Group-random algorithm to generate representative volume element models for composites

One of the most commonly used methods for characterizing the mechanical properties of discontinuous fiber reinforced composites (DFRC) is to establish a Representative Volume Element (RVE) model and perform finite element (FE) analysis. However, FE analysis on RVE models established by traditional sampling algorithms is often computationally expensive due to the large size of RVE that is required to be statistically representative of the composite. To address this issue, this paper proposes a new approach for constructing RVE models with more accurate description of fiber orientation, aiming to make the FE modelling more efficient by using an RVE with small size. When establishing RVE models with given target fiber orientation tensor, it is very challenging to accurately capture the orientation of fibers. In order to mitigate the error between the orientation tensor reconstructed by fibers generated in the RVE and the target orientation tensor, a group-random algorithm is proposed in the current work to generate RVE models. Unlike the traditional algorithm, in which fibers are sampled one by one in the RVE, the group-random algorithm samples a group of four fibers at one time in order to eliminate the error of the off-diagonal components of the reconstructed orientation tensor in the principal coordinate system. Then a modification tensor is further introduced to mitigate the error of the diagonal components of the reconstructed orientation tensor. Simulation results show that the orientation tensor error could be significantly reduced by the group-random algorithm even for the RVE with low number of fibers. The merits of the group-random algorithm are also witnessed by the stability and accuracy of predicting the elastic constants of composite materials through RVE modeling. It is thus concluded that the major advantage of this work is to provide an alternatively feasible strategy to substantially improve computational efficiency of RVE modelling.