A hybrid algorithm for developing third generation HRA methods using simulator data, causal models, and cognitive science

Over the past 10 years, there have been significant international efforts to modernize Human Reliability Analysis (HRA), with most efforts focused on one of two directions: developing new sources of HRA data from control room simulators, and developing new HRA methods based in cognitive science. However, these efforts have proceeded largely independently, and there has been little research into how to leverage these scientific advances in data together with the scientific advances in modeling and methods. This is a significant gap for HRA, and motivates a need for methodologies to unify the efforts of the modeling and data collection communities. In this paper we define a comprehensive hybrid algorithm for using causal models and multiple types of HRA data to provide a rigorous quantitative basis for cognitively based Human Reliability Analysis (HRA) methods such as PHOENIX and IDHEAS. The algorithm uses causal models built from and parameterized by a combination of data from cognitive literature, systems engineering, existing HRA methods, simulator data, and expert elicitation. The main elements of the hybrid algorithm include a comprehensive set of causal factors, human-machine team tasks and events, Bayesian Network causal models, and Bayesian parameter updating methods. The algorithm enhances both the qualitative and the quantitative basis of HRA, adding significant scientific depth and technical traceability to the highly complicated problem of modeling human-machine team failures in complex engineering systems.