A novel fatigue life prediction approach of a GH4169 superalloy-welded joint based on a physics-informed machine learning method

Defects were usually inevitable during welding process, the equivalent size of the maximum initial welded defects that perpendicular to the loading direction was usually introduced as the initial crack, whereas, the influence of morphology feature of the defects could not be well considered, and the fatigue life prediction issue of welded joint was usually challengeable according to the dispersion of the morphology, size, location, and quantity of the welding defects. Therefore, a physics-informed machine learning approach was constructed in order to captured the action mechanism of morphology features of welding defects in this study, the introduction of the additional physics information not only extended the initial training datasets, but also enhanced the interpretability of the lifetime prediction results, influence of the morphology detail of the defects was well considered through a modified physics fatigue prediction model. The final fatigue life prediction results revealed that physics-informed long short-term memory network approach was the best one compared with physics-informed convolutional neural network and the physics-informed random forest method, which exhibited the highest coefficient of determination and the most robust generalization ability.