A physics-informed data-driven multiaxial fatigue life prediction framework

Multiaxial fatigue-life prediction remains one of the most critical challenges in fatigue research. This study introduces a physics-informed and data-driven framework that maps arbitrary multiaxial load paths onto fatigue life through a pre-packaged damage kernel; every model constant is obtainable from standard fatigue-life equations. Instead of relying on prescribed non-proportional histories, the framework exploits any non-proportional test to calibrate the incremental damage constants by optimizing. The approach is validated against 754 data points extracted from 15 datasets (4 stress-controlled smooth-specimen series, 7 strain-controlled smooth-specimen series and 4 stress-controlled welded-joint series) encompassing 20 distinct load paths. Overall, 98.41 % of the predictions fall within the ±3 × scatter band, confirming the accuracy and versatility of the proposed model.