UNIVERSITY PARK, PA – The person looking back from the computer screen may not actually exist, thanks to artificial intelligence (AI) capable of generating compelling but ultimately fake images of human faces. Now that same technology could fuel the next wave of innovations in material design, according to scientists at Penn State.
“We hear a lot about deepfakes in the news today – an AI that can generate lifelike images of human faces that don’t match real people,” said Wesley Reinhart, assistant professor of human science and engineering. materials and professor at the Institute for Computational and Data Sciences. co-hire, at Penn State. “This is exactly the same technology that we used in our research. We are simply exchanging this example of images of human faces for elementary compositions of high performance alloys.
Scientists have formed a Generative Antagonist Network (GAN) to create new high-entropy refractory alloys, materials capable of withstanding ultra-high temperatures while maintaining strength and which are used in technology from turbine blades to rockets. .
“There are a lot of rules about what makes an image of a human face or what makes an alloy, and it would be really hard for you to know what all of those rules are or to write them down by hand,” said Reinhart. “The premise of this GAN is that you have two neural networks that are essentially competing with each other to learn what those rules are, and then generate examples that follow the rules.”
The team combed through hundreds of published examples of alloys to create a training dataset. The network has a generator that creates new comps and a reviewer who tries to determine if they look realistic to the training dataset. If the generator is successful, it is able to make alloys that the reviewer believes are real, and as this adversarial game continues over many iterations, the model improves, the scientists said.
After this training, the scientists asked the model to focus on creating alloy compositions with specific properties that would be ideal for use in turbine blades.
“Our preliminary results show that generative models can learn complex relationships to generate novelty on demand,” said Zi-Kui Liu, Dorothy Pate Enright professor of materials science and engineering at Penn State. “It’s phenomenal. This is really what we are missing in our general materials science computing community.
Traditional or rational design has relied on human intuition to find patterns and improve materials, but this has become more and more difficult as the chemistry and processing of materials become more and more complex, have said the researchers.
“When dealing with design issues, you often have dozens, if not hundreds, of variables that you can change,” Reinhart said. “Your brain just isn’t wired to think in 100-dimensional space; you can’t even visualize it. So one thing this technology does for us is compress it and show us patterns that we can understand. We need tools like this so that we can even tackle the problem. We simply cannot do it by brute force.
The scientists said their findings, recently published in the Journal of Materials Informatics, showed progress towards reverse alloy design.
“With a sound design, you have to take each of these steps one at a time; do simulations, check tables, consult other experts, ”Reinhart said. “The reverse design is essentially managed by this statistical model. You can ask for a material with defined properties and get 100 or 1000 compositions that could be suitable in a few milliseconds.
The model isn’t perfect, however, and its estimates have yet to be validated with high-fidelity simulations, but scientists have said it takes the guesswork out of it and offers a promising new tool for determining which materials to try.
Other researchers on the project were Allison Beese, associate professor of materials science and engineering and mechanical engineering; Shashank Priya, associate vice president of research and professor of materials science and engineering; Jogender Singh, Professor of Materials Science and Engineering and Senior Engineering Scientist; Shunli Shang, professor-researcher; Wenjie Li, assistant research professor; and Arindam Debnath, Adam Krajewski, Hui Sun, Shuang Lin and Marcia Ahn, doctoral students.
The Department of Energy and Advanced Energy-Research Projects Agency funded this research.