because due to flow separation in the aft body the cross flow is smaller
than the rigid body motion. For single screw vessels, k1 and k2 have been proposed as
0.36 and 0.66, but for twin screws this is not applicable as the cross flow depends on
the combination of signs of turning rate, location of propeller and its turning sense.
Furthermore, for twin screw vessels these factors depend also on the ratio of vy and rot,
where for course keeping problems, the rate of turn can be assumed to be close to zero.

3.5 Propulsion Control System

The model includes a propulsion control system for the calculation of propeller pitch
and rpm settings during each time step. The lever command sets initial values for rpm
and pitch according to the predefined combinatory curve, and both overload and wind
milling switches modify the rpm or pitch value. In case of transient maneuvers such as
crash stop astern, there are predefined engine slowdown and startup ramps. For the
course keeping analysis in the present context, only the predefined combinatory curve
is relevant for pitch and rpm control, as we compute time averaged values only.

4 External Loads

4.1 General
Wave loads, wind and current are acting on the ship, but in the present context, current
is disregarded. Although all force modules work in time domain, we have taken time
averaged values for the external forces and we search the course keeping equilibrium
for these static values. See state and wind parameters can be selected independently
from each other, but for the present analysis we have selected wave period and wave
height depending on the wind force according to the established ERN- concept by
DNV GL, if not stated otherwise.
 Safety vs. Sustainability 429

4.2 Wind Forces

Wind forces are computed on the basis of Blendermann’s coefficients [3], see.
Figure 5.

Fig. 5. Wind force coefficients for the investigated Bulk Carrier

4.3 Wave Drift Forces

Second order wave drift forces are computed in the present context by means of a strip
theory according to Augener [1]. The computation includes longitudinal forces,
transversal forces and the yawing moment. The longitudinal forces may limit the course
keeping ability with respect to the available power, the yawing moment and cross force
with respect to steering. Figure 6 shows the computed second order longitudinal force
time series for a JONSWAP spectrum with T1 = 11 s, H1/3 = 7 m of the investigated
bulk carrier.

Fig. 6. Computed time series for the second order longitudinal drift force for the investigated
Bulk Carrier for a JONSWAP- spectrum with T1 = 11 s, H1/3 = 7 m, encounter angle 150°. The
time average amounts to abt. 980 kN.
430 S. Krüger et al.

5 The Grounding Incident of a Bulk Carrier

On October 29th, 2017, the 75000 dwt Bulk Carrier Glory Amsterdam grounded in the
German bight due to severe weather conditions. The ship was at anchor when a storm
came up. The ship was dragging its anchor and the crew decided to start the main
engine. Despite the main engine running at full speed possible, the ship was still
dragging the anchor. A second anchor war brought out, but the ship was still drifting
backwards until she finally grounded. The accident investigations were carried out by
the German Federal Bureau of Marine Casualty investigations (BSU), and they are
summarized in the Report BSU 408/17 [2]. Figure 7, left shows the track of the ship
and Fig. 7, right, shows the ship in the final position after she has grounded.

Fig. 7. Track of the grounded bulk carrier (left) and the final floating position after she has
grounded. Note the setup of the two anchors.

During the investigation of the grounding accident, the question came up whether
the vessel had sufficient power to maintain its position against the waves. The wind was
strong with speeds of abt. 74 knots. The significant wave height was about 7 m at a
period of abt. 11 s.

Fig. 8. Mooring setup of the ship as analyzed by Josten (left) and an animation of the ship in
waves (right), accident condition.
 Safety vs. Sustainability 431

As the accident happened at an encounter angle of abt. 180°, steering forces are of
less relevance for the accident itself. They will be treated in the following sections. In
head wind of 74 knots and waves of T1 = 11 s, H1/3 = 7 m we compute abt. 980 kN
for the second order wave drift force (see Fig. 6) and of abt. 300 kN for the wind force.
The anchors could deliver a theoretical static design holding force of abt. 300 kN. This
value was theoretically obtained from the layout regulations of anchor equipment and
cross checked by catenary calculations for the actual anchor situation. Then, there
remain 980 kN which have to be delivered by the propeller. This case therefore clearly
shows that a ship must have sufficient power to be safely kept against the wind and
waves, as the anchors of marine merchant vessels are not designed for forces which
occur in bad weather conditions.

Fig. 9. Intersection of the computed propeller curve and the engine torque limit (left). Engine
rpm signal from the voyage data recorder (right). Source: BSU

Figure 9, left, shows the intersection of the propeller curve (including the external
forces) and the torque limit curve of the main engine. It shows an intersection at 62 rpm.
In this condition, the ship speed is nearly zero, which means that under these conditions,
the propeller could hardly deliver sufficient thrust to keep the vessel in position against
the combined action of wind and waves. Unfortunately, the ship has a barred speed
range between 52 and 62 rpm. Theoretically, at 62 rpm the propeller could hardly
produce the thrust to keep the ship against wind and waves, but the engine could not
increase its revolutions beyond 62 rpm. Consequently, the automation reduces the rpm
to get out of the barred speed range, and then it tries again to increase the rpm, as the
lever command is full ahead. This leads to engine hunting, where the resulting average
propeller thrust is clearly lower than 980 kN. This hunting effect results in the engine
rpm signal recorded by the ship´s voyage data recorder (Fig. 9, right). Consequently, the
ship must drag the anchors and she moves backwards with abt. 2 knots until she finally
grounds. From this grounding event, the following conclusions can be drawn:
• Our method is able to correctly compute the root cause of the grounding event,
which means that the computed forces seem to match with the reality.
• A ship must have sufficient engine power to be able to maneuver in harsh weather
situations, as the main engine (combined with the rudder) is the only means to keep
the position of the ship. The anchor forces are by far too small.
432 S. Krüger et al.

• In theory, the ship (including its anchors) could hardly keep its position against the
weather conditions, but without any margin. This shows that some safety margin
must be included in minimum power considerations, and the effect of barred speed
ranges needs to be taken into account.
In the following, we will study the effect of power limitations on the course keeping
problem of the ship in general.

6 Effect of Power Limitations on Course Keeping

The effect of a power limitation on the course keeping performance of the ship is twofold:
At first, the ability of the ship to keep her position against the combined action of wind and
waves depends of the generation of sufficient propeller thrust. As the rudder forces (for a
given rudder angle) depend solely on the propeller forces, also the steering capability of
the ship depends on the available power. Both effects are studied in this section.

Fig. 10. Computed drift angles (left) and rudder angles (left) of the bulk carrier in the weather
conditions of the accident.

Figure 10 shows the computed drift and ruder angles of the bulk carrier in the
weather conditions of the accident. As it is hardly possible to keep the course in
following seas due to the fact that the rudder is ineffective [7], we have restricted our
investigation to head seas. The radial rings show the ship speed, which was varied from
zero to 9 knots. The vessel is assumed to be in the center of the diagram, and the
encounter angle of the combined action of wind and waves varies from port beam to
starboard beam direction. The diagrams are slightly asymmetric, as the rudder is more
effective on the starboard side due to the right turning propeller. Figure 10, left shows
the computed drift angles, and it can be seen that the drift angle converges to 90° in
pure beam sea and wind conditions for a forward speed close to zero. The largest
rudder angles are required for bow quartering seas and beam seas at larger speed. In all
situations investigated, the rudder is able to provide the required steering forces. This
may indicate that the size of the rudder is sufficient. In the calculations, we have
assumed that the power is not limited. This is the reason why was at all possible to
perform the calculations. Because it was shown above that the ship could not even keep
zero speed as the installed power was too small. Therefore, each condition analyzed in
 Safety vs. Sustainability 433

Fig. 10 must be checked for the possible intersection of the resulting propeller curve,
see Fig. 9, left. If that would have been done, not a single point in the polar diagrams
would remain. Therefore, we have computed the course keeping limit for the ship by
reducing the environmental load (wind and waves) in such a way that for each com-
bination course and speed, the following criteria are met:
• The power demand must be within the torque limit of the engine overload limit (see
Fig. 8), and
• the rudder angle required for course keeping must be smaller than the 35° the
steering gear permits, and
• the transversal speed resulting from the drifting motion must not lead to a drifting
backward situation.
The computations result in the course keeping capability plot shown in Fig. 11.

Fig. 11. Course keeping capability plot for the bulk carrier. The rings denote the environmental
conditions (wind and sea state). The computations vary the encounter angle from 90° (beam sea
and wind) to 0° (head sea and wind).

Figure 11 shows the computed results in a course keeping capability plot. The rings
denote the weather condition. The wind force in BFT is connected to a corresponding
sea state according to the DNV ERN notation, and the combined action of wind and
waves has been taken into account. The encounter angle varies from beam seas to head
seas. The curve denoted by “Current engine power” shows the results for the vessel as
built. The ship was built in 2006 and does therefore not comply with the existing EEDI-
regulations. 9326 kW have been installed as main engine power. We have computed
the effect of the different EEDI stages on the powering of the bulk carrier, and the
results are the following:
434 S. Krüger et al.

• Presently, 9326 kW are installed in the bulk carrier.

• If the ship needs to comply with the present EEDI-regulations, 8490 kW could be
installed.
• If the ship would need to comply with the EEDI after 2020, 7197 kW could be
installed.
• If the ship would need to comply with the EEDI after 2025, 6019 kW could be
installed.
Our calculations show that quite a significant reduction in the powering of the ship
is to be expected. Based on these new powering figures, we have repeated the course
keeping capability computations and plotted the results in Fig. 11. It can be seen that
the EEDI has a significant impact on the course keeping ability of the ship. After 2025,
limiting weather conditions vary from BFT 6 in head seas to abt. BFT 8 in beam seas.
The grounding incident happened in the German Bight (North Sea). For this region,
the statistical distribution of the weather condition is known from the German BSH. If
the statistical distribution of the weather conditions is linked to the course keeping
capability of the ship, the results are the following:
• 13.7% of the weather conditions are more severe than the limiting conditions for the
ship with 9326 kW installed power.
• 14.9% of the weather conditions are more severe than the limiting conditions for the
ship with 8490 kW installed power.
• 18.8% of the weather conditions are more severe than the limiting conditions for the
ship with 7179 kW installed power.
• 21.3% of the weather conditions are more severe than the limiting conditions for the
ship with 6019 kW installed power.
The results show that after 2020, the probability of exceeding the course keeping
ability of the ship in the North Sea is significantly increased. Our computations have
not taken into account the effect of a barred speed range on the course keeping ability,
and we have also not taken into account that the effectiveness of propeller and rudder
could be decreased in situation s where they are not fully immersed.

7 Conclusions

A method was presented to compute the course keeping ability of a ship in adverse
weather conditions. The method is based on an existing maneuvering model which was
extended for course keeping problems. The method was applied to a grounding event
which took place in 2017 in the German Bight (North Sea). The root cause of the
grounding was the fact that the anchors did not provide sufficient holding forces to keep
the ship against the combined action of wind and waves. This shows that the course
keeping must be managed with the main engine and the rudder only. It turned out that
the main engine was not powerful enough to keep the ship against the weather con-
ditions. This was also due to the fact that main engine could not operate in the desired
area due to a barred speed range. The course keeping capability of the ship was
computed by our method, where sufficient course keeping was defined from three
 Safety vs. Sustainability 435

criteria. The results showed that due to future power reduction of the ship as defined by
the EEDI regulations, the probability of a comparable grounding event is significantly
increased after 2020.

References
1. Augener, P.: Computation of Wave Drift Forces for Dynamic Positioning within the Early
Design Stage. Ph.D-Thesis, Hamb. Univ of Techn. (2016)
2. The Federal Bureau of Marine Casualty Investigations: Report 408/17
3. Blendermann, W.: Practical Ship and Offshore Areodynamics. Schriftenreihe Schiffbau,
Rep. 669. Hamburg (2013)
4. Isay, W.H.: Propellertheorie. Springer, Heidelberg (1963)
5. Josten, M.: Trajectory prediction of moored vessels with reduced station capability due to
exceeded anchor load limits. In: OMAE, 2019, Glasgow, UK (2019)
6. Krüger, S.: Maneuvering – which results can be obtained from strip methods? In:
Proceedings of JSTG 2018, vol. 108, Hamburg (2018)
7. Billerbeck, H., Krüger, S., Lübcke A.: Safe maneuvering in adverse weather conditions. In:
Proceedings of IMDC 2018, Helsinki, Finland (2018)
8. Lübcke, A.: Dynamic positioning as a new aspect in the early design stage of ships –
calculations, model tests and results. In: Proceedings of JSTG, 2014 vol. 107, Hamburg
(2014)
9. Papanikolau, A., Zaraphonitis, G., Bitner-Gregersen, E., Shigunov, V., EL Moctar, O.,
Guedes Soares, C., Reddy, D.N., Sprenger, F.: Minimum propulsion and steering
requirements for efficient and safe operation (SHOPERA). In: Motorship P&EC Conference,
Hamburg, 4–5 March 2015 (2015)
10. Söding, H.: Body Forces. Manoeuvring Technical Manual. Seehafen Verlag, Hamburg
(1993)
11. Söding, H.: Limits of potential theory in rudder flow predictions. Ship. Techn. Res. 45, 3
(1998)
12. Vorhölter, H.: Numerische Analyse des Nachstroms und der Propellereffektivität am
manövrierenden Schiff. Ph.D-Thesis, Hamb. Univ of Techn. (2011)
 Sustainability Assessment of Small Scale
Fishing Vessel Operations: A Case Study
in Palabuhanratu, Indonesia

Vita R. Kurniawati1,2(&), Richard W. Birmingham2,

and Alan J. Murphy2
1
IPB University, Bogor 16680, Indonesia
[email protected]
2
Newcastle University, Newcastle NE4 7RU, UK

Abstract. The important factor in promoting sustainable fisheries is imple-

menting a sustainable fishing operation that is not only environmentally friendly
but also economically and socially acceptable. As most of the existing
improvement focused on the fishing gear, this study concentrates on the fishing
vessel. Investigation on the existing practices is a key to formulation of man-
agement measures, hence a sustainability assessment has been carried out on
small-scale fishing vessels by incorporating environmental and economic aspect
of fishing operations. A case study was conducted in Palabuhanratu, a region
with a significant number of small-scale fishing practices, which is typical of
many fishing regions in Indonesia and other developing nations. Assessment
was conducted on four principal types of vessels representing pelagic-ground
fishing and active–passive fishing. Amongst the studied vessels, the lift net
vessel which operates pelagic-passive fishing generates the least impact in both
environmental and economic indicators, whilst the largest impact is derived
from the trammel net vessel. In general, passive fishing has better performance
in comparison to active fishing. Compared to pelagic fishing, ground fishing
contributes higher impacts on environment, however in terms of economic, this
operation catches more valuable fish than its counterpart. This result suggests
the importance of trade-off when formulating management measures to promote
the sustainable fishing vessel operations.

Keywords: Fishing operation
Life cycle assessment
Small-scale fishing

vessel
Sustainability impact assessment

1 Introduction

As a food source for thousands of years, capture fishery is believed to play an important
role in human life. However, its sustainability is threatened by the fact that the world’s
fish stock has continued to decrease since the mid 1990’s followed by ecosystem
degradation, pollution and fuel inefficiency [1, 2]. One of the major contributors is the
excessive fishing efforts which are supported by some technical advancement to
enhance fishing efficiency such as marine propulsion system, vessel capacity, and fish

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 436–461, 2021.
https://doi.org/10.1007/978-981-15-4680-8_32
 Sustainability Assessment of Small Scale Fishing Vessel Operations 437

finding devices [3, 4]. This condition has developed people’s understanding that fish
stock should be maintained appropriately even though it is naturally replenished.
Sustainable development in fisheries is not only about maintaining fish stock but
also protecting the marine environment hence the gear and the vessel operated for
fishing should be environmentally friendly [5, 6]. Furthermore, as fishing is an eco-
nomic activity which involves human interaction with nature, the use of the gear and
the vessel should consider the profitability and the applicability within the fishing
community [7, 8]. In order to control the use of fishing gear, various regulations and
policies have been made, in addition to the development of selective and biodegradable
fishing gear through research. Regarding the vessels, the most common management
measures is regulating the catch capacity in order to control the fish stock. In fact, little
attention has been given to develop a sustainable vessel throughout its operational
lifetime, which inspires this research to focus on the development of sustainable fishing
vessel operations.
The implementation of sustainable development in fisheries requires sort of man-
agement measures which are formulated based on the performance of the existing
practices. This suggests the importance of impact assessment along the process. Given
that the sustainability concept is associated with the responsibility for future genera-
tions, the life cycle concept is essential in relation to delivering appropriate assessment
[9]. Therefore, the purpose of this study is to assess the sustainability of fishing vessel
operations by using a life cycle approach.
The investigation was concentrated on small scale fishing vessel (SSFV).
According to Food and Agriculture Organization (FAO), approximately 85% of the
global motorized fishing vessels operated in marine waters are small vessels [4]. The
domination of small vessels is found all over the world including Asia, Africa, Europe,
Latin America and the Caribbean, North America, and the Pacific and Oceania. Those
vessels typically associate with personal ownership and run artisanal fishing, a fishing
practice with labor intensive, low investment, and limited access to the fish resources
[10]. As a result, this kind of fisheries is being more vulnerable to the sustainability
issues. In this paper an SSFV is defined as a vessel which is 5 GT or less, traditionally
operated in coastal waters and aimed at supplying local markets.
A case study was conducted in Palabuhanratu West Java, Indonesia, located in the
western part of Java Island directly facing the Indian Ocean (Fig. 1). This region has
been bestowed with potential fisheries resources and evolved as one of the fishing
business centers in West Java Province, even Indonesia. It perfectly represents
Indonesian fisheries as the port is the home for at least 600 vessels, ranging from 2 GT
to 200 GT. About 80% of the fishing fleet is 5 GT or less and performs artisanal fishing
[11]. Those vessels operate eight types of fishing gears which are typically used by
Indonesian fishers, specifically, longline, troll line, purse seine, lift net, trammel net,
handline, pelagic Danish seine, and gillnet. Furthermore, the fishing port provides
sufficient data to support the fisheries research.
The role of the life cycle approach in sustainable development was acknowledged
by the introduction of Life Cycle Thinking in 1992 by the UN and refers to the
identification of the cycle of products or services along with their potential impacts [9,
13]. This approach allows the decision-makers to choose long-term actions by con-
sidering the impacts throughout the lifetime of the system. There are three life cycle
438 V. R. Kurniawati et al.

Fig. 1. Research site, Palabuhanratu-Indonesia [12]

techniques which can be employed in sustainability assessment: environmental life

cycle assessment (LCA), life cycle cost (LCC) and social life cycle assessment (S-
LCA) [9]. Whilst LCA is associated with the investigation of the potential environ-
mental impacts of the product or service throughout its life cycle, LCC calculates the
entire cost component, whereas S-LCA examines social performance.
Sustainability performance of each operation was quantified using a comparative
analysis which scores each vessel based on the ranking system. Even though con-
ducting different types of operation, the characteristics of the fishing community in
Palabuhanratu are mostly homogenous. They live in the same neighborhood with
interconnected occupations, being based in the same port, conduct fishing in the same
areas and have similar work patterns (one-day fishing). Furthermore, most vessels work
with the same vendors and the same buyers. Due to the similarity, the fishing com-
munity was regarded as one community and the social performance was assessed
without a comparative analysis. Therefore, this study concerns on the sustainability
assessment in terms of environmental and economic performance, whilst the social
performance will be discussed separately.

2 Scope of Analysis

2.1 Studied Vessels

The research focused on the four vessels equipped with pelagic Danish seine, trammel
net, handline, and lift net, which henceforth is referred to as pelagic Danish seiner (PD
vessel), trammel netter (TN vessel), handliner (HL vessel), and lift netter (LF vessel).
 Sustainability Assessment of Small Scale Fishing Vessel Operations 439

Those vessels are 5 GT or less and operate daily inside Palabuhanratu bay with a range
from 5 to 40 nm from the fishing port. Vessels operating other gear are not included
given that the vessels are larger than 5 GT, commonly conduct fishing outside the bay
at a distance of at least 60 nm from the fishing port, and the minimum fishing trip is
occurring once a week.
Those four studied vessels represent different operations incorporating active-
passive fishing and pelagic-ground fishing, as illustrated in Fig. 2. In active fishing, the
vessel moves to operate the gear, whilst in passive fishing, the vessel waits when the
gear is being operated. Furthermore, pelagic fishing captures the fish near the surface,
whilst ground fishing captures the fish near the seabed. The following sections describe
the characteristic of each vessel.

Fig. 2. Characteristic and classification of the studied vessels

Pelagic Danish Seiner. The vessel uses a Danish seine net, an encircled net consisting
of a conical bag net, two wings and two ropes. Being operated near the surface to catch
the small pelagic fish the net is referred to as a pelagic Danish seine net. The typical PD
vessel in Palabuhanratu is a 5 GT wooden boat with the principal dimensions being
approximately 12 m (LOA), 2.5 m (B) and 1.8 m (D) respectively. The vessel is
powered by a 40 HP marine outboard engine and using petrol fuel. The operation is on
a daily basis from dawn to dusk in the fishing areas located within 10–20 nm from the
port. The net is operated manually by 10–15 fishers, including the skipper.
440 V. R. Kurniawati et al.

Trammel Netter. The vessel operates a trammel net, a three layers of nylon net which
is vertically immersed or drifted in the water column to catch ground fish and crus-
taceans. A typical TN vessel in Palabuhanratu is a 9 m long wooden vessel and the
tonnage is approximately 4 GT. It is powered by a 24–30 HP inboard engine and
crewed by 3–4 people. Furthermore, the operation is conducted every day from dawn to
dusk. Basically, a TN vessel operates the net passively by transversely stretching the
net under the water to form a wall. However, in Palabuhanratu, a TN vessel primarily
conducts active fishing by encircling the net at a low speed. The net is circled in order
to sweep the seabed for roughly 30–60 min.
Handliner. The vessel operates a handline, a monofilament fishing line consisting of
branch lines which are arranged vertically. A typical HL vessel in Palabuhanratu is a
2 GT fibreglass hulled vessel with a length of between 7–9 m and powered either by a
single 15 HP marine outboard engine or two 4–5 HP multipurpose outboard engines.
The vessel is predominantly operated during the night, on a daily basis. When the gear
is deployed, the engine is deactivated and the vessel is anchored. The main catch
landed from HL vessel is hairtail fish which mainly inhabit the waters just above the
seabed.
Lift Net Vessel. A lift net is a bag shaped net which is horizontally immersed at a
certain depth. A typical lift net operated in Palabuhanratu is intended to catch small
pelagic fish. The net is deployed under a non-powered bamboo platform which is
anchored few miles away from the shore, hence a ferry is used to shuttle the fisher
between the platform and mainland. Therefore, an LF vessel usually consists of one
ferry and 8–10 platforms which are served on a regular basis. The ferry is a wooden-
hulled vessel about 14 m in length, powered by 100 HP marinized inboard engine and
the tonnage is approximately 5 GT. Meanwhile, the platform is approximately
9  9 m with the working deck is about 2 m above sea level. Each ferry and platform
is crewed by 1–2 people. Every day, at around 4 pm, the ferry conducts a return trip
between the port and platforms to shuttle the fishers. In the next morning, another
return trip is carried out to pick up the fishers and the catch from each platform.

2.2 System Boundary

A fishing vessel operation includes all activities started from preparation of the vessel
for going to sea until the vessel lands the catch and is ready for the next operation.
Therefore, investigation will include the impact generated from all fishing inputs and
outputs of the operation. The fishing input is the production factor which are required
to conduct the operation. Fishing inputs for one fishing vessel primarily consists of the
vessel itself, gear, engine, fuel container, fish container and fishing supplies. Mean-
while, the fishing output refers to the outcome of the operation, which is represented in
catch and revenue. The vessel’s performance in terms of catch quantity and monetary
value is not necessarily the same, therefore, in order to show the performance of the
fishing operation in both terms, two functional units are used e.g. per kg catch pro-
duction and per British pound sterling (£) revenue. Table 1 shows the annual catch and
revenue from each vessel which are used to do the calculation.
 Sustainability Assessment of Small Scale Fishing Vessel Operations 441

Table 1. Annual productivity of studied vessels

Fishing vessel The number of trips Catch (kg) Value (£)
PD vessel 218 58,060 31,941
TN vessel 241 3,194 6,488
HL vessel 236 4,182 6,257
LF vessel 216 90,456 34,977

Based on the number of fishing days per year, it can be seen that TN and HL vessels
have more fishing trips than other two vessels. In TN vessel this is primarily due to the
flexibility to switch the operational mode from active to passive, in order to adjust to
the seasonal patterns. Meanwhile, as the response to the seasonal changes, the HL
operations are sometimes conducted during the daytime with a very limited trip. In
contrast, the fishing operations in PD and LF vessels are not easily adjusting to the
weather changes. Furthermore, in LF vessel the operations are stopped for at least six
days per month during the full moon, because the bright moonlight prevents the fish
from approaching the lamps installed under the LF platform.
Regarding catch production, both PD and LF vessels are targeting small pelagic
fish, which migrate in large quantities. Hence, the amount of fish production by both
vessels are significantly higher than other vessels targeting ground fish. At this point, it
can be said that pelagic fishing is noticeably more productive than ground fishing.
Both LCA and LCC require information about fishing inputs and outputs
throughout the vessel’s lifetime. Figure 3 shows that the scope of assessment consists
of three stages e.g. production, use and end of life, and those stages are applied for each
fishing input. As a result, the assessment incorporates multiproduct and requires
intensive data collection. Data were collected from statistic reports, observation and in
depth interview with at least 34 stakeholders including the fishers, boat owners, fish
sellers, government employee and the influential figures. Data were also gathered from
the product information sheets published by the manufacturers.

Fig. 3. System boundary of the study

Regarding end of life treatment, assessments were made based on the way fishers
handle the fishing input when the lifetime is over. Different treatments are specified
442 V. R. Kurniawati et al.

depending on the type of product. For example, most of the wood from the vessels will
be reused or sold to the second-hand market, whilst most of the nets will be sent to
landfill. The proportion for each treatment is estimated based on the fisher’s judgment.
In order to ensure data accuracy, the information for each vessel was gathered from
several respondents and most data revealed similar values. When a significant differ-
ence was found, respondents were asked to explain their reasons and their opinion
about the conflicting data. In the end, the data entered into the calculation were those
which are agreed by the relevant respondents.
Collecting a comprehensive data which describes the lifetime of each input is
challenging, therefore, various exclusions were applied during the assessment:
1. Each fishing input is analyzed only based on the major components required during
each stage. This means the following variables were omitted due to data inaccuracy,
source limitation, and time restriction.
a. Resource and energy associated with transportation
b. Electricity consumption with exceptions for the production and maintenance of
the fishing vessel and ice production
c. The production of solid waste and emissions
2. Resource and energy required for engine production and end of life stages is
omitted for all types of the fishing vessel as no information related to engine
production was available. There are references relating to diesel engines, which are
applicable for the case of diesel-powered vessel [14, 15], however, since each vessel
uses a different type of engine, including the engine in only one vessel would lead
to a biased comparison. Even though the environmental load from engine was
neglected in the LCA, the monetary value was included in the LCC.
3. In the case of inadequate and uncertain information, available LCA result published
by other researchers were used, such us lamps [16] and wood adhesive [17].
LCC is conducted by accumulating the total expenses throughout the life time of
the product/process [18], hence the fishing inputs considered in the LCA are also used
for the LCC. The cost components considered in LCC include capital expenditure
(CAPEX), operation and maintenance costs (OPEX), disposal expenditure (DISPEX)
and externalities which consists of risk expenditure (RISKEX) and environmental
expenditure (ENVEX). It should be noted that not every component will be relevant to
the studied system, hence only CAPEX, OPEX and DISPEX are considered in this
study. Both ENVEX and RISKEX were omitted due to data unavailability.
As the capital cost, CAPEX are associated with the purchasing prices of the fishing
vessel, fishing gear, engine, and other fishing inputs. Subsequently, OPEX consists of
supplies, personnel, selling, and maintenance costs. Supplies costs refer to expenses
relating to fishing supplies including fuel, ice and bait, which are estimated based on
the fisher’s and vendor’s information. Personnel costs are the money paid to the
vessel’s crew and the selling cost is paid to the seller as the selling fee. Maintenance
costs incorporate costs for boat and gear repair in addition to the replacement of various
items. Little information related to the demolition of the fishing inputs was found, as no
particular mechanism was applied for waste treatment. Therefore, DISPEX only con-
siders the residual value attached to the fishing inputs that are saleable on the second-
hand market.
 Sustainability Assessment of Small Scale Fishing Vessel Operations 443

3 Environmental Life Cycle Assessment

3.1 Assessment Method
The LCA was calculated using the IMPACT2002+ method which is run by LCA
software SimaPro Classroom v8.5.2.0. Data for the background process for each
component of the product was primarily obtained from Ecoinvent 3 database, with
small portions carried out using Industry data 2.0 and European Life Cycle Database
(ELCD) databases. The background process is the production of generic materials,
energy, transport, and waste management defined by a third party [19]. In order to
ensure an even comparison between the four vessels, the same background process was
used across the fishing types. For example, sawn wood in any vessel used the same
Ecoinvent 3 database. There are extremely limited processes that are based on Asia
explicitly. Therefore, this study used the background processes that are based on the
valid average for every country in the world or the rest of the world, which in the
Ecoinvent data base it is denoted as GLO and RoW respectively.
The cumulative environmental impact is presented using endpoint categories,
which provide the result in four aggregated environmental problems, i.e. human health,
ecosystem quality, climate change and resource use. As those impacts are expressed in
different units, the value is converted into ecopoints, using the normalization and
weighting method defined in IMPACT 2002+. Originally ecopoint is denoted as Pt,
however due to the small fraction that found, milli-ecopoint is used denoted as mPt.

3.2 Inventory Analysis

Inventory analysis lists the real mass of components required for each fishing input
throughout the vessel’s lifetime, which the details are provided in the supplementary
material. The result is presented in the mass contribution which indicates the total
resources of each fishing input expressed in kg. For this calculation purpose, inputs
which cannot be converted into kg such as electricity were excluded. As seen in Fig. 4,
the result is shown per functional unit.
A substantial amount of the catch from the PD and LF vessels have resulted in
efficient vessel operations by showing a low input per kg catch. However, the catch is
actually low value fish, as in terms of £ revenue the input is increasing. Meanwhile, the
opposite result is shown in the TN and HL vessels, where the input per £ revenue is
lower than per kg catch. Furthermore, the figure reveals that in order to catch 1 kg fish
and to obtain £1 revenue both TN and HL vessels require more inputs than the PD and
LF vessels.
In each vessel, the operational supplies are responsible for the most substantial
input. On average, PD and LF vessels consume 90 and 105 L/trip respectively, whilst
HL and TN vessels bring about 10 and 25 L/trip. In addition, the PD vessel carries
twice as much ice as the supplies brought by the TN and HL vessels. Meanwhile, ice is
not required in the LF vessel.
As with the lifetime input, Table 2 summaries the composition of the inventory
result from each fishing input at different life stages. Given the operating supplies are
the major inputs for the fishing operation, the use stage takes the largest proportion in
444 V. R. Kurniawati et al.

Fig. 4. Inventory analysis for LCA calculated per kg catch and £ revenue

each vessel. Generally, it is followed by end of life, maintenance and production stages,
leaving other stages with minor contribution. This result is in line with the study
conducted on an anchovies fishing fleet [20]. The study reveals that the largest mass
contribution was derived from the use stage (69.50%), followed by end of life (9.60%),
maintenance (6.70%) and production (2.90%).

Table 2. Composition of each fishing input at different life stages

Life stage Fishing input PD vessel TN vessel HL vessel LF vessel
Production Fishing vessel 1.18% 2.19% 0.35% 1.26%
Fishing gear 0.05% 0.03% 0.01% 12.03%
Fish container 0.02% 0.11% 0.15% 0.54%
Fuel container 0.00% 0.01% 0.00% 0.04%
Lamps – – 0.00% 0.01%
Use: Supplies Fuel 50.09% 32.86% 21.72% 46.62%
Lubricant 1.99% 0.00% 0.27% 1.41%
Ice 41.71% 56.45% 76.10% 0.00%
Use: Maintenance Fishing vessel 1.51% 2.82% 0.35% 1.53%
Engine 0.07% 0.24% 0.14% 0.35%
Fishing gear 0.30% 0.12% 0.03% 10.39%
End of life Fishing vessel 2.69% 5.01% 0.70% 2.79%
Fishing gear 0.35% 0.03% 0.01% 22.43%
Fish container 0.02% 0.11% 0.15% 0.54%
Fuel container 0.00% 0.01% 0.00% 0.04%
Lamps – – 0.00% 0.01%
 Sustainability Assessment of Small Scale Fishing Vessel Operations 445

3.3 Result and Interpretation

Figure 5 shows the environmental impact of each vessel presented in endpoint cate-
gories. The assessment result indicates a drawback to the environment, hence the lower
point is the better. It can be seen that most of the impacts are associated with human
health and resource use, followed by climate change and ecosystem quality. Further-
more, the result is in agreement with the inventory analysis, where PD and LF vessels
shows a better performance in terms of mass allocation. When using the economic
allocation, HL and LF vessels lead the chart by producing less impact than their
counterparts. This result confirms that the revenue from the HL vessel plays an
important role in improving the environmental performance.

Fig. 5. LCA result calculated per kg catch and £ revenue

Regarding life cycle stage, Table 3 summaries the contribution of each fishing
input to the environmental impacts. The table shows that supplies is generating the
most significant impact in all vessel ranging from 86% to 97% of the total impact and
this large percentage is mainly due to its substantial input, as seen in Table 2. These
facts suggest that the potential solution to minimizing the environmental impact is
related to reducing the amount of supplies.
The following major impacts are resulted from the maintenance of the wooden boat
used for the PD, TN and LF vessels. The impacts of maintenance stage is driven by
repainting and re-planking activities. In regard to the repainting work, concern is given
to the painting materials due to the chemical substances included. The growth of
marine fouling is detrimental because it damages the hull besides increasing the drag
and fuel consumption. Consequently, repainting work is required every 3 or 4 months
in order to clean the hull of fouling and to apply new protective layers, especially to the
underwater area. In this case, repainting frequency is another aspect influencing the
environmental performance. However, it should be noted that changing the repainting
mechanism is a crucial decision as it requires further consideration to optimize the
amount of material, types of material and the period of maintenance. Regarding re-
planking, the main issue is with regards to electricity consumption. Seeing as the
446 V. R. Kurniawati et al.

Table 3. Composition of LCA result at different life stages

Life stage Fishing input PD vessel TN vessel HL vessel LF vessel
Production Fishing vessel 0.48% 1.39% 1.31% 0.61%
Fishing gear 0.22% 0.52% 0.18% 7.29%
Fish container 0.08% 0.56% 0.77% 0.01%
Fuel container 0.01% 0.05% 0.01% 0.06%
Lamps – – 0.11% 0.52%
Use: Supplies Fuel 72.38% 52.45% 43.44% 82.90%
Lubricant 3.67% 0.00% 0.69% 3.36%
Ice 20.79% 38.97% 52.41% 0.00%
Use: Maintenance Fishing vessel 1.77% 5.06% 0.58% 1.93%
Engine 0.13% 0.53% 0.35% 0.83%
Fishing gear 0.25% 0.00% 0.00% 1.25%
End of life Fishing vessel 0.17% 0.46% 0.14% 0.24%
Fishing gear 0.04% 0.01% 0.00% 0.97%
Fish container 0.01% 0.00% 0.00% 0.03%
Fuel container 0.00% 0.00% 0.00% 0.00%
Lamps – – 0.00% 0.00%

background process for electricity used in the assessment is specifically for Indonesia,
the data is relevant to the case study. Hence, minimizing the electricity consumption
will elevate the sustainability performance in the context of the Indonesian fisheries.
Electricity also generates a comparable impact to the wood during the production of
the wooden vessels, hence the improvement plan in the production stage should be
focused on the electricity and wood substitution. Assessment result from the HL vessel
which is a fiberglass boat suggest that fiberglass substitution could be the potential
solution because the vessel requires low maintenance and consumes less electricity
during the production stage.
In the case of the LF vessel, the major issue is not only the wood but also the
fishing gear. The addition of 10 platforms has resulted in a significant impact which is
essentially caused by the use of plastic drums as the platform base. Therefore, plastic
drums are also a possible area where the environmental performance of the LF vessel
can be improved.
The further top five contributors regarding environmental impacts are the fish
containers which are found in the TN and HL vessels. The type of fish container used in
both vessels is an expanded polystyrene (EPS) cool box, which is replaced on a
monthly basis. Therefore, in order to improve environmental performance, the devel-
opment should focus on the cool boxes material to reduce the frequency of
replacement.
Table 3 also reveals that the impacts derived from end of life stage is insignificant
(<1%), which in fact, according to Table 2, this stage contributes a substantial amount
of fishing inputs. Study conducted in another fishing vessel shows that maintenance and
 Sustainability Assessment of Small Scale Fishing Vessel Operations 447

production stages generate more impact than end of life stage, despite the smaller
inputs [20].
The background process used to calculate the impact is obtained from a database
that represents the global average and it is not explicitly in the Asian or Indonesian
context. Accordingly, the result of this assessment deals with substantial uncertainty
which is not performed in this study due to the technical restriction. Furthermore, there
are also certain limitations that restrict the data collection process and cause to some
omissions. However, it is important to note that this study compares different fishing
vessels operations from the same region with the focus on providing an equal
assessment. Therefore, despite some existing limitations, it is reasonable to argue that
by using the consistent data inputs, this study is able to present a proportional and valid
comparison.

4 Life Cycle Cost

4.1 Assessment Method
The LCC is carried out by accumulating the costs to operate a vessel throughout its
lifetime [21] using the following steps. Firstly, the cost information including invest-
ment, recurring and residual is available at the present value. Since the calculation
involves the future expenses, the costs were projected into the future value using the
Eq. 1. LCC was presented in the current price, therefore the costs were subsequently
discounted back to the present value using the Eq. 2. Finally, those values were added
to obtain the total costs, as described in Eq. 3. In this study, all the cost was calculated
in Indonesian Rupiah (Rp), and for final presentation, it was converted into £ using the
fixed exchange rate (1£ = Rp16,555).

FV ¼ PV:ð1 þ inf Þn ð1Þ

FV
PV ¼ ð2Þ
ð1 þ DisÞn

LCC ¼ C0 þ PVrecurring þ PVresidual ð3Þ

Where:
PV = present value
FV = future value
inf = inflation rate, to be constant at 7.08%
dis = discount rate, to be constant at 4.53%
n = the number of years
C0 = initial investment
PVrecurring = present value of all repeating expenses, including supplies, mainte-
nance and personnel cost
PVresidual = present value of residual value at the end of the lifetime
448 V. R. Kurniawati et al.

The period of analysis is 20 years, this being the average of vessels’ lifetimes.
Consequently, all the costs are adjusted to maintain for 20 years of operation, and
include with some uncertainties. Therefore, the following assumptions were made:
1. Future revenue is predicted by assuming that the catch production decreases by 2%
per year, according to the decreasing trend showed in the statistic reports.
2. The number of fishing inputs, maintenance costs per task, fishing supplies per trip
and the number of fishers per vessel remained constant.
3. Price changes throughout the studied year were estimated using the inflation rate.

4.2 Inventory Analysis

In the LCC, inventory analysis lists the price of each fishing input throughout its
lifetime. All the price is based on the market price in 2015. The LCC result is the
summation of the inventory which consider the time value of money, hence the results
are presented in the same unit price. The details of inventory analysis for the LCC is
provided in the supplementary material, whilst the result is presented in the following
section.

4.3 Result and Interpretation

Figure 7 shows the total cost per kg catch and £ income. In terms of kg catch, the cost
contribution in each vessel is remarkably diverse. It can be seen that low catch quantity
in the TN and HL vessels ranks them as the costliest operations. However, little
difference is shown in terms of £ revenue, which indicates that in order to produce £1 of
revenue, the cost spent in each vessel is nearly the same, regardless of the fishing
methods.
Concerning the life cycle stage of the fishing vessels, Table 4 shows that the largest
expenses are established in the use stage (OPEX), specifically operational costs
incorporating supplies, personnel and selling costs. According to the economic per-
formance of worldwide SSFV, domination of fuel and labor expenditure is found in the
fishing operations in most places [22]. Supplies cost is mainly associated with the fuel,
whilst personnel cost is linked to the sharing system and catch per trip. Therefore, it can
be said that the enhancement in LCC can be carried out through a constructive
adjustment in catch per trip, fuel consumption, fuel price, and the sharing system. This
concerns are relevant with all vessels, as the cost is a crucial issue in the fishing
business.
It is then followed by the maintenance cost. The figure shows that the percentage of
maintenance cost in the TN vessel is considerably higher than other vessels, which is
primarily due to the fishing gear. The gear used in the TN vessel is changed every year,
because of physical damage. The gear in the HL vessel also requires frequent
replacement, however, seeing that it is inexpensive, the maintenance cost is not sig-
nificantly affected.
Negative percentage shown in the table indicates the residual value of the
respective items. Residual value is gathered from the engine, vessel and gear prices
when sold on the second-hand market. It can be seen that the percentage is extremely
 Sustainability Assessment of Small Scale Fishing Vessel Operations 449

Table 4. Composition of LCC result at different life stages

Life stage Fishing input PD vessel TN vessel HL vessel LF vessel
Production Fishing 0.27% 0.86% 0.86% 0.33%
vessel
Engine 1.21% 3.65% 2.13% 0.51%
Fishing gear 0.28% 0.97% 1.38% 6.56%
Fish 0.04% 1.43% 0.55% 0.44%
container
Fuel 0.01% 0.04% 0.01% 0.01%
container
Lamps – – 0.08% 0.36%
Use: Supplies and Supplies 37.57% 28.20% 25.28% 34.74%
operational Personnel 44.64% 47.49% 56.44% 34.53%
Selling 11.45% 0.00% 11.93% 12.81%
Use: Maintenance Fishing 1.64% 3.54% 0.36% 1.61%
vessel
Engine 0.17% 0.62% 0.41% 0.96%
Fishing gear 3.08% 14.16% 0.96% 8.30%
End of life Fishing −0.22% −0.57% −0.29% −0.33%
vessel
Engine −0.09% −0.39% −0.10% −0.06%
Fishing gear −0.04% 0.00% 0.00% −0.77%
Fish – – – –
container
Fuel – – – –
container
Lamps – – – –

small. Buying a used boats and engines is common in the region, even though they are
not always in excellent condition, as there will be buyers who are ready to make
modification or sell on specific parts that are reusable.
Fiberglass boats were introduced into the region in 2008, so all the boats are less
than 10 years old. In this case, the residual value was estimated based on the price from
other regions with a similar boat design and size. Compared to other wooden vessels,
the residual value of the fibreglass boat is lesser. It is only 25% of the new vessel’s
price, whilst for other vessels, it is more than 40% (see supplementary materials).
450 V. R. Kurniawati et al.

Fig. 6. LCC result calculated per kg catch and £ revenue

5 Sustainability Assessment

In this paper, the sustainability performance of each studied vessel is analyzed by

comparing the environmental and economic impacts of fishing operations. This means
the LCA and LCC results were analyzed separately to produce the performance score.
In order to provide an equal comparison between environmental and economic
impacts, the results were normalized using the ranking method and weighted using an
equal weighting system, as illustrated in Fig. 7. In the ranking process, the impacts
derived from each vessel were ordered from the best (1st place) to the worst (4th place).
Next, the rank was reversed to define its score, where score 4 is for 1st place and score 1
is for 4th place. Subsequently, each indicator in the same level was weighted equally.
This means the equal weighting factor is applied for indicators under the same cate-
gories. Total weight for each indicator is obtained by multiplying the weight in each
level. Finally, sustainability performance was scored by multiplying the point of each
indicator with its weight.
The total score for sustainability performance is presented in Fig. 8. Unlike the
impact assessment, the scores here refer to the quality of sustainability, hence the
higher is the better. Accordingly, it can be seen that the best performance in both
environmental and economic aspects is the LF vessel, followed by the HL vessel, the
PD vessel and the worse is the TN vessel.
As with the main characteristics of fishing vessel operations, it is argued that the
environmental performance of the passive fishing is better than active fishing, whilst in
terms of economic, both vessels show an equal performance. Furthermore, the com-
parison in Fig. 8 suggests that pelagic fishing is not necessarily better than ground
fishing, since the other factors such as fishing method and fish price have more
influence on the sustainability performance of fishing vessel operations.
 Sustainability Assessment of Small Scale Fishing Vessel Operations 451

Fig. 7. Sustainability assessment method

Even though the sustainability of the LF vessel ranks at the first place, it does not
mean that other fishing operations should be transformed into LF fishing. This is
because each vessel belongs to a specific community with certain skills and fishing
habits. Furthermore, no conflict appears regarding the use of diverse fishing operations
in addition to the fact that all the gear is legal fishing gear. Therefore, the formulation of
management measures should be concerned on the continuity and improvement of the
existing fishing practices. For example, the national report showed that an opportunity
to exploit the ground fish and crustacean group is wide open due to the sustainable
stock level [23]. Accordingly, the development of passive ground fishing operation
could be the best option available, as the active ground fishing is less sustainable in
both the environmental and economic aspects.
The result of sustainability assessment is highly dependent on the scope of analysis.
Expanding the analysis by adding the social performance might result in different score
and recommendation, hence, when the social performance analysis is included, recal-
culation should be conducted in order to present the sustainability assessment in more
comprehensive result.
452 V. R. Kurniawati et al.

Fig. 8. Sustainability performance in terms of environmental and economic aspect

6 Conclusion

This research demonstrates that the sustainability performance of fishing vessel oper-
ations can be investigated comprehensively through a life cycle approach which
combines environmental and economic aspects. Both LCA and LCC have been per-
formed and the result shows that the largest impact contributor is in the use stage due to
substantial amount of supplies. The sustainability performance was scored using the
reverse ranking and equal weighting system, which reveals that the LF vessel generates
the least environmental and economic impact, whilst the largest impact is derived from
the TN vessel. In general, passive fishing has better performance in comparison to
active fishing. Furthermore, ground fishing contributes higher impacts on environment,
however in terms of economic, this operation catches more valuable fish than its
counterpart. This result suggests the importance of trade-off when formulating man-
agement measures to promote the sustainable fishing vessel operations. Finally, the
assessment method applied in this study enables the decision makers to formulate
management measures which focus on the areas for potential improvement, hence the
development of sustainable fishing operations can be promoted effectively.
 Sustainability Assessment of Small Scale Fishing Vessel Operations 453

Acknowledgement. This research is sponsored by Directorate General of Higher Education

(DIKTI), Ministry of Research Technology and Higher Education – Indonesia and Indonesian
Endowment Fund for Education (LPDP), Ministry of Finance – Indonesia.

Supplementary Materials

The values presented in the table are for one-cycle period followed by the information
regarding its lifetime, hence further adjustment is required to calculate the total input
either for one-year period or 20 years period.

Inventory for LCA

a. PD vessel
Lifetime/ End of life
Operation Assembly Resources Unit Input
Frequency Incinerated Landfill Reuse
1) Production
Sawn wood kg 7,500
Hull
Reinforcing steel kg 220
Polyvinyl Acetate kg 5
White cement kg 7
Fishing vessel 20 years 30% 20% 50%
Painting Talc Powder kg 5
Alkyd paint kg 11
Antifouling kg 5
Electricity Electricity kWh 290
Engine Marine outboard engine Yamaha E40GMHL, 40 HP unit 1 7 years 100%
Polyamide (Nylon) kg 144
Net
Fishing ge ar Natural Dye litre 100 10 years 80% 20%
Sinker Lead kg 17.5
Fish container Plastic container Plastic container 200 litres kg 38 5 years 20% 80%
Fuel container Plastic container Plastic container 35 litres kg 4.5 5 years 20% 80%
2) Operation
Petrol Petrol kg 16,494
Supplies Lubricant Lubricating oil kg 655 1 year n/a
Ice Ice block kg 13,734
3) Maintenance
Sawn wood kg 750
Hull Reinforcing steel kg 22 Every 2 years
Electricity kWh 72
Polyvinyl Acetate kg 5
Fishing vessel White cement kg 7 50% 50%
Talc Powder kg 5
Repainting Every 3 months
Alkyd paint kg 6
Antifouling kg 5
Electricity kWh 3.6
Engine Oil changing Lubricating oil kg 0.445 Every week n/a
Polyamide (Nylon) kg 12 Every year
Net
Natural Dye litre 100 Every 3 weeks
Polyethylene, HDPE kg 27
Fishing gear Ropes Polyethylene, HDPE kg 3.4 Every 5 years 70% 30%
Polyethylene, HDPE kg 7
Bamboo kg 80 Every year
Floats
Plastic container 35 litres kg 1.5 Every 5 years
454 V. R. Kurniawati et al.

b. TN vessel
Lifetime/ End of life
Operation Assembly Resources Unit Input
Frequency Incinerated Landfill Reuse
1) Production
Sawn wood kg 4,500
Hull
Reinforcing steel kg 150
Polyvinyl Acetate kg 3
White cement kg 5
Fishing vessel 20 years 30% 20% 50%
Painting Talc Powder kg 3
Alkyd paint kg 10
Antifouling kg 4
Electricity Electricity kWh 220
Engine Marine diesel engine Dongfeng S1115, 24 HP pcs 1 3 years 100%
Net Polyamide (Nylon) kg 1
Fishing gear 10 years 1% 99%
Sinker Lead kg 36
Fish container EPS cool boxes Cool box EPS insulation board kg 1 1 month 30% 70%
Fuel container Plastic container Plastic container 35 litres kg 6 5 years 20% 80%
2) Operation
Diesel fuel Diesel kg 3,507
Supplies 1 year n/a
Ice Ice block kg 6,025
3) Maintenance
Sawn wood kg 450
Hull Reinforcing steel kg 15 Every 2 years
Electricity kWh 50
Polyvinyl Acetate kg 3
Fishing vessel White cement kg 5 50% 50%
Talc Powder kg 3
Repainting Every 3 months
Alkyd paint kg 4
Antifouling kg 2
Electricity kWh 3.6
Lubricating oil kg 4
Engine Oil changing Every 3 months n/a
Diesel kg 2.5
Net Polyamide (Nylon) kg 2 Every year
Polyethylene, HDPE kg 23.6
Rope Every 5 years
Fishing gear Polypropylene kg 20 100%
Synthetic rubber kg 1.4 Every year
Floats
Plastic container 35 litres kg 3 Every 5 years

c. HL vessel

Lifetime/ End of life

Operation Assembly Resources Unit Input
Frequency Incinerated Landfill Reuse
1) Production
Fibreglass kg 350
Hull Sawn wood kg 187.5
Fishing vessel Reinforcing steel kg 5 20 years 30% 70%
Painting Antifouling kg 3
Electricity Electricity kWh 40
Multipurpose e ngine Ya maha MZ175, 4.7 HP pcs 2 5 yea rs 100%
Engine
Ge nera tor Ya maha ET-1, 780 wa tts/220v pcs 1 5 yea rs 100%
Mainline Polyamide (Nylon) kg 1.6
Fishing gear Sinker Le ad kg 1 10 ye ars 80% 20%
Roller Polypropylene kg 1.6
Fish container EPS cool boxes Cool box EPS insulation board kg 1 1 month 30% 70%
Fuel container Plastic container Plastic container 10 litres kg 0.5 5 years 20% 80%
Lamps Compact fluorescent lamp Compact fluorescent lamp kg 0.75 4 years 100%
2) Operation
P e trol P e trol kg 1,684
Supplies Lubricant Lubricating oil kg 21 1 year n/a
Ice Ice block kg 5,900
3) Maintenance
Fibreglass kg 15
Sawn wood kg 56.25
Hull Every 3
Fishing vessel Reinforcing steel kg 0.5 30% 70%
years
Electricity kWh 7.2
Repainting Antifouling kg 3
Engine Oil changing Lubricating oil kg 0.89 Every month n/a
Branchline Polyamide (Nylon) kg 2.4 Every year
Fishing gear 100%
Hook and Swivel Steel, low-alloyed kg 0.09 Every year
 Sustainability Assessment of Small Scale Fishing Vessel Operations 455

d. LF vessel

Lifetime/ End of life

Operation Assembly Resources Unit Input
Frequency Incinerated Landfill Reuse
1) Production
Sawn wood kg 9,000
Hull
Reinforcing steel kg 250
Polyvinyl Acetate kg 7
White cement kg 9
Fishing vessel 20 years 30% 20% 50%
Painting Talc Powder kg 7
Alkyd paint kg 19
Antifouling kg 7
Electricity Electricity kWh 300
Engine Marinised diesel engine Mitsubishi 4D31, 100 HP pcs 1 10 years 100%
Bamboo kg 3,635
Bamboo platform 10 years 49.20% 10.80% 40%
Polypropylene kg 90
Polypropylene kg 50
Fishing gear Net Polyethylene, HDPE kg 15 3 years 50% 50%
Polypropylene kg 15
Generator set Sumura, 1000 watts/220v unit 1 5 years 100%
Platform base Plastic container 200 litres kg 136.80 3 years 20% 80%
Fish container Bamboo basket Bamboo kg 20 1 year 90% 10%
Fuel container Plastic container Plastic container 200 litres kg 7.60 5 years 20% 80%
Lamps Compact fluorescent lamp Compact fluorescent lamp kg 1.88 4 years 100%
2) Operation
Supplies for ferry Diesel fuel Diesel kg 4,319
Petrol Petrol kg 1,291 n/a n/a
Supplies for platform
Lubricant Lubricating oil kg 52
3) Maintenance
Sawn wood kg 900
Hull Reinforcing steel kg 27 Every 2 years
Electricity kWh 72
Polyvinyl Acetate kg 7
Fishing vessel White cement kg 9 50% 50%
Talc Powder kg 7 Every 4
Repainting
Alkyd paint kg 8 months
Antifouling kg 4
Electricity kWh 4
Engine Oil changing Lubricating oil kg 10.70 Every month n/a
Bamboo kg 910 Every 2.5
Fishing gear Bamboo platform 88.40% 11.60%
Polypropylene kg 50 years
456 V. R. Kurniawati et al.

Inventory for LCC

a. PD vessel

Cost components Input (£) Lifetime/Frequency

1) Investment
Fishing vessel 1,510 20 years
Engine 997 3 years
Fishing gear 755 10 years
Fish container 29 5 years
Fuel container 11 5 years
2) Use: Supplies and operational
Petrol 1,303
Ice 182
1 year
Personnel 2,501
Marketing -
3) Use: Maintenance
Fishing vessel
Hull 227 Every 2 years
Repainting 56 Every 3 months
Engine 9 Every 3 months
Fishing gear
Net 725 Every year
Float 16 Every 5 years
Rope 88 Every 5 years
4) End of life
Fishing vessel 604
Engine 30
 Sustainability Assessment of Small Scale Fishing Vessel Operations 457

b. TN vessel

Lifetime/
Cost components Input (£)
Frequency
1) Investment
Fishing vessel 1,510 20 years
Engine 997 3 years
Fishing gear 755 10 years
Fish container 29 5 years
Fuel container 11 5 years
2) Use: supplies and operational
Petrol 1,303
Ice 182
1 year
Personnel 2,501
Marketing -
3) Use: Maintenance
Fishing vessel
Hull 227 Every 2 years
Repainting 56 Every 3 months
Engine 9 Every 3 months
Fishing gear
Net 725 Every year
Float 16 Every 5 years
Rope 88 Every 5 years
4) End of life
Fishing vessel 604
Engine 30
458 V. R. Kurniawati et al.

c. HL vessel

Cost components Input (£) Lifetime/Frequency

1) Investment
Fishing vessel 1,208 20 years
Engine 586 5 years
Fishing gear 72 5 years
Fish container 2 1 month
Fuel container 2 5 years
Lamps 16 4 years
2) Use: supplies and operational
Petrol 892
Lubricant 43
Ice 178
1 year
Bait 214
Personnel 2,963
Marketing 626
3) Use: maintenance
Fishing vessel
Hull 46 Every 3 years
Repainting 12 Every 3 years
Engine 22 Every week
Fishing gear 50 Every week
4) End of life
Fishing vessel 302
Engine 27
 Sustainability Assessment of Small Scale Fishing Vessel Operations 459

d. LF vessel

Cost components Input (£) Lifetime/ Frequency

1) Investment
Fishing vessel 3,624 20 years
Engine 1,510 10 years
Fishing gear
Bamboo 906 10 years
Net 91 3 years
Generator 151 5 years
Plastic drums 190 3 years
Fish container 12 1 years
Fuel container 2 5 years
Lamps 40 4 years
2) Use: supplies and operational
Diesel 1605
Petrol/Platform 684 1 year
Lubricant/Platform 104
Personnel 9,430 Every year
Marketing 3,498 Every year
3) Use: maintenance
Fishing vessel
Hull 379 Every 2 years
Repainting 83 Every 4 months
Engine 22 Every month
Fishing gear 227 Every 2.5 year
4) End of life
Fishing vessel 1,812
Engine 151
Fishing gear 211
460 V. R. Kurniawati et al.

References
1. Gartside, D., Kirkegaard, I.: A History of Fishing, vol. IV. Eolss Publishers Company
Limited, Oxford (2009)
2. Food and Agriculture Organisation (FAO): FAO Statistic: Fisheries and Aquaculture
Statistic 2015. Food and Agriculture Organisation of the United Nations, Rome (2015)
3. Christensen, J., Tull, M.: Historical Perspectives of Fisheries Exploitation in the Indo-
Pacific, vol. 12, pp. 1–12. Springer, New York (2014)
4. Food and Agriculture Organisation (FAO): State of the World Fisheries and Aquaculture
2016. Food and Agriculture Organization of the United Nations, Rome (2016)
5. United Nations (UN): Report of the United Nations Conference on Environment and
Development, Chapter 17. United Nations, Rio de Janeiro (1992). http://www.un.org/depts/
los/consultative_process/documents/A21-Ch17.htm
6. Food and Agriculture Organisation (FAO): Code of Conduct for Responsible Fisheries,
p. 41. Food and Agriculture Organisation of the United Nations, Rome (1995)
7. Charles, A.T.: Sustainable Fishery Systems. Blackwell Science, Oxford (2001)
8. Bjordal: The use of technical measures in responsible fisheries. In: Cochrane, K.L. (ed.) A
Fishery Manager’s Guidebook: Management Measures and Their Application, p. 231. FAO,
Rome (2002)
9. Kloepffer, W.: Life cycle sustainability assessment of products. Int. J. Life Cycle Assess. 13
(2), 89–94 (2008)
10. Borges, L.M.M., Maulin, G.C., Andriguetto, J.M.: Analysis of income sources of fishers’
families on the coast of the state of Parana, Brazil. J. Coast. Res. III, 1267–1271 (2006)
11. Palabuhanratu fishing port (PPN Palabuhanratu): Buku Laporan Tahunan Statistik Perikanan
Tangkap Tahun 2015 (Capture fisheries statistics of 2015). Palabuhanratu Pelabuhan
Perikanan Nusantara (PPN) Palabuhanratu (2015)
12. Google: Google Maps. Google LLC (2018). www.google.com/maps
13. Azapagic, A.: Appendix: life cycle thinking and life cycle assessment (LCA). In: Sustainable
Development in Practice, pp. 426–437. Wiley (2005)
14. Jiang, Q., Liu, Z., Li, T., Zhang, H., Iqbal, A.: Life cycle assessment of a diesel engine based
on an integrated hybrid inventory analysis model. Procedia CIRP 15, 496–501 (2014)
15. Reenaas, M.: Solid oxide fuel cell combined with gas turbine versus diesel engine as
auxiliary power producing unit onboard a passenger ferry: a comparative life cycle
assessment and life cycle cost assessment. Norwegian University of Science and Technology
(2005)
16. Ramroth, L.: Comparison of Life-Cycle Analyses of Compact Fluorescent and Incandescent
Lamps Based on Rated Life of Compact Fluorescent Lamp. Rocky Mountain Institute,
Colorado (2008)
17. Nilsson, B.: SPINE LCI dataset: production of wood adhesive PVAC 3316. SPINE@CPM,
Sweden (2000)
18. Utne, I.B.: Life cycle cost (LCC) as a tool for improving sustainability in the Norwegian
fishing fleet. J. Cleaner Prod. 17(3), 335–344 (2009)
19. Goedkoop, M., Oele, M., Leijting, J., Ponsioen, T., Meijer, E.: Introduction to LCA with
SimaPro. Creative Commons (2016)
20. Fréon, P., Avadí, A., Vinatea Chavez, R.A., Iriarte Ahón, F.: Life cycle assessment of the
Peruvian industrial anchoveta fleet: boundary setting in life cycle inventory analyses of
complex and plural means of production. Int. J. Life Cycle Assess. 19(5), 1068–1086 (2014)
21. Davis, M., Coony, R., Gould, S., Daly, A.: Guidelines for Life Cycle Cost Analysis.
Stanford University, Stanford (2005)
 Sustainability Assessment of Small Scale Fishing Vessel Operations 461

22. Tietze, U., Thiele, W., Lasch, R., Thomsen, B., Rihan, D.: Economic performance and
fishing efficiency of marine capture fisheries. FAO Fisheries Technical Paper Food and
Agriculture Organization of the United Nations, vol. 482, p. 66 (2005)
23. Ministry of Marine Affairs and Fisheries (MMAF): Estimasi potensi, jumlah tangkapan yang
diperbolehkan, dan tingkat pemanfaatan sumberdaya ikan di wilayah pengelolaan perikanan
negara republik indonesia (Estimation of potential yield, total allowable catch, and status of
fisheries resource in indonesian fishing areas). Ministry of Marine Affairs and Fisheries
Republic of Indonesia, Jakarta (2017)
 Assessing Port Congestion Using Ship
Movement Data: A Case Study of Tianjin Port

Guanlan Liu(&), Alessio Tei, and Alan J. Murphy

School of Engineering, Newcastle University, Newcastle upon Tyne, UK

[email protected]

Abstract. In this paper, a full year (2012) of pilot and traffic data is used to
perform an analysis of the vessel traffic movements for Tianjin port. The results
will include the berth-to-berth vessel movements, occupancy time of berths as
well as the utilisation rate of terminals. This study has identified that berth
utilisation in Tianjin port is inconsistent, ranging from 1% occupancy to 88%
occupancy over the year. These results will be used to inform further research on
policy and planning alternatives to reduce emission burdens and optimise port
operations, therefore to improve the efficiency and safety of port operation.

Keywords: Port performance
Port characteristic
Real data heuristic study

Tianjin port

1 Introduction

1.1 Background
In recent years, the number of vessels has been increasing steadily, resulting in port
congestion and frequent maritime accidents in coastal areas [1]. In 2012, a report
produced by the National Bureau of Economic Research concluded that longer trans-
port times dramatically reduce trade. They estimated that cargo is held in transit each
day worth 0.6% to 2% of the value of the goods [2]. In a position paper produced by
National Customs Brokers & Forwarders Association of America in 2015, investiga-
tion found out that at least seven out of ten America’s busiest ports are facing con-
gestion regularly. Vessels have to wait in port for days to unload. Trucks have to wait
in line for hours to pick up one single container, and customers’ shipment delays
lasting weeks. These studies show that congestion has become a significant trade
barrier for both imports and exports, as well as a corresponding negative impact on
economy [3].
Another significant problem caused by port congestion is environmental impact.
For instance, the longer a vessel stays in port the more fuel it will consume, which leads
to higher emissions. Vessel emissions have been estimated in various studies, but there
is a limited number of studies which contain estimations on in-port emissions [4].
According to ITF calculations, shipping emissions in ports are substantial, accounting
for 18 million tons of CO2 emissions, 0.4 million tons of NOX , 0.2 million of SOX and
0.03 million tons of PM10 in 2011. Around 85% of these emissions come from con-
tainerships and tankers, due to their dominant presence in terms of port calls, relatively

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 462–477, 2021.
https://doi.org/10.1007/978-981-15-4680-8_33
 Assessing Port Congestion Using Ship Movement Data 463

long turnaround time in ports for tankers and relatively short stays but high emissions
during the stays for containerships. 58% of CO2 emissions in ports from shipping are in
Asia and Europe [4].
There are two fundamental causes of port congestion: the port has a problem of
adapting to an increase in demand, or inefficient management of the berth. Many
researchers have studied the first cause using different simulation analysis to assess the
port capacity, such as De Weille and Ray [5], Kozan [6], El-Naggar [7], Ke et al. [8],
Sheikholeslami et al. [9] and Saeed and Larsen [10]. All of these studies show similar
result: for a port which has less capacity than its needs, port modification will help to
ease the port congestion. The solutions include increasing berth number, increasing the
access channel depth and increasing the length of the quay. For port in which the
capacity is higher than its actual needs, studies have been carried out on analysing the
way ships approach the berth. These studies focused on two links: anchorage-ship-
berth link and shore-side link (an internal transport link for moving cargo in port) [11].
To measure the performance of the first link, Berth Allocation Problem has been
addressed to deal with the utilisation of the quay space [12]. Effective allocation is a
critical issue which causes great concern to port operators and shipping companies
[13]. Researcher Murty et al. [14] solved the dynamic BAP to find out the most
appropriate berthing time and positions of a set of containerships arriving at the ter-
minal within a specified planning horizon. In BAP studies, researchers usually assume
an exact arrival time, a fixed arrival interval or a berth can only handle on vessel at one
time. Few studies have addressed the problem of disruptions of vessel arrival (e.g.
weather condition, marine traffic congestion and delays at upstream ports [13]) or other
real-time possibilities. Hence this study proposed a real port data heuristic study to
understand port congestion better, and the details of the case study will be described in
the next section.

1.2 Study Area

As one of the world’s ten busiest ports, Tianjin port is also the largest port in Northern
China. Its location, Beijing-Tianjin-Hebei region, has the most severe air pollution in
China, according to a datasheet of regional emissions by the Asian Development Bank
[15]. This makes it the ideal port for a case study. In this paper, a full year (2012) of
pilot and traffic data is used to perform an analysis of the vessel traffic movements for
Tianjin port. The results will include the berth-to-berth vessel movements, occupancy
time of berths as well as the utilisation rate of terminals.
Located on the western shore of Bohai Bay, Tianjin port is known as the maritime
gateway to the Beijing-Tianjin-Hebei Region. In 2012, Tianjin port handled 476
million tonnes of cargo and 12.3 million TEU of containers [16], making it the world’s
fourth largest port by throughput tonnage, and the tenth in container throughput [17].
The geographical area studied in this paper is mainly defined by the Vessel Traffic
Service area delimited by the Tianjin Port Authority. Figure 1 shows the layout of
Tianjin port in 2016. The studied areas are Dongjiang Port Area, Beijing Port Area and
Nanjiang Port Area. The studied types of terminals are dry-bulk, container and tanker,
which are shown in Table 1.
464 G. Liu et al.

