AI predicts the properties of complex metamaterials

Given a piece of 3D origami, can you flatten it without damaging it? Just by looking at the design, the answer is hard to predict, because every crease in the design should be consistent with flattening.

This is an example of a combinatorial problem. New research from the UvA Institute of Physics and the AMOLF research institute has demonstrated that machine learning algorithms can answer such questions accurately and efficiently. This should give a boost to the AI-assisted design of complex and functional (meta)materials.

In their latest book, published in Physical examination letters this week, the research team tested to what extent artificial intelligence (AI) can predict the properties of so-called combinatorial mechanical metamaterials.

Artificial materials

They are technical materials whose properties are determined by their geometric structure rather than their chemical composition. A piece of origami is also a type of metamaterial, whose ability to flatten (a physically well-defined property) is determined by the way it is folded (its structure), rather than by the type of paper it is from. do.

More generally, intelligent design allows us to precisely control where and how a metamaterial will bend, warp or bulge, which can be used for everything from shock absorbers to deploying solar panels on a satellite in space. .

A typical combinatorial metamaterial studied in the laboratory consists of two or more types or orientations of building blocks, which deform distinctly when a mechanical force is applied. If these building blocks are combined randomly, the material as a whole will generally not deform under pressure because not all blocks will be able to deform as they wish; they will get stuck.

Where a building block wishes to bulge outward, its neighbor should be able to smash inward. For the metamaterial to deform easily, all deformed building blocks must fit together like a puzzle. Just as changing a single ply can make a piece of origami unflattened, changing a single block can make a “floppy” metamaterial rigid.

Hard to predict

While metamaterials have many potential applications, designing a new one is a challenge. Starting from a particular set of building blocks, deducing the overall metamaterial properties for different structures often comes down to trial and error. These days, we don’t want to do everything by hand. However, because the properties of combinatorial metamaterials are so sensitive to changes in individual building blocks, conventional statistical and numerical methods are slow and error-prone.

Instead, the researchers found that machine learning could be the answer: even when they have only a relatively small set of examples to learn from, so-called convolutional neural networks are able to accurately predict the metamaterial properties of any building block configuration down to the finest detail.

“It far exceeded our expectations,” says Ph.D. student and first author Ryan van Mastrigt. “Prediction accuracy shows us that neural networks have in fact learned the mathematical rules underlying the properties of metamaterials, even when we don’t know all the rules ourselves.”

This discovery suggests that we can use AI to design complex new metamaterials with useful properties. More broadly, the application of neural networks to combinatorial problems raises many fascinating questions. Maybe they can help us solve (combinatorial) problems in other contexts. And conversely, the results can improve our understanding of neural networks themselves, for example by demonstrating how the complexity of a neural network relates to the complexity of the problems it can solve.