Dec 18 2020
In several industries, from energy production to aerospace, superhard materials are highly sought after. However, finding appropriate new materials has so far mostly been a matter of trial and error based on classical materials like diamonds.
Researchers have developed a machine learning model that can accurately predict the hardness of new materials, allowing scientists to more readily find compounds suitable for use in a variety of applications. Image Credit: University of Houston.
Scientists at the University of Houston (UH) and Manhattan College have developed a machine-learning model with the ability to precisely determine the hardness of new materials, thus enabling researchers to more readily discover compounds apt for use in a range of applications. The study has been described in the Advanced Materials journal.
Superhard materials—or materials with a hardness value surpassing 40 gigapascals on the Vickers scale, signifying that it would take over 40 gigapascals of pressure to leave an indentation on the surface of the material—are not so common.
That makes identifying new materials challenging. That is why materials like synthetic diamond are still used even though they are challenging and expensive to make.
Jakoah Brgoch, Study Corresponding Author and Associate Professor of Chemistry, University of Houston
One of the difficult factors is that the material’s hardness might differ based on the amount of pressure exerted, called load dependence. This renders testing a material experimentally complicated and making use of computational modeling almost impossible at present.
The model described by the team addresses this challenge by estimating the load-dependent Vickers hardness based entirely on the chemical composition of the material.
The team has identified over 10 new and potential stable borocarbide phases; research is ongoing to design and produce the materials so they could be tested in the laboratory.
Depending on the reported precision of the model, the prospects seem good. The team reported an accuracy of 97%.
According to Ziyan Zhang, the first author of the study and a doctoral student at UH, the database developed to train the algorithm depends on data related to 560 different compounds, each providing numerous data points. To find the data required to develop a representative dataset, the researchers had to go through hundreds of published academic papers.
All good machine learning projects start with a good dataset. The true success is largely the development of this dataset.
Jakoah Brgoch, Study Corresponding Author and Associate Professor of Chemistry, University of Houston
Brgoch is also a principal investigator with the Texas Center for Superconductivity at UH. Apart from Brgoch and Zhang, additional scientists on the project are Aria Mansouri Tehrani and Blake Day, both with UH, and Anton O. Oliynyk from Manhattan College.
Conventionally, scientists have employed machine learning to estimate a single variable of hardness, stated Brgoch. However, that does not account for the complexities of the property, such as load dependence, which according to Brgoch are still not understood clearly. That makes machine learning an ideal tool, though there have been some drawbacks earlier.
According to Brgoch, “A machine learning system doesn’t need to understand the physics. It just analyzes the training data and makes new predictions based on statistics.”
However, machine learning does have its own shortcomings.
The idea of using machine learning isn’t to say, ‘Here is the next greatest material,’ but to help guide our experimental search. It tells you where you should look.
Jakoah Brgoch, Study Corresponding Author and Associate Professor of Chemistry, University of Houston
Journal Reference:
Zhang, Z., et al. (2020) Finding the Next Superhard Material through Ensemble Learning. Advanced Materials. doi.org/10.1002/adma.202005112.