A Hybrid Bayesian Belief Network Model for Risk Modeling of Arctic Marine Operations Available to Purchase

Optimizing the design of offshore structures to withstand ice loads is a challenging task and various efforts are under way to develop robust concepts with acceptable structural safety. To ameliorate the deficiencies of structural design, as well as to reduce the costs of such Arctic offshore field developments, ice management operations may be considered to reduce the ice severity. Ice management in sea ice will typically involve use of 1 to 4 icebreakers depending on the operating environment. The ice management fleet is aimed at protecting the offshore installation by breaking the incoming ice into smaller pieces and by reducing the confinement in the ice cover. Failure of such a marine operation in the demanding Arctic environment can threaten the integrity of the offshore platforms or drilling units, e.g. by increasing the chances of failure of mooring lines in the occurrence of extreme events. Therefore, understanding the causes of such potential failure, as well as the factors influencing it is of crucial importance in order to plan for and mitigate the risks. Factors with an influence on the risk are called risk influencing factors (RIFs) and can be technical, organizational and human. RIFs are identified and structured in this study in a way that they affect the basic events of a conventional fault tree analysis and consequently the total risk. In this study, the RIFs are treated as uncertain variables. The established model is called a hybrid model because it is a merger of a Bayesian belief network (BBN) for the RIF structure and a conventional fault tree model. The Bayesian framework provides the opportunity for updating of the model constituents as more evidence becomes available over time. Case studies are defined to illustrate the methodology. Results show how the improvement in the status of the RIFs (better practices) can improve the reliability of the mooring lines of a floating unit and how the precision in data and other model parameters affect the results. At the end, an investment priority measure is proposed that can help in determining where among the various influencing factors the available limited resources should be spent in a way that it results in maximum gain in ultimate reliability.