Fiber-Reinforced Composites

Given the initial design domain and loading condition, we propose an algorithm that utilizes a mesh- independent neural-network-based representation to simultaneously optimize the matrix-topology, fiber-density, and fiber-orientation. This NN-based representation enables an end-to-end automatic differentiation, where we only need to define a loss function without manually deriving the sensitivity expressions. Further, this implicit representation allows realization of continuous fibers at high-resolution in the post-processing stage using the trained NN. Subsequently, the optimized FRC part can be fabricated using a 3D printer.