Kean Professor Receives Award from Army Research Lab for AI Materials Research

Kean University mathematics Assistant Professor Ensela Mema, Ph.D., has been awarded a grant from the U.S. Army Research Laboratory to use machine learning for materials science, offering students an opportunity to apply the latest science and techniques to materials development.  

The nearly $277,000 grant over four years will support research in Kean’s School of Integrative Science and Technology. The project explores microstructures, very small components of larger materials that play a role in the overall material’s shape and strength. Materials science focuses on understanding how the composition and structure of solid materials shape their properties and practical uses. 

“There is a pressing need to develop new predictive computational models of microstructure that are inexpensive, reliable and relevant to the real world,” Mema said. “The issue with computational models is that when microstructure is added into these models, the computational cost of solving them increases drastically. We don’t have analytical methods to solve these models, so we’re incorporating machine learning techniques to approximate the solution and gain insight into the actual behavior of the microstructures.” 

Kean students will play a key role in applying machine learning techniques to the problem. 

"This offers Kean students valuable research opportunities and an introduction to the STEM fields, computational science, math and physics,” said Louis Beaugris, Ph.D., chair of the Department of Mathematical Sciences at Kean. “Exposure to that type of real-world research early in their academic careers will motivate many of them to continue on to earn Ph.D.s and explore math and science careers.” 

The grant also positions Kean well on its path to R2 designation for high research activity under the Carnegie Classification of Institutions of Higher Education.  

“Part of the impact of this grant is introducing Kean on a larger scale to the Department of Defense,” said David Joiner, Ph.D., acting associate dean for integrative science and technology at Kean. “It puts Kean on a bigger platform as a university with potential collaborations as we seek Carnegie R2 designation.” 

Mema’s research targets a confluence of trends in materials development and new approaches to science. In recent years, demand has increased for new, advanced engineering materials that are stronger, durable, lightweight and cost-effective. Meanwhile, artificial intelligence tools have transformed science across the board, including the search for advanced materials.  

Traditionally, scientists have needed lab space, equipment and time to develop new materials and test the materials’ response to stress, temperature, magnetic fields and other factors. The in-person approach achieved reliable results but consumed more resources than computer-based testing. Using machine learning for materials research can produce less expensive solutions more quickly while still achieving the same reliable results.  

Understanding microstructures is a crucial part of materials research. Microstructures’ formation is what leads and guides the behavior of the overall material, Mema said.  

“They play a role in material’s overall strength. How brittle it is. Whether we can twist it,” she said. “The evolution of microstructure at such a small scale guides the behavior of material we see with human eyes.” 

Microstructures present a complex challenge, even for materials scientists leveraging advanced machine learning tools. While traditional computer modeling methods are available for studying microstructures, they are often costly and fail to accurately simulate real-world material behavior. 

Mema and her Army Research Lab collaborators have adapted deep learning algorithms to solve mathematical models in 1D and 2D that exhibit similar features to mathematical models with microstructure. 

“This will lead to quicker and reliable solutions to the predictive computational models of microstructure,” Mema said. “Once we have a reliable algorithm, we can then solve and gain insight into the actual behavior and evolution of the microstructures. Then the question becomes, can we use the algorithm to predict behavior at a small scale? If so, we could run simulations and gain insight into the behavior of the overall material under various stress conditions.”