Russian researchers have developed a machine learning system capable of significantly accelerating the development of new superhard materials based on boron and tungsten compounds. According to the press service of the Artificial Intelligence Institute AIRI, the new AI model made it possible to calculate the properties of a large number of variants of higher tungsten boride doped with various metals in just a few days, and to select the most promising ones.
The created system is not limited to working only with borides. It can also be used for other families of functional materials. This approach makes it possible to quickly and efficiently select structures for further laboratory research, bypassing the lengthy and resource-intensive quantum chemical calculations that are traditionally used in chemistry and materials science.
Together with specialists from Sber, Tomsk Polytechnic University, and Skoltech, scientists implemented a neural network model based on graph neural networks (GNN). These networks represent molecules and crystals as mathematical graphs, which allows for more accurate prediction of their properties. The team used an active learning method: only those structures on which the neural network made the largest errors were included in the training dataset. This reduced the computational complexity of the task and ensured high quality predictions based on only 200 training structures.
As a result, the neural network was able to evaluate the properties of about 375,000 compound variants, and the most promising was tungsten pentaboride doped with tantalum in a concentration of 20 to 60%. The synthesis of this material in the laboratory of Tomsk Polytechnic University confirmed its improved mechanical properties, which became an important proof of the effectiveness of using AI in the development of new superhard materials.
Read more on the topic:
Rostec Presents Russian Platform for Quantum Chemical Modeling at CIPR-2025 Conference
New Boron and Tungsten Based Catalyst Developed in Russia
New Class of Materials for Hydrogen Production Synthesized in Tomsk