ISTANBUL TECHNICAL UNIVERSITY GRADUATE SCHOOL OF SCIENCE ENGINEERING AND TECHNOLOGY
Ph.D. THESIS
JULY 2019
A DYNAMIC RISK ASSESSMENT METHODOLOGY (Dy-RAM) IN PORT WATERS
Ülkü ÖZTÜRK
Maritime Transportation Engineering Department
JULY 2019
ISTANBUL TECHNICAL UNIVERSITY GRADUATE SCHOOL OF SCIENCE ENGINEERING AND TECHNOLOGY
Ph.D. THESIS Ülkü ÖZTÜRK
(512142011)
A DYNAMIC RISK ASSESSMENT METHODOLOGY (Dy-RAM) IN PORT WATERS
Maritime Transportation Engineering Department
Maritime Transportation Engineering Programme
,
Deniz Ulaştırma Mühendisliği Anabilim Dalı
Deniz Ulaştırma Mühendisliği Programı
TEMMUZ 2019
İSTANBUL TEKNİK ÜNİVERSİTESİ FEN BİLİMLERİ ENSTİTÜSÜ
LİMAN SULARINDA DİNAMİK RİSK DEĞERLENDİRME (Dy-RAM) METODOLOJİSİ
DOKTORA TEZİ Ülkü ÖZTÜRK
(512142011)
Tez Danışmanı: Dr. Öğr. Üyesi Kadir ÇİÇEK
Jury Members : Prof. Dr. Metin ÇELİK ... Istanbul Technical University
Prof. Dr. Özcan ARSLAN ... Istanbul Technical University
Assist. Prof. İlker AKGÜN ... Marmara University
Ülkü ÖZTÜRK, a Ph.D. student of İTU Graduate School of Science Engineering and Technology student ID 512142011, successfully defended the thesis/dissertation entitled “A Dynamic Risk Assessment Methodology (Dy-RAM) In Port Waters”, which he prepared after fulfilling the requirements specified in the associated legislations, before the jury whose signatures are below.
Date of Submission : 29 May 2019 Date of Defense : 10 July 2019
Thesis Advisor : Assist. Prof. Dr. Kadir ÇİÇEK
Istanbul Technical University
Prof. Dr. Ş. İlker BİRBİL ... Erasmus University Rotterdam
This doctoral thesis can not be produced without the great support of my dear wife Ayça ÖZTÜRK ,
FOREWORD
I have been a member of Turkish Naval Forces since 2000 and a Navy Officer since 2008. The maritime dimension of the world is very familiar to me as a navy officer. However, detail maritime researches require further efforts. Because a navy officer always busy about his or her missions and assignments, unfortunately there is not so much time to perform this kind of further efforts. I want to thanks to the Turkish Naval Command, that allowed me to perform an academic research in Istanbul Technical University for two years. After this two year of academic studies, I took my research further. Maritime problems related with data analysis and machine learning takes my attention for five years. Therefore, I tried to solve a specific problem regarding the collision risk assessment in my doctoral study. However, this problem turned into a bit challenging work. These challenges have been unveiled thanks to my advisor. I want to thanks to my advisor Associate Professor Kadir ÇİCEK who support me in all time of my academic research. Another prominent supporter of my doctoral study is Proffesor İlker BİRBİL. I am very thanksful for his guidance. I also thanks to Professor Metin ÇELİK and teacher Oğuzhan GÜREL for their peerless support. In addition, the contribution of anonymous maritime pilots is very valuable for this study.
This doctoral thesis focuses on the navigational collision risk assessment of the vessels navigating in port waters. Ports are getting more and more complex as the maritime traffic increases. As we already know, the movement of goods is dependent on the shipping. Therefore, the maritime traffic will continue to increase because of the world financial dependencies until the safer, faster and economical way of transportation emerges.
Ports are the gateways of the maritime transporation. These gateways should be safe for navigation. Although there are lots of risk assessment methodology for open sea or restricted waterway navigation, there is no risk assessment methodology specialised for port navigation. Since there is no risk assessment methodology about this subject, I think that the proposed thesis will fill a significant gap in the maritime domain. There are two perspectiveS to consider the contribution of this doctoral thesis. The first one is the port management perspective. The port management, striving for the safe navigation in port waters in the limits of the international standards, can use this risk assessment methodology to assure that the port basin is safe for intended type of vessel. The second one is the shipping company perspective. The shipping company can monitor the risk degree of its vessels in port basins. Furthermore, the shipping company can evaluate the berthing/unberthing ability of its officers with this proposed risk assessment methodology.
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. I also wish to thank the anonymous experts for their time and considerations.
July 2019 Ülkü ÖZTÜRK
TABLE OF CONTENTS Page FOREWORD ... ix TABLE OF CONTENTS ... xi ABBREVIATIONS ... xiii SYMBOLS ... xv
LIST OF TABLES ... xvii
LIST OF FIGURES ... xix
SUMMARY ... xxi
ÖZET ... xxv
INTRODUCTION ... 1
LITERATURE REVIEW ... 7
INITIAL RESEARCH STUDIES ... 25
RESEARCH METHODOLOGY ... 41
APPLICATION ... 53
RESULT AND DISCUSSION ... 57
CONCLUSION AND RECOMMENDATIONS ... 61
REFERENCES ... 63
APPENDICES ... 69
ABBREVIATIONS
IMO : Internation Maritime Organization
COLREG : Convention of The International Regulations For Preventing
Collisions At Sea
ISO : International Standardization Organization
NCR : Navigational Collision Risk
S3VM : Semi Supervised Support Vector Machines
ARPA : Automatic Radar Plotting Aids
ECDIS : Electronic Chart Display and Information System
GPS : Global Positioning System
JMS : Japan Maritime Science
SYMBOLS
𝐅𝐓𝐖(𝒕) : Transverse wind forces in time t
𝐅𝐋𝐖(𝒕) : Longitudinal wind forces in time t
𝑭𝑾𝒊𝒏𝒅 : Total wind force affecting the ship
𝛒𝐀 : Density of the air in kg/m3
𝐀𝐋 : Longitudinal projected area of the vessel above the waterline 𝐕𝐖(𝒕) : Design wind speed (m/sec) in time t
𝐂𝐓𝐖 : Lransverse wind force coefficients for the aft and the bow 𝐂𝐋𝐖 : Longitudinal wind force coefficients for the aft and the bow
LIST OF TABLES
Page
Table 2.1 : Boundary of the research according to the characteristics . ... 10
Table 2.2 : Review protocol. ... 10
Table 2.3 : Search strings by databases. ... 10
Table 2.4 : Some parameters taken into account in NCR studies. ... 18
Table 2.5 : Occurrence of NCR parameters. ... 20
Table 2.6 : Aims and measurement models of the examined NCR assessment studies. ... 22
Table 3.1 : The details of the scenarios. ... 36
Table 3.2 : Descriptive statistics of raw data set. ... 39
Table 4.1 : An example NCR fuzzy rule table ... 43
Table 4.2 : Fuzzy inference table obtained from the survey ... 44
Table 4.3 : Comparison of the new rules against the rules obtained from the survey. ... 46
LIST OF FIGURES
Page
Figure 1.1 : Structure of the thesis. ... 5
Figure 2.1 : Systematic literature review framework. ... 7
Figure 2.2 : Flowchart of the review protocol. ... 8
Figure 2.3 : Number of selected papers organized by years of publication. ... 11
Figure 2.4 : Number of publications by journals. ... 11
Figure 2.5 : Distribution of the studies by risk approaches. ... 12
Figure 3.1 : Turning circle in port basin. ... 27
Figure 3.2 : Safe area of a ship within port region. ... 28
Figure 3.3 : Limited area of a ship within port region. ... 28
Figure 3.4 : Illustration of the area parameter. ... 29
Figure 3.5 : Wind force coefficients for (a) container, (b) VLCC, and (c) tanker and dry cargo ships... 30
Figure 3.6 : Fitted wind coefficient curves for container ships. ... 31
Figure 3.7 : Fitted wind coefficient curves for VLCC. ... 32
Figure 3.8 : Fitted wind coefficient curves for dry cargo and tankers. ... 33
Figure 3.9 : The longitudinal projected area of (a) the tankers and (b) the container ships... 34
Figure 3.10 : Fitted longitudinal projected area curves for container ships.. ... 34
Figure 3.11 : Fitted longitudinal projected area curves for tanker ships. (BSI, 2016). ... 34
Figure 3.12 : The number of scenarios by the ship tonnage. ... 36
Figure 3.13 : Boxplots of distance and wind parameters. ... 36
Figure 3.14 : The updated boxplots of the distance and the wind parameters. ... 37
Figure 3.15 : The correlation plot of the parameters... 38
Figure 3.16 : The histograms of the parameters... 38
Figure 4.1 : The general framework of the proposed NCR methodology. ... 42
