Artificial neural network modeling for anisotropic hyperelastic materials based on computational crystal structure data

摘要

Based on computational crystal structures data, a new modeling approach using artificial neural network for anisotropic hyperelastic materials is proposed. Compared to experiments, computational model for materials can simulate various behaviors of materials and provide information on all components of strain, stress and constitutive tensors. Classical hyperelastic modeling methods assume the strain energy density function, and obtain stress and constitutive tensor by differentiating it with respect to the strains. However, it is not easy to find the strain energy density function that describes the anisotropic hyperelastic materials such as complex hexagonal and tetragonal crystal structures. Bypassing the assumption of the strain energy density function, this study directly uses the data of the strain, stress, and constitutive tensor through machine learning technique. Artificial neural network (ANN) models are constructed for face-centered cubic (Cu), diamond cubic (Si) and hexagonal (ZnO) structures. Numerical simulations based on finite element method (FEM) are performed using the proposed ANN models in order to characterize macroscopic mechanical behaviors.