Stochastic Multiscale Characterization of Short-Fiber Reinforced Composites

A framework for stochastic modelling and optimization of materials with engineered microstructures is presented. Numerical methods for solving problems with short-fiber inclusions are discussed. Addition of fiber reinforcement has been shown to improve the performance of various materials in a number of applications. The response of fiber reinforced concrete under tensile stress is dependent upon several properties, including fiber geometry, fiber material properties, fiber length, and orientation of the fibers with respect to the applied load. In a real-world system the distribution of the fibers may be random, with orientation angle and configuration varying locally. Stochastic multiscale methods enable the connection of the scales to analyze the effect of randomly distributed short-fiber inclusions on the global response of the system. Randomly generated characteristic volume elements (CVE) are analyzed using the extended finite element method (XFEM) to capture local material response without the need for a mesh that conforms to the material morphology, ideal for situations with arbitrary fiber distributions. The variation observed in statistically equivalent CVE models is quantified. Correlation is determined between FRC descriptor variables and the tensile response of the composite. It is demonstrated that machine learning can be used to predict composite material properties of FRC to a reasonable degree of accuracy using information about the material microstructur