Abstract
Phononic crystal (PnC) topology optimization for wide and low-frequency bandgaps is computationally demanding due to repeated finite element method (FEM) eigenvalue analyses. In this work, a surrogate-assisted optimization framework combining a convolutional neural network (CNN) with simulated annealing (SA) is developed to accelerate two-dimensional PnC design. The CNN surrogate directly predicts dispersion relations from discretized unit-cell configurations, enabling efficient bandgap evaluation without repeated FEM calculations during optimization.Using this framework, multiple PnC structures are independently optimized, each targeting a specific bandgap order. The surrogate-assisted optimization achieves optimized topologies consistent with FEM-based SA results while reducing the computational time per optimization run from several thousand minutes to approximately one minute. The surrogate model further demonstrates good generalization capability across different filling ratios and structural configurations. Overall, the proposed CNN–SA framework provides an efficient and scalable approach for inverse design of phononic crystals.

This publication has 20 references indexed in Scilit: