Researchers from the Yale School of Engineering and Applied Sciences are analyzing the effectiveness of a machine learning tool in predicting the formability of bulk metallic glass.
Courtesy of Guannan Liu
Machine learning has been used for a wide range of tasks such as speech recognition, fraud detection, product recommendations, image recognition, and personalized medicine—however, its implementation has been limited when it comes to solving complex materials science problems.
One such problem is predicting the ability of an alloy to form glass, which is a mixture of one or more metals or metallic and non-metallic elements. A Yale-led study took this hurdle, exploring the use of a machine learning model to predict the formation of bulk metallic glass.
Bulk mineral bottles exhibit unique properties including high strength, high hardness, corrosion resistance and a large elastic stress limit. To predict the formability of these types of glasses, Yale researchers developed a machine learning model based on 201 alloy features created from a mixture of 31 elemental features, including atomic number, atomic weight, melting temperature, covalent radius, heat of fusion, and electrostatics. . This prediction was then compared to a model based on non-physical features, as well as a machine learning model based on human insights that they also developed.
“The nature of these different inputs is what sets this work apart, which ranges broadly from raw data to non-physical data to acquired human data,” said Guannan Liu GRD. PhD student in mechanical engineering and materials science at Yale University and the first author of the study.
Corey O’Hearn, A professor of mechanical engineering and materials science at Yale University confirmed that despite the success of machine learning tools in other fields, these methods have so far been unable to predict A new metal alloy for forming glass. Thus, there is an opportunity for future exploration.
“This work begins to address this question so that new machine learning methods can be developed for bulk metallic glass design,” O’Hern said.
The authors found that regardless of the nature of the data—raw, soft, and human-learned—the prediction accuracy of new alloys of similar composition from the training dataset was comparable between models.
However, the machine learning model based on 201 alloy features was found to produce worse results than the human learning based model in predicting new alloys whose compositions were very different from the training data set.
“It reveals a very powerful idea: complex materials science problems such as the formation of massive metallic glass require physical insights to develop efficient and predictable machine learning models,” said Liu.
Because a significant amount of the work has focused on comparing different machine learning tools in the past, the team’s approach allowed them to compare the machine learning approach to traditional computer-aided human learning, providing insight into the applications of machine learning in materials design.
Sung Woo Sohn, an associate research scientist in the Department of Mechanical Engineering and Materials Science at Yale University, dwelled on the difference in results between the study model and the human learning-based model, noting that the human learning-based model showed greater ability to extrapolate than the general machine learning model, “which provides accurate predictions only close to known data.”
Mark D. said: Shattuck, Professor of Physics at City College of New York and co-author of this study. “We’ve taken the first steps to identify this useful area of material design.”
According to Liu, the team aims to extend the use of machine learning to other areas, such as exploring the world of glass formation as well as the possibilities of new metallic glass.
The study appeared in the journal Acta Materia.