Fig. 1. The layout of the Tianjin port [18]

Table 1. Tianjin port terminal type

Terminal type Number of berths
Dry-bulk terminal 38
Container terminal 30
Tanker terminal 7

1.3 Dataset Information

This study focuses on assessing the Tianjin port performance based on a pilot station
recorded dataset. The dataset is a full year (2012) of traffic and pilot data in Tianjin
port, which contains 47625 trips. The dataset includes static and dynamic information,
which are shown in Table 2:

Table 2. The static and dynamic information in the dataset

Labels in dataset Meaning
Static information Pilot number No. of pilot
Name Ship name in Chinese
Nationality Ship nationality in Chinese
Length Ship length (m)
Width Ship width (m)
Start point The position where pilot embark
e.g. M/S - anchorage area, G1 - Berth G1
Action Instruction from pilot
e.g. Enter, Leave, Move
Destination The position where pilot disembark
e.g. M/S - anchorage area, G1 - Berth G1
Draught Ship draught (m)
(continued)
 Assessing Port Congestion Using Ship Movement Data 465

Table 2. (continued)
Labels in dataset Meaning
Dynamic information Agent plan time Estimate arrive time of the ship
e.g. 01/01/2012 15:01
Ship apply time Arrive time of the ship
e.g. 01/01/2012 15:01
Pilot embark time Pilot embark ship time
e.g. 01/01/2012 15:01
Pilot disembark time Pilot disembark ship time
e.g. 01/01/2012 15:01
Ship berthing time Ship berthing time
e.g. 01/01/2012 15:01

As indicated in IALA world VTS guide, all vessels must report their Estimated
Time Arrival (ETA) to the VTS centre at least 24 h before entry into the VTS area and
apply for authorisation [19]. Vessels shall also apply to the pilot station for pilotage
when navigating or shifting berth in port [19].
Figure 2 shows the stages a vessel has to go through to get in/out of the Tianjin
port. Notice that the dataset did not record the time vessels spend in anchorage area,
which means in this study waiting time will not be discussed. Vessels arrive at the pilot
station and wait for the pilot to embark. The next step is the berth for loading or
unloading, and the pilot disembarks after vessel berthing. When the vessel finished
work at berth, the pilot will embark again and disembark when the vessel leaves the
port. If the vessel has not finished the work but needs to move to follow the port
operator’s request, the vessel will move without the pilot. This process may repeat
several times before the vessel finished the work and leave the port.

Fig. 2. Vessel arrival process in Tianjin port

2 Method

Port performance can be defined in a variety of ways. A meaningful evaluation of a

port’s performance requires sets of measurements, relevant to the objectives of the
study. For example, these measurements was related to the quality of service to inland
466 G. Liu et al.

transport vehicles during their passage through the port, the quality of the cargo-
handling or the duration of a vessel’s stay in port [20]. Furthermore, in a number of
previous studies, researchers have used various units to represent the productivity of
berth, such as number of containers moved per hour [21], number of moves of a quay
crane per hour [22] and total tonnage to be discharged/loaded per vessel call [23].
Similarly, a port can be seen as offering a berth length for accommodating vessel
length for a period of time. In those terms the port is theoretically at full capacity if the
sum of the entire berth length offered is occupied for 100% of the time. Therefore in
this study, which is concerned with physically accommodating vessels for a period of
time and understanding the congestion caused by the physical presence of vessels in the
port, the value length*time of berth occupancy is defined as the measure of
performance.
For the port of Tianjin (the site of this study) the geographical extent of each berth
is pre-defined. Thus a number of usage scenarios, relevant to this study, could be
observed depending on the lengths of the vessels using the berths. These are defined as
follows:
• Berth entirely vacant – no vessel is in the berth
• Berth partially vacant – the sum of vessel(s) length is shorter than the offered berth
length
• Berth partially wasted – the vacant part of the offered berth cannot be used to serve
another vessel
• Ship overhang – the sum of vessel(s) length is longer than the offered berth length,
which indicates part of the vessel(s) length is in the adjacent berth
Table 3 shows the notations of the measures used in this study relevant to these
alternative scenarios.

Table 3. Measures of port performance

Measures Notation
The time a berth is entirely vacant Tv
The time a berth is partially vacant Tp
The time a berth is partially vacant but cannot be used – waste time Tw
The time a berth is shorter than the vessels length – overhanging time To
The duration of a berth start serving the first vessel till the last vessel in the whole Tt
year
Number of movements per vessel in berth during one visit m
The time a vessel spends in a berth Ts

The calculation formula of measures are as follows:

The time a vessel spends in a berth, Ts , expressed as:
 Assessing Port Congestion Using Ship Movement Data 467

8
<
ðTn berthing  Tn1berthing
 Þ  LOccupied ; Ship n enter berth
Ts ¼ Tn
berthing  Tn1 berthing  LOccupied ; Ship n move within berth ð1Þ
:
Tn embark  Tn1 berthing  LOccupied ; Ship n leave berth

When

Berth condition 6¼ 0 ð2Þ

min LShip \LOccupied \LBerth ð3Þ

The time a berth is partially vacant, Tp , expressed as:

8

< ðTn  Tn1berthing
Þ  LBerth  LOccupied
berthing  ; Ship n enter berth

 LOccupied ; Ship
Tp ¼ ðT
n berthing  Tn1berthing Þ  LBerth  n move within berth
:
Tn embark  Tn1 berthing  LBerth  LOccupied ; Ship n leave berth
ð4Þ

When

Berth condition ¼ 0 ð5Þ

The time a berth is entirely vacant, Tv , expressed as:

Tv ¼ ðTn berthing  Tn1 embark Þ  LBerth ð6Þ

When

Berth condition 6¼ 0 ð7Þ

LOccupied \min LShip ð8Þ

The time a berth is partially vacant but cannot be used – waste time, Tw , expressed
as:
8

< ðTn  Tn1berthing
Þ  LOccupied  LBerth
berthing  ; Ship n enter berth
Tw ¼ ðT
n berthing  Tn1berthing Þ  LOccupied

 LBerth ; Ship
 n move within berth
:
Tn embark  Tn1 berthing  LOccupied  LBerth ; Ship n leave berth
ð9Þ

When

Berth condition 6¼ 0 ð10Þ

LBerth \LOccupied ð11Þ

468 G. Liu et al.

The time a berth is shorter than the vessels length – overhanging time, To ,
expressed as:
8

<  ; Ship n enter berth
ðTn berthing  Tn1berthing
Þ  LOccupied  LBerth
To ¼ ðT
n berthing  Tn1berthing Þ  LOccupied

 LBerth ; Ship  n move within berth
:
Tn embark  Tn1 berthing  LOccupied  LBerth ; Ship n leave berth
ð12Þ

Where:
Tn berthing : Berthing time of No. n vessel (units: day), value taken from
database.
Tn1berthing : Berthing time of No. n−1 vessel (units: day), value taken from
database.
Tn embark : Pilot embark time of No. n vessel (units: day), value taken from
database.
LOccupied : Occupied length on berth (units: m), this is the sum of ship lengths
on a berth.
LBerth : Berth length (units: m)
Berth condition: Equals to zero when berth is empty, add one if vessel enter, minus
one if vessel leave, stay the same if vessel move within berth.
min LShip : The minimum length of vessel in the berth through the whole year

As a result, the berth occupancy can be calculated as Ts =Tt , which is the percentage
of time the berth is occupied by vessels. The percentage of partially vacant can be
calculated as Tp =Tt . The percentage of entirely vacant is Tv =Tt . The percentage of waste
time is Tw =Tt . Noticed that the sum of these four percentage is 1, it’s also a way to
validate the result.
In this paper, MATLAB software has been used to validate the dataset and simulate
the analytical model. The original dataset contains records of repeated trips or one way
trips due to human error, which need to be deleted before further analyse. The results
will be discussed in the next section.

3 Results and Discussion

3.1 Overall
For the Tianjin port in 2012, the total number of valid trips is 39,613. Figure 3 shows
the percentage of trips that are heading to, move within or leaving from container, dry-
bulk and tanker terminals.
Figure 4 shows the average percentage of berth occupancy, berth partially vacant,
berth entirely vacant and berth unable to use in different terminals. The results show
that the dry-bulk terminal is the most occupied and the container terminal is the least
occupied throughout the year. The average berth occupancy in the dry-bulk terminal is
 Assessing Port Congestion Using Ship Movement Data 469

Fig. 3. Percentage of trips for different terminal types

Fig. 4. The average percentage of berth usage in different terminals

56% and in container terminal is 32%. As mentioned before, the throughput of Tianjin
port in 2012 is 476 million tons of cargo and 12.3 million TEU of containers. The
statistics indicates that in 2012 Tianjin port has sufficient capacity to handle the
demand.

3.2 Berth Usage of Dry-Bulk Berths

Tianjin dry-bulk terminal has 38 berths, 16 berths handle general cargos, 16 berths
handle grain, coke, iron and coal cargos, and six berths handle general and dry bulk
cargos [24]. Figure 5 shows the berth occupancy and other performance measures in
dry-bulk berths. There are 25 out of 38 berths which berth occupancy percentage is
over 50% and 7 out of 38 berths which berth occupancy percentage is over 70%.
Results indicate in 2012 dry-bulk berths in Tianjin are busy, but still have the free
capacity to handle more cargos.
Figure 6 shows the berth overhang and number of vessel movement in berth and
Fig. 7 shows the layout of G1-G26 berths. The most occupied berth is G26, and this
berth also has the largest overhang. This is because its neighbour, G25, is not in use.
Hence vessels in G26 have more spaces and therefore no need to move. In contrast, for
G1-G5 berths, it can be seen from the map that they are neighbours and are all adjacent.
While G5, G3 and G1 are the second, third and sixth busiest berth handling general
cargos, their overhang is also significant. This compresses the available length of G2
470 G. Liu et al.

Fig. 5. Percentage of berth usage of dry-bulk terminal

and G4 berth. Nevertheless, the port operator moved vessels in G1-G3 several times to
achieve result. According to the Tianjin port website, No.1, No.2, No.4 and No.5
Stevedore Area are companies affiliated to Tianjin port holdings [25], which indicates
their berths can cooperate during the busy period.
In practice, once a berthing schedule is determined, the terminal manager will start
to assign resources to follow the requirements of the berthing vessels, including cargo
 Assessing Port Congestion Using Ship Movement Data 471

Fig. 6. Berth overhang and number of vessel movement of dry-bulk terminal

Fig. 7. Port layout of G1-G26 berths in Tianjin port [24]

472 G. Liu et al.

storage, facilities, labour and handling equipment. When disruptions occurred, for
example if the port operator wants the berth fully occupied, the manger must reallocate
the resources, which leads to additional costs onshore side and extra emission on ship
side. By optimising the berth allocation, there will be less vessel movement due to
berth reallocation, which means there will be fewer costs and emission in port. It can
also be seen in Fig. 7 that the entrance for G1-G26 berth is a long narrow waterway,
which means it is a waterway congested with vessels, and more vessels movement will
increase the probability of accidents. Therefore to reduce overhang and movements of
vessels, the port operator can increase the use of G10 and G11 berth, which are the
same type of berth as G1, G3 and G5 berth, since there is more than 45% berth space
which is entirely vacant and which has not been compressed by other berths.
For berths which handle grain, coke, iron, ore and coal, results show that for berth S8
(28% partially vacant) and S10 (44% partially vacant), they only have 2% and 0% berth
space that cannot be used. This means port operator managed to occupy the berth fully.
However, from Fig. 6 and Fig. 8 it can be seen that, with a very small overhang from
their neighbour berths, the port operator moved vessels in berth S8 and S10 very often to
achieve less berth space waste, this resulted in unnecessary costs and emissions.

Fig. 8. Port layout of S5-S15 berths in Tianjin port [24]

As for berths which handle general and dry bulk cargo, berth G23 is nearly empty
throughout the whole year, while G22 is 72% occupied and with a large overhang. This
is due to the fact that G22 needs to serve vessels which are longer than the berth. Hence
port operator may consider combining G22 and G23.
 Assessing Port Congestion Using Ship Movement Data 473

3.3 Berth Usage of Container Berths

Figure 9 shows the berth occupancy and other performance measures of container
berths. 23 out of 30 berths in the container terminal have an occupancy rate which is
under 40%. Less than 7% of container berths are over 50% occupied, and only one
berth has 73% occupancy throughout the whole year. Results show that Berth G33 only
has 1% occupancy. Although its neighbour berth G32 has an overhang, another
neighbour berth G34 has no overhang. This indicates that G33 is nearly empty
throughout the year. Furthermore, it can be seen in Fig. 10 that the most occupied berth
G8 also has the largest overhang among container berths, while its neighbour berth G7
has 17% occupancy. Therefore the port operator can consider combining them to
reduce the number of vessel movements.

Fig. 9. Percentage of berth usage in tanker terminal

474 G. Liu et al.

Fig. 10. Berth overhang and number of vessel movement in tanker terminal

Compare to dry-bulk berth occupancy, the low occupancy in container berths may
be caused by other types of vessels taking too much space in the waterway, hence
decreasing the number of containerships which have access to the berth. Further study
needs to be carried out to find the reason.

3.4 Berth Usage of Tanker Berths

Figure 11 shows the berth occupancy and other performance measures of all the tanker
berths. There are 4 out of 7 berths which berth occupancy percentage is over 40%, and
the lowest occupancy percentage is 23%. The types of cargo each berth handles are
shown in Table 4. Figure 12 shows the berth overhang and number of vessel move-
ment in tanker terminal. The results indicate that for berth S1 and S30, the berth length
is longer than the vessels. Take berth S1 as an example, the 54% partially vacant rate,
zero waste berth space and zero overhang show that all the vessels the berth served are
smaller than the berth length. The reason for this could be that there are no big tanker
ships that arrive at the berth, or due to waterway restrictions there is no access for big
tanker ships to reach the berth.
 Assessing Port Congestion Using Ship Movement Data 475

Fig. 11. Percentage of berth usage in container terminal

Table 4. Cargo types of tanker berths

Berth Cargo type
N1, N2 Ro-Ro, oil products, passengers, oil product
S1 Crude oil, dirty products, clean products and aviation fuel
S2 Clean products, chemical gases, aviation fuel and dirty products
S3 Dirty products, clean products and LPG
S4 Chemicals and petroleum products
S30 Crude oil and clean products

Fig. 12. Berth overhang and number of vessel movement in container terminal

This study analyses the port operation based on pilot station recorded traffic data
(2012) for the Tianjin port, one of the busiest ports in China. The major findings of this
study are summarised as follows:
Dry-bulk vessels contributed about 56% of the total trips. More than half of the dry-
bulk berths’ berth occupancy percentages are over 50%, which indicates a busy and
high utilisation rate dry-bulk terminal. Results also show that in order to achieve a high
utilisation rate the port authority moved the vessels around within the berth multiple
times. This will lead to additional costs, extra vessel emissions and a higher probability
of accidents in port.
476 G. Liu et al.

Container vessels contributed about 37% of the total trips. About 47% of berths had
a berth occupancy which was below 50%, and there was one berth which had 1% berth
occupancy.
Tanker vessels contributed about 7% of the total trips. There are 57% berths which
berth occupancy percentage was over 40%, and the lowest occupancy percentage was
23%. Results also show that the tanker terminal’s capacity is bigger than needed in
2012.

4 Conclusion

To the best of our knowledge, very few studies have linked the berth allocation
problem with real port data, the previous studies are mainly focusing on fixed vessel
arrival time or one vessel in a berth at one time. In this work, we assessed the existing
berth allocation scheme in Tianjin port by analysing the pilot station dataset. Results
also indicate that in practice, the port operator tends to put as many vessels as they can
in some berths to reach a high utilisation rate. The consequence of the operation
scheme is to increase the number of vessel movements and, in turn, the probability of
accident in port. Since there is alternative berth available to handle the vessels, the port
operator can consider increasing the usage of dry-bulk berth G10 and G11, container
berth G33 and G34. For tanker terminals, there is no evidence to show the berth
working hours, in this study we assumed the berth works based on a 24-h scheme. The
results indicate the tanker berths can serve bigger ships in Berth S1 and the tanker
terminal can handle more vessels.
In the future, more work needs to be done to estimate vessel emissions in port and
tanker terminal shore emissions. Another natural extension of the work done so far is to
discover the cause of current berth allocation problem by studying shore crane, land
traffic and storage area plans in port. It is also essential to find the balance between
increasing berth utilisation rate and maintain waterway safety in port. The data analysis
algorithm in this research can be applied to quickly identify port performance. The
result can also be used to benefit both the port operator and ship company.

References
1. Gao, X., Makino, H.: Analysis of anchoring ships around coastal industrial complex in a
natural disaster. J. Loss Prevent. Process Ind. 50(1), 355–363 (2016)
2. Georg, H., David, S.: Time as A Trade Barrier (2012)
3. NCBFAA Port Congestion: An Economic Threat page. http://www.ncbfaa.org/Scripts/
4Disapi.dll/4DCGI/cms/review.html?Action=CMS_Document&DocID=18042&MenuKey=
about. Accessed 14 Mar 2019
4. Merk, O.: Shipping Emissions in Ports. The international Transport Forum’s Discussion
paper. 2037 (2014)
5. De Weille, J., Ray, A.: The optimum port capacity. J. Transp. Econ. Pol. 8(3), 244–259
(1974)
6. Kozan, E.: Analysis of the economic-effects of alternative investment decisions for seaport
systems. Transp. Plann. Technol. 18(3), 239–248 (1994)
 Assessing Port Congestion Using Ship Movement Data 477

7. El-Naggar, M.E.: Application of queuing theory to the container terminal at Alexandria

seaport. J. Soil Sci. Environ. Manag. 1(4), 77–85 (2010)
8. Ke, G.Y., Li, K.W., Hipel, K.W.: An integrated multiple criteria preference ranking
approach to the Canadian west coast port congestion conflict. Expert Syst. Appl. 39(10),
9181–9190 (2012)
9. Sheikholeslami, A., Ilati, G., Yeganeh, Y.: Practical solutions for reducing container ships
waiting times at ports using simulation model. J. Marine Sci. Appl. 12(4), 434–444 (2013)
10. Saeed, N., Larsen, O.I.: Application of queuing methodology to analyze congestion: A case
study of the Manila International Container Terminal, Philippines. Case Stud. Transp. Pol. 4
(2), 143–149 (2016)
11. Dragović, B., Tzannatos, E., Park, N.K.: Simulation modelling in ports and container
terminals: literature overview and analysis by research field, application area and tool.
Flexible Serv. Manuf. J. 29(1), 4–34 (2017)
12. Meisel, F.: Container port planning, including berth and crane. In: Seaside Operations
Planning in Container Terminal. Contributions to Management Science. Physica-Verlag HD,
Heidelberg (2009)
13. Yan, S., Lu, C.C., Hsieh, J.H., Lin, H.C.: A dynamic and flexible berth allocation model with
stochastic vessel arrival times. In: Networks and Spatial Economics (2019)
14. Murty, K.G., Liu, J.Y., Wan, Y.W., Linn, R.: A decision support system for operations in a
container terminal. Decis. Support Syst. 39(3), 309–332 (2005)
15. Bank, A.D.: People’s Republic of China : Beijing – Tianjin – Hebei Regional Air Pollution
Control. October 2015
16. Ship Technology The world’s 10 biggest ports page. https://www.ship-technology.com/
features/feature-the-worlds-10-biggest-ports/. Accessed 14 Mar 2019
17. Shanghai International Shipping Institute. Global Port Development Report 2012 (2012)
18. Dongsheng, C., Yuehua, Z., Peter, N., Yue, L., Xiaotong, W., Ying, Z., Jianlei, L., Xiurui,
G.: Estimating ship emissions based on AIS data for port of Tianjin China. Atmos. Environ.
145, 10–18 (2016)
19. Tianjin P.R.China. IALA IAPH IMPA WORLD VTS GUIDE (2008)
20. Monie, G.De.: No. 6 Measuring and Evaluating Port Performance and Productivity (1987)
21. Briskorn, D., Drexl, A., Hartmann, S.: Inventory-based dispatching of automated guided
vehicles on container terminals. In: Kim, K.H., Günther, H.O. (eds.) Container Terminals
and Cargo Systems, pp. 195–214. Springer, Heidelberg (2007)
22. Meisel, F.: Seaside operations planning in container terminals. Physica-Verlag, Heidelberg
2009
23. Delovic, D., Mitrovic, D.M.: Some considerations on berth productivity referred on dry bulk
cargoes in a multipurpose seaport. Tehnicki Vjesnik-Technical Gazette 24(s2), 511–519
(2017)
24. Lloyd’s Register-Fairplay Ltd & IHS Fairplay. Ports & terminals guide 2003
25. Tianjin Port Holdings CO., LTD Company introduction page. http://www.tianjin-port.com/
port/Show.asp?m=1&d=931. Accessed 15 Mar 2019
 Numerical Simulation of Hybrid Platform
Supply Vessel (PSV) Fuel Consumption
for the Pre-Salt Layer in Brazil

Kazuo Nishimoto1(&), Claudio M. P. Sampaio1, Rodrigo J. Vale1,

and Felipe Ruggeri2
1
Dept Engenharia Naval e Oceânica,
University of São Paulo (USP), São Paulo, Brazil
[email protected]
2
Argonáutica Engenharia e Pesquisas, São Paulo, Brazil

Abstract. The new regulations regarding ship emissions based on IMO regu-
lations combined to the high fuel prices require the application of new tech-
nologies to improve ship efficiency and to reduce the CO2, NOX, SOX and BC
emissions. The application of LNG as fuel and battery packs to improve the
conventional engines performance are already a reality in offshore supply ves-
sels (OSV) around the world. However, the reduction in OPEX and payback
time of the additional investment is very dependent of the ship operational
profile, which are related to the FPSOs and port facilities particularities in
Brazil. In order to estimate the advantages of these technologies in the PSV
overall performance several different simulators regarding DP operation, sailing,
waiting and anchored mode are combined to obtain a digital twin of the vessel.
These simulators are combined to the real data monitored from the ships
regarding ship speed, position and heading collected in Brazil based on
AIS/AIS-Sat database. This database is combined to the environmental condi-
tions regarding wave, current and wind obtained based on global numerical
models to provide the environmental loads acting in the ship during the different
stages of the operation. The DP simulations are performed applying a thrust
allocation algorithm, the calm water resistance is obtained based on regression
models and the added resistance due to waves are computed based on strip
theory. The forces are converted into electric load by considering the propeller,
generators, switchboard, electric drive and engine efficiency curves combined to
a power management system (PMS) algorithm, which balance the loads in the
engines according to the operational mode. The baseline for comparison is the
current PSV without dual-fuel engines or battery pack.

Keywords: Platform Supply Vessel
Dual fuel
Ship efficiency

1 Introduction

The pre-salt support operations are becoming more and more frequent in Brazil due to
the exploration of the Pre-salt layer oil reservoir. In order to reduce the ship emissions
and fuel consumption a basic comprehension of the operational profile of these vessels,

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 478–497, 2021.
https://doi.org/10.1007/978-981-15-4680-8_34
 Numerical Simulation of Hybrid Platform Supply Vessel (PSV) 479

as the efficiency and possible modifications of the operational profile is required. This
task may be performed using a numerical simulator to reproduce the vessel operation.
The studied ship was defined as a PSV 4500 with the main characteristics shown in
Table 1, which was established based on similar vessels operating in the pre-salt of
Brazil. The selected vessel has 4 main generators that provides energy to two bow
tunnel thrusters and one retractable azimuthal thruster in bow that provide control in the
fore region while there are two azimuthal thruster located in the stern region used for
both propulsion and station keeping.

Table 1. Ship properties and power characteristics of PSV4500.

Dimension Value Equipment Power
LOA 86.8 m Main generator 6809 kW
Beam 18.8 m (4 x Gen) 1702 kW
Depth 7.96 m Main prop power 4319 kW
Draft 6.52 m 2 x Azimuth Nozzle 2159 kW
DWT 7500 ton Bow thruster 2275 kW
Speed 14.2 kn (2 x Tunnel + 813 kW
Displacement 7475 ton + 1 retractable) 650 kW

2 Environmental Conditions

2.1 Wave Conditions

The wave conditions were post-processed based on the several results from Wave
Watch III (WW3) model run directly by NOAA (National Oceanic and Atmospheric
Administration) and provided through a specific FTP. The numerical results are
available in several points in the Brazilian coast (Fig. 1) but the most important ones
are located near Santos and Campos Basins, since there are a large number of oil
platforms in these regions. Besides, Açu port (in white) was selected as base port for
PSV simulations because its very near location to a large cluster of oil platforms.

Fig. 1. Points provided by NOAA environmental conditions, buoys (left) and oil platforms
located in Santos and Campos Basin (yellow) and Açu port (white) (right).
480 K. Nishimoto et al.

The results regarding the wave conditions were download during the period
between August 1st, 2015 to January 1st, 2016 considering the simulated sea state with
a time sample of 1 h, providing 12473 records. For each sea state partition the sig-
nificant wave height is provided, as the peak period, the peak wave direction and the
directional spread. For each partition the numerical model also provides the fraction of
this wave component that is generated locally (sea) or propagated from other locations
(swell).
The several files were post-processed providing the wave record for Campos Basin
shown in Fig. 2 and Fig. 3, regarding the significant wave height and peak period. In
order to provide a comprehension between these wave conditions the results are pre-
sented for sea and swell conditions considering the two most energetic components
among the several partitions.

3.5
Sea
Swell
3

2.5
Hs(m)

1.5

0.5

0
Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb
2015/2016

Fig. 2. Time series of significant wave height in the period between August 1st, 2015 and
January 1st, 2016 for the Campos Basin point.

It can be observed significant wave heights over 3.5 m and peak periods up to 18 s,
which may induce large vessel motions during the operations. Since the Campos Basin
is located very far from the coast, the wind fetch is almost unlimited, therefore the
locally generated waves (sea) may reach large significant wave heights and peak
periods up to 12 s, which is uncommon compared to the sheltered locations (ex: ports
and harbors).
It can be verified that sea conditions generated by the local winds are provided most
from the northeast direction with peak periods usually below 7 s, although in almost
2% reaches 8 s. The significant wave height generated by the local wind reaches up to
2.0 m most of the times but for about 5% of the time the wave height reaches 2.5 m
and for less than 1%, 4.0 m height. On the other hand, the swell wave conditions reach
the Campos Basin from the east and south directions, the first one with more occur-
rences but the second one with more intensity in terms of significant wave height. The
waves coming from the east direction usually reach up to 2.0 m–2.5 m with peak
periods below 10 s while the waves from the south reach up to 3.5 m and 16 s, which
are long wave conditions (*400 m wavelength).
 Numerical Simulation of Hybrid Platform Supply Vessel (PSV) 481

18

16
Sea
Swell
14

Tp(s) 12

10

2
Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb
2015/2016

Fig. 3. Time series of peak period in the period between August 1st, 2015 and January 1st, 2016
for the Campos Basin point.

2.2 Wind Conditions

The wind conditions obtained from NOAA data for Campos Basin are shown, in terms
of average intensity and direction, in Fig. 4 where wind speeds up to 30 knots can be
verified, although the average speed during the year is about 15 knots.

40

30
Vw (knots)

20

10

0
Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb

400

300

200

100

0
Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb
2015/2016

Fig. 4. Time series of wind speed and direction between August 1st, 2015 and January 1st, 2016
for the Campos Basin point.

2.3 Current Conditions

Since the current conditions were not available in NOAA data, these data were
obtained from HYCOM project (HYCOM 2017). The data was extracted from the
webservice available in the project website and collected considering all water depth
layers from the sea bottom to the sea surface. However, since for the simulation only
the current acting on the ship’s hull is of interest, the mean values of the current were
computed between the mean sea surface and 6.0 m water depth (ship’s draft).
482 K. Nishimoto et al.

The time series of current intensity and direction is illustrated in Fig. 5, where the
current direction is conventionally defined as going to. It can be verified that current
intensity reaches up to 1.5 knots in extreme conditions, although most of the time the
values are below 0.8 knots. It can also be observed that the current direction is widely
spread in the south sector with predominant direction as southwest.

1.5

1
Vc (knots)

0.5

0
2014 2015 2016 2017 2018

400

300

200

100

-100
2014 2015 2016 2017 2018

Fig. 5. Time series of current speed and direction between August 1st, 2015 and July 1st, 2017
for the Campos Basin point.

3 Mathematical Models

Since the PSV operational activities change appreciably, several different models were
combined to provide an integrated simulator. The DP model computation is based on a
quasi-static approach where the total thrust required in the fore and stern regions are
computed in terms of time average loads that is, although an entire year of environ-
mental records on each set of wind, current and sea state conditions were used, only the
average thrust was computed. In order to control the degrees of freedom in the plane
(surge, sway and yaw), the environmental loads considered into the analysis are wind,
current, wave mean drift and wave-drift damping loads.
The wind and current loads are computed based on values presented in the literature
(Fossen and Perez 2009) and/or based on CFD (Computational Fluid Dynamics)
simulations. Other additional sources that could be used are the experimental results
published by (Blendermann 1994).
The wave mean drift are computed using the potential flow solver WAMIT® while
the wave drift damping loads are obtained based on the approximation proposed by
Aranha (Aranha 1994). Since most of these operations are performed in close prox-
imity to a FPSO (Floating Production Storage & Offloading Unit), the correct approach
would be to take into account the shadow effects imposed by the unit in terms of wind,
current and waves. However, in this simplified version these interactions was not
considered.
The several models are integrated to estimate the required thrust that must be
provided by the propulsion system in order to obtain the pre-established performance.
 Numerical Simulation of Hybrid Platform Supply Vessel (PSV) 483

It shall be highlighted that in the simulations the ship’s behavior is known through the
track records, allowing the characterization of engine loads for the computation of the
desired performance.

3.1 Conversion from Required Thrust into Engine Load and Fuel
Consumption
The diesel-electric system is modeled considering engines, generators, propellers,
switchboards, transformers and electric drives according to available manufacturers’
specification and guidelines. The modeled arrangement is depicted in Fig. 6 with its
various components, the electric system distributed into two switchboards that can be
connect by a tie.

Fig. 6. PSV diesel-electric configuration.

Fig. 7. Blocks diagram of the simulator.

484 K. Nishimoto et al.

The simulator’s diagram is sketched in Fig. 7, where it can be observed that the
vessel speed and the environmental conditions are the input information for the whole
computation.
The generators efficiency curve is based on manufacturers’ data, is defined as the
ratio between power-out and power-in (Vasques (2014)) and it is a function of power-
in, as shown in Fig. 8. It can be seen that the curve is almost flat for power levels above
1000 kW with an efficiency around 97%. On the other side, the generator efficiency
reduces appreciably for power levels below 400 kW achieving 87% for powers below
50 kW.

0.98

0.96

DF Engine 1 DF Engine 2 DF Engine 3 DF Engine 4

? KW ? KW ? KW ? KW

0.94 Battery 1
? kWh
720 rpm 720 rpm 720 rpm 720 rpm
Battery 2
? kWh
Diesel Engine
? KW
? C-rate ? C-rate 1800 rpm
=Pout/Pin

Gen 1 Gen 2 Gen 3 Gen 4 Harb Gen

1702 KW 1702 KW 1702 KW 1702 KW 500 KW
720 rpm 720 rpm 720 rpm 720 rpm 1800 rpm
690V / 60Hz 690V / 60Hz 690V / 60Hz 690V / 60Hz 690V / 60Hz

0.92
Bus 1 – 690V / 60 Hz Bus 2 – 690V / 60 Hz

690V / 480V / 690V / 480V /

220V 220V
690V / 690V 690V / 690V 690V / 690V 690V / 690V 690V / 690V

0.9

Bow Tunnel Thr. Retractable Thr. Bow Tunnel Thr.

813 KW Azimuth Thr. 650 KW Azimuth Thr.
813 KW
0 – 800 rpm 2159 KW 0 – 800 rpm 2159 KW
0 – 800 rpm
0.88 0 – 1200 rpm 0 – 1200 rpm

0.86
0 200 400 600 800 1000 1200 1400 1600 1800
Power in (kW)

Fig. 8. Generators efficiency curves.

The switchboard and transformer efficiencies are also defined as ratios of power-out
and power-in, however, as efficiency variation is low, these efficiencies were assumed
constant at 99.5% and 98.5%, respectively. The electric drive efficiency curve has
similar characteristics as the other devices as shown in Fig. 9. Based on manufacturer’s
catalogue data, its efficiency is function of rotation speed over rated speed. It can be
observed that its maximum efficiency is about 90% for rates above 60%, on the other
hand, for low rates, the efficiency rapidly falls, an aspect that should be carefully
evaluated.
The rotation is computed based on propeller-hull integration assuming modified B-
Troost curves that include the nozzle effect (Carlton 1994), this effect increasing the
bollard pull thrust at one hand, at the other reducing the propeller efficiency for high
advance coefficients. The KT and KQ coefficients and the propeller efficiency (η) are
defined, respectively, by Eqs. (1), (2) and (3), where T is the propeller thrust; Q,
propeller torque; q, salt water density; D, propeller diameter and J, the advance
coefficient.

T
KT ¼ ð1Þ
qn2 D4
 Numerical Simulation of Hybrid Platform Supply Vessel (PSV) 485

100

90

80

=Pout/Pin (%) 70 DF Engine 1

? KW
DF Engine 2
? KW
DF Engine 3
? KW
DF Engine 4
? KW
720 rpm 720 rpm 720 rpm 720 rpm
Battery 1 Battery 2 Diesel Engine
? kWh ? kWh ? KW
? C-rate ? C-rate 1800 rpm

60 Gen 1 Gen 2 Gen 3 Gen 4 Harb Gen

1702 KW 1702 KW 1702 KW 1702 KW 500 KW
720 rpm 720 rpm 720 rpm 720 rpm 1800 rpm
690V / 60Hz 690V / 60Hz 690V / 60Hz 690V / 60Hz 690V / 60Hz

50
Bus 1 – 690V / 60 Hz Bus 2 – 690V / 60 Hz

690V / 480V / 690V / 480V /

40 220V
690V / 690V 690V / 690V 690V / 690V 690V / 690V 690V / 690V
220V

30
Bow Tunnel Thr. Retractable Thr. Bow Tunnel Thr.
20 813 KW
0 – 800 rpm
Azimuth Thr.
2159 KW
0 – 1200 rpm
650 KW
0 – 800 rpm
Azimuth Thr.
2159 KW
0 – 1200 rpm
813 KW
0 – 800 rpm

10
0 10 20 30 40 50 60 70 80 90 100
speed n/nrated (%)

Fig. 9. Electric drive efficiency as function of the rate of speed rate per rated speed.

Q
KQ ¼ ð2Þ
qn2 D5

J KT
g¼ ð3Þ
2p KQ

The propellers characteristics were obtained from similar ships and manufacturers
general specifications (Rolls Royce 2014) and are summarized in Table 2. It should be
remarked that the azimuthal propellers have controllable pitch, thus the efficiency can
be optimized according to the required thrust and rotation.

Table 2. Propellers properties (Blade number = 4 and Ae/A0 = 0.55).

Propeller Bollard pull (kN) Diameter (m) Power (kW) Maximum rotation (rpm)
Stern azimuth 1 450 2.8 2159 1200
Stern azimuth 2 450 2.8 2159 1200
Bow tunnel 1 150 2.0 813 800
Bow tunnel 2 150 2.0 813 800
Bow azimuth 100 2.0 600 600

An example of conventional propeller diagram (KT, KQ and η curves) with nozzle is

shown in Fig. 10 for different pitch/diameter (P/D) ratios. However, from the simu-
lation point of view, it is more convenient to obtain a direct relation between propeller
thrust and torque, as torque can be converted to required propeller power by Eq. (4),
with the ratio between torque and thrust given by expression (5).

P ¼ 2pnQ ð4Þ
486 K. Nishimoto et al.

1.4
K (P/D=0.6)
T
10K (P/D=0.6) 1.2
Q
(P/D=0.6)
K (P/D=0.8)
T
1
10K (P/D=0.8)
Q
(P/D=0.8)
K (P/D=1) 0.8
T

K ,K ,
Q
10K (P/D=1)
Q

T
(P/D=1) 0.6
K (P/D=1.2)
T
10K (P/D=1.2)
Q 0.4
(P/D=1.2)
K (P/D=1.4)
T
10K (P/D=1.4) 0.2
Q
(P/D=1.4)
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
J=V/(nD)

Fig. 10. KT, KQ and η curves for PSV propeller.

KQ Q
¼ ð5Þ
KT TD

The fuel consumption is obtained based on manufacturers’ specific fuel oil con-
sumption (SFOC) of the power output curve of the engine, as depicted in Fig. 11 for
diesel and gas engine. Besides, in natural gas mode, dual-fuel engines require a pilot
diesel fuel for the beginning of the combustion process, this portion also included in the
graph. It can be observed that the SFOC is appreciably higher for loads below
1000 kW.

320
DF Engine 1
? KW
DF Engine 2 DF Engine 3 DF Engine 4
Diesel
? KW ? KW ? KW

Battery 1
? kWh
720 rpm 720 rpm 720 rpm 720 rpm
Battery 2
? kWh
Diesel Engine
? KW Gas
300 ? C-rate

Gen 1 Gen 2
? C-rate 1800 rpm

Gas+diesel (pilot)
Gen 3 Gen 4 Harb Gen
1702 KW 1702 KW 1702 KW 1702 KW 500 KW
720 rpm 720 rpm 720 rpm 720 rpm 1800 rpm
690V / 60Hz 690V / 60Hz 690V / 60Hz 690V / 60Hz 690V / 60Hz
Specific Fuel Cons. (g/kWh)

280 Bus 1 – 690V / 60 Hz Bus 2 – 690V / 60 Hz

690V / 480V / 690V / 480V /

220V 220V
690V / 690V 690V / 690V 690V / 690V 690V / 690V 690V / 690V

260
Bow Tunnel Thr. Retractable Thr. Bow Tunnel Thr.
813 KW Azimuth Thr. 650 KW Azimuth Thr.
813 KW
0 – 800 rpm 2159 KW 0 – 800 rpm 2159 KW
0 – 800 rpm

240
0 – 1200 rpm 0 – 1200 rpm

220

200

180

160
0 200 400 600 800 1000 1200 1400 1600 1800
Power out (kW)

Fig. 11. SFOC curves as function of the power-out.

 Numerical Simulation of Hybrid Platform Supply Vessel (PSV) 487

Since the PSV has more than a single engine, the SFOC of several engines shall be
combined to provide the minimum SFOC for the total required ship’s load. Since the
SFOC is higher at smaller loads for each individual engine, the total SFOC curve has
several jumps (Kjølleberg (2015)). In practice, there are several different ways to
distribute the engine loads to obtain the required power but, in order to provide a more
generic simulator, a non-linear optimization problem was solved where the fuel con-
sumption is minimized, as defined by Eq. (6), with SFOC changing according to the
engine load xe. Moreover, the total required power, PReq provided by the engines is
defined by constraint (7), each individual engine load limited to its maximum power,
given by constraint (8).
XNe
CT ¼ e¼1
SFOC ðxe Þ xe ð6Þ
XNe
x
e¼1 e
¼ PReq ð7Þ

xe  PMaxe ; e ¼ 1; 2;


; Ne ð8Þ

Table 3. Emission factors for different fuels.

Emissions (g/kg fuel)
Components Diesel Gas
CO2 3200 2700
CH4 0.3 31.2
NOX 60 13.7
SOX 10 0.6
BC 0.3 0.03
CO 10 10

Finally, the ship emissions are computed according to conversion factors presented
in Table 3 (Stenersen and Thonstad (2017), IPCC (2013), etc.), where they are rep-
resented as ratios of emission mass per fuel mass for CO2, CH4, NOX, SOX, black
carbon and CO constituents.

4 Environmental Loads

The interactions between vessel and wind, current and waves shall be computed for the
evaluation of the required thrust and later evaluation of the demanded electric power
and engine loads at the different operational conditions. It is important to highlight that
the main difference between DP and navigation modes is the forward speed, which is
appreciably higher for the last one.
488 K. Nishimoto et al.

4.1 Wind Loads

The wind loads are computed based on the coefficients presented by Fossen and Perez
(2009) with the longitudinal, transversal and moment loads computed based on
Eqs. (7), (8) and (9). In this equations AFrontal is the projected frontal area; ALateral, the
projected lateral area; CWX(hW), CWY(hW) and CWM(hW), the wind coefficients; hW, the
wind incidence angle (measured from the bow in counterclockwise direction); UW, the
wind speed and qair, the air density. The windage area is calculated based on the ship’s
profile, respectively, 550 m2 and 180 m2 for lateral and frontal areas. The wind load
coefficients are shown in Fig. 12.

1
x

0
C

-1
0 30 60 90 120 150 180 210 240 270 300 330 360
1
y

0
C

-1
0 30 60 90 120 150 180 210 240 270 300 330 360
0.2
m

0
C

-0.2
0 30 60 90 120 150 180 210 240 270 300 330 360
Angles (º)

Fig. 12. Wind coefficients: Top CWX (= Cx); middle, CWY (= Cy) and down, CWM (= Cm) –
(Fossen and Perez (2009)).

1
FWu ðhW Þ ¼ qair AFrontal CWX ðhW Þ UW
2
ð9Þ
2
1
FWv ðhW Þ ¼ qair ALateral CWY ðhW Þ UW
2
ð10Þ
2
1
MWz ðhW Þ ¼ qair AFrontal LPP CWM ðhW Þ UW
2
ð11Þ
2

4.2 Current Loads

The current loads are computed analogously to the wind ones by applying Eqs. (10),
(11) and (12), where UC is the current velocity; hC, the current incidence angle and
CCX(hC), CCY(hC) and CCM(hC), the current coefficients, as shown in Fig. 13.

1
FCu ðhC Þ ¼ qwater B T CCX ðhC Þ UC2 ð12Þ
2
 Numerical Simulation of Hybrid Platform Supply Vessel (PSV) 489

0.2

x
0

C
-0.2
0 30 60 90 120 150 180 210 240 270 300 330 360
1
y
0
C

-1
0 30 60 90 120 150 180 210 240 270 300 330 360
0.5
m

0
C

-0.5
0 30 60 90 120 150 180 210 240 270 300 330 360
Angles (º)

Fig. 13. Current coefficients: Top, CCX (= Cx); middle, CCY (= Cy) and down, CCM (= Cm) -
(Fossen and Perez (2009)).

1
FCv ðhC Þ ¼ qwater LPP T CCY ðhC Þ UC2 ð13Þ
2
1
MCz ðhC Þ ¼ qwater L2PP T CCM ðhC Þ UC2 ð14Þ
2

4.3 Mean Drift Forces

The mean drift forces are computed considering both sea and swell conditions and
combined as wave loads acting at the ship. The longitudinal and transversal forces and
yaw moment are computed using Eqs. (13), (14) and (15), with Sf(x, TP, HS) repre-

senting the standard Jonswap spectrum; Fxðx; 
hO Þ; Fvðx; 
hO Þ and Mzðx; hO Þ, the
RAOs (Response Amplitude Operators) of the longitudinal and transverse forces and
yaw moment and hO, the resulting incidence angle.
1
FOu ðhO Þ ¼ 2
Z
x ðx; hO Þjdx
Sf ðx; TP ; HS ÞjF ð15Þ
0

1  
Sf ðx; TP ; HS ÞFy ðx; hO Þdx
Z
FOv ðhO Þ ¼ 2 ð16Þ
0

1
MOz ðhO Þ ¼ 2
Z
 z ðx; hO Þjdx
Sf ðx; TP ; HS ÞjM ð17Þ
0

The RAOs are computed using the higher order method of the WAMIT® potential
flow solver taking into account for discretization angles between 0º (stern) and 180º
(bow), as the hull is symmetrical and there is no interaction with the FPSO. The mean
drift coefficients shown in Fig. 14.
490 K. Nishimoto et al.

20

10

0 0º
5 6 7 8 9 10 11 12 13 14 15
22.5º
40 45º
67.5º
20 90º
112.5º
0 135º
5 6 7 8 9 10 11 12 13 14 15 157.5º
400 180º

200

0
5 6 7 8 9 10 11 12 13 14 15
Period (s)

Fig. 14. Mean drift force coefficients computed using WAMIT Potential Flow code.

4.4 Wave Drift-Damping Forces

The wave drift damping forces are computed through Eqs. (16), (17) and (18) based on
Aranha’s approximation (Aranha (1994)).
  
x x
@F 
x coshO  2 @ Fx sin hO dx ð18Þ
1
Z
FWDDu ðhO Þ ¼ 2 Sf ðx; TP ; HS Þ x þ 4F
0
g @x @hO
  
1 x y
@F y
@F
FWDDv ðhO Þ ¼ 2
Z
Sf ðx; TP ; HS Þ x 
þ 4Fy coshO  2 sin hO dx ð19Þ
0
g @x @hO
  
x z
@M 
 z coshO  2 @ Mz sin hO dx ð20Þ
1
Z
MWDDz ðhO Þ ¼ 2 Sf ðx; TP ; HS Þ x þ 4M
0
g @x @hO

5 PSV Monitored Data

The AIS/AIS-Sat data were obtained for four different PSVs, however, as the vessel
Bram Bahia had more information available (10,279 records), it was selected for the
study. It should be noted that the ship’s data information is not equally distributed in
time therefore the data has to be resampled in the simulations, besides the removal of
incorrect/inconsistent data. The monitored positions of the vessels are illustrated in
Fig. 15, all vessels operating near the Brazilian coastline in Campos and Santos Basins.
The breakdown of ships’ speed is a key aspect to identify the maximum expected
efficiency improvement therefore the data was post-processed and segregated in six
different speed ranges, as shown in Fig. 16. The data records show that the vessels’
speed profile corresponds to low speed (<3 knots), which may be explained by the large
time spent on anchorage, waiting to begin DP operations and in DP mode. This fact is
particularly interesting for the study of hybrid power generation since diesel engines
have the worst fuel consumption performance when working far from the rated speed.
 Numerical Simulation of Hybrid Platform Supply Vessel (PSV) 491

°
°
45.0 W 42.5° W 40.0° W 37.5° W
47.5 W 35.0° W

BRAM BAHIA
BRAM BRASIL
BRAM BRAVO
BRAM BUCCANER

°
20.0 S

°
22.5 S

°
25.0 S

Fig. 15. Tracks of the monitored vessels.

90
BRAM BAHIA(2017-06-01 14:00-2017-08-30 12:24)
80 BRAM BRASIL(2017-06-01 16:36-2017-08-30 13:46)
BRAM BRAVO(2017-06-01 15:57-2017-08-30 13:57)
70 BRAM BUCCANER(2017-06-01 16:59-2017-08-30 16:00)

60
Occurrence (%)

50

40

30

20

10

0
0-3 3-6 6-9 9-12 12-15
Velocity (knots)

Fig. 16. Velocity distribution for the monitored vessels.

The vessel operational profile was split into five different conditions and for each
one a specific approach was applied to provide the required thrust and therefore obtain
the generator set loads and consumption. The characteristics of the five different states
are:
1) Waiting for DP operation: Closer than 750 m of an oil platform, ship speed above
1,0 knot and no DP2 requirement;
2) Operating in DP mode: Quasi-static balance model at close proximity to the plat-
form, with speed below 1.0 knot and DP2 requirement;
3) Sailing – Resistance assessment for speeds higher than 1.0 knots, estimated by
Holtrop formulation, for calm water conditions and strip theory, for added
resistance;
4) Anchored at port – Emergency generator plus one single engine;
5) Cargo Loading/Unloading – Only the emergency generator.
492 K. Nishimoto et al.

t=t+dt

No
t=tini t<tend END

Yes

Yes Sailing
Vship>1.0
mode

No

Lat/Lon at Yes Anchor

anchorage? mode

No
Yes Port
Lat/Lon at
Port? mode

No

Lat/Lon inside 750 m Yes

of any FPSO/FSO DP mode
Exclusion zone?
No

Waing for
DP operaon

Fig. 17. Algorithm structure to identify vessel status during the simulations.

The algorithm applied to identify the vessel status is illustrated in Fig. 17, the main
difficulty was to identify the waiting for DP operation condition, as this status is only
roughly characterized.

8000

7000

6000
Demanded Power (kW)

5000

4000

3000

2000

1000

0
0 2000 4000 6000 8000 10000 12000 14000
Time-steps

Fig. 18. Total required power during the Bram Bahia simulation.
 Numerical Simulation of Hybrid Platform Supply Vessel (PSV) 493

6 Numerical Results
6.1 PSV Performance Simulation
The performance simulation of Bram Bahia was performed considering the previously
described status. The demanded power timeline during the time-steps can be seen in
Fig. 18, where large variations can be observed, which are mostly related to the
sailing/navigating conditions with only two occasions in which the maximum available
power was almost reached.

2000
P1(kW)

1000
0
0 2000 4000 6000 8000 10000 12000 14000
2000
P2(kW)

1000
0
0 2000 4000 6000 8000 10000 12000 14000
2000
P3(kW)

1000
0
0 2000 4000 6000 8000 10000 12000 14000
2000
P4(kW)

1000
0
0 2000 4000 6000 8000 10000 12000 14000
500
P5(kW)

0
0 2000 4000 6000 8000 10000 12000 14000
Time-steps

Fig. 19. Engine loads distribution for the several PSV engines.

40

35

30

25
Time (%)

20

15

10

0
Navigating Anchored Port DP Waiting

Fig. 20. Time distribution for the several vessel operational modes for Bram Bahia.

The engine loads timeline distribution is given in Fig. 19, where it can be seen that
engine number 5 (P5) has a smaller maximum power compared to the others, its
optimum SFOC being achieved under a smaller load condition, which could be
advantageous if integrated with other engines for achieving the best compromise on
fuel consumption.
The ship’s operational profile was established based on the total time for each
operational mode as summarized in Fig. 20, where it can be seen that for almost 35%
of the time the vessel is navigating while, for about 27% of the time, the vessel is
494 K. Nishimoto et al.

70

60

50

Fuel consumption (%)

40

30

20

10

0
Navigating Anchored Port DP Waiting

Fig. 21. Fuel consumption for the several operational modes for Bram Bahia.

anchored. The port, DP operation and waiting time percentages complete the opera-
tional profile.
The fuel consumption breakdown for each operation mode can be seen in Fig. 21,
where it can be verified that the navigating condition represents almost 60% of the total
fuel consumption and is appreciably higher than the other ones. It clearly shows the
relevance of the high loads on the engine-generator sets used for navigation. On the
other hand, the DP mode is responsible for about 18% of the total fuel consumption
although corresponding only to 7% of the operational time.

Fig. 22. Total CO2, CH4, SOX, NOX, BC and CO emissions from Bram Bahia.

The total CO2 and other gases emissions for diesel engine and dual-fuel engine
(natural gas mode) operation are illustrated in Fig. 22. It is clearly observed that
operating on natural gas reduces CO2 emissions to about 16%, SOX in 77% and NOX
in 94%. By including a methane slip condition in the simulations, it is noted that the
CH4 emissions increase 104 times, its emission effect in the atmosphere being par-
ticularly severe and requiring a careful assessment of the power system design.
 Numerical Simulation of Hybrid Platform Supply Vessel (PSV) 495

Fig. 23. Relative ship speed, calm, added and total resistance from Bram Bahia simulation.

The simulation regarding the sailing conditions resulted in the total resistance
timeline distribution shown in Fig. 23, including calm and added resistance compo-
nents as well as the relative ship speed (with current correction). It can be observed that
there is a large variation in speed, which is directly correlated to the calm water
resistance. On the other hand, the added resistance component depends strongly on sea
state conditions, therefore it has a weaker correlation with speed, even though still
significant.

80

70

60
Resistance (kN)

50

40

30

20

10

0
Calm Added Total

Fig. 24. Ship resistance breakdown for Bram Bahia.

The mean ship resistance breakdown for the previous ship speed distribution is
given in Fig. 24, where the mean calm water resistance corresponds to about 63 kN,
496 K. Nishimoto et al.

while the mean added resistance is about 17 kN. It is interesting to note that the added
resistance component represents about 27% of the calm water resistance, appreciably
higher when compared to conventional merchant ship patterns in Brazil (*12%).

7 Conclusions

A simulator for PSV operations was developed taking into account ship particulars and
available data of several equipment provided by manufacturers. The ship operational
profile was established based on typical operation patterns and considering the
Brazilian scenario of several oil platforms, main ports and anchorage regions.
The methodology applied consisted of retrieving the ship data on available AIS and
AIS-Sat data from the internet and build a robust database with a specific script for
post-processing the information. An algorithm was developed to automatically identify
the ship status according to speed, latitude and longitude that, together with the
environmental conditions database (current, wind and wave), made possible to generate
a timeline of necessary information to simulate the operational profile of the ship.
Several numerical procedures were implemented and integrated. These include
procedures for estimating the environmental (wind, current and waves) loads, the ship
resistance (calm water and added resistance) components as well as for defining the
thrust and power requirements. The procedure for determining the requirements of the
tunnel and azimuth thrusters is based on a quasi-static approach applied at each time
step and an optimization algorithm of thrust allocation for minimizing the ship’s
required power. Furthermore, an additional algorithm was implemented for optimizing
the load distribution among the generator-engine sets for maximum ship energy effi-
ciency (less fuel consumption).
The simulation of Bram Bahia showed that the main fuel consumption occurs
during the navigation (*60%), this condition corresponding to only 35% of the total
time. On the other hand, although associated with only 7% of the time, the DP oper-
ation mode is responsible for almost 18% of the fuel consumption; the anchored and
port operation contributions being particularly small as the electric loads are minimal.
Finally, the simulations showed clearly the potential of reducing CO2 emissions with
dual-fuel engines with additional reduction of SOX and NOX emissions, although care
must be taken with the possibility of methane slip.

References
Aranha, J.A.: A formula for wave damping in the drift of a floating body. J. Fluid Mech. 275,
147–155 (1994)
Blendermann, W.: Parameter identification of wind loads on ships. J. Wind Eng. Ind. Aerodyn.
51, 339–351 (1994)
Carlton, J.: Marine Propellers and Propulsion. Butterworth-Heinemann Ltd., Oxford (1994)
Fossen, T.I., Perez, T.: Kalman Filtering for positioning and heading control of ships and
offshore rigs: estimating the effects of waves, wind and current. IEEE Control Syst. Mag. 29,
32–46 (2009)
 Numerical Simulation of Hybrid Platform Supply Vessel (PSV) 497

HYCOM. Hybrid Coordinate Ocean Model (2017). https://www.hycom.org/. Accessed 05 Apr

2017
IPCC 2013. Climate Change 2013: The Physical Science Basis, Contribution of Working
Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change
[Stocker, T.F., Qin, D., Platter, K., Tignor, M., allen, S.K., Boschung, J., Nauels, A., Xia, Y.,
Bex, V., Midgley, P.M. (eds.)] 1535 p. Cambridge University Press, Cambridge (2013)
Kjølleberg, J.: Weather Routing of Supply Vessels in the North Sea. Master Thesis, Department
of Marine Technology, Norwegian University of Science and Technology – NTNU,
Trondheim (2015)
Rolls Royce. Propulsion systems for optimal operations under all conditions – Control the Power
(2014). https://www.rolls-royce.com/*/media/Files/R/Rolls-Royce/documents/marine-
product-finder/propulsion-brochure.pdf. Accessed 30 June 2017
Siemens. IEC Squirrel-Cage Motors (2012). https://w3.siemens.com.br/drives/br/pt/motores/mo-
tores-bt/motores-abnt-ate-315l/Documents/Catalogo_de_Motores_IEC_D81.1_-_2008.pdf.
Accessed 12 Aug 2017
Sternersen, D., Thonstad, O.: GHG and NOX emissions from gas fuel engine. Report OC2017 F-
108, SINTEF Ocean AS Maritin (2017)
Vasqués, C.A.M.: A methodology to select the electric propulsion system for Platform Supply
Vessels (PSV). Master Thesis, Escola Politécnica, University of São Paulo (2014). (in
Portuguese)
 Optimization of River-Sea-Going Ship Type
Scheme and Operation Strategy Under
Emission Control Area in China

Xinjing Tan1,2, Wei Cai1,2(&), and Chao Liu1,2

1
School of Transportation,
Wuhan University of Technology, Wuhan, Hubei, China
[email protected]
2
Key Laboratory of High Performance Ship Technology of Ministry
of Education, Wuhan University of Technology, Wuhan, Hubei, China

Abstract. Throughout Chinese existing river-sea-going ships, there are wide-

spread problems such as chaotic ship types, small average tonnage, serious
aging of ships, and low energy efficiency. In order to cater to the requirements of
ship emission control areas implemented in China on January 1, 2016, research
is carried out on optimization of river-sea-going ship type scheme and operation
strategy under emission control area. The lowest cost model of single-ship
round-trip voyage is constructed and the ship operation strategies under the
emission control zone in China are analyzed. Through discussing the effects of
different operation strategies on ship speed, cost and emissions, some sugges-
tions are put forward for the operation of existing river-sea-going ships. At the
same time, combining the optimization of design speed with the optimization of
ship form schemes, a design speed model of river-sea-going liner based on
penalty cost is proposed, and a comprehensive evaluation index system of river-
sea-going liner form schemes based on technological advancement, environ-
mental protection and economy is constructed. Systematic optimization of river-
sea-going ship type scheme provides an effective way for the preliminary design
of the river-sea-going ship, in line with the current river-sea-going shipping
background of energy saving and emission reduction in China.

Keywords: River-sea-going ship
Emission control area
Ship type

optimization
Optimal speed
Carbon emission

1 Introduction

River-sea-going transportation is a transportation mode that directly transports goods to

the destination port without reloading between river vessels and sea vessels. Due to the
high efficiency, low cost and small cargo damage of river-sea-going transportation,
river-sea-going ships occupy an important position in the development of freight
transportation in the Yangtze River [1]. The large-scale and standardization of river-
sea-going ships have become the main trend of development [2]. However, in view of
China’s current river-sea-going ships, there are various problems such as ship clutter,
small average tonnage, low overall technical level, serious ship aging, and serious

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 498–515, 2021.
https://doi.org/10.1007/978-981-15-4680-8_35
 Optimization of River-Sea-Going Ship Type Scheme 499

pollution, hence it is hard to meet the requirements of today’s energy conservation and
emission reduction [3].
Since 2008, International Maritime Organization (IMO) has discussed the appli-
cation of the carbon emission reduction market mechanism in shipping industry for
many times. At present, seven different carbon emission reduction mechanisms have
been implemented in many countries [4]. In order to comply with the current trend of
energy conservation and emission reduction, China’s “Guidelines for the Evaluation of
Energy Efficiency Design Index (EEDI) for Inland Ships”, “Inland Green Ship Spec-
ification” and other policies and regulations have been issued. These regulations put
forward higher requirements on ship design standards and energy efficiency levels. On
January 1, 2016, “Implementation Plan for Ship Emission Control Areas (ECA) in the
Pearl River Delta, Yangtze River Delta, and Bohai Sea (Beijing-Tianjin-Hebei)
Waters” was promulgated to gradually reduce the emission of major pollutant gases
such as nitrogen oxides and sulfur oxides in these regions. In view of the ECA
implementation above, how to meet its demands depends on proper operation and
forward-looking design.
The carbon emissions and fuel consumption of a ship depend to a large extent on
ship speed. At present, there are three main methods to deal with ship carbon emissions
in combination with speed optimization: (i) controlling the total amount of carbon
emissions; (ii) adopting a carbon emission reduction market mechanism; (iii) combin-
ing relevant norms and policies. For the first method, Wang optimized the speed of
different legs in voyage by constructing the main engine fuel consumption model to
realize energy saving and emission reduction for ship transportation [5]. Xue et al. and
Xu et al. considered the impact of speed, and constructed a dual-objective optimization
model with the lowest cost and the lowest carbon emissions [6, 7]. Referring to the
second method, Wang and Xu adopted three different carbon emission reduction
market mechanisms to study the speed optimization model based on the lowest ship
transportation cost [8]. Regarding the third method, Kjetil Fagerholt et al. considered
the problem of ship routing selection and speed optimization in the ECA. This research
pointed out that in the ECA, the ship speed is lower in order to reduce the consumption
of low-sulfur oil; while in the non-ECA, the ship will accelerate the navigation to
meeting the time constraints of the shift, which may lead to an increase in carbon
emissions [9]. For the problem of ship speed optimization in the ECA, Chinese
researchers and scholars have not started relevant research, and there is still a lot of
work to be carried out.
Under the implementation of China’s ECA policy, an empirical analysis of oper-
ational countermeasures in the ECA for existing river-sea-going ships was conducted to
discuss the influence on the ship’s speed, cost and emissions, and eventually sugges-
tions were proposed to guide the existing river-sea-going ship operation. On the other
hand, a design speed model of river-sea-going liner based on penalty cost is proposed,
combined with the comprehensive evaluation index system of river-sea-going ship type
scheme to systematically optimized, which provides an effective way for the prelimi-
nary design of river-sea-going ship, and conforms to the current background of river-
ship-going shipping in China.
500 X. Tan et al.

2 Problem Description and Assumptions

With the continuous promotion of ECA policy, the price gap between low-sulfur oil
and heavy fuel oil will become larger and larger, resulting in a continuous increase on
fuel cost and the substantial increase of the total cost of ship operations. However,
incorrect operational strategies will contradict the original intention of the ECA
implement. For example, a ship may bypass the ECA and sail a longer distance, which
will cause the ship to consume more fuel and lead to higher carbon emissions. When a
ship must pass through the ECA, due to rising transportation cost, it may led to a cargo
transportation mode conversion from sea to land, which will also increase CO2
emissions [9]. Therefore, for the existing river-sea-going ship, specific measures must
be taken to deal with the ECA. In the following, the lowest cost model of single-ship
round-trip voyage is constructed to explore the relationship between speed, cost and
emissions under different operation strategies, and provide appropriate advices for the
operational of existing river-sea-going vessels.
For newly-built ships, design should be forward-looking. River-sea-going container
transportation belongs to liner transportation, in which routes, ports and schedules are
fixed. The stability of the schedule directly determines the service quality of the route,
and maintaining a high standard rate is significant for realizing long-term profitability
of the route. When ship is sailing in the ECA, proper deceleration can reduce the
consumption of low-sulfur oil and reduce the fuel cost. However, a ship cannot reduce
its speed without considering lower limit of speed and timeliness to guarantee the
efficiency in liner transportation.
In order to determine the speed of river-sea-going ships, a speed model based on the
penalty cost is proposed. At the same time, the implementation of the ECA has
improved the environmental performance requirements for new ships, and the opti-
mization of the design speed needs to be synchronized with the optimization of ship
type scheme. This paper proposes a comprehensive evaluation index system for ship
type schemes based on combination weighting method, and systematically introduces
the development and optimization process of ship type schemes for river-sea-going
ships.

2.1 Assumptions
The following assumptions are made in this paper:
(1) The cost of the installation of exhaust gas treatment device is a pre-investment, not
included in the voyage operating costs, and will be considered separately in sub-
sequent calculations.
(2) Under the two strategies, the fixed cost of round-trip voyage is the same, the
optimization objective can be simplified to the lowest voyage fuel cost.
(3) Since fuel consumption of auxiliary engines is independent of the speed, only fuel
consumption of main engines is considered.
(4) Calculation of fuel consumption per unit time of the ship’s main engine is sim-
plified to the cube of the ship speed [11].
 Optimization of River-Sea-Going Ship Type Scheme 501

(5) As the ship is in actual operation, only part of fuel consumption at discrete speeds
can be obtained (see Fig. 1). Based on three data points, this paper uses a weighted
linear combination method to fit the fuel consumption at a certain actual speed to
realize the linear transformation of nonlinear optimization problems [12].

Fig. 1. Diagram of fuel consumption data and Fig. 2. Cross section of a certain river-
weighted linear combination method of 424TEU sea-going container ship.
container ship.

2.2 Load Capacity Calculation Model

Ship’s container capacity is related to the ship’s principal dimensions and can be
estimated at the preliminary design stage. In order to increase the amount of containers,
river-sea-going container ships generally adopt a back-loading arrangement to elimi-
nate the impact of containers on driving sight. Besides, the hatches of most ship types
are set to an open form, which can maximize the use of the cabin space and load more
containers (see Fig. 2).

2.3 Fuel Cost Calculation Model for River-Sea-Going Ship

The fuel consumption of river-sea-going ships varies with the sailing season: during the
dry season (December to March) and the flood season (April to November), the loading
condition of the ship changes with the change of water depth conditions, and the fuel
consumption situation also changes. Due to the different water depth conditions during
the dry season (December to March) and the flood season (April to November), the
ship loading conditions will change, and so that causes the change of fuel consumption.
The calculation formula of the fuel consumption per unit time of a ship is given by
Chao Liu [13], so the detailed process will not be described here. What’s more, the
difference in water flow velocity during the dry season and the flood season will also
affect the ship’s speed. When calculating the fuel cost of river-sea-going vessels, the
influence of sailing season on fuel consumption should be considered. The key steps
are shown as follows:
502 X. Tan et al.

