Non-contact Boundary Condition Quantification using Data-driven Beamforming Approach

초록

In this contribution, we present a deep learning-based method for quantifying non-contact boundary conditions in a virtual manner. The acoustic beamforming method is a spatial filtering technique based on microphone array measurements, which can be beneficial for visualizing strength distributions of sound sources. However, it is widely known that the beamforming maps physically induce poor resolution at low frequencies, while severe side-lobe patterns occur at high frequencies. Acoustic beamforming has been studied from the standpoint of 'sound source localization,' which identifies locations and strengths of the sound sources. Recently, deep learning-based methods have been introduced to the field of acoustic beamforming, with promising results for overcoming the physical resolution limitations of conventional beamforming maps. In this study, beyond localizing the sound source's positions or strengths, we propose a novel method for quantifying certain boundary conditions that may arise in mechanical systems. We demonstrate that boundary conditions such as the plate's spring stiffness can be virtually quantified using our proposed deep learning approach without measuring them physically. © ICA 2022.All rights reserved

키워드