Physics-informed learning of fatigue crack growth in corroded steel: A Paris-law extension with corrosion degree and stress ratio optimized by Bayesian tuning

Fatigue degradation of corroded steels presents complex interactions between mechanical loading, stress ratio, and corrosion-induced defects, which are difficult to capture using empirical models. A Physics-Informed Neural Network (PINN) framework is developed, that embeds the residual form of Paris law into its loss function, enabling the prediction of fatigue crack growth rates in chloride-corroded Q690 steel. Then, A dimensionally consistent extension of the Paris equation is proposed by incorporating the corrosion degree (P) and stress ratio (R), calibrated through single-edge notched tension (SENT) tests. Furthermore, Bayesian Optimization (BO) is employed to automatically tune key hyperparameters of the PINN, improving convergence stability and predictive accuracy. The proposed model achieves superior agreement with experimental data compared with conventional Paris-type formulations and data-driven baselines, demonstrating the feasibility of combining physics-informed learning with corrosion-aware modeling for reliable fatigue life prediction in structural steels.