Bayesian approach for acoustic boundary admittance estimation

Abstract

It is essential for accurate acoustic predictions to represent the boundary properties of all interacting surfaces as accurately as possible. The acoustic boundary properties can be fully characterized by the complex-valued acoustic boundary admittance. Yet, it is a challenging task to determine this quantity in-situ, without measuring in laboratory conditions. A Bayesian approach is presented to estimate the acoustic boundary admittance based on a limited number of noisy sound pressure measurements at random observation points inside an acoustic domain. The sound pressure data are obtained utilizing the finite element method, which also serves as forward model in the inference process. This enables the application of the framework to arbitrarily shaped acoustic domains such as a car cabin. It is the benefit of the Bayesian approach, that it results in a probabilistic distribution for the boundary admittance and thus allows for uncertainty quantification of the estimation. The results prove, that the combined framework consisting of a FEM forward model and a Bayesian approach to solve the inverse problem is well-suited for the inference of the acoustic boundary admittance from noisy observations. It is particularly beneficial, if information about the uncertainty is desired.