Inverse design of a Helmholtz resonator based low-frequency acoustic absorber using deep neural network

Abstract

The design of low-frequency sound absorbers with broadband absorption characteristics and optimized dimensions is a pressing research problem in engineering acoustics. In this work, a deep neural network based inverse prediction mechanism is proposed to geometrically design a Helmholtz resonator (HR) based acoustic absorber for low-frequency absorption. Analytically obtained frequency response from electro-acoustic theory is deployed to create the large dataset required for training and testing the deep neural network. The trained convolutional neural network inversely speculates optimum design parameters corresponding to the desired absorption characteristics with high fidelity. To validate, the inverse design procedure is initially implemented on a standard HR based sound absorber model with high accuracy. Thereafter, the inverse design strategy is extended to forecast the optimum geometric parameters of an absorber with complex features, which is realized using HRs and a micro-perforated panel. Subsequently, a quasi-perfect low-frequency acoustic absorber having minimum thickness and broadband characteristics is deduced. Importantly, it is demonstrated that the proposed absorber, comprising four parallel HRs and a microperforated panel, absorbed more than 90% sound in the frequency band of 347–630 Hz. The introduced design process reveals a wide variety of applications in engineering acoustics as it is suitable for tailoring any sound absorber model with desirable features.