On one hand, the fuel consumption of a ship is:

eV ¼ 106  k
PS
g0
fV ð1Þ

where, eV represents the fuel consumption (t/h), k is a factor for considering the fuel
consumption of all fuel units, and can be taken as 1.1 to 2, PS represents the power of
main engines, g0 represents the fuel consumption rate (g/kwh) of the ship at rated
power, and fV is a correction coefficient of specific fuel consumption and load of
engine, which is given by Chao Liu [13].
On the other hand, considering the channel flow velocity, the container handling
efficiency, and the port berthing time, the round-trip voyage time of the ship is:


Teuv Teuv
tv ¼ ts1 þ ts2 þ2 þ þ 36 ð2Þ
100 50

where, with regard to the channel flow velocity, ts1 and ts2 represent the time of ship
sailing downstream and countercurrent (h) respectively. With regard to the container
handling efficiency and the port berthing time, they vary from port to port. Take a
container ship sailing on a river-sea-going route between Wuhan Port and Shanghai
Yangshan Port as an example, the container handling efficiency of Wuhan port is 50
TEU/h, while Shanghai Yangshan port is 100 TEU/h, and the fixed port berthing time
is taken as 36 h. Teuv represents ship’s container capacity.
In summary, the fuel cost of a ship’s round-trip voyage is:

Cv ¼ 104 cf
eV
tm ð3Þ

where, Cv denotes the cost of the ship’s round-trip voyage (ten thousand yuan), and cf
denotes the fuel price (yuan/ton).
The above calculations are carried out on river-sea-going ships in different seasons,
and finally the fuel cost of the whole year’s navigation can be obtained by adding the
fuel cost of each season.
In addition, transport efficiency indicator to measure the speed of ship transport
turnover is defined as the transport task completed by the ship in a unit time:

Teuv
EFF ¼ ð4Þ
tv

3 Model Construction

On one hand, in order to cope with the ECA policy, for the active ships, the minimum
cost model of single-ship round-trip voyage is constructed in “Strategy 1” and
“Strategy 2” respectively. On the other hand, for the purpose of improving the stability
of liner transportation, considering the transportation economy and liner transportation
efficiency, a punishment mechanism is proposed to impose a penalty cost on delayed
vessel, which characterizes the series of losses caused by the voyage delay. Based on
this, a speed optimization model based on penalty cost is constructed.
 Optimization of River-Sea-Going Ship Type Scheme 503

3.1 Construction of the Lowest Cost Model for Single-Ship Round-Trip

Voyage
There are three main strategies for the ECA policy [10]:
Strategy 1: Fuel conversion. That is, the ship uses low-sulfur oil in the ECA area
and heavy fuel oil in the non-ECA.
Strategy 2: Add an exhaust gas treatment device. The exhaust gas treatment de-vice
can filter out the sulfur oxides in the ship’s exhaust gas, so ships can continue to use
heavy fuel oil in the ECA.
Strategy 3: Use clean energy such as LNG. As a clean energy source, LNG can
greatly reduce emissions. Using LNG as a fuel naturally enables ships to meet
ECA’s emission requirements.
Since LNG technology is not yet mature, and the use of LNG as a fuel requires
sufficient filling stations as support, such measures have not yet been popularized. For
active ships, the first two strategies are more feasible.
Objective function of Strategy 1:
X 
min Z1 ¼ PnE
FmnE
xnE
m þ P
Fm
xm
E E E
ð5Þ
m2V

Constraints of Strategy 1:
X
xnE
m ¼1
m2V
X
xEm ¼ 1
m2V
ð6Þ
0  xnE
m 1
0  xEm  1
tV  tmax

Objective function of Strategy 2:

X
min Z2 ¼ PS
FmS
xSm ð7Þ
m2V

Constraints of Strategy 2:
X
xSm ¼ 1
m2V

0  xSm  1 ð8Þ

tVS  tmax
504 X. Tan et al.

where, Z1 and Z2 indicate the round-trip voyage cost under Strategy 1 and Strategy 2
respectively, xnE
m and xm represent the weight set of ship speed m when sailing in the
E

non-ECA and the ECA under the Strategy 1 respectively, and xSm represents the weight
set of the speed m under the Strategy 2. m 2 V ¼ fVl ; Vm ; Vh g represents the speed
interpolation point, which is taken as the minimum, median and maximum value of the
ship operating speed respectively. PnE and PE represent the fuel price for the non-ECA
and the ECA under Strategy 1 respectively, which is taken as 2,500 yuan/ton and 4,500
yuan/ton respectively, and PS represents the fuel price under Strategy 2, which is taken
as 2,500 yuan/ton. FmnE and FmE indicates the fuel consumption at the speed m in the non-
ECA and the ECA under Strategy 1 respectively, and FmS indicates the fuel consumption
at the speed m under Strategy 2. tV and tVS represent the round-trip voyage time under
Strategy 1 and Strategy 2 respectively, which can be calculated by the following:
X 
tV ¼ xnE
m
tm þ x m
tm
nE E E

m2V
X  ð9Þ
tVS ¼ xSm
tmS
m2V

where, tmnE and tmE respectively indicate the time taken by the ship for round-trip voyage
at the speed m in the non-ECA and the ECA under Strategy 1, and tmS indicates the time
taken by the ship for round-trip voyage at speed m under Strategy 2.

3.2 Construction of Speed Model Based on Penalty Cost

One concept related to the penalty cost is the “time window”, which can be understood
as the time limit for ship’s arrival. At present, the most used is the hard time window
constraint, that is, the ship must satisfy the time window constraint, and the soft time
window constraint is opposite.
The soft time window constraint allows the ship to violate the time window con-
straint to a certain extent, and imposes a certain fee on the ship as a punishment.
Studies have shown that using soft time window constraint can reduce the ship
transportation cost while maintaining high transportation efficiency [14]. Several
commonly used penalty cost functions are shown in Fig. 3. Where ½0; ta  is the internal
time window, indicating the best time for the ship to arrive at the port; ½0; tb  is the outer
time window, which the ship must not violate; ½ta ; tb  is the allowed time interference,
within which the ship is subject to the penalty fee.

Fig. 3. Graphics of several penalty cost function.

 Optimization of River-Sea-Going Ship Type Scheme 505

Referring to the above several penalty cost functions, the penalty cost function in
this paper is constructed as follows:

c
ttsbt a
ta  ts \tb
Pðts Þ ¼ ts ð10Þ
0 0\ts \ta

where, c is the penalty cost coefficient, which is taken as 50,000 yuan, the soft time
window ½ta ; tb  is mainly determined according to the schedule of river-sea-going ships.
Nowadays, ships are generally required to reach Shanghai within 72 h of departure
from Wuhan. Considering the influence of downstream and upstream flow velocities
and the uncertainties of navigation at sea, a certain margin of navigation time is
reserved as follow:
 
ð48; 60Þ Wuhan port to Shanghai Yangshan Port
ðta ; tb Þ ¼ ð11Þ
ð72; 84Þ Shanghai Yangshan Port to Wuhan port

Combined with the fuel cost calculation model established before, the Required
Freight Rate (RFR) of the liner based on the penalty cost can be established as follow:

1
RFRP ¼ ½Y þ P þ CR
ðI  PW
RÞ ð12Þ
Q

where, Q represents the number of containers carried by the ship each year, Y indicates
the annual operating cost of the ship, which consists of fixed costs such as crew fees,
insurance premiums, and variable costs such as fuel costs and port charges. P indicates
the annual penalty cost, which can be accumulated from the penalty costs of all
voyages, CR represents the capital recovery factor, I is the cost of the ship, PW denotes
the present value factor, and R denotes the residual value of the ship, which is taken as
10% of the ship price.

4 Solution Method

The optimal design speed of a ship is determined by economy, environmental pro-

tection performance, and technological advancement. Therefore, the optimization of
design speed should be synchronized with the optimization of ship type scheme. The
comprehensive evaluation index system of river-sea-going ship type based on com-
bination weighting method will be introduced below.
Comprehensive Evaluation and Optimization of Ship Type Scheme Based on
Combination Weighting Method
After calculating the relevant parameters of river-sea-going vessel series schemes with
different container capacity classes, it is necessary to evaluate the feasibility of the ship
type schemes and select the feasible schemes for subsequent comprehensive evaluation
and optimization. Feasibility evaluation indicators include ship type safety indicators
and technical indicators (Table 1).
506 X. Tan et al.

Table 1. Feasibility evaluation indicators of ship type scheme.

Feasibility evaluation Evaluation contents
indicators
Weight check Gravity and buoyancy balance
Initial stability Initial stability height GM > 0.3 m
Rolling period Rolling period > 8 s
Freeboard Meeting the specification requirements
Container capacity Within the range of ship’s container capacity at different container
capacity classes
Sailing time Satisfying time window constraints

For the feasible schemes, the system engineering idea is used to construct a
comprehensive evaluation index system to optimize the ship type scheme. Considering
the characteristics of ship type and liner shipping of river-sea-going vessels, the
comprehensive evaluation of the series of ship type schemes is carried out in terms of
technological advancement, environmental protection and economy. The comprehen-
sive evaluation index system frame is shown in Fig. 4.
Regarding the comprehensive evaluation index system proposed above, the weight
of each evaluation index is determined by the combination weighting method, which
combines subjective weights with objective weights. It reflects a compromise between
the subjective preferences of the designers and the objective distribution of each index.
The key steps are shown as follows:
Determine the Subjective Weight Vector by Analytic Hierarchy Process (AHP).
Firstly, the hierarchical structure model is established. The decision-making system is
divided into the target layer, the criteria layer and the scheme layer. Then, the index
weights of each layer are calculated, and the subjective weight vector of each index can
be obtained, as shown in Table 2.

Fig. 4. The comprehensive evaluation index system for river-sea-going container ship.
 Optimization of River-Sea-Going Ship Type Scheme 507

Table 2. Subjective weight of comprehensive evaluation indexes.

Criteria layer Weight Sub-criteria layer Weight Subjective weight x1
Technological advancement 0.261 C 0.435 0.113
R 0.063 0.016
Dto 0.132 0.035
EFF 0.37 0.096
Environmental protection 0.633 EEDI 1 0.633
Economy 0.106 RFRP 0.833 0.088
RFD 0.167 0.018

Determining the Objective Weight Vector by Variation Coefficient Method.

Calculations are as follow:

Vi ¼ rxii
x2 ðiÞ ¼ Pn
Vi
ð13Þ
Vi
i¼1

where, Vi , ri , and xi respectively represent the variation coefficient, the standard

deviation, and the mean value of index i; n represents the number of indexes; x2
represents the objective weight vector.
In this paper, the river-sea-going ship type scheme of 500TEU-capacity-class is
selected as the objective weight calculation sample, and the objective weight vector x2
is calculated (Table 3).

Table 3. Objective weight of comprehensive evaluation indexes.

Evaluation index Variation coefficient Objective weight x2
Technological advancement C 0.024 0.102
R 0.043 0.185
Dto 0.009 0.041
EFF 0.021 0.088
Environmental protection EEDI 0.059 0.251
Economy RFRP 0.019 0.082
RFD 0.059 0.251
508 X. Tan et al.

Determining the Combination Weight Vector by Combination Weighting

Method. Combination weight vector is calculated as follow [15]:

x ¼ a1 x1 þ a2 x2
ð14Þ
a1 þ a2 ¼ 1

where, a1 and a2 represent the combination coefficients of subjective and objective

weights respectively. If the decision maker is an empiricist, take a1 [ a2 ; if the
decision maker is a non-preferred neutral person, take a1 ¼ a2 ; if the decision maker is
a conservative, take a1 \a2 . Since the experience of designers plays a significant role in
river-sea-going ship design, it can be considered that the comprehensive evaluation
index system is largely dominated by designer’s experience, take a1 ¼ 0:6, a2 ¼ 0:4.
Eventually, the combined weights of the comprehensive evaluation index system
for river-sea-going container vessels can be obtained (Table 4).

Table 4. Combination weight of comprehensive evaluation indexes.

Evaluation index Subjective Objective Combination
weight x1 weight x2 weight x3
Technological C 0.113 0.102 0.109
advancement R 0.016 0.185 0.083
Dto 0.035 0.041 0.037
EFF 0.096 0.088 0.093
Environmental EEDI 0.633 0.251 0.480
protection
Economy RFRP 0.088 0.082 0.086
RFD 0.018 0.251 0.111

After the establishment of key technical models and the comprehensive evaluation
index system, it enables to optimize the series ship type schemes of river-to-sea vessels
at each container capacity class and obtain the optimum ship type scheme (Fig. 5).
 Optimization of River-Sea-Going Ship Type Scheme 509

Fig. 5. Optimization process of river-sea-going ship type scheme.

5 Case Study and Analysis

For existing ships: a typical active river-sea-going ship is selected as an example, and
the ship speed, cost and carbon emissions under two ECA strategies are compared and
analyzed. For newly-built ships: taking a certain container capacity class of river-sea-
going ship as an example, the optimization process of ship type elements and speed is
systematically introduced, and the rationality of the penalty cost function based on soft
time window constraints and its impact on ship form optimization results are further
verified according to the optimization results.
510 X. Tan et al.

5.1 Case-Based Strategy Analysis of ECA

A 424TEU river-sea-going ship (Table 5) on route from Wuhan port to Shanghai
Yangshan port (Table 6) is chosen as an example. Aiming at minimizing the operation
cost of a single-ship round-trip voyage, the optimal speed of the ship under Strategy 1
and Strategy 2 is decided. Besides, the effects of these two ECA strategies on ship
speed, cost and emissions are discussed below.

Table 5. Main Elements of 424TEU river-sea-going container ship.

LPP B D Td Ts V0 P
122.8 m 18.8 m 8.3 m 4.5 m 6.0 m 11.88 kn 1980 kW

Table 6. Legs Division on route from Wuhan port to Shanghai Yangshan port.
Leg Area type Navigation distance (n mile)
Wuhan to Nanjing Non-ECA 396
Nanjing to Shanghai Yangshan ECA 211

Table 7. Single-ship operation optimization results under different ECA strategies.

Strategy Ship speed (kn) Fuel cost (Ten Carbon
Non- ECA thousand yuan) emission
ECA (t)
Strategy 1: fuel conversion 9.04 7 9.63 100.59
Strategy 2: exhaust gas treatment 8.33 8.02 99.85
Strategy 3: without considering the 8.33 10.25 100.88
ECA policy

The optimal speed and single-ship round-trip fuel cost under two strategies are
calculated by using the linear optimization method in MATLAB. On the basis of fuel
consumption, the emission level of the ship under two strategies can be obtained
according to the carbon conversion coefficients of different fuels.
In order to further explore the impact of ECA implementation on single vessels, a
ship operates without considering the ECA policy will be considered as the third
option. In this case, it is assumed that the ship sails on the route without ECA and it
consumes heavy fuel oil throughout the route, so its optimal speed is the same as that of
Strategy 2. In fact, the ship consumes low-sulfur oil in the ECA, thus, the actual cost of
the ship under this strategy is larger than Strategy 2. The specific optimization results
are shown in Table 7.
The initial investment in the installation of exhaust gas treatment equipment on
ships is related to the power of the main engine. According to statistics from relevant
foreign institutions, the construction cost per kilowatt of main engine power is 118
 Optimization of River-Sea-Going Ship Type Scheme 511

Euros [16]. Considering the initial investment, the ship’s operating costs and carbon
emissions over time are in accordance with the optimal speed in Table 7 are as shown
in Fig. 6 and Fig. 7 respectively.
For the results above, it can be found that:
(1) When adopting the fuel conversion strategy, the ship will slow down sailing in the
ECA to reduce the consumption of high-priced low-sulfur oil, while it will
accelerate the navigation in non-ECA to make up for the loss of time.
(2) When adopting the exhaust gas treatment strategy, the cost of ship voyage is small,
but a large initial investment is required;
(3) If the implementation of ECA is not considered and no measures are taken, the
voyage cost and emissions are the largest;
(4) In the initial stage of ship operation, the cost of Strategy 2 is greater than that of
Strategy 1 (Fig. 6 and Fig. 7). When the ship operates in 2–3 years, the cost is
basically the same under this two countermeasures. Thereafter, the cost of Strategy
2 will be lower than Strategy 1. Moreover, emissions of Strategy 2 are always
slightly lower than that of Strategy 1.

Fig. 6. Variation of single-ship operating Fig. 7. Variation of single-ship carbon emis-

cost with time under different ECA strategies. sion with time under different ECA strategies.

Through the case-based ECA strategy analysis above, the following recommen-
dations can be made for the single-ship operation under the ECA policy: when the
vessel is in short-term operation, fuel conversion is recommended to adopt; if the vessel
has a longer service life, exhaust gas treatment should be adopted. With the passage of
time, the ship can achieve better economic benefits, and be conducive to achieve
energy conservation and emission reduction.
512 X. Tan et al.

5.2 Case-Based Ship Type Scheme Optimization Analysis

of River-Sea-Going Ship
A 900TEU-capacity-class river-sea-going container ship is chosen as an example, the
optimization process of its ship type elements and speed is introduced. Finally, the
optimal results of each container capacity class of ship type scheme are analyzed. The
key steps are as follows:
Select the Base Scheme. Since currently there are no 900TEU-capacity-class river-
sea-going ships on the route from Wuhan port to Shanghai Yangshan port, it is nec-
essary to carry out its preliminary design. The ship type parameters of the basic scheme
are shown in Table 8.

Table 8. Principal dimension of 942TEU river-sea-going ship.

LPP B D Td Ts Cb V0 P
139 m 25.6 m 10.05 m 4.5 m 6.0 m 0.8 11.5 kn 2*1103 kW

Generate a Series of Schemes. Based on the scheme above, grid method is used to
generate a series of ship type schemes (Table 9). The ship length LPP varies with the
gradient of standard container length plus clearance. According to the waterway
restriction and related specification requirements, the upper limit is taken as 150 m.
The ship breadth B varies with standard container width as step length, and the upper
limit is taken as 30 m. The breadth-depth ratio B=D ranges from 2.2 to 2.6. The design
draught mainly considers the seasonal variation of the minimum maintenance depth of
the waterway, ranging from 4.5 m to 6.0 m. The block coefficient Cb mainly considers
the hypertrophy characteristics of the river-sea-going ship type, ranging from 0.76 to
0.8. The speed range is set from 11 to 14 kn, according to the design speed of existing
typical river-sea-going vessels.
Choose the Optimal Scheme. Aseries of schemes are obtained by free combination of
the above main dimension variables. The relevant technical indicators of each scheme
are calculated by MATLAB programming, and the feasible schemes are obtained by
analyzing the feasibility according to Table 1. Eventually, based on the comprehensive
evaluation index system, the ship type schemes are evaluated comprehensively, and the
top 10 schemes are obtained as shown in Table 10.
In addition, in order to verify the rationality of the penalty cost function based on
the soft time window constraints and its impact on the ship type optimization results,
the preferred results of river-sea-going vessels at different container capacity classes
under no constraint and the hard time window constraint ðts 2 ð0; ta ÞÞ are compared
with that under soft time window constraint (Fig. 8, 9, 10 and 11).
 Optimization of River-Sea-Going Ship Type Scheme 513

Table 9. Partition of principal dimension variables based on the grid method.

LPP (m) B (m) B=D Td (m) Cb V0 (kn)
120.4 20.6 2.2 4.5 0.76 9
126.6 23.1 2.3 5.0 0.77 9.2
132.8 25.6 2.4 5.5 0.78



139 28.1 2.5 6.0 0.79 13.8
145.2 30 2.6 – 0.80 14

Table 10. Optimum selection results of 942TEU river-sea-going ship type schemes.
Ranking LPP (m) B (m) D (m) Td (m) Cb V0 (kn) P (kW)
1 145.2 25.6 10.24 4.5 0.8 11.4 2*1140
2 145.2 25.6 10.24 5.5 0.8 11.0 2*1030
3 145.2 25.6 10.24 6.0 0.8 11.0 2*1140
4 139.0 25.6 9.85 6.0 0.8 10.8 2*1030
5 145.2 25.6 10.24 6.0 0.8 10.8 2*1030
6 145.2 25.6 10.24 5.0 0.8 11.2 2*1140
7 145.2 25.6 10.24 6.0 0.8 10.6 2*1030
8 145.2 25.6 10.24 5.5 0.8 10.8 2*1030
9 139.0 25.6 9.85 5.5 0.8 11.0 2*1030
10 145.2 25.6 10.24 5.5 0.8 11.2 2*1140

Fig. 8. Speed comparison of optimal river- Fig. 9. Transport efficiency comparison of

sea-going ship type schemes at different optimal river-sea-going ship type schemes at
container capacity classes under different different container capacity classes under
time window constraints. different time window constraints.
514 X. Tan et al.

Fig. 10. RFR comparison of optimal river- Fig. 11. EEDI comparison of optimal river-
sea-going ship type schemes at different sea-going ship type schemes at different
container capacity classes under different time container capacity classes under different
window constraints. time window constraints.

It can be seen in Fig. 8, 9, 10 and 11 that:

(1) When the soft time window constraint is adopted, the RFR of the optimal ship type
scheme is slightly increased compared with the other two modes, but its trans-
portation efficiency and energy efficiency index can reach a better level. This shows
that for river-sea-going liner transportation, using soft time window constraint can
well balance the three indicators of cost, emissions and transportation efficiency,
and achieve more balanced results.
(2) With the increase of the container capacity of the ship type scheme, the three major
indicators have been optimized to different degrees, which also confirms the
advantages of large-scale ship type.

6 Conclusion

On the basis of fully understanding the implementation of China’s ECA policy, rea-
sonable suggestions are made for the operation of existing river-sea-going vessels by
constructing the lowest single-ship round-trip voyage cost model. Penalty cost is put
forward, and a speed model based on penalty cost under ECA is proposed. Addi-
tionally, as for the newly-built river-sea-going ships, the ship type schemes at different
container capacity classes are evaluated and optimized, focusing on the optimization
design for ships at large container capacity classes. This paper provides effective
recommendations for the design, development and operation of river-sea-going ships to
address the implementation of ECA policy in the Yangtze River Delta region in China.
However, this paper mainly considers the optimization of ship’s main parameters in the
design of river-sea-going ship, while the structure and layout of ship are ignored. In
order to make the designed ship type scheme truly applicable to newly-built ship,
detailed performance check and model test should be carried out for the preliminary
scheme.
 Optimization of River-Sea-Going Ship Type Scheme 515

Acknowledgement. This research is supported by Development of an Energy-saving and

Environmental-Friendly Type River-Sea-Going Demonstration Container (Research of Ministry
of Industry and Information Technology: High Technology Ship, No. 20151g0006), Classifi-
cation and Connotation of Green Ship Technology (Research of Chinese Academy of Engi-
neering, No. 20151g0118) and the Programme of Introducing Talents of Discipline to
Universities: High Performance Ship (No. B08031).

References
1. Huang, Z., Yang, B., Wang, H.: Research on the development and ship type characteristics
of river-sea-going ships. China Water Transp. (08), 29–30 (2015)
2. Ye, J., Zheng, D.: Present situation and development strategy of “River-Ocean Combined
Transportation” on the Yangtze River. J. Zhejiang Ocean Univ. (Hum. Sci.) 34(02), 19–24
(2017)
3. Jiang, W., Pei, Z., Wu, W.: Structural strength analysis of broad flat and full river-sea-going
ship. ZhongGuoShuiYun (01), 54–57 (2017)
4. Gu, W.: Market mechanism analysis on greenhouse gas emission reduction of international
shipping. J. Shanghai Maritime Univ. (03), 17–21 (2013)
5. Wang, Z.: Research on energy saving and optimization of sectional speed for inland vessels.
Dalian Maritime University (2014)
6. Xue, Y., Shao, J.: Fleet deployment for liner shipping in low-carbon economy. Navig. China
(04), 115–119 (2014)
7. Xu, H., Liu, W., Shang, Y.: Fleet deployment model for liners under low-carbon economy
and its algorithms implementation. J. Transp. Syst. Eng. Inf. Technol. (04), 176–181 (2013)
8. Wang, C., Xu, C.: Sailing speed optimization in zvoyage chartering ship considering
different carbon emissions taxation. Comput. Ind. Eng. 89, 108–115 (2015)
9. Fagerholt, K., Gausel, N.T., Rakke, J.G., et al.: Maritime routing and speed optimization
with emission control areas. Transp. Res. Part C Emerg. Technol. 52, 57–73 (2015)
10. Patricksson, Ø.S., Fagerholt, K., Rakke, J.G.: The fleet renewal problem with regional
emission limitations: case study from roll-on/roll-off shipping. Transp. Res. Part C: Emerg.
Technol. 56, 346–358 (2015)
11. Wang, S., Meng, Q.: Sailing speed optimization for container ships in a liner shipping
network. Transp. Res. Part E: Logist. Transp. Rev. 48(3), 701–714 (2012)
12. Andersson, H., Fagerholt, K., Hobbesland, K.: Integrated maritime fleet deployment and
speed optimization: case study from RoRo shipping. Comput. Oper. Res. 55, 233–240
(2015)
13. Liu, C., Cai, W., Mei, M., et al.: Establishment of a comprehensive fleet optimization model
for river-sea-going ships under low-carbon economy. Greece: Int. Soc. Offshore Polar Eng.
ISOPE-I-16-273 (2016)
14. Fagerholt, K.: Ship scheduling with soft time windows: an optimization based approach.
Eur. J. Oper. Res. 131(3), 559–571 (2001)
15. Hu, P.: Multi-objective optimization and decision making of conceptual ship design.
Comput. Digital Eng. (03), 390–394 (2014)
16. Jiang, L., Kronbak, J., Christensen, L.P.: The costs and benefits of sulphur reduction
measures: Sulphur scrubbers versus marine gas oil. Transp. Res. Part D: Transp. Environ.
28, 19–27 (2014)
 Estimation of the Collision
Risk on Planned Route

Makiko Minami1(&) and Ruri Shoji2

1
National Maritime Research Institute, National Institute of Maritime,
Port and Aviation Technology, Tokyo, Japan
[email protected]
2
Tokyo University of Marin Science and Technology, Tokyo, Japan
[email protected]

Abstract. In recent year, research on autonomous ships has progressed, and

collision avoidance is one of the important tasks among them. Collision
avoidance maneuvering includes judgment on uncertain events such as forecast
of motions of other vessels. Therefore, it is effective to improve safety by
navigating away from areas where a complicated encounter, in other words high
collision risk, is likely to occur even when establishing the collision avoidance
technology. Collisions tend to occur repeatedly in similar places. For this reason,
areas where complicated encounter situations are likely to occur are considered
to be areas where collisions occur frequently in the past. In addition, it is
considered that the area in which many abnormal approaches of the ship occur is
also an area where complicated encounter is likely to occur. From the analysis of
AIS data, the ship close point is extracted and shown on the map. Also, in order
to clarify the collision risk in the sea area, collision points and ship close points
were converted into continuous distribution by kernel density estimation method
and shown. Then, considering the time, the kernel density estimation method
was extended to three dimensions so that the collision risk of the target area can
be estimated. We used this to estimate the collision risk along the planned route.
It was found that even on the same route the collision risk greatly differs
depending on the time zone to be sailed. Even in existing vessels, the collision
risk on the planned route can be obtained using this system. Therefore, it is
possible to expect the accident prevention effect by allowing the operator to
navigate by grasping the high-risk sea area.

Keywords: Collision risk
Autonomous ships
AIS data
Kernel density

estimation method

1 Introduction

In recent year, research on autonomous ships has progressed, and collision avoidance is
one of the important tasks among them. It is because not only human lives may be lost
by the collision, but it also puts a load on nature by the outflow of fuel oil and cargo.
According to the annual of JTSB (the Japan Transport Safety Board), the collision
surveyed in 2017 hits 26.5% of number of all investigations and occupies the biggest
ratio [1].

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 516–532, 2021.
https://doi.org/10.1007/978-981-15-4680-8_36
 Estimation of the Collision Risk on Planned Route 517

In the collision avoidance maneuver, it is necessary to predict the movement of

other ship. If there is a ship other than the ship to be avoided and that disturbs own ship
course, navigation judgments become complicated. Therefore, even when a man
maneuvers a ship, collisions tend to occur frequently in sea areas where congestion or
ship courses are crossed. Even if collision avoidance technology is established, the risk
of collision and the avoidance method are judged based on the prediction as in the case
where a man maneuvering. Therefore, in order to improve the safety of navigation, it is
also effective to navigate away from areas where complex encounter situations occur
frequently.
It is said that human error has more than 80% of collision causes, but it is pointed
out that a collision tends to occur for a similar place and a cause repeatedly. In sea areas
where collisions frequently occurred in the past, there is a high possibility that com-
plicated encounter situations will occur. It is suggested that collision outbreak points
are plotted on a map, such as a hazard map of marine accidents made by JTSB, as
means to grasp an occurrence condition of collisions in a navigation area. In the sea
area where collisions occur frequently, a lot of points indicating the collision outbreak
are plotted. Therefore, a probability of collision of the target sea area can be shown
from that distribution. In this paper, a probability of collision is called collision risk.
In this research, in order to make it easy to grasp the distribution of collision
occurrence points, we considered conversion of point distribution to continuous one
using the Kernel density estimation method and display it as probability density dis-
tribution. The higher the density, the more likely it is that accidents are likely to occur,
or a complicated encounter situation is likely to occur. In other words, it can be said
that the collision risk is large. And then, it is said that there are many incidents behind
the accident occurrence. Therefore, we observed the movement of ship in Tokyo Bay
using AIS (Automatic Identification System) data, extracted the point where ships were
approach to each other and displayed on the map. And we also indicated close points of
ships by the probability density distribution.
From these studies it was assumed that the collision risk should be considered not
only for the area but also for the time. With autonomous ship, since the person is not on
board, it is possible to adjust the departure/arrival time by waiting offshore. Therefore,
we researched to lay a navigation planning with minimizing the collision risk as much
as possible, by calculating the probability density distribution of collision point and
ship close point along the planned route with different departure and arrival time, route
and speed.
518 M. Minami and R. Shoji

2 The Outbreak Condition of Collisions in Tokyo Bay

Collison occurrence points in Tokyo Bay are shown in Fig. 1. It shows 215 cases
involving merchant ship among collisions searched in August 2014 in J-MARISIS
(Japan-Marine Accident Risk and Safety Information System) [2]. We converted the
latitude and longitude of the collision occurrence point to plane rectangular coordinates
(x, y). The positive direction of x coordinate is
east, and the positive direction of y coordinate is
north. In addition, the color of the symbol indi-
cates an encounter situation. And red broken lines
indicate Uraga Suido Traffic Route, Nakanose
Traffic Route and Tokyo West Passage.
Collisions during anchoring or drifting often
occurred on quays and anchorage. And what
occurred near the center of the bay is collision
with a fishing boat under fishing. In addition,
collisions in the crossing situation occurred fre-
quently in off Yokohama, off Kawasaki and off
Haneda where the course crosses. Although the
number of occurrences of collisions in the over-
taking situation is small, most of them occurred in
the route. Collisions in the head on situation
occurred in relatively narrow places such as inside
the harbor. In particular, near of Yokohama Port
collision occurred frequently in the crossing situ-
ation and in the head on situation, because various
ship’s course such as a ship heading to Tokyo Port Fig. 1. Collision occurrence points
or a ship traversing Tokyo Bay crossed. in Tokyo Bay
The number of collisions by
encounter situation is 81 cases of
‘anchoring or drifting’, 76 cases
of ‘crossing’, 32 cases of ‘head
on’, 17 cases of ‘overtaking’, 9
cases of ‘others’. Two types of
‘crossing’ and ‘anchoring or
drifting’ occupy more than 70%
of the whole. In addition, Fig. 2
shows the number of occurrence
of collision by time. The number
of one is large in time zone from
5:00 to 8:00 (in the morning) and
from 16:00 to 19:00 (in the eve- Fig. 2. The number of occurrence of collision by time
ning). In the morning, the number
of collision in head on situation is larger than other time zone, and in the evening there
are many collisions in crossing situation.
 Estimation of the Collision Risk on Planned Route 519

3 Evaluation of Collision Risk of Target Area

In the previous chapter, the collision occurrence points are plotted on the map to show
the occurrence condition of accident. In this chapter, we tried converting the point
distribution to a continuous distribution in order to evaluate the collision risk in the
target area. There is a frequency distribution as a method of converting to continuous
distribution. Figure 3 shows the collision occurrence point shown in Fig. 1 by the
frequency distribution with the division sizes set to 1.0 nm. (a) and (b) are obtained by
shifting the origin of the boundary by 0.5 mile each in the x direction and the y
direction without changing the division size. The same data but gives the impression
that the collision occurrence situation differs depending on the boundary position.
Because whether the collision point is in the vicinity of the boundary or in the center, it
is handled in the same way. In this research, assuming that the point distribution like a
collision point follows a certain probability density function, we converted it into a
continuous distribution by estimating the original probability density function from the
actual point distribution using the kernel density estimation method [3].

Fig. 3. Frequency distribution of the number of collisions – compare the difference of origin

3.1 The Kernel Density Estimation Method

We expressed the occurrence condition of collision by probability density distribution
using kernel density estimation method. This method is as follows. First, the proba-
bility of collision occurrence around the collision point is assumed. Namely, in the
vicinity of the collision point, the collision occurrence probability is considered high,
520 M. Minami and R. Shoji

and a probability density function called a kernel with the collision occurrence point as
the maximum value represents the collision occurrence probability. Next, for all col-
lision occurrence points in the target area, the collision occurrence probability around
its point is expressed by the kernel function. Finally, all the kernel functions are
summed up as the probability density function of the target area. By means of this, we
can convert a point distribution such as a collision point into a continuous distribution
represented by probability density.
When the observed values {W1, W2, …, Wj, …, Wm}(Wj = (Wj1, Wj2, …, Wjd)T) is
given, the estimated value bf ðwÞ of the d-dimensional probability density function is
defined by (1) with kernel K and window width h.

X  
bf ðwÞ ¼ 1
m
1 
K w  W j : ð1Þ
mhd j¼1 h

The kernel function K(u) is a function defined for d-dimensional u, satisfying

Z
K ðuÞdu ¼ 1; for all u 2 Rd : ð2Þ
Rd

As a kernel function, Gaussian, Epanechnikov and Rectangular are generally used

as described in [3]. In this study, we used the Gaussian kernel which is easy to calculate
analytically. One-dimensional and two-dimensional Gaussian kernels are shown in (3)
and (4) respectively.

1
1d : K ðuÞ ¼ pffiffiffiffiffiffi eð1=2Þu
2
ð3Þ
2p
 
1 1
2d : K2 ðuÞ ¼ exp  uT u ð4Þ
2p 2

3.2 The Probability Density Distribution of Collision Occurrence Points

From collision occurrence points, we found the collision occurrence probability density
distribution using two-dimensional kernel density estimation method. The estimated
value of the probability density at x(x, y) is given by

Xn
x  xi y  yi
bf ðxÞ ¼ 1 Ks ; : ð5Þ
nh2s j¼1 hs hs

Where Ks is a kernel function based on the geographical distance, hs is the window

width and n is the number of collision points.
The estimation result of the probability density depends greatly on the window
width. It determines the spread of the distribution centered on the observation point.
The smaller the window width, the more the impact of its point on the surroundings
 Estimation of the Collision Risk on Planned Route 521

becomes limited. Several methods have been proposed for decision, but finally it is
recommended to determine the optimal value by trial.
Figure 4 shows examples of the probability density distribution obtained by
changing the window width for a collision in crossing situation. The number of target
points, n, is 76. As the widow width becomes large, the probability density distribution
is smoothed, making it difficult for the features of the target area to appear. When the
window width is 3704 m (2.0 nm), it is distributed in almost all areas of Tokyo Bay.
On the other hand, when the window width is 926 m (0.5 nm), the probability density
is high only around the accident point. In this study, it was judged that hs = 1852 m
(1.0 nm) was the most appropriated for trial calculations.
Figure 5 shows the occurrence probability density distribution by encounter situ-
ation of collision points shown in Fig. 1. In this study, the estimation result was
normalized by the number of points used for estimations, and it was shown in 10 levels
with the highest value being 10. As the level is higher, we judged that the collision risk
is high, because it indicates that many collisions have occurred in that sea area.

Fig. 4. Probability density distribution of collision in crossing situation – comparison of

window width

Fig. 5. Collision occurrence probability density distributions by encounter situation

522 M. Minami and R. Shoji

In head on situation, the density was high in Yokohama Port and Tokyo West
Passage, and in crossing situation, that was high in Tokyo Port and off Yokohama Port.
In overtaking situation, we can see that the density was high in the central part of
Tokyo Bay and Uraga Suido Traffic Route. However, since the number of collisions
was small, even the maximum is around Level 3. Also, during anchoring or drifting the
density was high around the harbor.
Figure 6 shows the change of collision occurrence probability density distribution
every 3 h. There were high density areas in 6:00–9:00, 9:00–12:00, 15:00–18:00 and
18:00–21:00. In the morning, the density was high in the range from Tokyo Port to
Yokohama Port, especially inside the harbor. On the other hand, in the evening, the
density was high in the range from Yokohama Port to the west side of Nakanose.
Between 3:00 and 6:00, the density was high in the sea area bound for Uraga Suido
Traffic Route from the entrance of Tokyo Bay. Depending on the time zone, the
magnitude and location of the probability density appears differently. In other words, it
is necessary to consider the time to evaluate the collision risk of the target area.

Fig. 6. Collision occurrence probability density distributions by time

 Estimation of the Collision Risk on Planned Route 523

4 Evaluation of the Collision Risk Considering Time

4.1 Spatiotemporal Distribution of Collision Points
To see a time variation in the occurrence condition of the collision, collision points are
shown in the spatiotemporal coordinates in Fig. 7. The symbols are color coded by the
encounter situation. In the vicinity of Yokohama Port, collisions occurred at any time
of day. And there were many collisions in crossing situation and during anchoring or
drifting. In addition, there were few collisions before dawn in the vicinity of Tokyo
Port, but collision in head on situation frequently occurred at 6:00–12:00. And collision
in overtaking situation occurred more frequently in the southern part of Tokyo Bay
around 6:00 than that of the other areas
and time zones.
However, in spatiotemporal coordi-
nates, collision points are expressed in a
state floating in the coordinate space.
Since it is difficult to specify the posi-
tion and time, it is difficult to grasp the
detailed collision occurrence situations.
In addition, although the total number of
collisions is large, the frequency of unit
area and time is low. So, it is difficult to
grasp the collision risk only with the
distribution of collision points. Fig. 7. Spatiotemporal coordinates of collision
points

4.2 Three-Dimensional Kernel Density Estimation Method

In order to more clearly display the features of the collision point distribution in three
dimensions, the probability density dis-
tribution was used. We extended the
two-dimensional kernel density estima-
tion method shown in Sect. 3.2 to three
dimensions considering temporal devi-
ation.
We obtain an estimate of the proba-
bility density at arbitrary time-space
coordinates (x, y, t). Since geographical
distances and temporal deviations are
independent dimensions, the product of
each kernel function is substituted into
(1) and it can be expressed as in (6).
Fig. 8. Three-dimensional probability density
distribution of collisions
524 M. Minami and R. Shoji

bf ðx; y; tÞ ¼ 1 X n
x  xi y  yi t  ti
2
Ks ; Kt ð6Þ
nhs ht i¼1 hs hs ht

where Kt is a kernel function based on the time deviations, ht is its window width. And
Ks, hs, and n are the same as in the previous section. The window width of time (ht) was
also adjusted by changing the value and it was judged that ht = 60 min. was most
appropriate to grasp the situation.
Figure 8 shows the estimation result by volume rendering with a three-dimensional
density distribution. If it is a sea area where a collision occurs regardless of the time
zone, the high-density area is shown in a columnar shape, but when a collision occurs
frequently in a specific time zone, it is shown in an island shape. In Fig. 8, since the
high-density area is represented not in the columnar shape but in the island shape, this
result shows that there are no sea areas where collision risk is always high. But some
sea areas the ship operators need to pay attention to collisions depending on the time.

4.3 Extraction of Vessel Close Points Using AIS Data

As mentioned in Sect. 4.1, since the densities in the unit area and the unit time were
low only with the collision point, it was difficult to grasp the collision occurrence
condition in detail. Therefore, it is said that there are many events that did not result in
accidents behind accidents [4], so we extracted the record the ship approached from the
AIS data and showed it as a probability density distribution like the collision point.
And then we examined whether it can be used for evaluating the collision risk.
The Method of Extraction. The AIS data includes information such as ship position,
SOG (Speed Over the Ground), heading, MMSI (Maritime Mobile Service Identity),
destination, and so on. Tracks are obtained by sorting data by ship by ship’s unique
information such as MMSI, then sorting by time. Conventional radar observation and
visual observation make it difficult to grasp movement of vessels over a long time in a
wide area and to obtain tracks from radar images, but by using AIS data, it became
easier to do it.
In this research, a truck was created from AIS data, and the approach situation was
extracted using the ship position, ship speed, and heading. The criteria for extraction of
the approach situation were CPA (Closest Point of Approach) analysis with TCPA
(Time to the Closest Point of Approach) less than 5 min and DCPA (Distance at the
Closest Point of Approach) less than 500 m. Currently, on board collision alarm criteria
are often less than 5 min for TCPA and 0.5 nm for DCPA. The collision alarm is issued
before becoming dangerous to prevent a collision. So, the criteria used in this study
have been more stringent. And the middle point of two ships which exceeded these
criteria was made a close point.
We used the data in Tokyo Bay for 7days from November 18 (Tuesday) to 24
(Monday) 2014 in JST (Japan Standard Time) obtained at Advanced Navigation
System in Tokyo University of Marine and Technology [5]. The number of vessels
subject to survey was 726, 761, 741, 764, 696, 667 and 733 on 18th, 19th, 20th, 21st,
22nd, 23rd and 24th, respectively. The average per day was 726.9, and the number of
vessels on Sunday on 23rd was the lowest among the data acquisition period.
 Estimation of the Collision Risk on Planned Route 525

In the extraction, the transmission interval of dynamic information such as the ship
position and speed of AIS data varies depending on the state and speed of the ship, so
we first asked for ship position every 5 min by the following procedure.
1. Sort data by ship with MMSI which is ship’s unique number,
2. Sort in time order and obtain for each ship track,
3. Calculate the ship position every 5 min by linear interpolation.
By the data organized at the
same time, TCPA and DCPA were
calculated and the close point was
extracted.
The Distribution of Ship Close
Points. Figure 9 shows the num-
ber of times of ship close points
extracted by time. The total num-
ber of points was 3564. Figure 10
shows the ship close point using
the spatiotemporal coordinates.
The symbols are color coded by
the encounter situation. Fig. 9. The number of ship close point by time
Ship approaches frequently
occurred at 6:00–8:00 and
16:00–18:00 in the same time
zone as the collision. From
Tokyo Port to Yokohama Port,
there were many ship close
points at 6:00–18:00. Also, at
Tokyo Port there were more
approaches in head on situation
than other areas, and at Yoko-
hama Port there were many
approaches in crossing situation.
In the vicinity of Uraga Suido
Trafic Route from Nakanose Fig. 10. Spatiotemporal coordinates of ship close points
Traffic Route, there were many
ship close points near 6:00 and 18:00, while there were few points around 12:00. In
addition, there were many cases of overtaking situation within the route, but there were
many approaches in head on situation near the intersection of Uraga Suido Traffic
Route and Nakanose Traffic Route. In this way, it was shown that the ship close point
was one of the features showing the characteristics of the sea area.
526 M. Minami and R. Shoji

4.4 Probability Density Distribution of Ship Close Points

The two-dimensional probability density distribution obtained from the ship close point
is shown in Fig. 11. High occurrence density was appeared all day near Uraga Suido
Traffic Route, and the density became high from Tokyo Port to Yokohama Port
depending on the time. Near Chiba and Ichikawa in Chiba Port, the occurrence density
was high in the morning, and in the afternoon the density was high even in the central
of the northern part of Tokyo Bay. Compared to the probability density distribution of
collision points, both of results, the occurrence density was high in the area from Tokyo
Port to Yokohama Port at 6:00–9:00, 9:00–12:00 and 15:00–18:00. On the other hand,
near Uraga Suido Traffic Route, the occurrence probability density of the approach was
high, but the one of collision was low.

Fig. 11. Ship approach occurrence probability density distributions by time

 Estimation of the Collision Risk on Planned Route 527

Figure 12 shows three-dimensional

probability density distribution of the
ship close point. The high-density area
was shown in an island shape, indicating
that the approaching situation was dif-
ferent depending on the time. The density
around Uraga Suido Traffic Route was
high throughout the day, but it was high
especially around 6:00 and around 18:00.
Due to the removal of the Dai3Kaiho,
which was a small artificial island, and
the dredging of the Nakanose Traffic Fig. 12. 3-D ship approach occurrence prob-
route have completed by 2008, traffic ability density distribution
flow near Dai2Kaiho was rectified but
the traffic volume remained high. The
frequency of collisions also increases as
the traffic volume increases with respect
to the width of the route [6, 7], and there
is also a questionnaire result that there
Dai2Kaiho[8]
are many near misses at the crossing of
the course. In other words, these sea
areas are places where danger of colli-
sion should be considered, it was not able
to evaluate by only the information of the
collision occurrence point. This result
shows that the ship close point extracted
from the AIS data can be a material to Fig. 13. Dai2Kaiho and Dai3Kaiho
evaluate the collision risk (Fig. 13).

5 Estimation of the Collison Risk

on Planned Route

5.1 Estimation Conditions

In the previous chapter, it was shown that it was necessary to
consider position and time to grasp the tendency of collision.
On the other hand, when planning a navigation plan, route,
departure time, speed and so on are decided, so navigation
position and time are determined. Therefore, we calculated
the collision occurrence density and approach occurrence
density encountered when the planned route and
departure/arrival time were decided.
The target ship is a domestic cargo ship, the Ship length
(Lpp) is 70.0 m, and the draft is 4.5 m. We considered the
following two kinds of routes. One is a route that a ship goes Fig. 14. Planned routes
528 M. Minami and R. Shoji

up north in Tokyo Bay and arrivals at Tokyo Port, and the other is a route that a ship
departure from Tokyo Port and goes south in Tokyo Bay. When navigating to go up
north in Tokyo Bay, if the Ship length is over 50 m and the water depth is less than
20 m, it is necessary to go through the Uraga Suido Traffic Route and Nakanose Traffic
Route. In addition, when navigating to go south in Tokyo Bay, we have to pass through
Nakanose west Sea Area and Uraga Suido Traffic Route. Therefore, northbound route
and southbound route are different. Planned routes are shown in Fig. 14, red line is
northbound route and blue one is southbound route. In addition, the traffic volume by
AIS is shown in the background. Ship speed was estimated by two kinds, one is set to
12 kt (6.19 m/s) for all route, the other is set to 15 kt (7.72 m/s) outside the traffic
route, because the upper limit of it in the traffic route is 12 kt.

Fig. 15. Time series variations of the normalized occurrence probability density (Northbound
route)

Fig. 16. Time series variations of the normalized occurrence probability density (Southbound
route)
 Estimation of the Collision Risk on Planned Route 529

5.2 Estimation Results

Figure 15 shows the time series variations of the normalized collision occurrence
density and approach occurrence density on northbound route and Fig. 16 shows them
on southbound route. The vertical axis shows the occurrence density in 10 levels, and
the horizontal axis shows the elapsed time from the start of calculation. As shown in
Fig. 15, the occurrence density of collision changes depending on the time. On the
other hand, the occurrence density of approach is large between 30 min and 90 min
after the start of the calculation, during which the ship navigates from the Uraga Suido
Traffic Route to the Nakanose Traffic Route. As shown in Fig. 16, the collision
occurrence density varies depending on the departure time. However, the level tends to
be large between 60 min to 90 min after the start of calculation sailing off Yokohama
Port. On the other hand, the approach occurrence density tends to be high-level around
120 min after the start of calculation sailing in the Uraga Suido Traffic Route.

(a) Northbound Route : Collsion points (b) Northbound Route : Ship close points
23:00 23:00
22:00 22:00
21:00 21:00
20:00 20:00
19:00 19:00
18:00 18:00
17:00 17:00
0.0- 16:00
0.0-
16:00
15:00 1.0- 15:00 1.0-
Arrival Time
Arrival Time

14:00 2.0- 14:00 2.0-

13:00 3.0-
13:00 3.0-
12:00 12:00
4.0- 4.0-
11:00 11:00
5.0- 10:00
5.0-
10:00
9:00 6.0- 9:00 6.0-
8:00 7.0- 8:00 7.0-
7:00 7:00 8.0-
8.0-
6:00 6:00
9.0 - 9.0 -
5:00 5:00
4:00 4:00
3:00 3:00
2:00 2:00
1:00 1:00
0:00 0:00
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%

(c) Southbound Route : Collsion points (d) Southbound Route : Ship close points
23:00 23:00
22:00 22:00
21:00 21:00
20:00 20:00
19:00 19:00
18:00 18:00
17:00 17:00
16:00
0.0- 16:00
0.0-
Departure Time
Departure Time

15:00 1.0- 15:00 1.0-

14:00 2.0- 14:00 2.0-
13:00 3.0- 13:00 3.0-
12:00 12:00
4.0- 4.0-
11:00 11:00
10:00
5.0- 10:00
5.0-
9:00 6.0- 9:00 6.0-
8:00 7.0- 8:00 7.0-
7:00 8.0- 7:00 8.0-
6:00 6:00
9.0 - 9.0 -
5:00 5:00
4:00 4:00
3:00 3:00
2:00 2:00
1:00 1:00
0:00 0:00
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%

Fig. 17. The ratio of the level on planned route

530 M. Minami and R. Shoji

Figure 17 shows the ratio of the level occupied by each route. Although the
magnitude of the level differed between the occurrence probability density obtained
from collision points and the one obtained from close points, the change with time
showed the same trend. Namely, on northbound route the level of occurrence density
encountered was small from 2:00 to 3:00, but it was large around 8:00 and in the
evening. On southbound route, the level of occurrence density encountered was large
from 4:00 to 8:00 and from 15:00 to 19:00.

Fig. 18. Time series variations of the normalized occurrence probability density (change ship
speed)

Table 1. Maximum level of probability density distribution of collisions

6:00 18:00 6:00 18:00
Arrival Arrival Departure Departure
Constant speed 3.97 9.71 5.47 7.04
Variable speed 4.49 9.78 5.65 6.81

Table 2. Maximum level of probability density distribution of ship approach

6:00 18:00 6:00 18:00
Arrival Arrival Departure Departure
Constant speed 6.58 2.31 8.33 7.69
Variable speed 6.63 2.38 8.56 8.16

Next, we obtained the occurrence probability density encountered when sailing at

15 kt on outside of the traffic route. Figure 18 shows the time series variations. Table 1
and Table 2 show the maximum value of the occurrence probability density obtained
from collision points and close points, respectively. There was also a difference in the
occurrence density encountered by changing the speed. We think that it is possible to
lay a navigation plan that reduces the collision risk encountered by adjusting the arrival
and departure time and ship speed.
 Estimation of the Collision Risk on Planned Route 531

5.3 Application to Existing Ship

Although it was shown that the occurrence proba-
bility density encountered is greatly different
depending on the time of departure and arrival, in
existing ships it is often not possible to change those
time from the relation of cargo handling. Figure 19
shows coloring with the level of normalized
occurrence probability density along the planned
route.
By showing to the operator the place where the
occurrence probability density is high and the time
of navigating the place, the operator can grasp the
characteristics of the planned route. Thereby we
believe that collision avoidance effect can be
obtained.

6 Conclusion

We analyzed the occurrence situation of collision Fig. 19. Collision risk on planned
related to merchant ship focusing the encounter route (southbound route)
situation in Tokyo Bay. There were many collisions
due to anchoring/drifting and crossing, and these
two types accounted for more than 70% of the total. Collision during anchoring or
drifting occurred frequently in anchorage and quay. Collisions in the crossing situation
frequently occurred off Yokohama and off Haneda, due to intersection of ship courses.
With the occurrence tendency by time, there was a peak in the number of collisions in
the morning and evening. In the morning, there was a relatively large number of
collisions in head on situation, and in the evening, there were many collisions in the
crossing situation.
In order to evaluate the collision risk of the target area, the collision points were
converted to a continuous distribution expressed by probability density using kernel
density estimation method. And then, we confirmed that the collision risk of the target
area is displayed in an easy-to-understand manner.
Since the number of occurrences of collision varies with time, the occurrence
distribution of collisions, which was often displayed in two dimensions, is displayed in
three dimensions using spatiotemporal coordinates. However, in the display of only the
collision points, the number of occurrences per unit area and unit time is small and it is
difficult to grasp the occurrence trend of the collision accident. Therefore, in order to
display the three-dimensional occurrence distribution in a more comprehensible man-
ner, the probability density distribution was displayed using the three-dimensional
kernel density estimation method. And the collision risk of the target area was eval-
uated. In addition, the ship close point extracted using the AIS data was displayed in
the spatiotemporal coordinates. We found that those points also indicated the feature of
the sea area. Then, we convert the close point of the ship to the probability density
532 M. Minami and R. Shoji

distribution by the kernel density estimation method and confirm that it indicates the
collision risk in the sea area clearly.
Using the probability density of both collision and approach, the risk of collision
was estimated along the planned route. And it showed that there is a big difference in
collision risk depending on departure and arrival time. The collision risk also changed
with a change in ship speed. In other words, if it is free to decide departure and arrival
times, it is possible to set a planned route with a smaller collision risk.
With regard to existing vessels, information such as the collision risk in the nav-
igation area targeted by this research is effective to prevent accidents for less experi-
enced operator, even if the skilled operator knows. Especially in accidents that tend to
occur repeatedly in the same place such as a collision, we thought that promoting
attention through information provision has a great effect of reducing accidents.

References
1. The Japan Transport safety Board: The Annual Report 2017, March 2019. http://www.mlit.
go.jp/jtsb/jtsbannualreport2017.html
2. The Japan Transport safety Board: Japan-Marine Accident Risk and Safety Information
System, March 2019. https://jtsb.mlit.go.jp/hazardmap/index_en.html
3. Silverman, B.W.: Density Estimation for Statistics and Data Analysis. Chapman and Hall,
London (1986)
4. Heinrich, H.W., Petersen, D., Roos, N.: Industrial Accident Prevention, 5th edn (1980). (trans.
Research Institute for Safety Engineering, KAIBUNDO, 1982)
5. Shoji, R., Kitazawa, F., Nishiyama, H.: Construction of advanced navigation system. In: 5th
International Conference on Emerging Trends in Engineering & Technology (ICETET 2012),
CD-ROM (2012)
6. Fujii, Y.: Effective collision diameter and collision rate of ship. J. Nautical Soc. Jpn 42, 1–8
(1969)
7. Itoh, H., Ishimura, E., Kudo, J., Mori, Y.: Estimation of ship encounter frequency at coastal
areas using AIS data. In: Proceedings of the Japan Society of Narval Architects and Ocean
Engineers, vol. 16, pp. 309–312 (2013)
8. Ministry of Land, Infrastructure, Transport and Tourism, Kanto Regional Development
Bureau, Tokyo Bay Waterway office, March 2019. https://www.pa.ktr.mlit.go.jp/wankou/
history/index4.htm
 Autonomous Route Planning for Ships
in Complex Waterways

Jinsong Xu1(&), Chunxiao Hou1, Rongwu Yang1,

and Ge (George) Wang2
1
Department of Naval Architecture and Ocean Engineering,
Shanghai Jiao Tong University, Shanghai, China
[email protected]
2
Seastel Marine System (USA) LLC, Houston, USA

Abstract. Route planning for ship navigation is a process to generate the

optimal route with the careful consideration of ship maneuvering constraints and
environmental conditions. This paper presents a study on a ship’s navigation
through a canal estuary area with the strong currents. An autonomous route
planning algorithm based on the modified RRT approach (rapidly-exploring
random tree) is developed to handle this increased level of complexity. A series
of navigation model tests were conducted to verify this route planning algo-
rithm. The optimal route generated from the proposed algorithm could ensure
safe navigation of the ship model through this canal estuary area without any
collisions with the dikes or the tidal barrier. Autonomous route planning built
upon scientific theories and operational experiences has the great advantage to
generate optimal routes even in complex waterways, which is possible to
minimize human errors in navigation decision.

Keywords: Autonomous route planning
Complex waterway
Optimal route

Modified RRT approach
Navigation model tests

1 Introduction

Route planning for ship navigation is a process to generate the optimal route with the
careful consideration of ship maneuvering constraints and environmental conditions.
The planned route consists of the waypoints, headings, and speed values for every
moment, which should be dynamically feasible for ship to follow the advised shipping
lane and avoid the static obstacles in a safe and efficient way. Traditionally, the task of
route planning has been heavily relied on the experiences of captain and mariners. But
in complex waterways with the restricted ship channels and harsh environmental
conditions, such traditional approach is not adequate to ensure risk-free navigation.
According to the “Annual Overview of Marine Casualties and Incidents 2017” [1],
there are 3145 maritime accidents reported by EU member states in 2016. Half of the
accidents are ship collisions and stranding, and more than 60% are caused by human
errors. Autonomous route planning built upon scientific theories and operational
experiences has the great advantage to generate optimal routes even in complex
waterways, which is possible to minimize human errors in navigation decision.

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 533–547, 2021.
https://doi.org/10.1007/978-981-15-4680-8_37
534 J. Xu et al.

There are two common approaches for autonomous route planning. The graph-
based approaches such as A* and its variants have been the most studied algorithms
[2, 3]. Their major challenge is the computation efficiency for real-time application.
With the increase in the dimensions of the state space and the complexity of the
environment, the computation time will increase significantly. The sampling-based
approaches such as rapidly-exploring random tree (RRT) and its variants [3] are rel-
atively new but more efficient. Their main advantage is the ability to easily integrate
dynamic constraints into the planning process, and efficiently generate a feasible route
in high-dimension state spaces [4]. The computation time will not increase significantly
with the environment complexity. However, the planned route from the conventional
RRT algorithm is feasible but not necessarily optimal, and sometimes even not smooth
enough due to random sampling procedure in the search [3].
This paper presents a study on a ship’s navigation through a canal estuary area with
the strong currents. A tidal barrier with the surrounding dikes is situated in the canal
mouth to prevent the strong tidal currents from flooding into the canal, which brings
extra difficulties to navigation. An autonomous route planning algorithm based on the
modified RRT approach is developed to handle this increased level of complexity. The
obstacle constraints, ship maneuvering constraints, strong current constraints, and route
optimality requirements could be satisfied simultaneously in the planning process. The
navigation model tests in a scaled waterway of canal estuary area are presented to
verify the proposed route planning algorithm. The planned optimal route could ensure
safe navigation of the ship model through this complex waterway without any colli-
sions with the dikes or the tidal barrier.
The study results presented in this paper demonstrate both the importance and
effectiveness of the proposed route planning algorithm, which is valuable for further
development of the autonomous route planning in complex waterways.

2 Autonomous Route Planning Algorithm

2.1 The Modular Architecture

As shown in Fig. 1, the autonomous route planning algorithm consists of the multiple
modules including the search guider on configuration space, the feasible route segment
generator, the currents effect adjuster, and the optimal route searcher. The obstacle
constraints, ship maneuvering constraints, strong current constraints, and route opti-
mality requirements are considered in different modules to ensure the optimal navi-
gation in complex waterways.

2.2 Search Guider on Configuration Space

Search guider is intended to find a set of collision-free paths considering only the static
obstacles between the starting point and goal point. The obtained path segments will
give guidance to the generation of the feasible route segments.
The search guider is based on the modified RRT approach (rapidly-exploring
random tree), which is an efficient algorithm to search non-convex, high-dimensional
 Autonomous Route Planning for Ships in Complex Waterways 535

Fig. 1. Modular architecture of the autonomous route planning algorithm

configuration spaces by randomly building a space-filling tree [3]. As illustrated in

Fig. 2, the modified RRT planning takes the current state as root node and uses Halton
random strategy to generate a sampling point wp in the unexplored search space. The
point wp is then connected to all existing tree nodes a, b, c of the RRT tree. All
connection lines which pass the collision detection are added into RRT tree to form the
multiple new branches, and the sampling point wp becomes a new node. This new node
is then connected to the goal point. If the connection line passes the collision detection,
it is also added as a RRT branch. Several collision-free path solutions from the root
node to the goal point are formed in Fig. 2. Repeating the above process, the RRT
nodes and branches are extended successively till a newly added node reaches the goal
point, and all possible collision-free path solutions are formed successfully.

Fig. 2. Illustration of search guider algorithm

536 J. Xu et al.

The feature of uniform random distribution in Halton random strategy enables the
RRT planning to explore the configuration space completely. Tracking the visited
sampling points in the form of newly added nodes could avoid infinite search loops in
the same state. Traversing all existing nodes for every sampling point could increase
the probability of containing the optimal path in the tree. The greedy algorithm of
connecting each new node with the goal point could greatly improve the efficiency of
path generation.

2.3 Feasible Route Segment Generator

Due to the weak maneuvering capability, ships could not follow the straight-line path
segments shown in Fig. 2. It is necessary to generate dynamically feasible route seg-
ments to replace the straight-line path segments between any pair of tree nodes.
In this paper, each feasible route segment is generated on the base of maneuver
automaton (MA) model in accordance with calm-water maneuvering equations and
velocity controller. MA model [5, 6] describes the vessel trajectory in terms of two
types of trajectory primitives. For the under-actuated ship with only yaw control and
surge control, the primitive trim is a finite-time steady cruising trajectory defined as (u,
r), which represents a combination of steady surge velocity u and steady yaw velocity
r. The trim library T is a trim set including all possible discrete velocity combinations
(u, r). The primitive maneuver is a finite-time transition trajectory from one trim (u, r)p
to another trim (u, r)q, which also represents a closed-loop step response of the velocity
controller from the initial state (u, r)p to target state (u, r)q. The maneuver library
M consists of all the maneuvers interconnecting any two different trims in library
T. The unknown state variables and uncontrolled sway motions in each primitive are
determined by ship maneuvering equations and vessel controller.
As shown in Fig. 3, the route segment from the initial state M0 to the target state Mf
can be expressed as a sequence of trajectory primitives. If the ship is specified to
approach the target state Mf in a steady straight-on way, the last leg of the primitive
sequence must be a straight-on trim. There must exist another steady turn trim in the
primitive sequence to complete course adjustment. In order to ensure the continuity of
position, heading, speed, and acceleration at any moment along the route segment, a
maximum of two maneuvers are also necessary in the primitive sequence to achieve
smooth switching between different trims. Therefore, the ship route segment from the
initial state M0 to the target state Mf could be generated by concatenating a maximum of
4 trajectory primitives: the first leg maneuver P1, the second leg turn trim with the
center O of turning circle, the third leg maneuver P3, and the last leg straight-on trim.
Selecting different trims from library T, a variety of different route segments could
be generated between two specified tree nodes. More primitives exist in library T and
M, more different route segments could be generated. Since the primitive trim and
primitive maneuver are all constructed in accordance with calm-water maneuvering
equations, the generated route segments naturally ensure the dynamical feasibility in
calm water.
 Autonomous Route Planning for Ships in Complex Waterways 537

Mf
12
straight-on trim
maneuver P3
turn trim
10 maneuver P1
ship
ship heading
ship course
8
turning circle
North(m)

O
6

M0

0
-4 -2 0 2 4 6 8
East(m)

Fig. 3. Generation of route segment from MA primitives

2.4 Currents Effect Adjuster

Since the canal estuary area always has the strong currents, the route segment generated
in calm water must be adjusted with currents effect.
As shown in Fig. 4, the calm-water route segment is discretized into a series of
waypoint P(n)static with a small time step Dt. If the current speed at this waypoint is
Up(n)current, and the ship is assumed to drift with the same speed Up(n)current for the
time interval Dt, then next waypoint in current P(n + 1)current could be estimated as
follows,

~
Pðn þ 1Þcurrent ¼ ~
PðnÞcurrent þ ½~
Pðn þ 1Þstatic  ~ ~PðnÞ
PðnÞstatic  þ U
Dt ð1Þ
current

Therefore, the calm-water route segment Pstatic would become the current-affected
route segment Pcurrent. If the starting points P(1)static and P(1)current are in the same
position, the original destination point in calm water PDes-static will shift to the new
destination point in currents PDes-current.
Going back to the route segment M0Mf in Fig. 3. The starting point P(1)static is set
on initial state M0, and the destination point in calm water PDes-static is set on the target
state Mf. The destination point in currents PDes-current must be shifted as shown in
Fig. 4. In the next iteration, the destination point in calm water PDes-static is adjusted
with the distance between Mf and PDes-current, and the destination in currents could be
updated again. Repeating the above iteration process, the final destination point in
currents PDes-current would match with the target state Mf. The corresponding current-
affected route segment M0Mf is the candidate segment for optimal route.
538 J. Xu et al.

Fig. 4. The adjustment of the route segment with currents effect

2.5 Optimal Route Searcher

For search of the optimal route, a cost function costR(x) is defined below to evaluate
the quality of each candidate route segment,

costRðxÞ ¼ a
tRðxÞ þ b
safetyRðxÞ ð2Þ

Here the first term represents economic cost, where tR(x) is the time duration of
route segment. The second term represents the safety cost, where safetyR(x) is the total
time duration of vessel invasion into the safety range of static obstacles. It takes infinity
when the candidate route segment would lead to collision with the static obstacles. The
weight coefficients a, b are adjustable parameters for balance consideration, which are
both set to be 0.5 in this research.
The complete search process for optimal route is based on the modified RRT
algorithm and illustrated in Fig. 5. The starting waypoint is taken as the root node, and
the target waypoint is taken as the destination node goal. There are already three
existing nodes a, b and c generated by RRT algorithm. In step 1, a new node d is
randomly generated in the unexplored search space based on Halton random strategy,
and then connected to the existing tree nodes b and c without collision. The branches
bd and cd constitute the straight-line path segments. In step 2, a set of current-affected
route segments between nodes c and d are generated from MA primitives. Based on the
cost function of each route segment, the optimal route segment with the minimal cost
value is selected to replace the straight-line path segment cd. In step 3, the path
segments bd and cd are all replaced by the optimal route segments. From the root node
to the new node d, two feasible routes are formed by concatenating the optimal route
segments. In step 4, after computing the route cost, one feasible route is selected as the
optimal route from root node to the new node d, and other routes are discarded. In step
5, the optimal route segment between the new node d and the destination node goal is
 Autonomous Route Planning for Ships in Complex Waterways 539

generated, and a feasible route from root node to the destination node goal is obtained.
In step 6, repeating the above process, more nodes e, f and more feasible routs are also
generated. Based on the total route cost, one feasible route with the minimal cost value
is finally selected as the optimal route from root node to the destination node goal.

Fig. 5. Illustration of complete search process for optimal route

3 Planning and Testing Results

3.1 Canal Estuary Description
This paper presents a study on a ship’s navigation through a canal estuary area shown
in Fig. 6. The green part is the restriction area including both the land area and shallow
water area, which is not allowed for ship navigation. A tidal barrier with the sur-
rounding dikes is situated in the canal mouth to prevent the strong tidal currents from
flooding into the canal, which brings extra difficulties to navigation. Two typical
current conditions are speed 1.50 m/s from north and speed 1.32 m/s from south.
The typical type of cargo ship on this waterway is 67.5 m in length and 10.8 m in
breadth. The maximum ship speed in calm water is 7.0 kn. The navigation line shown
in Fig. 6 is planned in accordance with the ship maneuvering performance in calm
water, but may not be feasible in strong current conditions.
540 J. Xu et al.

Fig. 6. Illustration of the canal estuary area

3.2 Model Test Setup

The 1:60 scaled model of canal estuary area was created for navigation tests as shown
in Fig. 7. The model area is around 50 m long and 35 m wide. The current generating
system was used to generate two typical current conditions, and the model-scale current
velocity distribution results are presented in Fig. 8 and Fig. 9.

Fig. 7. The 1:60 scaled model of canal estuary area

 Autonomous Route Planning for Ships in Complex Waterways 541

The autonomous ship model shown in Fig. 10 consists of the following modules:
(i) condition detection module to receive the real-time position data from the indoor
optics positioning system; (ii) condition analysis module to estimate the real-time
heading and speed; (iii) route planning module to specify the navigation route from the
starting point to goal point; (iv) action control module for the accurate tracking of the
planned route. The system framework of the autonomous ship model is illustrated in
Fig. 11.

Fig. 8. The model-scale current velocity distribution for heading currents

542 J. Xu et al.

Fig. 9. The model-scale current velocity distribution for following currents

Fig. 10. The autonomous ship model in navigation tests

 Autonomous Route Planning for Ships in Complex Waterways 543

Fig. 11. The system framework for autonomous ship model

3.3 Autonomous Route Planning Results

The autonomous route planning algorithm is applied to generate the optimal route in
this complex waterway. The locations of the dikes and tidal barriers, the restrictions of
land part and shallow water part, and current velocity distribution are fed into the
algorithm. The planned optimal routes for heading current condition and following
current condition are presented in Fig. 12 and Fig. 13, where the green part is
restriction area. They are different from the navigation line shown in Fig. 6.

