The data-driven approach for prediction of yield function of composites

摘要

The theories of plasticity are based on the hypothesis of the existence of a yield function. The traditional approach is to approximate the yield function by some analytical dependence. Due to the acute lack of experimental data, this approach is very limited for composite materials. An alternative to these approaches is performing yield function numerical prediction, which is based on the known properties of the fibers and matrix. The data-driven approach and big data processing technologies allow building a yield surface with high accuracy and get rid of arbitrariness in the analytical approximation. In the work, the data-driven approach is implemented for unidirectional reinforced carbon fiber composites with a hexagonal fiber packing scheme. The micromechanical analysis of the representative volume of the composite is performed using the finite element simulation. A combination of coordinate stresses is used as the yield strength of the composite. The value of these stress tensor components corresponds to plastic strains occurrence in the composite parts such as fiber or matrix. The numerical procedure for constructing a grid yield function under arbitrary loading trajectories is developed and its plane sections for various combinations of coordinate stresses are presented. The resulting multidimensional data arrays can be used in commercial software packages in the specific form of user-defined material models.