Human reliability in probabilistic risk assessment: A retrospective study

The aim of this report is to describe methods used in the human reliability analyses of five probabilistic risk assessments and the results achieved. The studies chosen are the German Risk Study, the Seabrook Study, the Oconee Study, the Calvert Cliffs Study and the Sizewell B Study. The studies are compared in respect of their human actions analysis and some general conclusions drawn. The main areas of the study are: scopes and objectives of the human action analyses, identification and classification of human errors, treatment of maintenance versus operational activities, impact of recovery actions, inclusion of psychological factors, human actions modelling, data sources used and impact of human activities on the core melt frequency. In addition, the uncertainty analyses connected with human reliability are reviewed. Similar and diverse ways to analyse human reliability are illustrated in several tables and figures. As the conclusion, it became evident that the quality of human actions analysis documentation varies significantly. Also the extent, to which the commission type human errors causing easily hazardous plant transient are taken into account, is usually low. The models utilized are mostly simple logical models suitable for calculation but hardly for detailled modelling purposes. Data for human reliability estimates is mainly taken from subjective sources. This procedure does not, however, explain all the differences in the results, but the degree of the inclusion of recovery has here an important role. Human actions have the key role in every study, in which they are addressed extensively, which is also reflected in the results.