Acoustic topology optimization using moving morphable components in neural network-based design

Abstract

In this study, we developed an acoustic topology optimization using moving morphable components (MMCs) for the design of two-dimensional sound reduction structures. MMC-based topology optimization has been developed for structural topology optimization; however, no extant study on the design of sound reduction structures has utilized MMC-based topology optimization. Instead of directly changing the distribution of pixel-wise materials to form the shape of a structure, MMC-based topology optimization changes the geometric and positional parameters of MMCs and forms the shape of a structure through the overlapping of MMCs. In this study, finite element analysis based on the Helmholtz equation was performed to calculate the acoustic performance of sound reduction structures. To complement the unsatisfactory performance of designs by local optimal points, we evaluated many designs optimized under different design conditions and optimization settings with respect to the original design condition. We also devised additional design procedures to improve the acoustic performance of sound reduction structures by exploring a lot of design samples modified from the designs based on MMC-based topology optimization. Owing to the rather long time required for repeated performance calculations, the performance was estimated by using a multilayer perceptron to roughly select the design samples that need to be evaluated by finite element analysis. Design examples for barrier structures and duct internal structures were considered to demonstrate the validity of the proposed approach.