Fig. 12. The planned optimal route in heading currents

544 J. Xu et al.

Fig. 13. The planned optimal route in following currents

3.4 Navigation Test Results

The first navigation test is in heading current condition. The navigation line shown in
Fig. 6 is selected as the planned route, and the actual trajectory of the ship model is
presented in Fig. 14. The ship model is very hard to turn the course in beam currents,
which results in the overshot to the upstream. Obviously, the navigation line in
accordance with the maneuvering performance in calm water is not feasible for tracking
in strong currents. The actual trajectory drifts away from the planned route and takes a
big turning in beam currents. The navigation process is dangerous and inefficient.
The second navigation test is also in heading current condition. The optimal route
shown in Fig. 12 is selected as the planned route, and the actual trajectory of the ship
model is presented in Fig. 15. Obviously, the optimal route generated from the
autonomous route planning algorithm is feasible for tracking. The navigation process is
safe and efficient.
The third navigation test is in following current condition. The optimal route shown
in Fig. 13 is selected as the planned route, and the actual trajectory of the ship model is
presented in Fig. 16. Obviously, the optimal route generated from the autonomous
route planning algorithm is feasible for tracking. The navigation process is safe and
efficient.
 Autonomous Route Planning for Ships in Complex Waterways 545

Fig. 14. The actual trajectory in heading currents by tracking the navigation line

Fig. 15. The actual trajectory in heading currents by tracking the planned optimal route
546 J. Xu et al.

Fig. 16. The actual trajectory in following currents by tracking the planned optimal route

4 Conclusions

This paper presents a study on a ship’s navigation through a canal estuary area with the
strong currents. A tidal barrier with the surrounding dikes is situated in the canal mouth
to prevent the strong tidal currents from flooding into the canal, which brings extra
difficulties to navigation. An autonomous route planning algorithm based on the
modified RRT approach is developed to handle this increased level of complexity. The
obstacle constraints, ship maneuvering constraints, strong current constraints, and route
optimality requirements could be satisfied simultaneously in the planning process.
A series of navigation model tests were conducted to verify this route planning
algorithm. A ship model with full autonomous functions was devised and a scaled
waterway of canal estuary area was created. The tests demonstrate that the optimal
route generated from the autonomous route planning algorithm could ensure safe
navigation of the ship model through this canal estuary area without any collisions with
the dikes or the tidal barrier. The selected route away from the optimal route may not be
feasible for tracking in strong currents, and the navigation process is usually dangerous
and inefficient.
The testing results demonstrate both the importance and effectiveness of optimal
route planning in complex waterways. Autonomous route planning built upon scientific
theories and operational experiences has the great advantage to generate optimal routes
even in complex waterways, which is possible to minimize human errors in navigation
decision.
 Autonomous Route Planning for Ships in Complex Waterways 547

References
1. European Maritime Safety Agency: Annual Overview of Marine Casualties and Incidents
2017, pp. 15–41 (2018)
2. Campbell, S., Naeem, W., Irwin, G.W.: A review on improving the autonomy of unmanned
surface vehicles through intelligent collision avoidance maneuvers. Annu. Rev. Control 36,
267–283 (2012). https://doi.org/10.1016/j.arcontrol.2012.09.008
3. Lavalle, S.M.: Planning Algorithms, 1st edn. Cambridge University Press, Cambridge (2006)
4. Chiang, H.L., Tapia, L.: COLREG-RRT: A RRT-based COLREGS-compliant motion planner
for surface vehicle navigation. IEEE Robot. Autom. Lett. 3, 2024–2031 (2018). https://doi.
org/10.1109/LRA.2018.2801881
5. Greytak, M.B., Hover, F.S.: Robust motion planning for marine vehicle navigation. In:
International Offshore and Polar Engineering Conference, Vancouver, vol. 2, pp. 399–406
(2008). https://doi.org/10.1.1.475.9757
6. Frazzoli, E., Dahleh, A., Feron, E.: Real-time motion planning for agile autonomous vehicles.
J. Guid. Control Dyn. 25, 116–129 (2002). https://doi.org/10.1109/ACC.2001.945511
Ocean Engineering
 Hydrodynamic Analysis of Offshore
Aquaculture Platform and Its Mooring Design

Jun Yu(&), Xiaoming Cheng, and Xiaofeng Wu

China Ship Scientific Research Center, Wuxi, Jiangsu, China

[email protected]

Abstract. Offshore aquaculture platforms, or fish cages, are increasingly

becoming popular in recent years in the effort of meeting the world’s demand for
seafood products. As a result, these platforms are being deployed further off-
shore in order to avoid causing pollution to near shore waters by densely
populated platforms. On the other hand, the dimensions of the platforms are also
becoming larger, reaching over 100 m in diameter in the recent designs. As the
aquaculture platforms are typically of frame type structures consisting of slender
beams which support the net around the frame, the hydrodynamic properties of
the platform is quite complex and yet is very important to the adequate design of
the platform and its mooring system as any significant under design or over
design of the platform and mooring system can incur serious economic conse-
quences. In this paper, a method based on the conventional potential flow theory
and Morison equation is proposed for the assessment of the hydrodynamic loads
on the aquaculture platforms including its frame and net. This method can take
into account the sheltering effect due to the surrounding net or beams but
neglects the energy dissipation due to the viscosity of the fluid. A Morison
model for an aquaculture platform is constructed and a mooring system is
designed in this paper followed by the time domain analysis to assess the
hydrodynamic forces, motion responses and mooring system performance in
various environmental conditions. The numerical results indicate that the motion
of the platform and the mooring tension can meet the design requirements in
survival environmental conditions, and the proposed method provides a useful
and efficient tool for the evaluation of the performances of the aquaculture
platforms.

Keywords: Aquaculture platform
Hydrodynamics
Morison elements

Mooring

1 Introduction

Overfishing in the offshore areas and the impact of offshore pollution have led to an
increasingly serious fishery resource reduction in most coastal countries, many tradi-
tional fishing grounds are gradually disappearing. The deterioration of the offshore
environment makes it urgent for all countries in the world to conduct comprehensive
and systematic research, utilize offshore fishery technology, develop offshore fishery
resources and drive the economic development. As a result, the fishery mode is
gradually changing from the traditional marine fishing to the development of fish

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 551–563, 2021.
https://doi.org/10.1007/978-981-15-4680-8_38
552 J. Yu et al.

farming or aquaculture, and the most widely used equipment for this is aquaculture
platform which is characterized by a cylindrical shaped frame structure surrounded by
fishing net.
The unique type of floating structure presents a new challenge to the researchers in
creating an adequate hydrodynamic model of the platform, which can be used to
calculate the hydrodynamic forces, motion responses with acceptable accuracy and
efficiency when the platform is exposed to waves and current. This issue also has a
significant impact on the design of mooring systems for aquaculture platforms as the
performance of a mooring system is heavily dependent on the hydrodynamics of the
floater. In the past two decades, many studies have been carried out in hydrodynamic
modelling of aquaculture platforms, including fishing net, and their mooring systems.
A model of a single net panel [1] was developed with the hydrodynamic coefficients for
each net element prescribed based on empirical data. The dynamic model of a 3D net
structure [2] was proposed and the structural forces were calculated by assuming that
each element consists of six nonlinear springs. Colbourne et al. [3] found that the
location of the mooring attachment and the style of the mooring could be adapted to
reduce the wave-induced motions of the fishing cage system. Colbourne et al. [4]
designed a new accessory of mooring system to reduce the deformation of marine
fishery cage. They also [5] developed a numerical model for irregular sea states to
simulate a single point mooring cage in an unsheltered open sea. The researchers
designed a mooring system consisting of three legs for a gravity-type cage, which is
made up of a taut vertical chain and an angled rope [6]. Mooring systems for wide
range of marine ranching were also designed [7], and the authors found it difficult to
achieve a satisfactory position keeping performance for all the cages of the offshore
aquaculture platforms. Siddiqui et al. [8] studied the motions of offshore cages and
predicted the mooring chain force by numerical simulation. Mikkola et al. [9] pointed
out that the design of moorings was one of the main technical challenges for offshore
fish farming in previous projects especially in deeper water areas. Some researchers
designed an innovative offshore aquaculture cage [10] and analyzed its hydrodynamic
behavior under different environmental conditions to prove its better seakeeping per-
formance. The authors showed that single and multiple-point mooring systems [11]
were vitally important to resist the excessive motions of aquaculture platforms for the
safety of fishes in the cage and demonstrated some existing mooring examples.
However, most of the previous work focused on relatively small fishery cages whose
hydrodynamic behavior differed from large aquaculture platforms.
This paper presents the study on a hydrodynamic modelling method and mooring
system design for an aquaculture platform with a diameter of 90 meters. A method
based on the conventional potential flow theory and Morison equation is proposed for
the assessment of the hydrodynamic loads on the aquaculture platforms including its
frame and net. In order to improve its stability and bearing capacity, a mooring system
is designed and time domain analysis performed to assess the hydrodynamic forces,
motion responses and mooring system performance in various environmental condi-
tions. The results show that the method proposed in this paper is relatively simple and
efficient with reasonable accuracy, and can be adopted in the initial design and per-
formance assessment of the cage type aquaculture platforms and their mooring systems.
 Hydrodynamic Analysis of Offshore Aquaculture Platform 553

2 Method of Hydrodynamic Analysis

2.1 Prototype Aquaculture Platform
The aquaculture platform studied in this paper has a shape of regular hexagonal prism with
a diameter of 90 m and height of 38.5 m, deployed in an offshore area with a water depth
of 50 m. A sketch of the platform is shown in Fig. 1. In addition to the frame structures,
there is a cylindrical buoyancy tank at the bottom of the platform with a diameter of 40 m
and a height of 6.2 m. There is also a cylindrical structure at the top of the platform for
housing the mechanical and electrical equipment necessary for the operation of the
platform. The main parameters of the aquaculture platform are listed in Table 1.

Fig. 1. Structure of aquaculture platform

2.2 Hydrodynamic Model of Aquaculture Platform

The main structure of an aquaculture platform consists of slender beams whose
diameters are in the order of 1–3 m, far smaller than wave length, which implies that
the wave diffraction effect is negligible and the wave force on the structure can be
calculated using Morison equation.
The equation below gives the expression of Morison equation for the hydrody-
namic force calculation for beam elements:

1
  : :
dF ¼ qDCd uf  us uf  us  þ qACm uf qAðCm  1Þ us ð1Þ
2
Where D is the diameter of the beam, uf is the velocity of fluid particle, us represents
the velocity of the beam itself, A is the cross section area of the beam, q is the density
of fluid, Cm is inertia coefficient which equals to 1 + Ca , where Ca is the added mass
coefficient of the beam, and Cd is drag coefficient of the beam.
554 J. Yu et al.

Table 1. Principal parameters of aquaculture platform

Items Working condition Survival condition
Displacement (T) 11810.9 11339.2
Draft (m) 37 32
Longitudinal location of the gravity center (m) 0 0
Transverse location of the gravity center (m) 0 0
Vertical location of the gravity center (m) 10.13 11.49
Longitudinal radius of inertia (m) 32.73 28.87
Transverse radius of inertia (m) 32.59 28.62
Yaw radius of inertia (m) 18.63 18.77

The hydrodynamic force calculation and motion analysis of the aquaculture plat-
form in the specified environmental conditions are conducted in time domain using an
ANSYS hydrodynamic analysis software AQWA. For each environmental condition, a
three hour time domain simulation is performed and the maximum force and mean
force on aquaculture platform are extracted and evaluated. It is worth mentioning that
the dimension of the buoyancy tank at the bottom of the platform is not small enough
and the Morison equation will not be applicable, instead a 3-D diffraction and radiation
theory is used for the calculation of the hydrodynamic forces on the tank.

2.3 Simulation of Fishing Net

An aquaculture platform is covered with synthetic fibre fishing net, sometimes with a
metal wire core to enhance its strength. As the diameter of the fibre wires of the fishing
net is in the order of millimeters, Morison equation is applicable to the calculation of
the hydrodynamic forces on the fishing net. The Morison elements are arranged in the
similar pattern as the fishing net, but with much bigger or coarser mesh so that sig-
nificantly fewer meshes are used in the Morison model than the actual fishing net, this
will reduce the complexity of the Morison model generation and the computation time.
In order to maintain the hydrodynamic properties of the actual net, the hydrodynamic
coefficients of the Morison elements have to be adjusted. On the other hand, the total
volume of the Morison elements is kept identical to that of the actual net so that the
buoyancy can be kept unchanged. In order to take into account the shielding effect of
the fishing net, extra Morison elements in the direction perpendicular to the net surface
are introduced. The parameters of the Morison model and the prototype fishing net are
listed in Table 2. This will result in a Morison model with its elements arranged in
three directions as shown in Fig. 2.
As the geometry properties of the Morison model are not identical to the prototype
net, the drag and wave inertia coefficients of the Morison model should be adjusted on
the basis of those for the prototype net so that the drag and wave inertia forces on the
Morison model will remain the same as for the prototype net. The way these coeffi-
cients are calculated for the Morison model is explained below.
 Hydrodynamic Analysis of Offshore Aquaculture Platform 555

Table 2. Basic geometry parameter of aquaculture platform

Items Prototype net Morison model
Mesh size (m) 0.0235 5
Diameter of element (m) 0.003 0.0399
Length of elements per unit area of net (m/m2 ) 85 12
Projected area of elements per unit area of net (m2 /m2 ) 0.255 0.255
Volume of elements per unit area of net (m3 /m2 ) 0.000602 0.000602

Fig. 2. 3D morison model of fishing net

First, the drag coefficient (Cd ) is calculated in the following equations:

q q 85
Fdx ¼ Cdxw   lx  dw ¼ 0:6    dw ð2Þ
2 2 2
q q
Fdz ¼ Cdzw   lz  dw ¼ 1:0   85  dw ð3Þ
2 2

For Morison model in Fig. 2:

q q
Fdx ¼ Cdxm   lxm  dm þ Cdzm   lzm  dm
2 2
ð4Þ
q 5 q 2
¼ Cdxm    dm þ Cdzm    dm
2 25 2 25
q q 10
Fdz ¼ Cdzm   lzm  dm ¼ Cdzm    dm ð5Þ
2 2 25

Where Cdxw , Cdyw and Cdzw are the prototype drag coefficients, Cdxm , Cdym , Cdzm are
the drag coefficients of Morison elements. When assigning the values for Cdxw and
Cdyw , the shielding effect can be taken into consideration. This effect is caused by the
upstream elements altering the flow field in the downstream region, and thus the drag
force on the downstream elements will be different from the upstream ones. In this
556 J. Yu et al.

study, reduced drag coefficient Cd is used for the downstream elements to account for
the shielding effect.
The inertia coefficients are calculated using the following equations:

Fax ¼ ð1 þ Caxw ÞqVdxw ¼1:5qVdxw ð6Þ

Faz ¼ ð1 þ Caxw Þq  2Vdxw ¼4qVdxw ð7Þ

For Morison elements:

Fax ¼ ð1 þ Caxm ÞqVdxm þ ð1 þ Cazm ÞqVdzm ð8Þ

Faz ¼ ð1 þ Caxm ÞqðVdxw þ Vdyw Þ ð9Þ

Where Caxw , Cayw and Caxw are the prototype inertia coefficients, Caxm , Caym , Cazm
are equivalent drag coefficients of Morison elements.

Table 3. The drag and inertia coefficients of Morison elements

Classification of Morison element Coefficient Value
Main structure of aquaculture platform Cd 1.0
Ca 1.0
Tangential direction in cage net Cdx , Cdy 16
Cax , Cay 1.4
Vertical direction in cage net Cdz −16
Caz −2.5

Based on the equivalence of prototype and Morison model coefficients, the above
equation can be solved and coefficients for Morison model obtained. The detailed
coefficients of all the Morison elements are shown in Table 3.

3 Numerical Simulation of Aquaculture Platform

3.1 Environmental Condition

The design environmental conditions for aquaculture platform are presented in Table 4,
in which the wave direction and current direction are defined as the directions the
waves and currents are going towards. Considering the symmetric configuration of the
aquaculture platform and the arrangement of mooring system below, only two repre-
sentative wave and current directions, i.e. 0° and 90° are considered. The significant
wave height is based on the measurement in the target area for the deployment of the
platform. As for peak spectral period, it is related to the significant wave height and is
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
in the range [12] of 13  Hs to 30  Hs empirically according to the energy
distribution of the wave spectrum.
 Hydrodynamic Analysis of Offshore Aquaculture Platform 557

Table 4. Designed ocean environmental conditions of aquaculture platform

Items Significant Peak spectral Wave Current Current Wind Wind
wave height Hs period Tp (s) direction velocity direction (m/s) direction
(m) (m/s)
Working 5.1 8.1, 9.6 12.4 0 1.1 0 24.6 0
condition 5.1 8.1, 9.6 12.4 30 1.1 30 24.6 30
Survival 7.3 9.4, 11.4, 0 1.8 0 32.3 0
condition 13.2
7.3 9.4, 11.4, 30 1.8 30 32.3 30
13.2

3.2 Complete Model of Aquaculture

With the creation of the diffraction model for the buoyancy tank, the Morison models
for both the frame of the platform and the fishing net, the whole aquaculture platform
model for hydrodynamic and motion analysis is completed. A coordinate system is also
defined with the origin at the mean water surface and z-axis upward. The aquaculture
platform will provide about 200000 cubic meter space for the growth of specially
selected fishes, bringing enormous economic benefits. The structure of the platform is
strengthened by a number of inclined struts and vertical columns. Figure 3 shows the
complete hydrodynamic model of the aquaculture platform.

Fig. 3. Complete aquaculture platform

558 J. Yu et al.

3.3 Mooring System Design

As the environmental conditions that the aquaculture platform will expose to can be
very harsh, the mooring system for the position keeping of the platform has to be
adequately designed to withstand the extreme wind, wave and current and also to
constrain the motion of the platform for the safety of the fish. Based on the above
hydrodynamic model of aquaculture platform, a hybrid mooring system is designed in
which each mooring line consists of a segment of synthetic fibre rope at the upper
section and a segment of steel mooring chain at the lower section. This type of mooring
composition has the advantage of more elasticity and hence lower extreme tension. The
layout of the mooring system is shown in Fig. 4. The mooring lines are numbered
counter-clockwise. There are 6 groups of mooring lines connected to separate columns
with an angle of 60° between the adjacent mooring groups, and each group contains
eight mooring lines. The environment direction is such defined that 0° is for wave and
current travelling towards the positive X-axis and the 90° for going towards positive Y-
axis. In order for the mooring system to function adequately as designed in the
specified environmental conditions, it is necessary to delve into the mooring properties
and behaviors of the synthetic fiber rope and the chain in complicated ocean envi-
ronment to determine the composition of the mooring lines. After a number of opti-
mization iterations the final mooring composition is obtained which is comprised of
synthetic fibre rope-chain1-chain2 (See. Fig. 5 and Table 5). The total length of each
mooring line is 295 m with a mooring radius (projected horizontal distance from
fairlead to anchor) of 290 m. This hybrid mooring line composition provides the
mooring system with an adequate stiffness so that the platform motion will be con-
strained under severe environmental conditions without causing excessive mooring line
tension.

mooring point on platform

mooring point on seabed

line 1
0° direction

line 8

line 9
30° direction

line 16
90° direction

Fig. 4. Arrangement of mooring system

 Hydrodynamic Analysis of Offshore Aquaculture Platform 559

Fig. 5. Composition of mooring line

Table 5. The detailed mechanical properties of mooring line

Subsection Length Diameter Wet weight Stiffness Breaking load
(m) (mm) (kg/m) (N) (N)
Synthetic fibre 28 181 −2 2.900E6 6.100E6
rope
Chain1 77 105 196.6 9.410E8 7.840E6
Chain2 190 93 149.6 7.380E8 6.440E6

4 Mooring System Performance

4.1 Mooring Intact Condition
The survival environmental condition is considered for the approval of the validity of
the mooring system. This condition includes a current speed of 1.8 m/s and an irregular
sea described by JONSWAP wave spectrum with a significant wave height of 7.3 m.
The wind is excluded from this study as it is not considered important to the platform
due to its small area exposed to wind. The main results are presented below and more
detailed results are listed in Figs. 6, 7, 8 and 9 for the highly loaded mooring groups.
From these results it can be observed that:

350 125
wave peried 9.4s wave peried 9.4s
wave peried 11.4s wave peried 11.4s
300 wave peried 13.2s wave peried 13.2s
100
250
Maximum tension(T)

75
Mean tension(T)

200

150 50

100
25
50

0 0
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
serial number of mooring line serial number of mooring line

Fig. 6. Maximum mooring tension in 0° Fig. 7. Mean mooring tension in 0° direction

direction
560 J. Yu et al.

300 100
wave peried 9.4s wave peried 9.4s
wave peried 11.4s wave peried 11.4s
250 wave peried 13.2s wave peried 13.2s

75
200
Maximum tension(T)

Mean tension(T)
150 50

100
25
50

0 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
serial number of mooring line serial number of mooring line

Fig. 8. Maximum mooring tension in 30° Fig. 9. Mean mooring tension in 30° direction
direction

(1) The mooring lines will be subjected to highest tensions in head sea (0°) than in
others directions. Although in head sea, the mooring tensions are more evenly
shared by the most tensioned lines than in other directions (Fig. 6 and 7), the
horizontal excursion of the platform is exactly in line with the direction of the
mooring lines, which results in higher mooring tension for the same movement of
the platform.
(2) The highest tension, under the significant wave height of 7.3 m and current speed
of 1.8 m/s in the 0° and 30° directions, can reach 265 tons (line 4 and 5) and 235
tons (lines 8), respectively. These results indicate that even for the structures with
high porosity, the hydrodynamic force can still be significant and lead to high
mooring tension. This is largely due to the shallow water effect which exacerbates
the hydrodynamic force and horizontal notion of the platform. Despite this, the
mooring strength still meets the design requirement with a minimum tension safety
factor of 2.3.
(3) From Fig. 6 and Fig. 8, it can be seen that the maximum mooring tension increases
with the increase of wave period. This is because the aquaculture platform moves
more under the action of long waves in shallow water. While Fig. 7 and Fig. 9
show that the mean mooring tensions are almost the same under different wave
periods. This may be an indication that the mean hydrodynamic force on the
aquaculture platform mainly comes from the current force and the variation of the
mean wave drift force with wave period in the survival condition is relatively
unimportant.

Table 6. Motion of aquaculture platform

Surge (m) Sway (m) Heave (m) Roll (m) Pitch (m) Yaw (m)
0° direction Min 3.987 0 −28.71 0 −0.6124 0
Max 11.53 0 −24.95 0 6.058 0
Mean 6.754 0 −26.4 0 1.108 0
30° direction Min 3.302 1.701 −28.46 −3.3 −0.6206 −0.2604
Max 9.813 5.506 −24.91 0.199 5.29 0.0185
Mean 5.802 3.178 −26.33 −0.6707 0.8572 −0.049
 Hydrodynamic Analysis of Offshore Aquaculture Platform 561

The motion of aquaculture platform is listed in Table 6. It is shown that the min-
imum and maximum motions are not change greatly from the mean location.

4.2 One Line Damaged Condition

It is required by classification societies that a mooring system should be able to survive
a situation when one mooring line is damaged in a survival environmental condition.
For the present mooring system, line 1 and line 4 are broken respectively for 0°
environment direction, and line 1 and 8 are broken respectively for 30° environment
direction. The main results for the one line damaged cases are presented below and
more detailed results are listed in Figs. 10, 11, 12 and 13. The results show that:
(1) With one line damaged, the mooring system can still effectively restrict the
movement of the platform for various wave and current directions and severities.
(2) The mooring tension becomes asymmetric as a result of the asymmetric mooring
system when one line is damaged. Generally speaking, mooring lines adjacent to
the damaged line will be subjected to higher tension than in the intact conditions.
The maximum mooring tension, under the significant wave height of 7.3 m and
current speed of 1.8 m/s in the 0° direction with line 1 broken and line 4 broken
can reach 288 tons (lines 2, 3, 4) and 285 tons (line 5) respectively. While the
maximum mooring tension, under the significant wave height of 7.3 m and current
speed of 1.8 m/s in the 30° direction with line 1 broken and line 8 broken can reach
241 tons (lines 8) and 251 tons (line 7) respectively. Comparing to line intact
condition, the maximum tension rises from 265 tons to 288 tons for 0° direction, a
8.7% increase. And the maximum tension rises from 235 tons to 251 tons for 30°
direction, an increase of 6.8%. Despite the increase of the maximum tensions in the
line damaged conditions, a minimum safety factor of 2.1 is still achieved for the
mooring system.
In general, the highest line tension is still far lower than the breaking load of lines
and therefore the mooring system presented in the paper is a feasible design.

350 350
wave peried 9.4s wave peried 9.4s
wave peried 11.4s wave peried 11.4s
300 wave peried 13.2s 300 wave peried 13.2s

250 250
Maximum tension(T)
Maximum tension(T)

200 200

150 150

100 100

50 50

0 0
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
serial number of mooring line serial number of mooring line

Fig. 10. Line 1 failure in 0° direction Fig. 11. Line 4 failure in 0° direction
562 J. Yu et al.

300 300
wave peried 9.4s wave peried 9.4s
wave peried 11.4s wave peried 11.4s
250 wave peried 13.2s 250 wave peried 13.2s

200 200
Maximum tension(T)

Maximum tension(T)
150 150

100 100

50 50

0 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
serial number of mooring line serial number of mooring line

Fig. 12. Line 1 failure in 30° direction Fig. 13. Line 8 failure in 30° direction

5 Conclusions and Recommendations

Numerical modelling and analysis have been conducted to assess the hydrodynamic
forces on an aquaculture platform and its safety in different environmental conditions.
A mooring system is also designed for its position keeping and to further strengthen its
safety in extreme environmental conditions. As the aquaculture platforms are typically
of frame type structures consisting of slender beams which support the net around the
frame, the hydrodynamic properties of the platform is quite complex and yet is very
important to the adequate design of the platform and its mooring system as any sig-
nificant under design or over design can incur serious economic consequences.
In this paper, a method based on the conventional potential flow theory and
Morison equation is proposed for the assessment of the hydrodynamic loads on the
aquaculture platforms including its frame and net. This method can take into account
the sheltering effect due to the surrounding net or beams but neglects the energy
dissipation due to the viscosity of the fluid. A Morison model for an aquaculture
platform is constructed and a mooring system is designed in this paper followed by the
time domain analysis to assess the hydrodynamic forces, motion responses and
mooring system performance for both mooring intact condition and mooring line
damaged condition for various wave and current directions and severities. The sim-
plified numerical model proposed in this paper can give a reasonable solution for the
hydrodynamic analysis of aquaculture platform. In addition, this model has the merits
of simple calculation and high calculation efficiency, which can be used to quickly
evaluate the influence of the complex ocean environments in the preliminary design
phase of aquaculture platforms.

Acknowledgements. The work summarized in this paper was partially supported by the Min-
istry of Industry and Information Technology with the research project in the fields of high-tech
ships ([2016]22).
 Hydrodynamic Analysis of Offshore Aquaculture Platform 563

References
1. Lader, P., Fredheim, A., Lien, E.: Dynamic behavior of 3D nets exposed to waves and
current. In: The 20th International Conference on Offshore Mechanics and Arctic
Engineering (2001)
2. Lader, P.F., Enerhaug, B., Fredheim, A., et al.: Modelling of 3D net structures exposed to
waves and current. In: 3rd International Conference on Hydroelasticity in Marine
Technology. Department of Engineering Science, The University of Oxford, Oxford, UK,
pp. 19–26 (2003)
3. Colbourne, D.B., Allen, J.H.: Observations on motions and loads in aquaculture cages from
full scale and model scale measurements. Aquacult. Eng. 24(2), 129–148 (2001)
4. Huang, C.C., Tang, H.J., Liu, J.Y.: Modeling volume deformation in gravity-type cages with
distributed bottom weights or a rigid tube-sinker. Aquacult. Eng. 37(2), 144–157 (2007)
5. Huang, C.C., Tang, H.J., Wang, B.S.: Numerical modeling for an in situ single-point-
mooring cage system. IEEE J. Oceanic Eng. 35(3), 565–573 (2010)
6. DeCew, J., Fredriksson, D.W., Bugrov, L., et al.: A case study of a modified gravity type
cage and mooring system using numerical and physical models. IEEE J. Oceanic Eng. 30(1),
47–58 (2005)
7. DeCew, J., Celikkol, B., Baldwin, K., et al.: Assessment of a mooring system for offshore
aquaculture. World Aquac. 43(3), 32 (2012)
8. Siddiqui, M.A., Nagarajan, V., Mukherjee, C.K.: Modeling the forces and motions of a
single point mooring marine aquaculture cage. In: The Twenty-fourth International Ocean
and Polar Engineering Conference. International Society of Offshore and Polar Engineers
(2014)
9. Mikkola, E., Heinonen, J., Kankainen, M., et al.: Multi-platform concepts for combining
offshore wind energy and fish farming in freezing sea areas: case study in the gulf of bothnia.
In: ASME 2018 37th International Conference on Ocean, Offshore and Arctic Engineering.
American Society of Mechanical Engineers, p. V006T05A029 (2018)
10. Jurado, A., Sánchez, P., Armesto, J.A., et al.: Experimental and numerical modelling of an
offshore aquaculture cage for open ocean waters. In: ASME 2018 37th International
Conference on Ocean, Offshore and Arctic Engineering. American Society of Mechanical
Engineers, p. V07AT06A053 (2018)
11. Buck, B.H., et al.: State of the art and challenges for offshore integrated multi-trophic
aquaculture (IMTA). Front. Marine Sci. 5, 165 (2018)
12. DNV-RP-C205, Environmental Conditions and Environmental Loads (2010)
 Prediction of Extreme Nonlinear
Hydrodynamic Responses and Mooring Line
Loads of Floating Offshore Structures

Dong-Hyun Lim1, Yonghwan Kim2(&), and Seung-Hoon Lee2

1
Hyundai Heavy Industry, Dong-gu, Ulsan 44032, South Korea
2
Seoul National University, Gwanak-gu, Seoul 08826, South Korea
[email protected]

Abstract. An accurate estimation of extreme hydrodynamic responses and

mooring line loads is important for the design of floating offshore structures.
However, this task is complicated due to nonlinearity of the dynamic system as
well as the non-Gaussian statistical behavior of the extreme response. In this
study, a new procedure for the prediction of extreme offset of the platform and
the mooring line tension is suggested. The procedure is composed of following
three steps. Step 1 is the prescreening analysis for identifying the simulation
conditions that are likely to give the large responses. In this step, a simplified
analysis method is used for rapid estimation of design values. Next, the extreme
offset of the platform is calculated by design wave analysis on the prescreened
simulation conditions. By adopting the design wave analysis concept, the
required time length to calculate the extreme value is greatly reduced, which
enables us to simulate much more conditions in the nonlinear time-domain
simulation. And finally, the full-length platform-mooring line coupled time-
domain analysis is performed for the verification of the design wave analysis
result. The accuracy of analysis method in each step is validated through
numerical simulations, and the procedure is applied to a semi-submersible
platform for the applicability test.

Keywords: Platform offset
Slow-drift motion
Mooring line tension
Design

wave analysis
Analysis procedure

1 Introduction

For the design of mooring lines in the floating offshore structures, accurate prediction
of the extreme line tension, which is closely related to the extreme offset of the
platform, is essential. However, estimating the design load on the mooring lines can be
difficult due to following factors. First, the environmental loads and the dynamic
system that governs the behavior of the platform and mooring lines are nonlinear. In
extreme conditions, the wave, wind and current loads acting on the platform and the
mooring lines are highly nonlinear, and the nonlinear restoring and damping forces
may affect the global behavior of the platform significantly. Therefore, complex non-
linear time-domain simulation is required to solve the problem. Many studies have
been conducted regarding this subject, which have shown that the numerical simulation

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 564–578, 2021.
https://doi.org/10.1007/978-981-15-4680-8_39
 Prediction of Extreme Nonlinear Hydrodynamic Responses 565

can accurately model the platform-mooring line coupled system in extreme conditions
[1–3]. Secondly, statistical behavior of the hydrodynamic response of the platform in
those extreme conditions is non-Gaussian. Especially, the slow-drift motion, which
occupies a large part of the platform horizontal offset, was observed to have a strongly
non-Gaussian characteristic in past studies [4, 5]. Hence, the time-domain simulation
should be performed for long time to obtain converged probability distribution of the
nonlinear responses, which increases the computational burden. In addition, the
number of simulation conditions is generally very large considering all possible
combinations of environmental and operational conditions, which is the third factor that
complicates the problem.
For these reasons, the hydrodynamic analysis for the extreme mooring line load is
usually conducted through a procedure that leads from a relatively simple analysis to a
precise one. Figure 1 is a diagram that summarizes the analysis procedure used in
current industry mentioned in API-RP-2SK [6]. According to this document, the
analysis procedure currently used consists of three major steps. The first step is the
frequency-domain radiation/diffraction analysis to calculate the RAO (response
amplitude operator) and QTF (quadratic transfer function) of the wave exciting force
and platform motion. In the second step, the prescreening analysis using a relatively
simple analysis scheme, such as the de-coupled frequency-domain analysis, coupled
frequency-domain analysis, or the semi-coupled time-domain analysis, is performed to
select the extreme simulation conditions among a number of conditions at initial stage.
And in the final step, the platform-mooring line coupled time-domain analysis is
performed for the selected extreme conditions to estimate the precise extreme value.
In other words, the current analysis procedure can be characterized by the sim-
plified analysis scheme used in the prescreening step. However, no matter which
analysis scheme is used in the prescreening analysis, there is uncertainty about the
accuracy due to the linearization of some or all of the nonlinearities in the dynamic
system. This may affect the reliability of selected extreme conditions and ultimately
increase the number simulation conditions that require time-domain coupled analysis.
That is, the dilemma of satisfying both high accuracy and fast computation speeds in
prescreening analysis can be regarded as a challenge to be solved to improve the
efficiency of the analysis procedure.
In this study, a new procedure for the extreme mooring line load analysis for
floating offshore structures is suggested to improve the efficiency of current industrial
practice. The suggested procedure can be characterized by the design wave analysis for
the extreme horizontal offset of the platform. Design wave analysis is a concept to
replace full-length time-domain simulations, but it can also be used for secondary
prescreening of extreme conditions due to considerably shorter calculation time than
full length analysis. The advantage of prescreening using design wave analysis is that it
can consider all nonlinearity in the prescreening step, securing higher accuracy than
existing prescreening methods. Details are explained in the following sections.
566 D.-H. Lim et al.

Fig. 1. Diagram of the current industrial practices of the dynamic analysis procedure for floating
offshore structures (API-RP-2SK)

2 Analysis Procedure

Figure 2 is a schematic diagram showing the overall procedure of the extreme mooring
line load estimation proposed in this study. The input data used for the analysis consist
of information about platform and mooring system, environmental condition and
operating condition, and transfer functions calculated through frequency-domain
radiation/diffraction analysis. The first step of the procedure is to perform the de-
coupled frequency-domain analysis for initial prescreening. Then, in the second step,
the extreme tension is estimated by the design wave analysis for the selected simulation
conditions. In the final step, the full-length time-domain coupled analysis is performed
for top few simulation conditions that are expected to give the greatest extreme
response through design wave analysis. If the accuracy of the design wave analysis
results is sufficiently guaranteed, the last step can be omitted (Path 2 in Fig. 2). Also, if
the design wave analysis is not applicable, the full-length time-domain coupled analysis
 Prediction of Extreme Nonlinear Hydrodynamic Responses 567

Fig. 2. Flow diagram of the suggested procedure for the dynamic analysis of floating offshore
structures

can be performed right after the de-coupled frequency-domain analysis (Path 3 in

Fig. 2). Details of each step are as follows.

2.1 Step 1: De-coupled Frequency-Domain Analysis

The de-coupled frequency-domain analysis of this study consists of static heading
angle analysis, static horizontal offset analysis, and de-coupled quasi-static analysis.
The static heading angle analysis is to calculate the equilibrium heading angle between
the environmental yaw moment and the restoring moment by mooring lines. In the
following static horizontal offset analysis and de-coupled quasi-static analysis, the
platform heading angle is assumed to be fixed at this equilibrium angle. In the static
horizontal offset analysis, the horizontal drift position where the wave drift force, static
wind and current loads and the mooring line restoring force are in equilibrium is
calculated. Also, the linearized restoring coefficients in platform’s longitudinal and
transverse directions at the equilibrium position are calculated for the de-coupled quasi-
static analysis.
In the de-coupled quasi-static analysis, the maximum horizontal offset of the
platform is calculated using the frequency-domain transfer functions, and the maximum
tension is estimated using the static restoring curve. The maximum horizontal offset is
calculated for the wave-frequency component and the low-frequency component sep-
arately. The variance of the wave-frequency motion is given by
Z 1  2
 ð1Þ 
r2WF;i ¼ Hi ðxÞ SðxÞdx; i ¼ 1; 2 ð1Þ
0

ð1Þ ð1Þ
where H1 and H2 represent the surge and sway motion RAOs, respectively, and S(x)
is the power spectrum of irregular incident waves. The variance of the low-frequency
motion is calculated using the power spectrum of low-frequency exciting forces.
568 D.-H. Lim et al.

The low-frequency exciting forces consist of the second-order difference-frequency

wave load and the dynamic wind load. The second-order difference-frequency wave
load spectrum is given by
Z 1  2
 ðÞ 
F;i ðxÞ ¼ 8
Swave Sðx þ lÞSðlÞHF;i ðx þ l; lÞ dl; i ¼ 1; 2 ð2Þ
0

ðÞ
where HF;i represents the difference-frequency wave load QTF in surge (i = 1) and
sway (i = 2) directions. The dynamic wind load spectrum can be calculated as follows,
under the assumption that the time-varying component is small compared to the mean
wind load.
 
Swind
wind 2 Swind
F;i ðxÞ ¼ qair Awind Ui i ðxÞ; i ¼ 1; 2 ð3Þ

iwind and Swind

In Eq. (3), U i ðxÞ represent the mean wind speed and wind spectrum in
surge (i = 1) and sway (i = 2) directions, respectively. Finally, the variance of the low-
frequency motion is calculated using the approximate solution as follows [7].
Z 1
F;i ðxÞ þ SF;i ðxÞ
Swave wind
r2LF;i ¼ dx; i ¼ 1; 2 ð4Þ
0 ½ci  ðm þ ai Þx2 2 þ b2i x2

In Eq. (4), ai, bi and ci represent the added mass, damping, and restoring coeffi-
cients in surge (i = 1) and sway (i = 2) directions, respectively.
Once the variance of each component of the horizontal motion is obtained, the
maximum value can be estimated using appropriate probability model. Finally, when
the maximum horizontal offset is calculated, the maximum tension is estimated by
static analysis of the mooring line at the position. More details on the de-coupled
frequency-domain analysis in this study can be found in [8].

2.2 Step 2: Design Wave Analysis for the Extreme Horizontal Offset
In this step, the extreme tension in the mooring line is estimated through design wave
analyses for the prescreened simulation conditions. The target response of the design
wave analyses is the slow-drift motion induced by difference-frequency wave load,
which occupies a dominant portion of the total horizontal offset of the platform.
Because most of the existing design wave methods are defined only for the first-order
hydrodynamic responses, they cannot be applied to the slow-drift motion which is of
second-order to the wave amplitude. Hence, Lim and Kim [9] developed a new design
wave method for the second-order hydrodynamic responses, and verified the applica-
bility of the method to the slow-drift motion problem. Following is the brief intro-
duction of the theory in [9].
 Prediction of Extreme Nonlinear Hydrodynamic Responses 569

The nonlinear equation of slow-drift motion is given as follows.

ð2Þ
ðM þ MA Þ€x þ BWD ðtÞx_ þ CH x þ f V ðx;
_ tÞ þ f M ðx; x;
_ €x; tÞ ¼ f W ðtÞ ð5Þ

In Eq. (5), x is the nonlinear slow-drift motion vector, and M, MA, and CH are the inertia
matrix, zero-frequency added mass matrix, and the hydrostatic restoring coefficient
matrix, respectively. Also, BWD and f(2)W represent the time-varying wave-drift damping
coefficient matrix and the second-order difference-frequency wave load respectively.
The viscous drag force acting on the platform is represented as fV, and fM stands for the
mooring line force calculated from nonlinear simulation of the mooring line dynamics.
Equation (5) is a nonlinear equation of motion, which is solved by time-domain
platform-mooring line coupled analysis. However, if the left-hand side of the equation
is linearized with proper restoring and damping coefficients, the equation of motion can
be solved in frequency-domain, which leads to the definition of slow-drift motion QTF.
The linearized equation of motion and the slow-drift motion QTF can be represented as
Eq. (6) and (7), respectively.

ð2Þ
ðM þ MA Þ€xL þ Bx_ L þ ðCH þ CM ÞxL ¼ f W ðtÞ ð6Þ
h i1
Hðx1 ; x2 Þ ¼ ðx1  x2 Þ2 ðM þ MA Þ þ iðx1  x2 ÞB þ ðCH þ CM Þ HF ðx1 ; x2 Þ
ð7Þ

In Eq. (6) and (7), B and CM represent the linearized damping coefficient and the
mooring line restoring coefficient, respectively. Also, HF represents the QTF of the
difference-frequency wave load. The slow-drift motion from the linearized equation of
motion is denoted by xL to distinguish from x in Eq. (5), and is termed hereafter the
linearized slow-drift motion.
The basic assumption of the design wave method in [9] is that the irregular incident
wave that induces a certain percentile of maximum linearized slow-drift motion also
induces the maximum nonlinear slow-drift motion of same percentile. The main
advantage of utilizing the linearized slow-drift motion is that its probability distribution
can be obtained analytically through eigenvalue analysis [4, 5]. For the surge motion in
long-crested sea, the incident wave elevation η(t) and the linearized slow-drift surge
motion xL(t) are given by

X
N   qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
gðtÞ ¼ Aj cos xj t þ ej ; Aj ¼ 2Sðxj ÞDx ð8Þ
j¼1

N X
X N    
xL ðtÞ ¼ Aj Ak H1 ðxj ; xk Þ cos ðxj  xk Þt þ ðej  ek Þ þ H1 ðxj ; xk Þ ð9Þ
j¼1 k¼1

where Aj, xj, and ej are the amplitude, frequency, and phase of individual discretized
wave components respectively, and H1 and H1 are the surge motion complex QTF and
its phase, respectively.
570 D.-H. Lim et al.

The eigenvalue problem for the probability distribution of xL(t) is then defined as
follows [4].
Z 1
K ðx1 ; x2 Þwj ðx2 Þdx2 ¼ kj wj ðx1 Þ ð10Þ
1
 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Sðjx1 jÞSðjx2 jÞH1 ðx1 ; x2 Þ; x1 x2  0
K ðx1 ; x2 Þ ¼ ð11Þ
0; x1 x2 \0

In Eqs. (10) and (11), kj and wj are the eigenvalue and eigenfunction, respectively,
of the Hermitian kernel K(x1, x2). Then, the first four statistical moments of xL(t) are
given by [10]

X
N X
N
1 X N
1 X N
mxL ¼ kj ; rxL ¼ 2k2j ; a3; xL ¼ 8k 3
; a 4; xL ¼ 3 þ 48k4j ð12Þ
j¼1 j¼1
r3xL j¼1 j r4xL j¼1

where mxL, rxL, a3,xL and a4,xL are the mean, standard deviation, skewness, and the
kurtosis of xL(t), respectively. Using these four statistical moments, the probability
distribution of xL(t) and its maximum value for certain period can be readily obtained
using the Hermite-moment method. Details can be found in [9–11].
Once the maximum value of the linearized slow-drift motion is obtained, the design
waves can be generated by forcing the maximum value to occur at the desired instant.
The basic concept of generating the design waves is to find the phase set {ej} that
induces xL(t) to have a peak of the specified percentile of its maximum value distri-
bution at the specified instant t = t0. The specified instant t0 should be set as small as
possible to minimize the computational cost, but it should be preceded by several
cycles to suppress the memory effect. The governing equations for {ej} with a peak of
the target value x0 at t = t0 are given by

f1 ðeÞ x ðt Þ  x0 0
fðeÞ ¼ ¼ L 0 ¼ ð13Þ
f2 ðeÞ x_ L ðt0 Þ 0

where e is the vector representation of the unknowns {ej}. Substituting Eq. (9) into
Eq. (13), the governing equations are alternatively represented as follows.
8
> P N 
N P    
>
> Aj Ak H1 ðxj ; xk Þ cos ðxj  xk Þt0 þ ðej  ek Þ þ H1 ðxj ; xk Þ ¼ x0
<
j¼1 k¼1
>
>
N 
N P
P    
>
: ðxj  xk ÞAj Ak H1 ðxj ; xk Þ sin ðxj  xk Þt0 þ ðej  ek Þ þ H1 ðxj ; xk Þ ¼ 0
j¼1 k¼1

ð14Þ

This system of equations is underdetermined, hence, infinite solutions exist. To solve

this nonlinear, underdetermined system of equations, the Newton-Raphson method
with the Lagrange-multiplier method is adopted. Details can be found in [9].
 Prediction of Extreme Nonlinear Hydrodynamic Responses 571

Because the peak value at t = t0 is the only constraint in the solution process above,
the design wave is not uniquely defined, and the induced profiles of linearized slow-
drift motion may significantly differ from each other. To reduce the difference, the
concept of the most likely extreme response profile method by [12] has been extended
to the second-order problem by [9] as follows.
According to the New Wave model, the expected profile of a standard normal
process u(t) near a local maximum u0 at t = t0 is given by [13]

_ 0 Þ ¼ 0 ¼ u0 Ru ðt  t0 Þ
E ½uðtÞjuðt0 Þ ¼ u0 ; uðt ð15Þ

where Ru(t) is the autocorrelation function of u(t). Also, according to Winterstein [11],
a non-Gaussian process, which in this case is xL(t), can be transformed into a standard
normal process u(t), and their autocorrelation functions satisfy following equation.
" #
2
X
4
RxL ðtÞ ¼ ðjrxL Þ Ru ðtÞ þ ðn  1Þ!c2n fRu ðtÞgn1 ð16Þ
n¼3

In Eq. (16), RxL(t) is the autocorrelation function of xL(t), and j and cn are the
coefficients defined by Winterstein [11]. RxL(t) can be obtained analytically using the
power spectrum of xL(t) given as follows.
Z 1
SxL ðxÞ ¼ 8 jH1 ðx  l; lÞj2 Sðjx  ljÞSðjljÞdl ð17Þ
0

Therefore, Ru(t) can be readily obtained by solving the cubic equation of Eq. (16).
Then, the final form of the average profile of xL(t) around its peak value x0 at t = t0 is
given by [9]

xML ðtÞ  E ½xL ðtÞjxL ðt0 Þ ¼ x0 ; x_ L ðt0 Þ ¼ 0

P4
ð18Þ
 mxL þ jrxL u0 Ru ðt  t0 Þ þ cn Hen1 ðu0 Ru ðt  t0 ÞÞ
n¼3

where Hen is the nth order Hermite polynomial. Details and application examples of the
design wave method for the slow-drift motion can be found in [9].

2.3 Step 3: Full-Length Coupled Time-Domain Analysis

After the prescreening analysis is complete, the full-length coupled time-domain
analysis is performed for the top few simulation conditions to estimate the extreme line
tension precisely. The simulation should be performed for 5 to 10 different random
phases of incident waves for one simulation condition to consider the randomness of
the maximum value. In this study, the coupled analysis program SML developed by
MIT has been used. Details of the program can be found in Kim and Sclavounos [2].
572 D.-H. Lim et al.

3 Procedure Application
3.1 Computational Model
The computational model used for the simulations is the semi-submersible platform
shown in Fig. 3. The platform is composed of 4 cylindrical columns and rectangular
pontoons, and 20 chain-polyester-chain mooring lines divided into 4 groups are
attached at the bottom of each column. The installation water depth is 3,048 m. Table 1
summarizes the details of the platform and the mooring lines.

2 1 20
19
3 18
4 17
5 2000
16

Y (m)
W E 0

S
6 15
7 14 -2000
8 13
9 12
10 11
-2000 0 2000
X (m)

Fig. 3. Semi-submersible model used in the simulations (left) and the mooring system
configuration (right)

Table 1. The main details of the platform and the mooring system
Cylinder diameter 29.0 m
Pontoon length 43.5 m
Draft 40.0 m
Displacement 143,528 ton
KG 33.0 m
Pre-tension in the mooring line 1,258 kN
Total length of one mooring line 4267.1 m
Surge natural period 231.1 s

3.2 Environmental Conditions

The environmental conditions used in the analysis are metocean data from the Gulf of
Mexico (GoM) Walker Ridge region. The metocean data, which is not provided in this
paper due to confidentiality, is given as 10-year, 50-year and 100-year return period
values of significant wave height, spectral peak period, wind speed and current speed in
8 directions of 45º interval. The simulation conditions have been defined by combining
the metocean data of different components in the following manner. If one component
corresponds to a return period of 100 years, one of the remaining two components is set
to 10-year, and the other component is set to 10-year or 50-year return period, so that
 Prediction of Extreme Nonlinear Hydrodynamic Responses 573

excessively strong environmental conditions is excluded. The correlation between

incident directions of the respective components was not taken into account in the
combination of the simulation conditions. In this case, the number of total simulation
conditions is 8,704. The JONSWAP spectrum of the peak enhancement factor 3.3 was
used for the wave spectrum, and the NPD wind spectrum was used for wind spectrum.
The duration of each environmental condition was set to 4.5 h.

3.3 Application Example

Figure 4 shows the results of the de-coupled frequency-domain analysis for 8,704
environmental conditions. In the figure, ‘ ’ indicates the mean drift position in each
condition and ‘ ’ symbol indicates the maximum drift position. Table 2 summarizes the
top 10 environmental conditions based on the maximum drift distance. A noticeable
feature of the top environmental conditions is that under most conditions the three load
components act in a similar direction. Also, most of the conditions have 100-year wave
or current, and the wind load appears to have minor effect on the extreme horizontal
offset of the platform.

Fig. 4. De-coupled frequency-domain analysis result for 8,704 environmental conditions

(diamond: mean offset, square: maximum offset)

Table 2. Top 10 critical environmental conditions in terms of maximum offset

Rank Wave Wave Wind Wind Current Current
direction return direction return direction return period
period period
1 S 50-year SE 10-year SE 100-year
2 S 100-year SE 10-year SE 50-year
3 S 50-year E 10-year SE 100-year
(continued)
574 D.-H. Lim et al.

Table 2. (continued)
Rank Wave Wave Wind Wind Current Current
direction return direction return direction return period
period period
4 S 100-year E 10-year SE 50-year
5 S 50-year S 10-year SE 100-year
6 S 100-year S 10-year SE 50-year
7 S 50-year SW 10-year SE 100-year
8 S 100-year SW 10-year SE 50-year
9 S 50-year SE 50-year SE 50-year
10 S 50-year NE 10-year SE 100-year

30 30
100 Slow-drift sway 100 Slow-drift sway
Wave elevation Wave elevation
20 20
80 80
Slow-drift sway (m)

Slow-drift sway (m)

Wave elevation (m)

Wave elevation (m)

60 10 60 10

40 0 40 0

20 20
-10 -10

0 0
-20 -20
0 100 200 300 400 500 0 100 200 300 400 500
t (s) t (s)

Fig. 5. Examples of design waves (left: 50-year wave, right: 100-year wave)

Next, let’s look at the results of the design wave analysis performed on the top 10
environmental conditions identified as the most dangerous through prescreening. The
coupled time-domain analysis program SML was used, and five design waves were
generated for each condition. The target value used in the generation of the design
wave was the 50%-percentile of the maximum value, and the target instant t0 was set to
450.0 s which is about twice the natural period of the slow-drift motion. Figure 5
shows an example of a design wave generated for a wave period of 50 years and 100
years in the south direction.
When the wind and current loads are considered together in the design wave
analysis, the target instant t0 should be away from the transient region due to static
wind and current load. To confirm this, the transient region test was performed for the
case where only the static wind and current loads exist. The test condition was set to
100-year current load and 50-year wind load from Southeast. Figure 6 shows the test
result. From this figure, it is observed that the transient region ends at around
t = 300.0 s. That is, the target instant of 450.0 s can be judged as an appropriate value.
Figure 7 shows the maximum horizontal offset and the maximum mooring line
tension obtained from the design wave analyses for the top 10 environmental condi-
tions. The abscissa of the graph represents the rank of each environmental condition
from the prescreening analysis. From Fig. 7, it is observed that the rank of the
 Prediction of Extreme Nonlinear Hydrodynamic Responses 575

50

40

Slow-drift sway (m)

30

20

10

0
0 100 200 300 400 500
t (s)

Fig. 6. Transient region test for steady wind and current loads

environmental conditions calculated from de-coupled frequency-domain analysis does

not remain identical in the design wave analyses. A noticeable feature of the design
wave analyses result is that the environmental conditions with 50-year wave and 100-
year current give more or less larger values than the opposite cases. This is because the
current load on the mooring lines was not properly considered in the de-coupled
frequency-domain analysis. However, in the comparison between the environmental
conditions with the same return periods, it can be seen that the ranking calculated from
the de-coupled frequency-domain analysis is maintained to some extent in the design
wave analyses. Also, it can be seen that the rank 1 condition from the de-coupled
frequency-domain analysis gave the highest value also in the design wave analysis.
Meanwhile, in all 10 environmental conditions, the maximum tension was observed in
line 11 (see Fig. 3).
As a final step of the analysis procedure, 4.5-h full-length coupled time-domain
analysis was performed for the top environmental condition from the design wave
analysis. The analysis was performed for 15 different random seeds for the incident
wave elevation and the time-varying wind speed. Figure 8 shows an example of the
time series of the surge motion and the mooring line tension in line 11. Table 3
compares the design wave analysis and the full-length simulation results. Since the
design wave analysis was performed for the 50%-percentile of the maximum value, the
full-length simulation result in Table 3 represents the median of the 15 repetitive test
results. From Table 3, it can be seen that the design wave analysis result is about 3–4%
larger than the full-length simulation results in both the maximum offset and maximum
line tension, which can be judged to be of reasonable accuracy.
576 D.-H. Lim et al.

180
Maximum offset (m)

160

140

120

100
1 2 3 4 5 6 7 8 9 10
Index
8000
Maximum tension (kN)

7000

6000

5000

4000
1 2 3 4 5 6 7 8 9 10
Index

Fig. 7. Maximum horizontal offset (upper) and the maximum mooring line tension (lower)
estimated by the design wave analysis in the prescreened environmental conditions

160
140
120
Platform sway (m)

100
80
60
40
20
0
0 2000 4000 6000 8000 10000 12000 14000 16000
Time (s)

8000
7000
Tension in line 11 (kN)

6000

5000
4000
3000
2000
1000
0
0 2000 4000 6000 8000 10000 12000 14000 16000
Time (s)

Fig. 8. Example time series of the platform sway motion (upper) and the tension in line 11
(lower) in the most extreme environmental condition
 Prediction of Extreme Nonlinear Hydrodynamic Responses 577

Table 3. Comparison of the design wave analysis and the full-length simulations on the 50%-
percentile of maximum responses
Design wave Full-length Ratio
analysis simulations
Maximum offset 157.9 m 152.8 m 103.5%
Maximum tension 7080 kN 6840 kN 103.3%

4 Concluding Remarks

In this study, a new hydrodynamic analysis procedure for estimating extreme mooring
line tension has been suggested. The procedure is composed of 3 main steps: the de-
coupled frequency-domain analysis, the design wave analysis for the slow-drift motion,
and the full-length coupled time-domain analysis.
To verify the applicability of the suggested procedure, a deep water semi-
submersible platform was selected as the test model. Totally 8,704 environmental
conditions were simulated through the analysis procedure. It was confirmed that the
design wave analysis can accurately predict the maximum response when wind and
current loads exist. Also, the procedure was identified to be able to predict the maxi-
mum line tension efficiently for a number of environmental conditions.

Acknowledgments. This work was supported under the framework of international cooperation
program managed by the National Research Foundation of Korea (2016K2A9A1A06929705).

References
1. Ormberg, H., Larsen, K.: Coupled analysis of floater motion and mooring dynamics for a
turret-moored ship. Appl. Ocean Res. 20, 55–67 (1998)
2. Kim, S., Sclavounos, P.D.: Fully coupled response simulations of theme offshore structures
in water depths of up to 10,000 feet. In: Proceedings of the 11th International Offshore and
Polar Engineering Conference, vol. 3, pp. 457–466 (2001)
3. Kim, M.H., Koo, B.J., Mercier, R.M., Ward, E.G.: Vessel/mooring/riser coupled dynamic
analysis of a turret-moored FPSO compared with OTRC experiment. Ocean Eng. 32, 1780–
1802 (2005)
4. Naess, A.: The statistical distribution of second-order slowly-varying forces and motions.
Appl. Ocean Res. 8(2), 110–118 (1986)
5. Lim, D.H., Kim, Y., Kim, T.: Stochastic analysis of the second-order hydrodynamic
quantities for offshore structures. Appl. Ocean Res. 64, 15–30 (2017)
6. API-RP-2SK, Design and Analysis of Stationkeeping Systems for Floating Structures.
American Petroleum Institute, USA
7. Faltinsen, O.M.: Sea Loads on Ships and Offshore Structures. Cambridge University Press,
Cambridge (1990)
8. Lim, D.H., Kim, Y., Kim, T.: Prescreening of environmental conditions for prediction of
severe operation condition of offshore structures. J. Adv. Res. Ocean Eng. 1(4), 252–267
(2015)
578 D.-H. Lim et al.

9. Lim, D.H., Kim, Y.: Design wave method for the extreme horizontal slow-drift motion of
moored floating platforms. Appl. Ocean Res. 71, 48–58 (2018)
10. Winterstein, S.R., Ude, T.D., Kleiven, G.: Springing and slow-drift responses: predicted
extremes and fatigue vs. simulation. In: BOSS 1994 7th International Conference on the
Behaviour of Offshore Structures, Massachusetts, USA (1994)
11. Winterstein, S.R.: Nonlinear vibration models for extremes and fatigue. J. Eng. Mech. 114
(10), 1772–1790 (1988)
12. Adegeest, L.J.M., Braathen, A., Løseth, R.M.: Use of non-linear sea loads simulations in
design of ships. In: Proceedings of Practical Design of Ships and Mobile Units, pp. 53–58
(1998)
13. Tromans, P.S., Anaturk, A.R., Hagemeijer, P.: A new model for the kinematics of large
ocean waves – application as a design wave. In: Proceedings of the 1st International Offshore
and Polar Engineering Conference, vol. 3, pp. 64–71 (1991)
 Design Automation of Mooring Systems
for Floating Structures

Bo Wu(&), Xiaoming Cheng, Ying Chen, Xinyun Ni, and Kai Zhang

China Ship Scientific Research Center, No. 222, East Shanshui Road,
Wuxi, China
[email protected]

Abstract. Mooring systems are widely used for the station keeping of floating
structures in order to maintain its position at a specified location with allowable
offset limits under given environmental conditions. A well designed mooring
system should have good station keeping ability, satisfy the relevant safety
requirements under both operational and survival conditions, low construction,
installation and maintenance costs etc. However the mooring design process
involves large number of mooring layout and parameter variations, and subse-
quent numerical simulations. The whole process is prolong and labor intensive,
and is heavily dependent on the designers’ experience. An attempt has been
made to develop an automated methodology and software tool that can be used
to generate an initial and feasible mooring system design automatically on the
basis of the user defined boundary conditions and constraints. This paper pre-
sents the initial progress achieved in the study of the automated mooring design
methodology. The Genetic Algorithm for optimization is adopted in the study to
obtain the optimal solution of the objective function. In order to prove the
efficiency and applicability of the automated optimal design procedure, a single
point mooing system for a FPSO and a spread mooring system for a semi-
submersible platform are designed using the proposed method and then analyzed
in time domain in different environmental conditions. The results indicate that
the proposed automated design method can be used to create a feasible mooring
system design which satisfies the requirements on mooring line tension and
structure offset.

Keywords: Mooring system
Automated design
Optimization
Genetic

algorithm

1 Introduction

Offshore floating structures are widely used in the offshore oil and gas industry,
including exploration and production platforms and FPSOs etc. [1–3], and their
capability of position keeping is mainly provided by mooring systems. At present, the
design of a mooring system is largely dependent on engineer’s experience, and the
balance between various design parameters is difficult to grasp. The mooring design
process requires repeated iterations, parameter adjustments and extensive numerical
simulations, which is time consuming and inefficient. It is in this perspective that an
effort is made in this study to develop an automated mooring design tool which can

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 579–594, 2021.
https://doi.org/10.1007/978-981-15-4680-8_40
580 B. Wu et al.

Fig. 1. Offshore floating structures

automatically generate an initial mooring system option which satisfies the user defined
requirements and boundary conditions (Fig. 1).
Single point mooring systems are widely used in floating production and oil storage
system (FPSO). The environmental forces of wind, wave and current make the floating
structure rotate around the mooring point, thus reducing the environmental load. Spread
mooring systems are commonly used mooring mode in offshore engineering, which is
adopted by semi-submersible platforms and Spar platforms, and some FPSOs as well
(Fig. 2).

Fig. 2. Spread mooring system and single-point mooring system

Research on mooring system design has been performed in the past. The prelim-
inary scheme of mooring system is designed based on a number of parameters, such as
the properties of the floating structure, operation requirements, water depth, seabed
slope, and environmental conditions, etc. A feasible and acceptable design of a
mooring system should meet the following key requirements: the maximum offset must
be within the specified limit, the maximum mooring tension must have a safety factor
not smaller than the value required by classification organizations. In general, the
radius of the mooring system and the general configuration are decided first, and on this
basis the mooring designers will select various mooring properties and parameters,
include mooring composition, material, length and sizes. Static, quasi-static or dynamic
analysis is performed to analyze the performance of the mooring system designed [4–
6]. The performance and general quality of the designed mooring system is heavily
dependent on the designer’s experience. Some researchers realized the importance of
design automation of mooring systems and made some efforts in this field. Different
parameters of a mooring system, such as mooring radius, mooring line length and the
 Design Automation of Mooring Systems for Floating Structures 581

layout of mooring lines, were studied to determine the optimization objectives and
constraints, and relevant algorithms were applied to optimize the mooring scheme [7,
8]. In this paper, an automated mooring system design method is presented which
combines the mooring analysis program, mooring material database and a generic
algorithm for the optimization of the mooring design.

1.1 Static Method

Static design is performed at the initial stage of the mooring system design.
Load/excursion characteristics for a single line and the whole mooring system are
established, ignoring the fluid dynamic forces on the lines. Static analysis is performed
by utilizing an algorithm to calculate the forces exerted on the platform from each
mooring line, given the line end-point coordinates on the platform and seabed together
with the density, length and elasticity of the mooring lines. The static analysis can be
performed based on the catenary mooring equations:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
T 2wZ
Th ¼ EA ð þ 1Þ2   EA ð1Þ
EA EA
Th wL Th L
X¼ sinh1 ð Þ þ ð2Þ
w Th EA

Tv ¼ wL ð3Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
T ¼ Th2 þ Tv2 ð4Þ

in which L, w, E, A represent the suspended length, wet weight, elastic modulus and
cross-sectional area of the mooring line. X and Z refer to the horizontal and vertical
dimension. T is the mooring line tension. Th and Tv represent the horizontal and vertical
components of the mooring line tension.
The length of the mooring line lying on the seabed should be checked to confirm
that there is no upward component of the force at the anchor. If there is insufficient line
length, the design and calculation should be repeated with increased line length or
increased line weight per unit length. The tension in the most heavily loaded line
should be compared with the breaking strength of the line. If the tension is too high, it
is necessary to change the mooring line material specifications, alter the line end co-
ordinates or the number of lines, and then repeat the calculation.

1.2 Quasi-static Method

The quasi-static design procedure consists of two types of calculations. The wave-
induced forces and responses at wave and drift frequencies of the platform are per-
formed in the time-domain simulation with the wind and current forces included, the
mooring tension is calculated statically without considering the line dynamics. The
quasi-static analysis method solves the equation of motion below:
582 B. Wu et al.

A€x þ B_x þ Bv x_ jx_ j þ Cx ¼ Fx ð5Þ

where the terms A, B, Bv refer to platform mass, linear and viscous damping, C rep-
resents the hydrostatic stiffness and Fx is the time varying external force.

F ¼ Fwave þ Fcurr þ Fwind þ Fmoor þ Fstatic þ . . . ð6Þ

Fwave ¼ Finc þ Fdiff þ Frad þ Fdrift ð7Þ

Fwave, Fcurr, Fwind represent the wave force, current force and wind force. Fmoor and
Fstatic refer to the mooring force and static force. Finc, Fdiff, Frad, Fdrift represent the
wave incidence force, diffraction force, radiation force and drift force.
Due to the random nature of the incoming irregular waves, the extreme values from
a time domain simulation are not usually statistically stable, therefore multiple seeded
simulations for the same sea state are required to achieve the statistically meaningful
results. It is a common practice to perform 10 seeded simulations for 3 h exposure time,
and then take the mean of 10 maximum values as the most probable maximum.

1.3 Dynamic Method

Fully dynamic analysis method takes into account the dynamic effects on mooring
tension, which include the hydrodynamic loads on the line and the inertia force of the
line due to its accelerations. Although the dynamic method is more accurate than the
quasi-static approach, it is more time consuming and hence less efficient. It is a
common practice to use the quasi-static analysis in the initial design stage, and use the
dynamic method for final approval of the design. In the numerical model for line
dynamic analysis, a mooring line is usually represented by a number of lumped mass
points which are connected in tandem by springs. The motions of the platform are
calculated. With the loads from the mooring system and the hydrodynamic damping
effect caused by the relative motion between the mooring lines and fluid are included in
the dynamic method. The line dynamic analysis is usually performed in time domain
which is more accurate than the frequency domain alternative. The time step should be
sufficiently small so that the highest frequency of the line dynamic oscillation can be
captured. Intelligent Optimization Algorithm.

