In their never-ending quest for better materials, researchers have found an unexpected ally — one that can scour through giant datasets with ease and compute how materials will behave at various temperatures and pressures. This ally, commonly known as Artificial Intelligence (or AI) could usher in a new age of material science.
Here’s the thing with materials: you have a lot of things that can be put together to obtain new materials with exciting properties, but it takes time, money, and effort. So instead, what researchers do before actually making a new material is creating a model of it on a computer.
The current prediction methods work well, and they’ve become quite standard — but it also takes a lot of computation power. Oftentimes, these simulations need supercomputers and can use up a lot of resources, which many researchers and companies just don’t have access to.
“You would typically have to run tons of physics-based simulations to solve that problem,” says Mark Messner, principal mechanical engineer at the U.S. Department of Energy’s (DOE) Argonne National Laboratory.
So instead, Messner and colleagues looked for a shortcut. That shortcut came in the form of AI that uncovers patterns in massive datasets (something which neural networks are particularly good at) and then simulates what happens to the material in extreme conditions using much less processing power. If it works, it’s much more efficient and fast than existing methods… but does it work?
In a new study, Messner and his team say it does.
AI, sort this out
In their new study, they computed the properties of a material 2,000 times faster than the standard modeling approach, and many of the necessary calculations could be performed on a common laptop. The team used a convolutional neural network — a relatively simple class of deep neural networks, most commonly applied to analyze images — to recognize a material’s structural properties.
“My idea was that a material’s structure is no different than a 3D image,” he said. “It makes sense that the 3D version of this neural network will do a good job of recognizing the structure’s properties—just like a neural network learns that an image is a cat or something else,” Messner said.
To put the approach to the test, Messner first designed a square with bricks, somewhat similar to how an image is built from pixels. He then took random samples of that design and used a simulation to create two million data points, which linked the design structure to physical properties like density and stiffness. These two million data points were fed into the neural network, and then the network was trained to look for the desired properties. Lastly, he used a different type of AI (a genetic algorithm, commonly used for optimization) to find an overall structure that would match the desired properties.
With this approach, the AI method found the right structure in 0.00075 seconds, compared to 0.207 seconds, which would have been the standard physics-based model. If the same ratio can be maintained for more complex computation, the approach could make it much easier for labs and companies with fewer resources to enter the material-making arena.
The potential is especially great in the field of renewable energy, where materials must withstand high temperatures, pressures, and corrosion, and must last decades. Another promising avenue is 3D printing materials — making a structure layer by layer allows for more flexibility than traditional measures, and if you can tell the machine exactly what you want it to produce.
“You would give the structure—determined by a neural network—to someone with a 3D printer and they would print it off with the properties you want,” he said. “We are not quite there yet, but that’s the hope.”
Messner and the team are even working on designing a molten salt nuclear reactor, which uses molten salt as a coolant and can operate at pressures far lower than existing nuclear reactors — but researchers first need to ensure that the stainless steel needed for the reactor will behave well under extreme conditions for decades.
The future of mechanical engineering looks bright. With ever-increasing computing power, 3D printing, and smarter algorithms, engineers can finely tune materials and produce the innovative materials industries need to thrive.
The study has been published in the Journal of Mechanical Design.
Was this helpful?