Modelling Causal Relationship Among Performance Shaping Factors Through Bayesian Network on Aviation Safety

Aviation safety is greatly influenced by pilot performance reliability. To assess the reliability, many human reliability analysis (HRA) methods are developed. Currently, in most HRA methods performance shaping factors (PSFs) are used to represent internal and external factors which contribute to human error. Up to now, the effects of PSFs are usually considered to be independent. However, more and more evidences show that causal relationships do exist among PSFs and neglecting that interrelationship will make the assessed human error rate to be too optimistic or too conservative. This paper builds a Bayesian network (BN) to represent interrelated relationships based on investigation of 50 human factor related aviation mishap reports of US Airforce from 2011 to 2019. The causal dependency of PSFs is derived from Human Factor Analysis and Classification System (HFACS) framework. And an Expectation–Maximization algorithm is used to quantify the dependency, which reduces the heavy reliance on expert judgement. Through sensitivity analysis we find that 2 key factors influencing cognitive error are negative state of operator and deteriorating technical condition, implying these factors have greater influence on the cognitive process of operator such as interpreting task demand and information perception. The proposed BN model can be used to identify the primary PSFs influencing pilot performance, providing targeted risk mitigating suggestions.