1.4 Optimization Algorithm

Originated in last century, intelligent optimization algorithms were established by
researchers with the development of technology, industry and computer applications. In
the 21st century, the rapid improvement of computer speed and the increase of engi-
neering demand provide a foundation for the in-depth implementation and effective
application of intelligent optimization algorithms. Complex engineering problems
provide broad application requirements for algorithms. Especially in recent years, many
basic algorithms have been applied in such fields as machine learning, automated
optimization and so on. Genetic algorithm, simulated annealing algorithm, ant colony
algorithm and particle swarm optimization algorithm are among the popular intelligent
optimization algorithms [10, 11].
 Design Automation of Mooring Systems for Floating Structures 583

1.5 Genetic Algorithm

Aimed at tackling the problems studied, the genetic algorithm is selected as the basic
algorithm in this paper. Genetic algorithm has great potential for solving complex
optimization problems, and has been applied in industry, biology, automation, artificial
intelligence and other fields. The genetic algorithm adopted in this paper is developed
for mooring system problem. The complete design process, intelligent design, evalu-
ation and analysis method, various complex constraints are included in the algorithm.
The main procedure of the algorithm is as follows: 1) Creating the initial popu-
lation, generating the size of individuals as the initial population P(0) according to the
characteristics of the mooring system and the boundary conditions. 2) Calculating the
fitness values of all individuals in the current generation population P(t) on the basis of
objective function. 3) Based on the gambling table selection method, the selection
probability of each individual is determined, in which the individual with high fitness
and excellent performance is inherited to the next generation of the population. 4) The
crossover operation is carried out according to the crossover probability of each
individual. Some chromosomes are exchanged between individuals to generate a new
population. 5) The mutation operation is performed to change individuals based on the
mutation probability. The mutation of alleles is used to ensure the coverage of the
solution space and the overall algorithm. 6) After the genetic operation, the fitness
values of the individuals in population P(t+1) are calculated again, and the fitness values
of the individuals in population P(t+1) are obtained. 7) The termination criterion is
judged, the genetic algebra of the algorithm is analyzed, and the convergence evalu-
ation is completed. If the termination criterion is not satisfied, the optimization process
will return to the second step. Otherwise, the individual with the largest fitness value is
output as the optimal solution (Fig. 3).

Begin

Initial population

Fitness calculation

Yes
Termination criteria Optimal result

No

Selection operation End

Crossover operation

Mutation operation

Regeneration population

Fig. 3. Genetic algorithm procedure

584 B. Wu et al.

1.6 Development and Verification of Algorithm

The Basic programming language is used to develop the algorithm and the binariztion
of parameters is realized by encoding and decoding. The objective function and con-
straints are studied first for testing, which will be improved in the subsequent devel-
opment of the intelligent design program for mooring system. The relevant genetic
operators such as selection, crossover and mutation are developed and realized
according to the basic theory of algorithm. In order to validate the effectiveness of the
algorithm developed, process analysis and result validation are performed. The prob-
lem of mooring system design is similar to a function with multiple extreme values, and
the best result needs to be analyzed. Therefore, for a complex multi-parameter problem
with multi-peak solution, the genetic algorithm is studied and verified. The surface of
the function and the convergence curve are shown in figures below (Fig. 4). The
objective function is given as follows:

f ðx; yÞ ¼ x þ 10 sinð5xÞ þ 7 cosð4xÞ þ y þ 12 sinð4yÞ þ 5 cosð5yÞ þ 30 0  x; y  10

ð8Þ

Fig. 4. Test function and convergence curve

Figure 5 shows the evolutionary process of the results calculated. The individual
values of the first generation of population are randomly distributed in the solution
space. With the increase of generations, the individual values approach the peak value
of the objective function. Through the operation of mutation, the algorithm can make
sure a global optimal solution being found without being trapped in an area of local
optimal solution. The results indicate that the optimal solutions of complex problems
can be obtained accurately and quickly by the intelligent genetic algorithm.
 Design Automation of Mooring Systems for Floating Structures 585

(a) generation 1 (b) generation 5 (c) generation 10

(d) generation 20 (e) generation 50 (f) generation 100

Fig. 5. The evolutionary process of the algorithm

2 Design Automation Method of Mooring Systems

2.1 Design Principles and Fitting Formulas

The initial mooring parameters are derived from a pure static mooring analysis and the
optimization algorithm developed in this study. This initial design and analysis process
is carried out on the basis of the given boundary conditions, necessary input infor-
mation and the design requirements, including the water depth, environmental condi-
tions, the size and type of the floater, and environmental force on the floater and its
maximum allowable static offset at equilibrium etc. In this study 10% of the water
depth is assumed as the maximum allowable static offset of the floater under the mean
environmental load, i.e. the sum of the mean wind force, current force and the second-
order wave force. Once an initial mooring design is obtained on the basis of the pure
static analysis, a more in-depth design, analysis and optimization process is conducted
using the automated software tool developed in this study. In this in-depth design stage,
the motion response of the floater due to wind, current and wave, including 1st order
and 2nd order wave forces, will be calculated in time domain, and the mooring tension
is calculated quasi-statically at the same time. The feasibility and validity of the
mooring parameters are checked against the maximum tension in terms of its safety
factor, the maximum allowable horizontal offset of the floater, assumed to be 20% of
the water depth in the study which includes the pure static offset and dynamic motion
response to wave frequency and low frequency excitation forces. In addition, the laid
586 B. Wu et al.

length of the bottom mooring lines is required to be greater than zero so that no anchor
uplift will occur in any situation.
An exploratory study is conducted on the feasibility of automated search for the
optimal mooring parameters, such as mooring line length and mooring radius.
A number sample points of mooring line lengths under different environmental loads
and water depths are obtained by using the static calculation program. As shown in
Fig. 6, for each mooring line length, different mooring wet weight, of 100 kg/m,
150 kg/m and 200 kg/m in this case, are calculated. The environmental load varies
from 50 t to 250 t, and the water depth varies from 100 m to 1000 m. By analyzing the
variation law of mooring line length in Fig. 6, it can be concluded that in the same
conditions, the larger the environmental load on the floating structure, the longer the
mooring line length needs to be. Also for deeper water depth and lighter mooing line,
longer line length is needed. For the mooring line material selection, the variation of
mooring line length shows consistency and regularity. Therefore, with environmental
load, water depth and mooring line weight as variables, the relevant formula can be
fitted by sufficient sample points and parameters. With the same method, mooring
radius and mooring line length of single point and spread mooring system are analyzed.

Fig. 6. Sample points of mooring line length

The surface fitting is carried out for the sample points, and the approximate formula
is fitted. For typical fitting problems, the least square method is effective and widely
used. The Levenberg-Marquardt method and Gauss-Newton method are improved
methods based on the least square method. In this paper, Levenberg-Marquardt method
is selected. Based on the sample points of mooring line length, the formula is fitted as
follows:

X
7 X
7
L ¼ p1 þ p2 lnðFÞi þ p3 lnðHÞ j ð9Þ
i¼1 j¼1

p1 ¼ 659:19  e0:006W ð10Þ

p2 ¼ 0:0103  e0:004W ð11Þ

 Design Automation of Mooring Systems for Floating Structures 587

p3 ¼ 0:0019  e0:001W ð12Þ

F, H, W and L represent environmental load, water depth, mooring wet weight per
unit length and mooring line length. The curved surfaces of the fitting formulas are
shown in Fig. 7. The fitting results are in satisfactory agreement with the sample points,
which is a verification of the fitting formula. With this method, the recommended value
of mooring line length is given by the fitting formula. And a certain range of the values
of mooring line length is also given within which the automated design procedure is
implemented.

Fig. 7. Fitting surface of mooring line length

2.2 Automated Design Methodology

The “automation” of the methodology is mainly embodied in: 1) The complete design,
analysis and main evaluation process is integrated into the methodology, from the input
of conditions to the output of results. The complete design process is realized auto-
matically without human intervention. 2) In the process of design and analysis, mon-
itoring of design conditions and constraints for mooring system are performed. 3) The
evolutionary operation and the algorithm cycle are performed until the output of the
optimal result. The automated design methodology consists of three main steps: input
of design conditions, main process of intelligent design and output of mooring system
scheme (Fig. 8).
On the basis of the input design conditions, the automated analysis is completed in
the first step. According to the environmental parameters, the steady environmental
loads and hydrodynamic coefficients of the floating structure are analyzed. Based on
the mooring material database, which includes mooring chain sizes, dry and wet
weights, elasticity and minimum breaking load, and the change of mooring line weight,
the material parameters are determined. The length and radius of mooring lines are
determined on the basis of the approximate formulas. The minimum breaking load of
mooring line is related to the mooring size, therefore the selection of mooring size must
also take the required safety factor for mooring tension into consideration. The mini-
mum weight of mooring line is selected and analyzed as the initial mooring design
scheme. For the recommended range of mooring line length and radius obtained in the
first stage, the genetic algorithm is applied to generate initial mooring schemes
588 B. Wu et al.

• Weight
Material Material • Diameter • Mooring line length
Mooring layout
selection database • Stiffness • Mooring radius
• MBL

• Water depth
Environmental • Wave • QTF
parameters • Wind • Steady wind/current force
• Current • Wind/Current coefficient

Floating
Input condition Floater type
structure

• Barge • length/width/height
• Semi-submersible • deck/pontoon/column
• Spar • deck/column/soft tank

• single-point • single-component
Mooring type
• spread • multi-component

Fig. 8. Logical relation of design parameters

Input conditions and analysis

Initial mooring scheme population

No
Constraints on
initial position

Yes

Constraints on No
Elimination schemes
equilibrium position

Yes

No
Constraints on
maximum offset

Yes

Fitness calculation

Population evolution of
mooring scheme

Selection operation

Crossover operation

Mutation operation

Population regeneration of
mooring scheme

Output optimal mooring scheme

Fig. 9. Procedure of the automated design methodology

 Design Automation of Mooring Systems for Floating Structures 589

randomly according to the population size and the number of individuals required to be
calculated. The calculation, evaluation and evolution are subsequently performed.
Taking the energy of the mooring system, which equals to the area under the force-
excursion curve, as the fitness function, the performance of each mooring system
scheme is analyzed in combination with the engineering constraints.
The genetic operator of the algorithm is used to evolve the population, retaining the
individual of mooring scheme that meets the conditions and has high fitness value, and
eliminating the individual of mooring system scheme that does not satisfy any con-
dition. Individuals with relatively higher fitness value and satisfying more conditions
can generate the next generation of mooring scheme population, which can be iterated
and evolved continuously until the algorithm converges and finally outputs the optimal
mooring system scheme. The proposed mooring system scheme obtained from the
automated design is finally output with detailed parameters, including mooring system
layout and mooring material parameters (Fig. 9).

3 Application

The automated design and analysis of typical mooring systems are performed in the
study. The environmental conditions are shown in Table 1. The JONSWAP wave
spectrum is adopted in the simulation of irregular waves. The NPD spectral wind is also
applied in the simulation.

Table 1. Environmental conditions

Environmental parameters Condition
Wave Wave spectrum JONSWAP
Significant wave height (Hs) 8m
Spectral peak period (Tp) 12.8 s
Wind Wind spectrum NPD
Wind velocity (Vw) (10 min mean) 46 m/s
Current Current velocity (Vc) 1 m/s

3.1 Single-Point Mooring System

Taking a FPSO as an example, the automated design of single point mooring system is
studied to verify the feasibility of the method. The numerical analysis is performed in
time domain using hydrodynamic analysis software AQWA. Ten 3-h simulations are
performed each with a different wave spectral seed, which is used to reduce the
uncertainty in the extreme values obtained from the time domain simulation in a
random process. Figure 10 shows the hydrodynamic model of the FPSO and mooring
system (Table 2).
590 B. Wu et al.

Fig. 10. Computational model

Table 2. Design scheme

Results Diameter Wet Weight Stiffness MBL Mooring Mooring Total Fitness
(mm) weight (kg/m) (kN) (t) length radius weight value
(kg/m) (m) (m) (t)
None 66 75 – – – – – – –
None 76 100 – – – – – – –
Scheme 1 85 125 144 617015 548 984.6 912.9 141.8 42936206
Scheme 2 93 150 172 738625 645 903.8 828.3 155.5 44128742
Scheme 3 101 175 203 871165 747 839.9 761.5 170.5 46235625
Scheme 4 108 200 232 996106 841 762.5 679.0 176.9 46929149

The horizontal motion response curve and the mooring line tension curve are
shown in Fig. 11. The static equilibrium positions of the four schemes are from 17.3 m
to18.8 m. The value is less than 10% of the water depth, which meets the assumed
engineering design requirements. The motion response of the floating structure with the
four mooring schemes ranges from −3.5 m to 42 m, and the maximum value is about
20% of water depth, which also satisfies the maximum offset constraint. The maximum
tension of mooring line is about 230 t, and the minimum safety factor is 2.4, which
meets the safety requirement for quasi-static calculation in classification society stan-
dard. Although all the four output mooring system schemes satisfy the design condi-
tions and requirements, there are differences in the mooring material properties and
mooring radius, further decision needs to be made on the basis of their respective
merits before a final scheme is selected (Tables 3 and 4).

Fig. 11. Motion response and mooring tension

 Design Automation of Mooring Systems for Floating Structures 591

Table 3. Motion response

Results Equilibrium position (m) Motion response
(m)
Min Max Mean
Scheme 1 18.8 −3.3 42.6 22.6
Scheme 2 17.8 −3.6 42.1 21.8
Scheme 3 17.4 −3.5 41.6 20.5
Scheme 4 17.3 −3.3 41.8 20.6

Table 4. Mooring line characteristic

Results Tension
Pre-tension (t) Max (t) Safety factor
Scheme 1 61.6 226.4 2.4
Scheme 2 68.8 227.9 2.8
Scheme 3 77.2 233.4 3.2
Scheme 4 81.4 233.7 3.6

3.2 Spread Mooring System

The automated design of the spread mooring system is performed with a semi-
submersible platform as an example. The water depth is assumed to be 500 m, and a
chain-wire-chain mooring composition is adopted in order to reduce the weight of the
suspended part of the line and thus reduce the pre-tension. For simplicity, the usual
4  2 or 4  3 mooring layout for a semi-submersible platform is simplified to a
4  1 layout, in which 1 mooring line representing the properties of 2 or 3 lines. The
motion response of the floater and the characteristics of mooring system are evaluated.
The hydrodynamic model of the semi-submersible platform and mooring system are
shown in Fig. 12 (Table 5).

Fig. 12. Computational model

 592
B. Wu et al.

Table 5. Design scheme

Results Mooring Mooring Top segment - Chain Middle segment – Wire rope Bottom segment - Chain Total Total MBL

length(m) radius(m) weight length (t)

Diameter Wet weight Length Diameter Wet weight Length Diameter Wet weight Length
(t) (m)
(mm) (kg/m) (m) (mm) (kg/m) (m) (mm) (kg/m) (m)

None — — 142 350 — 144 72.1 — 142 350 — — — —

None — — 147 375 — 149 77.1 — 147 375 — — — —

Scheme 1 1739.6 1559.4 152 400 15 153 81.3 608.8 152 400 1115.8 519.9 608.8 1500

Scheme 2 1734.3 1554.1 157 425 15 157 85.7 604.2 157 425 1115.1 554.3 604.2 1581

Scheme 3 1633.6 1447.6 162 450 15 161 90.1 499.3 162 450 1119.3 592.4 499.3 1662

Scheme 4 1584.0 1385.2 166 475 15 164 93.5 374.1 166 475 1194.9 663.5 374.1 1727

Scheme 5 1532.7 1330.9 170 500 15 167 96.9 362.7 170 500 1155.0 672.9 362.7 1792
 Design Automation of Mooring Systems for Floating Structures 593

The results of motion responses are shown in Table 6. The static offset at equi-
librium position is about 10% of water depth. The range of motion response is close to
each other for different mooring schemes, and the maximum horizontal offset is less
than the maximum allowable motion. Table 7 shows the results of the mooring line
tensions calculated in time domain. The maximum tension is between 524 t and 620 t,
and the safety factor is 2.9, which satisfies the classification society’s standard.
Compared with the full-chain mooring line in the same conditions, the pre-tension of
the chain-wire-chain mooring line is reduced (Fig. 13).

Fig. 13. Motion response and mooring tension

Table 6. Horizontal motion response

Results Equilibrium position (m) Motion response
(m)
Min Max Mean
Scheme 1 50.4 0.0 79.5 57.5
Scheme 2 47.6 0.0 76.1 54.5
Scheme 3 47.9 0.0 76.0 54.7
Scheme 4 49.4 0.0 77.7 53.5
Scheme 5 49.0 0.0 77.5 51.8

Table 7. Mooring line characteristic

Results Minimum laid length (m) Tension
Pre-tension (t) Max (t) Safety factor
Scheme 1 481.9 202.2 524.0 2.9
Scheme 2 499.0 217.5 537.1 2.9
Scheme 3 438.4 278.2 575.1 2.9
Scheme 4 441.9 326.2 598.9 2.9
Scheme 5 403.7 339.2 618.2 2.9
594 B. Wu et al.

4 Concluding Remarks

In order to realize the automation of mooring system design, a study has been per-
formed in intelligent design of mooring systems. On the basis of the analysis results, it
can be concluded that:
• The automated design methodology of mooring system is established in this paper.
The procedure includes the complete design process of mooring systems, combined
with performance evaluation and intelligent optimization algorithm. Mooring
schemes meeting all the design conditions and with satisfactory performances are
designed based on the methodology.
• The computer program for automated design of mooring system is developed. With
this program, the single point and spread mooring systems are studied, and the
feasibility of the program is verified.
• The methodology proposed in this studied improves the efficiency of mooring
system design. It provides an engineering solution and application tool for the
automated design of mooring systems.

References
1. Chakerabarti, S.K.: Handbook of Offshore Engineering. Offshore Structure Anakysis Inc.,
Plainfield (2005)
2. Karimirad, M.: Offshore Energy Structures. Springer, Cham (2014)
3. Wang, C.M., Wang, B.T.: Large Floating Structures. Springer, Singapore (2015)
4. Det Norske Veritas: DNV-OS-E301. Position Mooring (2010)
5. Det Norske Veritas: DNV-RP-C205. Environmental Conditions and Environmental Loads
(2010)
6. American Petroleum Institute: API RP 2SK. Design and Analysis of Stationkeeping Systems
for Floating Structures (2005)
7. Mirzaei, M., Maimun, A., Priyanto, A., Fitriadhy, A.: Mooring pattern optimization using a
genetic algorithm. J. Biomed. Opt. 66(2), 189–193 (2015)
8. Girón, A.R.C., Corrêa, F.N., Hernández, A.O.V., et al.: An integrated methodology for the
design of mooring systems and risers. Marine Struct. 39, 395–423 (2014)
9. Wu, B., Cheng, X.M., Tian, C., Ni, X.Y.: Mooring system design and assessment for a
multi-moduled very large floating structure. In: 12th International Conference on Hydro-
dynamics, Egmond aan Zee, The Netherlands (2016)
10. Kramer, O.: Genetic Algorithm Essentials. Springer, Heidelberg (2017)
11. Sanghamitra, B., Sankar, K.P.: Classification and Learning using Genetic Algorithms.
Springer, Berlin (2007)
 A Study of Riser Oscillations Caused by Slug
Flows During Subsea Petroleum Production

Sergio Nascimento Bordalo(&) and Celso Kazuyuki Morooka

Division of Petroleum Engineering of the Department of Energy,

Faculty of Mechanical Engineering, University of Campinas (Unicamp),
Campinas, Brazil
[email protected]

Abstract. The theoretical influence of the internal slug flows on the motion of
suspended slender pipes is presented here through two examples of computer
simulations applied to two types of risers employed in offshore petroleum
production – a steel catenary riser (SCR) and a steel lazy wave riser (SLWR)
hanging from a floating platform. The results presented here provide strong
evidence that slug flows may induce hazardous oscillation, even in the absence
of other excitations. The exploitation of hydrocarbons from underground
reservoirs in open sea is undertaken amidst very adverse environmental con-
ditions, below ultra deep waters and usually very deep reservoirs, as in the pre-
salt and carbonate formations discovered in the brazilian southeast atlantic coast
basin. In this scenario, large production rates of oil and gas are required to
achieve economically feasible operations; which means successfully maintain-
ing an uninterrupted flow of reservoir fluids under safe conditions from many
subsea wells. The multiphase petroleum flowing from the wellhead at the sea
bottom to the floating process facility at the sea surface is conducted by lengthy
pipelines lying on the seabed up to the point where the flow ascends to the sea
level through very long and slender pipes – named risers – suspended by its top
end at the floating production unit. Risers may be flexible or made from steel
rigid pipes and are often subjected to sea current and wave loads and to the
movements of its supporting ship or platform, which are dangerous to the pipe
integrity. Risers are also excited to oscillate due to the multiphase flow inside
them. It is of paramount importance to understand and determine the dynamic
behavior of the suspended pipes due to those loads, in order to achieve a service
life as extensive as economically possible. The method employed in the present
research consists of modifying a computer dynamic simulator of pipeline
movement, adding the internal fluid loads arising due to the weight and
momentum of the flowing masses of liquid and gas. The simulations show that a
steady slug flow may alone produce oscillatory displacements of the riser of
rather significant amplitudes, without any other external cause; particularly
when the slug’s excitation is near the natural frequencies of the riser. The results
indicate that the oscillations are strongly dependent on the slug frequency. The
modified simulator is flexible enough to compute the effect of a wide range of
slugs’ configurations (geometry, fluids, flow rates etc.), allowing the analysis of
fatigue damage and risk of failure under many production conditions.

Keywords: Offshore petroleum risers
Slug flow
Oscillation dynamics

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 595–613, 2021.
https://doi.org/10.1007/978-981-15-4680-8_41
596 S. N. Bordalo and C. K. Morooka

1 Introduction
1.1 Objective
The exploitation of hydrocarbons from underground reservoirs in open sea is under-
taken amidst very adverse environmental conditions, below ultra deep waters and
usually very deep reservoirs, as it happens in the pre-salt and carbonate formations
discovered in the brazilian southeast atlantic coast basin. In this scenario, large pro-
duction rates of oil and gas are required to achieve economically feasible operations;
which means successfully maintaining an uninterrupted flow of reservoir fluids under
safe conditions from many subsea wells. The multiphase (oil, gas and water) petroleum
flow from the wellhead at the sea bottom to the floating process facility at the sea
surface is conducted by lengthy pipelines lying on the seabed up to the point where the
flow ascends to the sea level through very long and slender pipes suspended by its top
end at the floating production unit (FPU, Fig. 1). These suspended pipes are known as
risers, which may be flexible or made from steel rigid pipes. Normally, risers are simply
deployed from the sea surface to the sea bottom, but, sometimes, for steel rigid risers,
buoys are placed along the length of the riser in order to reduce the total weight
supported by the production ship or platform. Offshore risers are often subjected to sea
current and wave loads and to the oscillatory movements of its supporting ship or
platform, which impose motions to the suspended pipe. Aside from those environ-
mental loads, risers are excited to oscillate due to the multiphase petroleum mixtures
flowing inside them. A proficient domain of the dynamic behavior of suspended pipes
due to those loads is fundamental to the design of petroleum risers, because one aims
specially for a service life as extensive as economically possible.

FPU
Sea Level
Water Depth

Catenary
Riser

Seabed
Petroleum Pipeline

Fig. 1. Illustration of a catenary riser configuration.

In this paper, the theoretical influence of the internal slug flows on the motion of
suspended slender pipes is presented through examples of computer simulations
applied to different types of risers employed in offshore petroleum production – a steel
catenary riser (SCR) and a steel lazy wave riser (SLWR) hanging from a floating
platform. In this study, two kinds of fluid loads are considered arising from the
 A Study of Riser Oscillations Caused by Slug Flows 597

interaction of the internal flow and the pipe; one is due to the weight of the fluid and the
other is due to the curvature of the flow path. In the case of slug flows, the flow pattern
of the two-phase mixture of oil and gas develops as an intermittent streamwise dis-
tribution of mass which may induce an oscillatory movement on the pipe. The internal
fluid dynamic loads depend on the flow rates of both liquid and gas phases, fluid
densities, flow section area, and also on the trajectory of the flowline, thus requiring the
calculation of the riser movement under flow conditions, which, in turn, depends on the
magnitude of the flow forces. Therefore, the simulations demand an interactive cou-
pling of the computer modules that determine the motion of the pipe body and the
algorithm that predicts the fluid forces. In the present paper, a brief description of the
mathematical flow model and flow loads is discussed, before presenting the effect of
the slug flow frequency on the amplitude of oscillation of the riser. For this study, the
pipe is not subjected to external excitations; i.e., sea currents, sea waves and the
movement of the floating platform are not considered; therefore demonstrating that the
oscillations are solely due to the slug flow. The simulations show that the effect of the
internal intermittent flow of liquid and gas on the pipe is significant, especially when
the slug flow frequency approaches one of the pipe’s natural frequencies, incurring in
relatively high bending amplitudes. Such results raise concerns regarding fatigue
damage during long term operations.

1.2 Background
In the field of fluid mechanics, there is a huge effort in the characterization and pre-
diction of two-phase flow patterns, in the calculation of pressure gradients and flow
rates, and in the determination of slug lengths and frequencies (Shoham 2006; Govier
and Aziz 2008; Johnson 2016; Michaelides et al. 2016), but studies on the effect of slug
flows on pipe motion have recently become a matter of great interest for petroleum
production.
Risers are subjected to loads due to its own weight, buoyancy, sea currents, waves,
and the movement of the FPU (Sparks 2007). The forces arisen from such loads
combined with the elastic and inertial properties of the riser result in a behavior that can
be described as comprised of dynamic (time dependent) displacements superimposed
on a static (time independent) deformation. Such movements of the riser body generate
stresses – static and dynamic – that must be considered in the design of the riser for a
lifetime of operations. The effects of sea currents, waves, and the movement of the
floating production unit on the dynamics of risers have been the subject of numerous
studies, both experimental and with the aid of numerical simulations on computers
(Chakrabarti and Frampton 1982; Chakrabarti 1987; Gonzales et al. 2005; Morooka
and Tsukada 2013).
A short list of the previous works on pipe-slug interaction will omit many
researchers; but one might start with the well known work of Paidoussis (1970) on a
vertical straight pipe with small deflections conveying one-phase flow; include Moe
and Chucheepsakul (1988) first use of a dynamic load model for one-phase flow; Patel
and Seyed (1989) implementation of a sine function to represent two-phase flow; Wu
and Lou (1991) analytical linearization technique for one-phase flow; Paidoussis
(1998) compilation of the sate of knowledge at the time; Pollio and Mossa (2009)
598 S. N. Bordalo and C. K. Morooka

follow up on Patel and Seyed (1989); Yamamoto and Murai (2010) numerical study of
the effect of the internal flow on riser’s mechanics; Bordalo et al. (2015) description of
the intermittent forces generated by slug flows.
Recent developments have been undertaken to introduce two-phase flow effects
into the study of marine petroleum pipelines. For instance, Bossio et al. (2014) ana-
lyzed a straight horizontal pipe over a free span under the effect of vortex induced
vibration (VIV) and slug flow, employing a one-dimensional equations approach
similar to straight beam approximations. It was found that, because the slug mass
affected the system weight and inertia, the pipe oscillations induced by the VIV were
modified in the presence of slugs. Su (2015) studied a vertical straight riser subjected to
cross flow sea current and internal slug flow, and, as Bossio, used one-dimensional
beam simplifications. It was observed that the amplitude of oscillation depended on the
flow rates of liquid and gas. Van der Heijden et al. (2014) studied the impact force of
one slug on a jumper U-bend, where the bend was divided into three straight pipes and
the slug was treated as a bullet of liquid passing through. This study focused on fatigue
analysis at a point of the pipe due to fluctuations generated by the passing slug. Reda
et al (2012) compared a moving force model and a mass moving model representing
one slug coursing through a horizontal straight pipe over a free span, with a length less
than one hundred diameters. The differences between the results of those two models
were classified in terms of two nondimensional variables related to the slug speed and
slug mass. Ortega et al. (2013) and Gundersen et al. (2012) analyzed the combined
effects of waves and slugs on a riser, employing a marine pipeline wave response
simulator and a slug simulator. It was observed, in both papers, that the presence of the
slugs had an impact on the pipe response generated by the waves. Jia (2012) attempted
to simulate complex transient slugging (slug pulses or slug generation) and its inter-
action with pipelines using a full computational fluid dynamics simulator (CFDS),
which was relatively heavy and slow, limiting the amount of analysis and parametric
variation. Safrendyo and Srinil (2018) studied the concurrent effects of VIV with slug
flows. The present research deals with large oscillations induced by slug flow forces on
various geometries of long slender suspended pipelines.

2 Methodology and Model

2.1 Model for the Flowing Mass Distribution

The first ingredient of the proposed model of the flow forces is the fluid mass distri-
bution along the length of the pipe. The geometry of the mass distribution of two-phase
flows has autodetermined configurations, called flow patterns, which depend on the
flow rates of each phase, their physical properties and the geometry of the pipeline
(Beggs and Brill 1986). Flow patterns – bubbles, slugs, churning and annular – are a
function of the liquid and gas flow rates (QL and QG) for a pipe internal flow area
A. Bubble flows and annular flows have a pretty uniform distribution of mass compared
to the intermittent characteristic of the slug flows, therefore, bubble flows and annular
flows almost do not give rise to major oscillations (although minor high frequency
vibrations may be present). On the other hand, the intermittency of slug flows is the
 A Study of Riser Oscillations Caused by Slug Flows 599

primary source of variable loads, as indicated by the exploratory experiments carried

out by Bordalo et al. (2008) on a laboratory-scale model, which suggested that a
whipping motion may arise on risers due solely to slug flows, even when there are no
relevant currents of the surrounding media, neither waves nor movement of the top of
the riser. This innate oscillation inducing characteristic of slug flows is the main
concern of the present flow load model.
Here, the slug pattern is represented as a sequence of fluid cells (packages) – each
containing certain amounts of liquid and gas in the form of liquid plugs and gas pockets
(Fig. 2) – travelling at a specified velocity vc (Dukler and Hubbard 1975). As time goes
by, these fluid masses are tracked as they move along the pipe, so that one may
determine the flow properties at any moment and lengthwise position.

s
vc
φ(s,t)
Gas Liq

Lc
Fig. 2. Travelling slug cells.

The length of the fluid cell Lc is determinant for the frequency imposed by the slug
flow fs (Eq. 1).
vc
fs ¼ ð1Þ
Lc

The density of the fluid mixture at each point along the flow would be composed of
the density of the liquid phase (qL) and the gas phase (qG) according to

qðs; tÞ ¼ /:qL þ ð1  /Þ:qG ð2Þ

in which, the mass distribution / may be any appropriate function F(n), where n
describes the translational movement of the slugs, such as
 
s  vc :t
/¼F ð3Þ
Lc

where t is time, and s is the position along the pipe length. In this paper, a “slug
function”, periodical in space (Fig. 2), was prescribed through a subroutine attached to
the computer code that calculated the internal flow forces. The / function varies
between 0 and 1 – “0” when the pipe section is full of gas, and “1” when it is full of
liquid; also, there may be a film of liquid flowing in the gas pocket between two liquid
plugs.
The complete description of the slug load model depends on given variables from
the production case under analysis. Besides entering the usual physical and geometrical
600 S. N. Bordalo and C. K. Morooka

properties of the pipeline and its environment in the simulator input data files, the
analyst must also supply the necessary data of the internal flow: liquid and gas densities
(qL and qG), liquid and gas flow rates (QL and QG), total length of the pipeline (Lt),
diameter of the internal flow cross-section (Di). The velocity of the slug cell (vc) and the
volume fractions of liquid and gas relative to the volume of the slug cell (respectively,
VL/Vc = HL, named “hold up”, and VG/Vc = HG = (1 − HL), named “void fraction”)
are related to the flow rates according to QL = vc.Ai.HL and QG = vc.Ai.HG, hence they
are calculated from

QL þ QG
vc ¼ ð4Þ
Ai

QL
HL ¼ ð5Þ
QL þ QG

The length of the liquid plug (LL) and of the gas pocket (LG = Lc − LL) are not
equal, they are derived from the volumetric relationship HL = KL + jf (1 − KL), where
KL = LL/Lc and jf is the volumetric fraction of liquid film inside the gas pocket (Vf/VG)
depicted in Fig. 2, yielding

HL  jf
KL ¼ ð6Þ
1  jf

These results are mathematically accurate for the travelling mass distribution
employed here. In case other features are added to the modeling of the slugs – such as a
slip velocity ratio between the slugging layer and the liquid film, the previous equations
should be modified accordingly.
The slug frequency fs or the length Lc of the slug cells are still an issue not fully
resolved, as a literature search reveals (Zabaras 2000; Shoham 2006; Rodrigues et al.
2007; Govier and Aziz 2008). In the present work, the user is left with a choice of
correlations from three available formulations – Gregory and Scott (1969), Heywood
and Richardson (1979), or Fossa et al. (2003); other correlations may be implemented
in the future. Alternatively, the user may specify the frequency of the slugs or the
length of the slug cells (one or the other but not both), in the input data file, at will or
based on some field information. These two variables, fs and Lc, are made consistent by
the use of Eq. 1. In the case of the bubble and annular flow patterns, the mass dis-
tribution is represented by a constant average mixture density along the pipe; which is
of no great concern here because the homogeneous distribution does not present
intermittency, and therefore, it does not induce time varying loads.

2.2 Model for the Slug Loads

The second component of the proposed computer model consists of the magnitude and
direction in space of the flow induced loads on the pipeline. The internal flow forces of
more relevance, in the present cases, are the force of gravity and the force due to the
curvature of the flow trajectory (Coriolis effects are presently neglected). The force of
 A Study of Riser Oscillations Caused by Slug Flows 601

gravity is simply the weight of fluid at a pipe section; it depends on the density of the
fluid q (Eq. 2), the area of the internal flow cross-section Ai and the intensity g of the
gravity field (Johnson 2016). For an infinitesimal transversal slice element of the pipe,
of thickness ds and volume dV = Ai.ds, the mass of fluid is dm = q.dV, so the mag-
nitude of the weight dFg = g.dm becomes dFg = q.g.Ai.ds; therefore the vector
describing the gravitational load, in terms of force per unit length of pipe, is

~
Fg0 ¼ q:g:Ai :~
ug ð7Þ

in which, ~ ug is the unit vector of the direction of the gravity field.

In the case of slugs, the intermittency of the mass distribution q(s,t) along the pipe
gives rise to a variable weight load which, coupled to the movement of the slugs,
generates a frequency of excitation, given by Eq. 1. In fact, if the frequency of exci-
tation is close to the natural frequency of the pipeline structure, the amplitude of the
oscillation imposed by the internal flow may become quite severe.
The force induced by the curvature of the flow trajectory (which may be a curved duct
or a bent pipe, as in the case of jumpers, suspended risers etc.) is generated by the variation
of the flow momentum due to the change in flow direction constrained by the path of the
pipeline (Johnson 2016), and acts against the pipe wall, normal to the local direction of the
flow (which is tangent to the trajectory), in the outward direction with respect to the radius
of curvature r (that is, away from the center of curvature). The curvature force depends on
the density of the fluid q, the area of the internal flow cross-section Ai and the intensity of
the centripetal acceleration ac = v2/r (the flow velocity is v). The magnitude of this inertial
force dFc = ac.dm becomes dFc = q.v2/r.Ai.ds for an infinitesimal pipe segment, and the
vector describing the curvature load, in terms of force per unit length of pipe, is
2
~ v
Fc0 ¼ q: :Ai :~
ucf ð8Þ
r
in which, ~
ucf is the unit vector of the centrifugal direction. Again, the variation of
density with time and space q(s,t) gives rise to a variable load.

2.3 Methodology for Flow Trajectory and Pipe Dynamics

The third cornerstone of the proposed computer model is the trajectory of the flowline.
The radius of curvature and the centrifugal vector (appearing in Eq. 8) depend on the
instantaneous geometric configuration of the pipeline body, which may be constantly
changing as the pipeline is moving under the influence of the acting forces. Mathe-
matically, the radius of curvature is given by Eq. 9; pragmatically, this equation may be
used in its finite difference form, and, therefore, r may be obtained from the ratio of dh
and ds, where ds is the length of the curved path and dh is the angle between the tangent
vectors (change in direction) at the extremities of that path, for small enough ds.

1 dh
¼ ð9Þ
r ds
602 S. N. Bordalo and C. K. Morooka

The centrifugal unit vector may be derived from the vector ~

P that specifies the 3D-
space coordinates of the points composing the longitudinal extension of the pipeline –
the pathline, which also defines the trajectory of the flow inside the pipeline. The
derivation follows the sequence shown as Eq. 10.

ð10Þ

For the numerical simulations developed in the present work, the pipeline body is
described by a sequence of moving nodes and the derivatives in Eq. 9 and Eq. 10 are
computed by 3D vector finite-differencing of the pathline, thus providing generality in
dealing with any 3D geometry, with displacements in all three global directions (X,Y,Z),
as needed in the cases of moving submerged pipelines, specially for the different types of
risers.
Since the pathline is continuously changing, as the pipeline oscillates, the internal
flow forces’ subroutines must maintain a constant two-way dialog with the dynamic
integrator of the pipeline movement simulator (which is solving the equations of
motion for the pipeline). At each time step, the simulator must call upon the internal

Data Data

SIMULATOR Internal Flow

Loop – Movement Dynamics
Coordinates of
(for “n” Time Steps) Curvature
Position along
Forces on the Pipeline the Pipeline
Mass
Distribution
Movement Simulation
Internal Flow
(One Time Step)
Forces Gravity Force
Curvature Force
Coordinates of Positions
along the Pipeline

Output

Fig. 3. Flowchart of the interaction between the pipeline simulator and the internal flow forces’
subroutines (“Curvature”, “Mass Distribution” and “Gravity Force, Curvature Force”). Arrows:
flow of information – from the simulator to the subroutines (red), and from the subroutines to the
simulator (green).
 A Study of Riser Oscillations Caused by Slug Flows 603

flow forces’ subroutines, passing the current vector position ~ P to them. The internal
flow forces’ subroutines evaluate the curvature of the pathline and the magnitude and
direction of the internal flow forces, returning these to the simulator. Then, the sim-
ulator computes the movement and new position of the pipeline for the next time-step,
based on these forces and on the inertia, elasticity and energy dissipation of the pipe.
This process is illustrated in Fig. 3. These subroutines modules were created, and
wedged into a version of an existing simulator (Mourelle 1993).

3 Results
3.1 Steel Catenary Riser (SCR)
As an example of application to offshore oil fields, the following results refer to a steel
catenary riser (SCR) set up with a configuration very much alike those employed in real
life (Fig. 1). Risers are built from rigid pipes, but may present large oscillations
because they are very long and slender, hanging loose between its ends. Table 1
presents the data used in the simulation. Many different slug flows may be found in the
fields; herein a case was selected from one of many possibilities.
The simulations provide insight into many dynamical variables – 3D coordinates of
the movement of points of the riser body, rotation of pipe elements, axial tension, shear

Table 1. Steel catenary riser case data

Riser length Lr [m] 3100
Horizontal projection Yp [m] 1918
Vertical projection Zp [m] 1990
Water depth Hw [m] 2000
TDP from the top projection Ytdp [m] 1412
TDP from the anchor Yp–Ytdp [m] 506
Suspended length Sr [m] 2594
External diameter De [m] 0.212
Internal diameter Di [m] 0.165
Elasticity modulus E [kN/m2] 207  106
Specific weight of pipe cs [kN/m3] 77.7
Slug cell length Lc [m] 100
Liquid length LL [m] 50
Gas length LG [m] 50
Specific weight of liquid cL [kN/m3] 6.76
Specific weight of gas cG [kN/m3] 4.63
Slug velocity vc [m/s] 3.0
Slug frequency fc [Hz] 0.0303
Slug period Tc [s] 33.0
TDP: touchdown point
604 S. N. Bordalo and C. K. Morooka

forces and nodal moments, axial, bending and shear stresses etc. For the sake of space
restrictions and objectiveness, only a few samples are presented herein.
Sample time-histories are shown for the case at hand, displaying the results for two
selected nodes (that are positioned as depicted in Fig. 4). The displacement of the
nodes, in the Y and Z direction, are referenced to a configuration denominated “static”,
which is the position of the nodes when no dynamic loads are acting on the pipeline. It
can be seen, in Fig. 5, that, even though no external environmental forces are present,
the pipeline oscillates due solely to the excitation imposed by the internal flow. In the

Steel Catenary Riser

2000
time = 2600s sea level riser top

Selected Nodes
1500

Z (m)
1000

node 1500
500

anchor node 0600

sea bed
0
-2000 -1500 -1000 -500 0
Y (m)

Fig. 4. Steel catenary riser (SCR) geometry.

present case, the slug weight is the prevalent flow load compared to the curvature
forces, because the radius of curvature is large and the flow velocity is not high enough.
Node 0600 (600 m from the anchor point) is close to the touchdown region, where
the pipe approaches a horizontal plane; the Y displacement is smaller than the Z
displacement, since there is more freedom to move in the Z-direction, which is almost
perpendicular to the line body, and less freedom to move in the Y-direction, which is
almost along the axis of the line body. Node 1500 (1500 m from the anchor point) is
almost midway in the suspended segment of the riser; therefore the transversal dis-
placements with regard to the length of the body, which are easier to be accomplished,
translate into significant displacements in the Y and Z directions; in the present case,
oscillation amplitudes of the order of 50 mm (Fig. 5b) represent 24% of the external
diameter of the riser.
Figure 6 shows the effect of the internal flow load on the bending stress (the graph
shows the longitudinal bending stress on the external wall of the cross section of the
pipe); the slug flow imposes an alternating stress on top of an average value, therefore
raising concerns about fatigue of the pipeline. The evaluation of the alternating stress is
the starting point for fatigue analysis in the design of the pipeline, regarding its failure
and useful service life. As expected, the node closer to the touchdown point (TDP –
 A Study of Riser Oscillations Caused by Slug Flows 605

Y-displacement - Node 0600 Z-displacement - Node 0600

100 100

Z - displacement (mm)
Y - displacement (mm)

50 50

0 0

-50 -50

-100 -100
2000 2500 2000 2500
Time (s) Time (s)

(a)

Y-displacement - Node 1500 Z-displacement - Node 1500

100 100
Z - displacement (mm)
Y - displacement (mm)

50 50

0 0

-50 -50

-100 -100
2000 2500 2000 2500
Time (s) Time (s)

(b) .

Fig. 5. a. Time-history: node 0600, displacement in Y and Z. b. Time-history: node 1500,

displacement in Y and Z.

contact with the seafloor) presents higher average bending stress compared to the node
far above in the suspended stretch of the riser body, because the static curvature of a
catenary geometry is greater closer to the TDP (Fig. 4). For these two nodes, the
alternate component of the bending stress is also larger at the node closer to the
TDP. Close to the TDP (node 600), the vertical upward movement of the nodes in the
suspended side of the pipe leads to a large increase in the curvature at the TDP (while
vertical downward movement leads to a decrease in the curvature), because the nodes
on the side of the anchored line do not have as much freedom to move due to the
contact with the ground. The nodes in the suspended segment are free to move in the
transversal direction of the pipeline, and act as a free span long beam with moderate
bending. Approaching the riser top, bending gets even smaller due to the hinged end,
but rotation gets larger.
606 S. N. Bordalo and C. K. Morooka

Bending Stress - Node 0600 Bending Stress - Node 1500

33 13
Bending Stress (MPa)

Bending Stress (MPa)

32 12

31 11

30 10

29 9
2000 2500 2000 2500
Time (s) Time (s)

Fig. 6. Time-history: nodes 0600 and 1500, bending stress (external wall).

In order to deeply investigate effect of the slug frequency, and enlarge the scope of
the present discussion, simulations were run with slugs flowing in the range of 2.4 m/s
to 4.0 m/s. The results are shown in Fig. 7, displaying “Zmax − Zmin” – the full
amplitude in the Z direction of the displacement of three selected nodes of the riser

Zmax - Zmin
120
Node 0700
100 Node 1450
Node 2500
Zmax - Zmin (mm)

80
Natural Freq
60

40

20

0
20 25 30 35 40 45
Frequency (mHz)

Fig. 7. Effect of the slug frequency on the Z displacement (vertical) of the SCR: nodes 0700,
1450 and 2500.

(these nodes lie in zones of large amplitude). It is observed that the effect of the slugs
on the displacement of the riser is amplified close to one of the natural frequencies of
the pipeline, which is 0.03018 Hz (period of 33.14 s).
This behavior is well known for mechanical elastic systems forced by oscillatory
loads. It should be pointed out that the observed amplitudes are not to be neglected; for
scale reference, the external diameter is 212 mm, so amplitudes of the order of 5% to
 A Study of Riser Oscillations Caused by Slug Flows 607

50% of the diameter are shown in the range of Fig. 7. Such oscillations may be even
worse for other flow rates of oil and gas in this pipeline.

3.2 Steel Lazy Wave Riser (SLWR)

Steel Lazy Wave Riser

2500
Ɵme = 2800s sea level Top
2000
Selected Nodes

1500
node 5500

Z (m)
1000

node 2500
500
<Anchor node 0400
sea bed 0
-3000 -2000 -1000 0
Y (m)

Fig. 8. Steel lazy wave riser (SLWR) geometry.

A final example of application to offshore oil fields is the steel lazy wave riser (SLWR)
shown in Fig. 8. Simulation data is presented in Table 2. The hump in the line
geometry (around node 2500) is achieved by a series of buoys attached to the pipe in
that region, providing positive floatability.
Figure 9 displays the results for two selected nodes (that are located as shown in
Fig. 8). As before, the displacement of the nodes, in the Y and Z direction, are ref-
erenced to the “static” configuration. Again, it is observed that the pipeline oscillates
without any external cause, but the excitation induced by the internal slug flow. Similar
to the SCR case, the displacements in the Z-direction close to the TDP (node 0400) are
larger than in the Y-direction. With less constrains due to the geometry, node 2500 (in
the floaters segment) is allowed to move with more freedom in the Y-direction than
node 0400. Time history of the bending stress is shown in Fig. 10; it is noted that the
slug flow causes a variable stress do develop which would induce a daily damage to the
structure (the damage rate is the rate at which the life of the structure is consumed by
fatigue inducing loads). Since damage is correlated to the amplitude of the alternating
stress, it is of interest to check the difference between maximum and minimum values
(DMM) of the stresses along the length of the pipe, as done in Fig. 11. The series of
buoys in the hump of the riser generate a sequence of high DMMs in that region, as
shown in the zoom window of Fig. 11.
Finally, the riser response to the slug frequency is presented in Fig. 12, in the range
of 0.015 Hz to 0.020 Hz, for three nodes that lie in the zones of large amplitudes. The
608 S. N. Bordalo and C. K. Morooka

Table 2. Steel lazy wave riser case data

Riser length Lr [m] 4560
Horizontal projection Yp [m] 2969
Vertical projection Zp [m] 2129
Water depth Hw [m] 2140
TDP from the top projection Ytdp [m] 2033
TDP from the anchor Yp–Ytdp [m] 935.4
Suspended length Sr [m] 3626
External diameter De [m] 0.212
Internal diameter Di [m] 0.165
Elasticity modulus E [kN/m2] 207  106
Specific weight of pipe cs [kN/m3] 77.7
Slug cell length Lc [m] 139
Liquid length LL [m] 72
Gas length LG [m] 67
Specific weight of liquid cL [kN/m3] 6.76
Specific weight of gas cG [kN/m3] 4.63
Slug velocity vc [m/s] 2.32
Slug frequency fc [Hz] 0.0166
Slug period Tc [s] 59.9
TDP: touchdown point

full amplitude of vertical displacement “Zmax − Zmin” is plotted for three selected
nodes. Again, it is verified that the oscillation of the pipe increases as the slug fre-
quency approaches one of the natural frequencies of the riser; in this case, the full
amplitude at node 1000 reaches close to 2.8 De.
The present research included the simulation of many slug configurations, and may
be extended to many more as desired. However, due to space limitations in the paper
only one slug distribution was presented here for each riser. A comparison between
SCR and SLWR should be made for a more extensive range of operational conditions,
analyzing many possible slug flows resulting from various flow rates of liquid and gas
to be transported by the riser, during its operational life, and employing a design
criteria, probably weighted by economic values, to rate the SCR and SLWR designs.
The two examples given in this paper are just a representation of what one may find
amidst innumerous possible operational states in the set under analysis. These two
examples show how different a SCR and a SLWR behave, due to their dynamical
characteristics (natural frequencies etc.); therefore, each one may be better fitted for a
different design scenario: range of operation slug flow frequencies, flow rates, water
depth etc.
The purpose of the presented calculation examples was to demonstrate how to
include slug flows and fluid loads into the riser analysis, and to demonstrate the utter
importance of slug flows to the behavior of risers under production conditions.
 A Study of Riser Oscillations Caused by Slug Flows 609

Y-displacement - Node 0400 Z-displacement - Node 0400

1000
Y - Displacement (mm)

Z - Displacement (mm)
-5500

500 -6000

-6500
0 2000 2500
2000 2500
Time (s) Time (s)

(a)

Y-displacement - Node 2500 Z-displacement - Node 2500

Z - Displacement (mm)
Y - Displacement (mm)

19500 -30500

19000 -31000

18500 -31500
2000 2500 2000 2500
Time (s) Time (s)

(b)

Fig. 9. a. Time-history: node 0400, displacement in Y and Z. b. Time-history: node 2500,

displacement in Y and Z.

Bending Stress - Node 0400 Bending Stress - Node 2500

65 173
Bending Stress (MPa)

Bending Stress MPa)

63 171
61 169
59 167
57 165
55 163
2000 2500 2000 2500
Time (s) Time (s)

Fig. 10. Time-history: nodes 0400 and 2500, bending stress (external wall).
610 S. N. Bordalo and C. K. Morooka

SLWR: DMM - bending stress

20 20

15
DMM (MPa)

10
10

5 0
1700 1800 1900

0
0 1000 2000 3000 4000 5000
Position along the pipe (m)

Fig. 11. SLWR: bending stress DMM; slug frequency 0.0167 Hz (period 60 s). The lengthwise
coordinate runs from the anchor to the top of the riser.

Zmax - Zmin
700
Node
600 1000
Node
Zmax - Zmin (mm)

500
5000
400 Node
5300
300
200
100
0
15 16 17 18 19 20
Frequency (mHz)

Fig. 12. Effect of the slug frequency on the Z displacement (vertical) of the SLWR: nodes 1000,
5000 and 5300.

4 Conclusion

The results presented here provide strong evidence that slug flows may induce haz-
ardous oscillation to suspended petroleum risers, even in the absence of other excita-
tions, such as those incurred by sea currents, sea waves and motion of the sustaining
floating platform. The simulations show that a steady slug flow may alone produce
oscillatory displacements of the riser of rather significant amplitudes, without any other
external cause. The results also indicate that the oscillations are strongly dependent on
 A Study of Riser Oscillations Caused by Slug Flows 611

the slug frequency, especially when the slug flow frequency approaches one of the
pipe’s natural frequencies, incurring in relatively high bending amplitudes. Such results
raise concerns regarding the physical integrity of the pipeline due to fatigue damage
during long term operations.
This type of engineering problem will depend on the mechanical properties of the
pipeline and on the characteristics of the flow. No generic mathematical solution exists
in the literature, due to the complexity of the problem and the inherent non-linearity of
the governing equations.
However, the results indicate that the methodology presented here is promising, for
it opens a way for the engineer of subsea pipelines to analyze many different designs
under various internal flow conditions arising from the submarine oil wells. The pro-
posed methodology adds the internal flow loads due to the weight and momentum of
the fluid to the computer dynamic simulation, and considers the mass spatial distri-
bution of liquid and gas in slug flows. Consequently, this modified simulator is flexible
enough to compute the effect of a wide range of slugs’ configurations – geometry,
fluids types, production flow rates of liquid and gas etc.; allowing the analysis of
fatigue damage and risk of failure under many production conditions.
The flow model for steady regular periodic slugs employed in this paper works well
and should be quite useful for the designer of pipelines as it is. However, it is indeed a
“first order” approximation in the sense that it recreates the major features of the flow
while leaving a few issues out, such as the more detailed slug features (noses, tails,
dispersed bubbles and mist) and transient behaviors of the internal flow that appear
under some circumstances; for instance, irregular slugs, transient slugging, streamwise
slugs variation with pressure or pipe inclination, flow pattern transition, churning etc.
These cases would require further work to tailor the current computational models
for them. Still, one might keep the present methodology to track the liquid and gas
masses as they would evolve along the pipe, and add these flow details to the moving
mass distribution employed in the present work.
Finally, in an offshore oil field scenario, the study of the internal flow forces should
be combined with loadings from sea currents, ocean waves, motion of floating units
connected at the top of the riser, and their concurrent effects, such as drag, vortex
induced vibrations etc. The interaction between any two of these categories of loads is
very complex; the combination of each effect is nonlinear, that is, they cannot be
computed independently and simply superposed.
The time varying curvature of the pipeline induced by movement of the FPU, or by
sea currents and waves, would originate time varying curvature loads on a riser due to
the internal flow. The modeling algorithms employed in the present work may be
inserted in almost any pipeline dynamics simulator with the proper adaptations, in
order to carry out coupled simulations of the internal flow with all other external
induced loads, and to calculate the combined damage.

Acknowledgment. Special thanks are due to Dr. Ricardo Franciss and Dr. Marcio M. Mourelle
(Petrobras/Cenpes), for sharing their knowledge about Anflex© and helpful comments on risers
dynamics.
612 S. N. Bordalo and C. K. Morooka

References
Beggs, H.D., Brill, J.P.: Two-Phase Flow in Pipes, 5th edn., Tulsa (1986)
Bordalo, S.N., Morooka, C.K., Cavalcante, C.C.P., Matt, C.G.C., Franciss, R.: Whipping
phenomenon caused by the internal flow momentum on the catenary risers of offshore
petroleum fields. In: 27th International Conference on Offshore Mechanics and Artic
Engineering. ASME, Estoril (2008)
Bordalo, S.N., Morooka, C.K., Oliveira, A.P.: Modeling the forces induced by the two-phase
flow on deep water production risers. In: SPE Artificial Lift Conference - Latin America and
Caribbean. SPE, Salvador (2015)
Bossio V, B.M., Blanco A, A.J., Casanova M, E.L.: Numerical modeling of the dynamical
interaction between slug flow and vortex induced vibration in horizontal submarine pipelines.
J. Offshore Mech. Arctic Eng. 136(4) (2014). (Online July, printed November, ASME)
Chakrabarti, S.K., Frampton, R.E.: Review of riser analysis techniques. Appl. Ocean Res. 4(2),
73–90 (1982)
Chakrabarti, S.K.: Hydrodynamics of Offshore Structures. Computacional Mechanics Publica-
tions, Southampton (1987)
Dukler, A.E., Hubbard, M.G.: A model for gas-liquid slug flow in horizontal and near horizontal
tubes. Ind. Eng. Chem. Fundam. 14(4), 337–347 (1975)
Fossa, M., Guglielmini, G., Marchitto, A.: Intermittent flow parameters from void fraction
analysis. Flow Measur. Instrum., (14), 161–168 (2003)
Gonzales, E.C., Mourelle, M.M., Mauricio, J., Lima, T.G., Moreira, C.C.: Steel catenary riser
design and analysis for Roncador field development. In: Annual Offshore Technology
Conference (OTC), Houston. EUA (2005)
Govier, G.W., Aziz, K.: The Flow of Complex Mixtures in Pipes, 2nd edn. SPE - Society of
Petroleum Engineers, Richardson (2008). (ISBN 9781555631390)
Gregory, G.A., Scott, D.S.: Correlation of liquid slug velocity and frequency in horizontal
cocurrent gas-liquid slug flow. AIChE J. 15(6), 933–935 (1969)
Gundersen, P., Doynov, K., Andersen, T., Haakonsen, R.: Methodology for determining remnant
fatigue life of flexible risers subjected to slugging and irregular waves. In: 31st International
Conference on Offshore Mechanics and Artic Engineering. ASME, Rio de Janeiro (2012)
Heywood, N.I., Richardson, J.F.: Slug flow of air-water in a horizontal pipe: determination of
liquid holdup by c–ray absorption. Chem. Eng. Sci. (34), 17–30 (1979)
Jia, D.: Slug flow induced vibration in a pipeline span, a jumper and a riser section. In: Offshore
Technology Conference, May, Houston, Texas, USA (2012)
Johnson, R.W.: Handbook of Fluid Dynamics, 2nd edn. CRC Press (2016). (ISBN
9781439849552)
Michaelides, E.E., Crowe, C.T., Schwarzkopf, J.D.: Multiphase Flow Handbook, 2nd edn. CRC
Press (2016). (ISBN 9781498701006)
Moe, G., Chucheepsakul, S.: The effect of internal flow on marine risers. In: 7th International
Conference on Offshore Mechanics and Artic Engineering, February, Houston, Texas, USA
(1988)
Morooka, C.K., Tsukada, R.I.: Experiments with a steel catenary riser model in a towing tank.
Appl. Ocean Res. 43, 244–255 (2013)
Mourelle, M.M.: Dynamic analysis of structural systems composed of risers and anchor lines. Ph.
D. thesis, Federal University of Rio de Janeiro, Rio de Janeiro, Brazil, December 1993
Ortega, A., Rivera, A., Larsen, C.M.: Flexible riser response induced by combined slug flow and
wave loads. In: 32nd International Conference on Offshore Mechanics and Artic Engineering.
ASME, Nantes (2013)
 A Study of Riser Oscillations Caused by Slug Flows 613

Paidoussis, M.P.: Dynamics of tubular cantilevers conveying fluids – part 1: theory. J. Mech.
Eng. Sci. 12(2), 85–96 (1970). (Plus: Part 2: Experiments. J. Mech. Eng. Sci. 12(2), 96–113)
Paidoussis, M.P.: Fluid-Structure Interactions - Volume 1: Slender Structures and Axial Flow.
Elsevier (1998). (ISBN 978-0-12-544360-9)
Patel, M.H., Seyed, F.B.: Internal flow induced behaviour of flexible risers. Eng. Struct. 11(4),
266–280 (1989)
Pollio, A., Mossa, M.: A comparison between two simple models of a slug flow in a long flexible
marine riser. ARXIV, Physics – Fluid Dynamics, 0911.1873 (2009). (Cornell U., November,
Ithaca, NY, EUA. Also in: Flexible pipe behaviour investigation using two models of internal
slug flow regime. FLUCOME 2007, Tallahassee, Florida, USA)
Reda, A.M., Forbes, G.L., Sultan, I.A.: Characterization of dynamic slug flow induced loads in
pipelines. In: 31st International Conference on Offshore Mechanics and Artic Engineering.
ASME, Rio de Janeiro (2012)
Rodrigues, H.T., Morales, R.E.M., Mazza, R.A., Rosa, E.S.: A comparative study of closure
equations for gas-liquid slug flow. In: 19th COBEM – International Congress of Mechanical
Engineering. ABCM, Brasilia, November 2007
Safrendyo, S., Srinil, N.: Slug flow-induced oscillation in subsea catenary riser experiencing
VIV. In: 37th International Conference on Ocean, Offshore and Arctic Engineering. ASME,
Madrid (2018)
Shoham, O.: Mechanistic Modeling of Gas-Liquid Two-Phase Flow in Pipes. SPE - Society of
Petroleum Engineers, Richardson (2006). (ISBN 9781555631079)
Sparks, C.P.: Fundamentals of Marine Risers Mechanics: Basic Principles and Simplified
Analysis. PennWell, Tulsa (2007)
Su, J.: Vibration behavior of marine risers conveying gas-liquid two-phase flow. In: 34th
International Conference on Offshore Mechanics and Artic Engineering. ASME, St. John’s
(2015)
Van der Heijden, H.E.J., Smienk, H., Metrikine, A.V.: Fatigue analysis of subsea jumpers due to
slug flow. In: 33rd International Conference on Offshore Mechanics and Artic Engineering.
ASME, San Francisco (2014)
Wu, M.C., Lou, J.Y.K.: Effects of rigidity and internal flow on marine riser. Appl. Ocean Res. 13
(5), 235–244 (1991)
Yamamoto, M., Murai, M.: A numerical analysis of the internal flow effect on the riser’s
mechanical behavior. In: 29th International Conference on Ocean, Offshore and Arctic
Engineering. ASME, Shanghai (2010)
Zabaras, G.J.: Prediction of slug frequency for gas-liquid flows. SPE J., 252–258 (2000)
 A Simplified Nonlinear Analysis Procedure
for the Flexible Risers in Subjected
to Combined Loading Effects

Jeong Du Kim1 and Beom-Seon Jang2(&)

1
Department of Naval Architecture and Ocean Engineering, Seoul National
University, Seoul, South Korea
2
Research Institute of Marine System Engineering, Department of Naval
Architecture and Ocean Engineering, Seoul National University,
Seoul, South Korea
[email protected]

Abstract. A dynamic analysis of flexible risers in large-scale domain remains

the offshore industry’s most challenging problems. The cross section of flexible
risers, consisting of multiple metallic and plastic layers, experiences inter-layer
slippage during cyclic bending. This results in nonlinear and hysteretic bending
behavior with rapid dissipation of energy. For the dynamic analysis for flexible
risers in the large scale domain, this study replaces all layers of risers with
simple beam elements with a nonlinear beam model. This simplified beam
model traces the curvature to bending moment relation and captures the varia-
tion of stiffness due to the onset of slippage. The suggested methods formulate
both the change of configuration due to large deflection and contact between
risers and seabed, therefore more realistic and effective analysis can be achieved.
In the current study, the details of the analysis method are described. Main
formulation methods of the total Lagrangian framework and the interaction
between risers and seabed are described as well as the nonlinear bending model
for flexible risers. The validity of the developed method is examined through
some case studies, and the results show how the bending nonlinearity of flexible
risers affects the overall behavior of riser systems during its operation.

Keywords: Flexible riser
Dynamic analysis
Riser-seabed interaction

1 Introduction

Flexible risers are widely used pipelines in the offshore industry to transfer oil products
from the well to the platform. The metallic layers of flexible risers are wounding the
internal bores with helical shape, in which the interlayer slippage between the layers
maximizes the bending flexibility in critical locations such as touch down zone. Recent
developments of high-tensile wires further contribute to the extension of service life
and to the prevention of radial buckling. Although these advantages, the design of
flexible risers requires uncertain and time-consuming from the early development
phases. The helical wires inside tensile armor wires generally exert cyclic stick-
slippage phenomena, therefore the cross-section of flexible risers undergo rapid energy

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 614–630, 2021.
https://doi.org/10.1007/978-981-15-4680-8_42
 A Simplified Nonlinear Analysis Procedure for the Flexible Risers 615

dissipation during bending. This makes it hard to predict the dynamic behavior of
flexible risers both in the local-scale and large-scale domain.
Many studies have dealt with the dynamic analysis of flexible risers. Feret and
Bournazel (1987) formulated the axial and radial displacements of the riser subjected to
tension and pressure loads. They derived an explicit form for the stress and displace-
ment fields for each layer, in which the overall fields are expressed by several cross-
sectional variables. In their works, this equation provides a constitutive matrix so that
the static problem of small-scale flexible risers can be linearized. Witz (1996) examined
the effectiveness of theoretical approaches through some experiments on the internal
diameter of 10-in. flexible risers. The validation against the results obtained by theo-
retical approaches showed that theoretical approaches provide a proper model for
flexible risers. Feret et al. (1995) assumed that movements of helical wires always
follows the torus surface of their contact surface. Accordingly, the normal stress of
wires is expressed as a differential form using the Darboux frame. With the derived
stress field, the authors accomplished to take into account the interlayer slippage for the
moment calculation. Kebadze and Kraincanic (1999) presented a similar approach in
which the deformed helical wires are assumed to follow the helical path. He treated
tensile armor layers as a homogeneous tube, presuming that all helical wires are
contacted with surrounding layers until a certain threshold. The authors noted that each
wire loses its friction after a critical curvature, so the nonlinear bending hysteresis curve
can be obtained by deriving the equilibrium of sliding force and friction.
The dynamic analysis for flexible risers has been another interest among many
researchers. Fylling and Bech (1991) investigated the effect of the nonlinear bending on
the dynamic behavior of flexible risers. They found that friction along the contact
surface of helical wires dominates curvature responses. Dib et al. (2013) employed a
nonlinear dynamic substructuring technique to the dynamic analysis of flexible risers.
The presented method quickly estimates the responses of all riser segments based on
the several deformation modes and the coordinate transformation between local and
large scale model.
Previous researchers have suggested various methods in the prediction of dynamic
responses of flexible risers. However, in the previous studies, many simplifications are
made in the modeling of flexible risers and the detailed effects of the interlayer slippage
have been overlooked. In view of this, some improvements are still required to achieve
an accurate dynamic simulation of flexible risers.
In order to address the current issue, a new dynamic analysis method for flexible
risers, referred to as an OPFLEX, are being developed. The purpose of this develop-
ment is to take into account the interaction between helical wires and surrounding
layers in the global dynamic analysis, employing an accurate nonlinear bending model
in the modeling of riser systems. The nonlinear model of each element captures the
onset of the slippage during dynamic analysis, therefore the bending behavior of
flexible risers can be simulated in the large-scale domain. The element of riser models
is formulated based on the Euler beam model. The deflection of this model is basically
solved under the total Lagrangian framework with a high order strain and stress basis,
therefore both the large deflection and nonlinear bending effects of flexible risers can be
incorporated into the solver. This study, as a first phase of the developments, aims at
the validation of the presented dynamic analysis solver and the investigation of the
616 J. D. Kim and B.-S. Jang

influence of the interlayer slippage of flexible risers on the dynamic behavior of riser
configurations.
The present study unfolds as follows. Firstly, the details of the formulation of the
dynamic analysis solver will be introduced. The static and dynamic analysis method
based on the total Lagrangian framework and time marching scheme estimates the
structural behavior of riser structures subjected to various load conditions. Then, the
theoretical approach for flexible risers and its implementation on Euler beam models
are introduced. Finally, the influence of dynamic responses of flexible risers will be
investigated through the comparative study.

2 Detailed Formulation of Riser Dynamic Analysis

2.1 Total Lagrangian Formulation for Euler Beam

The dynamic analysis of risers generally requires a large displacement analysis due to
its geometric characteristics. In the large-scale domain, the vibration of risers can be
predicted with the catenary equations, but this neglects the bending stiffness of each
riser segments, causing some inaccuracy during time domain analysis.
In order to describe the kinematics of the risers experiencing large displacements,
the present work utilizes the total Lagrangian formulation in the time-domain dynamic
analysis. This method provides an exact strain and stress basis by relating the current
and initial configuration. The strain components of the body subjected to configuration
changes are described as Eq. (1). The strain components, Eij , referred to as Green-
Lagrange strain tensor provides a consistent differential operator to obtain the magnitude
of deformation in the body. The stress conjugate to this stress is called as Piola-Kirchoff
2nd stress tensor. Both the strain and stress components are symmetric regardless of the
coordinate since the strain measurement is derived from the deformation gradient tensor.


1 @ui @uj @uk @uk
eij ¼ þ þ ð1Þ
2 @xj @xi @xi @xj

sij ¼ Cijkl ekl ð2Þ

Where:
i; j; k : base unit in initial configuration
u : displacement in initial configuration.
x : spatial coordinate in initial configuration
Cijkl : constitutive matrix.

