Probabilistic modeling of unified S-N curves for mechanical parts

Abstract

This article aims at modeling unified S-N curves based on multifailure modes and stress ratios. Heterogeneous mixture distributions are used to describe life distribution of multifailure modes. Bayes clustering is applied to cluster failure data to avoid convergence to global optima. Derived from regression analysis, failure data are compensated by nonrandom error. The relationship between distribution parameters and stress is established, and the ultimately unified probabilistic S-N curve is modeled. Two assumptions are presented to modify an original S-N curve model with varied stress ratios. The first one is to take into account the effects of low-amplitude load “exercising,” which we call results in the “pseudo-fatigue damage.” The other one is to set up probability density function of fatigue limit considering uncertainty. Combined with material parameters modification, thus a new modified and unified S-N curves model is available for not only researchers, but engineers as well.