Determination of Johnson-Champoux-Allard model parameters with machine learning techniques

Acoustic modelling of materials is a key aspect to perform reliable predictions of untested configurations and to implement optimal strategies, especially nowadays, due to the development of 3D printing techniques allowing to generate deterministic microstructures in porous materials. However, this modelling is not straightforward due to the number of parameters characterizing porous materials. Each one of these parameters can be estimated experimentally with its own setup and procedure, not always easy to execute. On the other hand, the use of inverse methods for the estimation of parameters relies on physical approximations and requires experimental procedures that are strongly sensitive to boundary conditions. Consequently, experimental determination of the parameters of porous materials may become financially expensive and very time consuming. The purpose of this work is to investigate the performances of machine learning techniques in determining the parameters of Johnson-Champoux-Allard model, in order to understand whether missing values can be predicted just relying on a set of already performed experimental characterizations when not all the experimental setups are available. In particular, Gaussian processes are used on a training set describing porous materials in terms of Johnson-Champoux-Allard parameters and the values of acoustic indices at fixed frequencies.