Artificial intelligence will aid in developing new radiation-resistant materials for nuclear energy.

Researchers from the N. L. Dukhov All-Russian Institute of Automation and the National University of Science and Technology MISIS have proposed a new model based on artificial neural networks to predict the emergence of defects in nuclear reactor materials. The findings are beneficial for developing materials that can withstand radiation over extended service periods.

2024-12-19 10:01:08https://naked-science.ru/article/column/radiatsionno-stojkie-mate

Defects form in the cladding of heat-generating elements in nuclear reactors during operation. One of the main issues is radiation swelling, which refers to the gradual increase in the volume of the material upon irradiation, negatively impacting its strength and durability. For the cladding of heat-generating elements in modern fast neutron reactors, austenitic heat-resistant steel is utilized. This material must maintain its mechanical properties under high doses of radiation, with permissible deformation limited to a few percent.

There are two "classic" approaches for predicting radiation swelling. The first is empirical models. These are reliable but not universal, as they are limited to specific materials and conditions. The second method is multiscale modeling, which takes into account physical processes at various levels, from atomic to macroscopic. However, it is currently not accurate enough for predictions in real-world conditions.

"A promising method is machine learning. Artificial intelligence can predict material behavior based on the steel composition and irradiation conditions," said Pavel Korotaev, an expert at the Laboratory of Modeling and Development of New Materials at MISIS National University of Science and Technology.

Using this method, researchers have forecasted the complete swelling profile during irradiation with fast neutrons, depending on the radiation dose, reactor temperature, and steel composition.

"Previously, no one had predicted the complete 'dome' of swelling using machine learning. To train our model, we examined dozens of materials that can swell up to 50 percent. As a result, we can predict swelling with high accuracy. This has helped clarify how various alloying materials affect radiation resistance. For instance, elements such as nickel, titanium, phosphorus, silicon, and carbon reduce swelling, but only up to a certain limit," added Pavel Korotaev.

In the future, scientists plan to enhance the model's capabilities in the field of prediction.

Details of the research are published in the scientific journal Computational Materials Science (Q1). The work was conducted with the support of the Russian Science Foundation.