PhD Defense - Zijian NIU

A gridless method for three-dimensional acoustic imaging based on unconstrained sonons

Abstract

Three-dimensional acoustic imaging with microphone arrays is a challenging task due to the numerous degrees of freedom of the physical processes at stake. For instance, aeroacoustic measurements always encounter many challenges. On the one hand, aeroacoustic phenomena involve objects immersed in fluid motion, resulting in highly intricate mechanisms for generating sound sources. On the other hand, the acoustic source must be identified within a three-dimensional environment, including both volumetric and surface elements, some of which may have considerable dimensions. Indeed, limitations arise when performing three-dimensional measurements due to the large number of required microphones and the exhaustive searches associated with grid-based methods.

In this context, instead of meshing a region of space to identify acoustic source distributions, the aim of this work is to propose a novel gridless method to reconstruct the sound field by an equivalent set of point sources with unrestricted coordinates, named "acoustic sonons". Their radiation is expected to reproduce the same acoustic properties as the measured acoustic field, such as its pressure level, directivity and spatial coherence by recovering phase information. It is found that the proposed concept of sonons, while being very flexible, effectively provides equivalent representations for a variety of source distributions. The simulation results show that the method can be adapted to diverse acoustic field situations, whether they are composed of elementary sources or intricate volumetric source densities. The problem is formulated within a probabilistic framework, through a hierarchical Bayesian model inferred by a dedicated Markov chain Monte Carlo algorithm. The performance of the method is evaluated on analytical radiation problems and its ability to reconstruct the directivity is tested for two far-field radiation cases and an industrial case.