Quantitative Evaluation of Human Reliability for Operating Tasks in Nuclear Power Plants

Human reliability can be computed quantitatively provided that the operational data are put in the proper format and several information matrices are available. Both one dimensional and multidimensional trends can be easily determined by counting the number of operational errors that belong to one or more specific categories. Two examples are given for one dimensional trends in the operational tasks in 65 nuclear plants over a period of ten years. In the first example, the classification of events is made according to the cue used in error discovery. The results show that observation of unannunciated displays represents 41% of the total number of discovery cues. The second example shows that by classification of events according to the human failure mode, 50% of the total operation errors are due to omission. Recommendations and restrictions are given in detail. Assuming a constant failure rate (w.r.t. time) and an exponential human reliability model, failure rates (per hour and per demand of frequency of use) for valve mispositioning in two nuclear systems are computed. Comparison of the results obtained with other estimates of human error probabilities is made. Also, the results show that the unavailability of valves in those systems where human errors involved mispositioning is not significantly greater than that due to mechanical causes.