Development of a human reliability database. Proposition of human reliability data collection and updating of human error probability

The nominal human error probabilities (N-HEPs) used in the current human reliability analysis (HRA) are mostly based on European & U.S. experts' judgement or on human reliability data from petrochemical industries, etc. As it doesn't seem to be realistic to use the N-HEPs for the HRA conducted in Japanese nuclear industry, we propose a method of collecting human reliability data through simulator training at each nuclear power plant and updating existing N-HEPs based on this data. For estimating N-HEPs, it is required to determine the number of human errors for each type of error mode, and the number of trials including success during simulator training. The latter one can be obtained from the HRA task analysis results that assume errors when a certain task is performed. If the tasks for which HRA has been conducted are realistically performed in simulator training, the number of errors occurred during the training can be counted from the record and observations. However, the observation HEPs calculated with the data through the training are probability values under the condition with stress and recovery from error. Since the existing N-HEPs are values without effects of these factors, the converted HEPs used to update the N-HEP should be calculated backward from the observation HEP, using the HEP equation to account for the effects of stress and recovery. Finally, the N-HEPs can be updated even with a small amount of collected data obtained from the converted HEP, by using Bayesian update with the prior distribution of HEP parameters of the existing HRA method. (author)