A METHOD FOR INFERRING CONDITIONAL STOCHASTIC FAILURE RATES FROM THE TIME-HISTORY OF OBSERVED FAILURES

摘要

Observed failures, rather than first principles, are used to estimate fatigue rates probabilistically conditioned on operating conditions. The method developed assumes that a normal random variable may be used to approximate the damage limit (remaining lifetime) of components subjected to cumulative damage and that when a component fails, its damage limit has vanished at a rate proportional to the amount of time spent at each operating condition experienced during its lifetime. By considering differences in cumulative damage between pairs of failed components, we obtain the relative rates at which damage is accumulated for each observed operating condition. When the differences in component lifetimes are dominated by variations in experienced conditions, it is possible to estimate absolute rates. Otherwise, variations in initial damage limits dominate and it is only possible to estimate the mean and variance of this distribution. We demonstrate the procedure on synthetic data, including a test for the dominant source of lifetime variations.