Prediction of Mohs hardness with machine learning methods using compositional features

Published: 27 January 2019| Version 1 | DOI: 10.17632/jm79zfps6b.1
Contributor:
Joy Garnett

Description

Hardness, or the quantitative value of resistance to permanent or plastic deformation, plays a very crucial role in materials design in many applications, such as ceramic coatings and abrasives. Hardness testing is an especially useful method as it is non-destructive and simple to implement to gauge the plastic properties of a material. In this study, I proposed a machine, or statistical, learning approach to predict hardness in naturally occurring materials, which integrates atomic and electronic features from composition directly across a wide variety of mineral compositions and crystal systems. First, atomic and electronic features from composition, such as van der Waals and covalent radii as well as the number of valence electrons, were extracted from composition. In this study, the author trained a set of classifiers to understand whether compositional features can be used to predict the Mohs hardness of minerals of different chemical spaces, crystal structures, and crystal classes. The dataset for training and testing the classification models used in this study originated from experimental Mohs hardness data, their crystal classes, and chemical compositions of naturally occurring minerals reported in the Physical and Optical Properties of Minerals CRC Handbook of Chemistry and Physics and the American Mineralogist Crystal Structure Database. The database is composed of 369 uniquely named minerals. Due to the presence of multiple composition combinations for minerals referred to by the same name, the first step was to perform compositional permutations on these minerals. This produced a database of 622 minerals of unique compositions, comprising 210 monoclinic, 96 rhombohedral, 89 hexagonal, 80 tetragonal, 73 cubic, 50 orthorhombic, 22 triclinic, 1 trigonal, and 1 amorphous structure. An independent dataset was compiled to validate the model performance. The validation dataset contains the composition, crystal structure, and Mohs hardness values of 51 synthetic single crystals reported in the literature. The validation dataset includes 15 monoclinic, 7 tetragonal, 7 hexagonal, 6 orthorhombic, 4 cubic, and 3 rhombohedral crystal structures. In this study, the author constructed a database of compositional feature descriptors that characterize naturally occurring materials, which were obtained directly from the Physical and Optical Properties of Minerals CRC Handbook45. This comprehensive compositional-based dataset allows us to train models that are able to predict hardness across a wide variety of mineral compositions and crystal classes. Each material in both the naturally occurring mineral and artificial single crystal datasets was represented by 11 atomic descriptors. The elemental features are number of electrons, number of valence electrons, atomic number, Pauling electronegativity of the most common oxidation state, covalent atomic radii, van der Waals radii, ionization energy of neutral

Files