Figure 4.2 : Fuzzy membership functions. ... 43
Figure 4.3 : Distribution of expert’s choices by each rule. ... 45
Figure 4.4 : The scatter plot of S3VM results ... 47
Figure 4.5 : The correlation plot of raw data ... 48
Figure 4.6 : Error rates of the random forest model... 50
Figure 4.7 : Error rates of the random forest model with the noisy dataset. ... 50
Figure 4.8 : Error rates of the neural network model. ... 51
Figure 4.9 : Error rates of the gradient boosting model. ... 51
Figure 4.10 : Importance of the parameters. ... 52
Figure 5.1 : Scenario application of the proposed model. ... 53
Figure 5.2 : Indicators of the scenario application. ... 54
Figure 5.3 : Proposed parameters of the scenario application. ... 55
Figure 6.1 : Partial plots of the input parameters relating to the “Very Low” risk class. ... 58
Figure 6.2 : Partial plots of the input parameters relating to the “Very High” risk class. ... 59
Figure A.1 : Partial plots of the input parameters relating to the “Low” risk class. . 70 Figure A.2 : Partial plots of the input parameters relating to the “moderate” risk
class. ... 70
A DYNAMIC RISK ASSESSMENT METHODOLOGY (Dy-RAM) IN PORT WATERS
SUMMARY
The underpinning motivation of researching complex maritime environment includes not only never-ending demands of maritime navigation but also emergency of safety of navigation. Navigational safety is one of the main concerns in the maritime industry. While merchant vessels become bigger and waterways become congested, marine officers still have the overall responsibility with his/her experience and judgment. Some maritime regulations such as COLREG and IMO recommendations introduce useful, practical and reliable methods to provide navigational safety. However, these regulations don’t have concrete boundaries about safety limits. For example, COLREG Rule 7 describes collision risk insufficiently; such risk shall be deemed to exist if the compass bearing of an approaching vessel does not appreciably change. COLREG also relies on good seamanship in the case of collision avoidance, which is recommended to act in ample time. Consequently, it seems that maritime accidents can’t be prevented satisfactorily by regulation, navigational aids and education. Both the standardization efforts of IMO and the decision making/supporting systems base on the risk assessment process, eventually. The collision risk assessment is the fundamental pillar of these supporting systems, and it is a basic and significant concept in autonomous ship navigation. Therefore, an effective systematic model to assess collision risk continually by monitoring parameter states is necessary for both the management and safety capabilities. This necessity can be more tangible as the data become bigger.
Number of scientific studies were proposed in order to complement the shortcoming of maritime rules by introducing navigational collision risk assessment approaches. Each study has its advantages and disadvantages. However, due to feature parameter numerousness, inherently complex marine environment, uncertainty and lack of information, there is no common way to define collision risk. Furthermore, no references have been made to the foundation of risk analysis in the maritime application area. The selection of appropriate and consistent collision risk assessment methods commonly depends on each unique situation. These studies have particularly paid attention to two major maritime areas; open sea and restricted or congested waterways. As yet, study considering port approach manoeuvring including berthing and unberthing haven’t been presented. In this regard, this study aims to define a structural collision risk assessment framework concerning port approach manoeuvring including berthing and unberthing scientifically. Furthermore, this proposed navigational collision risk assessment is a moderate realist approach of the maritime transportation risk analysis. A detailed literaure review about navigational collision risk assessments has been conducted for that purpose. Although there are review articles of development of collision avoidance and ship safety domains studies,
evolution of individual navigational collision risk assessment studies has not been introduced yet. This attempt has revealed the shortcomings of the current level of navigational collision risk assessment studies. In addition, literature gap was highlighted precisely in order to elicit roadmap for proposed doctoral thesis. On the other hand, potential parameters for the evaluation of collision risk in port approach manoeuvring were listed in accordance with literature review. The way how environmental conditions (wind, current and wave force) can be employed was presented. As these parameters have been employed rarely and basically in the literature, proposed doctoral study aims to establish environmental forces on a ship in accordance with international standards. This doctoral thesis proposes three new parameters specialsed for port approach manoeuvrings to assess navigational collision risk. These new parameters, distance, wind and area, introduce a practical way of monitoring navigational environement within port basins. In addition to these three new parameters, speed of ownship is the fourth parameter employed in the navigational collision risk assessment methodology. The navigational risk assessment methodology uses these parameters to label the observations, however the log parameters of the simulations are used in the training process.
In the research phase of this doctoral thesis 140 port approach (berthing/unberthing) manoeuvring scenarios has been conducted in a JMS (Japan Maritime Science) Ship Handling Simulator. JMS Ship Handling Simulator has all navigation aids (ARPA radar, ECDIS, rudder, tugboat, communication equipment, GPS, alarm panels, and so on) that a real merchant ship has. These scenarios have been conducted in a way of which it can embrace all possible scenario within port approach manoeuvrings. Then, log data has been extracted in R environment in order to preprocessing/examining the dataset. This approach has introduced a remarkable dataset within collision risk assessment methodology. Furthermore, examining the dataset reveals the structure of parameters unique to port approach manoeuvring. Data examining steps includes detecting missing values, deleting outliers, standardization. This dataset can be available also for further researches.
At last, in order to determine the technique, which will be employed in the study, potential techniques in this area was introduced. It was observed that wide variety of the moldes exist from linear to nonlinear. However, the most important part of the risk assessment is the determining the labels of the intended model. While the navigational risk assessment studies in the literature determine the labels of the models in variaous way, the labels of each observation have been determined by a methodology including fuzzy and semi-supervised learning approach in this proposed doctoral thesis. The fuzzy clustering method separates the parameter to be easily defined with linguistic variables. One of the basic ways of determining the lables for training is constructing a fuzzy rule table with expert knowledge. However, constructing all rules from the expert knowledge can be insurmountable since there are 1440 rules combination in this study. Therefore, only twenty rules have been selected for the survey. Prepared survey has been filled by 20 experienced maritime pilots. All pilots have minimum 10 years of experience and twelve of them have more than 20 years of experience. Observations without the labels are labelled with the semi-supervised support vector machines (S3VM). The labels obtained by S3VM along with the raw data of the simulation results are used for training a random forest to obtain the NCR degree. On the other hand, two statistical models, neural network and gradient boosting, have also been applied to the same dataset with parameter tuning. Random forest has a good performance also with the noisy dataset. which shows that random forest is robust.
