Multifactorial prediction of corrosion fatigue crack growth in aluminum alloys using physics-informed neural networks

Abstract

In-service corrosion fatigue cracking in aluminum alloy structural components poses a significant threat to the structural integrity of aircraft, making accurate crack propagation prediction essential for both safety and maintenance planning. Traditional machine learning models, such as Random Forest, Extreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN), rely primarily on data-driven methods and often neglect the underlying physical mechanisms, resulting in reduced prediction accuracy in complex environments. To overcome this limitation, physics-informed neural network (PINN) is used, which integrate physical laws (the Walker crack growth model) with data-driven learning. This hybrid approach effectively captures critical factors that influence crack propagation, such as initial crack length, stress ratio, and environmental conditions (e.g., pH, temperature, and chloride ion concentration). By embedding physical knowledge into the network, PINN significantly improves both the accuracy and generalizability of crack growth prediction. Experimental validation on various aluminum alloys, including 2024, 7075, and LY12, demonstrates that PINN outperforms traditional models, achieving higher prediction accuracy and faster convergence. The study underscores the potential of PINN for crack growth prediction, advancing fatigue life prediction and contributing to improved safety and durability of aircraft components.