Predicting low-cycle-fatigue and very high-cycle-fatigue life under creep and oxidation using physics-informed machine learning

This study proposes a physics-informed machine learning (PIML) based approach to predict the combined low and very-high-cycle fatigue (CCF) life considering the effects of creep and oxidation. In order to reveal the combined low and very-high-cycle fatigue failure mechanism, CCF and very high-cycle fatigue (VHCF) tests were performed at room-temperature (RT) and 600°C. The experimental results indicate that the coupling effect between low-cycle fatigue (LCF) and VHCF significantly reduces the fatigue life. During CCF at elevated temperature, creep and oxidation contribute to the failure of GH4169. Based on the CCF behaviour characteristics, a novel normalized damage parameter was established to quantify damage accumulation. Furthermore, Monte Carlo simulation (MCs) was used to overcome data sparsity, and PIML models were developed for VHCF, LCF and CCF life prediction under high temperature. The combined use of MCs and PIML resulted in predictions that lie almost entirely within a scatter band of a factor of three. Experimental validation demonstrates the high accuracy and broad applicability of the proposed prediction model.