The proposed NCR methodology incorporates parameters concerning to the port approach manoeuvring and, evaluates how these parameters related with navigational collision risk. Unlike other approaches, employing navigation specific paramters such as DCPA and TCPA, it deals with close distances with the help of parameters relevant to collision risk in port approach manoeuvring. The results of the model show that the proposed distance, area and wind parameters are proper for the NCR assessment in port approach manoeuvrings. Furthermore, it has been observed that the result of the clustering of the speed parameter is in parallel with the ISO standard. These parameters can give useful and practical perception about the navigation safety. Additionally, employing the log data of the simulations instead of the proposed parameters can increase the applicability of the model, and be used for operational purposes in the real life. This may contribute to the navigational safety efforts in port basins eventually. One of the speed vectors has the highest relation with the NCR according to the random forest parameter importance result. However, it is important to find out that wind parameter has also high impact on NCR emphasising the significance of the hydro-meteorological data. The partial plots of the parameters corresponding risk classes show that each parameter has different effect on determining the risk classes. These partial effects can also be incorporated for evaluating the navigational safety within port basins.
Since the shipping traffic has increased, the number of the port approach manoeuvrings has also increased. Therefore, proposed model can contribute to the monitoring the navigation safety in port basins. Navigators can determine the preventive action with respect to the severity levels accompanying with the parameter significance. Furthermore, the significance of the parameters can be employed in prioritizing the preventive actions.
LİMAN SULARINDA RİSK DEĞERLENDİRME (Dy-RAM) METODOLOJİSİ
ÖZET
Denizdeki acil ihtiyaçlar ve emniyet hususunun yüksek öneme haiz olması, denizde güvenli bir şekilde seyir yapmanın gereksinimlerini oluşturmakta ve by yöndeki çalışmaları teşvik etmektedir. Denizde seyir güvenliği ve emniyeti, denizcilik endüstrisindeki ana kaygılardan ve üzerinde çalışılmakta olan konulardan biridir. Dünya, nüfus ve ticaret artışına paralel olarak ticari gemilerin boyutları gittikçe artmakta ve su yolları buna bağlı olarak yoğunlaşmaktadır, ancak buna rağmen gemi zabitanları seyir güvenliğine ilişkin tek sorumluluğa sahip bireyler olmaya devam etmektedirler. Bu sorumluluğun üstesinden gelmelerine yardımcı olan en temel araç ise elde ettikleri deneyimlerdir. IMO standardizasyon çalışmaları; COLREG, seyir emniyetini ve güvenliğini sağlamak için kullanışlı, pratik ve güvenilir yöntemler sunmaktadır. Ancak, COLREG Kural 7, çarpışma riskini yetersiz bir şekilde tanımlamıştır; yaklaşan bir geminin nisbi kerterizinin kayda değer bir şekilde değişmemesi durumunda riskin var olduğu kabul edilmektedir. COLREG, çatışamadan kaçınma durumlarına yönelik sade ve ayrıntılı tavsiyeler vermemekle birlikte daha çok iyi denizcilik uygulamalarına atıf yapmaktadır. Sonuç olarak, köprü üstündeki seyir yardımlarıcıları ve zorunlu denizcilik eğitim/sertifikasyonu deniz kazalarını tatmin edici bir şekilde önleyememiştir.
IMO'nun standardizasyon çabaları gibi diğer karar destek sistemlerinin temel dayanaklarından bir tanesi risk değerlendirme sürecidir. Bu nedenle, seyir çarpışma riski değerlendirmesi bu tip destek sistemlerinin temel dayanaklarından bir tanesidir. Ayrıca, seyir çarpışma riski değerlendirmesi son zamanlarda deniz otonom seyri kapsamında sıklıkla kullanılan önemli bir kavramdır. Bu nedenle, seyir çarpışma riskini sürekli olarak değişkenlerin durumlarını izleyerek değerlendirmek için etkili bir sistematik model gereklidir. Hem seyir yönetim hem de seyir güvenlik ve emniyet yeteneklerinin arttırılması için bu özelliklere sahip bir modele ihtiyaç duyulabilir. Bu gereklilik, gemi seyrine ilişikin veriler arttıkça daha da somut olabilir.
Denizcilik düzenlemelerinin seyir çarpışma riskinin açık bir şekilde tanımlamasının eksikliği bir çok bilimsel çalışma ile giderilmeye çalışılmıştır. Herbir bilimsel çalışmanın avantajları ve dezavantajları vardır. Bununla birlikte, her bir deniz alanının (açık deniz, dar kanallar, liman sular vb.) farklı özellikleri olduğundan dolayı seyir çarpışma riskini bulmanın tek bir yolu yoktur. Diğer yandan, bilimsel anlamda deniz seyir çarpışma riskine yönelik belirli ve ortak kabul gören bir risk tanımlaması da yapılmamıştır. Uygun ve tutarlı bir çarpışma riski değerlendirme yönteminin seçimi, genellikle her bir deniz alanı için değişebilir. Bu alandaki bilimsel çalışmalar bugüne kadar özellikle iki ana deniz alanı ile ilgilenmiştir; açık deniz ve kısıtlı ya da yoğun su yolları. Fakat, liman bölgesinde aborda ve avara manevralarını da içeren bir seyir çarpışma risk değerlendirmesi çalışması henüz yapılmamıştır. Bu kapsamda, bu
doktora çalışması, liman sularında avara ve aborda manevraları dahil olmak üzere bir seyir çarpışma riski değerlendirme metodolojisi ortaya koymayı hedeflemektedir. Ayrıca, bu önerilen seyir çarpışma riski değerlendirmesi ılımlı ve gerçekçi bir risk değerlendirme yaklaşımdır. Ilımlı ve gerçekçi risk değerlendirmesi ise kısmen uzman görüşlerine dayanan bir risk değerlendirme metodolojisini ifade etmektedir. Bu amaç için seyir çarpışma riski değerlendirmeleri hakkında detaylı bir yazın taraması yapılmıştır. Seyir risk değerlendirmesi kapsamında geminin güvenli alanı ve çarpışmadan kaçınma manevralarına ilişkin yazın taraması çalışmaları olmasına rağmen konumuz olan bireysel seyir risk değerlendirme çalışmalarına odaklanan bir yazın çalışması henüz yapılmamıştır. Bu doktora tezinde yazın açığının bulunması maksadıyla ayrıntılı ve sistematik bir yazın taraması yapılmıştır. Öte yandan, liman yaklaşma manevralarının risk değerlendirmesinin yapılabilmesi maksadıyla mevcut diğer seyir risk değerlendirme çalışmalarında kullanılan değişkenler incelenmiştir. Diğer çalışmalarda nadiren kullanılan ve bu sebeple bazı çalışmalar tarafından eleştirilen çevresel koşulların (rüzgar, akım ve dalga kuvveti) nasıl kullanılabileceği de gösterilmiştir. Bu değişkenler önceli çalışmalarda nadiren ve basit bir şekilde kullanıldığı için bu doktora çalışması, çevresel koşulları uluslararası standartlara uygun olarak ortaya koymuştur. Bu doktora tezi, seyirsel çarpışma riskini değerlendirmek için liman yaklaşımı manevraları için tasarlanmış üç yeni değişken önermektedir. Bu yeni değişkenler, mesafe, rüzgar ve alandır. Bu parameterler liman havzalarındaki seyir risk durumunu izlemek için pratik bir yol sunmaktadır. Bu üç yeni değişkene ek olarak, gemi hızı seyir çarpışma riski değerlendirme metodolojisinde kullanılan dördüncü değişkendir. Seyir risk değerlendirme metodolojisinde risk etiketlerinin belirlenmesi için önerilen bu değişkenler kullanılmakla birlikte modeli eğitmek için gemi üzerindeki algılayıcılardan elde edilebilecek ham veriler kullanılmaktadır.
