ABAQUS-based Simulation and Deep Learning Optimization of the Working Mechanism of Ceramic Granule Recycled Concrete Diaphragm Wall

This paper proposes an integrated optimization framework for ceramic recycled concrete partition walls by combining ABAQUS simulation and a deep learning model. A 3D model was developed in ABAQUS with appropriate boundary conditions and loading steps to simulate real conditions. Performance data from physical tests under various configurations were used to train an enhanced feed-forward neural network. Bayesian optimization was applied to refine design parameters. Results indicate strong agreement between ABAQUS simulations and experimental data, and the deep learning model shows high accuracy in predicting key performance indicators of the partition wall.