Bayesian network modeling of potential patterns in maritime safety performance

Although major maritime accidents occur rather rarely, they can produce severe consequences. Safety management aims at preventing such accidents. For monitoring and improving safety, safety management requires knowledge on the safety performance. Information on various variables which are potentially related to safety and a description of how the variables interact, or which types of patterns the interactions constitute, could be beneficial for increasing that knowledge. However, these patterns might not be apparent, as maritime traffic and its safety are complicated systems. Utilizing Bayesian network modeling approach, this thesis explores potential patterns between variables related to maritime safety. The decision-makers can then exploit this information as a starting point for analyses of the mechanisms which have generated these patterns and of what the patterns tell about safety. The thesis models safety-performance patterns from different, complementing angles. The work begins with examining the feasibility of maritime accident causation pattern models for the exploration of safety performance. This includes analyzing an existing causal collision model and assessing the feasibility of accident and incident data for collision and grounding cause pattern learning. The focus then shifts to patterns present in multiple safety indicator data, before the analysis is extended to safety management patterns and safety management dependencies with safety performance. While causal pattern modeling is found questionable, Port State Control inspections have potential to act as a valuable data source for safety performance information. However, the inspections in Finnish ports have resulted in few deficiencies and thus the data contains only weak patterns. It might be worth evaluating whether the Port State Control could be developed so that the inspections would produce data on more detailed safety performance differences between different ships. On the other hand, maritime safety management seems to be a rather tightly coupled system with several properly functioning subareas but inadequate as a whole. Regarding the application of Bayesian networks for the pattern modeling problem, the thesis concludes that with their capability to express uncertain, rather complex dependencies and to combine data with expert knowledge, Bayesian networks offer an attractive tool for such a task. As the amount of collected and shared data is slowly increasing within the maritime community, the Bayesian network models can be easily updated with new information, and thus their quality and worth for decision-support could be improved.

Department dc:contributor.department

Sovelletun mekaniikan laitos

Grantor dc:publisher

Aalto University

Year dc:date.issued

2015

Author dc:creator

• Hänninen, Maria

Advisors dc:contributor.advisor

• Kujala, Pentti, Prof., Aalto University, Department of Applied Mechanics, Finland
• Kuikka, Sakari, Prof., University of Helsinki, Finland

Contributors dc:contributor

• Aalto-yliopisto
• Aalto University

Language dc:language.iso

en

Repository record dc:identifier.uri

https://aaltodoc.aalto.fi/handle/123456789/15104

OAI record oai:identifier

https://aaltodoc.aalto.fi/server/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Aaaltodoc.aalto.fi%3A123456789%2F15104