Efficient acoustic finite element simulation and optimization through inverse matrix prediction by neural network: Learning-based estimation of inverse system matrix

Abstract

This study proposes a novel deep learning framework to enhance the efficiency of acoustic simulations within the framework of the static condensation approach. To this end, a pixel-based representation and the static condensation method are employed. The static condensation scheme inherently involves computationally intensive matrix inversions. By leveraging deep learning, the condensed matrices that require these inversions are predicted directly, thereby accelerating the finite element procedure. Furthermore, the proposed method is applied to topology optimization for binary structures, which demands efficient solution strategies.