A new method of finding rare earth compounds using artificial intelligence could lead to discoveries that revolutionize personal electronics, experts say.
Researchers from Ames Laboratory and Texas A&M University trained a machine-learning (ML) model to assess the stability of rare-earth compounds. Rare earth elements have many uses, including clean energy technologies, energy storage, and permanent magnets.
“New compounds may enable future technologies that we cannot even fathom yet,” Yaroslav Mudryk, the project supervisor, told Lifewire in an email interview.
To improve the search for new compounds, scientists used machine learning, a form of artificial intelligence (AI) driven by computer algorithms that improve through data usage and experience. Researchers also used high-throughput screening, a computational scheme that allows researchers to test hundreds of models quickly. Their work was described in a recent paper published in Acta Materialia.
Before AI, the discovery of new materials was mainly based on trial and error, Prashant Singh, one of the team members, said in an email to Lifewire. AI and machine learning let researchers use material databases and computational techniques to map both chemical stability and physical properties of new and existing compounds.
“For example, taking a newly discovered material from lab to market may take 20-30 years, but AI/ML can significantly speed up this process by simulating material properties on computers before setting foot in a lab,” Singh said.
“AI is revolutionizing how we think about solving many of these high-dimensional complex problems, and it opens a new way to think about future opportunities.”
AI beats older methods for finding new compounds, Joshua M. Pearce, the John M. Thompson Chair in Information Technology and Innovation at Western University, said in an email interview.
“The number of potential compounds, combinations, composites and novel materials is mind-blowing,” he added. “Rather than take the time and money to make and screen every one for a specific application, AI can be used to help predict materials with useful properties. Then scientists can focus their efforts.”
Markus J. Buehler, the McAfee Professor of Engineering at MIT, said in an email interview that the new paper shows the power of using machine learning.
“It’s a dramatically distinct way to make such discoveries than what we have been able to do previously—discoveries are now faster, more efficient, and can be more targeted to applications,” Buehler said. “What is exciting about the work by Singh et al is that they combine cutting edge materials tools (Density Functional Theory, a way to solve quantum problems) with tools of material informatics. It’s definitely a way that can be applied to many other materials design problems.”
Rare earth compounds are found in many high-tech products such as cell phones, watches, and tablets. For instance, in displays, these compounds are added to endow materials with highly targeted optical properties. They are also used in your cell phone’s camera.
Conceptual smart glasses with information displayed on the lenses like mail, cloud storage, and a stock ticker.
Olemedia / Getty Images
“They are, in some way, a sort of wonder material that serves as an important element in modern civilization,” Buehler said. “There are challenges, however, in how they are mined and how they are supplied. Hence, we need to explore better ways to either use them more effectively or to replace the functions with new combinations of alternative materials.”
It’s not just mineral compounds that can benefit from the machine learning approach used by the authors of the new paper. AI can be applied to many areas where the problems are so complex that scientists cannot develop conventional solutions via mathematics or simulations of known physics, Buehler said.
“After all, we do not yet have the right models to relate the structure of a material to its properties,” he added. “One area is in biology, specifically protein folding. Why do some proteins, after having a small genetic change, lead to disease? How can we develop new chemical compounds to treat disease or develop new drugs?”
Another possibility is finding a way to improve the performance of concrete to reduce its carbon impact, Buehler said. For example, the material’s molecular geometry could be arranged differently to make materials more effective so that we have more strength with less material use and that the materials last longer.
“AI is revolutionizing how we think about solving many of these high-dimensional complex problems, and it opens a new way to think about future opportunities,” he added. “We are just at the beginning of an exciting time.”