AI predicts structural properties of metamaterials in a new study
Researchers used artificial intelligence to predict certain properties of combinatorial mechanical metamaterials.
The example of a combinatorial problem given was determining which side of an origami could be flattened without damage. A combinatorial problem is an algorithm related to a given number of elements.
Using machine learning to discover properties of metamaterials
The new study was led by the UvA Institute of Physics and research institute AMOLF. The research was published in the journal Physical Review Letters.
Metamaterials are materials that do not occur in nature and whose properties result from artificial structure rather than chemical composition. The research conducted by UvA and AMOLF showed that machine learning algorithms can accurately answer questions, such as the combinatorial issue with the origami. The team tested how well artificial intelligence could predict the properties of certain metamaterials.
The combinatorial mechanical materials are engineered materials. The origami mentioned is considered a type of metamaterial with the ability to be flattened determined by how it’s folded, or its structure, rather than the material it was made with.
The researchers also mentioned that smart design lets them control where and how a metamaterial will bend, allowing it to be used for a variety of things, from “shock absorbers to unfolding solar panels on a satellite in space.”
Building blocks of metamaterial
In the study, researchers studied a typical piece of combinatorial metamaterial built with two or more building blocks, which collapse when a specific amount of force is applied. In order for all of the building blocks to buckle and not jam or get stuck when pressure is applied, the contorted building blocks need to fit together like puzzle pieces. Changing just one block in the metamaterial can make bendable metamaterial inflexible and sturdy.
Although metamaterial can be applied and used in various ways, designing a new metamaterial proves to be difficult for the research team. Creating the metamaterial properties for different structures comes down to trial and error. “In this day and age, we do not want to do all of this by hand.
However, because the properties of combinatorial metamaterials are so sensitive to changes to individual building blocks, conventional statistical and numerical methods are slow and prone to mistakes,” the researchers stated.
Neural networks predict with accuracy
Due to the concept of trial and error, the researchers discovered that machine learning could be accurate and efficient. They found out that neural networks could accurately predict metamaterial properties of any pattern or structure with great detail and exactness. The machine learning was able to learn the complexities of metamaterial properties.
“This far exceeded our expectations,” said Dr. Ryan van Mastrigt, the first author of the study. “The accuracy of the predictions shows us that the neural networks have actually learned the mathematical rules underlying the metamaterial properties, even when we don’t know all the rules ourselves.”
This finding suggests that artificial intelligence could be used to design new complex metamaterials with suitable properties in the future. Also, in the study, the researchers said the findings could improve the comprehension of neural networks and demonstrate the problems that they can solve.
Propulsion technology has matured considerably, culminating in the New Space Age today.