Speaker
Description
andgap materials play a critical role in controlling wave propagation for applications in vibration isolation and acoustic filtering. Traditional topology optimization methods can effectively maximize bandgaps but are computationally expensive and limited in scalability. Data-driven approaches, including generative models and supervised learning, accelerate design discovery but remain constrained by the distribution of the training data and often lack strong generalization to unseen design environments. In this work, we propose a discrete soft actor–critic (SAC)-based deep reinforcement learning framework to autonomously design elastic structures with broadband bandgaps. The agent is first trained to optimize the bandgap in a 4 cm × 4 cm design space and subsequently evaluated in a smaller 3 cm × 3 cm environment to assess its generalization capability. The results demonstrate that the proposed reinforcement learning framework successfully discovers structures withbroadband bandgaps and generalizes to unseen design domains without additional training. Full-wave harmonic simulations confirm that the bandgap pre-dictions obtained from unit-cell analysis are consistent with the frequency response of the corresponding periodic structures. Compared with conventional generative modeling approaches, the proposed method learns the underlying relationship between geometry and bandgap formation through environment interac-tion, enabling improved versatility and physics-aware design discovery.