Porous material characterization with Bayesian and machine learning method

Two main classes of characterization methods are usually employed for acoustic parameters of porous media, namely the analytical inversion method or the numerical fitting procedure. While analytical inverse methods suffer from a high sensivity to the operator, the numerical fitting methods can lead to non physical sets of parameters. The fitting procedure can be bounded by the analytical inversion. The Bayesian method, which is presented in this work, is the appropriate tool to couple these methods. It enables to carry out a numerical fitting procedure using the analytical inversion as prior estimation while increasing the confidence in the resulting uncertainties. In a second part, machine learning method and data mining are used as an alternative characterization method which could be a good prior estimation for the Bayesian method. Results and performances will be shown for various porous media.