ASU researchers to unlock materials design efficiency

Yi “Max” Ren and Yang Jiao received an NSF award to reveal how a material’s microstructure influences its properties

Materials scientists can be likened to Gordon Ramsay.

Like Ramsay uses beef tenderloin, pâté, duxelles and puff pastry to make his famous beef Wellington dish, materials scientists combine various materials to make systems for special applications.

Just as different factors, such as the ingredients and cooking techniques, can influence the deliciousness of a meal, there are key elements, otherwise referred to as materials design variables, that influence the behavior of a complex materials system. An example of such a variable is the amount of metal in a polymer matrix.

Unlike cooking, due to the complex and disordered nature of materials systems, it’s nearly impossible to predict the exact variables that influence their properties. This limitation prevents researchers from creating new materials systems for emerging industries such as robotics and space exploration.

Yang Jiao and Yi “Max” Ren have been awarded a $465,000 National Science Foundation grant to uncover how a materials system’s random microstructure influences its macro-level properties, unlocking the ability to design novel products efficiently.

Ren is an associate professor of aerospace and mechanical engineering in the School for Engineering of Matter, Transport and Energy, part of the Ira A. Fulton Schools of Engineering at Arizona State University. He develops machine learning models for a wide range of applications.

Jiao, a professor of materials science and engineering in the Fulton Schools, is a computational material theorist, focusing on mathematical frameworks for designing materials systems.

The two researchers are teaming up to develop a new set of artificial intelligence, or AI, tools that can predict how a materials system’s internal structure influences its properties, making materials design more efficient and affordable. By dramatically reducing the time and cost of materials discovery, their work could accelerate breakthroughs that make renewable energy cost less, electronics more powerful and life-saving medical devices more accessible to people worldwide.

Their work focuses on solving three key challenges: Predicting the microstructure needed to achieve a specific materials property; determining a materials system’s properties by analyzing its microstructure; and explaining the physics rules governing the relationship between a materials system’s microstructure and its properties.

Making materials design less complex

Ren compares the materials design process to making bread. Just like you can control how much salt goes into making focaccia but can’t control exactly where in the bread each grain ends up, there are variables that influence material properties but remain impossible to control.

Take a resin-metal composite as an example. If the desired property is conductivity, then how a conductive metal, such as aluminum, gets distributed throughout a polymer matrix, such as polyester, determines the composite’s overall conductivity. However, aluminum’s physical distribution is not the only variable influencing the system’s conductivity.

“We’re embarking on a mission to find a delicate way to quantify other factors influencing a material’s properties beyond what we can see,” Jiao says. “We plan to use large language models to provide a physical interpretation of exactly what’s going on.”

Physics fizzling out

Ren explains that current physics theories used to predict properties have strong limitations.

“The existing statistical physics theories for predicting a composite’s overall properties rely on assumptions that often fail in reality,” Ren says. “They also don’t apply to several of the complex materials systems that we are interested in.”

Jiao and Ren state that existing equations treat microstructural arrangements as having infinite possible configurations, making it computationally expensive to test theoretical predictions against experimental results.

To address this challenge, the team plans to combine AI and first-principle physics models to reliably predict and explain only the most influential variables for specific properties. By narrowing the focus from infinite possible configurations to the most influential variables, the team could design materials to create new kinds of products at a fraction of the current cost.

Their ambitious plan comes with considerable challenges.

“Currently, AI models, such as OpenAI’s o3, are unable to perform the math necessary to arrive at the same conclusions as us,” Ren says. “The current AI systems cannot reason deeply about the first principles of a materials system’s complex structure-property relationships.”

By combining the fundamental laws of physics and experimental data, the team intends to build a model with access to accurate knowledge about the inner workings of complex materials systems. This capability will enable the model to reason beyond information published in literature and identify the variables most likely to influence specific properties.

“Our goal is to build a model that is inspired by physics, but at the same time, is adaptive based on our experimental data,” Ren says. “We want a model that can work for all kinds of underlying governing equations and is not limited by existing theoretical assumptions.”

By the end of the funding’s three-year duration, the team is determined to create a set of AI tools that will not only lower materials design cost but also explain the exact relationship between a materials system’s microstructure and its properties.

“Helping us do the computation accurately and quickly is key, but more importantly, we want to develop an AI model that can interpret the mechanism behind a material’s properties,” Jiao says. “Understanding that is a critical aspect of this research.”