MIT’s AI system reveals the internal structure of materials from surface observations

Maybe you can’t tell a book from its cover, but according to researchers at MIT you may now be able to do the equivalent for materials of all sorts, from an airplane part to a medical implant. Their new approach allows engineers to figure out what’s going on inside simply by observing properties of the material’s surface.

The team used a type of machine learning known as deep learning to compare a large set of simulated data about materials’ external force fields and the corresponding internal structure, and used that to generate a system that could make reliable predictions of the interior from the surface data.

The results are being published in the journal Advanced Materials, in a paper by doctoral student Zhenze Yang and professor of civil and environmental engineering Markus Buehler.

“It’s a very common problem in engineering,” Buehler explains. “If you have a piece of material — maybe it’s a door on a car or a piece of an airplane — and you want to know what’s inside that material, you might measure the strains on the surface by taking images and computing how much deformation you have. But you can’t really look inside the material. The only way you can do that is by cutting it and then looking inside and seeing if there’s any kind of damage in there.”

One potential application of the new method is nondestructive testing; you no longer have to open a metal pipe, for instance, to detect defects. Credit: Courtesy of the researchers

It’s also possible to use X-rays and other techniques, but these tend to be expensive and require bulky equipment, he says. “So, what we have done is basically ask the question: Can we develop an AI algorithm that could look at what’s going on at the surface, which we can easily see either using a microscope or taking a photo, or maybe just measuring things on the surface of the material, and then trying to figure out what’s actually going on inside?” That inside information might include any damages, cracks, or stresses in the material, or details of its internal microstructure.

The same kind of questions can apply to biological tissues as well, he adds. “Is there disease in there, or some kind of growth or changes in the tissue?” The aim was to develop a system that could answer these kinds of questions in a completely noninvasive way.

Achieving that goal involved addressing complexities including the fact that “many such problems have multiple solutions,” Buehler says. For example, many different internal configurations might exhibit the same surface properties. To deal with that ambiguity, “we have created methods that can give us all the possibilities, all the options, basically, that might result in this particular [surface] scenario. “

The techniques they developed involved training an AI model using vast amounts of data about surface measurements and related internal properties. This includes not only uniform materials, but also combinations of different materials. “Some new planes are made of composite materials, so they are intentionally designed to have different topologies,” Buehler says. “And of course, in biology, all kinds of biological substances are made up of multiple components, and they have very different properties. For example, bones have very soft proteins, and then very contains hard mineral matter.”

The technique works, he says, even for materials whose complexity is not fully understood. “For a complex biological tissue, we cannot understand exactly how it behaves, but we can measure its behavior. increase.”

Yang says the method they developed has broad applicability. “This is not limited to solid mechanics problems, but can be applied to fluid mechanics and many other types of engineering disciplines,” Buehler said. They add that it can be applied to determine various properties of magnetic fields, such as magnetic fields in fusion reactors. It is “very universal, not only for different materials, but also for different fields.”

Yang says he first started thinking about this approach when he was studying data for materials that were partly blurred in the images he was using, and that he wanted to “fill in the blanks” for missing data in the image. “I wondered how this could be done,” he said. Blurry area. “How can I recover this lost information?” he wondered. Further reading revealed that this is an example of the widespread problem of trying to recover lost information, known as the inverse problem.

Developing this method involved an iterative process of having the model make preliminary predictions, comparing it to actual data on the material in question, and further fine-tuning the model to match that information. . The resulting models were tested for cases where the materials are well understood enough to calculate their true internal properties, and the predictions of the new method agreed well with those calculated properties.

The training data includes not only images of the surface, but also various other types of measurements of surface properties such as stress, electric field, and magnetic field. Researchers often used simulated data based on their understanding of the underlying structure of a particular material. And even if the new material has many unknown properties, this method can produce good enough approximations to give engineers general direction on how to pursue further measurements.

As an example of how this methodology is applied, Buehler notes that it is impractical today to inspect an entire aircraft, so expensive methods such as X-rays are used to scan several representative planes. He points out that aircraft are often inspected by inspecting sensitive areas. “It’s a different approach and a much cheaper way to collect data and make predictions,” Buehler says. “From there, you can decide where you want to look and even test with more expensive equipment.”

First, he expects the method, which is freely available to anyone through the GitHub website, to be applied primarily to laboratory environments, such as testing materials used in soft robotics applications.

“We can measure what’s on the surface of these materials, but we don’t know what’s going on inside them because they’re made of hydrogels, proteins, and biomaterials for actuators.” because there is no theory.” For that reason. So this is an area where researchers have the potential to use our technology to predict what’s going on inside and design better grippers and better composites,” he added. I was.

Reference: “Fill in the Blank: Transferrable Deep Learning Approaches to Recover Missing Physical Field Information” by Zhenze Yang and Markus J. Buehler, March 19, 2023. advanced materials.
DOI: 10.1002/adma.202301449

This research was supported by the U.S. Army Research Service, Air Force Office of Scientific Research, the Google Cloud Platform, and MIT Quest for Intelligence.