Research cooks up stronger, more resilient engineering materials

If you’re baking a pie, it’s best to follow a tried and true recipe. You measure precise amounts of ingredients, mix them in a particular order, and bake.

But what if you change a variable or two, like baking the pie for two hours instead of one? Will it yield the same result? 

Similarly, for researchers using high-level prediction models to discover new materials, a slight change in processing methods can dramatically affect the outcome of a recipe. But artificial intelligence (AI) has made it easier than ever to accurately predict the outcome.

In a new commentary in Nature Reviews Materials, researchers from Virginia Tech and Johns Hopkins University highlight how AI and machine learning are significantly increasing the rate of discovery for new materials and alloys by analyzing what happens when you tweak variables in processing techniques.

Maitreyee Sharma Priyadarshini, assistant professor in the Kevin T. Crofton Department of Aerospace and Ocean Engineering, is a co-author on the commentary. With research expertise in machine learning, hypersonics, and computational materials discovery, she explained how most prediction models often overlook the critical role that the processing method plays in the development of new materials.

What is materials discovery? 

Just as a pie depends on the right ingredients and proportions, materials are combinations of elements whose properties depend on composition. For example, solar cell materials must maximize energy conversion efficiency, while hypersonic materials must withstand temperatures up to 3000 kelvins without eroding away. Materials discovery is essentially “cooking” the best material for an intended application.

How has the use of AI and machine learning changed the field of materials discovery? 

The manner in which a material is processed is just as important as the composition. You can have the right ingredients in the right proportions, but if you bake the pie at the wrong temperature or length of time, it won’t turn out well. Similarly, materials must be processed under the right conditions.

The challenge is that processing variables are continuous: temperatures can range from 50 degrees Celsius to 500 degrees Celsius, and the amount of time you bake the materials can range from minutes to days. The number of possible combinations is enormous, making exhaustive experimental testing essentially impossible. This is where AI and machine learning help. By analyzing existing data, our computational models can recommend promising processing conditions, such as specific temperatures and times, that are likely to yield better material properties.

The problem is akin to finding a needle in a haystack. We need some sort of a guiding light in this really wide space, telling us to go in a specific direction to find the optimal recipe. Using AI methods gives us that guidance, offering a better way of determining the right conditions for processing as well as the optimal ingredients to develop a chosen outcome.  

What excites you most about using AI tools in your research?

What we have noticed is that AI often provides recommendations outside of our chemical intuition. As researchers working with a particular material for a long time, we might expect one result, but the AI suggests something completely different. When we test it, the properties are better than anything we’ve seen before. 

We don’t exactly know why it is working so well. Trying to understand “the why” so that we can learn from and improve our methods is something that is very, very exciting for me. 

And this leads to development of new materials that are stronger, more durable, or more resistant than anticipated?

Exactly. For example, for structural applications, we discovered materials not found in literature with mechanical strength exceeding what was seen in our training datasets, as shown in our recent arXiv article. In the Nature Review article, we discuss biodegradable metals for implants, which need to remain strong and hold up against extremely corrosive biological media. Our models identified processing conditions that produced higher hardness and lower corrosion rates than expected. These “out‑of‑distribution” discoveries save time, money, and resources while yielding exotic materials with superior properties.

What current research are you working on at Virginia Tech?

We're developing frameworks that optimize multiple material properties simultaneously. Past methods focused on a single objective, such as mechanical strength, but in aerospace applications, materials must balance strength, oxidation resistance, and other factors. Multi‑objective optimization is a key direction.

My group is also building multi-physics frameworks and high‑fidelity models for material-flow interactions for hypersonic applications. For example, we study high-temperature ceramics used in hypersonics, analyzing how they degrade under extreme conditions and integrating that knowledge into large‑scale computational fluid dynamics simulations. This helps predict how spacecraft or aircraft materials will perform at hypersonic speeds.

This research is surprising and exciting, showing the power of AI to help uncover solutions beyond traditional intuition.