[SY-D6] Machine learning assisted by first-principles calculations for designing intermetallic-typed metallic glasses
Metallic glasses have interesting properties such as low Young’s moduli, high corrosion resistance and high wear resistance and are considered new materials in many fields. The factors that contribute to the glass-forming ability (GFA) are an important consideration in the design of new metallic glasses. Ternary metallic glasses are classified into three types on the basis of their atomic size distribution. The factors that affect the GFAs of the metallic glasses are presumed to vary among the three types. However, metallic glasses with two alloying elements exhibit irregular composition ratios or atomic radii; thus, in these cases, the contributing factor is unclear. The binary metallic glasses usually crystallize only intermetallic crystalline phase after heating. For example, Ti-Cu and Ti-Ni binary metallic glasses crystallized as only TiCu and Ti2Ni intermetallic compounds. Therefore, these metallic glasses such as Ti-Cu and Ti-Ni are able to be called as intermetallic-typed metallic glasses (IMG). In this work, we aimed to elucidate the design criteria of IMG and design new metallic glasses based on the machine learning assisted by the first-principles calculations. First, the first-principles calculations were performed on intermetallic compounds and then, explanatory variables were calculated. Second, a regression systems of GFA was constructed using artificial neural networks (ANN) and logistic regression analysis (LRA). The results of ANN validation showed very high accuracy than the results of LRA validation. The partial regression coefficients of LRA indicated the design criteria of intermetallic-typed metallic glasses. The obtained regression systems were applied to intermetallic compounds whose GFA are unknown, and the alloy systems of a novel metallic glass were predicted.