A predictive data mining approach for microstructure optimization

We ask and seek answers for these questions: can we identify the microstructure that is theoretically predicted to yield a desired property demanded by a selected application? What if what’s demanded is a combination of multiple properties? What if there’re more than one microstructure that yield the same property? Can we identify the complete space (or as much of it as possible) of them? Traditionally, we search (by means of global optimization), but it’s slow and incompetent in a lot of ways. Now, how can modern data mining facilitate this process?

This data mining based approach improves on the traditional search-based scheme by inserting three additional steps: random data generation, search path refinement and search space reduction. The last two steps can be processed in parallel to save the learning time.


The application problem involves design of magnetoelastic Fe-Ga alloy microstructure for enhanced elastic, plastic and magnetostrictive properties. While theoretical models for computing properties given the microstructure are known for this alloy, inversion of these relationships to obtain microstructures that lead to desired properties is challenging. The challenges are primarily due to the high dimensionality of microstructure space, complexity of the constraints and non-uniqueness of solutions. These challenges render traditional search-based optimization methods incompetent in terms of both searching efficiency and accuracy.

We experiment with five polycrystalline alloy design problems each with a different property to optimize. The properties are either a singular or a composite of the following: Young modulus (E), Yield strength (Y) and magnetostrictive strain (ms).