Evidential FMEA method for human reliability assessment

Abstract

Evidence theory is a useful tool for modeling and reasoning uncertain information inherent in experts’ evaluations, which is not handled efficiently in traditional failure mode and effects analysis (FMEA). This study proposes an integrated FMEA method that incorporates evidence theory and is applied to human reliability assessment. The human error information of a human-machine system in FMEA is described as a directed graph by a Bayesian network (BN) to assess the dependence among potential human-related failure modes. The BN is extended to propagate the epistemic uncertainty of FMEA team members, where belief mass is applied to model uncertainties in team members’ knowledge and to convert their subjective cognition into varying levels of uncertainty. Risk indexes for occurrence, severity and detection from multiple sources are defined as a special assessment state. The combination of the belief mass of different failure modes is performed using extended Dempster’s rules to avoid the influence of conflicting evidence. Finally, an application in the healthcare system is provided to verify the effectiveness of our model. A comparison with other fuzzy FMEA methods is also conducted, demonstrating the advantages of the proposed model in dealing with decision-makers’ epistemic uncertainty and potential failure mode interdependencies.