Dependency analysis in reliability studies

The reliability of engineering systems, especially safety systems, is acknowledged to be significantly affected by the dependency of the components and subsystems. The factors that cause the dependency range from the design of the system to the maintenance procedures employed. Assessing the effect of dependency has been a concern in both reliability theory and statistics. Most interest has centred on common-cause, or common-mode, failure but other forms, such as time between failures, can play a major role. The engineering and statistical approaches to dependency have tended to diverge, and misconceptions in both assessment and modelling have arisen. This paper constructively reviews the varying approaches employed, and indicates future benefits that might accrue by the proper use of statistical methodologies. In this exploration, the pitfalls associated with some of the techniques are highlighted. One of the major problems encountered in assessing dependency is that the information used for assessment comes from maintenance records or incident reports which are not collected specifically for dependency studies. Other aspects are the naivety of some of the models employed, which include cut-off assessment, the beta-factor, and multiple-parameter models. Another feature, which seems to be overlooked by practitioners, is failures that are time-dependent. Complementary to these practical approaches, there are a number of statistical methods that could be applied. In the recent literature, there has been considerable work on developing multivariate models. There are also a variety of stochastic processes that might be employed. The paper examines how such models can be practically applied to the data available, to produce a more comprehensive understanding of dependency in the field of reliability. Particular emphasis is given to methods for identification of dependency between components and systems, especially those capable of detecting time-dependent failures.