The microstructure of a printed lightweight high-temperature aluminum alloy, especially its inclusions, is characterized in a multimodal way connecting the findings of light and electron microscopy. Eventually, machine learning-based post-processing leads to reproducible, operator-independent results.
Nuclear energy production critically depends on the reliable performance of a wide variety of materials working in concert under unique and often extreme operating conditions. Understanding and optimizing this performance, qualifying materials, and preventing premature failure similarly depends on a comprehensive understanding of these materials at the microstructural level. Due to the characteristics of the materials used, however, analyzing materials for nuclear energy production presents unique challenges. In this talk we will discuss some of the unique solutions that enable characterization of these materials with advanced microscopy techniques, including electron, ion, and x-ray microscopy, and present some examples highlighting novel characterization workflows enabled by these capabilities.

Key Learnings:

• Machine learning-based classification can be trained on one dataset, creating a model
• The model is then applied across multiple samples to give repetitive, non-subjective results
• The ability to perform object classification based on features, such as textural information rather than just local greyscale values, has the potential to extract information from datasets in a reproducible way.