Physics-informed: ML and GWO optimization for damping design of laminated structures

Abstract

The design of laminated structures with enhanced damping capacity has become a major challenge in advanced engineering applications, where lightweight construction, vibration mitigation, and structural durability must be achieved simultaneously. In this work, a robust hybrid framework is proposed to predict and optimize the damping behavior of multilayer laminated glass plates by combining refined finite element (FE) modeling, machine learning (ML), and metaheuristic optimization. A refined FE model is first developed for a laminated structure composed of two glass face sheets, a polyvinyl butyral viscoelastic core, and ultra-thin adhesive films, while explicitly accounting for interfacial shear transfer and displacement discontinuities. The numerical formulation is validated through experimental modal analysis, showing good agreement between the predicted and measured modal characteristics. Based on the validated FE simulations, a high-fidelity dataset is generated to train and compare five advanced ensemble regression models, namely random forest, gradient boosting, XGBoost, LightGBM, and CatBoost. The comparative analysis demonstrates that these models accurately capture the nonlinear relationships between the design variables and the damping response, with the best-performing model achieving a coefficient of determination of R 2 = 0.959 on the testing dataset. Furthermore, the selected surrogate model is coupled with the Gray Wolf Optimizer, leading to an optimal modal damping factor of 0.253. To the best of the authors” knowledge, this study is the first to integrate a refined physics-based FE model, experimental validation, ensemble ML, and metaheuristic optimization into a unified framework for damping-oriented design of laminated glass structures incorporating ultra-thin adhesive interlayers. The proposed methodology substantially reduces the computational effort associated with iterative optimization while preserving high predictive accuracy, providing an efficient and reliable tool for the optimal design of vibration-sensitive laminated structures.