Integration of symbolic regression and domain knowledge for interpretable modeling of remaining fatigue life under multistep loading

Abstract

This research work aims to explore the integration of data-driven symbolic regression (SR) and domain knowledge to model the remaining fatigue life under multistep loading. To this end, six classical semiempirical damage models are analyzed to distill reliable domain knowledge as the restrictions on the structures of SR formulas. Meanwhile, a total of 194 experimental results involving fifteen materials and structures as well as three kinds of loading spectrums are collected for data support. As a major contribution of this research work, a novel model without including fitting parameters is successfully discovered for remaining fatigue life estimation under two-step loading. This model can be interpreted in the framework of conventional damage models, and shows good extendibility to multistep loading through proper definitions of the damage indicator and the damage transition. Extensive model evaluations demonstrate that the discovered model is better than five existing damage models in terms of predictive accuracy and application scope, showing great applicability for remaining life estimation under multistep loading.