In the total Lagrangian formulation, the framework remains consistent since the
effects of changes in configuration are already formulated in the formulation of the
strain and stress tensor. The real virtual works of the deformed body, as shown in
Eq. (3), is energetically equal to that obtained with Green-Lagrange strain and Piola-
Kirchoff 2nd stress, note Eq. (4). The changes in configurations are successfully
 A Simplified Nonlinear Analysis Procedure for the Flexible Risers 617

expressed in the fixed forms by rebuilding the strain and stress. Nevertheless, the
integration of internal energy still contains higher order terms, which makes the
problem nonlinear to displacement fields. The complete form of virtual work, Eq. (4),
thus linearized and re-expressed in iterative form, note Eq. (5).
Z
sIJ deIJ dV ¼ R ð3Þ
V
Z
t þ Dt
sij dt þ Dt eij dv ¼ t þ Dt R ð4Þ
vt þ D
Z Z Z
Dt
Cijkl Dt
0
ekl dDt
eij dv þ t Dt
0 Sij d0
gij dv ¼ t þ Dt R  Dt
0 Sij d0

t
eij dv ð5Þ
v v v

Where:
Dt
 

e : linear strain increment, 12 Dt ui;j þ Dt uj;i þ t uk;i Dt Dt
0 uk;j þ uk;i uk;j
t

t : time
v; V : volume of body in initial, deformed configuration, respectively
Cijkl : constitutive matrix.
I; J : base unit in deformed configuration
R : external force
Dt : time increment
e; e : Cauchy strain and Green-Lagrange strain
Dt

g : nonlinear strain increment, 12 Dt uk;i Dt uk;j
s; S : Cauchy stress and Piola-Kirchoff2nd stress

The riser element model is formulated on the base of total Lagrangian as shown in
Fig. 1. The element contains two nodes having 6-degree of freedom and the third order
Hermite polynomials are utilized to shape functions, Fig. 1(a). The solution algorithm
proceeds with the Newton-Raphson algorithm in which the stiffness and internal
vectors are iteratively updated until the equilibrium with external force, Fig. 1(b).

Fig. 1. Kinematics of riser segments in total Lagrangian formulation: (a) 6-degree of freedom
element and (b) force equilibrium during iterative solution
618 J. D. Kim and B.-S. Jang

The test result for the validation of solver is depicted in Fig. 2. The test structure
has a fixed clamp in the left end and subjected to vertical load until the end section
reaches vertical Fig. 2(a). The exact solution to this problem is well derived by Chen
(2010). He suggested an integral form for the cantilever structure under the large
deflection and the presented equation showed good agreement to the experiment
results. The numerical analysis results on the deflection of 120 m cantilever beam
structure are graphed on and compared with exact solutions, Fig. 2(b). The result
shows that the present formulation well predicts the large deflection of slender, even
though the midsection vertically folds with severe deflection.

Displacement graph, 120m cantilever

(OPFLEX vs. exact solution)
y N ⋅ Punit 1.2

1
θm
Non-dimensional disp.

0.8

0.6
δ
0.4

0.2 l/L (Chen, 2010) δ/L (Chen, 2010)

sinθm (Chen, 2010) l/L (OPFLEX)
l 0
δ/L (OPFLEX) sinθm (OPFLEX)
0.0 5.0 10.0 15.0 20.0
L K=PL2/EI
(a) (b)

Fig. 2. Large deflection test result: (a) cantilever beam for test and (b) result of validation
against exact solution

2.2 Seabed Interactions

The interaction between riser structures and seabed results the intensive bending stress
along the touchdown region, whereby the normal force induced by the penetration of
the riser into seabed surface should be reflected into the static and dynamic analysis. In
the present work, the riser-seabed interaction model is established under the basis of the
node to a contact algorithm.
The contact algorithm reflects the interaction between the riser structures and
seabed in an iterative way. During the dynamic analysis, the algorithm measures the
clearance distance of all nodes to search for potential contact regions. The clearance
distance is defined as the vertical distance between the node and the seabed. When
some nodes become below the elevation of the seabed, the algorithm supplements the
normal stiffness and additional restoring force induced by the seabed.
Figure 3 illustrates how the contact algorithm reflects the interaction. The normal
force on the beam element after the penetration is in proportional to the contact area of
 A Simplified Nonlinear Analysis Procedure for the Flexible Risers 619

Fig. 3. Riser-seabed interaction model: (a) estimation of contact surface and (b) pure penalty
method

the outer surface. Presuming that the cross-section of flexible risers maintains a small
ovality, the surface area is in proportional to the penetration depth of the element. The
corresponding contact force is derived by the multiplication of the contact surface with
normal stiffness of the given seabed. This value determines the additional normal
stiffness per unit length, so the beam element subjected to the penetration re-formulate
the constitutive matrix.
The normal stiffness of each beam element is determined based upon each node’s
penetration status, as indicated in Fig. 3(b). Individual cases categorize the potential
penetration status of beam elements during the analysis. The first case represents that
two nodes are buoyant state, in which the element maintains the positive seabed
clearance. On the contrary, case 2 shows a partially penetrated beam element, where
some part of the element is penetrating the seabed but the other part maintains the
clearance. In this case, the area of the penetrating part is approximated by the vertical
displacement of two nodes and the corresponding bending stiffness is assigned in
accordance with the penetration state. The buoyant part of the element is not affected
by the seabed, but the normal stiffness is also supplemented to this part since the beam
element contains two nodes. The last case indicates the full penetrating state, which
indicates that two nodes are below the elevation of the seabed and their downward
displacement is controlled by the seabed stiffness. Likewise in case 2, the normal
stiffness and the restoring force from seabed highly depend on the contact area and the
depth of the penetration. Accordingly, the distribution of contact surface and normal
stiffness along the length of the beam element is approximated by the position of two
nodes.
The seabed stiffness of the beam elements are assumed to exert the restoring force
in the normal direction of the axis, therefore the supplemented matrix is rotated in
accordance with the rotated angle. At the same time, the restoring force from the
interaction between the seabed and riser structures are directly estimated by multiplying
620 J. D. Kim and B.-S. Jang

the penetrated displacement with the generated seabed stiffness in each iteration.
Therefore, the solution algorithm continuously updates the contact states until it
reaches equilibrium states.

Fig. 4. Seabed interaction benchmark test

The simple test on contact algorithm is conducted as shown in Fig. 4. The test
model is a simple pipeline structure having the length of 120 m and laid out straight
towards the axial direction. It is identified that the current contact algorithm shows
good agreement to that from the commercial program, Orcaflex.
The accuracy of developed solver accompanying the interaction of riser and sea-
bed is validated against the commercial dynamic analysis software, Fig. 5. The test is a
vertical upload of 120 m steel riser with 400 ton on the free-hang point. The seabed has
60 kN normal stiffness considering the property of a mud. The results computed by the
current solver show good agreement with those by Orcaflex. Although the suggested
riser-seabed interaction method, differently from Orcaflex, is based on the pure penalty
method, it is concluded that the solver can obtain the distribution of cross-section
resultant without inaccuracy.

2.3 Time Marching Scheme

The dynamic analysis of riser structures normally entails various nonlinear character-
istics induced by structural and environmental causes. In example rapid rotation of riser
elements generates continuous changes in global stiffness, therefore they undergo
dynamically non-Gaussian vibration. These intrinsic nonlinearities of riser structures
necessitate the use of direct time marching schemes rather than modal analysis.
Therefore, the current research utilizes the Newmark-b scheme to integrate the time
derivatives of displacement fields. The Newmark-b scheme formulates the governing
equations by approximating the unknown values from several critical points. The
velocity and displacement at the time, ti þ 1 , is approximated from the acceleration,
Eq. (6) and (7). The dynamic equation of the system, Eq. (8), is therefore denoted in
terms of displacement increments considering the inertia and damping effects, Eq. (9).
 A Simplified Nonlinear Analysis Procedure for the Flexible Risers 621

Fig. 5. Results of benchmark test against Orcaflex: (a) deformed shape, (b) shear force,
(c) tension, and (d) bending moment

  
1
xðti þ 1 Þ ¼ xi þ Dtx_ i þ Dt2  b €xi þ b€xi þ 1 ð6Þ
2

x_ ðti þ 1 Þ ¼ xi þ Dt½ð1  cÞ€xi þ c€xi þ 1  ð7Þ

M€xðtÞ þ Cx_ ðtÞ þ KxðtÞ ¼ f ð8Þ




6 3 2 6
þ C þ K dx i ¼ df  dR þ 3M þ C €
x i þ M þ 3C x_ i ð9Þ
Dt2 Dt Dt Dt

Where:
f : external force
x; x_ ; €x : displacement, velocity, acceleration
C : damping coefficient
K : stiffness
M : inertia coefficient
b : beta-coefficient
c : gamma-coefficient
Dt : time increment
622 J. D. Kim and B.-S. Jang

The schematics of the nonlinear dynamic analysis process is described in Fig. 6.

From every time increment, the dynamic solver assumes the initial displacement and
relevant velocity and acceleration data. The stiffness is updated under the recursive
process to meet both the dynamic and contact equilibrium. The process continuously
updates every condition of riser status until the residual force bellows a predetermined
convergence level.

Fig. 6. Schematics of nonlinear dynamic analysis process

2.4 Nonlinear Bending Model for Flexible Risers

The contribution of helical wires on the cross section resultant is categorized by the
contact status between tensile armor and adjacent anti-friction tape. The helical wires
generate bending stress under the relatively low level of curvature. After the overall
curvature exceeds the critical level, referred to as critical curvature, some part of wires
loses their frictional forces due to the excessive sliding force. The contribution of
helical wires to bending moment consequently diminishes and the bending stiffness of
the riser section only depends on the other layers.
The theoretical model proposed by Kraincanic and Kebadze (2001) provides an
adequate solution to obtain the critical curvature. When helical wires wounding the
internal anti-friction tape bent under the critical curvatures the friction forces over-
comes the sliding force along the wires, leading the elevation of bending stress. The
normal stress of helical wire is derived as Eq. (10). The sliding force along the wires is
identical to the differential of axial force towards the helical path, Eq. (11). The slip of
 A Simplified Nonlinear Analysis Procedure for the Flexible Risers 623

helical wires initiates when the sliding force just overcomes the friction force. The
friction force on the helical wires in helical direction is in proportional to both side
contact force, Eq. (12). Thus, the critical curvature is the equilibrium point of sliding
force and friction force, Eq. (13).

e0 ¼ Rsin2 asinhj b ð10Þ

R
dN0 ¼ Ebtcos2 acoshb dh ð11Þ
sina
 R
dF ¼ P1 k1fr þ P2 k2fr b dh ð12Þ
sina
  
bcr ¼ P1 k1fr þ P2 k2fr = Etsinacos2 cosh ð13Þ

Where:
b : breadth of wire
t : height of wire
k1fr ; k2fr : friction coefficient
E : modulus of elasticity
F : friction force
N0 : normal stress on the section of helical wires
P1 ; P2 : contact pressure on the tensile armor layer
R : radius of tensile armor layer
e0 : normal strain on the section of helical wires
b : curvature
bcr : critical curvture
a : laying angle
h : angular position of wire.

The contact pressure which restricts the slip of helical wires can be estimated
through the static analysis of the riser section under the axisymmetric load. The radial
responses for internal and external pressure are obtained based on Eqs. (14)–(17).
The internal strain energy of the polymer sheath is treated as a homogeneous tube,
Eq. (14). The overall behavior of internal carcass and pressure armor having highly
twist-ed profiles can be approximated by alternating the layer as an orthotropic tube, as
shown in Eq. (15). The constitutive matrix for these layers shall be determined in
accordance with the FE result of the theoretical approaches. The strain energy of tensile
armor layers is calculated through the integration of infinitesimal energy along the
cross-section surface, see Eq. (16). The stress of the surface of helical wires is the
multiplication of modulus of elasticity and strain expressed in Eq. (10).
624 J. D. Kim and B.-S. Jang

Z
1 E  2 
Uiso: ¼ Ee1 þ 2vEe1 e2 þ Ee22 þ Gc212 dv ð14Þ
v 1v
2 2

Z

1 C11 e21 þ 2C12 e1 e2 þ 2C16 e1 c12
Uortho: ¼ dv ð15Þ
2 v þ C22 e22 þ 2C26 e2 c12 þ C26 e2 c12
Z
1 Xm
Uwire ¼ r1 e1 dv ð16Þ
2 j¼1
v

Where:
C : constitutive matrix
E : modulus of elasticity
c : normal shear strain
e : normal strain
r : normal stress.

Fig. 7. Validation results of nonlinear bending model for 7-in. internal diameter flexible riser

The nonlinear bending model based on the theoretical approach is validated against
FEA. The bending hysteresis curve for 7-in. internal diameter flexible riser subjected to
internal and external pressure is shown in Fig. 7. The result computed by the bending
model shows good agreement to that from FEA. It is concluded that the nonlinear
bending model can be considered into the dynamic analysis in a large-scale approach.
The bending nonlinearity of flexible risers is adopted on the dynamic analysis in
accordance with the update process shown in Fig. 8. The element stiffness in bending
contains stick and slip state value whereby the dynamic solver changes the element
property depending on the nodal curvature. The element maintains the high resistance
in bending when the nod-al curvature stays below the critical curvature. Otherwise, the
helical wires are treated as the slip state, and the iterative update process reassigns the
internal force with a reduced bending moment.
 A Simplified Nonlinear Analysis Procedure for the Flexible Risers 625

Fig. 8. Update process of nonlinear bending during dynamic analysis

3 Dynamic Analysis Test

In this study, the dynamic analysis considering the nonlinear bending stiffness of
flexible risers is demonstrated. Furthermore, the effect of nonlinearity in the large-scale
analysis is investigated by comparing the results from the nonlinear model with those
from the linearized beam model. The nonlinear bending property originates the pres-
sure distribution in the seabed, thus the riser element changes its bending stiffness after
the critical curvature. The test is conducted for the internal diameter of 7-in. flexible
riser. The details of the test riser are described in Fig. 9.

Fig. 9. Details of internal diameter of 7-in. flexible riser

3.1 Pure Vibration Test

Pure vibration test on the relatively small size of the flexible riser is conducted as
Fig. 10. The right end side of the riser is clamped, which causes progressive bending
moment growth along the length. The left side, end tip, is subjected to the dynamic
626 J. D. Kim and B.-S. Jang

vertical force with 1.0 kN and 10 s period, Fig. 10(a). The linear bending model
assumes that no friction between the helical wires and the other layers. Consequently,
the bending stiffness of this model is consistent regardless of exposed curvature, and
only the geometric nonlinearity is involved during the dynamic analysis. Meanwhile,
the non-linear bending model maintains relatively high bending stiffness unless the
riser element exceeds the critical curvature. After the curvature grows more than the
critical curvature, the bending stiffness dramatically reduces.

Fig. 10. Schematics of pure vibration test: (a) Test model and bending property of comparing
two models, (b) dynamic response at the end tip

The time history of the vertical displacement at the end tip shows the effect of
bending nonlinearity of the flexible riser, see Fig. 10(b). The dynamic responses of
linear bending model smoothly increase with the low bending stiffness. In case of the
nonlinear bending model, unlike to those of linear bending model, the displacement of
the end tip remains relatively lower than in the bending model since its bending
property. The dynamic responses suddenly increase after the critical curvature, due to
the exceedance of curvature level.

3.2 Free-Hang Vibration Test

Free-hang vibration test is demonstrated to test the dynamic behavior of flexible riser
interacting the seabed. Linear and nonlinear bending model also used in this test for the
investigation. The models have 120 m length and dynamically loaded by using the
dynamic boundary condition at the fairlead.
The initial configuration of the risers are obtained through the static analysis in
order to reflect the actual installation procedure, Fig. 11. The riser structure is first laid
down on the seabed surface which set to be 300 m from the mean water line. The
deformed state of the riser with its weight is calculated, therein most force from the
 A Simplified Nonlinear Analysis Procedure for the Flexible Risers 627

riser weight is supported by the seabed. After the weight is loaded the vertical
uploading of fairlead point is conducted. Using the Lagrange multiplier, the position of
the fairlead point is adjusted to the predetermined value. Fairlead point is finally
adjusted to meet the user-determined value. The dynamic analysis initiates when the
initial configuration of the riser is obtained.

Fig. 11. Load sequence in analysis test

The static responses of nonlinear bending model and the linear bending model are
compared in Fig. 12. The configuration after the vertical uploading of fairlead and
horizontal adjusting is represented in Fig. 12(a) and (b), respectively. The linear model,
different from the nonlinear bending model, shows much different deformed shape
across the touchdown zone. This implies that even though both models have an
identical property in bending after the critical curvature, the nonlinear bending model
results in some difference from linear bending model.

Fig. 12. Comparative study results: (a) static analysis results after vertical uploading and
(b) static analysis result after horizontal adjusting
628 J. D. Kim and B.-S. Jang

Some discrepancies between two models are also identified in bending hysteresis
curve, Fig. 13. The curvature of the comparing models increases during the vertical
uploading and decreases while the horizontal adjusting stage. The cross-section
bending moment in the nonlinear bending model grows much higher than the linear
bending model due to the contribution of the friction stress in helical wires. The slope
in bending moment reduces after the critical curvature since the friction between the
helical wires and other layers is neglected based on the slip initiation.

Fig. 13. Bending hysteresis curve of riser segment near touch down zone

Time domain history of critical points is shown in Fig. 14. The deformed shape
after the horizontal adjustment in the nonlinear bending model shows different results
than in the linear bending model. The results show that the nonlinear bending model
maintains stable vibration during the dynamic analysis. Meanwhile, during the

Fig. 14. Vertical displacement time domain history: (a) displacement near touch down zone and
(b) displacement near fairlead
 A Simplified Nonlinear Analysis Procedure for the Flexible Risers 629

vibration, the amplitude of displacement in the nonlinear bending model is almost

similar to that of linear bending model. Different from those in the pure vibration test,
the range of dynamic responses of two models rather similar to each other. Observing
the vertical displacement of the sample point near the touchdown zone, the nonlinear
bending model results in approximately 0.033 m of vertical vibration. The linear
bending model shows about 0.038 m of vertical vibration near the touchdown zone,
which seems similar to the results of the nonlinear bending model.
It is concluded that the riser vibrating under relatively small amplitude loading
results in similar dynamic responses because the curvature responses remain small, and
the curvature level of most element exceeds the critical curvature level.

4 Conclusion

The current study has dealt with the development of the global dynamic analysis for
flexible risers. Based on the theoretical approaches, the dynamic analysis solver for
flexible risers, OPFLEX is developed and demonstrated for the validity. The beam
elements of riser models are generated by the total Lagrangian formulation to describe
the variation of configuration due to large deflections. To model the detailed effects of
the interaction between risers and seabed, the contact algorithm is developed and
involved in the analysis procedure. During the analysis, the contact algorithm updates
the contact status of every beam element and restoring force induced by the penetra-
tion, so the movements of risers are fully coupled with the seabed. The nonlinear
behavior of flexible risers is modeled by updating the property of beam elements in
accordance with the critical curvature. When the curvature of beam elements exceeds
their threshold, the bending stiffness rapidly decreases considering that all helical wires
undergo slippage. For the validity of the developed analysis solver, several case studies
are carried out and the results are discussed. Through the case studies, it is identified
that the nonlinear bending behavior of flexible risers reduces the curvature responses
until they overcome the critical curvature but maintains similar responses to that of the
linearized model after the slippage. The suggested method is still under development.
The nonlinear bending model of beam element should be improved to describe pro-
gressively decreasing bending stiffness and be coupled with rapidly changing axial
force.

References
Chen, L.: An integral approach for large deflection cantilever beams. Int. J. Non-Linear Mech. 45
(3), 301–305 (2010)
Dib, M.W., Cooper, P.A., Bhat, S., Majed, A.: A nonlinear dynamic substructuring approach for
efficient detailed global analysis of flexible risers. In: Offshore Technology Conference.
Offshore Technology Conference, May 2013
Feret, J.J., Bournazel, C.L.: Calculation of stresses and slip in structural layers of unbonded
flexible pipes. J. Offshore Mech. Arct. Eng. 109(3), 263–269 (1987)
630 J. D. Kim and B.-S. Jang

Feret, J., Leroy, J.M., Estrier, P.: Calculation of stresses and slips in flexible armor layers with
layers interaction (No. CONF-950695). American Society of Mechanical Engineers,
New York (1995)
Fylling, I., Bech, A.: Effects of internal friction and torque stiffness on the global behavior of
flexible risers and umbilicals. In: Proceedings of the 10th International Conference on
Offshore Mechanics and Arctic Engineering (1991)
Kebadze, E., Kraincanic, I.: Non-linear bending behaviour of offshore flexible pipes. In: The
Ninth International Offshore and Polar Engineering Conference. International Society of
Offshore and Polar Engineers (1999)
Witz, J.A.: A case study in the cross-section analysis of flexible risers. Mar. Struct. 9(9), 885–904
(1996)
 Modal Parameter Identification of an Electrical
Submersible Pump Installed in a Test Well
Using Drop Tests

Ulisses A. Monteiro1(&) , Ricardo S. Minette2,

Ricardo H. R. Gutiérrez1 , and Luiz A. Vaz1
1
Department of Naval and Ocean Engineering,
Federal University of Rio de Janeiro (UFRJ), Rio de Janeiro, Brazil
{ulisses,rhramirez,vaz}@oceanica.ufrj.br
2
Center for Research and Development (CENPES/PETROBRAS),
Rio de Janeiro, Brazil
[email protected]

Abstract. Electrical submersible pumps (ESPs) are used in the petroleum

industry to pump large amount of fluids from subsea deep wells. Currently,
ESPs are responsible for approximately 10% of the world’s crude oil production.
When high flow rates are required, gas lift and the ESP are the only available
technologies that can be applied. ESP systems are much more complex and less
reliable than the gas lift method, but they are more efficient and able to yield
higher flow rates and pressure. ESPs are installed inside or near production
wells, with the entire auxiliary infrastructure for power supply and control
system. This means that maintenance is prohibitive because of production losses
and the need for expensive and unavailable drill ships. In order to increase
ESP’s reliability, is highly important to know the equipment dynamic behavior
while it is operating. The objective of this work is to identify the modal
parameters of an ESP with 62 stages and 10.42 m in length, weighing 616.8 kg
and installed 9.02 m deep in a test well circuit using water as working fluid.
Drop tests were carried on and the vibration responses collected by water proof
accelerometers were used to perform an output-only modal analysis using the
Enhanced Frequency Domain Decomposition (EFDD) technique. Results were
compared with modal parameters obtained using impact tests and showed that
the drop test is an useful method for modal testing of ESP’s in test wells.

Keywords: Electrical submersible pump
Test well
Drop test
Output-Only

modal analysis
Enhanced Frequency Domain Decomposition

1 Introduction

Electrical submersible pump (ESP) installed in production wells is an efficient and

reliable system of artificial elevation capable of bring up great amounts of oil from its
natural reservatories, located hundreds of meters down to the sea surface. These vol-
umes vary from a minimum level of 150 barrels per day to a maximum of 150.000 bpd

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 631–638, 2021.
https://doi.org/10.1007/978-981-15-4680-8_43
632 U. A. Monteiro et al.

(24 to 24,600 m3/d). However, ESP with varying velocity controllers may yet extend
significantly this interval, which implies that these systems has become the most used
technology in the world and a natural choice for petroleum reservatories [1].
Initially, EPS’s used to have applicability only in terrestrial production fields
(onshore) and, with the pionerism of the Brazilian company, Petrobras, it was initiated
successful prototypes tests in maritime offshore fields in 1994, in the RJS-221 well, on
Campos Basin, being the first one to install the system in deep water, in 1998 [2].
In order to increase ESP’s reliability, is highly important to know the equipment
dynamic behavior while it is operating. To simulate the real operating conditions, the
ESP’s are tested in a test well, where electrical and mechanical tests can be performed
to evaluate the dynamic behavior of the equipment.
The objective of this work is to identify the modal parameters of an ESP with 62
stages and 10.42 m in length, installed 9.02 m deep in a test well and weighing
616.8 kg. Drop tests were carried on and the vibration responses collected by water
proof accelerometers were used to perform a modal analysis using the EFDD technique.

1.1 Enhanced Frequency Domain Decomposition (EFDD)

Output-Only Modal Analysis was developed in order to bypass the existing limitation
of the traditional method and, in this way, to offer the possibility of modal parameters
assessment only from dynamic response information. It can be mentioned that this
analysis allows to estimate these parameters in cases of unknown excitation forces (in
terms of its magnitude and frequency), applied to linear and time-invariant systems.
This modal identification technique is an enhanced version of the Frequency
Domain Decomposition (FDD) method [3]. Its main advantage, related to the FDD
method, is the possibility of a more precise assessment of the natural frequencies and
the identification of damping rate.
The essence of the EFDD method is based on the Singular Value Decomposition
(SVD) of the Power Spectrum Density (PSD) function obtained from the system
response, which graphical representation allows to identify the system natural fre-
quencies and corresponding vibration normal modes.
The estimated PSD matrix, Gy(f), formed by the estimated PSD’s for all the sen-
sors, may be decomposed using the Singular Value Decomposition (SVD), which for a
complex hermitian and positive definite matrix has the following form:

Gy ð f Þ ¼ USUH ð1Þ

where the diagonal matrix S is the singular value matrix and the unity matrix U = [u1,
u2,…], contains the singular vectors which are interpreted as the modal vectors of the
system and the superscript “H” denote the Hermitian transpose operation. It must be
mentioned that the PSD decomposition, shown in Eq. (1), does not fully correspond to
the theoretical decomposition of the PSD matrix and, hence, the EFDD method is an
approximate solution supposing that the modal coordinates are not correlated.
It is necessary to use some criterion in order to recognize the peaks in the singular
value graphical presentation. With the region of analysis selected using Modal
Assurance Criterion (MAC), each vibration mode is correlated to a SDOF system with
 Modal Parameter Identification ESP 633

the corresponding Autocorrelation Function, which is calculated through the Inverse

Fast Fourier Transform (IFFT). Next, a triangular time window is applied in order to
enhance the Autocorrelation Function estimative. Finally, through the logarithmic
decrement it is possible to estimate the corresponding damping rate and with the Zero
Crossing curve it is possible to calculate the undamped natural frequency of the SDOF
system.

2 Experimental Tests

The tested ESP set has a 10.42 m length and 62 stages. Figure 1 shows the schematic of
the test well circuit, using water as working fluid, installed in Mossoró/RN-Brazil, with
the ESP’s dimensions and the local where the impacts were applied due to the drop test.
A more detailed technical description may be found in [2].

Fig. 1. Test well circuit.

The drop tests were carried on in the production pipe, near the place where the ESP
set is supported and, as a matter of safety, it was not operating during the drop test.
About 30 impacts were applied on the structure and the vibration signals were origi-
nally acquired with a 25 kHz sampling rate [2].
634 U. A. Monteiro et al.

The accelerometers were attached to a stainless steel plate positioned on the ESP
case by a metal belt, as can be seen in Fig. 2.

Fig. 2. Accelerometers attachment on the ESP.

Three (3) accelerometers were installed on each of the ESP’s components (pump,
seal/protector, motor), in three different transversal sections, except in the pump on
which it was installed two accelerometers, totalizing 8 measuring points (S1, S2,…,S8).
For the modal analysis, it was only considered the accelerometers with the impact
response representing a well-defined free vibration data with low noise distortion.
The signals were filtered with a 500 Hz low pass filter and were re-sampled with a
1 kHz sampling rate. The ESP operating range is from 1800 to 3720 RPM (30 to
62 Hz), but only the modal parameters in the range of 0 to 100 Hz was analyzed.

3 Experimental Results

Figure 4 shows the acceleration signal from the accelerometer corresponding to S2 in

Fig. 3.
Table 1 shows the results obtained with the EFDD technique. It is verified that the
modal parameters corresponding to the 8th vibration mode was not identified with the
drop tests. Results obtained previously in [5], using impact hammer, and reproduced in
Table 2 shows that there is an 8th vibration mode in 48.55 Hz.
The reason becomes clear when singular values near these frequencies are ana-
lyzed, which show these peaks having low energy and high noise. In order to improve
the analysis, noise minimization techniques and enhanced PSD estimation (using others
estimators) are suggested by some researchers [4].
 Modal Parameter Identification ESP 635

Fig. 3. Accelerometers distribution in the structure.

Fig. 4. Free decay response in the accelerometer S2.

Table 1. Modal parameters identified with drop test

Mode Natural Damping rate [%]
frequency [Hz]
1 3.97 5.69
2 7.16 2.68
3 10.58 3.21
4 15.51 2.15
5 19.75 1.98
6 26.48 4.10
(continued)
636 U. A. Monteiro et al.

Table 1. (continued)
Mode Natural Damping rate [%]
frequency [Hz]
7 40.68 2.59
8 – –
9 58.87 1.63
10 69.28 1.76
11 75.09 1.86

4 Results Discussion

Table 2 show the results obtained using impact hammer to test the ESP´s structure [5].
The impacts were located near the ESP’s supporting rig, at the wellhead. The test was
carried out using a rubber point impact hammer model 086D50.
Table 3 shows the differences between the estimated modal parameters in Table 1
and Table 2.

Table 2. Modal parameters identified with impact tests [5]

Mode Natural Damping rate [%]
frequency [Hz]
1 4.73 4.66
2 7.95 2.79
3 11.46 3.42
4 16.48 2.15
5 20.96 3.17
6 26.88 5.24
7 41.04 1.43
8 48.55 3.29
9 58.01 2.49
10 70.51 3.13
11 79.33 4.86

In the ESP’s operation range, between 1800 and 3720 RPM (30 and 62 Hz), the
differences in the estimated undamped natural frequencies were lower than 6%. The 1st
vibration mode identified with drop test presented the highest difference, showing that
the structure is not well excited in this frequency band.
While the natural frequencies and the vibration normal modes are easily calculated
from the experimental tests, the accurate estimative of the damping factors still be open
to research, as shown in [6].
Finite element models previously developed for the ESP under test showed that
there are vibration modes under 4 Hz that were not experimentally identified. This
leads to the conclusion that neither impact nor drop tests are capable of exciting the
 Modal Parameter Identification ESP 637

Table 3. Absolute differences (%) between modal parameters identified with drop test and
impact test
Mode Natural Damping rate [%]
frequency [%]
1 16.1 22.1
2 9.9 3.9
3 7.7 6.1
4 5.9 0.0
5 5.8 37.5
6 1.6 21.7
7 0.9 81.1
8 – –
9 1.5 34.5
10 1.7 43.8
11 5.3 61.7

lower frequency region up to 4 Hz. This is mainly due to the position where the
excitation is applied, near the wellhead, where the ESP is supported.

5 Conclusions

In this paper, an ESP set of 62 stages installed in a test well had its modal parameters
experimentally identified using the EFDD technique and the structure’s vibration
responses obtained through drop tests while the ESP set was out of operation. Results
were compared with those obtained in [5] using impact hammer and EFDD technique.
Results showed that with the drop test it is possible to obtain an estimate of the
ESP’s modal parameters but, for more accurate results, the impact test should be used.
The damping rate is the most challenging parameter to estimate, as can be seen in
Table 1 and Table 2, with differences reaching 81.1%. This problem is independent of
the estimation method and still is object of actual researches. Hence, the use of the
estimated damping rates must be cautious.
On the other hand, identification of natural frequencies does not face such problem.
To take into account the rotordynamic effect, the Operational Modal Analysis
(OMA) should be considered, where the ESP is tested during normal operation,

Nomenclature
ESP = Electrical submersible pump
EFDD = Enhanced Frequency Domain Decomposition
FDD = Frequency Domain Decomposition
Gy(f) = estimated PSD matrix
H = superscript denoting the Hermitian transpose operation
IFFT = Inverse Fast Fourier Transform
MAC = Modal Assurance Criterion
638 U. A. Monteiro et al.

OMA = Operational Modal Analysis

PSD = Power Spectrum Density
S = singular value matrix
SDOF = Single Degree of Freedom
SVD = Singular Value Decomposition
U = unity matrix, contains the singular vectors

Acknowledgements. This study was possible due to sponsorship of Petrobras through its
Research and Development Center (CENPES), under Grant 5850.0106375.17.9 and the Elec-
trical Submersible Pump Specialized Workshop (UE-BCS), located in Mossoró-RN, Brazil.
Their technical group was preponderant to developing this work.

References
1. Takacs, G.: Electrical Submersible Pumps Manual: Design, Operations and Maintenance.
Gulf professional publishing, Houston (2009)
2. Minette, R.S., Silvaneto, S.F., Vaz, L.A., Monteiro, U.A.: Experimental modal analysis of
electrical submersible pumps. Ocean Eng. 124, 168–179 (2016). https://doi.org/10.1016/j.
oceaneng.2016.07.054
3. Brincker, R., Zhang, L., Andersen, P.: Modal identification from ambient responses using
frequency domain decomposition (2000)
4. Brincker, R.: Introduction to Operational Modal Analysis. Wiley, Hoboken (2015)
5. Huamani, J.A., Monteiro, U.A., Gutiérrez, R.H.R., Troyman, A.C., Vaz, L.A.: Output-only
modal analysis of an electrical submersible pump installed in a mock well using time and
frequency domain techniques. In: Proceedings of XXXVIII Iber. Latin-American Congress
Computer Methods Engineering, p. 13 (2017)
6. Brincker, R., Zhang, L., Andersen, P.: Modal Identification of Output-Only Systems Using
Frequency Domain Decomposition. Smart Mater. Struct. 10, 441–445 (2001)
 Fatigue Analysis of Vortex-Induced Vibration
for Marine Risers with Top-End Platform
Motion Excitations

Yuchao Yuan1,2 , Hongxiang Xue1,2(&), Wenyong Tang1,2,

and Jun Liu1,2
1
State Key Laboratory of Ocean Engineering, Shanghai Jiao Tong University,
Shanghai, China
{godyyc,hongxiangxue,wytang,jliu}@sjtu.edu.cn
2
Collaborative Innovation Center for Advanced Ship and Deep-Sea
Exploration, Shanghai, China

Abstract. Vortex-induced vibration (VIV) of marine risers is a typical fluid-

structure interaction (FSI) phenomenon. When considering the top-end platform
motions, the nonlinearity of risers’ VIV would strengthen significantly, making
it become a worthwhile research frontier in ocean engineering. Platform heave
leads to the riser’s axial tension fluctuating with time varying (i.e. parametric
excitation), while platform surge results in an equivalent unsteady flow field
induced by the riser’s forced oscillatory movements. To investigate the VIV
response of a full-scale production riser with both platform heave and surge
excitations, an alternative time domain numerical analysis procedure is estab-
lished in this paper. The platform motion response is predicted with the help of
three-dimensional potential flow theory, and the riser’s VIV is simulated based
on force-decomposition model. By updating the structural stiffness matrix and
improving the VIV excitation force formula, the platform heave and surge
excitations are taken into account respectively. Four types of cases including
steady flow case, heave case, surge case and combined case are calculated to
reveal their respective response characteristics as well as the coupling effect
between the platform heave and surge excitations. Associated with rainflow
counting algorithm and S-N curve method, the stress components of the riser’s
VIV under different cases are analyzed, and the structural fatigue damage is
evaluated in detail. The obtained conclusions could provide some references to
the engineering field at the design stage of marine risers.

Keywords: Marine risers
Vortex-induced vibration
Platform motion

excitation
Response characteristic investigation
Fatigue damage evaluation

1 Introduction

With the oil and gas exploitation spreading to deeper sea, the safety of longer marine
risers has been attracting increasing attention over the recent years. Vortex-induced
vibration (i.e. VIV) is a fluid-structural interaction phenomenon commonly existing in
ocean engineering, which is also a significant detrimental factor to the marine risers’

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 639–656, 2021.
https://doi.org/10.1007/978-981-15-4680-8_44
640 Y. Yuan et al.

safety. When ocean current flows through, vortex would shed periodically around the
riser, making the riser subjected to nonlinear hydrodynamic forces. As a flexible
structure, the riser vibrates as a result of the oscillatory hydrodynamic forces, then
disturbs the surrounding flow and corresponding hydrodynamic forces. More impor-
tantly, the vortex shedding may synchronize with the riser’s motion, and the response
amplitude would enlarge evidently, causing severe fatigue damage even structural
failure.
Under the actual marine environment, the encountered excitations of the riser are
usually more complicated than most of the existing laboratory experiments and
numerical simulations. Figure 1 is the sketch of platform-riser-seabed system under
complicated marine environment. The riser is connected to the floating platform
through the tensioner at top-end and its bottom is fixed at seabed. Due to wave and
current load, the platform would experience diverse degrees of freedom motion
response. Platform heave leads to the compression and stretch of the tensioner, then
causing the axial tension of the riser fluctuating with time. The time-varying tension
would affect the boundary condition and structural stiffness property of the riser.
Platform surge drives the riser to oscillate in water, and the riser’s local oscillation
amplitude decreases along the axial direction because of the bottom constraint. As a
result, the riser would withstand an equivalent sheared-oscillatory flow field. Both the
time-varying tension and equivalent unsteady flow have significant impacts on the
riser’s dynamic analysis, therefore taking the platform motion excitations into con-
sideration is necessary to achieve a deeper insight of the riser’s VIV response and
fatigue characteristics under complicated marine environment.

Fig. 1. Sketch of platform-riser-seabed system under complicated marine environment

Most of the existing experimental and numerical researches on risers’ VIV mainly
focus on constant tension and steady flow cases. Over the last decades, the effects of
time-varying tension and unsteady flow filed on risers’ dynamic response are receiving
increasing attention, and they have been investigated separately and preliminarily.
 Fatigue Analysis of Vortex-Induced Vibration 641

Franzini et al. [1] carried out the laboratory experiments of a 2.552 m riser’s VIV with
time-varying tension. Karniadakis et al. [2] made some efforts to investigate the VIV
with varying structural bending stiffness with the help of spectral methods for com-
putational fluid dynamics (CFD). Josefsson et al. [3] investigated the transverse VIV of
a variable tension riser by combining CFD and computational structural dynamics
(CSD) methods. There are also several alternative numerical models developed by da
Silveira et al. [4], Tang et al. [5], Yang et al. [6] and Chen et al. [7], in which VIV is
mostly simulated with wake oscillator models. When it comes to unsteady flow cases,
some relevant research work has been published as well. Fu et al. [8] investigated the
VIV response features of a 4 m riser under oscillatory flows with laboratory experi-
ments, and Wang et al. [9] discussed the VIV fatigue damage of a 23.71 m truncated
steel catenary riser (SCR) riser model with its top-end heaving in still water. Chang
et al. [10] predicted a riser’s VIV under the unsteady flow generated by the platform
heave motion. Thorsen et al. [11, 12] developed a semi-empirical prediction approach
and compared their numerical results with the test measurements of a top-tensioned
riser (TTR) and an SCR under oscillatory flows. Besides, the riser’s VIV under pure
oscillatory flow cases and uniform combined with oscillatory flow cases are simulated
with CFD technique [13–15]. Above experimental and numerical researches all indi-
cated that the time-varying tension and unsteady flow field induced by top-end platform
motions both affects the riser’s VIV response evidently.
The authors have developed a time domain numerical model available for riser’s
VIV prediction, and the prediction method has been validated against the published
experimental results under time-varying tension cases and oscillatory flow cases
respectively [16, 17]. As a subsequent research, the multi-degrees of freedom motion
excitations of the top-end platform are taken into consideration simultaneously in this
paper. The platform heave and surge motions are simulated as multi-frequency time-
varying tension excitation and equivalent sheared-oscillatory flow field excitation
respectively, so that a full-scale deepwater riser’s VIV fatigue damage characteristics
under complicated marine environment could be investigated in detail.

2 VIV Numerical Model Methodology

The adopted VIV numerical model in this paper has been published by the same
authors before, so here just describe it briefly for the sake of conciseness, and more
details could be found in Reference [16]. The time domain VIV prediction procedure is
based on Euler-Bernoulli beam theory and force-decomposition model, and the
hydrodynamic force coefficient database originates from the typical cylinder forced
vibration experimental data [18]. If the structural dynamic response is out of the
database range, an empirical hydrodynamic damping model proposed by Venugopal
[19] will be used. The structural damping coefficient cs is expressed as cs ¼ 4pmf n,
where m is the riser mass, f is the response frequency and n is the structural damping
ratio. The lock-in region is set as the non-dimensional frequency range of [0.125, 0.25].
When lock-in occurs, the riser element will be synchronized onto the structural natural
frequency closest to the non-dimensional frequency of 0.17, which corresponds to the
largest excitation force coefficient. A second-order digital control loop, consisting of a
642 Y. Yuan et al.

phase error detector and a second-order tracker, is adopted to guarantee that the
excitation force is truly in phase with the local velocity of the riser. The riser’s gov-
erning differential equation of motion is discretized in space by finite element method
(FEM). The whole numerical model is based on direct-integration dynamic analysis,
and the structural kinetic equation is solved with Hilber-Hughes-Taylor (HHT) method
[20]. The VIV of marine risers is a typical fluid-structure interaction issue, with strong
nonlinearity. Compared with explicit dynamic analysis, the implicit dynamic analysis
usually shows better performance when solving such complicated mechanism issues
due to its better convergence and stability.
This paper takes the multi-degrees of freedom motions of the top-end platform into
consideration. Platform heave would lead to the time-varying axial tension of the riser,
so the structural stiffness matrix is updated according to the real-time tension distri-
bution at the beginning of each time step. Platform surge would force the riser’s top-
end to oscillate in the original steady background current, causing an extra sheared-
oscillatory flow field. Therefore, the VIV excitation force calculation is based on the
time-varying velocity and direction of the complicated combined flow. The flow-chart
of the riser’s VIV time domain analysis with platform motion excitations is shown in
Fig. 2. Note that, the proposed numerical model has been well verified applicable for
the VIV prediction of the marine risers under unsteady flow cases and time-varying
tension cases in References [16, 17] respectively. It is reasonable to investigate the
riser’s VIV response and fatigue characteristics with platform motion excitations based
on the developed numerical analysis procedure.

Fatigue
Damage Platform
Evaluation Start Motion
Prediction

Structural
Response Platform
Calculation Heave
Excitation

VIV
Stiffness Matrix Real-time
Update Axial Tension

Frequency Amplitude

Damping Force Time-varying current

Added Mass Lock-in NO
Coefficient
velocity & direction
Coefficient Database
YES
Excitation Force
Coefficient Platform
Database Surge
Inertia Excitation
Force
Excitation Damping Next Step
Force Force

Fig. 2. Flow-chart of riser’s VIV time domain analysis with platform motion excitations
 Fatigue Analysis of Vortex-Induced Vibration 643

3 Top-End Platform Motion Excitations

To investigate the effect of the platform motion excitations on the riser’s VIV, this
section firstly predicts the time domain motion response of a tension leg platform
(TLP) under the complicated marine environment, and then integrates the equivalent
excitations of the top-end platform heave and surge into the riser’s VIV analysis.
Figure 3 shows the FE model of top-end platform for hydrodynamic calculation.

Fig. 3. FE model of TLP for hydrodynamic calculation

3.1 Time Domain Motion Response Prediction

The motion response prediction of the top-end platform is based on three-dimensional
potential flow theory and indirect time domain method. A TLP serving in 1500 m
water depth is chosen, and its relevant parameters are listed in Table 1 and Table 2.
There are eight tension legs connected with the floating platform to provide necessary
constraint stiffness. The original length and outer diameter of the tension legs are
1474.04 m and 0.9235 m respectively, and the axial stiffness is 1968 t/m. The fixed
point coordinate of tension legs at TLP is X = ±39.809 m & Y = ±33.477 m or
X = ±33.477 m & Y = ±39.809 m, and the vertical position is 5.44 m. The main
parameters of the TTR are shown in Table 3.

Table 1. Main parameters of TLP

Parameter Value
Platform height from baseline (m) 54.0
Column height (m) 48.0
Draft (m) 28.0
Center distance between columns (m) 58.0
Column diameter (m) 20.0
Pontoon height (m) 9.5
Pontoon width (m) 10.72
644 Y. Yuan et al.

Table 2. Parameters of TLP under work case

Parameter Value
Water depth (m) 1500
Displacement (t) 51375.3
Total pretension of tension legs (t) 9160
Total pretension of top-tension risers (t) 4540.5
Platform weight (t) 37674.8
Gravity center height without TLs and TTR (m) 45.25
Inertia radius of roll without TLs and TTR (m) 33.20
Inertia radius of pitch without TLs and TTR (m) 33.72

Table 3. Main parameters of TTR

Parameter name Value
Length (m) 1500
Outer diameter (m) 0.346
Inner diameter (m) 0.286
Young’s modulus (GPa) 210
Poisson ratio 0.3
Wet mass per unit length (kg/m) 187.10
Axial stiffness (MN/m) 6196
Top-end pretension (t) 449.05

This paper selects the one-year-return-period sea state of China South Sea as the
simulated marine environment. The parameters of the corresponding irregular wave are
listed in Table 4. JONSWAP spectrum is chosen to describe the irregular wave, and the
spectral shape parameter c is taken as 3.3 for extremely deep water. Liu [21] gave the
current profile distribution of South China Sea, as shown in Fig. 4.

Table 4. Parameters of irregular wave under one-year-return-period sea state

Parameter name Value
Significant wave height (m) 8.7
Average zero-crossing period (s) 9.92
Spectral peak period (s) 12.30
Spectral peak frequency (rad/s) 0.51
Spectral peak magnitude (m2s/rad) 20.77
 Fatigue Analysis of Vortex-Induced Vibration 645

1500

1200

Water depth (m)

900

600

300

0
0.0 0.5 1.0 1.5 2.0
Current velocity (m/s)

Fig. 4. Current profile distribution of South China Sea [21]

The hydrodynamic behaviors (including added mass coefficient, damping coeffi-

cient and load Response Amplitude Operator i.e. RAO) of the TLP are obtained from
the frequency domain prediction. Then they are used to calculate the platform’s time
domain motion response under combined current and irregular wave. The time histories
of platform heave and surge are presented in Fig. 5 and Fig. 6, where only the dynamic
components are included. The calculation time interval is set as 10800 s, which is long
enough to minimize the effect of the randomness induced by the irregular wave
spectrum.

2
Heave (m)

-1

-2

-3

-4
0 2000 4000 6000 8000 10000 12000
Time (s)

Fig. 5. Time history of TLP heave displacement under one-year-return-period sea state
646 Y. Yuan et al.

15

10

5
Surge (m)

-5

-10

-15
0 2000 4000 6000 8000 10000 12000
Time (s)

Fig. 6. Time history of TLP surge displacement under one-year-return-period sea state

3.2 Equivalent Excitation Simulation

The riser’s top-end is connected to the platform through the tensioner. The tension of
the tensioner T is a function of the piston stroke s, approximately following the non-
linear formula based on gas theory [22], as shown in Eq. (1).

T0
T¼ ð1Þ
ð1 þ s=s0 Þc

where T0 is the top-end pretension of the riser, s0 is the motion length interval of the
piston, c is the gas constant which could be set as 1.0–1.3 according to the gas property
of different types. s0 in this paper is taken as 11.2 m (down stroke sdown = -5.6 m, and
up stroke sup = 5.6 m), and c is set as 1.1. The relation curve between tension and stroke
is shown in Fig. 7. Associated with the platform heave response in Fig. 5, the time
history of the riser’s tension at top-end could be easily obtained, as shown in Fig. 8.

10

8 Significant Variation
T (MN)

6 Up stroke sup

Down stroke sdown

Insignificant
T0
4 variation

2
-6 -4 -2 0 2 4 6
s (m)

Fig. 7. Tension-stroke curve of tensioner

 Fatigue Analysis of Vortex-Induced Vibration 647

6.5

6.0
Top-end tension (MN)

5.5

5.0

4.5

4.0

3.5

3.0
0 2000 4000 6000 8000 10000 12000
Time (s)

Fig. 8. Time history of deepwater riser top-end tension under one-year-return-period sea state

The combination of a sheared-oscillatory flow field is adopted to simulate the

equivalent effect of the platform surge excitation. The derivation of the displacement
time history in Fig. 6 with respect to time is obtained as shown in Fig. 9, which denotes
the platform surge velocity. Based on the background steady flow, an extra sheared
time-varying flow field, whose top-end velocity is equal to the platform instantaneous
surge velocity, is included when predicting the riser’s dynamic response. The rigid
body swing movement of the riser induced by platform surge is taken into consider-
ation so that the encountered velocity is always perpendicular to the riser’s axial
direction. The combined flow field presents obvious non-uniformity in both time and
spatial domains.

1.5

1.0
Surge velocity (m/s)

0.5

0.0

-0.5

-1.0

-1.5
0 2000 4000 6000 8000 10000 12000
Time (s)

Fig. 9. Time history of riser’s top-end equivalent velocity under one-year-return-period sea state
648 Y. Yuan et al.

4 Dynamic Response Investigation

Four different types of load cases are simulated in this paper. Steady flow case only
includes the steady background ocean current excitation, heave case includes the
excitations of steady flow and platform heave motion, and surge case includes the
excitations of steady flow and platform heave motion, while combined case includes
the excitations of steady flow, platform heave as well as surge motions.

4.1 Local Response Characteristic

VIV displacement time histories at the riser’s midpoint under four cases are shown in
Fig. 10. VIV local response under steady flow case is relatively stable with visible
periodicity, presenting the feature of the linear superposition of multi-frequency
components. Above response features are similar with those under pure sheared flow
cases. For the rest three cases, the nonlinearity of the displacement response is sig-
nificant. Especially when t is around 700 s under heave case and when t is around
450 s under surge as well as combined cases, the riser presents different vibration
processes from the adjacent time intervals. According to Fig. 9, there is obviously a
sudden increase of the platform’s surge velocity near t = 450 s, which could explain
the peak displacements at t = 450 s for the Surge case and Combined case. Compared
among the first three cases, the time-sharing and strongly nonlinear response charac-
teristics induced by platform surge excitation are more evident than those induced by
platform heave excitation. Figure 11 gives the VIV displacement amplitude spectra at
the riser’s midpoint under four cases. The first row follows the typical features under
spatial non-uniform steady flow, i.e. multiple frequency peaks are excited discretely.
Under heave case, some extra frequency components around the peak frequencies
excited under steady flow case are also excited evidently. Associated with Reference
[17], these weak but clearly existent frequencies are believed owing to the sub-
harmonics induced by the axial multi-frequency parametric excitation. Under surge
case, there is an apparent difference from steady flow case, i.e. the low-frequency
components (less than 0.5 Hz) are excited additionally and the response energy con-
centrating at the dominant frequency of 0.7 Hz becomes much weaker. This is because
the velocity variation range of the incoming current under surge case turns so broad
that the riser would locate in the low-frequency response excitation region during some
enough time intervals. While under combined case, the displacement amplitude
spectrum covers the respective features of the first three cases. The sub-harmonic
frequency components induced by platform heave excitation and the low-frequency
response induced by platform surge excitation appear at the same time, and the
dominant frequency of 0.7 Hz turns furtherly weaker.
 Fatigue Analysis of Vortex-Induced Vibration 649

Steady flow case

Displacement (m)
0.2
0.1
0.0
-0.1
-0.2
200 700 1200 1700 2200
Time (s) Heave case
Displacement (m)

0.2
0.1
0.0
-0.1
-0.2
200 700 1200 1700 2200
Time (s) Surge case
Displacement (m)

0.2
0.1
0.0
-0.1
-0.2
200 700 1200 1700 2200
Time (s) Combined case
Displacement (m)

0.2
0.1
0.0
-0.1
-0.2
200 700 1200 1700 2200
Time (s)

Fig. 10. VIV displacement time histories at the riser’s midpoint under four types of cases

0.04 Steady flow case

Amplitude (m)

0.03
0.02
0.01
0.00
0.0 0.5 1.0 1.5
Frequency (Hz)
0.04 Heave case
Amplitude (m)

0.03
0.02
0.01
0.00
0.0 0.5 1.0 1.5
Frequency (Hz)
0.04 Surge case
Amplitude (m)

0.03
0.02
0.01
0.00
0.0 0.5 1.0 1.5
Frequency (Hz)
0.04 Combined case
Amplitude (m)

0.03
0.02
0.01
0.00
0.0 0.5 1.0 1.5
Frequency (Hz)

Fig. 11. VIV displacement amplitude spectra at the riser’s midpoint under four types of cases
650 Y. Yuan et al.

4.2 Overall Response Characteristic

The envelops of VIV maximum displacement along the riser under four cases are
compared in Fig. 12. The envelops of the first two cases are approximate, and heave
case is a little larger than steady flow case within z/L between 0.45 and 0.82. When
platform surge excitation is included, VIV maximum displacement enlarges obviously.
In particular, the response amplitude could reach twice the corresponding values of
steady flow case and heave case within the riser’s middle segment. The maximum
displacement envelops along the riser present lower order modal vibration shape than
steady flow case and heave case as well. Under combined case, the maximum dis-
placement is larger than pure heave or surge cases. It means the riser’s VIV response
with multi-degrees of freedom platform motion excitations is more severe than that
without or only with single-degree of freedom platform motion excitation. From the
point of view in engineering application, the coupling aggravation effect between the
platform’s different degrees of freedom motion excitations on the structural dynamic
response should be taken into consideration for the riser’s stiffness design.

Steady flow case

1.0 Heave case
Surge case
Distance from bottom/Riser length

Combined case
0.8

0.6

0.4

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0
Maximum displacement/Riser diameter

Fig. 12. Envelops of VIV maximum displacement along the riser under four types of cases

5 Fatigue Damage Evaluation

In this section, based on the predicted stress time history at the calculation location, the
loop number of the alternating stress with different amplitudes is counted by rainflow
counting algorithm. Then, the fatigue analysis of the deepwater riser with platform
motion excitations is carried out with S-N curve method and Miner-Palmgren cumu-
lative fatigue damage theory.
 Fatigue Analysis of Vortex-Induced Vibration 651

5.1 Stress Component Analysis

Figure 13 gives the envelops of the maximum bending stress along the riser under four
cases. Steady flow case and heave case have approximate stress level, and the maxi-
mum stress under surge case turns slightly larger, while the corresponding value reach
peak for combined case. Above magnitude relation is quite similar with the maximum
displacement in Fig. 12, which indicates again that the combination with platform
heave and surge excitations trends to cause more severe riser stress level than steady
flow case and single motion excitation cases. According to the envelop shape, steady
flow case and heave case basically present the 18th mode dominating the riser’s
dynamic response, while once platform surge excitation is included under both surge
and combined cases, the dominant modal order becomes not clear enough and irregular
maximum stress distribution appears, which is because more frequency components are
excited with current velocity variation range broadening.

1.0
Distance from bottom/Riser length

0.8

0.6

0.4
Steady flow case
Heave case
0.2 Surge case
Combined case

0.0
0 3 6 9 12
Maximum bending stress (MPa)

Fig. 13. Envelops of maximum stress along the riser under four types of cases

Apart from the extreme stress level, the loop number of alternating stress would
also importantly affect the structural fatigue damage. From Fig. 13, the response stress
mainly fluctuates within [0 MPa, 10 MPa]. Here sets 0.2 MPa as a constant gradient to
in turn count the loop number of the alternating stress with different amplitudes among
[0 MPa, 0.2 MPa], [0.2 MPa, 0.4 MPa], [0.4 MPa, 0.6 MPa] and so on by rainflow
counting algorithm. Figure 14, Fig. 15 and Fig. 16 represent the statistic results at three
different axial locations of z/L = 0.1, 0.3 and 0.5 respectively.
For above three different calculation locations, the statistic results present identical
regularity, namely, the loop number of alternating stress decreases with the stress
amplitude increasing. It means the alternating stress induced by VIV is mainly dom-
inated by high-frequency and small-amplitude components. Four cases could be
classified into two groups. The stress loop numbers under steady flow case and heave
case (i.e. Group i) are basically approximate, which is also true for surge case and
combined case (i.e. Group ii). Compared with Group i, Group ii has broader variation
range of stress amplitude. The loop numbers of small-amplitude alternating stress are
larger and the ones of large-amplitude stress are smaller. For Group ii, although the
652 Y. Yuan et al.

Steady flow case

2048 Heave case
1024 Surge case
Combined case
512
256
Loop number

128
64
32
16
8
4
2
1
0 1 2 3 4 5 6 7 8 9 10
Cyclic stress amplitude (MPa)

Fig. 14. Bending stress loop number of different loop stress amplitudes at z = 150 m

Steady flow case

2048 Heave case
1024 Surge case
Combined case
512
256
Loop number

128
64
32
16
8
4
2
1
0 1 2 3 4 5 6 7 8 9 10
Cyclic stress amplitude (MPa)

Fig. 15. Bending stress loop number of different loop stress amplitudes at z = 450 m

Steady flow case

Heave case
2048
Surge case
1024
Combined case
512
256
Loop number

128
64
32
16
8
4
2
1
0 1 2 3 4 5 6 7 8 9 10
Cyclic stress amplitude (MPa)

Fig. 16. Bending stress loop number of different loop stress amplitudes at z = 750 m
 Fatigue Analysis of Vortex-Induced Vibration 653

large-amplitude alternating stress that causes larger fatigue damage single time has less
loop number than Group i, the small-amplitude stress have much more loops which
might also lead to larger structural damage. Therefore, which group has more severe
total damage is still unknown before the cumulative fatigue analysis with S-N curve
and Miner-Palmgren theory is carried out (Figure 15).

5.2 Fatigue Characteristic Comparison

The envelops of VIV cumulative fatigue damage along the riser under four cases are
shown in Fig. 17, where the relevant material parameters of the adopted S-N curve are
taken as m = 3.0 and A = 1011.687. Steady flow case has the maximum fatigue damage,
and heave case ranks the second, while the corresponding value of combined case is
larger than surge case but smaller than heave case. Associated with the statistic results
of the alternating stress loop number in Fig. 14, Fig. 15 and Fig. 16, it could be
deduced that VIV fatigue damage is dominated by the large-amplitude stress compo-
nents, namely, the case under which the large-amplitude alternating stress has more
loops trends to accumulate more severe total structural damage. Compared to the three
cases with platform motion excitations, the nonlinearity of VIV hydrodynamic forces
under steady flow case is relatively weak, so the structural response is more stable,
which is beneficial to accumulate fatigue damage. Note that, when the platform motion
excitations come from single-degree of freedom to multi-degrees of freedom, the riser’s
fatigue damage may increase or decrease. In fact, the nonlinearity of load excitations
under combined case is obviously stronger than heave case and surge case, causing the
structural dynamic response being more unstable, so it acts as a detrimental role to
accumulate severe fatigue damage. However, according to the maximum stress com-
parison in Fig. 13, the extreme stress response under combined case is larger than both
heave case and surge case due to the coupling aggravation effect between the platform
different degrees of freedom motion excitations, making the fatigue damage easier to
enlarge. Who would dominate the final cumulative damage for the above two con-
tradictory influence factors is still not clear enough under the condition that only single
sea state is simulated in this paper.

1.0
Distance from bottom/Riser length

0.8

0.6

0.4
Steady flow case
Heave case
0.2 Surge case
Combined case

0.0
1E-4 1E-3 0.01 0.1
Fatigue damage (1/year)

Fig. 17. Envelops of VIV fatigue damage along the riser under four types of cases
654 Y. Yuan et al.

Compared with the results between four types of load cases, the effect of platform
surge excitation on the riser’s VIV fatigue analysis is much more significant than that
of platform heave excitation. For a TLP, the heave stiffness provided by tension legs is
usually large, which means platform heave amplitude is relatively smaller than its surge
or sway response under the actual marine environment. So the parametric excitation
induced by platform heave would not be very intensive. Moreover, platform surge
excitation directly affects the riser’s encountered current velocity as well as hydrody-
namic forces, and VIV response is more sensitive to the obvious variation of flow field
than the slight fluctuation of structural stiffness property. Therefore, it is reasonable that
platform surge excitation has a stronger impact on the deepwater riser’s fatigue
characteristics.

6 Conclusion

An alternative time domain numerical model available for deepwater riser’s VIV
prediction in consideration of the top-end platform multi-degrees of freedom motion
excitations is proposed in this paper. Firstly, the wave-induced motions of a TLP are
predicted with three-dimensional potential flow theory. Secondly, platform heave
excitation is simulated as the riser’s time-varying tension, and platform surge excitation
is equivalent to the unsteady combined flow field for the riser. Then, a full-scale marine
riser’s dynamic response is investigated under steady flow case, heave case, surge case
and combined case. Finally, the riser’s fatigue characteristics under four types of load
cases are analyzed in detail. Some conclusions are drawn as follows.
(1) Under heave case, surge case and combined case, the nonlinearity of structural
response displacement is significant, the response pattern changes with the variation of
time and the amplitude modulates irregularly during the whole time interval. Compared
with platform heave excitation, the nonlinear and time-sharing response characteristics
of VIV with platform surge excitation are more apparent. Under combined case, both
the sub-harmonic response induced by platform heave excitation and the low-frequency
response induced by platform surge excitation occur simultaneously. The riser’s VIV
with platform multiple degrees of freedom motion excitations would be more intensive
than that without or only with single degree of freedom motion excitation. Taking the
coupling aggravation effect between the platform’s different degrees of freedom motion
excitations into account is necessary in the engineering design of marine risers.
(2) For the diverse components of VIV stress, the loop number of alternating stress
decreases exponentially with the increase of stress amplitude. Compared with steady
flow case and heave case, platform surge excitation would broaden the variation range
of stress amplitude, increase the loop number of small-amplitude alternating stress and
decrease the loop number of large-amplitude alternating stress under surge case and
combined case.
(3) VIV fatigue damage is mainly dominated by the large-amplitude & low-
frequency alternating stress among four types of cases. The fatigue damage under
steady flow case is largest among all the four cases, followed by heave case, while the
fatigue damage under combined case is larger than that under surge case but smaller
than heave case. The effect of TLP heave excitation on riser’s VIV fatigue damage is
 Fatigue Analysis of Vortex-Induced Vibration 655

relatively weaker than that of platform surge excitation. The magnitude relation
between riser’s VIV fatigue damage with single and multiple degrees of freedom
motion of top-end platform is still unclear, and the direct calculation is necessary to
effectively evaluate the structural cumulative fatigue damage.
Overall, this study would make efforts to promote the VIV response analysis and
fatigue damage evaluation of the deepwater risers under complicated marine envi-
ronment. The obtained conclusions could provide some references to the engineering
design field. Under the actual marine environment, the encountered wave condition is
more complicated than the simulated one by this paper, the long term probability
distributions of wave direction as well as spectral shape parameter should also be taken
into account. More detailed and further investigations need to be carried out in the
future to precisely evaluate the deepwater riser’s fatigue life with platform multi-
degrees of freedom motion excitations.

Acknowledgments. This paper is based on the projects supported by the National Natural
Science Foundation of China (Grant No. 51579146, 51490674) and Shanghai Rising-Star Pro-
gram (Grant No. 16QA1402300).

References
1. Franzini, G.R., Pesce, C.P., Gonçalves, R.T., Mendes, P., Fujarra, A.L.C.: Experimental
investigations on Vortex-induced vibrations with a long flexible cylinder. Part II: effect of
axial motion excitation in a vertical configuration. In: Proceedings of the 11th International
Conference on Flow-Induced Vibration, The Hague, The Netherlands (2016)
2. Karniadakis, G.E., Sherwin, S.J.: Spectral/hp Element Methods for Computational Fluid
Dynamics. Oxford University Press, New York, USA (2005)
3. Josefsson, P.M., Dalton, C.: Vortex-induced vibration of a variable tension riser. In:
Proceedings of the 26th International Conference on Offshore Mechanics and Arctic
Engineering, San Diego, California, USA, paper No. OMAE2007–29200 (2007)
4. da Silveira, L.M.Y., Martins, C.d.A., Leandro, D.C., Pesce, C.P.: An investigation on the
effect of tension variation on VIV of risers. In: Proceedings of the 26th International
Conference on Offshore Mechanics and Arctic Engineering. San Diego, California, USA.
OMAE2007–29247 (2007)
5. Tang, Y.G., Shao, W.D., Zhang, J., Wang, L.Y., Gui, L.: Dynamic response analysis for
coupled parametric vibration and vortex-induced vibration of top-tensioned riser in deep-sea.
Eng. Mech. 30(5), 282–286 (2013)
6. Yang, H.Z., Xiao, F.: Instability analyses of a top-tensioned riser under combined vortex and
multi-frequency parametric excitations. Ocean Eng. 81(2), 12–28 (2014)
7. Chen, W.M., Li, M., Guo, S.X., Gan, K.: Dynamic analysis of coupling between floating
top-end heave and riser’s vortex-induced vibration by using finite element simulations. Appl.
Ocean Res. 48, 1–9 (2014)
8. Fu, S.X., Wang, J.G., Baarholm, R., Wu, J., Larsen, C.M.: Features of vortex-induced
vibration in oscillatory flow. J. Offshore Mech. Arct. Eng. 136(1), 1–10 (2013)
9. Wang, J.G., Fu, S.X., Baarholm, R., Wu, J., Larsen, C.M.: Fatigue damage of a steel
catenary riser from vortex-induced vibration caused by vessel motions. Marine Struct.
39(39), 131–156 (2014)
656 Y. Yuan et al.

10. Chang, S.H., Isherwood, M.: Vortex-induced vibrations of steel catenary risers and steel
offloading lines due to platform heave motions. In: Proceedings of the Offshore Technology
Conference. Houston, Texas, USA (2003)
11. Thorsen, M.J., Sævik, S., Larsen, C.M.: Time domain simulation of vortex-induced
vibrations in stationary and oscillating flows. J. Fluid Struct. 61, 1–19 (2016)
12. Thorsen, M.J., Sævik, S., Larsen, C.M.: Non-linear time domain analysis of cross-flow
vortex-induced vibrations. Marine Struct. 51, 134–151 (2017)
13. Zhao, M., Cheng, L., An, H.W.: Numerical investigation of vortex-induced vibration of a
circular cylinder in transverse direction in oscillatory flow. Ocean Eng. 41(5), 39–52 (2012)
14. Zhao, M., Kaja, K., Xiang, Y., Yan, G.R.: Vortex-induced vibration (VIV) of a circular
cylinder in combined steady and oscillatory flow. Ocean Eng. 73, 83–95 (2013)
15. Deng, Y., Huang, W.P., Zhao, J.L.: Combined action of uniform flow and oscillating flow
around marine cylinder at low Keulegan-Carpenter number. J. Ocean Univ. China 13(3),
390–396 (2014)
16. Yuan, Y.C., Xue, H.X., Tang, W.Y.: Numerical analysis of Vortex-Induced Vibration for
flexible risers under steady and oscillatory flows. Ocean Eng. 148, 548–562 (2018)
17. Yuan, Y.C., Xue, H.X., Tang, W.Y.: A numerical investigation of Vortex-Induced Vibration
response characteristics for long flexible cylinders with time-varying axial tension. J. Fluid
Struct. 77, 36–57 (2018)
18. Gopalkrishnan, R.: Vortex induced forces on oscillating bluff cylinders. Doctoral thesis,
Massachusetts Institute of Technology, Cambridge, USA (1993)
19. Venugopal, M.: Damping and response of a flexible cylinder in a current. Doctoral thesis,
Massachusetts Institute of Technology, Cambridge, USA (1996)
20. Hilber, H.M., Hughes, T.J.R., Taylor, R.L.: Improved numerical dissipation for time
integration algorithms in structural dynamics. Earthq. Eng. Struct. Dyn. 5(3), 283–292
(1977)
21. Liu, H.: Concept design and analysis of deepwater drilling riser. Doctoral thesis, Harbin
Engineering University, Harbin, China (2009)
22. Wang, K.P., Xue, H.X., Tang, W.Y.: Numerical simulation of hydropneumatic tensioner for
coupled deepwater platform system. J. Shanghai Jiao Tong Univ. 46(10), 1652–1657 (2012)
 Real Time Estimation of Local Wave
Characteristics from Ship Motions Using
Artificial Neural Networks

Bulent Duz(B) , Bart Mak, Remco Hageman, and Nicola Grasso

Maritime Research Institute (MARIN), Wageningen, The Netherlands

{b.duz,b.mak,r.hageman,n.grasso}@marin.nl

Abstract. Wave characteristics such as significant wave height, peak

period and main direction have an important effect on all aspects of ship
operations. Yet, estimation of these characteristics in real time in an
accurate, robust and efficient manner poses a significant challenge. Cur-
rent approaches include using wave buoys, wave radars, weather forecasts
and regression based on ship motion statistics. External sources are not
always available or reliable, and wave radars are expensive. Statistics
based methods work reasonably well for wave height, but fail to estimate
peak period and relative wave direction.
In this study the attention is focused on estimating the wave charac-
teristics from ship motions using artificial neural networks. The inputs
to the neural networks were the time histories of 6-DOF ship motions,
and the outputs were the wave characteristics. Using time series instead
of statistics preserves phase differences between signals. Especially when
estimating the relative wave direction, these phase differences encode
essential features. Therefore, the problem was treated as multivariate
time series regression.
Two different data sources were used with neural networks: in-service
measurement data and numerical simulation data. In-service measure-
ment data was collected from a frigate type vessel for a period of two
years. The numerical simulation data was produced by using a time-
domain seakeeping and manoeuvring software.
In this paper, we present methods to successfully implement neural
networks for real time estimation of local wave characteristics. The paper
will include details about the measurement data, numerical simulations,
the networks architectures and their performances on the measurement
and simulation data.

