Estimation of sound absorption coefficient of layered fibrous material using artificial neural networks. (1st December 2020)

• Record Type: Journal Article
• Authors: Jeon, Ju Hyun; Yang, Sung Soo; Kang, Yeon June
• Is Part Of: Applied acoustics. Volume 169(2020)
• Publication Date: 2020-12-01
• Subjects: Artificial neural network -- Transfer matrix method -- Sound absorption coefficient -- Non-acoustical parameter; Acoustical engineering -- Periodicals; Periodicals; 620.2
• DOI: 10.1016/j.apacoust.2020.107476
• Languages: English
• ISSNs: 0003-682X

Abstract

Highlights: Estimation of sound absorption coefficient of multi-layered acoustical materials. Artificial neural network (ANN) was used. ANN considered thickness, density, specific airflow resistivity, and sequence. Proposed ANN model showed better estimation accuracy than the transfer matrix model. Abstract: The feasibility of a data-based artificial neural network (ANN) for the estimation of the sound absorption coefficient of a layered fibrous material is investigated in this study. The sound absorption coefficient of a four-layered fibrous material was estimated using a well-trained ANN model with only one non-acoustical parameter: the airflow resistivity ( σ). The results indicated that the ANN model exhibits a good correlation between the estimated and measured absorption coefficient. The training data sets were built by carrying out experimental measurements using a two-microphone impedance tube with 230 combinations of four-layered fibrous materials. The results of the ANN are compared in three different cases with the transfer matrix method (TMM), which is the conventional method of estimating the sound absorption coefficient of multi-layers using several non-acoustical parameters. The sound propagation model in acoustical material for the TMM was used by two models proposed by Delany-Bazely (one non-acoustical parameter) and Johnson-Champoux-Allard (five non-acoustical parameters). By comparing the estimated sound absorption coefficient from the ANN and TMM with measured values, it was demonstrated that the model developed in this paper gives more accurate results within the defined conditions. The results were compared in the frequency range of 3000–6000 Hz, and the error of the ANN model was less than 1.67%.