Read the news 👉 Artificial intelligence discovers the ‘Toledano steel’ of the future 👈
As the research team led by Ziyuan Rao states in their publication, in the past, alloy discovery had a significant serendipity component. In order to improve this search process, thermodynamic design rules began to be used, which had major limitations when it came to extrapolating results. Now, thanks to the work of this group of researchers, the power of artificial intelligence has been put at the service of high entropy alloys, opening up a new panorama in this area of knowledge.
In the field of new materials, it is common to take years to obtain new candidates that meet certain attractive specifications for industry or research. This is because, if we think of all the possible combinations of elements typically used in alloys, we would have up to 1050 options, and as almost always when we encounter problems that “explode” at the combinatorial level, AI is an ideal tool to respond to these challenges. In this way, the arrival of artificial intelligence in this process will reduce these times, the associated costs and will serve as a catalyst for obtaining materials with better properties such as low coefficients of thermal expansion or corrosion behaviour.
How did they do it?
The team collected data from many different alloys and trained a Machine Learning model that identified relationships between properties and elements present in the alloys. Given that the approach has the limitations of the data that have been incorporated into these models, an iterative process is incorporated that makes use of active learning through which data from promising candidates for which there is no data, theoretical calculations (thermodynamic and DFT (density-functional theory)) and re-training on the dataset are incorporated.
What are their results?
In the two to three months required for the entire process, the researchers have discovered two high-entropy Invar alloys with extraordinarily low coefficients of thermal expansion.
This discovery highlights the suitability of artificial intelligence as a tool to support the discovery of new materials and opens up a promising horizon in this field. It is also true that the discovery is only the first step on a much longer road, as synthesising or producing the material will present its own challenges.
