Human Error Classification and Analysis in Nuclear Power Plants

As process and control systems become larger, the man-machine interactions become more complex, causing an increase in the importance of human factors on system function. Fractions of failures resulting from human error in various large systems have varied in percentage from 10 to 60 according to the statistics summarized by Meister(1). Recent statistics on the error causes in commercial reactors showed a high percentage (30%) of errors being due to personnel(2)(3). It is therefore useful to analyze human errors in nuclear power plants and to highlight areas in which improvements could be made to achieve better man-machine interfaces and higher human operational efficiency. One of the key problems in analyzing human error is its classification. There is, however, no definite way to do this. Rouse & Rouse proposed classification of human error from the viewpoint of the human information processing(4). Rasmussen et al. also adopted same point of view to classify errors in nuclear power plants(5). They proposed the human error classification categorized into 18 aspects each of which had about 20 items. On the other hand, Licensee Event Reports (LER's) solely emphasize what occurred and, thus it is difficult to analyze human errors reported in LER's from that viewpoint. Rasmussens' classification is too detailed to be adopted for the analysis ofLER. In this study, classification suitable for human error analysis in LER's is proposed. An approach to extract the points for plant reliability increase from the classified data is also discussed.