Design Of 2D Phononic Crystals Via Deep Learning

Phononic crystals are artificial materials with a periodic structure that have the property of completely blocking or significantly attenuating the propagation of sound waves or elastic waves,known as bandgap.Due to the vibration reduction properties of bandgap,phononic crystals have a wide range of applications in engineering vibration reduction,noise reduction,and sound wave control.Traditional phononic crystal design methods usually rely on repeated geometric adjustments and iterative simulations to gradually approach the target bandgap.This process is not only computationally expensive,but also relies on the experience or intuition of the designer,and cannot achieve the design of phononic crystal configurations from a specific range of bandgap requirements.Combining deep learning technology with phononic crystal configuration design can solve this problem well.This article focuses on the application of deep learning techniques in the design of phononic crystals.The goal is to use an optimized cWGAN-gp model to reverse-design the configuration of phononic crystals,and to validate the generated configurations using finite element analysis.The article first introduces the theoretical basis of deep learning models,including the structure of neural networks,training,and types of deep learning models.Then,the article details a method for collecting a two-dimensional phononic crystal sample set,including the collection of configuration and band gaps,and performs forward prediction.Subsequently,the article proposes a conditional Wasserstein GAN(cWGAN-gp)model with gradient penalty to reverse-design the configuration of phononic crystals,and validates the model using finite element analysis.Finally,the article improves the cWGAN-gp model by using a variational autoencoder to improve the generator network and using residual connections to optimize network complexity,in order to improve its training speed and generation performance.The accuracy of the phononic crystal configurations generated by the deep learning model is verified by comparing the band gaps calculated by finite element analysis of the generated configurations with the label band gaps.This study provides a new approach and method for the design of phononic crystals,and also provides a new way of thinking and practical basis for the application of deep learning in the field of materials.