Methodology for analysis of operator errors of commission during nuclear power plant accidents with application to probabilistic risk assessment.

摘要

Current Human Reliability Analysis has been criticized in recent years for treating operator error in a simple, behavioral fashion. The models used to assess human error rate do not adequately address the cause of operator error and do not account for dependencies and coupling factors that influence operator behavior. These models do not formally analyze contributions from emergency operating procedures (EOPs). Recent efforts have led to artificial intelligence based models that are still under development. These are too complex and model unconstrained operator decision making. Their developers have made little effort to effectively integrate the models with the probabilistic framework.; This dissertation describes a realistic, yet implementable modeling approach to simulate operator errors of commission (EOCs). The presentation includes a modeling section and an implementation section. The modeling section contains the bases for error, and error classifications and definitions. The bases were identified as the plant need and procedural instruction. Errors are identified at the level of expression, and traced through causal chains to identify possible causes. Classifications included for error expressions as well as diagnosis, since diagnosis is an important error cause.; The modeling strategy employed to model the errors consists of identification of performance influencing factors (PIFs) and development of generic mapping rules to map PIF on errors. Dependencies among actions at different times are accounted for by operator-related PIFs, and used to develop operator action sequences. PIFs related to plant, EOPs are also identified.; The methodology to incorporate EOCs in PRAs uses results from event trees (ETs) to generate Human Interaction Time{dollar}underline{lcub}rm LINES{rcub}{dollar} (HITLINEs). Combination of ET sequences, instrument failure, and time-independent PIFs are screened to select important ones for further analysis. HITLINEs are then constructed for each combination. Mapping rules are used to generate sets of PIFs given appropriate inputs, and to generate errors given these PIFs. The forward branching structure of HITLINEs is managed by merging, truncation, and termination. Similar end states in terms of hardware status on the ET, are combined within and among HITLINEs. The results thus obtained are used to update ETs by reordering sequences or including additional top events.