A team of young SFU researchers used machine learning algorithms to analyze the spectrum of thermal (infrared — IR) rays. To train artificial intelligence, a database was created containing 470 simulated infrared spectra for 230 different zeolites — promising nanoporous materials and a number of their modified crystal structures. Based on these data, connections are established between structural details and spectral features that can be used to quickly determine the type of zeolite and modifications of its structure during technologically important processes, based only on the analysis of IR spectra. Due to the availability and low cost of IR spectroscopy, the proposed method will be useful in petrochemistry, medicine and biotechnology.

Zeolites are porous materials that are widely used in various fields, particularly as catalysts in important reactions in the petrochemical industry.

The depletion of traditional energy sources and the desire of society to develop “green” chemistry requires constant improvement of reactions, the catalysts themselves, as well as an understanding of the processes occurring with these materials at the atomic level during their operation. To optimize catalytic processes, an unambiguous determination of the structural features of the catalyst is required, and some of the methods used in production for the synthesis of zeolites are Raman spectroscopy and infrared spectroscopy. These methods make it possible to establish what types of bonds, functional groups and structural units are present in the material. However, firstly, they do not allow a quantitative assessment of the structural features, and secondly, in a number of cases, a clear correspondence between the observed vibrational modes and the structural units of zeolites has not yet been established.

In a recent study, young scientists of the International Research Institute of Smart Materials of the Southern Federal University of Physics and Mathematics. Oleg Usoltsev , Doctor of Physical and Mathematical Sciences Aram Bugaev , Ph.D. Sergei Guda and master’s student Bogdan Protsenko under the guidance of Ph.D. Alina Skorynina studied porous materials and established how changes in the structure of zeolites affect their vibrational spectra. To do this, scientists calculated the theoretical basis of infrared spectra and applied machine learning algorithms to it.

The scientists created a database of 470 simulated IR spectra of zeolites, correlated with their structural parameters. They then used machine learning to predict structural elements and establish correlations with the spectrum. Based on their results, they proposed a new and improved mechanism for studying porous materials using infrared spectroscopy and machine learning.

“Our method can be useful for studying zeolites in situ/operando (under real technological conditions), that is, allowing one to monitor the evolution of the structure in real time in a controlled atmosphere during catalytic reactions with control of the output products. The research results can be applied in various industries such as petrochemicals, medicine and biotechnology. This is due to the availability and low cost of infrared spectroscopy, for which the proposed approach was developed,” said Alina Skorynina, Ph.D., research engineer at the Moscow Institute of Mathematics and Mathematics.

In addition, the development will allow us to limit ourselves at the stage of initial diagnostics of zeolites to measuring IR spectra and eliminate more labor-intensive methods, which will significantly speed up the process of analyzing ongoing structural changes.

“The main difference between this study and previous searches for correlations between spectral data and structural features is that a large set of different zeolites was considered, consisting of 230 known porous structures with their uniformly calculated theoretical spectra. The proposed solution to the classification problem makes it possible to determine the presence of structural subunits only from the vibrational spectrum. The solution of the regression problem establishes the relationship between some structural features and observed vibrations in the frequency range 1400–200 cm–1. The established correlations can be especially useful for in situ observations of the evolution of zeolite frameworks under external conditions (temperature, pressure, reaction substrate) using IR spectroscopy,” noted Alina Skorynina.

The project “Establishing the patterns of vibrational spectra of zeolites using quantum chemical modeling and machine learning methods” is carried out with the support of the Russian Science Foundation (project No. 22-23-01). The results of the first year of research are presented in an article in the highly rated journal Inorganic Chemistry (Q1).

According to scientists, in the future it is planned to expand the database with Raman spectra for the same zeolite structures, as well as to include experimental data for natural and synthetic zeolites in the database, which will improve the accuracy of structure prediction.

The team of the International Research Institute of Smart Materials of Southern Federal University is actively developing research in the field of creation and nanodiagnostics of new materials, takes leading global positions in the field of X-ray absorption spectroscopy and takes an active part in the implementation of major federal programs. In particular, scientists are working on the project “Full-cycle technologies for the express development of functional materials under the control of artificial intelligence” of the “Priority 2030” program (National Project “Science and Universities”) of SFU, which will make it possible to achieve a big breakthrough in domestic science.