Applicazione di algoritmi di machine learning su leghe amorfe

In recent years, there has been a lot of interest in finding new methods for the development of new materials possessing desired properties. A topic of current interest is Bulk Metallic Glasses (BMGs) a novel family of materials which have a unique set of properties making them very attractive for some industrial applications. In this work, machine-learning (ML) algorithms (or models) were used to identify compositions at which BMGs with suitable properties can be obtained. The algorithms were trained on two different datasets, the first one with 8412 alloys (BMG, Ribbon and crystalline) used for a classification task, i.e. to qualitatively predict when it is possible to form BMG, and the second one with 495 alloys Fe-based (only BMG) used for a regression task, i.e. to estimate the maximum size of the BMG that can be obtained. We used ML models (Random Forest, Support Vector Machine, XGBoost) to train a classifier capable of distinguishing alloys in three different classes BMGs, Ribbons and Crystalline alloys, according to their molar compositions. For the regression task, the dataset was composed of two different type of features, molar compositions and thermophysical parameters obtained in part from Thermo-calc software and in part from empirical equations, and we used ML models (Multiple Linear Regression, XGBoost, Support Vector Machine) to predict the critical casting diameter (Dmax) which quantitatively represents the glass-forming ability (GFA) of amorphous alloys.