Bayesian network approach to human reliability analysis (HRA) at offshore operations

Abstract

This thesis presents a quantitative approach to human reliability analysis (HRA) in offshore emergency conditions. Most of the traditional HRA methods use expert judgment techniques as human performance data for emergency situations are not readily available. Expert judgment suffers from uncertainty, incompleteness and when collected from multiple experts, may have conflicting views. This thesis investigates these limitations and presents a proper aggregation method to combine multiple expert judgments using Fuzzy Theory to handle the uncertainty and Evidence Theory to handle the incompleteness and conflict. Furthermore, the traditional approaches of HRA suffer from the unrealistic assumption of independence among different performance shaping factors (PSFs) and associated actions. This thesis addresses this issue using the Bayesian network (BN) approach which can represent the interdependencies among different PSFs and associated actions in a direct and structured way. The integration of Fuzzy Theory and Evidence Theory to the BN approach gives an HRA model that can better estimate the success or failure likelihood of personnel in offshore emergency conditions. Incorporation of environmental factors makes the model applicable for offshore emergencies occurring in harsh environments. Finally the thesis presents a new methodology to collect human performance data using a virtual environment. Using the collected data, a simplified BN model of offshore emergency evacuations is tested and verified.