A novel fatigue and creep-fatigue life prediction model by combining data-driven approach with domain knowledge

A high-precision and concise-form life prediction model is of essential significance to the deterministic life design and reliability assessment of high-temperature components in structural integrity field. In this work, a new model involving only one model parameter is developed to predict fatigue and creep-fatigue life under strain-controlled conditions, during which two physical parameters are extracted by intervening the data-driven symbolic regression results with domain knowledge. A dataset comprising 224 groups of high-temperature low-cycle fatigue and creep-fatigue interaction data for Inconel 718 alloy has been summarized from our prior works to facilitate model development. An additional 125 groups of Inconel 718 data with varying strain ratios and test temperatures are sourced from literature to highlight the advantages of the new model. Model comparisons are conducted with three physics-driven models and four data-driven models. It demonstrates that the new model exhibits superior comprehensive performance according to Bayesian information criterion in comparison to physics-driven models. Moreover, the new model achieves a better extrapolation of life prediction on unseen data set relative to data-driven models. Finally, the data sets of three different materials are collected to further evaluate the universality of the new model in this work.