Estimation of acoustic field in 1-D duct using artificial neural networks

Anonymous

Keywords: Helmholtz equation, Acoustic pressure, Optimization, Feedforward neural network

TL;DR: Solving 1-D Helmholtz equation using neural networks

Abstract: Estimation of sound field in one-dimension finds extensive applications in many areas such as speech, automotive, aerospace and biomedical industries. Traditionally, it is obtained by solving Helmholtz equation using analytical and numerical methods (finite difference, finite element, etc.). This paper discusses a neural network methodology to solve 1-D Helmholtz equation subjected to some constraints. Unlike other governing equations, Helmholtz equation poses a biasing problem with the loss functions at higher frequencies. In the current work, an automatic weight update algorithm is proposed to bypass this difficulty. The results obtained from the proposed methodology are compared with those of the analytical method. A good correlation has been observed between the two methods. The robustness of the methodology with respect to the frequency is also verified.