Machine learning enables new approach to solve the boiling crisis
Machine learning enables new approach to solve the boiling crisis lead image
Nuclear systems, computer chips and other devices that generate heat are designed to avoid melting by dissipating heat using a boiling fluid. However, this can be complicated by the boiling crisis – a formation of a vapor layer that restricts a surface from dissipating more energy.
For more than 70 years, scientists have tried to predict this crisis. Ravichandran et al. leveraged high-resolution data with machine learning to provide a solution.
The authors analyzed thousands of gigabytes of high-resolution infrared thermometry data of the bubbling dynamics on three surfaces as they reached the boiling crisis with an artificial neural network model.
Once trained, the model was successfully able to predict when the surface would reach the critical heat flux – the point after which a surface cannot get rid of enough heat and can melt down – and identify the most important characteristics determining the boiling crisis.
“My hope is that approach we used will open a new way to do certain kinds of research, particularly in thermal transfer,” said author Matteo Bucci. “Hopefully, this will convince the rest of the community that we have to explore more kinds of approaches like this.”
The authors are continuing their work with machine learning to create a system that could be used in real time to monitor the temperature of a device and report when it might hit a boiling crisis. Such a system could have a wide range of uses, including in nuclear technology.
“Predicting the boiling crisis is beneficial for many industries,” Bucci said. “It will allow systems to operate with more confidence, efficiency and reliability at higher power.”
Source: “Decrypting the boiling crisis through data-driven exploration of high-resolution infrared thermometry measurements,” by Madhumitha Ravichandran, Guanyu Su, Chi Wang, Jee Hyun Seong, Artyom Kossolapov, Bren Phillips, Md Mahamudur Rahman, and Matteo Bucci, Applied Physics Letters (2021). The article can be accessed at https://doi.org/10.1063/5.0048391 .
