Auto-encoder-based inverse characterization of Transport properties of acoustic foams

The characterization of non-acoustic parameters is critically important for understanding the acoustic property and structural design of polyurethane (PU) foams. However, inverse characterization of acoustic PU foams through experiments and simulations often results in prolonged cycles and high resource wastage. To address the above issue, an innovative approach based on the Auto-encoder (AE) was proposed in this paper. In the AE approach, the decoder module was utilized for the forward prediction part, while the encoder was used for the inverse characterization. A sample database of 96,730 data sets covering PU foams’ sound absorption coefficients at 500–6000 Hz was established to train the AE model. To verify the effectiveness of the trained model, a comparative experiment with numerical simulations was firstly conducted. The results revealed that the coefficient of determination (R²) of forward prediction module surpasses 0.99, while the prediction time is significantly rapid, averaging 0.0005 s per sample, which is 1/22,000 of numerical simulation time. Another comparative experiment was conducted between the inverse characterization results of the machine learning model and the experimental data from real samples. The results showed that the average error of the characterization parameters (non-acoustic parameters and material thickness) is about 8.70 %. In summary, this study provides an intelligent inverse characterization method for targeted sound absorption of PU foams, with potential extensions to the inverse characterization of other acoustic porous materials.