Development of bayesian models for marine accident investigation and their use in risk-based ship design

Historical marine accident/incident data remain severely underused in regulatory development as well as during design and operation. It is widely recognized that this is mainly the result of underreporting in commercially available databases and in databases maintained by national authorities. A factor further signifying this underuse is the evident improper reporting because most data are maintained as textual information requiring significant amounts of time and effort to distill and use the essential characteristics of accidents. Compounded with improved accessibility to an ever increasing amount of historical records, there is a need to develop the means that all the available information from marine accident/incidents is fully used in decision-making during development of new regulations, design, and operation. This article elaborates on the underlying causes for the current unsatisfactory state of affairs and details the description of the structures adopted for the development of appropriate marine accident databases using Bayesian Belief Networks as the platform for translating the information contained in the databases to probabilistic risk-based knowledge-intensive models. The article further explains the use of these models within a risk-based ship design framework, concluding with an example case study of application for fire safety onboard passenger ships.