Low frequency sound absorption metasurface optimization design method based on deep learning

Low-frequency noise has long been a challenge in noise control, and acoustic metasurface has emerged as one of the most viable solutions for low-frequency acoustic control by virtue of its ultra-thin geometry. Acoustic metasurface is generally controlled by several structural parameters. Normally, traditional metasurface design methods typically involve extensive theoretical analysis and numerical simulations to achieve desired performance. This process requires frequent adjustment of structural parameters and a lot of computing resources. In this paper, we introduce a method to rapidly achieve on-demand design of low-frequency acoustic absorbing metasurface using an autoencoder-like neural network (ALNN), which can significantly accelerate the inverse design process after the training of ALNN. We have successfully applied the method to achieve on-demand design of acoustic metasurface in a very low frequency range (50Hz-100Hz) and can optimize the design according to the maximum bandwidth, minimum thickness and minimum weight with limited time. The effectiveness of the method is validated through numerical simulation and experiment. Our work contributes to accelerating the design of metasurface absorbers, especially providing new avenues for low-frequency acoustic absorption applications and research.