JMS (Japan Maritime Science) Gemi Köprü Üstü Simülatöründe toplam 140 liman yaklaşma (yanaşma / ayrılma) manevra senaryosu uygulanmıştır. Tüm senaryolarının yapıldığı Gemi Komuta Simülatöründa gerçek bir ticari geminin sahip olduğu tüm navigasyon cihazları (ARPA radarı, ECDIS, dümen, römorkör, haberleşme teçhizatı, GPS, alarm panelleri vb.) bulunmaktadır. Bu senaryolar, liman yaklaşımı manevralarında olabilecek olası tüm durumları kapsayacak şekilde gerçekleştirilmeye çalışılmıştır. Bu senaryolardan elde edilen ham veriler R programlama diline aktarılarak gözden geçirilmiş ve istatksel analize uygun hale getirilmiştir. Elde edilen veri kümesi liman yaklaşma manevralarının incelenmesi ve başta risk değerlendirmesi olmak üzere daha bir çok analizin yapılabileceği benzersiz bir imkanın ortaya çıkmasını sağlamıştır.
Seyir risk değerlendirmesi için bugüne kadar uygulanan teknikler incelenmiş olup doğrusal ve doğrusal olmayan çeşitli yöntemlerin olduğu tespit edilmiştir. Fakat, risk değerlendirme sürecinde en önemli adımın değişkenlerin ve etiketlerin belirlenmesi olduğu ortaya çıkmıştır. Yazındaki çalışmalar risk etiketlerini bir çok farklı yöntemle elde etmektedirler. Bu doktora tezinde, her bir gözlemin etiketleri, bulanık ve yarı denetimli öğrenme yaklaşımını içeren bir metodoloji ile belirlenmiştir. Bulanık kümeleme yöntemi sayesinde değişkenler sınıflandırılarak sözlü olarak ifade edilme imkanı yaratılmıştır. Risk etiketlerinin belirlenmesinde en temel yöntemlerden bir tanesi uzman bilgisine başvurmaktır. Bununla birlikte, bu çalışmada 1440 kural birleşimi olduğu için uzman bilgisinden tüm kuralları oluşturmak zor olabilir. Bu nedenle, uzman bilgisinin çalışmaya aktarılamsı maksadıyla oluşturulan ankette sadece yirmi kural belirlenmiştir. Hazırlanan anket 20 deneyimli kılavuz tarafından
doldurulmuştur. Tüm kılavuzlar en az 10 yıllık deneyime sahip olmakla birlikte bunların on iki tanesi en az 20 yıllık deneyime sahiptir. Geriye kalan etiketsiz gözlemler ise, yarı denetimli destek vektör makineleri (S3VM) ile etiketlenmiştir. S3VM ile elde edilen etiketler simülasyona ait ham veriler ile birleştirilmiş ve rastgele orman modeli ile eğitilmiştir. Öte yandan, başka iki istatistiksel model, yapay sinir ağı ve gradyan artırma, aynı veri kümesi kullanılarak eğitilmiştir. Rastgele orman modelinin şaşıtıcı doğruluk oranına sahip olmasından dolayı veriye gürültü eklenerek tekrar yeni bir eğitim veri kümesi oluşturulmuştur. Model, gürültülü veri kümesi ile de iyi bir performans göstermiştir. Modelin sonuçları, mesafe, alan ve rüzgar değişkenlerinin liman yanaşma manevralarında seyir risk değerlendirmesi için uygun olduğunu göstermektedir. Ayrıca, hız değişkeninin kümelenmesi sonucunun ISO standardına paralel olduğu görülmüştür. Bu değişkenler seyir güvenliği hakkında faydalı ve pratik bilgiler verebilir. Metodoloji her ne kadar konu ile ilgili özelleştirilmiş değişkenler ile kurulmuş olsa da algoritmanın eğitimi ham verilerle yapılmış ve gayet iyi sonuçlar alınmıştır. Önerilen değişkenler yerine simülasyonların ham verilerinin kullanılması, modelin uygulanabilirliğini artırabilir ve gerçek hayatta operasyonel amaçlar için de kullanılabilirliğini arttırabilir. Bu durum sonuçta liman havzalarında seyir güvenliği çabalarına da katkıda bulunabilir.
Gemi hız vektörlerinden birinin risk derecesi ile en yüksek ilişkiye sahip olduğu beklenen bir sonuçtur. Fakat, rüzgar hızının ikinci en yüksek ilişkiye sahip değişken olarak bulunması, seyir çarpışama riski üzerinde hidro-meteorolojik verilerin önemini vurgulaması açısından önemlidir. Her bir risk sınıfının belirlenmesindeki değişkenlerin kısmi etkilerini gösteren kısmi grafikler çıkarılmış olup değişkenlerin liman içerisinde risk üzerindeki etkileri incelenmiştir. Ortaya konan seyir risk değerlendirme metodolojisinin uygulanan örnek senaryo üzerinden de değerlendirilerek liman yaklaşma manevralarında gayet başarılı bir şekilde çalıştığı görülmüştür.
Sonuç olarak, Küresel anlamda deniz trafiği artmakta ve buna bağlı olarak liman manevralarının sayısı da artmaktadır. Bu nedenle, önerilen risk değerlendirme modeli sadece seyir çarpışma risk değerlendirmesi için değil liman havzalarındaki seyir trafiğinin güvenliğinin izlenmesine de katkıda bulunabilir. Ayrıca, seyir çarpışma riskinin seviyelerine göre önleyici eylemlerin belirlenmesine de parametrelerin önceliklendirmesi aracılığıyla katkıda bulunulabilir.
INTRODUCTION
Maritime trade is increasing in parallel with the world population. Maritime technology is striving to supply this demand by various way including bigger ship dimension. Therefore, manoeuvring of the ships in congested or busy waters have become more dangerous with the emergence of increasingly larger ships. On the other hand, autonomus navigation systems lead the big data eco system in the maritime domain. Collision avoidance at sea, highly related with autonomous navigation, has become fundamental pillar of navigational safety in all dimension of waterways. Maritime regulations such as International Maritime Organization (IMO) recommendations and International Regulations for Preventing Collisions at Sea (COLREG) have already provided some methods regarding navigational safety. In fact, these methods are applicable and reliable to ensure navigational safety according to navigation experience. COLREG introduces a number of practical ways to avoid collision. IMO also gives a number of recommendations to support navigation safety. Maritime education and training (MET) is also supported by various national and international organizations to decrease the number of navigational collisions caused by human factor. Despite these efforts, there are 1296 maritime incidents in the last 5 years in the word 383 of which are collision and grounding according to IMO. Furthermore 80 per cent of these incidents are related about human erroneous action. It is clear that the organizational and regulational efforts are not sufficient enough to reduce maritime incident to the intended level. Because of this predicament, decision making/supporting systems turn into an important facilitating equipment to increase collision avoidance capability and navigational safety. These decision-making systems are the unique candidates to reduce navigational errors arised from human interactions. Therefore, a systematic navigational collision risk assessment model can be practical, reliable and necessary to monitor navigational collision risk instantly in all kind of waterways.
