Physics-informed Neural Networks for characteristic identification of noise control systems

Abstract

Recently, Physics-informed Neural Networks (PINNs) were proposed as useful numerical simulation methods for inverse analysis such as property identification. PINNs are methods to obtain physically satisfactory solutions by introducing a loss function with respect to the governing equations into a neural network. In this study, PINNs based on the wave equation was constructed for application to noise control systems. In this paper, we report a method to identify the acoustic attenuation parameters in an acoustic tube by recording the resonance state in the tube with a microphone. The obtained results are evaluated by comparison with the finite difference method.