Application of Causal Human Reliability Operator-Centered Models to Maintenance Errors

SUMMARY & CONCLUSIONSHuman reliability analysis (HRA) seeks to identify, characterize, and quantify potential errors in complex engineering systems. The applications of HRA span many industries, including nuclear power generation, maritime, aviation and space, and petrochemical. Analysts leverage HRA to predict and mitigate potential catastrophic human failure events (HFEs) in these systems. However, very little HRA work has been conducted on maintenance actions, although they contribute to a large portion of human-related errors. It is essential to examine maintenance tasks as part of a thorough analysis as they present myriad opportunities for difficult-to-detect latent errors.Our previous work developed CHROMO, a set of Bayesian network (BN) models that document all major dependencies between performance influencing factors (PIFs), human failure mechanisms, and crew failure modes (CFMs). The models are informed with reliability engineering data and literature, capable of providing multipurpose insights, and can be flexibly updated with further data, which satisfy the ideal qualities that a third-generation HRA method should embody.In this paper, we demonstrate the application of the CHROMO models to a maintenance-related adverse event in the maritime shipping industry: the 2024 engine room fire aboard the cargo vessel Stride. The event was the unfortunate culmination of several HFEs during refueling maintenance operations. To apply the CHROMO models to this scenario, we set evidence of the PIFs in the BNs and updated the model to receive posterior probabilities about the likelihood of the relevant CFMs. This gives insight into the most likely causes of human error, based on the conditions present during the event.The results show that it is possible to model relevant contextual factors of the scenario and obtain error probabilities using CHROMO; including this as evidence results in a significant increase in the probabilities of each CFM occurring, an effective increase from the order of 10-2to 10-1. Additional related CFMs were identified and observed to significantly increase in likelihood, reflecting other potential progressions of events. Some factors had larger effects than others; a lacking human-system interface was the largest contributing factor to the information-gathering failure and poor information use behavior was the largest contributor to the decision-making failures. We would recommend mitigation of these PIFs to prevent future similar occurrences.