Machine learning can help create new energy materials
With the help of machine learning, expensive and time-consuming research can be made more efficient, which should help accelerate the development of new materials for green energy. In a new study, researcher Lindsay Merte shows how the process can be achieved.
In order to be able to develop green energy technologies like fuel cells that are more efficient and economical, better catalyst materials are needed. This in turn requires the ability to measure and control the surface structures of solids at the atomic level. But due to technical limitations and high costs, this is a big challenge.
Now that we know the structure, we can start looking at its chemical properties and try to figure out what makes it so good as a catalyst.
“Catalysts are a large part of the transition to green energy. If we can control how the atoms arrange themselves on solid surfaces, then we can develop better catalysts,” says Merte, associate professor of physics at Malmö University.
A catalyst is a material that accelerates a chemical reaction that is otherwise too slow to be useful. Catalysts are used in 80 per cent of all chemical production and in fuel cells that convert chemical energy into electricity. The study was conducted on an alloy of platinum and tin that acts as a catalyst. During operation, it is found that layers of tin oxide form on the surface of the alloy.
“These oxides are thought to improve the performance of the material, but their structures were not known. To understand more, we have to know how the atoms are arranged on the surface, because it is through changes there that we can influence how a catalyst works.
“For 3D materials, we have very good methods for determining atomic structures, but these cannot be used as effectively for surfaces, and so the process becomes very difficult and time consuming. With these tin oxides, we were unable to find the atomic arrangement with experiments alone,” says Merte.
The exact atomic structure is crucial for how the catalyst works, and therefore the researchers want to understand how the surfaces change when they are exposed to liquid or gas. This sometimes requires several years of experimentation, great computing power and is very costly. What Merte and research colleagues from Aarhus, Vienna and Lund have shown is how machine learning – which is a form of artificial intelligence – can be used to speed up the process.
With machine learning methods, the computer can learn to predict whether a structure will be stable or not without having to perform so many expensive simulations. This allows us to test possible structures much faster, which is what we need to find the right one.
The researchers showed how machine learning could be used to find the structure of the platinum-tin alloy. They were able to prove that it was the correct one by using atomic scale microscopy and X-ray scattering.
“Now that we know the structure, we can start looking at its chemical properties and try to figure out what makes it so good as a catalyst. And with this information, we can hopefully design catalysts that are even better,” says Merte.
Text: Magnus Jando & Adrian Grist