An efficient optimization based microstructure reconstruction approach with multiple loss functions

Stochastic microstructure reconstruction involves digital generation of microstructures that match key statistics and characteristics a (set of) target microstructure(s). This process enables computational analyses on ensembles without having to perform exhaustive costly experimental characterizations. Statistical functions-based deep learning-based methods are among the stochastic approaches applicable wide range material systems. In this paper, we integrate statistical descriptors as well feature maps from pre-trained neural network into an overall loss function for optimization based procedure. helps us achieve significant efficiency in reconstructing retain critically important physical properties microstructure. A numerical example bi-phase random porous ceramic demonstrates proposed methodology. We further detailed finite element analysis (FEA) reconstructed calculate effective elastic modulus, thermal conductivity hydraulic conductivity, order analyse algorithm’s capacity capture variability these with respect those method provides economic, efficient easy-to-use approach multiphase materials 2D which has potential be extended 3D structures.