Neural Network Based In-situ Method to Determine Surface Impedance and Absorption Coefficient of Porous Materials (vor Ort)
* Presenting author
At recent conferences we proposed a free-field or in situ wave field analysis method for determining acoustic material parameters of porous absorbers based on measurements over finite samples using trained neural networks to model the edge effect for inverse material estimation. The approach avoids estimation errors resulting from simplifying assumptions concerning the sound field or the type of material reaction, so that also angle-dependent surface impedances of non-locally reacting materials can be determined. Using numerically simulated measurements it was shown that the finite sample effect can be neutralized very well and accurate material estimates can be obtained even with a simple measurement setup. Building on the knowledge gained so far, this contribution presents further investigations concerning ways to improve the neural networks for sound field prediction as well as the assessment of the ability to extrapolate, i.e. to estimate material parameters in frequency ranges above the intervals learned in training. Furthermore, the entire method is applied to real measurements in the hemi-anechoic chamber, the additional adjustments required for this are discussed, and the resulting estimation results are evaluated.