Machine Learning for Modeling the Edge Effect of Porous Absorbers
Master Thesis of Herrmann, Sven
Theoretical models for the sound field above porous absorbers are usually based on the assumption of an infinite material layer in at least one dimension. Those analytical approaches are only rough approximations and of limited suitability for practical applications like the estimation of material or surface properties derived from sound field measurements above finite absorber samples. Therefore, an approach for modeling the edge effect based on supervised machine learning is introduced in this work and tested with support-vector machines as well as with neural networks. The deviations between the Allard et al. model and the FEM due to diffraction are used as target values for the training. In addition to flow resistivity, material thickness and edge length of the sample, the frequency as well as the positions of source and receiver are also varied. The models trained this way showed a good performance in all investigated cases and provide better results than the basic Allard et al. model without correction for the edge effect.