Prior knowledge informed neural network based on Basquin’s formula for predicting fatigue life

This study introduces a prior knowledge-informed neural network (PKiNN) for predicting material fatigue life. The prior knowledge module generates data adhering to S-N curve characteristics, while a pure data-driven module assesses factors like stress concentration from notches. By merging these outputs through a single-layer fully connected neural network, fatigue life predictions are obtained. Experimental validation confirmed PKiNN’s superior extrapolation capabilities compared to traditional machine learning (ML) approaches. Moreover, PKiNN’s predictive performance is less dependent on specific training data set choices, distinguishing it from current informed ML methods.