A fatigue life prediction framework for CFRP/Al hybrid (riveted/bonded) joints and bonded joints under hygrothermal-load conditions based on physics-guided machine learning

The widespread use of lightweight materials has presented challenges in dissimilar material joining technologies. Hybrid (riveted/bonded) joints (HJs) and bonded joints (BJs), common in carbon fiber reinforced polymer/aluminum alloy (CFRP/Al) structures, exhibit favorable fatigue performance. However, the coupling of hygrothermal conditions and mechanical loads complicates fatigue failure mechanisms. Fatigue life prediction remains difficult due to limited experimental data, multiphysics coupling, and pronounced nonlinearity. To address these issues, this study proposes a physics-guided machine learning (PGML) framework. A hygrothermal fatigue experiment was conducted to construct a dataset and to develop a nonlinear damage accumulation model that accounts for both hygrothermal and load effects. Guided by this model, a residual neural network was designed to correct systematic prediction errors, combining the strengths of physics and machine learning. The approach achieved an R² of 0.97, with most predictions falling within 2 times error band. Sensitivity analysis revealed feature contributions across different structures and indicated directions for improving the physical model. The PGML framework developed in this study enables efficient and accurate fatigue life prediction for CFRP/Al joints under hygrothermal-load coupling conditions and provides an extensible methodology for related fatigue life prediction problems.