Probability of failure of safety-critical systems subject to partial tests

Abstract

A set of general formulas is proposed for the probability of failure on demand (PFD) assessment of MooN architecture (i.e. k-out-of-n) systems subject to proof tests. The proof tests can be partial or full. The partial tests (e.g. visual inspections, partial stroke testing) are able to detect only some system failures and leave the others latent, whereas the full tests refer to overhauls which restore the system to an as good as new condition. Partial tests may occur at different time instants (periodic or not), up to the full test. The system performances which are investigated are the system availability according to time, the PFD average in each partial test time interval, and the total PFD average calculated on the full test time interval. Following the given expressions, parameter estimations are proposed to assess the system failure rates and the partial test effectiveness according to feedback data from previous test policies. Subsequently, an optimization of the partial test strategy is presented. In the 2oo6 system given as example, an improvement of about 10% of the total PFD average has been obtained, just by a better (non-periodic) distribution of the same number of partial tests, in the full test time interval.