Study produces unparalleled predictive criterion for glass forming ability

With superior mechanical, physical, and chemical properties, metallic glass has emerged as a popular material used in a range of applications, including electronics, biomedical instruments, and nuclear waste disposal. It’s also an effective reinforcing element in concrete, plastic and rubber.

However, the material’s complex structure presents challenges for determining glass forming ability (GFA), primarily characterized by critical cooling rates and critical casting diameters that are difficult to measure. Researchers would like to identify new criteria with more easily measurable parameters.

Tan et al. introduced machine learning methods with symbolic regression and artificial neural network models to develop a new criterion that exhibits stronger GFA predictive performance than its predecessors.

Usually, each metallic glass exhibits three characteristic temperatures during preparation: glass transition temperature (T_g), onset crystallization temperature (T_x), and liquidus temperature (T_l). The GFA is represented by critical casting diameter (D_max).

“Our job is to find a formula that establishes a relationship between the three characteristic temperatures and D_max, so that we can predict the D_max by the temperatures,” said author Yong-Chao Liang.

The scientists also used neural network models to generate more data to test predictive efficacy for other alloys.

“In short, we found supercooled liquids do not crystallize between T_g and T_x, but tend to crystallize between T_x and T_l. Therefore, T_x-T_g should be maximized and T_l-T_x should be minimized in the ideal expression for GFA,” he added.

Source: “Discovery of a new criterion for predicting glass-forming ability based on symbolic regression and artificial neural network,” by Baofeng Tan, Yong-Chao Liang, Qian Chen, Li Zhang, and Jia-jun Ma, Journal of Applied Physics (2022). The article can be accessed at http://doi.org/10.1063/5.0105445 .