Using genetic algorithms to evolve three-dimensional microstructures from two-dimensional micrographs

The article describes work to bring together the topics of evolutionary computing and stereology and asks the reader to judge whether such an approach can be genuinely useful or just represents a clever application of computer science. The problem we address is that of constructing three-dimensional (3-D) microstructures from two-dimensional (2-D) micrographs. Our solution is a computer program called MicroConstructor that evolves 3-D discrete computer microstructures, which are statistically equivalent to the 2-D inputs in terms of the microstructural variables of interest. The core of MicroConstructor is a genetic algorithm that evolves the 3-D microstructure so that its stereological parameters match the 2-D data. MicroConstructor uses a general method of pattern construction, the EmbryoCA, that does not require intervention from the user and is highly evolvable. This article presents initial results from successful experiments to evolve 3-D two-phase microstructures from 2-D input microstructures. The advantages and disadvantages of the method are discussed, and we conclude that the method, though delightfully elegant and full of potential, has yet to prove itself capable of constructing 3-D microstructures that would interest experimentalists and computer modelers.