A sound source reconstruction approach based on the machine vision and inverse patch transfer functions method

摘要

The inverse patch transfer functions (iPTF) method can realize local reconstruction of a vibrating structure in non-anechoic environments, and the geometrical shape of the reconstruction area could be irregular as long as it is known. However, the shape of the non-planar reconstruction area is not easy to be obtained in practice, which is adverse to its application in the situ tests. To achieve the sound source reconstruction of the local area with unknown shape in non-anechoic environments, a hybrid method that combines the machine vision and the iPTF method is proposed. The machine vision technology is applied to facilitate the boundary modeling of a target radiation segment. It makes use of the BundleFusion method with an RGB-D camera to accomplish the 3D reconstruction of the target area, as well as the localization of microphones. A virtual cavity consisting of the target area, virtual measurement surface, and the gap between them could be constructed automatically. Consequently, the impedance matrix of the virtual cavity is obtained based on the boundary integral equation with the adoption of the free field Green's function. A microphone array mounted up with a rigid masker is applied in the iPTF method, which can form the rigid Neumann boundary condition on the masker. A simulation is first carried out to investigate the influence of 3D reconstruction errors of the machine vision technology on the sound source reconstruction. Then, two experiments are performed to validate the feasibility and effectiveness of the reconstruction in noisy environments. One is conducted in the semi-anechoic room with a cylindrical radiator, and a loudspeaker is put aside as the interference source. The other is conducted in the cabin of an aircraft under cruising condition. It is demonstrated that the proposed method can realize the normal velocity reconstruction without prior geometry knowledge of the target area, and the magnitude and distribution of the dominant sound source could be reconstructed accurately. (C) 2021 Elsevier Ltd. All rights reserved.