Keywords: Wave characteristics · 6-DOF ship motions · Multivariate

time series regression · Machine learning

1 Introduction
Sea state information is important for operational decisions or advice on all
activities at sea. Personal and environmental safety require correct interpreta-
tions of the actual conditions. Efficiency suffers from incorrect interpretations
c Springer Nature Singapore Pte Ltd. 2021
T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 657–678, 2021.
https://doi.org/10.1007/978-981-15-4680-8_45
658 B. Duz et al.

as well. In the longer run, having statistics on encountered conditions helps in

designing structures and ships. This results in higher levels of safety and more
realistic lifetime expectations. Hence, many methods have been proposed in the
past few decades to infer sea state information from indirect measurements.
Traditionally, sea state characteristics are measured using wave buoys or
satellite data. These methods result in a rough estimation, limited by the dis-
tance from the buoys and the resolution of the satellite data. Occasionally the
sea state is monitored using a wave radar. Although this gives reliable results,
it is not used very often because of the high costs.
An alternative is to use the motions of the vessel itself to infer sea state
information. Methods have been developed by Tannuri [1] and Iseki [2] match-
ing the measured motion spectrum with a predicted spectrum. The matching
spectrum is found either by Bayesian optimization or an iterative method. Both
approaches use estimated response functions as mapping from sea state to ship
motions. This idea has been further optimized by Nielsen [3] to include sailing
vessels and more efficient sea state estimation [4].
In this study 6-DOF ship motion data is used to build a parameterized model
for direct sea state estimation. No knowledge of the ship is used to build the
model. Instead, a parameterized machine learning model is trained using time
series data of the ship motions together with the known sea state in a supervised
learning setup. Being able to capture time series data in this way allows the use
of local phase differences in the motion signals for the sea state estimation. In
this study it is shown that specific neural network structures can be used to do
so and that the resulting models perform well in a wide variety of sea states.
Such a data driven method can be used when sufficient data of good quality is
available for training. For numerical simulation data, this might not be an issue,
but for measurement data it can be a prohibitive factor. This issue limits the
applicability of a data driven method in two ways. First of all, such data is scarce.
Not only are the ship motions needed, but also measured sea states are required.
Secondly, since the model is trained for the ship the measured data came from,
its performance on another ship might deteriorate. However, generalization over
multiple ship geometries was investigated in [5] using simulated data, and very
good results were obtained.
The ideal solution would only use simulated data, removing the need for
any measured data, other than for validating the method. This aspect is also
investigated in this study. Although the proposed method does not work well
enough for this, further investigation shows that the features that are used from
the simulated data are compatible with the features present in the measured
data. From this, some promising future work is proposed.

2 Machine Learning

In this study, neural networks were used to infer sea state information from the
6-DOF ship motions. Both simulated and measured data were used to train the
neural networks. The data was recorded in the time domain, which captures
 Estimation of Wave Characteristics via Machine Learning 659

local phase differences, as opposed to frequency domain data. This way, datasets
were built with ship motion time series and the corresponding sea state. Details
about the datasets can be found in Sect. 3.
Some neural networks, especially Convolutional Neural Networks (CNN) and
Recurrent Neural Networks (RNN), are useful for time series regression. Specific
neural network structures were designed for the problem at hand. Detailed dis-
cussion on motivation and designs can be found in [6]. The next few sections
briefly describe two networks used in this study.

2.1 Multivariate LSTM-CNN (MLSTM-CNN)

MLSTM-CNN network was mainly adopted from [7]. It consists of a fully con-
volutional block and a Long Short-Term Memory (LSTM) block as shown in
Fig. 1. The depth of the LSTM layer is 16 with a dropout rate of 0.25. The first
convolution layer has 128 filters of size 6 × 12. The second and third convolution
layers have 192 and 128 filters of size 1 × 1, respectively. The dense layer has 128
nodes and the dropout rate is 0.25.

Fig. 1. The architecture of the MLSTM-CNN network.

2.2 Sliding Puzzle Network

The Sliding Puzzle Network is designed to respond to individual features, with-

out temporal relations between them. This is achieved by selecting patches at
random locations from an input sample that have the same length as the filters
in the network. Position independence is achieved by reducing filter activations
to statistics, in this case mean, minimum and maximum. The diagram of this
network is shown in Fig. 2.
The network uses 64 filters in the temporal direction, with size 1×25, followed
by 128 filters that combine the 6 channels, with size 6 × 1. The dense layer has
30 nodes and no dropout is used.
660 B. Duz et al.

Fig. 2. The Sliding Puzzle network. Note that two convolutional layers are used instead
of one. The first convolutional layer only works in the time direction, the second com-
bines the 6 channels. The receptive field of this combination is the same as the receptive
field of a single convolution with combined dimensions. The number of trainable weights
is less though, which leads to better generalization.

3 Data Sources and Treatment

Two types of data were used, namely simulated data and measured data. The
details of the two datasets are explained in the next three sections.

3.1 Setup of the Numerical Simulations

Numerical simulations were carried out in 6-DOF with a time-domain seakeep-

ing and manoeuvring tool called FREDYN v16.1.1 [8,9]. A strip theory based
seakeeping tool in the frequency domain named SHIPMO v17.2.2 was used as
a preprocessor to calculate added mass and damping coefficients and diffrac-
tion forces. The time step size was 0.25 s, and the fourth-order Runge-Kutta
time integration scheme was used. The ship was controlled in heading mode;
the autopilot maintained the heading of the vessel rather than its track. The
simulations were carried out at a ship speed of 15 kn for a duration of 600 s. A
frigate-type naval vessel was used as the geometry. The vessel was the same as
the one on board of which the in-service measurement data was collected.
Ten thousand simulations were carried out for a wide range of significant
wave height (Hs), peak period (Tp) and main direction (Dp). Hs, Tp and Dp
were uniformly distributed over the ranges of [0.5 m, 10.5 m], [6.5 s, 15.5 s] and
[0◦ , 360◦ ), respectively. Latin hypercube sampling (LHS) was then used to sample
10000 points from this 3D input domain. Figure 3 shows the result from LHS.
The waves were realized using Jonswap spectrum with γ = 3.3, and the spectrum
was discretized using 80 Fourier components.
 Estimation of Wave Characteristics via Machine Learning 661

Fig. 3. Latin hypercube sampling (LHS) with 10000 points for the uniformly dis-
tributed Hs, Tp and Dp.

3.2 Treatment of the Numerical Simulation Results

The results of the numerical simulations were first investigated. Any simulation
where capsizing occurred was ignored. The surge, sway and yaw motions were
filtered to be divided in a wave-frequent (WF) and a low-frequent (LF) part.
The input to the neural networks was a multivariate time series X ∈ 6×d×1 ,
where 6 refers to the 6-DOF motions and d is the length of the motion signals.
The value of d was obtained as a result of a study where the effect of the duration
and the sampling rate of the motion signals on the performance of the neural
networks was examined. In the end, d was chosen to be 200, and the resulting
signal duration was 2.5 min. One sample was extracted from each simulation
resulting in a total sample size of 10000.

3.3 Treatment of the In-service Measurement Data

The in-service measurement data was collected for a period of two years on board
a frigate-type naval vessel. The 6-DOF motions at the center of gravity of the ship
were measured using accelerometers, and the wave characteristics were measured
via a wave scanning radar mounted on the ship. The data was saved as 30 min
long pieces. The sampling rate of the ship motions was 20 Hz, and the sampling
rate of the ship speed was 1 Hz. The wave scanning radar provided two-minute
data files which were used to derive statistical parameters, such as Hs, Tp and
Dp. The accuracy of peak period and directionality of the wave was generally
good. However, because the radar was mounted on a moving platform the wave
height was not very accurate. In order to compensate this inaccuracy, a wave
data fusion approach was executed to improve the wave height assessment [10].
The in-service measurement data required preprocessing and was unbalanced;
Fig. 4 shows the histograms of Hs, Tp and Dp, where the unbalanced nature
of the data can be clearly observed. More information about how the measured
data was prepared for training is provided in [6]. Furthermore, the ground truth,
662 B. Duz et al.

3500 3000 2000

3000

Number of samples
Number of samples
2500
Number of samples

1500
2500
2000
2000
1500 1000
1500
1000
1000 500
500 500

0 0 0
1 1.5 2 2.5 3 3.5 4 4.5 5 6 7 8 9 10 11 12 0 60 120 180 240 300 360
Hs [m] Tp [s] Dp [deg]

(a) Hs (b) Tp (c) Dp

Fig. 4. Histogram of the wave characteristics from the in-service measurement data.

i.e. wave characteristics, contained contributions from wind and swell seas and
possibly other sources such as currents.
Similar to the numerical simulation results discussed in Sect. 3.2, the input
to the neural networks was a multivariate time series X ∈ 6×d×1 where d
was chosen to be 200, and the resulting signal duration was 2.5 min. In total,
20120 samples were extracted from the in-service measurement data. Figures 5
and 6 illustrate the mean and standard deviation of each sample for 6-DOF ship
motions, while Fig. 7 illustrates the Hs, Tp, Dp and ship speed (Vs) for each
sample. Note that the samples were chronologically ordered in the sense that
the sample with the index value of n was collected later than the sample with
index n − 1, and earlier than the sample with index n + 1.

4 Results

4.1 Training Methodology

We have trained our networks with stochastic gradient descent (SGD) [11] utiliz-
ing the Keras library [12] with the TensorFlow-GPU backend [13] and PlaidML
[14] running on a NVidia GeForce GTX 1080 with 2560 CUDA cores and 8 GB
memory. The processor was the Intel Xeon CPU E5-1630 v4 with 3.7 GHz. In
SGD, we used a momentum of 0.9 with a decay of 0. The loss function was the
mean squared error (MSE). Learning rates of 0.06 and 0.03 were used with the
simulated and measured data, respectively.
For each neural network, a hyperparameter tuning study was carried out
to optimize the performance and efficiency of the networks. The hyperparam-
eters included both network architecture parameters and training parameters,
such as learning rates. Since the architecture of the networks were different, the
hyperparameter tuning study varied between the networks in required effort.
The results are reported as the average (or combined in some cases) of the 5-
fold cross-validation. In a cross-validation study, the original data is partitioned
into a training set to train the model, and a validation set to evaluate it. In
5-fold cross-validation, the original data is partitioned into 5 equal size pieces.
From the 5 pieces, a single piece is retained as the validation data to evaluate the
 Estimation of Wave Characteristics via Machine Learning 663

Fig. 5. Mean of the ship motions of each sample from the in-service measurement data.
664 B. Duz et al.

Fig. 6. Standard deviation of the ship motions of each sample from the in-service
measurement data.
 Estimation of Wave Characteristics via Machine Learning 665

Fig. 7. Hs, Tp, Dp and Vs of each sample from the in-service measurement data.

model, and the remaining 4 pieces are used as training data. The cross-validation
process is then repeated 5 times (the folds), which allows each of the 5 pieces
being used exactly once as the validation data. The 5 results from the folds can
then be averaged (or otherwise combined) to produce a single estimation. The
advantage of this method is that all samples in the original data are used for
both training and validation, and each sample is used for validation exactly once.

4.2 Evaluation Metrics

The results are reported as the 95% error level. This means that 95% of the
predictions errors is less than or equal to this value. Additionally, mean and
standard deviation of the prediction error are provided.
666 B. Duz et al.

4.3 Results from the In-service Measurement Data

The 5-fold cross-validation with the measurement data was carried out in two
fashions:

• Chronological: The original data was first ordered chronologically. Without

shuffling, the 5-fold cross-validation procedure was then followed.
• Shuffled: The original data was shuffled first, and afterwards the 5-fold cross
validation was carried out.

Figure 8 shows the convergence of the training and validation losses from the
two neural networks for Dp. These losses are presented as the average and spread of
5-fold cross validation. The histograms in Fig. 9 show how the validation errors for
Dp are distributed. It can be seen that the errors are distributed symmetrically, and
the neural networks show better performance for the shuffled data. Figures 10 and
11 show the scatter plots of truth and prediction values on the shuffled and chrono-
logical measured data, respectively. Figure 12 illustrates the validation error of Dp
with respect to Hs and Tp on the shuffled measured data using MLSTM-CNN. It
can be seen that the validation error of Dp does not have any bias with respect to
Hs or Tp. Similar observations were made for the validation errors of Hs and Tp
as well. Table 1 lists the results for the evaluation metrics from the two networks
applied on the shuffled and chronological measured data.

(a) MLSTM-CNN on the shuffled data (b) Sliding Puzzle on the shuffled data

(c) MLSTM-CNN on the chronological (d) Sliding Puzzle on the chronological

data data

Fig. 8. Training and validation losses from the two networks on the shuffled and chrono-
logical measured data for Dp. Note the much larger spread in validation losses for the
chronological data.
 Estimation of Wave Characteristics via Machine Learning 667

Table 1. Results from the two neural networks applied to the shuffled and chronological
measured data.

Train./Valid. Wave 95% level val. Mean of val. error Std of val. error Network name
data characteristic
Shuffled Dp [deg] 24.86 −0.05 13.59 MLSTM-CNN
21.67 −0.20 11.71 Sliding puzzle
Hs [m] 0.41 0.0 0.20 MLSTM-CNN
0.39 0.0 0.19 Sliding puzzle
Tp [s] 1.01 0.0 0.50 MLSTM-CNN
0.95 0.0 0.46 Sliding puzzle
Chronological Dp [deg] 37.46 −0.68 19.99 MLSTM-CNN
38.29 0.53 19.35 Sliding puzzle
Hs [m] 0.51 0.0 0.27 MLSTM-CNN
0.51 0.0 0.27 Sliding puzzle
Tp [s] 1.40 0.01 0.68 MLSTM-CNN
1.48 0.01 0.69 Sliding puzzle

(a) MLSTM-CNN on the shuffled data (b) Sliding Puzzle on the shuffled data

(c) MLSTM-CNN on the chronological (d) Sliding Puzzle on the chronological

data data

Fig. 9. Histograms of the validation errors from the two networks on the shuffled and
chronological measured data for Dp.
668 B. Duz et al.

180 150 180 150

135 135

90 90
100 100
45 45
Truth [deg]

Truth [deg]
0 0

-45 -45
50 50
-90 -90

-135 -135

-180 0 -180 0
-360 -270 -180 -90 0 90 180 270 360 -360 -270 -180 -90 0 90 180 270 360
Prediction [deg] Prediction [deg]

(a) Dp (MLSTM-CNN) (b) Dp (Sliding Puzzle)

4.5 150 4.5 150

4 4

3.5 3.5
100 100
Truth [m]

Truth [m]

3 3

2.5 2.5
50 50
2 2

1.5 1.5

1 0 1 0
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Prediction [m] Prediction [m]

(c) Hs (MLSTM-CNN) (d) Hs (Sliding Puzzle)

12.5 150 12.5 150

11.5 11.5

10.5 100 10.5 100

Truth [s]

Truth [s]

9.5 9.5

8.5 50 8.5 50

7.5 7.5

6.5 0 6.5 0
5.5 6.5 7.5 8.5 9.5 10.5 11.5 12.5 5.5 6.5 7.5 8.5 9.5 10.5 11.5 12.5
Prediction [s] Prediction [s]

(e) Tp (MLSTM-CNN) (f) Tp (Sliding Puzzle)

Fig. 10. Truth and prediction values for Dp, Hs and Tp from the two networks on the
shuffled measured data.
 Estimation of Wave Characteristics via Machine Learning 669

180 150 180 150

135 135

90 90
100 100
45 45
Truth [deg]

Truth [deg]
0 0

-45 -45
50 50
-90 -90

-135 -135

-180 0 -180 0
-360 -270 -180 -90 0 90 180 270 360 -360 -270 -180 -90 0 90 180 270 360
Prediction [deg] Prediction [deg]

(a) Dp (MLSTM-CNN) (b) Dp (Sliding Puzzle)

4.5 150 4.5 150

4 4

3.5 3.5
100 100
Truth [m]

Truth [m]

3 3

2.5 2.5
50 50
2 2

1.5 1.5

1 0 1 0
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Prediction [m] Prediction [m]

(c) Hs (MLSTM-CNN) (d) Hs (Sliding Puzzle)

12.5 150 12.5 150

11.5 11.5

10.5 100 10.5 100

Truth [s]

Truth [s]

9.5 9.5

8.5 50 8.5 50

7.5 7.5

6.5 0 6.5 0
5.5 6.5 7.5 8.5 9.5 10.5 11.5 12.5 5.5 6.5 7.5 8.5 9.5 10.5 11.5 12.5
Prediction [s] Prediction [s]

(e) Tp (MLSTM-CNN) (f) Tp (Sliding Puzzle)

Fig. 11. Truth and prediction values for Dp, Hs and Tp from the two networks on the
chronological measured data.
670 B. Duz et al.

4.5 12

4
11

3.5
10
3
Hs [m]

Tp [s]
9
2.5

2 8

1.5
7
1
-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150
Validation error of Dp [deg] Validation error of Dp [deg]

Fig. 12. Validation error of Dp versus Hs and Tp on the shuffled measured data using
MLSTM-CNN.

The results show that the machine learning approach works, that there are
some differences in performance between the two networks and that the way the
data is used to train and evaluate the performance has quite some impact on
the results.
To fully evaluate the results, these aspects will be discussed in the next three
sections, together with a comparison to state of the art methods for sea state
estimation.
Comparison to Spectral Sea State Estimation from Ship Motions
The model driven spectral methods use known response functions for mapping
between sea state and ship motions. Although this approach works well for simu-
lated data, it has some difficulty to find the relative wave direction for measured
data. Our best methods seem to compare well to the sea trial results reported
in [4,15].
The main difference is the need for data. While the model driven method
could work for any ship in any condition, the performance of our data driven
method depends heavily on the availability of good quality measured data. How-
ever, this dependency may change in the future. Neural networks have the capa-
bility of generalization, meaning that, for this problem, data coming from multi-
ple different ships can be used to train a general network that can be used with
ships that were not in the training set. This idea is explored in [5] for simulated
data.
Neural Network Structures
MLSTM-CNN and Sliding Puzzle perform similarly on both datasets. Even
though MLSTM-CNN adds complexity in a parallel track, the global average
pooling greatly reduces the complexity of the convolutional part. Furthermore,
the parallel track supplies a different view on the data, counteracting overfitting
on features as well. The Sliding Puzzle network also deliberately reduces the
complexity by limiting the convolutional stage and collapsing filter results by
 Estimation of Wave Characteristics via Machine Learning 671

different forms of global pooling. This way both methods use a natural form of
regularization, fitting the properties of the data1 .
Chronology of the Data
The measured data was collected over two years, in which the ship encountered
many different sea states. Within the data set, the wave characteristics are esti-
mated accurately though the performance for Tp is worse than for Hs and Dp.
However, it is important to know how good the predictions will be when new
data is seen. Since no samples have been duplicated, there is no contamination
of the validation sets. Still there is a problem.
Keeping the data ordered chronologically before splitting into a training and
validation set shows rather different validation results. Since this is a clear mea-
sure of the expected performance on new, unseen data, the previous case must
have had some sort of contamination. Since we are using time series data for
regression, we enter a realm in which the rules are different. Since the conditions
the ship sailed in gradually change over time, the situation we have here is best
described by a simple analogy. Suppose a network needs to be trained to classify
different breeds of cats in images. The data set that is given contains an equal
number of images for each breed, grouped per breed. If the network is trained
on at least some images for every breed, it will be able to accurately classify any
image from the validation set. If some breeds were absent in the training set,
performance on the validation set will deteriorate. This will happen if the vali-
dation set consists of consecutive samples from the original chronological set. A
common solution is to shuffle the dataset before splitting. Similarly, if we do not
shuffle the ship motion/sea state dataset before splitting, some weather patterns
will not be present in the training set, as they simply only occurred at specific
times.
In a classification problem, the dataset is supposed to contain all classes. In
our regression problem, with a dataset that grows over time, we do not have that
guarantee. So, splitting before shuffling gives an estimate of the performance on
new data. If the performance is significantly worse than for the shuffled set, the
data set is not complete yet, turning this into an important measure. The score
on the shuffled set is still a good indication of the possible performance on a
complete set.

4.4 Results from the Numerical Simulations

Scatter plots of the truth and prediction values from the two networks are shown
in Fig. 13, and the results for the evaluation metrics are listed in Table 2. The two
networks exhibit an almost uniform performance in terms of prediction accuracy
over the entire range of Hs, Tp and Dp. Between the two, MLSTM-CNN performs
better for Dp, while Sliding Puzzle performs better for Tp. The networks perform

1
Designing neural networks to general properties of the data is different from data
snooping, as no sample knowledge is used, just knowledge about the way the data
is acquired.
672 B. Duz et al.

similarly for Hs. Similar to the performance on the measured data, the scatter
around the truth values is larger for Tp compared to Hs and Dp.

Table 2. Results from the two neural networks applied to the simulation data.

Train./Valid. Wave 95% level val. Mean of val. error Std of val. error Network name
data characteristic
Simulation Dp [deg] 1.86 0.02 0.98 MLSTM-CNN
3.63 0.0 1.90 Sliding puzzle
Hs [m] 0.51 0.0 0.24 MLSTM-CNN
0.58 −0.01 0.27 Sliding puzzle
Tp [s] 1.06 −0.02 0.51 MLSTM-CNN
0.74 0.0 0.36 Sliding puzzle

4.5 Reusing the Models Trained on Simulated Data

for the Measured Data
In this section, results are presented from the two networks that were trained
on the simulated data and subsequently reused on the in-service measurement
data. Two approaches were adopted when reusing the models on the measured
data:
– Transfer learning: Here, the decoder layers (the last two dense layers in the
networks) of the trained models were further trained on the in-service mea-
surement data. Note that apart from the decoders, the learned parameters
of all the other layers were not changed. Transfer learning allows using the
learned features of the model trained on the simulated data, and repurposing
these features to a second target model to be trained on the in-service mea-
surement data. If the learned features are general in the sense that they are
suitable on both data sets, then this approach is expected to work properly.

– Direct application: Here, the models trained on the simulated data were
directly applied on the in-service measurement data without changing any
parameter of the model.
Figures 14 and 15 illustrate the results from the two networks when the direct
application and transfer learning approaches are adopted, respectively. Table 3
lists the results for the evaluation metrics. It can be observed that the trans-
fer learning approach results in a considerably better performance with both
networks compared to the direct application approach. This indicates that the
features learned by the filters on the simulated data can be indeed used with
the measured data if they are decoded correctly. Between the neural networks,
Sliding Puzzle performs less worse than the MLSTM-CNN with the direct appli-
cation approach, while MLSTM-CNN performs better with the transfer learning
approach. The 95% levels of validation errors for Dp of 30.62◦ , Hs of 0.42 m
and Tp of 1.13 s are acceptable results opening the door for further studies with
reusing the models trained on simulated data for the measured data.
 Estimation of Wave Characteristics via Machine Learning 673

180 180

135 135

90 90

45 45
Truth [deg]

Truth [deg]
0 0

-45 -45

-90 -90

-135 -135

-180 -180
-180 -135 -90 -45 0 45 90 135 180 -180 -135 -90 -45 0 45 90 135 180
Prediction [deg] Prediction [deg]

(a) Dp (MLSTM-CNN) (b) Dp (Sliding Puzzle)

11 11
10 10
9 9
8 8
7 7
Truth [m]

Truth [m]

6 6
5 5
4 4
3 3
2 2
1 1
0 0
0 1 2 3 4 5 6 7 8 9 10 11 0 1 2 3 4 5 6 7 8 9 10 11
Prediction [m] Prediction [m]

(c) Hs (MLSTM-CNN) (d) Hs (Sliding Puzzle)

16 16
15 15
14 14
13 13
12 12
Truth [s]

Truth [s]

11 11
10 10
9 9
8 8
7 7
6 6
6 7 8 9 10 11 12 13 14 15 16 6 7 8 9 10 11 12 13 14 15 16
Prediction [s] Prediction [s]

(e) Tp (MLSTM-CNN) (f) Tp (Sliding Puzzle)

Fig. 13. Truth and prediction values for Dp, Hs and Tp from the two networks on the
simulation data.
674 B. Duz et al.

180 150 180 150

135 135

90 90
100 100
45 45
Truth [deg]

Truth [deg]
0 0

-45 -45
50 50
-90 -90

-135 -135

-180 0 -180 0
-360 -270 -180 -90 0 90 180 270 360 -360 -270 -180 -90 0 90 180 270 360
Prediction [deg] Prediction [deg]

(a) Dp (MLSTM-CNN) (b) Dp (Sliding Puzzle)

4.5 150 4.5 150

4 4

3.5 3.5
100 100
Truth [m]

Truth [m]

3 3

2.5 2.5
50 50
2 2

1.5 1.5

1 0 1 0
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Prediction [m] Prediction [m]

(c) Hs (MLSTM-CNN) (d) Hs (Sliding Puzzle)

12.5 150 12.5 150

11.5 11.5

10.5 100 10.5 100

Truth [s]

Truth [s]

9.5 9.5

8.5 50 8.5 50

7.5 7.5

6.5 0 6.5 0
5.5 6.5 7.5 8.5 9.5 10.5 11.5 12.5 5.5 6.5 7.5 8.5 9.5 10.5 11.5 12.5
Prediction [s] Prediction [s]

(e) Tp (MLSTM-CNN) (f) Tp (Sliding Puzzle)

Fig. 14. Truth and prediction values for Dp, Hs and Tp from the two networks with
the direct application approach.
 Estimation of Wave Characteristics via Machine Learning 675

180 50 180 50

135 135
40 40
90 90

45 45
Truth [deg]

Truth [deg]
30 30
0 0

-45 20 -45 20

-90 -90
10 10
-135 -135

-180 0 -180 0
-360 -270 -180 -90 0 90 180 270 360 -360 -270 -180 -90 0 90 180 270 360
Prediction [deg] Prediction [deg]

(a) Dp (MLSTM-CNN) (b) Dp (Sliding Puzzle)

4.5 50 4.5 50

4 4
40 40
3.5 3.5

30 30
Truth [m]

Truth [m]

3 3

2.5 2.5
20 20

2 2
10 10
1.5 1.5

1 0 1 0
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Prediction [m] Prediction [m]

(c) Hs (MLSTM-CNN) (d) Hs (Sliding Puzzle)

12.5 50 12.5 50

11.5 11.5
40 40

10.5 10.5
30 30
Truth [s]

Truth [s]

9.5 9.5
20 20
8.5 8.5

10 10
7.5 7.5

6.5 0 6.5 0
5.5 6.5 7.5 8.5 9.5 10.5 11.5 12.5 5.5 6.5 7.5 8.5 9.5 10.5 11.5 12.5
Prediction [s] Prediction [s]

(e) Tp (MLSTM-CNN) (f) Tp (Sliding Puzzle)

Fig. 15. Truth and prediction values for Dp, Hs and Tp from the two networks with
the transfer learning approach.
676 B. Duz et al.

Table 3. Results from the two neural networks with the direct application and transfer
learning approaches.

Train./Valid. Wave 95% level val. Mean of val. error Std of val. error Network name
data characteristic
Direct application Dp [deg] 87.72 −4.91 40.46 MLSTM-CNN
62.83 −1.86 31.60 Sliding puzzle
Hs [m] 3.81 1.50 1.19 MLSTM-CNN
2.70 1.0 0.91 Sliding puzzle
Tp [s] 4.58 2.35 1.32 MLSTM-CNN
4.73 2.85 1.16 Sliding puzzle
Transfer learning Dp [deg] 30.62 −0.06 16.20 MLSTM-CNN
34.23 0.02 18.32 Sliding puzzle
Hs [m] 0.42 0.0 0.21 MLSTM-CNN
0.46 0.0 0.22 Sliding puzzle
Tp [s] 1.13 −0.01 0.57 MLSTM-CNN
1.37 0.0 0.66 Sliding puzzle

5 Conclusions

In this study, machine learning approaches were employed to estimate wave

characteristics from time histories of ship motions. For that purpose, both sim-
ulated and measured data were used to train the models. Two neural network
architectures were considered and their performances were compared. Further-
more, the networks trained on the simulated data were reused for the in-service
measurement data by adopting two approaches, direct application and transfer
learning. The performance of the neural networks with these approaches was
also investigated.
With the measurement data, the methods presented in this paper show good
results and compare well to established methods. By including convolutional
filters to encode the phase relations between the 6-DOF motions, the adopted
neural networks were capable to estimate the wave characteristics accurately
especially for the significant wave height and main direction. The data used
to achieve this does show some shortcomings though and does not cover the
full spectrum. More good quality data is needed to improve methods based on
measured data.
The way the data is used for training is crucial when evaluating method per-
formance. We have seen that there is quite a difference in performance between
using the data chronologically and shuffled. Even though shuffling is common
practise, it is to be used carefully when using data acquired over time. The
advice is to use both training strategies to get insight in both the possible per-
formance and the actual performance when applied to new data.
With the simulation data, the results indicate that the neural networks show
a very good performance in general. Similar to the results on the measured data,
the performance of the networks was better for Hs and Dp compared to Tp.
 Estimation of Wave Characteristics via Machine Learning 677

Finally, the neural networks trained on the simulated data were reused on the
in-service measurement data. The results show that the learned features of the
models trained on the simulated data were indeed suitable for the measured data
as well. Especially when the transfer learning approach was adopted, the neural
networks were able to estimate the wave characteristics accurately. However, this
accuracy level was lower than the one obtained when the neural networks were
directly trained on the measured data itself.
Future work will include more complex numerical simulation studies with
short-crested waves and combined wind and swell sea spectrums. Model test
data will also be utilized for validation. Furthermore, the neural networks will be
trained to estimate the full directional spectrum of the waves (encoder-decoder
networks).

References
1. Tannuri, E.A., Sparano, J.V., Simos, A.N., Cruz, J.J.D.: Estimating directional
wave spectrum based on stationary ship motion measurements. Appl. Ocean Res.
25, 243–261 (2003)
2. Iseki, T., Ohtsu, K.: Bayesian estimation of directional wave spectra based on ship
motions. Control Eng. Pract. 8, 215–219 (2000)
3. Nielsen, U.: Estimations of on-site directional wave spectra from measured ship
responses. Marine Struct. 19, 33–69 (2006)
4. Nielsen, U.D., Brodtkorb, A.H., Sørensen, A.J.: A brute-force spectral approach
for wave estimation using measured vessel motions. Marine Struct. 60, 101–121
(2018)
5. Mak, B., Düz, B.: Ship as a wave buoy - using simulated data to train neural
networks for real time estimation of relative wave direction. In: Proceedings of
the ASME 2019 38th International Conference on Ocean, Offshore and Arctic
Engineering OMAE2019, paper OMAE2019-96225, Glasgow, Scotland (2019)
6. Mak, B., Düz, B.: Ship as a wave buoy - estimating relative wave direction from
in-service ship motion measurements using machine learning. In: Proceedings of
the ASME 2019 38th International Conference on Ocean, Offshore and Arctic
Engineering OMAE2019, paper OMAE2019-96201, Glasgow, Scotland (2019)
7. Karim, F., Majumdar, S., Darabi, H., Harford, S.: Multivariate LSTM-FCNs for
time series classification. arXiv 1801.04503v1 (2018)
8. McTaggart, K., De Kat, J.O.: Capsize risk of intact frigates in irregular seas. Trans.
SNAME 108, 147–177 (2000)
9. Van’t Veer, R., De Kat, J.O.: Experimental and numerical investigation on progres-
sive flooding and sloshing in complex compartment geometries. In: Proceedings of
the 7th International Conference on Stability for Ships and Ocean Vehicles (STAB
2000), February, Launceston, Australia, vol. A, pp. 305–321 (2000)
10. Stredulinsky, D.C., Thornhill, E.M.: Ship motion and wave radar data fusion for
shipboard wave measurement. J. Ship Res. 55, 73–85 (2011)
11. Kiefer, J., Wolfowitz, J.: Stochastic estimation of the maximum of a regression
function. Ann. Math. Stat. 23(3), 462–466 (1952)
12. Chollet, F., et al.: Keras (2015). https://github.com/fchollet/keras
13. Abadi, M., et al.: TensorFlow: large-scale machine learning on heterogeneous
systems (2015). https://www.tensorflow.org/. (Software available from tensor-
flow.org)
678 B. Duz et al.

14. Intel.ai: PlaidML (2018). https://www.intel.ai/plaidml

15. Bispo, I., Filho, A., Tannuri, E., Simos, A.: Motion-based wave inference: monitor-
ing campaign on a turret FPSO. In: Proceedings of the ASME 2016 35th Interna-
tional Conference on Ocean, Offshore and Arctic Engineering OMAE2016, paper
OMAE2016-54956, Busan, South Korea (2016)
 Nonlinear Wave Loads on Deep-Water
Semisubmersibles

Rahul Manohar1, Xin Wang2(&) , Velizar Prampakov1,

and Arun Dev2
1
Frigstad Engineering Singapore Pte Ltd.,
10 Hoe Chiang Road, Singapore 089315, Singapore
2
Faculty of Science, Agriculture and Engineering, Newcastle University,
Newcastle upon Tyne NE1 7RU, UK
[email protected]

Abstract. Among various types of offshore drilling rigs, the semisubmersible

platform is the most popular one for the mid and deep-water exploration
activities thanks to its good motion characteristics. With the oil exploration
activities moving into deeper water and variable deck load capacity growing, the
size of semisubmersible units has increased significantly. Because of the
increased usage of these large units in harsh environments, the prediction of the
extreme wave loads has become more important. The critical responses of the
large semisubmersibles include the longitudinal shear force between pontoons,
the split force between pontoons, and torsional moment. For these giant deep-
water units, the traditional linear analysis overestimates the wave loads, which
leads to a substantial increase of the lightship weight and hence the cost of the
rig. In this paper, we compare the experimental and linear numerical results of
the wave loads of a 7th generation deep-water semisubmersible drilling unit.
The linear numerical calculation uses a three-dimensional panel method in the
frequency domain. The important findings in this paper are: 1. The wave loads
measured in the model tests were 10 to 15% less than the linear analysis, and the
torsional moment showed a maximum difference as about 30%. This result
indicates potentials of a substantial reduction of the lightship weight by using
nonlinear analysis. 2. The linear and nonlinear results also showed different
characteristics to different wave directions in some cases. This difference will
affect the selection of the critical design wave and thus influence the global
structural analysis and the design of the platform.

Keywords: Nonlinear wave-induced loads
Semisubmersible drilling unit

Tank tests
Linear radiation-diffraction calculation
Design wave method

1 Introduction

Among various types of offshore drilling rigs, the semisubmersible platform is the most
popular one for the mid and deep-water exploration activities thanks to its good motion
characteristics. The semisubmersible design has changed a lot since the debut of its first
generation in the early 1960s. In the past few years, the rapid increase in upstream
costs, capital and operating costs put many projects at risk and more vulnerable to the

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 679–696, 2021.
https://doi.org/10.1007/978-981-15-4680-8_46
680 R. Manohar et al.

oil price drop. Cost escalation has been one of the key reasons for project deferments in
recent years. Hence the design optimization without limiting any operational functions
becomes crucial.
A Drilling semisubmersible unit consists of a deck which mainly carries drilling
and other supporting equipment, and living quarters that are supported by vertical
columns, horizontal pontoons, and bracings. The number and arrangements of pon-
toons and columns distinguish many configurational variants employed in the evolution
of the semisubmersibles. Only the four, six, and eight column configurations continue
until today. Similarly only the twin pontoon and the closed array pontoon arrangements
are currently used. A three-column, closed array pontoon (triangular) arrangement has
been proposed for both Semisubmersible Floating Production System and Tension-Leg
Platform applications, and offers a steel reduction opportunity, but these designs have
not been successful, perhaps because of the more complex deck arrangements [1].
Because of the deep-water exploration and requirement to carry more variable deck
loads, the overall semisubmersible sizes have increased a lot. Due to the bigger size and
usage of these units in the harsh environments, there is a significant amount of increase
in the dynamic response of the rigs and the wave-induced loads. The increased sec-
tional forces affects the hull strength and hence the lightship weight. For these giant
deep-water units, the traditional linear analysis overestimates the wave loads, which
leads to a substantial increase of the lightship weight and hence the cost of the platform.
The nonlinear wave-induced hull girder loading has been an important topic for
ship structural analysis [2]. Many numerical and experimental researches have been
carried out focusing on the nonlinear effects of the wave-induced loads for ships [2–8].
In recent years, many researches have also been conducted on the nonlinear wave-
induced loads of offshore floating wind turbines [9–11]. Most of the past researches
have been focused on ships or offshore platforms with simple designs (mainly floating
wind turbines), and there is a lack of published researches on the nonlinear wave-
induced loads of semisubmersible platforms. Besides, our understanding on how the
nonlinearities of the wave-induced loads affect the structural design of the semisub-
mersibles is also not complete.
The main purpose of this paper is to investigate the nonlinearities of the wave loads
and its impact on the “critical design wave” parameters. Linear sectional forces are
derived from linear wave diffraction and radiation theory and compared with the model
test data of one of the biggest semisubmersible drilling units.

2 “Design Wave” Method for Global Response Analysis

2.1 Characteristic Wave Load Responses

When considering the global strength of the semisubmersibles, the following responses
are often considered (see Fig. 1).
1. Longitudinal shear force between the pontoons, FL
2. Split force between pontoons, FS
3. Torsion moment about a transverse horizontal axis, Mt
4. Longitudinal acceleration of deck mass, aL
 Nonlinear Wave Loads on Deep-Water Semisubmersibles 681

5. Transverse acceleration of deck mass, aT

6. Vertical acceleration of deck mass, aV
In oblique waves, the longitudinal forces on the pontoons may introduce opposite
longitudinal displacement for each pontoon, and thus causes bending moment (S-
moment) and longitudinal shear force on the transverse braces. Split force between
pontoons often produces the maximum axial force in the transverse horizontal braces of
a twin pontoon semi-submersible. Critical values of the split force are normally found
at beam-sea with wave length equals to twice of the outer breadth between the pon-
toons. Torsional moment normally gives the maximum axial force in the diagonal
horizontal and diagonal vertical braces. The critical values often occur in oblique waves
with the wave length equals to the diagonal distance between the pontoon ends.
Longitudinal and transverse accelerations of the deck mass introduces shear forces and
corresponding bending moments for the columns connecting the upper and lower hulls.

Fig. 1. Definitions of sectional loads and deck accelerations.

A “design wave” approach is often adopted for semisubmersibles when assessing

the maximum stress under the above loads. The global structural response is assessed
using the “critical design wave” condition. The method for evaluation of the parameters
of the “critical design wave” is explained in the following sections.

2.2 Hydrodynamic Response in Regular Waves

The wave load response in regular waves, RAOðx; bÞ, is obtained for a range of wave
frequency x with various wave headings, b. To investigate the nonlinearities of the
wave loads, linear numerical analysis and model tests are adopted to obtain the linear
and nonlinear responses.
Linear Numerical Calculation. The linear wave loads are calculated using the
commercial software Sesam (DNVGL), which is based on a linear frequency domain
panel method for the three-dimensional radiation-diffraction problem. The actual hull
geometry and weight distribution are represented in the numerical model. The weight
682 R. Manohar et al.

distribution of the places above the upper hull like quarters and the deck houses are
modelled with beams only for simplification of the finite element model. Figure 2
shows the mesh used for the linear numerical computations. The sectional forces and
moments are calculated by integrating the forces acting on all members located on one
side of the center plane.

Fig. 2. The mesh used for numerical computations.

Model Tests. A model of the semi-submersible at a scale of 1 to 50 was used for the
model tests. The model was fitted with a splitting loads frame to measure the wave
splitting loads between the port side and starboard side. The sectional forces and
moments are taken from the centre of the platform by using loads measured at two
cross sections with one near to the design draft level and another at the junction of the
column and upper hull as shown in Fig. 3. During the model tests, the free decay tests
were also conducted to obtain the damping coefficients. These values are also adopted
in the linear numerical computations.

Fig. 3. Setup of the load frame in experiment.

2.3 Environmental Conditions and Response Spectrum

The environmental conditions depend on the locations where the platform is intended
to operate through the entire lifespan. These generally include parameters defining
waves, currents, winds, ice, etc. In this research, only the wave conditions are
 Nonlinear Wave Loads on Deep-Water Semisubmersibles 683

considered, and the JONSWAP spectrum is used to represent the irregular wave field.
The JONSWAP wave spectrum is a peak enhanced Pierson-Moskowitz spectrum and
takes into account of energy flow in a sea state. This type of spectrum is usually applied
for the structural strength analyses of a platform that will operate in harsh environ-
mental conditions. The JONSWAP spectrum SJ ðxÞ is obtained by modifying the
Pierson-Moskowitz spectrum for a developing sea state (Eq. 1).

2 
xxp
exp 0:5 rxp
SJ ðxÞ ¼ Ac SPM ðxÞc ð1Þ

where
h   i
x 4
SPM ðxÞ ¼ 16
5

HS2 x4p
x5 exp  54 xp ;
c = non-dimensional peak shape parameter;
r = spectral width parameter;
Ac ¼ 1  0:287 lnðcÞ is a normalizing factor;
xp ¼ 2p=Tp is the angular spectral peak frequency.
The response spectra of wave loads can be calculated using the amplitude of the
response transfer function, RAOðx; bÞ and the defined wave spectrum SJ ðxÞ (Eq. 2).
The response spectra are obtained for all selected wave spectra and headings. For the i
th wave spectrum and the j th wave heading, the response spectrum is calculated as
below.

ði;jÞ    2 ðiÞ
SR ðxÞ ¼ RAO x; b ¼ bj
SJ ðxÞ ð2Þ

For each response spectrum, the maximum response corresponding to the 90% fractile
in the short-term extreme value distribution can be obtained using Eq. 3.
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi


ði;jÞ
¼ rs 0:5 ln 1  pN ¼ 2:12
rs
1
Respp¼0:9 ð3Þ

N = number of waves (1080 in a 3 h storm);

p = fractile level, 90%;
1
pffiffiffiffiffiffiffiffi 1 2
rs ¼ 2m0 ¼ 2 SR ðxÞdx :
R

2.4 Critical Design Waves

The critical design waves are the waves where the structure produces maximum sec-
tional forces and moments. It is assumed that the structure will give the worst global
stresses under these conditions. The period Td , amplitude ad and heading bd of the
critical design wave are selected corresponding to the maximum responses. Here we get
684 R. Manohar et al.

two sets of critical design waves: one is based on the linear numerical calculations, and
the other one based on the nonlinear model test results.
Firstly, the maximum
n probable
o response among all wave spectrum and headings is
ði;jÞ
obtained as MAX Respp¼0:9 . The corresponding wave heading is also obtained as
bmax . The design wave heading is then chosen as bd ¼ bmax . Secondly, the wave period
Td , of the critical design wave can be evaluated from the response transfer function
(RAOðx; bd Þ) obtained in Sect. 2.1. Td corresponds to the peak-wave-period of the
RAO curve for b ¼ bd . Finally, the design wave amplitude is calculated using Eq. 4.
n o
ði;jÞ
MAX Respp ¼ 0:9
ad ¼ ð4Þ
RAOðT ¼ Td ; b ¼ bd Þ

2.5 Summary
The flowchart of the “design wave” method explained above is summarized in Fig. 4.
The red dotted boxes indicate various comparisons presented in this paper. The global
structural analysis in the flow chart is not included in the scope of this paper.

Fig. 4. Flow chart of the “Design Wave” method.

 Nonlinear Wave Loads on Deep-Water Semisubmersibles 685

3 Platform Particulars and Environmental Conditions

A deep-water 7th generation semisubmersible drilling unit designed by Frigstad

Engineering Ltd is used in this research. The survival draft condition is considered. The
drilling unit is specially designed for deep-water exploration activities. The unit con-
sists of four columns, two pontoons, two connecting braces between each pair of
columns, and the upper hull deck. The columns supporting the upper hull provide the
stability to the structure and submerged pontoons and submerged part of column
provide the required buoyancy to the hull.
The main particulars and design parameters at survival draft condition of the
platform is presented in Table 1. As the platform is symmetric about the x and y axes,
the wave headings in one quarter are considered. Seven wave headings are studied in
this paper, from 0 to 90° with 15° interval. The wave spectra used in this research is
shown in Table 2.

Table 1. Main particulars and design parameters at survival draft.

Parameter Value Unit
Length 117 m
Breadth 81 m
Draft 18 m
Displacement 61,506 T

Table 2. Irregular wave spectra used in this research.

No. Hs (m) Tp (s) c No. Hs (m) Tp (s) c
1 1.40 3.73 5.0 9 13.48 14.03 3.64
2 2.50 4.97 5.0 10 14.98 15.69 2.71
3 3.90 6.21 5.0 11 15.80 15.40 2.50
4 5.62 7.45 5.0 12 15.80 18.43 2.50
5 7.14 8.69 5.0 13 15.80 19.75 2.50
6 8.78 9.94 5.0 14 15.80 21.07 2.50
7 10.42 11.18 5.0 15 15.80 22.38 2.50
8 12.00 12.48 4.72 16 15.80 23.70 2.50

4 Results and Discussions

4.1 Responses in Regular Waves

We first compare the linear and nonlinear wave loads in regular waves. The sectional
forces/moments and deck accelerations explained in Sect. 2.1 are presented and
compared in Fig. 5, 6, 7, 8, 9 and 10. The results are presented as the ratio over the
maximum nonlinear results among all headings measured in model tests. The
686 R. Manohar et al.

horizontal axis is defined as T=T0 , where T0 is the wave period corresponds to the
maximum wave load measured in the model tests.
Longitudinal Shear Force Between the Pontoons, FL (Fig. 5). The linear and
nonlinear results show similar trend. The maximum longitudinal shear force occurs at
45° heading for both the linear and nonlinear results. The maximum value of the linear
calculation is about 10% higher than the nonlinear results from the model tests.
Split Force Between Pontoons, FS (Fig. 6). The linear and nonlinear results show
similar trend. The maximum split force occurs at 90° heading for both the linear and
nonlinear results. This is as expected that the beam wave induces the maximum split
force between the pontoons. In this case the bracings plays a critical role and have
compressions and tensions forces within the bracing structure. The maximum value of
the linear calculation is close to 20% higher than the nonlinear results from the model
tests.
Torsion Moment About a Transverse Horizontal Axis, Mt (Fig. 7). Some dis-
crepancies are seen in the trend between the linear and nonlinear results. The maximum
torsion moment occurs at 60° heading for the linear results and 45° for the nonlinear
results. The maximum value of the linear calculation is about 30% higher than the
nonlinear results from the model tests, which may have a great impact on the structural
design.
Longitudinal and Transverse Acceleration of Deck Mass, aL and aT (Fig. 8 and
Fig. 9). For both the longitudinal and transverse accelerations, the linear and nonlinear
results show identical trend. Maximum longitudinal acceleration occurs at 0° while
maximum transvers acceleration occurs at 90°. The maximum value difference between
the linear and nonlinear results is about 10% to 15%.
Vertical Acceleration of Deck Mass, aV (Fig. 10). The linear and nonlinear results
show quite difference trends. The maximum vertical acceleration occurs at 90° heading
for both the linear and nonlinear results. However, the corresponding peak wave
periods are different between the linear and nonlinear results. The maximum value of
the linear calculation is about 15% higher than the nonlinear results from the model
tests.
Summary. The maximum values from the linear numerical calculations and nonlinear
model tests are summarized and compared in Fig. 11. The results are presented as the
ratio of the linear sectional loads over the nonlinear results. In summary, the linear
results are averagely 10%–20% larger than the nonlinear results. Particularly for the
torsion moment and vertical deck acceleration, discrepancies can also be observed in
the curve trends. At the survival draught, the variation of waterline plane area, the water
on top of the pontoons and the penetration of the bracings will introduce nonlinear
effects that are not taken into account in the linear radiation-diffraction calculations.
The above results indicates that this nonlinearity may have a significant impact on the
structural design. The reduction of the wave-induced loads measured in model tests
suggests the structural design of the platform can be further optimized in terms of steel
weight.
 Nonlinear Wave Loads on Deep-Water Semisubmersibles 687

Fig. 5. Comparison of longitudinal shear force between the pontoons.

688 R. Manohar et al.

Fig. 6. Comparison of split force between pontoons.

 Nonlinear Wave Loads on Deep-Water Semisubmersibles 689

Fig. 7. Comparison of torsion moment about a transverse horizontal axis.

690 R. Manohar et al.

Fig. 8. Comparison of longitudinal acceleration of deck mass.

 Nonlinear Wave Loads on Deep-Water Semisubmersibles 691

Fig. 9. Comparison of transverse acceleration of deck mass.

692 R. Manohar et al.

Fig. 10. Comparison of vertical acceleration of deck mass.

150%
125%
100%
75%
50%
25%
0%
F_L F_S M_t a_L a_T a_V

RAO-MAX (Exp-nonlinear) RAO-MAX (Num-linear)

Fig. 11. Comparison of maximum sectional loads and deck accelerations.

 Nonlinear Wave Loads on Deep-Water Semisubmersibles 693

4.2 Maximum Short-Term Response and Critical Design Waves

The maximum short-term response among all headings can be obtained following the
method explained in Sect. 2.3. The comparison between the linear calculation and the
nonlinear model test results are presented in Fig. 12. For the longitudinal shear force,
the linear and nonlinear results of the maximum response show negligible difference.
However, for the other sectional loads and deck accelerations, the linear calculation
results are significantly larger than the nonlinear model test results. Figure 13 shows
the comparison of the wave headings corresponding to the maximum response. Only
the torsion moment has a different maximum response wave headings.

150%
Maximum Response

125%
100%
75%
50%
25%
0%
F_L F_S M_t a_L a_T a_V

Max. Resp. (Exp-Nonlinear) Max. Resp. (Num-Linear)

Fig. 12. Comparison of maximum short-term responses.

100
80
Wave Heading
(Max. Resp.)

60
40
20
0
-20
F_L F_S M_t a_L a_T a_V

Heading (Exp-Nonlinear) Heading (Num-Linear)

Fig. 13. Comparison of wave headings corresponding to maximum responses.

Following the approach explained in Sect. 2.4, the parameters of the critical design
waves now can be derived. The derived critical design wave parameters based on linear
and nonlinear analysis are presented in Fig. 14, 15 and 16. Regarding the critical
design wave heading, only the torsion moment shows some difference. The critical
heading is 45° based on the nonlinear results and 60° using linear analysis. The critical
wave periods are identical for the linear and nonlinear analysis. In Fig. 16, it is evident
694 R. Manohar et al.

that the nonlinear analysis gives a significantly smaller amplitude of the critical design
wave. The only exception is the torsion moment where the nonlinear analysis gives a
slightly higher wave amplitude for the nonlinear analysis.

100
80
Wave Heading

60
40
20
0
-20
F_L F_S M_t a_L a_T a_V

Experiment-nonlinear Numerical-Linear

Fig. 14. Wave headings of the critical design waves.

14
12
Wave Period (s)

10
8
6
4
2
0
F_L F_S M_t a_L a_T a_V

Experiment-nonlinear Numerical-Linear

Fig. 15. Wave periods of the critical design waves.

120%
Wave Amplitude Rao

100%
80%
60%
40%
20%
0%
F_L F_S M_t a_L a_T a_V

Experiment-nonlinear Numerical-Linear

Fig. 16. Wave amplitude ratios of the critical design waves.

 Nonlinear Wave Loads on Deep-Water Semisubmersibles 695

5 Conclusions

In this paper, the nonlinear effects of the sectional forces/moments and deck acceler-
ations of an ultra large semisubmersibles and their impact on the “design wave”
selection have been investigated. The important findings are summarized as below.
• In general, the nonlinear effects are significant. The linear calculation over estimates
the sectional loads and deck acceleration by 10% to 30%. As the current structural
design is mainly based on a linear assumption, it is possible that structural design
can be further optimized to reduce the total weight of the platform if a nonlinear
analysis approach is adopted.
• It is evident that the nonlinear effects are particularly strong for the torsion moment
and the vertical acceleration of the deck. Besides significant differences in the
response amplitude, the linear and nonlinear results also show clear differences in
the trend of the response curves.
• The nonlinear effects does not affect significantly the headings and the wave period
of the critical design wave, except for the torsion moment.

6 Future Works

The global strength analysis can be conducted based on the nonlinear responses of the
wave-induced loads. The structural design then can be optimized to reduce the total
lightship weight of the platform.
Besides, an in-depth study may be conducted for the designs of future semisub-
mersibles by classifying the designs into a few categories. As we know from the study
that some particular periods are more critical for the design of semisubmersible which
can be thoroughly checked in the model tests. We can have results with more accuracy
including non-linear effects which may be standardized to some extent by considering a
factor based on the type and the category of the Semisubmersibles.
A special consideration for fatigue prone areas is required if the overall thickness of
hull panels has been reduced from the original designs. Hence it is really important to
consider the fatigue life too.

References
1. Chakrabarti, S.K.: Handbook of Offshore Engineering. Elsevier, Amsterdam (2005)
2. Jensen, J.J.: Committee VI.1. Extreme hull girder loading. In: Ohtsubo, H., Sumi, Y. (eds.)
Proceedings of the 14th International Ship and Offshore Structures Congress 2000,
Nagasaki, Japan, ISSC 2000, vol. 2, pp. 263–320 (2000)
3. Watanabe, I., Ueno, M., Sawada, H.: Effects of bow flare shape to the wave loads of a
container ship. JSNAJ 166, 259–265 (1989)
4. Chen, R., Du, S., Wu, Y., et al.: The experiment on extreme wave loads on a flexible model
of the s 175 container ship. China Ship Scientific Research Center Report (1999)
696 R. Manohar et al.

5. Xia, J., Wang, Z., Jensen, J.J.: Non-linear wave loads and ship responses by a time-domain
strip theory. Mar. Struct. 11(3), 101–123 (1998)
6. Jensen, J.J., Pedersen, P.T.: Wave-induced bending moments in ships - a quadratic theory.
Trans. RINA 121, 151–165 (1979)
7. Fonseca, N., Guedes Soares, C.: Time-domain analysis of large-amplitude vertical motions
and wave loads. J. Ship Res. 42(2), 139–153 (1998)
8. Watanabe, I., Sawada, H.: Effects of elastic responses to the longitudinal bending moment in
two-directional irregular waves. Naval Archit. Ocean Eng. 24, 91–102 (1986)
9. Philippe, M., Courbois, A., Babarit, A., et al.: Comparison of simulation and tank test results
of a semi-submersible floating wind turbine under wind and wave loads. In: 32nd
International Conference on Offshore Mechanics and Arctic Engineering, vol. 8. ASME
(2013)
10. de Bruijn, R., Huijs, F., Bunnik, T., et al.: Calculation of wave forces and internal loads on a
semi-submersible at shallow draft using an iVOF method. In: 30th International Conference
on Offshore Mechanics and Arctic Engineering, vol. 7. ASME (2011)
11. Rivera Arreba, I.: Computation of nonlinear wave loads on floating structures. Master thesis,
NTNU (2017)
 Experimental and Numerical
Comparisons of a Floating Absorber
Wave Energy Converter in Regular Waves

A. Novás-Cortés2 , L. Santiago-Caamaño1 , M. Miguez-González1 ,

and V. Dı́az-Casas1(B)
1
Integrated Group for Engineering Research, University of A Coruña, Ferrol, Spain
[email protected]
2
Rotary Wave S.L., Valencia, Spain
http://www.udc.es

Abstract. One of main problems of wave energy development is high

cost of devices development. Thus the time required to simulate these
devices with Computational Fluid Dynamics difficulty the early stages of
design process. In this paper the used of simplificated numerical models
are used to reduce this time in presented.
The aim of this paper, is the comparison and validation of floating
absorber of wave energy, between open source software simulations, based
in potential theory and tests in a wave tank.
Therefore, in the presented cases mathematical models in frequency
domain have been computed through Nemoh to obtain the hydrodynamic
coefficients, (added mass and damping matrix) and these coefficients are
included in BEMIO and WEC-Sim, which execute numerical model in
time domain, to tests simulation.

Keywords: Wave energy converters · Marine energy

1 System Description
The present document describes the study of dynamic behavior of one float,
which form part of Butterfly WEC. The Butterfly wave energy converter, is an
attenuator wec, and its composed by a floats with transversal arm coupled to a
freewheel which makes rotates a longitudinal shaft when the wave passed.
The tested model is 2,570 m long, 1,222 m width, 0,397 m high and 74,6 kg
weighty. The model has nine floats in both sides, i.e., the model has two shafts,
The research that has led to this publication has been funded by CDTI and sup-
ported by the Ministry of Economy, Industry and Competitiveness through the project
“DEVELOPMENT OF AN INNOVATIVE SYSTEM FOR THE GENERATION OF
ELECTRIC ENERGY AND DRINKING WATER FROM MARINE ENERGY AND
ITS APPLICATION TO REMOTE ZONES IN COLOMBIA” with file number INNO-
20171035 of the INNOGLOBAL 2017 call.
c Springer Nature Singapore Pte Ltd. 2021
T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 697–706, 2021.
https://doi.org/10.1007/978-981-15-4680-8_47
698 A. Novás-Cortés et al.

coupled to nine floats each one. The shafts are coupled to hydraulic pump, which
pressurize water and sent it to shore.
The present paper analyzes the numerical simulation, of the system Arm-
Float of Butterfly WEC, as it showed in Fig. 2, and compare the results with
wave tank experiences results (Fig. 1).

Fig. 1. Butterfly WEC

Fig. 2. System to study

The arm-float characteristics are the following:

Parameter Value Unit

Mass 0,5322 kg
XG 0,0070 m
YG 0,1424 m
KG 0,053 m

2 Numerical Modelling and Simulations

The general motion equation, in time domain, is given by:

mξ¨ = Fe (t) + Fr (t) (1)

 A Coruña University 699

where ξ¨ is the float elevation, Fe (t) is the excitation force, which could be
descomposed into two components: the Froude-Krylov force(FF K ) and diffrac-
tion force (FS ). Fr (t) is the radiation force.

mξ¨ = FF K + FS + Fr (t) (2)

The objective is obtain ξ¨ from software simulation and compare it with wave
tank experiences [1–3].
The main used software, WEC-Sim is an opensource time domain code that
solves the for the system dynamics of WECs. The code is based in Boundary Ele-
ment Method, it calculate the dynamic response solving the equation of motion
based in Cummins equation [4]. The WEC-Sim software calculate the excita-
tion force (Fe (t)), the radiation force (Fr (t)) using hydrodynamic coefficients
provided by the frequency-domain BEM solver, in this case Nemoh.
OpenWEC software in an open source WEC Simulator, which allow perform
numerical simulations of custom made WEC shapes in all kind of wave condi-
tions. The hydrodynamic parameters are calculated using the open-source BEM
solver Nemoh, which assume the potential flow theory and use the first order of
waves theory (fully linear) [5,6].
The first step was mesh the system and simulate it in frequency domain
in OpenWEC, the results were include it in BEMIO code to obtain the .h5 file,
necessary to WECSim simulation [7–9]. Bemio computes the following equations
of Radiation impulse response, Eq. 3 and excitation impulse response, Eq. 4.

2 ∞ Bi,j (ω)
K̄r,i,j (t) = cos(ωt)dω (3)
π 0 ρ
 ∞
1 Xi (ω, β)eiωt
K̄e,i,β (t) = dω (4)
2π −∞ ρg
The Simulink blocks model is composed by Hydrodynamic body (Float), a
translation constraint, which is the connection between float and arm, a non-
hydrodynamic body (Arm), other constraint, the shaft and the frame, as is shown
in Fig. 3 (Fig. 4).

Fig. 3. Arm-Float block scheme in Simulink

700 A. Novás-Cortés et al.

Fig. 4. Arm-Float Sim-mechanical model

3 Wave Tank Experiences

The wave tank experiences, are to validate the new design of Butterfly WEC
device in relevant conditions The comparative experiences were did it in regular
wave conditions, which main features of waves are the following:

Parameter Value Unit

Direction 180 deg
Type Regular
Height 0,10 m
Period 1,1 s
Frequency 5,712 rad/s
λ 1,889 m
Sw 0,05 m

Wave Tank Description

The experiences were made in the wave tank that A Coruña University has in
Esteiro Campus. The hydrodynamic test basin is 56 m long, 3.80 m wide and
2.00 m deep, equipped with a generator of regular and irregular longitudinal
waves, and a tracking carriage.
The wave generator is of the pivoting flap type, has the ability to generate
both regular and irregular waves according to different typical spectra (Jonswap,
Bretscheneider, TMA), as well as wave spectra defined by the user. Likewise,
the generator has an active wave absorption system, which allows it to greatly
reduce waiting times between different wave tests. Likewise, the generator control
software has a data collection system for shovel movement and wave height at
 A Coruña University 701

different points of the channel, through wave probes located both on the carriage
and on the tank itself.
Relative to the carriage of monitoring and tracking, this is a motorized plat-
form, whose movement supports both a programmable remote control as direct
manual control, allowing accurate tracking of the models or vehicles tested. This
platform has an approximate weight of 250 kg, a race that covers 45 m of the
total 55 m of channel length and dimensions of 5000 × 1130 × 355 mm. Its drive
is electric, and allows a maximum speed of 3 m/s.

Data Collection System

Data collection system is composed by different sensors. On wave tank plat-

form is located a system of motion measurement cameras, capable of analyzing
the dynamics of the models that are tested in said channel, and a wave height
measurement system, in charge of the processing of wave height data generated
(Fig. 5).

Fig. 5. Balls to detect with cameras the float elevation

The camera movement measurement system consists of three high-speed cam-

eras (QUALYSIS Oqus), capable of determining the movement parameters of a
series of markers located on the object whose dynamics you want to know. By
processing the data obtained through the cameras, the system is able to obtain
movement data in 6 degrees of freedom without the need for physical interac-
tion, which is especially useful when the movements you want to know are of
high amplitude.
In addition to the camera system, there is also a wireless measurement inertial
system, consisting of a set of IMUs (Inertial Measurement Units) of the man-
ufacturer XSENSE, composed of MEMS sensors that include magnetometers,
gyroscopes and accelerometers, which provide the speeds angular, accelerations,
magnetic field, pressure and orientation in 6 degrees of freedom of the system to
be tested.
702 A. Novás-Cortés et al.

4 Results and Conclusions

4.1 Hydrodynamic Coefficients

The hydrodynamic coefficients in heave and pitch degrees of freedom are showed
in Graphs 6 and 7. According with OPENWec simulations, at experiments fre-
quency, 5,712 rad/s (0,909 Hz in graphs), the coefficients are around this values:

Added mass Hydrodynamic damping

Heave 1,73 3,4
Pitch 0,077 0,012

The main degree of freedom, to evaluate in this case in the heave.

Fig. 6. Hydrodynamic coefficients in Heave

Fig. 7. Hydrodynamic coefficients in Pitch

 A Coruña University 703

4.2 Numerical Results

From Nemoh computing we obtain, the components of equation of motion exci-

tation force and radiation force (Fig. 8), in 5,712 rad/s, experimental wave con-
dition. In addition, in Fig. 8 is showed the added mass coefficient matrix.

Fig. 8. Excitation force, Radiation force, and Added mass coefficients

4.3 Simulation Results

The simulation results from Bemio computing are the following, graphs. The main
values to compare with experimental values are in the range of frequencies between
5–6 rad/s, in experiences the frequency is 5,7 rad/s (Figs. 9, 10, 11 and 12).
704 A. Novás-Cortés et al.

Fig. 9. Normalized Added mass in heave and pitch dof

Fig. 10. Normalized Excitation in heave and pitch dof

Fig. 11. Normalized Radiation Impulse function in heave and pitch dof

Fig. 12. Radiation Damping in heave and pitch dof

In WEC Sim simulations the data obtained is showed in Fig. 13.

 A Coruña University 705

Fig. 13. WEC Sim results

Experimental Results

The wave tank results are showed in Graph 14. As it consider in simulations
results the main variability dof is heave (mm), which it is in medium range
of variation around 20 mm, considering that wave height is 10 mm, the float
elevation is the double of the wave height. The pitch variation is around 5◦ .

Fig. 14. Wave tank results

References
1. Folley, M.: Numerical Modelling of Wave Energy Converters. State of Art for Single
Devices and Arrays (2016). https://doi.org/10.1016/0022-2836(81)90087-5
2. Soares, C.G.: Renewable Energies Offshore (2015). https://doi.org/10.1007/
11823285 121
3. Price, A.A.E., Dent, C.J., Wallace, A.R.: Frequency domain techniques for numerical
and experimental modelling of wave energy converters (2009)
4. Gao, Z., Bingham, H.B., Nicolls-Lee, R., Adam, F., Karmakar, D., Karr, D.G.,
Catipovic, I., Colicchio, G., Sheng, W., Liu, P., Slätte, J., Shin, H., Mavrakos, S.A.,
Jhan, Y.T., Ren, H.: Offshore Renewable Energy (2015)
5. Li, Y., Yu, Y.-H.: A Synthesis of Numerical Methods for Modeling Wave Energy
Converter-Point Absorbers (2012)
706 A. Novás-Cortés et al.

