Knowledge-based reconstruction of random porous media

Abstract

An evolutionary optimization technique is used to reconstruct digitized material models of 300 3 nm 3 size for mesoporous two-phase systems. The models are adapted to the two-point probability (TPP) and to a volume-based pore-size distribution (PSD) which were derived from SANS and adsorption experiments and which carry statistical information about morphology and topology of the pore system. To avoid extreme update-costs, the bulk of mutations are assessed by means of a suitable approximation of the PSD; it is demonstrated that a sporadic insertion of the PSD suffices to drive the algorithm towards satisfactory models in acceptable time. Our approach is knowledge-based in the sense that (i) the mutations are restricted to expedient exchanges of phase-voxels by a heuristic rule, and (ii) the sporadic calculation of the PSD from the current state of the model, in essence, provides an efficient self-control for the evolutionary process. We applied the method to reconstruct periodic models of the xerogel Gelsil 200. Such reconstructs of real mesoporous solids could be utilized, for instance, to verify theories of adsorption and capillary condensation.