Data-Driven, Physics-Based, or Both: Fatigue Prediction of Structural Adhesive Joints by Artificial Intelligence

Here, a comparative investigation of data-driven, physics-based, and hybrid models for the fatigue lifetime prediction structural adhesive joints in terms complexity implementation, sensitivity to data size, accuracy is presented. Four data-driven (DDM) are constructed using extremely randomized trees (ERT), eXtreme gradient boosting (XGB), LightGBM (LGBM) histogram-based (HGB). The physics-based model (PBM) relies on Findley’s critical plane approach. Two (HM) were developed by combining approaches obtained from invariant stresses (HM-I) stress (HM-F). A dataset 979 points four adhesives employed. To assess split into three train/test ratios, namely 70%/30%, 50%/50%, 30%/70%. Results revealed that DDMs more accurate, but sensitive size compared PBM. Among different regressors, LGBM presented best performance generalization power. HMs increased predictions, whilst reducing size. HM-I demonstrated datasets sources can be utilized improve predictions (especially with small datasets). Finally, showed highest an improved