Bayesian modelling for human error probability analysis in CREAM

Abstract

Cognitive reliability and error analysis method (CREAM) has been developed and gradually complemented with novel quantitative techniques to enhance its inherent perspective human error probability (HEP) analysis. While some probabilistic models are used to improve the accuracy of analyzing HEP, they ignore the characteristics of the easiness, visibility and integrity of the traditional deterministic method in CREAM. This paper aims to establish a novel Bayesian network model, which is capable of providing the instant and precise estimate of HEP given the updated information about a dynamic context without compromising the easiness and visibility features of the traditional method. The mathematical procedure of developing the Bayesian network is described in a 6-step methodology, including definition of primary effects of common performance conditions (CPCs), adjustment of dependency of CPCs, new grouping of CPCs, distributions of prior conditional probabilities, integration of positive and negative CPCs and estimate of HEP. Main contributions of this paper lie in its original intention of eliminating irrelevant neutral primary effects in the process of adjusting CPC dependency and separately treating the numbers of positive and negative effects of CPCs. The proposed quantitative HEP analysis methodology can be widely applied to various industries to facilitate the associated human error reduction and operational safety improvement.