Data-Driven and Sustainability Analysis of Bulk Metallic Glasses

Abstract

Metallic glasses (MGs) have attracted significant attention for their outstanding mechanical and physical properties, making them promising candidates for next-generation engineering applications. However, their adoption is limited by the difficulty of forming a stable amorphous structure and the reliance on trial-and-error experimental methods, which are costly and inefficient. In this work, we introduce a novel data-driven approach for predicting the glass-forming ability (GFA) of MGs and accelerating the discovery of new bulk metallic glasses (BMGs).

Two extensive datasets were compiled for this purpose. The first dataset consists of 2,441 alloys with critical temperatures (CTs), including glass transition temperature (Tg), crystallization onset temperature (Tx), and liquidus temperature (Tl), along with their associated maximum critical diameter (Dmax) and 25 thermal parameters derived from the literature. A cluster-based method utilizing unsupervised K-means model was used to cluster the dataset into three distinct regions, then several ML models were built with Extra Tree model outperforming other models. Interpretability was ensured through SHAP analysis and Kernel Density Estimation (KDE) plots. The second dataset comprises 7,544 alloys characterized by elemental compositions and Dmax. This dataset was used to develop regression models for Dmax prediction. Leveraging advanced machine learning (ML) techniques, we identified a Random Forest (RF) regression model as the best-performing model, achieving RMSE of 0.8097 mm for Dmax prediction on the testing set. Additionally, a variational autoencoder was deployed to propose new alloys with good GFA. Also, ternary diagrams were generated for the top 3 ternary alloy systems in the dataset that show the highest predictive performance using the RF model, aiming at proposing new compositions with enhanced GFA. Validation using the experimental data confirmed the reliability of the RF model for the ternary alloy systems design.

Building upon the compositional insights gained from the data-driven analysis, the second part of the thesis evaluates the sustainability of MGs as core materials in transformer applications. The environmental, economic, and social performance of the transformer is assessed through a comparative life cycle assessment (LCA) of three low-voltage transformer configurations: ferrite, silicon steel, and amorphous (MG) core transformers. A cradle-to-gate LCA is performed for four distinct production scenarios of MG and several (Fe-B-Si) amorphous alloys to assess their sustainability. Furthermore, a cradle-to-use LCA is conducted over an operational period of 40 years to measure the long-term environmental advantage of amorphous transformers. The social aspect is integrated by analyzing human well-being, health implications, and noise pollution of transformers. The findings show that amorphous core transformers offer superior sustainability across environmental, economic, and social dimensions, paving the way for greener power systems and more sustainable transformer design.