Per-State Signatures and Mean Time-to-Failure: An Algorithm for Systems With Simultaneous Failures

Abstract

We present an algorithm that computes the structural signature and the mean time-to-failure (MTTF) of a monotone binary system at every one of its states, rather than only at the state in which all components work. It makes no assumption on the structure function and evaluates it exactly once per state, as many times as needed for the signature of the complete system alone, so computing the MTTF everywhere costs only a factor polynomial in the number of components over computing it once. As a by-product, it identifies all minimal path sets. We assume exchangeable component failure times whose survivor count is a continuous-time Markov chain. This setting admits simultaneous failures, as in the Lévy-frailty Marshall-Olkin, binomial failure rate and lethal-shock models, where the simplifications available for single failures do not apply. We use the per-state MTTF to define threshold repair policies. We show that the per-state signature is a strictly finer decision criterion than the MTTF: signature dominance at a common failure count implies MTTF dominance, but not conversely. Under the Lévy-frailty Marshall-Olkin model the MTTF is monotone in the state, which yields the long-run cost rate of every threshold policy in a single pass. Experiments on 22 benchmark networks with up to 30 components and 31 failure models show that the best MTTF-threshold policy has a median excess cost of 0.36% over the optimal state-based policy, and that it is never worse than the best policy that repairs after a fixed number of failures.