Scientific studies have employed various type of methods to assess navigational collision risk. Macduff (Macduff, 1974) and Fuji et al. (Fujii, Yamanouchi, and
Mizuki, 1974) are the pioneers of the collision probability models. These models base on the appearance frequency of the vessel in the specific water area. One of the shortcomings of these studies is the assumption of the randomly distributed ships over the specified water area. It means that they employ the ships frequency to assess the collision risk in the considered water area. These kind of probability models evaluate the potential accident situations in the given waterway like Gulf of Finland (Jakub Montewka, Hinz, Kujala, and Matusiak, 2010), Yangtze River (D. Zhang, Yan, Yang, Wall, and Wang, 2013) and Strait of Istanbul (Ulusçu, Özbaş, Altıok, and Or, 2009). This kind of evaluation models base on the frequency of the vessels. While probabilistic models are bound to specific water area, individual collision risk models can evaluate any type of waterway (open sea, restricted waters, etc.) in a macroscopic way. Furthermore, individual collision risk models, producing a risk degree, are a unique provider for collision avoidance models. Therefore, in the rest of the study, NCR has referred to the individual navigational collision risk which can be deemed as the link between the maritime transportation risk analysis and the collision avoidance. There are different data sources such as AIS, IMO, EQUASIS, EMSA, simulations and Local Maritime Organizations to conduct maritime researches. However, underreporting is also very common in maritime industry. Annex 28 of IMO Resolution MSC.115 (73) states that GPS receiver has the performance accuracy of position within 10 meter and 35 meters with 95 per cent confidence interval in differential and non-differential modes, respectively. Due to the importance of close distances in port approach manoeuvrings, I chose simulation data rather than AIS data to construct risk assessment model. In order to process the proposed NCR methodology, experienced pilots were called to conduct 140 approach manoeuvrings (lasting approximately 110 hours in total) in a JMS Ship Handling Simulator. After conducting these approach manoeuvrings in the JMS Ship Handling Simulator, a unique data set to perform analysis was obtained free from any error or missing fields. If we consider the lacking data in maritime domain into consideration, this data set presents amount of opportunities to investigate other aspects of port basins as well. In this doctoral thesis, I employed this unique and case specific data set to process the NCR assessment methodology. In addition to these, I introduce three new parameters to assess NCR in ports basins according to simulation data set. All three new parameters contribute to determining the NCR of ships properly and consistently. The
proposed NCR assessment methodology includes three learning layers; unsupervised learning (fuzzy clustering), semi supervised learning (semi supervised support vector machines), and supervised learning (random forest). This proposed NCR assessment methodology gives quite significant results in parallel with the maritime standards and regulations, and can enable stakeholders of maritime industry to instantly evaluate the NCR of ships in the port approach manoeuvrings.
Research Background
Port basins are very critical both for world trade and ship navigation, as they are the gateways of merchant fleets. Port authorities recommends safety procedures and limits in port approach manoeuvrings to ensure navigation safety. Furthermore, ISO standards (BSI, 2016) proposes some limitation for parameters such as; speed, wind speed etc. for port approaches manoeuvrings. In addition to parameters of ship, port layout recommendations (Del Estado, 1999) are also proposed for safe navigation in port basins. These recommendations include not only safe berthing speed, turning circle and passage sizes but also other important layouts and environmental limits. Both port layout recommendations and navigation limitation within port basins can play a reference role in risk assessment methodology. Therefore, I try to put together these safety standards and navigation risk assessment methodology to construct a consistent and meaningful collision risk prediction tool.
While there are various approaches to determine the collision risk (Debnath and Chin, 2010; Hilgert and Baldauf, 1997; Kijima and Furukawa, 2003), there is no consensus on defining NCR because of the multi dimensional charactheristic of the maritime environment. There is number of collision risk assessment studies in the literature. Each study has a contribution to the collision risk assessment. However, because of the multi dimensional maritime environment and uncertainty in ship navigation, there is no common way to define collision risk. Risk analysis in the maritime application area (Goerlandt and Montewka, 2015) is not so widely classified according to risk approaches. While understanding of the risk can be changed by how the risk concept is defined, Goerland and Montewka (2015) have extended that to eight classes for maritime transportation applications; strong realist, moderate realist, moderate realist with uncertainty quantification, scientific proceduralist, precautionary constructivist, moderate constructivist with uncertainty evaluation, moderate constructivist and
strong constructivist. Realist risk approaches emphasise the underlying true risk as an outcome of a technology, constructive risk approaches hold that risk as a mental construct of the experts. Other than these, proceduralist risk approaches consider different stakeholders, such as scientists and risk-affected laypersons. This doctoral study employs a moderate realist risk approach to assess the navigational collision risk.
Aim of the Thesis
The main purpose of this doctoral thesis is to present an applicable, reliable and validated NCR assessment methodology in port approach manoeuvrings including berthing and unberthing. This has been achieved by focusing on the NCR assessment of ships encountering with land, shallowness, buoys and other land obstacles until berthing in port manoeuvring basins. The applicability, reliability and validation of the proposed methodology is based on the raw data utilization, maritime standards/recommendations and simulations applications, respectively, as it will be focused in the subsequent chapters. The focus of this doctoral thesis is not the open sea or restricted waterways introduced by former NCR assessment studies. Therefore, this doctoral study is the only candidate for the NCR assessment methodology in port manoeuvring areas.
Structure of the Thesis
I already mentioned that the focus of this doctoral thesis is determining the navigational collision risk in port waters including berthing and unbderhing. This has been achieved by a number of steps; systematic literature review, proposing new parameters for specialised for port basin, simulation experiments, a detailed survey and a three layers’ statistical analysis framework. The systematic literature review enables us to reveal the current navigational collision risk structures and parameters. By this way, the parameters, can be employed in this study, have been evaluated. On the other hand, simulation experiments have provided a unique dataset specialised for port approach manoeuvrings. A detailed survey filled by the maritime experts gives sample navigational collision risk degrees for selected features. At last, I have constructed a three-layer statistical analysis methodology including fuzzy logic, semi-supervised support vector machines and random forest to predict NCR degree. Figure 1.1 represents the structure of the thesis.
LITERATURE REVIEW
The research methodology of the literature review is structured with a systematic way proposed by Tranfield et al. (2003). The general framework of the systematic literature review employed by the proposed doctoral thesis has been shown in Figure 2.1. The first stage, “Define” includes two steps: Ghe Identification of need for a literature review and the Development of a literature review protocol. The second stage “Collect and Select” includes two steps: The Identification of documents and the Selection of relevant documents. The third stage “Analyse” includes the Categorization of documents and Data extraction steps. Finally, “Result” stage includes the Document findings step. The Document finding step presents the information extracted from the existing knowledge.
Figure 2.1 : Systematic literature review framework (Tranfield et al., 2003). • Step 1: Identification of need for a literature review • Step 2: Development of a literature review protocol
STAGE 1: DEFINE
•Step 1: Identification of documents •Step 2: Selection of relevant documents
STAGE 2: COLLECT AND SELECT
•Step 1: Categorization of documents •Step 2: Data Extraction
STAGE 3: ANALYSE
•Step 1: Document findings
STAGE 4: RESULTS
Figure 2.2 : Flowchart of the review protocol (Tranfield et al., 2003).
Technological advancement in collision avoidance system necessitates a systematic literature review for NCR assessment because there is various risk assessment application in maritime industry. Therefore, this study can support the future works by introducing a systematic literature review considering methods and categorizations of each NCR assessment studies.
The literature and discussions within the review team (Tranfield et al., 2003) has been done in order to identify the keywords and search terms. Scoping study has also a signigficant role in determining the keywords. This is a process which needs an approval from team members to conduct subsequent research as shown in Figure 2.2. This doctoral study aims to take into all related studies into consideration. Furthermore, the maritime transportation industry has always close links to scientific studies. Technological advancements trigger the maritime transportation optimization needs. Because of these reasons, the time span for literature review is from 1970 to October of 2018.
