BAYES-HEP: Bayesian belief networks for estimation of human error probability

Abstract

Human errors contribute a significant portion of risk in safety critical applications and methods for estimation of human error probability have been a topic of research for over a decade. The scarce data available on human errors and large uncertainty involved in the prediction of human error probabilities make the task difficult. This paper presents a Bayesian belief network (BBN) model for human error probability estimation in safety critical functions of a nuclear power plant. BBN is a powerful tool and has been widely used for risk and reliability analysis framework under the conditions of uncertainty. BBNs are joint probability distribution between multiple variables (i.e., performance shaping factors, error modes, error mechanisms, etc.) expressed as directed acyclic graphs consisting of nodes and arcs. Nodes represent system components, and arc represents relationship among them with conditional probability tables. This network functions as an engine for calculation of probability of events given the observation of other events in the same network. BBN can be used to model the uncertainty parameters in a system. Sensitivity analysis can also be performed to study how uncertainty in the model output can be attributed to different sources of uncertainty in the model input. It is frequently applied in real-world situations such as diagnosis, forecasting, and environment, but received less attention in the area of human reliability. In view of its natural architecture, BBN is found to be more appropriate when there is scarce, multi-source data available as in the case of human error data. Further, the probabilistic approach of BBN is best suited for safety assessment to predict system reliability and estimate the probability of consequence of an event. BBN has been adapted to model human factors by taking into account the different parameters and their mutual influences. BBN can also be used to identify the potential human factors leading to significant reduction of accident probability during the operation of NPP. Several researchers have developed advanced HRA models to estimate the error probability but most of the HRA model requires expert judgment at several stages and hence it is subjective. Other models such as HCR and CREAM are not applicable for all scenarios. Limited studies are carried out to eliminate the need of human expert in the human reliability methods. This paper introduces a method that uses the available historical data and provides efficient technique for automated evaluation of human error probability using BBN. In this respect, human cognitive reliability (HCR) has been identified as a suitable model to estimate the mean time required for human intervention and in the selection of human action to estimate HEP in the model. The developed model using BBN would help to estimate HEP with limited human intervention. A step-by-step illustration of the application of the method and subsequent evaluation is provided with a relevant case study and the model is expected to provide useful insights into risk assessment studies.