A Novel Surrogate Model for Predicting Drilling-Affected Stress State and Bolt Insertion Damage of Carbon Fiber-Reinforced Polymer/Titanium Based on Physics-Informed Neural Network

In aerospace manufacturing, the performance of Carbon Fiber Reinforced Polymer (CFRP)/Titanium (Ti) interference-fit bolted joints depends critically on two key processes: drilling and bolted joining. There is a compelling need to analyze the stress and damage state across these multi-stage processes to enhance the mechanical performance of the laminated structure. This paper proposed a novel surrogate model based on Physics-Informed Neural Network (PINN) to predict the drilling-affected stress state and bolt insertion damage. The model integrates the drilling-induced damage factor at the entrance/exit of CFRP and the Lekhnitskii radial extrusion pressure as feature variables. The predicted stress and displacement are constrained by constitutive and geometric equations derived from Classical Lamination Theory embedded into the loss function. The proposed PINN is demonstrated to have the ability to effectively predict the stress, displacement, and bolt insertion damage in the CFRP overlay region under varying drilling-induced damage conditions, bolt shank diameters, and interference-fit amounts. Results indicate that the average accuracy of predicting the stress state of CFRP could reach 94%, improved by 70.4% compared with the Finite Element model (FEM). The bolt insertion damages of PINN are consistent with the optical image of the damaged region and the average predictive accuracy of the damaged area at the entrance/exit of CFRP could reach 92% and 91%, respectively, improved by 59% and 91% compared with FEM. The model provides solid support for assembly process design and contributes significantly to improving assembly efficiency.