Data-driven surrogates via regression and CNN-LSTM models with metaheuristic optimization for frequency control in nonlocal porous laminated composite plate resting on elastic foundation

The study presents a comprehensive machine learning-based surrogate modelling and optimization framework to minimize the non-dimensional frequency of nonlocal porous laminated composite plates resting on elastic foundations. A data is generated using a Higher Order Shear Deformation Theory based on finite element method incorporating nonlocal elasticity, porosity, and Pasternak elastic foundation effects. Five regression models Fine Tree, Medium Gaussian SVM, Bagged Trees, Matern Gaussian Process Regression and Wide Neural Network were trained and evaluated. Among them, the Bagged Trees model achieved the highest accuracy with an R2 of 0.998. In parallel, a hybrid CNN-LSTM model is used to capture hierarchical local features and complex global dependencies within structured data and regression, yielding an R2 of 97.8%. The Rao-3 metaheuristic algorithm is employed to minimize the NDF by optimizing input parameters through the trained surrogate models and GA optimisation is also employed for side-by-side comparison. The surrogate-based framework demonstrates a robust integration of data-driven modelling and optimization for efficient prediction with in the data and control of vibrational performance in advanced composite structures.