Methodical treatment of dependent failures in risk analyses

In this report the state-of-the-art regarding dependent failures is compiled and commented on. Among others the following recommendations are infered: The term 'common mode failures' should be restricted to failures of redundant, similar components; the generic term is 'dependent failures' with the subsets 'causal failures' and 'common cause failures'. In risk studies, dependent failures should be covered as far as possible by 'explicit methods'. Nevertheless an uncovered rest remains, which should be accounted for by sensitivity analyses using 'implicit methods'. For this the homogeneous Marshall-Olkin model is recommended. Because the available reports on operating experiences only record 'common mode failures' systematically, it is recommended to additionally apply other methods, e.g. carry out a 'precursor study'. (orig.)