A new machine learning platform instantaneously detects and quantifies radiation-induced defects in parts and testing materials in nuclear reactors. The technology demonstrated by University of Michigan and Theia Scientific researchers provides real-time image-based detection and quantification of radiation damage in different materials.

The system was tested by exposing samples of iron, chromium and aluminum to a krypton beam. The process creates radiation defects, allowing researchers to quickly replicate the damage sustained after years or decades of use in a nuclear reactor. Defects are observed as black dots in electron microscope images that are recorded as video during the test.

The real-time quantification and visual overlay can be seen in the top monitor where the researcher tries to quantify the number of large black dot defects as an evolution of radiation damage. Source: Kevin Field, University of Michigan Nuclear Oriented Materials and Examination GroupThe real-time quantification and visual overlay can be seen in the top monitor where the researcher tries to quantify the number of large black dot defects as an evolution of radiation damage. Source: Kevin Field, University of Michigan Nuclear Oriented Materials and Examination Group

Machine learning software based on a convolutional neural network displays the results in graphics overlaid on the electron microscope imagery. The defects are labeled as to size, number and other parameters to provide a measure of structural integrity. This capability allows for instantaneous quantification of radiation-induced defects and negates the need to download video and manually count every defect in selected frames.

In addition to improving the development of components for advanced nuclear reactors, the platform may also find application in the transportation and biomedical sectors.

To contact the author of this article, email shimmelstein@globalspec.com