科研之友
微信
新浪微博
Facebook
分享链接
摘要
Inversely designing and optimizing topological structures of phononic crystals that dominate extraordinary wave characteristics has become a research hotspot. In this study, a joint framework combining a data-driven deep learning model with the semi-analytical two-dimensional periodic spectral finite element method is applied to achieve the inverse design and optimization of topology. A convolutional neural network and a generative adversarial network are trained for inverse design. Meanwhile, the semi-analytical periodic approach is utilized to analyze the wave characteristics of two-dimensional phononic crystals. Through the proactive tuning of the band structures, the topologies of phononic crystals are inversely on-demand designed and optimized for the anticipated partial bandgap or complete bandgap, respectively, which revealed the unidirectional wave transmission or vibration isolation characteristics within the desired frequency segment. The unidirectional wave propagation and vibration isolation performance are validated through numerical simulations. This work holds the potential to benefit the design, optimization, and application of metamaterials.
