Machine learning assisted probabilistic creep-fatigue damage assessment

In order to investigate the probabilistic damage distribution under creep-fatigue interaction, machine learning framework with the divide-and-conquer methodology is proposed to expand the creep-fatigue life sample size of each load condition. The optimized deterministic life prediction model, strain energy density exhaustion model (SEDE), is selected to take material variability into account. Subsequently, random accumulated creep and fatigue damage are obtained by the combination of probabilistic SEDE life model and creep-fatigue life distributions through the Latin hypercube sampling (LHS) simulation. A relative scatter factor depicted in the creep-fatigue interaction diagram is introduced to reveal the dominance of scatter in creep/fatigue on life scatter. Consequently, a probabilistic creep-fatigue damage assessment diagram with involving probabilistic equipotential line for safety evaluations is established. Such probabilistic damage assessment may provide reference and has promising potential in the further creep-fatigue life design for reliability.