Bayesian Optimization–Based Physics‐Informed Neural Network for Adaptive Prediction of Multiaxial Fatigue Life in Metallic Materials

ABSTRACT Multiaxial fatigue life prediction holds significant application value in engineering fields, including aerospace, energy, and the automotive industries. However, traditional fatigue life prediction methods often encounter limitations when dealing with complex loading paths and material behaviors. To address this, the present study proposes a fatigue life prediction method that integrates Bayesian optimization with physics‐informed neural networks (BO‐PINN). This method incorporates the critical plane model as a physical constraint and dynamically adjusts the physical constraint weight via Bayesian optimization, thus achieving an optimal balance between data fitting and physical consistency. This significantly enhances the model's prediction accuracy and robustness under complex multiaxial loading conditions. The method is validated through experimental data from three distinct materials. The results demonstrate that BO‐PINN outperforms MLP and traditional PINN models in terms of prediction accuracy, stability, and cross‐material generalization ability, particularly under nonproportional loading and complex path conditions. It also effectively mitigates prediction uncertainty in small datasets. Moreover, BO‐PINN offers probabilistic predictions with confidence intervals, addressing the gap in uncertainty quantification that traditional methods fail to resolve.