Uncertainties and Transformation Principles in PSA

摘要

When evaluating the reliability of systems, e.g. by fault and event tree, usually, classical probabilistic tools are applied. Doing so, one, however, can be faced with sincere difficulties. First, the probabilites may not be exactly known. Second, the dependencies between elements may be unknown (often, statistical independence is assumed what may lead to erroneous results). Third, information on failure probability of components may be available merely in the form of experts' evaluations what often can be described by fuzzy numbers. Hence, there is a need of grasping insufficient information (leading to inexact probabilities) and information from different sources. In the paper we present unifying approaches based on belief and plausibility functions (Dempster-Shafer Theory) enabling the creation of hybrid models.