High-Fidelity Modeling of Stochastic RVE Based on Learning Strategies for Unidirectional Fiber Reinforced Composite Material

High-fidelity modeling of representative volume element (RVE) is crucial for accurately predicting the mechanical properties of unidirectional fiber-reinforced composites. In this study, the distribution deviation technique is innovatively proposed to quantify the deviation of numerically generated fiber distribution and combined with particle swarm optimization algorithm to develop a method that can help generate high-fidelity random fiber distribution. This method can effectively reproduce the short-range regularity and long-range randomness in the fiber distribution of unidirectional composites and eliminate unreasonable matrix-rich corners. Meanwhile, the computational efficiency has been greatly enhanced by avoiding the time-consuming trial-and-error process of obtaining appropriate parameters. Compared with other up-to-date methods, the random fiber distribution generated by this method can more accurately characterize the initiation and evolution process of damage in composite materials, significantly reducing the error range of the predicted stress–strain curve and thereby providing more accurate predictions for the damage initiation, strength, and failure strain of composite.