A new method for evaluating system-level importance measures based on singular value decomposition

Abstract A new method is described for evaluating probabilistic importance measures of a system or function that is a naturally defined cluster of components and that is intermediate in a hierarchy of levels of performance measures in a nuclear power plant. It is based on the singular value decomposition (SVD) in linear algebra. For the sample problem, the importance ranking of the systems determined by the SVD importance measure is different from those obtained by existing methods.