Physics Informed CNN.

[NNadir]

If you thought this was about the news network, go no further. My standard oft repeated joke is that one cannot get a degree in what is still called "journalism" if one has passed a college level science course with a grade of C or better.

It's about "convolutional neural network" (CNN) analysis of temperature fields, AI for kriging:

Physics-informed CNN for temperature field monitoring in advanced reactors

Some excerpts:

Advanced reactors are promising energy systems that can enable the world to transition to a more sustainable energy matrix. These concepts are potentially more fuel efficient and safer, compared with previous generations of nuclear reactors. Many designs, like high-­temperature gas reactors (HTGRs) and molten salt fast reactors (MSFRs), target high outlet temperatures, allowing for their operation in processes where high heating is required, such as for hydrogen production and desalination.

Their deployment, however, faces technological challenges, including the need to address specific instrumentation requirements. For example, local temperature measurements remain challenging in higher temperature regions of these reactors, where values can exceed 1,000 K. Despite many advances, there are still unique challenges in available thermometry technologies when dealing with extreme temperatures and irradiated environments,1, 2 which are typical in these reactors. Furthermore, developing economically viable monitoring approaches is another aspect that should be considered to ensure the economic competitiveness of these reactors...

(I personally think to get the best out this sustainable energy source, we need temperatures higher than 1000K, and they are feasible.)

...Here, we employ an innovative, physics-­informed machine learning (ML) approach to measuring temperature distributions in physical domains indirectly, adopting the convolutional neural network (CNN) algorithm from previous works.3–5 This algorithm leverages the Kirchhoff-Helmholtz integral equation6 to reconstruct the spatial distribution of diffusive and advective fields over domains with arbitrary geometry. The present work demonstrates that this method enables less invasive or indirect measurement techniques for advanced reactors...

It gets cooler, I mean hotter:

...The proposed methodology is geometry agnostic and can be applied to a broad range of conditions, provided the underlying numerical datasets are validated with benchmark data. Furthermore, reference benchmark data must be used to ensure the quality of the generated numerical data.

Here, we present recent advances in using the CNN field reconstruction approach targeting an application in the two nuclear systems mentioned above. This work showcases how the CNN-based field reconstruction technique developed can perform temperature monitoring for advanced reactors. In the HTGR case, the CNN reconstructs the temperature fields within the solid material of a prismatic fuel assembly (i.e., graphite and fuel compacts). Interestingly, the CNN can perform satisfactory predictions for multiple gas mass flow rates and also account for a blockage in one of the noninstrumented channels...

I like that locution, "geometry agnostic."

This is based on a presentation given at the ANS meeting, which I did not attend (it's not really my field), but the references are provided, and I hope to pick up those accessible to me.

I trust you'll have a pleasant weekend.