Study on the Sound Absorption Performance of Three Cycle Minimum Surface Based on Convolutional Neural Network

Three Cycle Minimum Surface (TPMS) structures have been extensively studied in the field of sound absorption and noise reduction, but there is relatively little research on its sound absorption. Traditional learning can no longer meet the demand, and combining deep learning for sound absorption performance prediction is an inevitable trend. In the paper, the acoustic absorption properties of TPMS structures based on Convolutional Neural Networks (CNN) model are proposed. This paper utilizes the implicit modeling method to visualize the TPMS-G-type structure in nTopology software, and takes the two-dimensional image of the TPMS-G structure as input image. Meanwhile, the sample size of the datasets is enlarged by data enhancement techniques. Its acoustic performance of TPMS-G 3D model is simulated by COMSOL simulation software, which is set as the output value. Finally, CNN modal is constructed as the prediction model. The confusion matrix, residual graph, training loss curve, and ROC curve are used to evaluate model performance. By adjusting the depth of the network model and various hyperparameters, the accuracy of the optimized model is as high as 92.52%.