6. Babarit, A.: Theoretical and numerical aspects of the open source BEM solver
NEMOH (2015)
7. Blanco Ilzarbe, J.M., Davidson, J., Engsig-Karup, A.-P., Eskilsson, C., Goteman,
M., Guanche Garcia, R., Harris, J., Jansson, J., Lannes, D., Nava, V., Shao, Y.,
Stansby, P., Yu, Y.-H., Zabala, I.: Hydrodynamics of Wave Energy Converters (2017)
8. Perez, T., Fossen, T.I.: Parametric time-domain models based on frequency-domain
data (2007)
9. Yu (NREL), Y.-H., Coe, R., Ruehl (Sandia), K.: WEC Design Practices and Tools
(2018)
 Study on Motion Response of a Floating
Offshore Turbines’ Unit with OWC

Motohiko Murai1(&) and Ryohei Mochiduki2

1
Yokohama National University, Yokohama, Kanagawa 2408501, Japan
[email protected]
2
Offshore and Engineering Division, Department of Gas Project,
Gas Project Group, Japan Marine United Corporation, Yokohama, Japan

Abstract. Recently the problem of the global warming rises an importance of

using the renewable energy. Wind energy is recognized as one of the key
sustainable energy source. In fact, there are rapidly increase the number of a
wind power plant in land and now the power plant extends to offshore area.
Therefore, a floating offshore wind turbine (FOWT) system which can be
applied to various sea area is attracted. One of the problem is that FOWT must
oscillate in waves. Generally, the FOWT system is horizontal axis system as
same as those in land, however, the motion of FOWT in waves causes relatively
large acceleration at a nacelle. And it is concerned that the acceleration disturbs
the maintenance work.
Because the vertical axis type FOWT has a nacelle, which many of the
mechanical system is in, set near the water level though its efficiency in aero-
dynamics has disadvantage, it is possible that the maintenance work will be
easier. The authors has proposed a concept of twin vertical axis type FOWT. In
our past experiment we have confirmed that the twin system can cancel a gyro
moments induced by each rotation and the suitable distance of the twin turbine
increases the rotational speed of the turbine than the single turbine. On the other
hand, the shape of the underwater structure making the needed buoyancy and
reducing the passive external hydrodynamic force has relatively small stability
inducing acceptable but relatively large pitch motion in waves. So, to reduce the
pitch motion while generating the electrical energy, we propose a new concept
of a floating offshore twin vertical axis turbines system with oscillated water
columns (OWCs). The OWC is one of the type of wave energy converter
(WEC).

Keywords: OWC
WEC
FOWT
Hydrodynamic motion
Power generation

1 Introduction

Most of us as a residence of the earth are beginning to act against the world warming
issue, the Paris agreement is adapted in December 2015 is one of the epoch catching
the trend. To change the energy resource from the fossil fuel to other source is one of
the positive action on these. Then the business on the green/the renewable energy such
as wind energy is attracted as a new progressing business field.

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 707–731, 2021.
https://doi.org/10.1007/978-981-15-4680-8_48
708 M. Murai and R. Mochiduki

In fact, as the change and installation of the renewable energy use facility is rapidly
and ambitiously promoted in the recent years, the ratio of the newly constructed
renewable energy source facility to the total newly constructed energy facility occupied
more than 60% in 2014. Now the occupied ratio of the power by the renewable energy
source for the world power generation is already enhanced to more than 20%. IEA
(International Energy Association) reported that the total amount of the power gener-
ation by using the renewable energy will growing to be able to cover the sum of the
energy demand in USA and EU (7600 TWh) by 2021 [1].
In such world trend, the ratio of the renewable energy source in Japan at last
achieves to 4% due to the increased our conscious to the environment and safeness of
the energy generation after the accident of the nuclear power plant by the huge
earthquake. It is, however, that we have no choice but to say that the introduction
achievement of the renewable energy is quite poor comparing with the international
trend even though Japan has a huge unused EEZ and its potential of the marine
renewable energy. Of course, in Japan several project on FOWT [2] is carrying out and
the introduction of wind generator is gradually increase [3]. But it could say an
acceleration of the introduction achievement is still weak.
The reason of such low introduction achievement of the renewable energy is lack of
the suitable area in which foreign countries mainly use for a mega scale wind farm in
land. Here, “suitable” means not only the wind field environment but also the relation
between the developer vested interest and the candidate of the development site. By the
consideration of the complex landlord vested interest or the wind field problem, to
examine of using an ocean space has room. And as a technology for the unused ocean
space, there is a floating offshore wind turbine system. In the present, some experi-
ments in the presentation and the demonstration stage on the floating offshore wind
turbine system are being carried out in the actual offshore site.
In nowadays, the FOWT in the presentation/demonstration stage is almost hori-
zontal axis type FOWT. This is because that the experience on a large scale wind
turbine system including a nacelle and blade technology has been matured based on the
experience in land. In land, the wind turbine technology is stood on the system fixed to
the land. It would be natural to have a question whether to adapt simply the land based
technology to the ocean space utilization is a best solution or not. One of the reason is
the dynamic motion of a floating structure in waves. To adapt a floating system has a
lot of merit for example it is not depend on the water depth of the related site. On the
other hand, it cannot escape from the dynamic motion induced by wave loads. It in the
horizontal axis type is apt to concentrate the mass which is in a generator and a gear
box to the top side of the structure. And the top heavy system must increase the scale of
the structure in horizontal for increasing restoring force and in vertical for making the
gravity center lower down. As a result, the larger floating system push up the cost of the
construction and its mooring system for mooring the heavier.
Furthermore, when we consider the maintenance work of the floating system, it will
be better that the devices specially frequently needed the maintenance work is set on
the lower for easy to access from the sea level. In such consideration, it might be a
possibility that a vertical axis system of a FOWT which has not so much experience in
land is a good candidate.
 Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 709

As for the vertical axis type wind turbine system which is often categorized into the
Savonius type using drag force the or the Darius type using lift force is often to say its
power generation efficiency is lower than that of the horizontal axis type wind turbine [4].
So, the authors have been already proposed the floating offshore wind turbine
system with twin vertical axis wind turbines shown in Fig. 1 [5]. In the system, it has
been confirmed that the rotational speed of the turbine is increased than the single one
by setting the twin turbine appropriate distance through our experiments [5–7]. This
directly affects the efficiency increase of the power generation. And one of other merit
is that the gyro moment induced by the rotational turbines can be canceled in the
system by rotating both turbines in the opposite direction.

Fig. 1. An image of the floating offshore wind turbine system with twin vertical wind turbines.

One the other hand, we designed its shape within the sufficient stability to reduce
the wave forces, but we notice the importance of providing an additional restoring force
becoming itself more safety.
In this paper, we discuss on a possibility of the new concept in which OWCs
(Oscillating-Water-Column) are attached to the already proposed floating offshore wind
turbine system with the twin vertical turbines. As well known, WEC generates the
energy from the wave energy by acting as a damper system. And there has been many
researches on OWC itself via theoretical, numerical and experimental approach [8–11]
and some projects of WEC based on the OWC concept had been done [12]. In this new
concept, we consider that the OWCs can reduce pitch and heave motion of the floating
system by its restoring effect and its damping effect while generating the electrical
energy in waves. The reduction of the pitch and heave motion in waves should increase
safety and the efficiency of the power generation of the twin wind turbine systems. One
of the feature points of this concept attaching the OWCs at the corners of the original
structure has a lever which can increase the larger potential of the OWCs’ local vertical
motion which is the source of generating power due to the heave motion and the pitch
motion.
710 M. Murai and R. Mochiduki

2 Analyses Method

The assumed model used in this research is a FOWT system with 4 OWCs attached to
the floating wind turbine system with twin vertical axis turbines already proposed by
the authors as shown in Fig. 2. As for the analyses, we used parallel and appropriately
the in house code based on the singularity distribution method [13–17] and a com-
mercial software ANSYS-AQWA.

Fig. 2. An image for numerical analyses by ANSYS-AQWA of the floating offshore wind
turbine system with twin vertical wind turbines.

First of all, we explain the dynamic equation including the force by the OWCs. We
assumed that the floating structure is a rigid body and the dynamic motion in waves are
treated as 6-DOF. The definition of the 6-DOF is shown in Fig. 3. Subscription 4, 5,
and 6 are the rotational angle around x axis in counter clockwise from y axis, the
rotational angle around y axis in counter clockwise from x axis, and the rotational angle
around z axis in counter clockwise from x axis respectively.
Here, the dynamic equation of the floating body in regular waves of radius fre-
quency x on the steady harmonic mode is written as following:

 X 6
  
Fj ¼ x20 Mij þ mij  ix0 Nij þ Cij ½ni  ð1Þ
i¼1

n is a complex amplitude of the relevant dynamic mode. When the effect of the
OWCs is additionally taken into account to the dynamic equation (1), the equation is
given as the follows using the additional force by OWC Fo .

FO ¼ CO n_ i þ KO n , FO ¼ ðixCO þ KO Þn ð2Þ
 Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 711

Fig. 3. A coordinate system and the definition of motions for the floating structure.

As the equation shows that CO and KO are defined as the coefficients related to the
damping term and the restoring term by the OWCs respectively. Then we assume and
define the force induced by the OWC motion as follows:

CO ¼ qgAwo
a ð3Þ

KO ¼ qgAwo
b ð4Þ

Here, Aw is the water line area of OWC from outer view, and a and b are the
arbitrary coefficients. The meaning of the equation is that if b = 0 is no restoring force
given by treated the OWC as a tubular shape, and b = 1 is the OWC treated as a solid
shape with no hole. So, the realistic possible range of b is defined as 0  b  1.
As for a, generally it would be changed by the performance and function of the
assumed OWC mechanism and given as the parameter of the OWC. But, in this paper,
we want to examine that an acceptable or possible range of a given as the virtual OWC
performance. The image of the function or role of a and b is illustrated in Fig. 4.
The equation considering these is as follow.

 X 6
     
Fj ¼ x20 Mij þ mij  ix0 Nij þ CO;ij þ Cij þ KO;ij ½ni  ð5Þ
i¼1

Next, we explain a prediction method for a generated power by the OWC due to its
local heave motion. The output power generated by OWC is written as follows.
 
P ¼ FO  n_ OWC;i ¼ CO
n_ OWC;i þ KO
nOWC;i  n_ OWC;i ð6Þ
712 M. Murai and R. Mochiduki

Fig. 4. A sketch of the role of a and b in the numerical model of the OWC.

Here, nowc,i means that the local heave motion of the OWC due to i-th mode motion
of the whole structure. Because we consider that heave motion and pitch motion affect
the local heave motion of OWC, each nowc,i is given as follows.

nOWC;3 ¼ n3

nOWC;3 ¼ tan n5  ‘OWC ð7Þ

When n5 is enough small to be able to treat as satisfying the linear relation tan
n5 = n5, the equation can be modified as the follows.

nOWC;3 ¼ n5  ‘OWC ð8Þ

Then we obtain the expectation of generated power by integrating the power

equation for 1 period (T = 2p/x) as shown follows.

ZT ZT  
1 1 1  2
P¼ Pdt ¼ CO n_ OWC;i þ KO nOWC;i
n_ OWC;i dt ¼ x2 CO nOWC;i  ð9Þ
T T 2
0 0

After here, the generated power expectation is obtained by these equations.

3 Numerical Examination
3.1 In Regular Waves
In this section, we carry out the numerical examples. The model used in the numerical
experiment is shown in Fig. 5 and Table 1.
 Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 713

Fig. 5. Sketch of the assumed model.

Table 1. Particulars of the calculation model

The floating structure Total weight 6909 [ton]
Displacement 6740 [m3]
Depth 35 [m]
Length 75 [m]
Width 45 [m]
Center of buoyancy (KB) 20.8 [m]
Center of gravity (KG) 22.0 [m]
Meta center height (GM) 1.52 [m]
Water line area 50 [m2]
Radius of inertia (Kyy) 20.4 [m]
OWC Length 2.5 [m]
Width 2.5 [m]
Water line area 25 [m]
X directional distance from the center of gravity 25 [m]

At first, we examine a relation between C0, K0 and a, b. As already said in the

previous section, it can be guessed that K0 which relates restoring force has stronger
relation to b and C0 which relates damping term has stronger relation to a. So, we first
examine the case of which we fix a = 0 and change b and we fix b = 0 and change a.
The RAO of them are shown in Fig. 6.
714 M. Murai and R. Mochiduki

(a) Heave (α=0)

(b) Pitch (α=0)

Fig. 6. RAO of the heave and pitch motion of the model in various a and b.
 Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 715

(c) Heave (β=0)

(d) Pitch (β=0)

Fig. 6. (continued)

As shown in Fig. 6, b in constant a changes the natural frequency and a in constant

b suppresses the amplitude of the motion. And it is as along our insight. The frame
shape structure under the water line act as a lower hull and the whole structure has a
wave free frequency in heave and pitch motion (in this case, 32 s in heave and 38 s in
pitch). As shown in the figure, we understand that the wave free frequency is not so
changed by a and b.
716 M. Murai and R. Mochiduki

(a) Power by its heave motion (α=0.08)

(b) Power by its pitch motion (α=0.08)

Fig. 7. RAO surface of power due to heave and pitch motion in constant a.
 Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 717

(c) Power by its heave motion (α=0.16)

(d) Power by its pitch motion (α=0.16)

Fig. 7. (continued)

The wave free frequency is directly affected by the shape of the structure under the
water line. Therefore, the restoring and damping factor changed by the OWCs are not
so much influence to the wave free frequency. And it seems that the influence of the
existence of the OWC to the pitch motion suppress is relatively larger than those to the
heave motion because of its lever’s length.
One of the object of this research is confirmation of that the motion suppression of
the whole structure by attaching OWCs. The results indicate enough its possibility if
718 M. Murai and R. Mochiduki

the parameter of the OWC can be set within the appropriate range. As for the range, it
is confirmed that the suppression rate of the floating structure is converged and almost
constant in the range over than b = 0.2 through our numerical results. So, in the later
discussion and calculation, we set the range of a being from 0 to 0.2.
Next we explain the effect of OWC to the generated electric power. Figure 7 shows
a RAO surface of the generated power. In the figure, we set a 0.08 and 0.16 constantly
and change b and incident wave period. The RAO surface is separately shown the
power by the heave motion and by the pitch motion in the figure. Observed the overall
trend of the figure on the heave and the pitch motion in Fig. 7, we notice that the pitch
motion affects the generated electric power much than the heave motion.
This trend has been already appeared in the RAO of the motion of the structure.
That is, the larger suppression effect by the OWCs’ in the pitch motion indicates the
more power generation due to the pitch motion. However, since the RAO of the pitch
used for power generation output calculation is the angle per unit amplitude [rad/m],
numerically there is a tendency for the RAO to become greatly around shorter periods’
range. In this regard, it must carefully treat the result.
As for the displacement, the pitch motion is attenuated even if a is small compared
to the heave motion, but it can be seen that the power generation by the pitch motion in
the vicinity of 5 s has a larger peak than the heave motion. That is regardless of b.
The feature point of the power generation by the heave motion from reading the
figure is that of course it is quite high correlation with the heave motion but the range in
which it can generate the certain degree is wide even though the peak is not so high.
As already said, it should be mentioned that because the generated power is depend
on square of x, the expectation of the generated power for unit wave height in longer
wave period apt to be evaluated relatively smaller.
In the next, we observe Fig. 7 from the view point how change the power gener-
ation with b in the same a. b changes the peak width of the power generation by the
heave motion as same as the motion response. However, b does not changed much that
by the pitch motion. That is little bit differ from the motion response. It suggests that
L_owc which acts as lever length affect much to the expectation n of the power gen-
eration by the pitch motion than the change of the motion response. Therefore we
understand that setting b considering the heave motion has a priority from the view
point of controlling the power generation output in the assumed case.
In the next, on the contrary, we observe Fig. 8 from the view point how change the
power generation with a in the same b. It seems that a clear peak appears due to a
change. As shown these, the peaks are always observed in a < 0.1 in any b within the
numerical examination except Fig. 8(a). It is shown both in the heave and the pitch.
And in the case of Fig. 8(a), the absolute value of the generated power is relatively
small. As observed the detail of the shape of the peak, a spike like peak appears in the
heave motion of relatively large b and small a. These results indicate that we set a less
than 0.1 is almost reasonable in any set of b for maximizing the power generation.
These results indicate that we set a for maximizing the power generation less than 0.1 is
almost reasonable in any set of b.
 Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 719

(a) Power by its heave motion (β=0.1)

(b) Power by its pitch motion (β=0.1)

Fig. 8. RAO surface of power due to the heave and pitch motion in constant b.
720 M. Murai and R. Mochiduki

(c) Power by its heave motion (β=0.9)

(d) Power by its pitch motion (β=0.9)

Fig. 8. (continued)

This means that becoming the motion amplitude too small due to a the generated
power becomes smaller because that the generated power given by the square of its
local heave motion speed n_ OWC which is superposition of the heave motion and the
pitch motion. That is that a > 0.1 is too large to generated the electric power from the
dynamic motion even the damping effect increases with the absorbed wave energy in
the damper.
 Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 721

And as shown in the power generation response surface by the pitch motion, the
maximum power is observed vicinity a = 0.05, although the dynamic motion is
maximized at a = 0. This trend is also observed in the heave result especially in wave
period 5–10 s in which the peak makes gentle. This means that there is an optimum
combination of a and b which can reduce the motion in waves but not too small to
generate electric power.

(a) α=0.08

(b) α=0.16

Fig. 9. Total Power RAO surface in various a (Fixed) and b

722 M. Murai and R. Mochiduki

The difficulty of the optimization is that the generated electric power is determined
by the local OWC motion which is instantaneously coupled with the heave and the
pitch motion. And furthermore, there is 90° phase gap between the pitch motion and
the heave motion in generally. So, the total generated power cannot to be a simple sum
of the generated electric power by the heave motion and by the pitch motion and
sometimes those are canceled each other. Therefore, we use the following equation to
estimate the power generation taking into the coupled effect of the heave and pitch
motion.

nOWC;35 ¼ n3 þ ‘OWC
n5
 
Pall ¼ Fg
n_ OWC;35 ¼ Cg
n_ OWC;35 þ Kg
nOWC;35
n_ OWC;35 ð11Þ

Numerical result of the total generated power re-obtained by the equation are
shown in Fig. 9.
The results shown in Fig. 9 are obtained in the condition of a = 0.08 and 0.16.
Looking at the figure, there is a peak of the power generation output in the wave
period of around 5 s, and it can be seen that the range of the periodic band in which the
power generation is relatively increased expands with the increase of b. As for this, it is
thought that the pitch motion as the whole floating body contributes mainly as inferred
from the consideration so far.
In addition, Fig. 10 shows the response surface of the power generation output to a
and the incident wave period when b is fixed. Looking at the figure, there is a main
gentle peak in the vicinity of incident wave period of 7 s and a = 0.08, and another
steep peak are also observed in the vicinity of incident wave period of 15 s and small a.
For the later one, it seems that the natural period of heave motion as the whole
floating structure affects. In addition, it can be seen that as a increases, the relative
heave motion is suppressed.
On the other hand, from the viewpoint of the coupling effect of motion, in this
model, we cannot observe the response that the power generation significantly and
obviously drops while there were a possibility of that its heave and pitch motion are
canceling each other. To confirm this point, the phase of the heave motion and the pitch
motion is shown in Fig. 11.
Looking at this, it can be confirmed that the phase difference between the heave and
the pitch is surely changing around 90° in the relatively long period range. But in the
frequency band where substantial power generation can be expected, the phases fluc-
tuate from each other, a coupling that the heave and pitch motions cancel each other has
not appeared. Because a as a damper for power generation causes forces like action and
reaction to act on the motion interference between the heave and the pitch, it is possible
that such particular wave period is not conspicuous as differ from in the case where
OWC is not attached. As a result, it seems that results like adding together the power
output by the heave and the pitch motion are coming out.
Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 723

(a) β=0.4

(b) α=0.8

Fig. 10. Total Power RAO surface in various a and b (Fixed)

724 M. Murai and R. Mochiduki

(a) Heave ξ3 (b) Pitch ξ5

(c) OWC part: ξOWC,35 (d) Heave-Pitch

Fig. 11. RAO of phase of the motions and the phase difference between Heave and Pitch
motion.

3.2 Expected Value of the Motion and the Power Generation in Irregular
Waves
Up to the previous section we have made calculations assuming regular waves.
However, in actual oceans various waves and waves with wave periods are mixed
together to form a complicated sea surface, and it is important to evaluate the response
in irregular waves. Therefore, we estimate the expected value of the motion (heave,
pitch) of the floating structure and power generation amount in irregular waves and
examine in real sea area. In order to predict the response spectrum SH in the irregular
wave, by multiplying the spectral distribution SQQ of the ocean wave (irregular wave)
by the square of the response H (x) in the regular wave as shown below.

SH ðxÞ ¼ jHðxÞj2
SQQ ðxÞ ð12Þ

In this paper, we calculate the response spectra SH of the floating structure motion
and SP of the power generation amount per unit width respectively by using the
response H (x) in regular waves obtained in the previous section and the standard wave
spectrum of JONSWAP as the irregular wave spectrum SQQ.
JONSWAP spectrum is given as follows using the significant wave height: H1/3 and
the significant wave period: Ts.
 
ðTS =f Þ2
SQQ ðxÞ ¼ aH1=3
TS
ðTS
f Þ eð4ðTS
f Þ Þ
c
4 e
5 5
2r2
ð13Þ
 Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 725

0:0624
a¼ 
0:23 þ 0:0336
c  1:9
0:185
þc

8
>
> x 1
>
< 0:07



2p TS
r¼ ; c ¼ 3:3; TS ¼ 1:05T1=3
>
> x 1
>
: 0:09


[
2p TS

In order to evaluate the floating structure motion and the power generation amount
based on the response spectrum SQQ obtained by the above equation, from the response
spectrum obtained here, the representative values rn, rP.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uZ1 uZ1
u u
u u
rn ¼ t SH ðxÞdx; rP ¼ t SP ðxÞdx ð14Þ
0 0

By using the representative value rn showing the response of the floating motion,
we confirm the response change by a and b for each significant wave period of T1/3. In
order to confirm the effect of suppressing the floating motion by the OWC which is also
the object of this research, here we try to evaluate using value ratio r divided by the
representative value of the floating motion without OWC. That is, when the value of
ratio r exceeds 1, it can be evaluated that the floating body fluctuation is increased by
OWC. The result when T1/3 = 8.0 s is shown in Fig. 12.
Looking at Fig. 12, like the trend at in regular waves, the pitch motion compared to
the heave motion it can be seen that the relative change of the floating structure motion
from when not have a OWC is greater. In addition, as the value of b increased, the
tendency of floating motion to increase due to attaching OWC also appeared. This is
because the natural period of the pitch motion shifts to the short period side due to
addition of OWC and increase of b. Also, paying attention to the heave motion, it is
visually observed as the white area showing the boundary line where the floating
structure with OWC becomes unshakable than the floating structure without OWC. The
white area spreads obliquely like that b changes with a in the range of a = 0.04 to 0.1.
This is probably because the significant wave period is close to the natural period of its
heave motion and its influence to occur easily tends widely in the range of the vicinity
periodic bands.
Conversely, paying attention to the pitch motion, the white area does not depend on
the value of b, the range of the value of a is narrow, and the inclination of the white
area is larger than the heave. Like as the consideration in the regular wave, it is thought
that the influence of the lever to the pitch motion is large and the damping effect to the
pitch motion appears sufficiently with relatively small a.
726 M. Murai and R. Mochiduki

(a) Heave

(b) Pitch

Fig. 12. Relative motion ratio comparing with the original one by change of a and b.
(T1=3 ¼ 8:0 s)
 Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 727

Fig. 13. RAO surface of rP changed by a and b (T1/3 = 8.0 s).

Next, in this section, using the expected power generation value r P obtained by a
and b, confirm the change in response to the significant wave period T1/3. The result
when T1/3 = 8.0 is shown in Fig. 13. As you can see from this, it can be seen that the
power generation expectation value is distributed in a gentle hill shape at the apex of
a = 0.05, b = 0.9 differing from the tendency of relative motion. The change in the
power generation expectation value due to the change in b is hardly observed in the
range where the period of the significant wave period is small, but it tends to be a
noticeable change as b increases. This is because, because of the nature of the wave
spectrum, the period band widens as the period of the significant wave increases, so it is
evaluated that power generation can be expected in a wide wave period when obtaining
the power generation expectation value.

3.3 The Power Generation Expectation Value and Safety of the Floating
Structure with OWC in the Actual Sea Generation
Here we will evaluate the safety at installing a wind turbine in the actual sea area. There
are many evaluation items of safety in the windmill, but here we evaluate by the
acceleration caused by the sway of the windmill. Acceleration is used for evaluation in
living environments and the like, and in marine structures such as windmills, the work
environment is largely affected mainly. From the above, we evaluate how much the
acceleration decays due to the effect of the OWC, compared with the floating structure
without OWC.
The evaluation value of the acceleration in a certain sea area is obtained by mul-
tiplying the spectral distribution SQQ of the ocean wave used also in the previous
section by the square of the acceleration response in the regular wave to obtain the
acceleration response in the irregular wave.
728 M. Murai and R. Mochiduki

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uZ1
u
u
racc ¼ t Sacc ðxÞdx ð15Þ
0

In addition, here we evaluate how much acceleration is relatively attenuated by

dividing by the representative value racc changed by OWC at the representative value
racc, 0 of the acceleration of the floating body without OWC attached.
An assumed waters is off the coast of Fukushima Prefecture where the demon-
stration experiment is being carried out, and evaluation examples as shown in Fig. 14
and Fig. 15 were prepared using “Waves and wind database in the waters close to
Japan [18]”.

Fig. 14. Effect of the reduction of the acceleration of floating body by attaching the OWCs.

Fig. 15. Change of the expectation value of generated electric power by the attached OWCs.
 Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 729

Although not shown in this paper, the acceleration of the floating body was surely
decayed around a = 0.04 –.05 in all the waters where evaluation was made from the
overall trend in all of the several assumed waters analyzed this time. This means that
the response is suppressed with a damping force of about 4 to 5% of the restoring force
of the OWC, which means that the trend is not influenced much by the significant wave
period or the significant wave height. In Fukushima prefecture offshore in which the
significant wave period and he significant wave height are 6.8 [s] and 1.67 [m]
respectively shown in Fig. 12 and Fig. 13, it is confirmed that the rotational acceler-
ation of the floating body can be attenuated in the range of a > 0.05. On the other hand,
looking at the change in the amount of power generation in each OWC, high power
generation can be expected around 0.03 < a < 0.06.
The choice of the optimum value varies depending on the weighting of factors in
each case, but in this model, a *0.06 might be a candidate that we can expect the
amount of power generation while suppressing the acceleration by about 20%. In the
case of this model, because it is expected that power generation is possible by sup-
pression of its dynamic motion, and performance and safety as a wind power gener-
ation floating system can be expected to be improved by suppressing acceleration.
Through the discussion, the combination of OWCs and floating offshore wind turbine
systems is considered to be effective.

4 Conclusions

In this study, we assumed a model with OWC mounted on a floating structure part of a
floating twin vertical type wind turbine proposed by the authors. Then we examined
how the floating motion and the electric power generation by its heave and its pitch in
the regular waves change by the parameters on force of OWC. Furthermore, we dis-
cussed the power generation expectation value in irregular waves. Here, we summarize
the knowledge obtained from an example of calculation results.
1. C0 has the effect of suppressing the motion amplitude of the floating body, and K0
confirmed the effect of changing the natural period of the floating body to the short
wave period side respectively. Based on this effect, we introduced the coefficients of
a and b and investigated the motion suppression effect. As a result, a = 0.2: that is,
if we can give 20% damping force of the restoring force by OWC, it is enough for
floating motion to suppress.
2. As a result of checking the influence of parameters simulating the OWC perfor-
mance on the amount of electricity generation, unlike the floating motion, relatively
high power generation output can be obtained at the power generation output in the
vicinity of 8 [s] where its pitch motion becomes large. As for the tendency, although
the heave motion cannot produce a higher power output than the pitch motion, we
confirmed the effect of widening the wave period band in which power generation
can be performed by setting an appropriate b. Even in the power generation output
when the heave and the pitch were coupled, the tendency of obtaining high power
generation output at around 8 [s] did not change, and it was also confirmed that the
power generation effective band is broadened by setting an appropriate b.
730 M. Murai and R. Mochiduki

Moreover, from the transition of the value of a, it was confirmed that by using the
OWC mechanism to generate a damping force of about 6.0 to 8.0% of the restoring
force by the OWC, it is possible to achieve both power generation performance and
motion suppression performance.
3. In the case of suppressing the motion of the heave in the irregular waves, it was
found that the correlation between the restoring force and the damping force by the
OWC strongly appears. In addition, the effect of OWC strongly appears in its pitch
motion compared with its heave motion, so it was possible to attenuate the floating
motion largely by a slight OWC damping force. It was also confirmed that very high
power generation expected value can be obtained when the significant wave period
is 6.0 [s] or more in the power generation expectation value.
4. As a result of investigating improvement of safety and improvement of power
generation ability by adding OWC to the floating structure assuming the sea area
certified as a demonstration field, when the damping force by the OWC is set about
4 to 5% of the restoring force of the OWC, it was confirmed that the hydrodynamic
motion of the floating structure was suppressed by about 80% of the floating
structure without the OWC. And as for the acceleration, we confirmed that choosing
effective parameters reduce its acceleration to 40–60% while generating/absorbing
energy more than 50 kW by the attached OWCs.

References
1. International Energy Agency. http://www.iea.org/. Accessed 05 Mar 2019
2. Fukushima offshore wind consortium. http://www.fukushima-forward.jp/. Accessed 24 May
2019
3. NEDO. http://www.nedo.go.jp/library/fuuryoku/. Accessed 24 May 2019
4. NEDO: Guidebook of Installation of Wind Turbines (2008)
5. Murai, M., Hashimoto, G.: A basic study on the motion of a twin-floating vertical wind
turbines in waves. In: Proceedings of Annual Meeting of JASNAOE, vol. 20, pp. 41–44
(2015)
6. Dabiri, J.: Potential order-of-magnitude enhancement of wind farm power density via
counter-rotating vertical-axis wind turbine arrays. J. Renew. Sustain. Energy 3, 043104
(2011)
7. Kinzel, M., Mulligan, Q., Dabiri, J.: Energy exchange in an array of vertical-axis wind
turbines. J. Turbul. 13(38), 1–13 (2012)
8. JSME, Fluids Engineering Division: Development of OWC type WEC. http://www.jsme-
fed.org/newsletters/2008_12/no5.html#ctop. Accessed 05 Mar 2019
9. de O Falcão, A.F., et al.: OWC wave energy devices with air flow control. Ocean Eng. 26
(12), 1275–1295 (1999)
10. Ikoma, T., et al.: Primary conversion efficiency of OWC type WECs installed on a large
floating structure. In: OMAE 2012-83337, pp. 31–39 (2012)
11. López, I.: Optimisation of turbine-induced damping for an OWC wave energy converter
using a RANS–VOF numerical model. Appl. Energy 15, 105–114 (2014)
12. NEDO: The State of the arts and roadmap of WEC technology. http://www.nedo.go.jp/
content/100107274.pdf. Accessed 24 May 2019
 Study on Motion Response of a Floating Offshore Turbines’ Unit with OWC 731

13. Kim, W.D.: On the harmonic oscillation of a rigid body on a free surface. J. Fluid Mech. 21
(3), 427–451 (1965)
14. Garrison, C.J., et al.: Interaction of waves with submerged objects. Journal of the Water-
Ways, Harbors and Coastal Engineering Division ASCE paper No. 811, pp. 259–276 (1971)
15. Faltinsen, O.M., Michelson, F.C.: Motions of large structures in waves at zero froude
number. In: Proceedings of International Symposium on the Dynamics of Marine Vehicles
and Structures in Waves, pp. 91–106 (1974)
16. Van Oortmerson, G.: The motion of a ship in Shallow Water. J. Ocean Eng. 3, 221–255
(1976)
17. Murai, M., et al.: The hydroelastic response of very large floating structure supported by
footing-columns (2nd report). J. Soc. Naval Archit. Jpn 192, 299–308 (2002)
18. MNRI: Statistical Database of winds and waves around Japan. https://www.nmri.go.jp/
wwjapan/namikazemain.html. Accessed 05 Mar 2019
 Development and Design of a Floating Type
Ocean Current Turbine System

Yasushi Dodo1(&), Shigeki Nagaya1, Tetsuo Okada2,

Makoto Toyoda1, and Akio Ito1
1
IHI Corporation, Kanagawa, Japan
[email protected]
2
Faculty of Engineering, Yokohama National University, Yokohama, Japan

Abstract. Ocean currents are expected as future renewable energy resources.

Recently various ideas are studied all over the world to utilize the power of this
ocean current for generation of electricity. Application of various types of tur-
bines has been proposed for this purpose, but very little examples have been
reported with regard to actual demonstration in the natural ocean current
environment.
The Kuroshio Current which flows through near Japan coast is one of the
strongest ocean currents in the world. To utilize this plentiful energy, we
developed the floating type ocean current turbine system “KAIRYU” under the
support of NEDO (New Energy and Industrial Technology Development
Organization), and we completed the 100 kW order turbine demonstration test
in the natural Kuroshio Current.
In this paper, we report design methodologies and the results of the test.
Especially we focus on the structural design to enable stable floatation in the
water, the control of weight distribution and the center of floatation for balance,
and the pressure-resistant design of the shell structure to secure water-tightness
in the deep sea.

Keywords: Marine renewable energy
Offshore structure
Power generation

Ocean current
Kuroshio
Floating type ocean current turbine
Experiment in
sea
KAIRYU

1 Introduction

Japan’s Exclusive Economic Zone (EEZ) is the sixth largest in the world. Proceeding
development of utilizing ocean renewable energy in the EEZ is required actively in
terms of the reduction in greenhouse gas emissions and energy security. Ocean current
such as Kuroshio flows stably almost all year near Japan. Accordingly, by using this
Japanese own natural enormous energy, it is strongly expected to establish new clean
and stable source of electric power.
On the other hand, there have been many studies on turbine system to utilize the
power of ocean current for generation of electricity [1–5]. Some types were installed on
the seabed, but to make the most of the Kuroshio Current spreading widely near the
water surface, majority of the proposed systems was floating type tethered to the sea

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 732–755, 2021.
https://doi.org/10.1007/978-981-15-4680-8_49
 Development and Design of a Floating 733

floor. Among them, Chen [1] discussed a large platform fitted with dozens of turbines.
Finkl et al. [2] reported an open-center turbine application in the Straits of Florida,
while Shirasawa et al. [3] proposed an energy farm consisting of many single rotor
floaters with torque compensation using a float and a counterweight. Lo et al. [4] and
Lai et al. [5] studied a floating system fitted with a large foil floater at the upper part of
the floating body to induce dynamic lifting force.
In this study, we propose a twin-turbine floating system moored from the bottom of
the sea (Fig. 1), attaching importance to the stable performance on electricity generation
as well as the efficient installation and maintenance. A floating type turbine tethered to
the sea floor enables easy maintenance by uplift to the surface of the sea through
deballasting as shown in Fig. 1. In addition, because the turbine is located well beneath
the water surface in the operating condition, it does not affect ship navigation in the area,
and also it is not influenced by the wave forces, which may be very severe on the surface
of the sea in case of a typhoon. Furthermore, by floating around the relevant depth of the
sea, the turbine can effectively utilize a wide and deep Kuroshio Current.
In this paper, we first present the basic design of the system and describe the pilot
studies with regard to the design concept and the strength of the turbine blades and the
pressure vessel under the deep sea pressure. Next, design and construction of the
100 kW demonstration machine are explained. Demonstration tests were carried out
with regard to the performance of electricity generation, mooring, and strength of the
floating body. In the final sections, we describe the test results, and especially elaborate
on the results of strain measurements in comparison to the structural design and the
finite element analysis, followed by the concluding remarks.

Sea surface
Mooring rope
50 m

Ocean current flow

De-ballasting for maintenance

Sinker or anchor
Turbine system
Riser cable Subsea cable

Fig. 1. Concept view of floating type ocean current turbine system

2 Basic Design
2.1 Design Concept
As explained in the Introduction, the ocean current turbine system operates beneath the
sea surface moored from the bottom of the sea, and is uplifted to the water surface
when maintenance is required. Being installed in deep water, it can tap into the
enormous ocean current energy from the wide ocean area available for power
generation.
734 Y. Dodo et al.

The system adopts horizontal-axis turbine generators, and as such, it is necessary to

cancel the rotor torque so that the floater attitude is effectively controlled during
operation. So we adopted two-turbine system so that the torque in the left and the right
cancels each other. The turbines and the floater were designed to the downwind type,
rotating around the mooring anchor point according to the flow direction. Therefore,
this system can be applied in any current direction without active control (see Fig. 2).
Thus, the ocean current turbine system can achieve stable floatation in the middle
layer of the sea, without significant influence from the surface waves.

Fig. 2. Floating type ocean current turbine system

2.2 Pilot Study on Design Concept

A scale model pilot test was conducted to confirm the design concept as explained in
the previous section. A model with a 1/25 scale about the blade diameter (Fig. 3) was
fabricated and moored in the sea, and its behavior in the water was observed.
The dimensions of this scale model are about 1.8 m in length and 2 m in width,
with turbine blades of 1.5 m in diameter. This scale model is equipped with a buoyancy
adjustment device, an attitude adjustment device and a blade pitch control device to
study the movement of the demonstration machine.
This pilot test was carried out both in a water tank and in the actual sea. Firstly, the
water tank test was conducted at IHI’s towing tank to confirm the stability in the
stationary flow (The angle within ±5° deviation from the target and the depth within
25% deviation from the target). Then, the actual sea test was conducted at the sea area
of Numazu in Shizuoka Prefecture to confirm the stable performance of the floating
body in waiting at the surface of the sea, ballasting under the sea and starting of the
 Development and Design of a Floating 735

turbine. As a result of this study, it was found that the basic concept works. In addition,
it was found that the floating body becomes unstable when the center of gravity is close
to the center of floatation, and thus the geometrical relationship of the center of gravity
and the center of floatation is very important in the water.

Fig. 3. 1/25 scale design concept model

2.3 Pilot Study on Turbine Blades

The material of the turbine blades was planned to be G-FRP. However, there was no
established strength data of a very large blade similar to that used for windmills,
working in the ocean current. So static strength test and destruction test of a 5 m scale
turbine blade (Fig. 4) were carried out, and strength enough was confirmed. In the
static strength test, the model of the turbine blade was simply supported by rigid walls,
and loaded by two hydraulic cylinder at the two points between the supports to simulate
the actual bending moment in the operation.
A turbine blade is designed for the purpose of getting energy from an ocean current.
Its large deformation deteriorates the performance, and should be avoided even in the
intact condition. From our preliminary study we assumed that the deformation should
be within 5% of the span length. As a result of this test, the deformation was about 2%
under the operational load, showing that no performance reduction is expected due to
deformation. The cost of the blades was also confirmed to be within an expected level.
736 Y. Dodo et al.

Fig. 4. Static strength test for 5 m scale turbine blade

2.4 Pilot Study on Performance of the Pressure Vessel in the Deep Sea
To achieve stable floatation in the water, the pressure-resistant design of the shell
structure to secure water-tightness in the deep sea is very important. First, taking account
of the heavy pressure in the deep sea down to 100 m depth, strength of the shell structure
is of primary importance. In addition, there are several flange joints on the shell for human
access and maintenance of the equipment. Actually a man hole is necessary to enable
human access for maintenance, and in some cases large equipment should be taken out or
carried in for maintenance. Thus, water-tightness of these flange joints is very important.
Therefore, to confirm strength and water-tightness of the pressure vessel, a mock-
up model of the main structure was fabricated with the same diameter and the same
structural and joint arrangement as the target demonstration machine, and the pressure
test was carried out at the depth of 100 m (Fig. 5). It was confirmed that there are no
problems with regard to strength and water-tightness of the shell structure. Strain
gauges were arranged around the shell structure of this mock-up model, and the results
were compared with the results of finite element analysis. Details of this comparison
will be explained in Sect. 5.2.

Fig. 5. Pressure test with using mock-up model

 Development and Design of a Floating 737

3 Development of the Demonstration Machine

3.1 Design
From the feedback of the pilot studies as explained in the previous chapter, we
established the basic design of the demonstration machine. To keep stabilized attitude
of the floater in the water, the demonstration machine was fitted with two side turbine
pods and a center pod as shown in Fig. 6. The center pod with a water ballast tank was
arranged at a higher position than the two turbine pods, so that the floating center is
located sufficiently higher than the center of gravity to maintain the self-stability in the
water.
Because we confirmed sufficient water-tightness of the flange connections in the
pilot study, the same flange connections were arranged on the shell structure of the
demonstration machine. The two side turbine pods were connected by a cross beam
structure. This cross beam structure was designed so that it has sufficient strength
against the torque from the two generator turbines. In addition, finite element analysis
was carried out to verify the strength against various loading conditions, including not
only this torque from the turbines, but also the water pressure, the mooring force, the
construction loads such as lifting load, and the wave loads in case of afloat and towing
conditions.
The demonstration machine should have equivalent weight versus buoyancy to be
efficiently operated beneath the sea surface. In addition, the positions of the center of
gravity (and the center of buoyancy) is very important to maintain sufficient attitude
stability of the floating body. Therefore, control and optimization of the weight dis-
tribution of the body and equipment is extremely important.
To achieve this, we employed a 3-dimensional CAD system for the design.
Figure 6 shows the generated 3-dimensional model, comprising all the structures and
equipment, such as structural members, welding bead, electrical controller, bolts and
nuts, lubrication oil and so on. Fully utilizing this model, we could correctly assess the
center of gravity of the entire floating body, and the layout design of the equipment
considering stability was made possible.

Fig. 6. 3-dimentional model of demonstration machine

738 Y. Dodo et al.

3.2 Components
The demonstration machine has some heavy components, which cannot be easily
installed inside the pods, because they are too heavy for manual handling and also the
narrow tunnel-like pod tube structure prohibits usage of cranes. So we divided the
overall body into some sub units which can be assembled outside the floating body, and
realized work reduction for assembly in the tunnel. An example of sub unit division is
indicated in Fig. 7.

Sub unit of Attitude Adjust

Sub unit of main shaft

Sub unit of controller

Fig. 7. Example of sub unit division (inside view of left side pod)

These main components and sub units are indicated below and in Fig. 8. Figure 9
shows the overview picture of the demonstration machine “KAIRYU” in the sea.
• Buoyancy adjustment device; ballast water management system controls the weight
of demonstration machine.
• Attitude adjustment device; ballast shifting among three pods changes the position
of center of gravity to prevent trim and heel.
• Generator; permanent magnet synchronous generators can make variable torque,
which controls the turbine rotation speed at any current velocity.
• Blade pitch control device; pitch angles of 4 turbine blades can be controlled
independently.
• Mooring line; HMPE rope has high strength, low elongation and light weight.
• Power cable; Electricity is transmitted through riser cable and subsea cable. They
also include optical fiber cable. Therefore, monitoring of signals from each
instrument and actuator is possible at the shore side.
 Development and Design of a Floating 739

Generator (Both side )

Control unit (Both side ) Turbine(GFRP)

Transformer

Buoyancy adjustment
device

Attitude adjustment device

( 3 portion )

Mooring equipment (Both side)

Blade pitch control system
( Both side ) Power cable

Fig. 8. Main components of demonstration machine

Fig. 9. Demonstration machine of ocean current turbine system, “KAIRYU”

740 Y. Dodo et al.

3.3 Specifications and Capacity of Demonstration Machine

We designed and constructed the demonstration machine based on the considerations
as mentioned above. This demonstration machine was named “KAIRYU” after voting
by the children living near the field of the demonstration test. The specifications of this
demonstration machine are indicated below, and Fig. 10 shows the overall picture of
the demonstration machine.

Specifications and Capacity

– Power generation capacity: Max.100 kW (50 kW  2)
– Rated current velocity: 1.5 m/s
– Assumed current velocity range for power generation: 0.5 – 2.0 m/s
– Turbine diameter: 11 m
– Length of main body: About 20 m
– Width of main body: About 20 m
– Total weight of body: About 380 tons
– Floating depth: 20–50 m/s
– Transmission voltage: 6600 V
– Distance of the moorage from shore: Max. 10 km
– Assumed Service life: 20 year
– Assumed maintenance cycle: 1 year

Fig. 10. Picture of the demonstration machine manufactured actually

 Development and Design of a Floating 741

4 Demonstration Test
4.1 Overview of the Demonstration Test
Towing tests and mooring tests were performed using the demonstration machine
“KAIRYU”. In the towing test conducted first, sensors and actuators are adjusted to
make sure that the balance of the machine in the water keeps stable. In this test, we
collected the data on the generated electric power from the turbines up to its maximum
magnitude of 100 kW.
In the subsequent mooring test, we installed “KAIRYU” in the actual Kuroshio
Current, and confirmed its practical power generation capability on site. As a result, we
could obtain the rated output and the performance of the designed power generation
(about 30 kW at the current speed of 1 m/s).
Moreover, various data were collected in the demonstration tests for further use and
feedback to the future design of larger practical application. These include confirmation
of the performance of the attitude control system, confirmation of the installation and
removal method of the floating body, to name a few.

4.2 Towing Test (Performance Test)

Arrangement of the towing test is shown in Fig. 11. A gravity base anchor was hung
from the barge at the depth of about 100 m. In order that relative arrangement will be
same as that of the mooring test, KAIRYU was tethered to the anchor, and was towed
by a tug together with the barge. This towing test was conducted in the west area of
Kagoshima Prefecture (Fig. 12).

Fig. 11. Arrangement of towing test

742 Y. Dodo et al.

Fig. 12. Towing test area

4.3 Mooring Test

The site of the mooring test was the northern area of Kuchinoshima Island (Fig. 13).
This area is designated as the “demonstration field of the ocean renewable energy” by
the government, and is suitably organized by the local government for this kind of

Fig. 13. Mooring test site

 Development and Design of a Floating 743

Fig. 14. Arrangement of mooring test

demonstration test. The arrangement of the demonstration test is shown in Fig. 14.
KAIRYU was installed at the test point and moored by line anchored to the bottom of
the sea. The electric power generated by turbine flows through the riser cable and the
subsea cable, and is finally transmitted to the base on the barge.

4.4 Test Results

Power generation performance at various velocities was obtained in the towing test.
The relation between the power generation and the current velocity in the towing test is
shown in Fig. 15. The obtained results are observed to agree well with the designed
power curve. In the mooring test, KAIRYU generated approximately 30 kW output at
the current velocity of approximately 1.0 m/s in the actual Kuroshio Current. This also
agrees well with the designed performance.
With regard to the floating stability performance in the Kuroshio Current, we
confirmed the stability of depth and attitude in this demonstration tests. The time
history of the measured depth and attitude is shown in Fig. 16, recorded in the mooring
test when the operation phase changed from the waiting mode at the sea surface to the
power generating mode. According to the results, we can confirm that KAIRYU has an
excellent capability to follow the designated depth. In addition, the maximum angle of
attitude inclination was kept within approximately 5°, which means high stability.
Throughout the mooring test, we could control KAIRYU’s depth within the target
range and attitude inclination angle within ±10°. This high stability comes from the
positional relation between the center of gravity and the center of buoyancy. Thus,
active control by the attitude adjustment device was not used continuously, but only
when it is needed.
744 Y. Dodo et al.

Fig. 15. Comparison of power curve between measured result and designed value

Fig. 16. Depth and attitude (roll/pitch)

Regarding the installation of the mooring test, KAIRYU was conveyed to an area
near Kuchinoshima and installed with gravity base anchor, mooring line and power
cable (see Fig. 17). The process of installation including laying down the subsea cable
was completed in 3 days. This quick installation was achieved by the simple mooring
configuration of this floating type power generation system.
 Development and Design of a Floating 745

Fig. 17. Installation beside Kuchinoshima Island

5 Study on the Structural Strength

Strength design of the floating body structure is very important to lead the demon-
stration test to a success without any troubles. To ensure this, we conducted a mock-up
test and finite element analysis before the more large-scale demonstration test, as we
have already touched upon in Sect. 2.4. In this chapter, we review the strength aspect
of the turbine floater. First, the results of the mock-up test is discussed in comparison to
the results of finite element analysis. Then, the strain measurement results of KAIRYU
in the demonstration tests are shown and discussed again comparing with the results of
finite element analysis.

5.1 Mock-Up Test of Structure for Water-Tightness

The pressure-resistant design of the shell structure to secure water-tightness in the deep
sea is very important. Water-tightness of the flange connections is also essential, while
several flange connections are inevitable for access man holes and other parts on the
shell structure to enable human access and installation and/or removal of equipment.
The mock-up model was fabricated to be tested for the strength and water-tightness as
shown in Fig. 18. The diameter of the tube, structural arrangement and flange
746 Y. Dodo et al.

3D-model of KAIRYUI

Pressure-resistant
Pressure-resistant Maintenance hatch dome
flange

Photo of mock-up Body

Drawing of mock-up

Fig. 18. Mock-up model of the shell structure

connections were determined to be as similar to the demonstration machine as possible.

Then, this mock-up model was put into the pressure test at the water depth of 100 m.

5.2 Results of Mock-up Test

The mock-up model was instrumented with strain gauges inside the shell, and was
submerged into the deep sea of about 100 m depth. Figure 19 shows the arrangement
of the strain gauges. Strains were measure continuously throughout the process of
changing the water depth of the mock-up, where the depth was changed as shown in
Fig. 20. Submersion trials were carried out twice in this procedure as shown in the
figure. The measured strain data were collected after the mock-up was uplifted
onboard. Figure 21 plots the measured strain at each sensor location during the first
submersion trial, taking the water depth in the abscissa.
 Development and Design of a Floating 747

0° Ch9 to16 Ch8 Ch7 Ch3,4

Ch9,10 Ch1

Ch5,6
Ch17(X),
270° Ch15,16 90° Ch18(Y)
Ch11,12 y
Ch19(Z)90°,
Ch20(Y) 90° z
y Ch13,14 Ch21(Z)270°,
x
Ch23(X) Ch22(Y) 270° Ch2
Ch24(Y)
x 180°
z

Fig. 19. The arrangement of stress sensors

First Try Second Try

10
0
-10
-20
-30
Depth [m]

-40
-50
-60
-70
-80
-90
-100
-110
-120
0:00 0:30 1:00 1:30 2:00 2:30 3:00

Time [h/min]

Fig. 20. Time chart of depth

We can observe in Fig. 21 that the measured strain changes in proportion to the
water depth. Figure 22 shows the stress contour and the displacement of the structure
obtained by finite element analysis conducted under the water pressure at the largest
depth of 113 m during the test. Comparison between the results of FEM and the actual
measurement is shown in Fig. 23 on the same condition of the water depth of 113 m.
The afloat condition (the condition where the mock-up is afloat still on the water) was
regarded as the initial condition and the value of the strain was calibrated to be zero in
the initial condition.
748 Y. Dodo et al.

400
CH001
300
CH002
200
CH003
100
CH004
0
Strain με

CH005
-100
CH006

-200
CH007

-300 CH008

-400

-500
-120 -100 -80 -60 -40 -20 0 20

Water depth (m)

200
CH009
100
CH010
0
CH011
-100
CH012
Strain με

-200
CH013
-300
CH014

-400
CH015

-500 CH016

-600

-700
-120 -100 -80 -60 -40 -20 0 20

Water depth (m)

300
CH017
200
CH018

100 CH019

0 CH020
Strain με

CH021
-100
CH022
-200
CH023
-300
CH024

-400

-500
-120 -100 -80 -60 -40 -20 0 20

Water depth (m)

Fig. 21. Relation between measured strain and water depth

 Development and Design of a Floating 749

Fig. 22. Stress contour and displacement (FEM)

800
FEM Center of Pressure-
600 Joint of hatch
Test resistant flange
400
Dome weld line
neighborhood General part
200 Top of dome
0
Plate thickness change part
Strain με

-200 (Circumferential direction)

-400

-600

-800 Plate thickness change part Cylindrical body of

(The axial direction) Pressure-resistant flange
-1000
[με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με] [με]
Ch 1 Ch 2 Ch 3 Ch 4 Ch 5 Ch 6 Ch 7 Ch 8 Ch 9 Ch 10 Ch 11 Ch 12 Ch 13 Ch 14 Ch 15 Ch 16 Ch 17 Ch 18 Ch 19 Ch 20 Ch 21 Ch 22 Ch 23 Ch 24
Ch. No.

Fig. 23. Comparison of strain between FEM and the actual measurement

We can observe from Fig. 23 that the measured strain agrees well with that of FEM
in way of the general part of the body and the hemispheric end part of the body.
However, agreement was relatively worse in way of the locations close to the welding
line of hatches and flanges or where the shell thickness changes. With regard to the
locations close to the welding line, the strain gauges were instrumented at the location
1.5 times plate thickness apart from the weld toe, while stresses from FEM were picked
750 Y. Dodo et al.

out at the location 1.5 times plate thickness apart from the intersection of the crossing
shell elements because the plate thickness and the weld bead are not modeled in FEM.
This may have caused the discrepancies in the strain between measurement and FEM,
especially when the plate thickness is large. In this sense, the results of FEM give
conservative results.

5.3 Strength Design of the Floating Body

In the previous section, it was confirmed that we can conservatively assess strength of
the floating structure using FEM. Then, we apply the identical method to the structural
design of the demonstration machine. Design load cases are indicated in Table 1. To
the conducted demonstration test, it is considered that the condition “Normal opera-
tion” applies. Therefore, comparison is made between the results of strain measurement
and the results of FEM assuming normal operation condition under the current velocity
of 1.5 m/s corresponding to the rated output and water depth of 50 m. An example of
the analysis results in normal operation is shown in Fig. 24.

Table 1. Load cases

No. Condition Location of the Loading condition
float
1 Start-up Surface of the Zero water pressure and
sea Full of ballast water
2 Normal Undersea Max. water pressure and
operation Max. thrust load of blade
(Electricity
generation)
3 One side Undersea Max. water pressure with one side turbine
failure blades lost
4 Stop Undersea Max. water pressure and
Zero thrust load of blade
5 Stand-by Surface of the Zero water pressure and
sea Zero thrust load of blade
6 Maintenance Land 1) Lifting load
or installation or on the barge 2) Lifting + wind load
3) Static load standing still on land
4) Heeled condition on the barge
5) Longitudinal inclination on the barge
 Development and Design of a Floating 751

Top of center pod

Cross beam

Bottom of center pod

Fig. 24. Stress contour of FEM (The red circle shows the strain measurement points)

5.4 Locations of Strain Measurement

As shown by the red circles in Fig. 24 and Fig. 25, the strain measurement was
conducted on the shell of the central pod and the cross beam inside the floating body.
FBG (Fiber Bragg Grating) sensors were applied to avoid adverse effect of electro-
magnetic noise in the measurement.

#2016

Fig. 25. Measurement location of the floating body (central pod and cross beam)
752 Y. Dodo et al.

5.5 Results of Strain Measurement

As to the magnitude of the measured strain, zero point is not clear, and therefore, only
the change in the magnitude should be evaluated. The primary factor exerting large
strain in the floating body is the static water pressure in case of this demonstration test.
Therefore, we calculated differences between the measured strains at two different water
depths, and compared them with the results of FEM under the water pressure equivalent
to this difference in water depth. Here we use the measured strains corresponding to two
water depths 40 m and 10 m in the mooring test. Time history of the floating body water
depth is indicated in Fig. 26. Measured strains in water depth of 40 m and water depth of
10 m are tabled in Table 2 in combination with the analysis results. The timings of the
two different water depths were chosen as close as possible as indicated in Fig. 26 to
exclude influences of the fluctuation of the conditions other than the water depth.
Figure 27 shows the relation between the floating body water depth and the
measured strain. We can observe linear relationship between the water depth and the
strain as a whole. Nonlinearity is observed when the water depth is small. This may be
attributed to the action of waves and floater motions.
Furthermore, strain gauges were also installed at some other locations including
structural discontinuity with large stress concentration. Among them, the maximum von
Mises stress was about 52 MPa at the water depth of 45 m in way of the strain gauge
#2016 located at the right end of Fig. 25, showing sufficient margin to the yield stress.

(Test.1)
Water depth (m)

Time

Water depth 40m equivalence Water depth 10m equivalence

10

(Test.2)
0
Water depth (m)

-10

-20

Water depth 10m equivalence

-30

-40
Water depth 40m equivalence
-50
Time

Fig. 26. The floating body depth chart as time change

 Development and Design of a Floating 753

Table 2. Measured and analyzed strains at each depth

(1) Water depth 40 m (unit: micro-strain)
Portion Test.1 Test.2 Analysis
Circumferential direction of bottom of center pod –200 –202 –132
(#2003)
Axial direction of bottom of center pod(#2004) –72 –78 17.6
Axial direction of top of center pod(#2002) –14 –16 –5.6
Top of beam structure(#2017) 28 18 81.3
Bottom of beam structure(#2018) 25 24 66.3
(2) Water depth 10 m (unit: micro-strain)
Portion Test.1 Test.2 Analysis
Circumferential direction of bottom of center pod –73 –80 –33
(#2003)
Axial direction of bottom of center pod(#2004) –60 –60 4.4
Axial direction of top of center pod(#2002) 4 5 –1.4
Top of beam structure(#2017) –21 –31 20.3
Bottom of beam structure(#2018) –25 –23 16.6

20

-20

-40

-60
Measured strain με

-80

-100

-120 Axial direction of top of center pod

Circumferential direction of bottom of center pod
-140
Axial direction of bottom of center pod
-160 Top of beam
Bottom of beam
-180
-40 -35 -30 -25 -20 -15 -10 -5 0
Water depth (m)

Fig. 27. Relation between the floating body water depth and the strain (test 1)
754 Y. Dodo et al.

5.6 Comparison Between Strain Measurement and Analysis

Figure 28 shows the comparison between the measured strain and the results of FEM.
The vertical axis shows the difference of the strains between the water depth of 40 m
and the water depth of 10 m, corresponding to the water head of 30 m at the center of
the generator pods. The FE analysis was also carried out under the static pressure of
30 m head at the center of the generator pods, in combination with the other loads such
as thrust forces at 60% of the rated output.
From this figure, we can observe that the measured strains agree quite well with the
results of the FEM. Regarding the axial stress of the center pod, although the absolute
difference in the strains is not so significant between the measurement and the analysis,
the ratio of the strains between them is far from 1.0. At this location, uniform axial
compressive strain and local tensile strain on the inner surface of the shell plate due to
the bending under outer pressure cancel each other, and the resulting strain is very
small. This may have caused this relatively larger error at this location.
In this chapter, we carried out comparison between the strains measured in the
mock-up test as well as the demonstration mooring test and the strains obtained by the
FE analysis. As a result, good agreement between them was observed, and we could
confirm the validity of the design method.

100

50

FEM
Test1
0
Strain(με)

Test2

-50

-100

-150

Portion
Fig. 28. Comparison of analysis results and measured values
 Development and Design of a Floating 755

6 Conclusion

We designed and constructed the floating type ocean current turbine system
“KAIRYU”, and conducted a world’s first 100 kW class demonstration test in the
actual Kuroshio Current site. As a result, the following concluding remarks are made.
• It was confirmed that KAIRYU can be controlled stably in the natural Kuroshio
Current with regard to required depth and attitude.
• Actual power output of KAIRYU in the towing test and in the ocean current was in
good agreement with the designed power generation.
• Strain on the structure was measured in the mock-up in the static pressure condition
as well as KAIRYU in the operating condition, and was compared with the results
of FEM. Good agreement was observed in general, and it was confirmed that the
strength assessment method is adequate to be used for further development of a
larger practical turbine floater.

References
1. Chen, F.: Kuroshio power plant development plan. Renew. Sust. Energ. Rev. 14, 2655–2668
(2010)
2. Finkl, C.W., Charlier, R.: Electrical power generation from ocean currents in the Straits of
Florida: Some environmental considerations. Renew. Sust. Energ. Rev. 13, 2597–2604
(2009)
3. Shirasawa, K., Tokunaga, K., Iwashita, H., Shintake, T.: Experimental verification of a
floating ocean-current turbine with a single rotor for use in Kuroshio currents. Renew. Energ.
91, 189–195 (2016)
4. Lo, H.Y., Chen, J.H., Hsin, C.Y., Chiu, F.C.: Dynamics of floating Kuroshio turbine system
in deployment, 29th Asian-Pacific Technical Exchange and Advisory Meeting on Marine
Structures (TEAM 2015), pp. 169–176. Vladivostok, Russia (2015)
5. Lai, S.A., Hsin, C.Y., Lee, C.F., Lin, T.Y., Chen, J.H.: The floating Kuroshio turbine blades
geometry design with consideration of the structural strength, 31st Asian-Pacific Technical
Exchange and Advisory Meeting on Marine Structures (TEAM 2017), pp. 671–677. Osaka,
Japan (2017)
Author Index

A Dotta, Raul, 91
Amarante, Rodrigo M., 91 Duz, Bulent, 657
Aoyama, Kazuhiro, 321
Arai, Daisuke, 209 E
Arifin, Mohammad Danil, 287 Egorov, Alexander, 120
Egorov, Gennadiy, 106, 120
B Elenbaas, Marcel, 51
Bäck, Thomas, 38 Evegren, Franz, 51
Berg-Jensen, Jan Henrik, 220
Birmingham, Richard W., 436 F
Bordalo, Sergio Nascimento, 595 Furustam, Jan, 38
Byun, Seok-Ho, 80
G
C
Gotoh, Koji, 243, 258
Cai, Wei, 137, 498
Grasso, Nicola, 657
Calisal, Sander, 173
Grudniewski, Przemyslaw, 156
Chai, Wei, 397
Chen, Ying, 579
H
Chen, Zhan, 137
Hageman, Remco, 657
Cheng, Xiaoming, 551, 579
Ham, Seung-Ho, 237
Chernikov, Dmitriy, 106
Hamada, Kunihiro, 287
Choung, Joonmo, 373
Han, Donghwa, 373, 383
Chun, Do-Hyun, 237
Hayashi, Kiyoharu, 273
Colbourne, Bruce, 409
Heo, Jaekyung, 220
Coller, James A., 3
Hirata, Noritaka, 287
Collette, Matthew, 24
Hou, Chunxiao, 533
D
Daley, Claude, 409 I
de Bruijn, Arnold, 51 Ichinose, Yasuo, 209
de Winter, Roy, 38 Ihara, Kai, 287
Dev, Arun, 679 Ilnitskiy, Igor, 120
Díaz-Casas, V., 697 Imasaki, Shogo, 273
Dodo, Yasushi, 732 Ito, Akio, 732

© Springer Nature Singapore Pte Ltd. 2021

T. Okada et al. (Eds.): PRADS 2019, LNCE 65, pp. 757–759, 2021.
https://doi.org/10.1007/978-981-15-4680-8
758 Author Index

J Murphy, Alan J., 338, 436, 462

Jang, Beom-Seon, 614 Muzhoffar, Dimas Angga Fakhri, 287
Jiang, Yuyan, 349
Josten, Michal, 423 N
Nagaya, Shigeki, 732
K Narimatsu, Hisao, 243, 258
Kam, Yu K., 91 Ni, Xinyun, 579
Kaneko, Azumi, 209 Nilva, Alexander, 106
Kang, Joong Kyoo, 363 Nishimoto, Kazuo, 478
Kang, Kuk-Jin, 373 Novás-Cortés, A., 697
Katayama, Yasuo, 273
Kil, Woo-Sung, 80 O
Kim, Hyun-Soo, 383 Oka, Masayoshi, 308
Kim, Jeong Du, 614 Okada, Tetsuo, 732
Kim, Ki-Su, 190 Osawa, Naoki, 273
Kim, Sung-Pyo, 363 Ozden, Ali Erinc, 383
Kim, Yonghwan, 564
Krause, Matthias, 51 P
Krüger, Stefan, 423 Paboeuf, Stéphane, 51
Kurniawati, Vita R., 436 Pan, Zhiyuan, 220
Kuwada, Masakazu, 243 Park, Daehwan, 220
Park, Ji-In, 363
L Park, Mi-Ho, 80
Lee, Hye-Won, 237 Pazouki, Kayvan, 338
Lee, Jeong-Youl, 80 Prampakov, Velizar, 679
Lee, Sahng-Hyon, 363
Lee, Seung-Hoon, 564 R
Leira, Bernt J., 397 R. Gutiérrez, Ricardo H., 631
Lim, Dong-Hyun, 564 Rashed, Sherif, 273
Lim, Ho-Jeong, 363 Roh, Myung-Il, 190, 237
Lim, Serena, 338 Ruggeri, Felipe, 478
Liu, Chao, 498
Liu, Guanlan, 462 S
Liu, Jun, 639 Sampaio, Claudio M. P., 91, 478
Liu, Wenzhang, 198 Sano, Masanori, 273
Santiago-Caamaño, L., 697
M Savasta, Thomas, 156
Ma, Chong, 308 Sazidy, Mahmud, 409
MacKay, John, 409 Shi, Wei, 397
Maeda, Toshimitsu, 258 Shoji, Ruri, 516
Mak, Bart, 657 Singer, David J., 3, 64
Manohar, Rahul, 679 Sinsabvarodom, Chana, 397
Masui, Takayoshi, 209 Sobey, Adam, 156
Miguez-González, M., 697 Son, Myeong-Jo, 80
Minami, Makiko, 516 Souflis, Leonidas, 423
Minette, Ricardo S., 631 Sun, Zhe, 349
Mochiduki, Ryohei, 707
Monteiro, Ulisses A., 631 T
Moriyama, Atsuo, 258 Tahara, Yusuke, 209
Morooka, Celso Kazuyuki, 595 Takami, Tomoki, 209, 308
Muller, Thijs, 38 Tan, Xinjing, 498
Murai, Motohiko, 707 Tanabe, Yoshikazu, 321
Murakami, Mikikazu, 273 Tanaka, Yoshiki, 321
Murakawa, Hidekazu, 273 Tang, Wenyong, 639
Author Index 759

Tei, Alessio, 462 Wu, Wenwei, 198

Torigoe, Takumi, 258 Wu, Xiaofeng, 551
Toyoda, Makoto, 732 Wu, Yigang, 137
Toyoshima, Junichi, 273
X
U Xu, Jinsong, 533
Uchino, Issei, 258 Xue, Hongxiang, 639
Uemura, Takamori, 258
Y
V Yang, Biye, 349
Vada, Torgeir Kirkhorn, 220 Yang, Rongwu, 533
Vale, Rodrigo J., 478 Yin, Xuewen, 198
Vaz, Luiz A., 631 Yotsuzuka, Takayuki, 321
Vieira, Daniel P., 91 Yu, Jun, 551
Yuan, Hao, 64
W Yuan, Yuchao, 639
Wang, Ge (George), 533
Wang, Xin, 679 Z
Wang, Yazhong, 137 Zago, Lariuss, 91
Wincott, Claire, 24 Zhang, Guiyong, 349
Wu, Bo, 579 Zhang, Kai, 579