Deep learning system explores the inside of materials from the outside

Can you tell from the outside what the inside of a material looks like? That is technically possible in principle, for example with X-ray technology. Or if destruction is not an issue, the material can simply be cut open. A new method based on AI now takes advantage of the fact that much of what happens inside a material also has an influence on the surface. To do so, a team of researchers at MIT used Deep Learning to compare a large set of simulated data about the external force fields of materials with the corresponding internal structure to develop a system that can make reliable predictions about the interior based on the surface data. The results were published by PhD student Zhenze Yang and Professor of Civil and Environmental Engineering Markus Bühler in the journal Advanced Materials.

When surface structures refer to the interior

According to Markus Bühler, this is a common problem in engineering: "If you have a piece of material - maybe a car door or a part of an airplane - and you want to know what's inside the material, you can measure the strains on the surface by taking pictures and calculating how much deformation you have. But you can't really look inside the material. You can only do that by cutting it and then looking inside to see if there's any damage there." X-ray technology, on the other hand, is expensive and requires bulky equipment. "So we basically asked ourselves the question: Can we develop an AI algorithm that looks at what's going on on the surface, which we can easily see either with a microscope or a photograph, or just measure things on the surface of the material, and then try to figure out what's going on inside?" This internal information could include damage, cracks or stresses in the material, or details of the internal microstructure. The same kind of questions can apply to biological tissue, Markus Buehler adds. "Is there a disease there, some kind of growth or changes in the tissue?" The goal was to develop a system that could answer these kinds of questions in a completely non-invasive way.

Tracking down the inner life of materials with deep learning system

"To achieve this goal, we had to deal with complex issues, including the fact that there are multiple solutions to many of these problems," says Bühler. For example, many different internal configurations can have the same surface properties. To deal with this ambiguity, "we developed methods that gave us all the possibilities, basically all the options that could lead to this particular [surface] scenario."

The technique they developed involved training an AI model using large amounts of data on surface measurements and their associated internal properties. This included not only uniform materials, but also those containing different materials in combination. "Some new aircraft are made of composite materials, so they are intentionally made of different phases," Buehler says. "And, of course, in biology, too, any kind of biological material is made of several components that have very different properties, as in bones, where there are very soft proteins and very rigid minerals."

Widely applicable method

The technique even works for materials whose complexity is not yet fully understood, says Markus Bühler. "With complex biological tissue, we don't understand exactly how it behaves, but we can measure the behavior. We don't have a theory for it, but once we've collected enough data, we can train the model."

Zhenze Yang says the method they developed has broad applicability. "It is not limited to problems in solid mechanics, but can also be applied in other engineering disciplines such as fluid dynamics and other fields." Buehler adds that it can be used to determine a wide range of properties, not only stress and strain, but also fluid or magnetic fields, such as the magnetic fields in a fusion reactor. It is "very universal, not only for different materials, but also for different disciplines."

Yang says he first thought about this approach when he was examining data on a material where part of the images he was using were out of focus, and he wondered how it might be possible to "fill in" the missing data in the blurred area. "How can we recover this missing information?" he wondered. As he read further, he realized that this was an example of a common problem known as the inverse problem, which attempts to recover missing information.

How the deep learning system for material properties was developed

The development of the method was an iterative process in which the model made preliminary predictions, compared them to actual data about the material in question, and then further refined the model to incorporate this information. The resulting model was tested on cases where the materials were known well enough to calculate the actual internal properties, and the predictions of the new method matched well with the calculated properties.

Training data included images of the surfaces, as well as various other measurements of surface properties, including stresses and electric and magnetic fields. In many cases, the researchers used simulated data based on an understanding of the underlying structure of a particular material. And even if a new material has many unknown properties, the method can produce an approximation good enough to give engineers a general direction for further measurements.

The two researchers assume that this method, which is available via the website GitHub is freely accessible to everyone, will initially be applied primarily in laboratory environments, for example when testing materials for soft robotics applications.

Source: Techexplore.com