A predictable glass forming ability expression by statistical learning and evolutionary intelligence

This paper demonstrates how principal component analysis of multivariate BMG alloy data and the genetic programming of the extracted features in the form of principal components can be used to develop a meta-modeling scheme for GFA expression. The proposed GFA model can estimate the glass forming potential of an alloy from its composition data, unlike the characteristic temperature based glass forming ability expressions, consisting of Tg, Txand Tl. The BMG alloys have been described by means of generic attributes of the constituent elements and corresponding composition of the alloy yielding a multi-dimensional descriptor space for a 594 BMGs compiled from literature. The PCA model of the data base plausibly reduced the dimensionality into a two dimension in terms of two extracted features by first two principle components capturing the 82% of the data knowledge. Successively, these principle components are used to develop a constitutive model for glass forming ability using genetic programming. The combinatorial analysis of the meta-model for GFA expression is applied to the prediction of potential compositional zone in five different experimentally explored ternary systems. The predicted composition zones are discussed in the context of available experimental data in literature and the energy of formation of the stable phases in respective alloy systems.