A Black-Box Time-Discretization Scheme for Time-Dependent Reliability Problems

We propose a time-discretization scheme for time-dependent reliability problems. It takes a general, “black-box” limit-state function (LSF) with a fine temporal mesh and a small number of realizations, and reduces the temporal discretization to a few critical time instants where failure is more likely to take place. No assumption is made about the underlying structure of the auto-covariance function of the LSF as an stochastic process, rather, the correlation structure is empirically estimated and exploited to reduce redundant information. The scheme is a first step in the development of general time-discretization procedures that can be combined with active-learning strategies and surrogate reliability to tackle complex time-dependent reliability problems. Two representative problems are solved, with crude MCS, to illustrate the concept and show the consistency of results.