Statistical Evaluation of Design-Error Related Nuclear Reactor Accidents

Abstract

A general methodology for the statistical evaluation of design-error related accidents is proposed that can be applied to a variety of systems that evolves during the development of large-scale technologies. The evaluation aims at an estimate of the combined “residual” frequency of yet unknown types of accidents “lurking” in a certain technological system. A special categorization of incidents and accidents is introduced to define the events that should be jointly analyzed. The resulting formalism is applied to the development of U.S. nuclear power reactor technology, considering serious accidents (category 2 events) that involved, in the accident progression, a particular design inadequacy. The analysis of the five events in that category indicates a drastic reduction (by more than a factor 50) in the combined residual frequency between the Experimental Breeder Reactor I and the Three Mile Island 2 accidents.