Physics-data fusion driven model for fatigue life prediction of titanium alloys under proportional and nonproportional loading

Abstract

Traditional physical models and purely data-driven approaches often face significant challenges in predicting multiaxial fatigue life of titanium alloys due to small sample sizes and the complex nonproportional hardening effects under various loading paths. To overcome these challenges, this study proposes a physics-data fusion framework that integrates the critical-plane-based physical model with a support vector regression (SVR) error compensation module to improve fatigue life predictions for titanium alloy. Validated against multiaxial fatigue test data of TC4 titanium alloy, all predictions yielded by the proposed fusion model fall within the two-fold scatter band. The coefficients of determination for the training and test sets are 0.9905 and 0.9620, with mean squared errors of 0.0038 and 0.0137, respectively. These results are substantially superior to those of either the pure physical model or the pure SVR model, confirming that the proposed approach offers a robust and practical tool for accurate multiaxial fatigue assessment of titanium alloys.