General regression neural network for prediction of sound absorption coefficients of sandwich structure nonwoven absorbers

In this paper, we propose a more general forecasting method to predict the sound absorption coefficients at six central frequencies and the average sound absorption coefficient of a sandwich structure nonwoven absorber. The kernel assumption of the proposed method is that the acoustics property of sandwich structure nonwoven absorber is determined by some easily measured structural parameters, such as thickness, area density, porosity, and pore size of each layer, if the type of the fiber used in nonwoven is given. By holding this assumption in mind, we will use general regression neural network (GRNN) as a prediction model to bridge the gap between the measured structural parameters of each absorber and its sound absorption coefficient. In experiment section, one hundred sandwich structure nonwoven absorbers are particularly designed with ten different types of meltblown polypropylene nonwoven materials and four types of hydroentangled E-glass fiber nonwoven materials firstly. Secondly, four structural parameters, i.e., thickness, area density, porosity, and pore size of each layer are instrumentally measured, which will be used as the inputs of GRNN. Thirdly, the sound absorption coefficients of each absorber are measured with SW477 impedance tube. The sound absorption coefficient at 125 Hz, 250 Hz, 500 Hz, 1000 Hz, 2000 Hz, 4000 Hz and their average value are used as the outputs of GRNN. Finally, the prediction framework will be carried out after the desired training set selection and spread parameter optimization of GRNN. The prediction results of 20 test samples show the prediction method proposed in this paper is reliable and efficient.