Human reliability analysis using virtual emergency scenario via a Bayesian network model

Abstract

Human reliability assessments (HRA) are typically completed by eliciting expert opinion. Data used are subjective and are prone to uncertainty and errors. This thesis outlines an HRA method using a Bayesian network (BN) model to evaluate human performance in emergency scenarios using a virtual environment (VE). VE can be used to simulate emergency situations to evaluate human performance in an environment that is controlled and safe and gives access to data that is based on an experimental method, rather than expert opinion. This method involves selecting appropriate performance shaping factors (PSFs) that are varied into different states to create credible scenarios in the VE to observe human performance. The virtual experimental technique provides a way to collect data to quantify a BN. The BN approach is suited to the assessment of human reliability due to its ability to 1) characterize dependency among different performance shaping factors (PSFs) and human errors, 2) incorporate new evidence as it becomes available, and 3) quantify the impact of different PSFs on different individuals. This paper presents an extension of the work done by Musharraf et al. (2014) by introducing PSFs that were purposively selected based on the ability to implement them in the VE, their relevance to real-life situations, and whether they could be controlled to minimize the effects of variables other than the chosen PSF. The PSFs used in this paper are complexity, stress, and uncertainty.