How machine learning can speed up the discovery of new minerals

According to new research, machine learning could help develop new types of materials with valuable properties such as resistance to extreme temperatures and resistance to rust. Properties that can be useful for a whole range of sectors: for example, metals that resist extremely low temperatures can make it possible to improve spacecraft. Corrosion-resistant metals can be used in boats and submarines.

Scientists today usually perform experiments in the lab to find ways to combine metals to create new ones. They start with a known element, such as iron, which is cheap and malleable, and add one or two more to see the effect on the parent material. It is a tedious process with many trial and error that inevitably leads to more hiccups than meaningful results.

The new articlePublished in the magazine Science This month suggests that by using artificial intelligence (AI), researchers can more accurately predict which groups of minerals will bode well.

Using this method, researchers at the Max Planck Institute have succeeded in identifying 17 new minerals with promising properties. The team was looking for low-level “Invar” minerals. It is the degree to which materials expand or contract when exposed to high or low temperatures.

Read also

Get ready for the next generation of AI

Metals with low Invar do not change in size even when faced with extreme temperatures. They are commonly used in industries such as natural gas transportation and storage, says Xiuan Rao, a materials science researcher at the Max Planck Institute and co-author of the paper published in Science.

The team was able to find these new minerals through a combination of laboratory experiments and artificial intelligence. It first had to overcome a major challenge: the lack of existing data to train machine learning models. To do this, they used the several hundred data points available to them to describe the properties of existing metal alloys. The AI ​​system used this data to predict which new minerals would show lower Invar.

The researchers then created these minerals in the lab. They evaluated the results before returning them to the machine learning model. The process continued like this – the model came out with groups of minerals, the researchers tested them and relayed the data – until 17 new minerals with promising properties appeared.

These findings could pave the way for greater use of machine learning in materials science. An area that still relies heavily on laboratory experiments. In addition, the technique of using machine learning to make predictions that are then verified in the lab could extend to other fields such as chemistry and physics, materials science experts believe.

To understand why this technology is such an important development, it’s worth looking at the traditional method for creating new compounds, says Michael Titus, assistant professor of materials engineering at Purdue University, who was not involved in the study in question. The DIY lab process is laborious and inefficient.

“It’s really like looking for a needle in a haystack,” Michael Titus compares. The latter often tells his new students that there are at least a million potential new substances waiting to be discovered. Machine learning can help researchers determine what triggers it to follow.

Izu George, a University of Tennessee materials science and engineering professor who was not involved in this research, was surprised by what the team was able to accomplish with this new technology.

“It’s very impressive,” he admits.

Going forward, the research team wants to use machine learning to discover new alloys with more than one desired property. Ezo-George agrees that computational-based methods will be critical to the future of materials science.

“It is possible that the machine learning approach will be dominant because people have tried to scan very large areas experimentally, but it is very time-consuming and expensive,” he explains. “The challenge is whether we find something useful.”

Article by Tami Shaw, translated from English by Cozy Pastakia.

Read also

How robotic bees can help this insect survive