Realist, constructivist and proceduralist risk approaches (Bradbury, 1989; Rosa, 1998) are the general risk approaches employed in the risk analysis studies. However, Goerlandt and Montewka (Goerlandt and Montewka, 2015) have further classified these risk approaches for the maritime domain. The study of Goerlandt and Montewka (2015) have classified maritime transportation risk analysis studies considering the following characteristics;
(i) the analysis aims and scope; (ii) the applied definition of risk;
1. Scoping Studies
* Relevance and size of literature * Delimination of the subject area *Creation of inclusion and
exclusion criteria
* Identification of keywords and search terms 3. First Results * Number of collected papers * Brief check CONDUCT RESEARCH Approve 2. Systematic Search * Protocol development * Search string definition * Databases selection * Search string application
(iii) the applied tools to measure risk (quantitative, qualitative, fuzzy number, probability, etc.);
(iv) whether events, or events and consequences are accounted for;
(v) the applied types of evidence (data, models, expert judgement, lay person judgment);
(vi) the consideration of contextual attributes (fear, voluntary exposure, equity, etc.); (vii) the adopted approach to risk analysis science;
It has been observed from the historical overview of the adopted maritime transportation risk analysis studies (Goerlandt and Montewka, 2015) that three approaches (strong realist, moderate realist and moderate constructivist) are more common among the maritime transportation risk analysis studies. It is important to note that there are tangible differences among the studies employing different tools to measure risk. For example, risk analysis studies employing frequentist probability (Goerlandt and Kujala, 2011; Klemola et al., 2009; Macduff, 1974; Merrick et al., 2002; Merrick and van Dorp, 2006; J Montewka et al, 2011; Montewka et al, 2010; van Dorp et al, 2001) as a risk measurement tools aims to determine the transportation safety of a specific area with historical data. However, there is other objetives, one of which is the focus of this doctoral study, such as collision avoidance, trajectory planning, decision support for vessel traffic service (VTS) operators and officer training. Individual collision risk assessment studies focusing on these objectives uses different risk measurement tools (subjective probability, modelled probability, quantitative indicator, qualitative indicator, fuzzy number, etc.).
The boundary of the systematic literature review is presented in Table 2.1, also representing the characteristics of the corresponding risk analysis studies. The studies adopting English language has been reviewed. The review protocol to process the review has been presented in Table 2.2. The objectives of collision risk and collision avoidance studies have similar characteristic, and are nested. Therefore, the search strings have been determined consistently. This study employs two online databases to search published articles and conference papers: Web of Science and Elsevier. Different type of search strings enabled us to reveal unexplored studies since these online databases have similar search engines. Search strings by databases have been shown in Table 2.3.
Table 2.1 : Boundary of the research according to the characteristics (Ozturk and
Cicek, 2019).
Characteristic Measure
The analysis aims and scope Individual navigational collision risk The applied definition of risk All
The applied tools to measure risk
Quantitative/qualitative indicator, fuzzy numbers, subjective/modelled probability
Whether events, or events and
consequences are accounted for Events
The applied types of evidence Data, model and expert judgments The consideration of contextual attributes None
The adopted approach to risk analysis science
Strong realist, moderate realist and moderate constructivist
Table 2.2 : Review protocol (Ozturk and Cicek, 2019).
Subject Description
Keywords navigation; collision; risk; ship Search field title, abstract, keywords
Boolean Operators OR, AND Exclusion criteria See table 1.
Publication type Article and conference papers Publication language English
Time interval 1970 - October 2018
Table 2.3 : Search strings by databases.
Database Search string
Scopus Title: (“Collision” OR “Avoidance”) AND Abstract: (“Ship” AND
“Navigation”), TO (EXACTKEYWORD, "Ships")), (LIMIT-TO (SUBJAREA, "ENGINEERING")), (LIMIT-TO(DOCTYPE , "article ") OR LIMIT-TO (DOCTYPE , "cp")), Language: (“ENGLISH”)
Web of Science
Title: (“Collision” OR “Risk” OR “Avoidance”) AND Abstract(“Ship” AND “Navigation”) Research Area: ENGINEERING, Type: (ARTICLE OR PROOCEEDING), Language: (“ENGLISH”)
Science Direct
Title: (“Collision” OR “Avoidance”) AND Abstract: (“Ship” AND
“Navigation”), Keywords: (“Risk”)
The proposed literature review process reveals 237 individual collision risk studies (Ozturk and Cicek, 2019). After removing irrelevant studies meeting the exclusion criteria, literature review process has determined a total of 34 studies to review. Figure 2.3 and Figure 2.4 presents the classificaiton of these 34 navigational collision risk studies according to years of publication and publication journal, respectively. Figure 2.3 states that the number of published NCR analysis studies has increased after 2005. Furthermore, Figure 2.4 states that Ocean Engineering, The Journal of Navigation, and
Safety Science journals has published the 46 per cent of the NCR studies (Ozturk and Cicek, 2019). A full review of NCR studies enabled these NCR assessments studies to be categorised by risk approaches mentioned above. Figure 2.5. shows that strong realist risk analysis approach is the most common approach among these NCR assessment studies.
Figure 2.3 : Number of selected papers organized by years of publication (Ozturk
and Cicek, 2019).
Figure 2.4 : Number of publications by journals (Ozturk and Cicek, 2019).
0 1 2 3 4 5 1 9 9 3 1 9 9 7 2 0 0 0 2 0 0 1 2 0 0 3 2 0 0 4 2 0 0 5 2 0 0 6 2 0 0 7 2 0 0 9 2 0 1 0 2 0 1 1 2 0 1 2 2 0 1 3 2 0 1 4 2 0 1 5 2 0 1 6 2 0 1 7 0 2 4 6 8 2016 IEEE TrustCom-BigDataSe Advances in Mechanical Engineering Applied Ocean Research Expert System with Apllication German Journal of Hydrograpy ICTIS 2011 IFAC Manoeuvring and Control of Marine Craft Infor. Processing and Security Sys. Int.Shipbuilding Progr. J. Cent. South Univ. Mathematical Problems in Engineering Neurocomputing Proc. of the Fifth Int. Conf. on Mach. Learning and Cyb. Risk Analysis The 3rd Int. Conf. On Transportation Infor. and Safety J. Mar. Sci. Technol. Transnav Journal Safety Science The Journal of Navigation Ocean Engineering
Figure 2.5 : Distribution of the studies by risk approaches (Ozturk and Cicek, 2019).
Studies on navigational risk analysis
A systematic literature review framework, conducted above, enabled NCR assessment studies to classify by risk analysis approaches. The motivation behind this classification is to provide insight into which risk approach the NCR models have adopted. It has been assumed that this classification approach may also contribute to the study of Goerland and Montewka (2015). This section provides some details about the focused NCR studies.
