This article presents a novel concept of deep reinforcement learning (DRL) to facilitate the reverse design of layered phononic crystal (PC) beams with anticipated band structures focusing on the band structure analysis of thermoelastic waves propagating. To this end, we define the reverse design of phononic crystals (PCs) as a game for the DRL agent. To achieve the desired band structure, the DRL agent needs to obtain the topological system of PC. We trained a DRL agent called deep deterministic policy gradient (DDPG). An environment is developed and used to simulate the reverse design of layered PCs with the acquisition of a reward function. The presented reward function encourages the agent to achieve the desired bandgaps. The trained DDPG agent can maximize the game’s score by attaining the desired bandgap. The presented concept allows the user to instantly generate the design parameters through the trained DDPG agent without unnecessary search over the design space. We demonstrated that the DRL agent could perform very well for the automated design of PCs with hundred design cases.

Keywords