Material Properties Are Couched in the Language of Artificial Intelligence

Stone, iron, steel, silicon. What will the next transformative material be?  Prasanna Balachandran, University of Virginia assistant professor of materials science and engineering with a joint appointment in mechanical and aerospace engineering, narrows the search for the best possible material to meet a pressing need or purpose.

Balachandran has earned a Young Faculty Award from the Defense Advanced Research Projects Agency to better target research and development of high entropy alloys that perform well in extreme environments. It is a daunting task.

“We can design materials from any combination of elements, on the order of millions of millions. The number of alloys that might endure high-temperature environments is computationally huge, and experimentation is painstaking and expensive,” Balachandran said. He meets the challenge with a data-driven approach combining artificial intelligence and quantum mechanics.

“My computational models anticipate which alloys are most likely to perform well in these extreme, high-temperature environments, so that each experiment advances the ball further down the field,” he said.

In contrast to the scientific method’s “do-see” experimentation, models driven by artificial intelligence integrate the desired rules and the salient physics at the get-go. The algorithms search and select a single or small subset of alloys for experimentation that meet threshold criteria for performance, functionality and cost. Balachandran then ploughs the experimental findings back into the model so the next round of experiments tills the most fertile ground with higher yields.  

“Data science and artificial intelligence are tools we use to make the search for suitable alloys more productive and cost effective,” Balachandran said. “These tools help us ask the right questions to understand what one can do with materials. These tools uncover trends and patterns of material properties and behaviors that are beyond what an individual scientist can observe or easily comprehend.”

Balachandran is one of three UVA faculty to earn the prestigious Young Faculty Award this year. The award is designed for rising stars in junior research positions and provides them support to develop their ideas in areas that will be useful for national security.

Balachandran’s research in materials science has always intersected with artificial intelligence. He credits his mentor Murugananth Marimuthu, more affectionately known as “Dr. Ananth,” for introducing him to the field while an undergraduate at the PSG College of Technology in Coimbatore, India.

“My interaction with Dr. Ananth was truly fortuitous; it helped me clarify my own research aims in my undergraduate thesis,” Balachandran said.

Dr. Ananth passed along the knowledge and enthusiasm he himself gained as a student of Sir Harshad “Harry” Bhadeshia, University of Cambridge Tata Steel Professor of Metallurgy. Bhadeshia proved the value of mathematical models to predict microstructural features of steels and to alloy steel to improve its mechanical properties.

Inspired by their example, Balachandran explored the use of artificial neural networks, computing systems that mimic the way in which synapses and neurons function in the human brain, to parse through infinite design possibilities for aluminum alloys and bulk metallic glasses.

Eager to continue this interplay of materials and artificial intelligence for his graduate education, Balachandran set his sights on Iowa State, the first and at the time the only university in the United States with a mature program in this cross-disciplinary field. Balachandran joined a research group led by Krishna Rajan, who pioneered research in materials informatics.

“One of the best things about working with Krishna was his wide-ranging reach for ideas. He expected us to read works outside the materials science literature and find synergies across multiple disciplines,” Balachandran said. Rajan’s pluralistic approach helped Balachandran formulate his own principles and ideas on how to leverage data science to conduct meaningful research in materials science.

Balachandran earned his Ph.D. from Iowa State during the nation’s renaissance in artificial intelligence for materials science research. “The year 2011 was a special one for me. The U.S. formally launched a new research initiative that put a bright spotlight on data-driven materials science, which was the core of my dissertation,” Balachandran said, referring to the Materials Genome Initiative for Global Competitiveness.

“The whole idea of using machine learning and artificial intelligence in materials science took off. Conferences were jam-packed. The field really blossomed,” he said.

Balachandran realized that in order to make a career in materials informatics, he would need to embed himself in sponsored research with real-world applications. “To fully exercise my creativity and innovate the field, I needed to both pose and prove my hypotheses in my own research group,” Balachandran said.

Balachandran found the perfect fit as a post-doc at Drexel University, joining a research group led by James Rondinelli, now associate professor and director of the materials research and design group at Northwestern University’s McCormick School of Engineering. Kismet may have played a role in Balachandran’s post-doc research; Rondinelli himself earned a DARPA Young Faculty Award, which brought Balachandran to Drexel University.

In Rondinelli’s group, Balachandran learned to connect machine learning with physics-based simulations. “Working with Dr. Rondinelli catapulted my professional development and helped shape my vision for materials informatics,” Balachandran said

Balachandran’s dissertation research and post-doc enabled him to drill down and discover resources within the field of materials informatics—to make the most of his discoveries. He also appreciated the opportunity to forage for ideas outside of materials science, working alongside physicists, computer scientists, statisticians and experimentalists.

Whereas his dissertation and post-doc research identified “what will work,” his research at Los Alamos focused on why materials behaved in standard, special or spurious ways. “Our group published good papers and validated AI’s role in materials discovery. We also showed we can fail; there’s no magical formula for discovering new materials, and no substitute for expert knowledge,” Balachandran said.

These early career experiences leave no doubt that Balachandran is a materials scientist at his core. As an assistant professor in UVA’s Department of Materials Science and Engineering, Balachandran found a group of like-minded researchers and educators. In 2017, UVA Engineering recruited Balachandran for its Multifunctional Materials Integration initiative, which brings together researchers from multiple engineering disciplines to formulate materials with a wide array of functionalities.

Balachandran connected with the Virginia Nano Computing Group led by Avik Ghosh, UVA professor of electrical and computer engineering and physics. Ghosh’s nanomagnetism research team is working on a new paradigm to engineer tiny information-carrying bits, called skyrmions, to simultaneously increase memory, processing speed and power economy for conventional memory and unconventional computing.

Engineering skyrmions requires significant materials discovery, as the set of naturally occurring stable materials is limited, Ghosh explained. They also impose conflicting requirements, to get their size, speed, lifetime and readability on target in a single device platform. Fast computational materials discovery and optimization are needed along this path.

“Dr. Balachandran plays a critical role in this regard. He operates at multiple levels – using machine learning algorithms for fast superficial predictions of overall trends, as well as detailed models that are slow and thorough and look closely into the underlying atomic chemistry,” Ghosh said.

Balachandran also enjoys working with experimentalists. Within UVA Engineering he has created a positive feedback loop. “We couch what we know and don’t know in the language of mathematics, which is just another way of talking about uncertainty. If there is a lot of uncertainty around a class of materials or a materials system, physics allows us to understand and picture that landscape,” Balachandran said.

“Good science, visibility and encouragement from peers is in the secret sauce of leading a research group,” Balachandran said. Balachandran embraces his role to nucleate ideas, give talks, and to be an ambassador for materials science within the academy and the nation’s encompassing research enterprise.

“I try to empathize with the new generation of scholars,” Balachandran said. He aims to make his own research program as inclusive as possible, to help each student realize his or her own vision. “Every student is an innovator.”

Balachandran points his own research team toward societal imperatives in materials science, from sustainable transportation and renewable energy to quantum computing.

“I don’t know which materials will help us do these things,” he said. “Our group doesn’t have the capability to deliver a ready-made solution. But our capability can rapidly distill the available information and identify knowledge gaps. We can recommend promising research trajectories and formulate flexible AI strategies that can learn from both successes and failures.”