LS-DYNA Machine Learning–Based Multiscale Method for Nonlinear Modeling of Short Fiber–Reinforced Composites

Short-fiber-reinforced composites (SFRC) are high-performance engineering materials for lightweight structural applications in the automotive and electronics industries. Typically, SFRC structures manufactured by injection molding, which induces heterogeneous microstructures, resulting nonlinear anisotropic behaviors challenging to predict conventional micromechanical analyses. In this work, we present a machine learning-based multiscale method integrating molding-induced material homogenization, Deep Material Network (DMN) finite element simulation software LS-DYNA analysis of SFRC. DMN is physics-embedded learning model that learns microscale morphologies hidden representative volume elements through offline training. By coupling with elements, have developed highly accurate efficient data-driven approach, predicts composite at computational speed orders-of-magnitude faster than high-fidelity direct numerical simulation. To industrial-scale products, transfer utilized generate unified database, effectively captures effects fiber orientations fractions on overall properties. Numerical examples presented demonstrate promising performance modeling.