A data-physics integrated approach to life prediction in very high cycle fatigue regime

Abstract

The defects created in metallurgical and manufacturing processes generally play a decisive role in very high cycle fatigue life of engineering structures. By taking stress level, defect size and location into account, the physical Z-parameter model on fatigue life prediction was combined with artificial neural network as a new data-physics integrated approach for fatigue life prediction of 15Cr and FV520B-I steels in this work. The original data from tests were expanded based on the Z-parameter model, and the physics-informed loss function featuring Z-parameter was integrated into artificial neural network as the constraint. Results showed that the physics-informed neural network established in this work could be applied for life prediction in the very high cycle fatigue regime, and the model came with higher predictive accuracy than the physical Z-parameter model and the Mayer’s model did.