A PINN–XGBoost Hybrid Model for High‐Cycle Fatigue Life Prediction of Mg Alloy Welded Joints

Mg alloy welded joints are increasingly used in lightweight structures, yet accurate prediction of their high‐cycle fatigue life remains challenging, particularly under small‐sample conditions. This study proposes a physics‐informed PINN–XGBoost hybrid framework for fatigue life prediction. Physics‐guided constraints derived from the Basquin relation and Goodman mean stress correction are incorporated into the PINN loss function to guide the model toward physically consistent fatigue trends under small‐sample conditions. A static fusion model (Hybrid V1) and a dynamic physics‐guided fusion model (Hybrid V2) are further developed to integrate PINN and XGBoost predictions. The proposed Hybrid V2 employs a learnable weighting network to enable sample‐wise adaptive fusion. Results demonstrate that Hybrid V2 achieves superior prediction stability, generalization capability, and physical consistency across different fatigue regimes. Robustness is validated through different train–test splits and independent experimental fatigue data.