Computational microstructure characterization and reconstruction for stochastic multiscale material design

There are two critical components of connecting material and structural design in a multiscale design process: (1) relate material processing parameters to the microstructure that arises after processing, and (2) stochastically characterize and subsequently reconstruct the microstructure to enable automation of material design that scales upward to the structural domain. This work proposes a data-driven framework to address both the above components for two-phase materials (composites with two materials mixed together, each having distinct material properties) and presents the algorithmic backbone to such a framework. In line with the two components above, a set of numerical algorithms is presented for characterization and reconstruction of two-phase materials from microscopic images: these include grayscale image binarization, point-correlation and cluster-correlation characterization, and simulated annealing algorithm for microstructure reconstruction. Another set of algorithms is proposed to connect the material processing parameters with the resulting microstructure by mapping nonlinear, nonphysical regression parameters in microstructure correlation functions to a physically based, simple regression model of key material characteristic parameters. This methodology that relates material design variables to material structure is crucial for stochastic multiscale material design.