Strong realist models presume that quantitative analyses of the system are the representation of the absolute risk. Expert judgment and stakeholders are not considered in these models. For example, Hilgert and Baldauf (1997) have constructed a rule-based collision risk model of ship to ship encounters for open sea waterways. Collision risk model based on CPA (Closest Point of Approach) and range with proposed critical limits: hydrodynamic safe passing distance (𝐶𝐻), safe passing distance (𝐶𝐴), safe range (𝑅𝐴), critical range (𝑅𝑐) and manoeuvring range (𝑅𝑀). A few years later, Smierzchalski (2005) proposed a hexagon-shaped area of danger to state navigational risk for trajectory planning. The author has formulated six distance from the centre of the ship to cover the ship domain corners with parameters such as TCPA, DCPA, own speed, relative speed, ship length, ship breadth. But stern and port side safe distances have been assumed to be more than 0.5 miles. It has been suggested that the safe distance in front of the bow should be 2 to 3 miles, under which collision risk emerges. However, Szlapczynski (2006) considers that TCPA and DCPA parameters are pointless to assess NCR. Then, temporary approach factor (𝑓𝑚𝑖𝑛), the ratio of the distance between two encountering ships to minimum acceptable distance, has been proposed for integrating with different ship domain model when convenient. The
56% 23% 21% Strong Realist Moderate Realist Moderate Constructivist
reason of proposing this factor is the obscure perspective of DCPA in ship domain violation. Unlike the mentioned study (Śmierzchalski, 2005), Wang (2010) introduced a quaternion ship domain model employing the analytical ship domain of Kijima and Furukawa (2003). The study has determined the size of ship domain by four radii (fore, aft, starboard, and port), which can be elliptical or convex to overcome the limitation of analytical domains. Comparison of the proposed model with existing ship domain models in simulation has given more coherent (in reaction time) collision risk awareness for each encounter situation. Modifying existing ship domain models is a distinguished perspective, although it can lead to heavy dependency on preferred ship domain model. Proposed approach in that study overcomes the limitation of traffic geometry. Some authors have presented navigation-specific variables to take NCR assessment further. For example, Bukhari et al. (2013) proposed variance of compass degree (VCD), which is the difference between two consecutive bearings. In addition to VCD, DCPA and TCPA have also been employed to construct fuzzy rule-based NCR assessment model. Smolarek and Sniegocki (2013) have assumed that wind speed is an important aspect of the safety of the manoeuvring in restricted waters. Authors have examined the risk in the approach channel considering the speed, wind speed and wind direction parameters. Relation between parameters and the risk degree has not been examined but it can be seen that speed parameter has the highest linear relationship with the risk degree. Other than the above mentioned models, nonlinear support vector machines (SVM) has been employed by Gang et al. (2016) to determine collision risk index for both mariners and autonomous systems. The collision risk estimation methodology of Ahn et al. (2012) and Gang et al. (2016) are similar. However, Gang et al. (2016) have employed fuzzy theory without expert knowledge. Other than the proposed limits of (Hilgert and Baldauf, 1997), Wen et al. (2016) evaluated SPD as “safe distance of approach” (SDA) regarding COLREG Rule 8 and proposed the limit values for ship encounters. In fact, int this study collision risk has been assumed when the distance of two ship is below 8 nm. On the other hand, authors have considered the 12 times ship length as the distance of last minute avoidance. Moderate realist models are similar as the strong realist approach but they consider the expert judgment as a source of evidence. For example, Kao et al. (2007) have introduced a fuzzy logic approach to solve overloading problem of VTS systems in
congested harbour entrances. Speed, length of ship and sea state fuzzy membership functions determines the radius of a guarding ring. The fuzzy guarding ring is similar to the concept of the ship domain. The model is recommended as a tool for monitoring the ships in the congested harbour entrance. However, the multi-ship encounters have not been examined. Mou et al. (2010) have benefited from Safety Assessment Model for Shipping and Offshore on the North Sea (SAMSON) accident number estimation model to evaluate collision risk in busy waters. Some parameters (TCPA, CPA and encounter angle) are combined with the SAMSON’s model. That model comprising maritime accident dataset has been proposed as a helpful tool for monitoring the risk of maritime traffic. Questionnaires can be quite subjective for incorporating perceived risk level. Therefore, Ahn et al. (2012) have proposed an interviewee-independent method deemed a pioneering work. This has been achieved by both rule-based collision risk table modified from Hara and Hammer (1993) and fuzzy membership functions of DCPA (Koyama and Yan, 1987) and TCPA. The study of Hara and Hammer (1993) has obtained the knowledge of watching officers by conducting experiment in a ship handling simulator. Ahn et al. (2012) has employed adaptive network-based fuzzy inference system to increase the number of membership functions of DCPA and TCPA of Hara and Hammer (1993). A neural network model uses the the output of the fuzzy rules as output label and own ship’s speed, target ship speed, the own ship’s heading, target ship heading, the distance between the two ships, and ship domain (Zhao et al., 1996) as input vector. Restricted visibility (Cockcroft and Lameijer, 2012), topographical characteristics (Lee and Rhee, 2001a) and high speed constraints have also adopted in this NCR evaluation to elicit complex maritime environment. The results of neural network model in the simulations have shown that neural network model is more realistic than fuzzy inference system. This study also designated different ship domain for restricted and open sea areas, having particular idiosyncrasies. The NCR assessment methodology of this study is quite feasible for maritime environment. The severity of navigation can be assessed by specific parameters like DCPA, TCPA and VCD but the parameters like speed, heading and distance are the system parameters affecting the navigational situations. On the other hand, Zhang et al. (2016) have taken the previous (Zhang et al., 2015) study forward by including a new parameter called minimum distance to collision ( Montewka et al., 2010) to the model. Vessel size is employed in the model rather than distance between two ships. As a matter of fact, the study has ignored the risk below one nautical mile
encounters. VCRO values of scenarios are confirmed by expert judgments can be considered as a source of evidence. The former model of Zhang et al. (2016) has employed a static ship domain, and is impotent to give any information about the dynamic state of the ship. Therefore, Zhang et al. (2017) applied different domain model (Kijima and Furukawa, 2003) to display the dynamic state of NCR. The merged VCRO (average VCRO rank of two encountered ships), the VCRO change rate and the ship pair size have been proposed with fuzzy clusters. The authors used expert knowledge with fuzzy clusters to assess NCR. The new VCRO model of Zhang et al. (2016) have distinguished from prior studies because of the ship manoeuvring characteristics obtained by the ship domain model of Kijima and Furukawa (2003). These characteristics are the time to 90-degree heading and, the tactical diameter, representing the response time and pattern of a ship in changing direction in case of emergency. Monitoring the dynamic changes in encounter situation is another superiority of this VCRO models over prior VCRO models
The moderate constructive models assume the risk as a mental construct of an expert in constrast to strong and moderate realist models. The data has come from the navigational situation but the risk degree has been obtained from the expert knowledge according to the monitoring data. For instance, Inoue (2000) has presented an environmental stress model based on linear regression, employing the time to collision (TTC) parameter to estimate ship handling difficulty in restricted and congested waterways. TTC parameter can be assumed to be the time to reach any obstacle or ship. Pedersen et al. (2003) used this approach to compare two navigational aid (ARPA and recommended Collision Danger Presentation). In forthcoming years, Chin and Debnath (2009) proposed ordered probit regression model to assess NCR in port waters. Knowledge of 160 pilots has been obtained to fit NCR model based on TCPA and DCPA variables for night/day condition and four different vessel class. The results have shown that DCPA is more significant in determining the NCR level than is TCPA. Navigational Traffic Conflict Technique (Debnath and Chin, 2010) has employed this approach to overcome the limitation of the number of collision records in a given water way like port anchorages (Debnath and Chin, 2016). On the other hand, Perera et al. (2011) have proposed a fuzzy approach to increase the navigational safety of vessels in open seas. Distance, relative bearing of target vessel, encounter angle and relative speed are the input parameters of the proposed NCR model.
