A team of MSU students and postgraduates presented the MaterialsGen AI platform for managing the stages of R&D of multicomponent materials, originally developed for concrete mixes.
One of the key tasks of the platform is to predict the properties of concrete regulated by the requirements of GOST and SNiP (compressive strength, mobility) based on the initial data on the composition of the mixture: specific gravity of sand, crushed stone, cement, water and various additives per 1 m3. Properties are predicted for any "age" of concrete from 1 to 90 days after mixing, which allows you to set the time to reach the required ranges of properties even before the experiment in a matter of seconds.
"The traditional method of testing concrete mixes requires 28 days of hardening a concrete cube to assess the final strength, while there is an acute shortage of highly qualified technologists to work with complex compounds, and the work of laboratories and workshops in production facilities is usually not digitized," says Maxim Smirnov, project manager, graduate student at the Chemical Faculty of Lomonosov Moscow State University.. – In addition, there is a systematic overspending of raw materials due to unstable quality and different sources of raw materials. It turns out that the current cycle of expert formulation selection takes a month of manual labor, while its accuracy cannot be called satisfactory, it is only 64% (the percentage of actual concrete strength in the range predicted by an expert technologist on a sample of more than a thousand compounds)."
The second and main task of the project is the generation of new formulations of concrete mixtures (calculation of the ratio of components and determination of the mixing technique) according to the specified ranges of desired properties. This significantly expands the space of known formulations and allows us to offer new formulations for various climatic conditions and applications, including 3D printing.
The platform is based on Bayesian models of artificial intelligence and neuroevolutionary algorithms that provide prediction of properties and generation of formulations with a small amount of training data. The project also developed a number of application modules aimed at digitalizing processes within the production circuit, including an electronic laboratory journal and a chat with an AI technologist. The solution was created by a team within the framework of the Design Laboratory of the Intellect Foundation.
During the project, some of the formulations generated by the MaterialsGen platform were experimentally validated at our partner's production site. During the 28-day test cycle, the recipes gained compressive strength in the range of 80-99% of the predicted.
Using the system allows you to close the blind spots of traditional ERP and LIMS systems used in many Russian industries. The team is currently implementing a pilot solution on large Russian banks with an estimated IRR of 144% and a payback period of one calendar year with an economic impact of 41.5 million rubles per year.
The MaterialsGen project took the 3rd place at the V Youth Competition of Innovation and Entrepreneurship of the SCO countries, which was held in Qingdao (China). The expert jury of the competition noted the high level of project development, the technological viability of the service and its applied importance for accelerating and optimizing research in the field of new materials. Special attention was paid to the potential for further development of the platform and the possibilities of its application. According to Maxim Smirnov, the universal AI core of MaterialsGen can be adapted to any multicomponent mixtures in other industries (for example, in non-ferrous metallurgy, agriculture, medicine and biotechnology). The team is open to finding investors and business customers to develop solutions for optimizing multicomponent mixtures and their formulations, as well as digitalizing production.
In the Design Laboratory, MSU students and postgraduates are working on practice‑oriented solutions based on AI and digital technologies. Each project is implemented under the guidance of industry mentors, allowing you to combine a strong scientific component and business logic, which implies the possibility of rapid implementation and obtaining measurable results in a short time.
The MaterialsGen project team consists of Maxim Smirnov, Project Manager, Graduate student of the Chemistry Faculty of Lomonosov Moscow State University, Roman Chupin, Product Manager, graduate of the Economics Faculty of OmSU, Mikhail Nikulin, ML Engineer, Junior Researcher at Lomonosov Moscow State University, graduate of the Faculty of Mechanics and Mathematics; Babken Beglaryan, Responsible for R&D Materials – postgraduate student at the Faculty of Chemistry of Lomonosov Moscow State University; Imil Mamin, ML- engineer – 4th year student at the Faculty of Physics of Lomonosov Moscow State University.
Reference
The non-profit Foundation for the Development of Science and Education "Intellect" provides grant and scholarship support to students and young scientists of Moscow State University, promotes the creation of new educational courses and programs, increases the number of high-level publications and the development of infrastructure for educational and scientific activities. The Foundation's mission is to support science and education in the field of artificial intelligence and its application in scientific research. The Intellect Foundation was established in 2020 by Oleg Deripaska, the founder of the Volnoe Delo Foundation, a graduate of Lomonosov Moscow State University.
