Analysis Of Accidents And Terrorist Acts Using A Posteriori Fault Trees

摘要

Fault trees are conventionally used to predict the future probability (a priori) of failure of complex systemsbased on data for failure of lower level components or parts of the system. This lower level failure data isbased on collected data on past failures. In this paper we modify the technique to analyze the events thatled up to an accident or terrorist act which has occurred (a posteriori). In general there is a paucity ofobjective data concerning such events and various subjective techniques must be used to estimate theprobabilities. New techniques are developed to use subjective information for estimating the probabilities.The top-down approach fixes the top event probability at a value close to one and estimates the ratios ofoccurrence probabilities of the lower level events. The bottom-up approach assumes there is some data onlower level events and augments it with a subjective model based on a rectangular or triangular distributionusing a simple minimum variance estimator. This simple estimator is simpler than a Bayesian estimate andis called pseudo-Bayesian.These methods are applied to a simple accident example and a supposed terrorist act.