Convolutional neural network for homogenization of particulate composite materials based on finite element data

This study develops a convolutional neural network model to predict the apparent mechanical properties of particulate composite materials based on finite element data. The particulate composite material is considered with random inclusions in size and position. The datasets for training and testing processes are generated by using a validated finite element simulation. Various parametric studies are then investigated, including model efficiency and uncertainty propagation. Moreover, the influence of the constituents and microstructure is numerically revealed based on the proposed convolutional neural network model. It is shown that the developed convolutional neural network model is capable of capturing the microstructural features and provides accurate predictions of apparent mechanical properties of particulate composite materials.