Given a piece of 3D origami, can you fold it without breaking it? Just by looking at the structure, the answer is difficult to predict, where every crease in the design will be suitable for flattening. This is an example of a combination problem. New research from the UvA Institute of Physics and the AMOLF research institute has shown that machine learning algorithms (AI predicts the properties of complex metamaterials) can answer such questions effectively and efficiently. This should provide an improvement in the AI-assisted design of complex and functional (meta) objects.
In their latest work, published this week in Review Letters, the research team tested how artificial intelligence (AI) can predict the properties of so-called combinatorial mechanical metamaterials.
Uses
They are technical materials whose properties are determined by their geometrical properties rather than their chemical composition. Origami paper is also a kind of metamaterial, whose breaking strength (a well defined object) is determined by the way it is folded (its structure), rather than by the type of paper it is made of. In general, intelligent design allows us to control exactly where and how the metamaterial will bend, warp or load, which can be used for everything from shock absorbers to deliver solar panels to satellites in space.
A common type of metamaterial that is studied in the laboratory consists of two or more shapes or orientations of building blocks, which flex well when a load is applied. If these building materials are combined randomly, the whole thing will not break under pressure because not all things will be able to break as they want; they will stick. Where a building needs to be pushed out, its neighbor will be able to push inward. In order for the metamaterial to transform easily, all the building blocks with deformation must fit together like a puzzle. Just as changing one ply can make a piece of origami less brittle, changing one block can make a “floppy” metamaterial stronger.
It’s hard to tell
Although metamaterials have many potential applications, creating new ones is a challenge. Starting from a basic set of building blocks, deriving 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 very sensitive to changes in individual building blocks, statistical and numerical methods are slow and prone to errors.
Instead, the researchers found that machine learning could be the answer: even if they only have a small sample to learn from, so-called neural networks are able to accurately predict features The metamaterial of each building block structure down to the finest detail.
“It far exceeded our expectations,” says PhD student and first author Ryan van Mastrigt. “Predictive accuracy shows us that neural networks have indeed learned mathematical rules underlying the properties of metamaterials, even when we don’t know all the rules ourselves.”
This research shows that we can use AI to design complex metamaterials with useful properties. More broadly, applying neural networks to problem solving raises a number of interesting questions. But maybe they can help us solve problems (combinatorial) in other situations. And on the other hand, the results can improve our understanding of the neural networks themselves, for example by showing how the complexity of the neural network is related to the complexity of the problems it can solve.
