- Статьи
- Science and technology
- Sophisticated inspection: AI will help create devices to inspect ships and reactors
Sophisticated inspection: AI will help create devices to inspect ships and reactors
Russian scientists have developed an innovative method of searching for substances that convert X-rays into visible radiation. They made their discovery on the basis of AI algorithms. These compounds are used in X-ray television equipment as materials that capture the radiation transmitted through objects. The proposed technique will help to create large-size equipment of this kind, which will make it possible to study the internal structure of large objects. For example, ships, buses, subway cars, nuclear reactors, submarines.
How inspection devices are organized
Researchers from Siberian Federal University, L.V. Kirensky Institute of Physics of the Krasnoyarsk Scientific Center of the Siberian Branch of the Russian Academy of Sciences and South China University of Technology have developed an innovative method of searching for substances that effectively convert X-rays into visible radiation on the basis of artificial intelligence algorithms. The work was carried out under a joint Russian-Chinese grant from the Russian Science Foundation.
As the scientists explained, such compounds are called scintillators. They absorb ionizing radiation (e.g., X-rays, gamma quanta, alpha particles) and emit light in the optical range. Such properties are in demand, for example, to register radiation transmitted through light-conducting objects.
In particular, in X-ray-television equipment designed to study the internal structure of materials, diagnose human and animal organs or reveal the hidden contents of cargo. The technique proposed by the scientists will make it possible to identify substances that will help to create translucent equipment in a wide range of shapes and sizes.
- Physics would not have developed without scintillators, because invisible radiation must be visualized somehow. Nowadays there is a developed line of medium class devices for this purpose. Special single crystal scintillators are grown for them. However,there is a need for the creation of and larger translucent equipment, - told "Izvestia" the developer of the method, associate professor of the basic department of solid state physics and nanotechnology SFU and senior researcher at the laboratory of crystallophysics IF SB RAS Maxim Molokeev.
He explained that such equipment is needed to detect defects in the internal structure of airplanes, train cars, ships, submarines, nuclear reactors and other large-sized equipment. It is impossible to grow crystals for devices of this size, so other approaches must be developed.
How AI is helping to discover new materials
Organometallic compounds are a promising direction for the search for scintillators, explained Maxim Molokeev. They contain both metal ions and organic molecules. Chemists have already developed a number of effective organometallic compounds, which are successfully used in practice. At the same time, they have to rely more on intuition for synthesis than on systematic data analysis.
- We are the first group to use artificial intelligence to show methods for finding such compounds. Moreover, since statistics are not enough, we used non-standard ways of preparing data for machine learning," Maxim Molokeev said.
In particular, he explained, it was hypothesized that the high brightness of radiation contributes to a certain distance between metal atoms. At the next step, the scientists collected a database of 296 open organometallic compounds with known distances between metals and used it to train the AI. The program then applied the identified dependencies to search for new compounds. Thus, the problem was solved on a large amount of data, which increased the reliability of the prediction.
Finally, the researchers tested the obtained rules and algorithm and found five promising scintillator compounds. One of them was synthesized by Nikolay Golovnev, professor of the Department of Physical and Inorganic Chemistry at SFU, and four more were obtained at the South China University of Technology in Guangzhou. All the obtained compounds showed good characteristics and turned out to be suitable for instrumentation. In particular, both in terms of resolution and luminescence intensity they meet the requirements for the production of medical equipment accepted worldwide.
Open use of the developments
Maxim Molokeev told us that the obtained compositions are powdery substances that can be easily applied to surfaces. Scientists used one of them to make an experimental flaw detector for checking electrical devices. In the course of the tests, the experts X-rayed a USB-connector with a housing filled with plastic, and on a plate covered with a scintillator, a detailed image of its internal structure - a system of wires and metal components.
- We proposed a new approach to using machine learning in materials science. In the future, by extending these tools, it is possible to predict new materials with unique properties in various industries. In particular, the developed method will help to find scintillators for other applications as well. For example, for microscopic transmission devices, which will help to study hard-to-reach areas of organisms or artificial mechanisms, - believes the expert.
He noted that the proposed methodology is open and any researcher can freely use it for their development and scientific research.
- AI methods are being developed to predict the structure, composition and properties of new compounds and materials. At the same time, of great importance are the works that reveal hidden regularities and supplement the researcher's chemical intuition and lead to the synthesis of new compounds with useful functional properties," Alexey Korovin, a senior researcher in the group "Design of New Materials" at the AIRI Institute, told Izvestia.
At the same time, he noted that a significant challenge is associated with the limited experimental samples with the simultaneous complexity of the chemical space. Therefore, in order to build effective models , it is desirable to formulate the problem correctly in advance, which will allow for more accurate selection of appropriate algorithms, modeling and experimental data collection, and validation of the results.
- Since the data sets are small, the key in the proposed methodology is rather the intuition of the researcher: how he collected and prepared the data, what limitations he laid down at the stages of training and search, rather than the approach itself," says Alexei Samosyuk, Project Manager of the Laboratory of Hybrid Information Systems at the Moscow Institute of Physics and Technology.
According to him, the proposed solution is a standard approach. Similarly, for example, in biology to search for new drugs - a neural network is trained on data about known compounds, and then there is a search for similar structures. Often with additional restrictions on the search.
According to Nikolay Brilliantov, professor of the Skoltech AI Center, the proposed approach illustrates how artificial intelligence methods can accelerate the development of new functional materials and reduce research costs, even if the approach itself is no longer something fundamentally new.
- It is now becoming more or less a routine approach. In the work of scientists proposed another application of artificial intelligence methods in materials science, in this case to predict the structure of photoluminescent materials, - commented the expert.
He emphasized that the researchers used a small but carefully selected array of experimental data to train an algorithm to identify patterns between structural parameters (distances between metal ions) and luminescence efficiency. The results of the AI were tested and confirmed in practice: the selected compounds demonstrated high luminosity of emission.
Переведено сервисом «Яндекс Переводчик»
