On-Demand Inverse Design of Metamaterials Using Deep Neural Networks with Bayesian Optimization

08/19/2025

Recent advances in metamaterial design highlight methodologies tailoring metamaterials to achieve target behaviors in specific applications. This investigation presents a data-driven design approach using deep neural networks for the on-demand inverse design of metamaterials to address limitations inherent in traditional methods. We propose an efficient Bayesian optimization framework for deep neural network hyperparameter optimization.

The desired properties being the target bandgap width and bandgap midfrequency, the design model proposes candidate unit cell topologies. The methodology demonstrates exceptional accuracy in both the forward prediction and inverse design of 2-dimensional metamaterial structures, with particular emphasis on bandgap characteristics. Statistical analysis reveals R² coefficients exceeding 0.99, validating the model’s predictive capabilities. The demonstrated framework represents a substantial advancement in computational metamaterial design, offering potential applications across multiple materials science and engineering domains.

Authors:

Original publication:

Than V. Tran, S.S. Nanthakumar, Timon Rabczuk, Xiaoying Zhuang. On-Demand Inverse Design of Metamaterials Using Deep Neural Networks with Bayesian Optimization. IINTELLIGENT COMPUTING, 2025. DOI: 10.34133/icomputing.0139

Published by Team CSST

Go to list