Study on the Mechanism and Interpretability of Defect Features on Fatigue Damage in 6061 Aluminum Alloy

In order to obtain a damage assessment method that can clearly express the influence mechanism of defect characteristics on fatigue damage, an integrated analytical framework combining the Shapley Additivity Interpretation (SHAP) method with the Extreme Gradient Boosting (XGBoost) model is established. Based on this framework, a high-accuracy fatigue damage prediction model was first established using the XGBoost model. Its accuracy and reliability are rigorously evaluated using the coefficient of determination R2 and root mean square error RMSE. Results demonstrate that the model performs exceptionally well on both the training and test datasets (R2 values of 0.999 and 0.846, respectively, with RMSE values of 0.0136 and 0.3727), establishing a reliable foundation for subsequent damage mechanism interpretation. Furthermore, by comparing defect feature importance between the XGBoost model and the SHAP interpretability model, it is revealed and cross-validated that the dominance order of defect features affecting the stress concentration factor k is Smax > P > Sarea > N > Vmax, confirming the physical stability of this dominance relationship. Additionally, leveraging the powerful local interpretability of the SHAP method, the contribution of each defect feature to the value in any sample was quantitatively analyzed, establishing a mathematical relationship between defect feature variables and the stress concentration factor k. Ultimately, based on the results of interpretability analysis and guided by fracture mechanics theory, the damage variable D is constructed by coupling multiple defect characteristics. This approach quantitatively reveals the intrinsic relationship between this variable and material fatigue damage. It provides a scientific basis for maintenance and repair decisions of aluminum alloy components in engineering applications, significantly enhancing their safety and reliability during use.