Estimation of sound absorption coefficient of composite structured aluminum foam by radial basis function neural network. (1st January 2022)

• Record Type: Journal Article
• Authors: Liang, Lisi; Mi, Han; Guo, Wenlong; Zhang, Yu; Ma, Hongyue; Zhang, Ziheng; Li, Linbo
• Is Part Of: Applied acoustics. Volume 185(2022)
• Publication Date: 2022-01-01
• Subjects: Radial basis function neural network -- Open-cell aluminum foam -- Sound absorption coefficient -- Composite structure -- Transfer matrix method; Acoustical engineering -- Periodicals; Periodicals; 620.2
• DOI: 10.1016/j.apacoust.2021.108414
• Languages: English
• ISSNs: 0003-682X

Abstract

Abstract: In order to better understand the acoustic performance of composite structured open-cell aluminum foam in the low and medium frequency bands, this paper presents a method for predicting the absorption coefficient of a given structure in the low and medium frequency bands using radial basis function (RBF) neural networks. The method uses a number of easily measurable structural parameters to obtain the sound absorption coefficient of composite structural aluminum foam and then compares the predictions with those of two conventional transfer matrix methods (TMM). Therefore, in the experimental part, 165 composite structured open-cell aluminum foams were, firstly, designed using three different thickness of aluminum open-cell foam with a density of 0.914 kg/m 3 and three different thickness of aluminum open-cell foam with a density of 0.711 kg/m 3 . Next, the two main structural parameters of airflow resistivity and porosity of each layer were measured as inputs to the radial basis function neural network. Seventeen of these data sets were then used as a test set to compare the results of the absorption coefficients with the experimental values of the other two numerical models at 10 low frequency frequencies, respectively. Finally, a maximum root mean square error of 0.023, and a mean root mean square error of 0.012 were derived using radial basis function neural networks, which were numerically much lower than the other two TMM models, and RBF had higher accuracy, indicating that the method was effective.