Deep Learning Model Overcomes the Challenge of Real-World Measurements of Isotope Production Target Cooling Systems

The Science

When a particle beam hits a target in an isotope production facility, it generates heat that is removed by water channels. The transfer of heat between the target and the flowing water produces subcooled flow boiling. This is a situation where evaporation and condensation occur simultaneously. Without sufficient cooling, temperatures can reach a critical point, melting and destroying the target. Researchers wanted to test the limits of the cooling systems in an isotope production facility. However, the radiation levels present during target irradiation are too high to allow real-world measurements. Instead, the researchers built a mock apparatus to collect temperatures and high-speed video of boiling. They analyzed the measurements with deep-learning tools and used the results to validate a model that predicts boiling in these complex systems.

The Impact

Isotope production facilities use targets to create isotopes that support medical imaging, cancer therapy, and other applications. With that connection to biology in mind, researchers used a deep-learning tool originally designed to track biological cell activity to examine water cooling. The tool analyzed bubbles in the water as an indicator of boiling, tracking bubble formation, size, and movement. The researchers combined detailed analysis of bubble behavior with temperature data to create complete boiling curves. Next, they developed a model based on the curves to predict cooling in target systems. This enables higher beam intensities without risking target failure. The model can be expanded to potentially benefit other particle accelerator target applications as well as isotope production facilities worldwide.

Summary

Particle accelerators and isotope production facilities depend on cooling for the proper function of target systems during irradiation. The target systems at Los Alamos Neutron Science Center’s (LANSCE) Isotope Production Facility rely on a series of water channels in the target system to remove heat as targets are irradiated. The high heat flux between the target surface and the flowing water in the channels likely produces subcooled flow boiling. With concerns about target failures caused by reaching critical heat flux and boiling crisis, researchers wanted to determine maximum operating conditions, but radiation levels present during target irradiation prevent live monitoring. Instead, researchers built an experimental apparatus that mimics the target system to collect temperature data and high-speed video of bubble activity in the mock cooling channel.

Researchers adapted a deep-learning tool that was originally developed to detect cells in biological images. With the adapted algorithm, the team extracted critical bubble parameters from the high-speed video, enabling the researchers to develop and validate a framework to predict a complete boiling curve. The researchers then confirmed that current operations are well below the predicted critical heat flux. This model can be expanded for full-scale modeling of complex multi-target and cooling channel geometries and potentially benefits other particle accelerator target applications and isotope production facilities that lack in-beam monitoring at different accelerators worldwide.

Funding

The research was supported by the Department of Energy Isotope Program managed by the Office of Science for Isotope R&D and Production. The research was carried out by Los Alamos National Laboratory.