It is significant to determine the macroscopic mechanical properties of composite materials with complex microstructure efficiently and accurately in many fields. We propose a deep learning method based on three-dimensional convolutional neural network (3D CNN) to predict the elastic coefficients of composite materials with inclusions of arbitrary sizes, shapes and material parameters. 3D datasets are generated, and a storage algorithm is proposed to reduce great storage costs in 3D. A general framework for 3D CNN models is constructed, and numerical experiments are carried out using 3D CNNs of various scales. Our results demonstrate that the scale of full connection part is the key factor of prediction ability of 3D CNNs in this task. We also demonstrate that our method can effectively save computational time compared with traditional numerical methods such as the finite element method in large-scale prediction tasks.
展开▼
