Prediction of elastic properties of short fiber reinforced composites based on machine learning

The elastic and mechanical properties of short fiber reinforced composites are significantly affected by their internal structure and the properties of the underlying materials, and the parametric analysis of these effects requires extremely high experimental or numerical analysis costs. In order to solve this problem, this paper combines the numerical homogenization method based on periodic representative volume units (RVE) and artificial neural network (ANN) to construct three forms of mechanical property prediction surrogate models of short fiber reinforced composites: spatial random distribution, intralayer random distribution and aligned distribution, respectively. Each surrogate model can quickly predict the equivalent elastic properties of composites under different parameter combinations (fiber length, aspect ratio, volume fraction, and fiber and matrix material properties), and the goodness of fit R² is above 0.98, the calculation time is negligible compared to conventional simulation calculations, which greatly saves experimental and computational costs and creates important conditions for the design and customization of short fiber-reinforced composites.