However, DCPA and TCPA parameters are criticized (Goerlandtet al., 2015; Zhang et al., 2015) as they can lead to misleading risk evaluation. This idea arises from the fact that low level of CPA doesn’t mean risky encounter if the bow-cross range (distance the second vessel passes ahead of the bow of the first vessel) is wide. In other words, port to port or starboard to starboard encounter types can be safe although CPA value is low. Zhang et al. (2015) introduced a vessel conflict ranking operator (VCRO) to assess the safety of two ships encounter to overcome that problem. The model excludes the harbour approaches from the application. This exclusion is an indicator of the idiosyncrasies of different waterways (Ozturk and Cicek, 2019). Lopez-Santander et al. (2017) has adopted relative situation (head-on, crossing, overtaking), colour (red, green) and trajectory variability (erratic, not erratic) in addition to TCPA and DCPA to assess NCR. While the results have shown that DCPA has the most significant effect on the perceived risk with the highest absolute coefficient value (0,395), TCPA coefficient has the lowest value (0,0249), which is a surprising result for maritime environment. Since it is insurmountable to examine all NCR studies in detail, information of other NCR assessment studies (Chen et al., 2014; Perera and Guedes Soares, 2015; Ren et al., 2011, etc.) has been introduced in Table 2.4 and Table 2.5.
Parameters in the studies
The occurance rates of the navigational parameters in NCR studies have been examined in this section comprehensively. Firstly, the NCR parameters employed in each NCR studies have been presented in Table 2.4. Furthermore, Table 2.5 introduces the number and rate of occurence of the NCR parameters.
The distribution of some parameters can be seen in Table 2.4. It can be observed from the Table 2.4 that DCPA and TCPA parameters have been employed by most of the NCR studies. There is no increasing or decreasing trend in the parameters used in the NCR studies. However, it is important to note that most studies give priority to use DCPA and TCPA, in spite of their alternatives recommended by other studies. It has been observed that there is no noteworthy distinction among different area-specific studies (open sea and restricted waters) in respect to employed parameters (Ozturk and Cicek, 2019). DCPA, TCPA and relative bearing are the most employed risk indicators in the NCR studies as can be seen in Table 2.5. It is absolutely clear these parameters are highly corelated with the mariners’ judgment about collision risk (Ozturk and
Cicek, 2019). As a matter of fact, COLREG Rule 7 (b) states that radar equipment, which determines TCPA and DCPA, can be used as early warning of collision risk. In addition, steady relative bearing is a collision indicator as stated in rule 7 (d). However, environmental parameters such as wind speed, current speed, wave height, water depth have rarely been employed as can be seen in Table 2.5 (Ozturk and Cicek, 2019). Another inference from the Table 2.5 is that there is no individual collision risk assessment study regarding port approach manoeuvrings.
Table 2.4 : Some parameters taken into account in NCR studies (Ozturk and Cicek, 2019). Authors RA AA M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 M11 M12 M13 M14 M15 M16 M17 M18 M19 M20 Lopez-Santander and Lawry, (2017) 3 NM * * * Zhang et al., (2017) 2 O,R * Gang et al., (2016) 1 NM * * * * Wen et al., (2016) 1 O * * * Zhang et al., (2016) 2 O,R * * * Zhao et al. (2016) 2 NM * * * * Chen et al., (2015) 1 O * * * * Goerlandt et al., (2015) 3 O * * * *
Perera and Guedes
Soares, 2015 1 O,R * * Zhang et al., (2015) 3 O,R * * * Shu et al., (2014) 2 R * * * * Bukhari et al. (2013) 1 R * * Li and Pang, (2013) 1 R * * * Smolarek and Sniegocki, (2013) 1 R * * Ahn et al., (2012) 2 R * * * * * Balmat et al., (2011) 1 O * * * * * * *
Table 2.4 (continued) : Some parameters taken into account in NCR studies (Ozturk and Cicek, 2019). Authors RA AA M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 M11 M12 M13 M14 M15 M16 M17 M18 M19 M20 Perera et al., (2011) 3 O * * * Ren et al., (2011) 1 NM * * * Meng et al., 2009 1 NM * * Mou et al., (2010) 2 R * * * Tam and Bucknall,
(2010) 1 R * * *
Wang, (2010) 1 O,R * * *
Xu et. Al (2010) 2 O * * * * Balmat et al.
(2009) 1 O * * * * * * *
Chin and Debnath,
(2009) 3 R * * *
Kao et al., (2007) 2 R * * *
Liu and Liu,
(2006) 1 NM * * Szlapczynski, (2006) 1 NM * * * * Smierzchalski, (2005) 1 R * * Lisowski, (2004) 1 NM * * * * * * Kijima and Furukawa, (2003) 1 NM * * * *
Lee and Rhee
(2001) 1 O,R * * * *
Inoue, (2000) 3 R Hilgert and
Baldauf, (1997) 3 R * *
*1=Strong realist model, 2=Moderate realist model, 3=Moderate constructivist model; O= Open sea, R=Restricted waters, NM=Not mentioned, RA=Risk approach, AA=Application area.
Table 2.5 : Occurrence of NCR parameters (Ozturk and Cicek, 2019).
Key Parameter Number of
Occurrences Rate of Occurrence
M1 TCPA 19 54,3% M2 DCPA 18 51,4% M3 Relative bearing 13 37,1% M4 Distance 12 34,3% M5 Speed 7 22,1% M6 Length 5 15,2% M7 Relative speed 4 11,4% M8 Visibility 4 11,4% M9 Breadth 3 8,6% M10 Day/Night 3 8,6% M11 Sea State 3 8,6% M12 Wind Speed 3 8,6% M13 Advance 1 3,0% M14 Encounter type 2 5,7% M15 Flag 2 5,7% M16 Heading 2 5,7%
M17 Kind of water region 2 5,7%
M18 Number of companies 2 5,7% M19 Number of detentions 2 5,7% M20 Ship domain 2 5,7% M21 Tactical diameter 1 3,0% M22 Tonnage 2 5,7% M23 Velocity ratio 2 5,7% M24 Years of construction 2 5,7% M25 Relative distance 1 2,9% M26 Approach factor 1 2,9%
M27 Bow cross range 1 2,9%
M28 Color 1 2,9%
M29 Encounter angle 1 2,9%
M30 Minimum distance to collision 1 2,9%
M31 Geographic position 1 2,9%
M32 Rate of change of relative direction 1 2,9%
M33 Shipping evaluation 1 2,9%
M34 Speed change 1 2,9%
M35 Temporary approach factor 1 2,9%
M36 Trajectory variability 1 2,9%
M37 TTC 1 2,9%
M38 VCD 1 2,9%
M39 VCRO 1 2,9%
M40 VCRO Change rate 1 2,9%
Methodologies in the studies
Risk assessment applications of the maritime navigation are lacking the clarity about foundational issues and key terminology concerning the scientific methods for risk analysis (Goerlandt and Montewka, 2015). The diversified data sources, case specific approaches and unclear boundary of the safety limits in NCR analysis can be the primary reasons of that issue. There are various data sources for navigational risk assessment. The leading data sources are extracting expert knowledge views via surveys (Chin and Debnath, 2009; Inoue, 2000; Lopez-Santander and Lawry, 2017); simulator experiments (Ahn et al., 2012) and navigational databases like AIS (Debnath et al., 2011; Mou et al., 2010). These NCR studies generally employ mathematical functions (exponential, trigonometrical, logarithmic etc.) and fuzzy membership functions or statistical methods. Then, the methodology determines the navigational collision risk by classifing the outcomes. The classifications on the limit values or a rule-based approach. However, this systematic bases on the perennial experience of the maritime environment and not refers to any tangible rule or standard. Determining the limit values is the most critical step of the collision risk assessment as they are the tresholds for decision making. COLREG also doesn’t give anystandard about the safe passing distance of two encountered ship in diversified states of encounters although it is the fundamental pillar of maritime traffic. This situation is also stated by various authors (He et al., 2017; W. Zhang et al., 2015). Because of this standard safety limit gap, NCR assessment models prefer to design these navigational context (ship domain, safet distance and others) by mathematical functions. Aims and risk measurement models of each NCR assessment study have been presented in Table 2.6.
