A Bayesian model to treat within-category and crew-to-crew variability in simulator data for Human Reliability Analysis

The models adopted in Human Reliability Analysis (HRA) characterize personnel tasks and performance conditions via categories of task and influencing factors (e.g. task types and Performance Shaping Factors, PSF). These categories cover the variability of the operational tasks and conditions affecting performance, and of the associated Human Error Probability (HEP). However, variability exists as well within such categories, for example because of the different scenarios and plants in which data is collected, as well as of the operating crew differences (within-category and crew-to-crew variability). This paper presents a Bayesian model to mathematically aggregate simulator data to estimate failure probabilities, explicitly accounting for the specific tasks, scenarios, plants and crew behavior variability, within a given constellation (i.e. combination) of task and factor categories. The general aim of the proposed work is to provide future HRA with reference data with stronger empirical basis for failure probability values, both for their nominal values as well as for their variability and uncertainty. Numerical applications with both artificially-generated data and real simulator data are provided to demonstrate the effects of modelling variability in HEP estimates, to avoid potential overconfidence and biases. The applicability of the proposed model to ongoing simulator data collection programs is also investigated.

Automated summary

This paper presents a Bayesian model to mathematically aggregate simulator data to estimate failure probabilities, explicitly accounting for the specific tasks, scenarios, plants and crew behavior variability within a given constellation of task and factor categories. The aim of the proposed work is to provide future HRA with reference data with stronger empirical basis for failure probability values.