Classification and reconstruction of three-dimensional microstructures using support vector machines

Reconstruction of three-dimensional (3D) microstructures is posed and solved as a pattern recognition problem. A microstructure database is used within a support vector machines framework for predicting 3D reconstructions of microstructures using limited statistical information available from planar images. The 3D distributions of the grain size of the reconstructed polyhedral microstructures exhibit qualitative agreement with stereological predictions. Amenability of the approach for studying microstructure–property relationships is shown by comparing the computed properties of reconstructed microstructures with available experimental results. Combination of classification methodology and principal component analysis for effective reduced-order representation of 3D microstructures is demonstrated. The pattern recognition technique discussed uses two-dimensional microstructure signatures to generate in nearly real-time 3D realizations, thus accelerating prediction of material properties and contributing to the development of materials-by-design.