Convolutional neural networks for estimating transport parameters of fibrous materials based on micro-computerized...

This study proposes a method for estimating the transport parameters of fibrous materials from x-ray micro-computed tomography (CT) images using convolutional neural networks (CNNs). Two-dimensional (2-D) micro-CT images and numerically obtained transport parameters were used to train the CNNs; Stokes flow and potential flow were used to numerically obtain the transport parameters using geometrical models extracted from the raw CT images. Then, analogously to constructing a three-dimensional image of the fibrous material by stacking the 2-D slice images, the volumetric transport parameters of the fibrous materials were calculated using the parameters of each 2-D image predicted by the trained CNN models. The transport parameters of the fibrous volume predicted by the CNN models showed good agreement with the measured values. In addition, the sound absorption coefficient was calculated by applying both the predicted and measured transport parameters to the semi-phenomenological sound propagation model and compared with the measured sound absorption coefficient. The results of the study confirm the feasibility of predicting transport parameters of fibrous materials using a neural network model based on raw micro-CT images.