Deep Learning Techniques for Design of Locally Resonant Metamaterials

Abstract This study investigates a 2D elastic metamaterial featuring a hexagonal honeycomb unit cell with embedded intracellular resonators. A deep feedforward neural network is trained as a surrogate dynamic model, using synthetic data generated via an orthotropic plate formulation and the plane wave expansion method. The surrogate is then employed to optimize design parameters that maximize the bandgap width while targeting a specified mean frequency. The optimized design is fabricated using 3D printing and experimentally validated through laser scanning vibrometry.