Can you flatten a 3D origami object without causing it harm? Because every fold in the design has to be compatible with flattening, it is difficult to guess the solution just by looking at it.
An example of a combinatorial issue is this. Recent studies conducted by the UvA Institute of Physics and the research centre AMOLF have shown that machine learning algorithms can swiftly and correctly respond to inquiries of this kind. Artificial intelligence-assisted design of complicated and useful (meta)materials is anticipated to benefit from this.
The study team examined how effectively artificial intelligence (AI) can anticipate the characteristics of so-called combinatorial mechanical metamaterials in their most recent work, which was published in Physical Review Letters this week.
artificial components
These materials are designed, and rather than their chemical makeup, their geometrical structure governs how they behave. The capacity of an origami piece to flatten (a physically well-defined attribute) is governed by the way it is folded (its structure), not by the kind of paper it is composed of. Origami is another sort of metamaterial. In a broader sense, the smart design enables us to precisely regulate where or how a metamaterial will bend, buckle, or bulge, which may be employed for a variety of applications, from shock absorbers to unfolding solar panels on a spacecraft in orbit.
A typical combinatorial metamaterial under study in the lab is composed of two or more different kinds of orientations of building blocks, each of which deforms differently in response to an external mechanical force. Because not all of the building pieces will be able to bend in the desired manner when put together randomly, the material as a whole typically won’t give way under pressure. Instead, it will jam.
A neighbouring construction block should be able to squeeze inward where one wants to protrude outward. All of the distorted construction components must fit together like a jigsaw puzzle for the metamaterial to quickly buckle. A ‘floppy’ metamaterial may become stiff by modifying a single block, much as altering a single fold can render an origami piece unflattenable.
elusive to foretell
Although there are several possible uses for metamaterials, creating a new one is difficult. It typically comes down to trial and error to determine the general metamaterial characteristics of various structures starting from a specific set of building blocks. We don’t want to carry out all of this work by hand in the modern world. However, typical statistical and numerical approaches are sluggish and prone to errors because the characteristics of combinatorial metamaterials are so sensitive to changes to individual building blocks.
Instead, they discovered that machine learning could hold the key: even when given a sparse sample size to train from, so-called convolutional neural networks are able to precisely anticipate the metamaterial characteristics of any arrangement of building blocks. Ryan van Mastrigt, the first author and a doctoral student, adds, “This well surpassed our expectations.” Even if we don’t fully understand all of the mathematical principles behind the metamaterial qualities, the forecasts’ accuracy demonstrates that the neural networks have really mastered them.
This study shows that artificial intelligence (AI) may be used to create new complex metamaterials with valuable features. More generally, we can ask a lot of intriguing questions by using neural networks to solve combinatorial problems. They could help us solve (combinatorial) issues in other situations. Conversely, the discoveries may deepen our knowledge of neural networks by, for example, showing how a neural network’s complexity corresponds to the complexity of the issues it can handle.
