Physics-informed explainable artificial intelligence framework for corrosion-fatigue life prediction, reliability assessment, and inspection planning of welded steel structures

Atmospheric corrosion combined with cyclic loading is one of the most critical degradation mechanisms affecting the fatigue life of welded steel structures. Although numerous machine learning models have demonstrated the capability to accurately describe crack propagation processes, the direct application of these models to probabilistic analysis and life-cycle management often requires considerable computational effort. This study proposes a Physics-Informed Explainable Artificial Intelligence (XAI) Framework for corrosion-fatigue life prediction, reliability assessment, and inspection planning of welded steel structures, aiming to improve prediction accuracy while supporting engineering decision-making throughout the corrosion-fatigue process. The proposed framework is developed based on the Newman-Raju crack propagation model combined with the Paris-Walker law and a corrosion-induced thickness reduction model. Latin Hypercube Sampling is employed to generate a database consisting of 9,571 simulation samples. Subsequently, the dataset is divided into training and testing subsets, with 7,656 samples used for training and 1,915 samples reserved for independent testing, while the XGBoost algorithm is adopted to predict corrosion-fatigue life and SHAP is utilized to interpret the influence of input variables. The results indicate that the proposed model achieves a coefficient of determination of 𝑅 2 = 0 . 9 7 5 7 5 on the independent testing dataset, while five-fold cross-validation gives a mean 𝑅 2 = 0 . 9 7 6 4 0 . The corresponding MAPE is 3.49%. SHAP analysis identifies the stress concentration factor (SCF), corrosion acceleration factor ( 𝜂 𝑐 𝑜 𝑟 𝑟 ), and initial crack depth ( 𝑎 ⁢ ₀ ) as the dominant variables affecting corrosion-fatigue life, consistent with their respective roles in local stress amplification, environmental crack-growth acceleration, and initial crack severity within the underlying fracture-mechanics formulation.