Advanced nuclear reactors offer enhanced efficiency and safety compared to the long-standing conventional reactors in use. This is achieved by changing either the type or the rate of nuclear reactions within the reactor core. However, these changes can lead to increased radiation exposure for core materials, like austenitic stainless steels, which were not originally designed to endure such conditions.

An alternative is a special type of steel called Ferritic-Martensitic (FM) steel because it is more resistant to damage caused by nuclear radiation. But a variety of FM steels can be made by changing the composition and processing conditions, and they behave differently under different levels of radiation exposures at different temperatures. Only a small subset of these steels has been experimentally studied so far, mostly because conducting experiments in extreme environments brings its own challenges – scarcity of nuclear testing facilities, large expenses, and safety issues.

It is therefore important to thoroughly investigate the effects of neutron irradiation on FM steels to identify the most suitable option for a specific irradiation level in a given reactor. One approach is to use physics-based models, but they require extensive defect characterisation data as input, which is missing in most experiments reported in literature. As an alternative, a collaborative team from IISc, led by Ankur Chauhan from the Department of Materials Engineering, and the University of Wisconsin-Madison, USA, has developed machine learning (ML) models. These models forecast the impact of neutron irradiation on the strength of FM steels, employing input parameters such as composition, processing conditions, and testing variables such as radiation dose and temperature.

The team used an algorithm called SHAP to pinpoint the most important input parameters/variables influencing the strength of FM steels upon irradiation. Using these variables, they deployed four ML algorithms to predict the strength of different FM steels subjected to varied radiation levels and temperatures. Their research demonstrates that these predictive models can significantly reduce the time and cost needed for conducting experiments in challenging conditions and accelerate the development of materials for advanced nuclear reactors.