Click here to log in or
Author: Matthew Hutson, Massachusetts Institute of Technology
Boiling is not just to heat up dinner. It is also used to cool things. Converting liquids into gases can remove energy from hot surfaces and prevent everything from nuclear power plants to powerful computer chips from overheating. But when the surface becomes too hot, they may experience a so-called boiling crisis.
In a boiling crisis, bubbles will form quickly, and before they separate from the heated surface, they will stick together to form a vapor layer that isolates the surface from the cooling fluid above. The temperature rises faster and may lead to disasters. Operators want to predict such failures, and new research uses high-speed infrared cameras and machine learning to provide insight into this phenomenon.
Matteo Bucci, Assistant Professor Norman C. Rasmussen in the Department of Nuclear Science and Engineering at MIT, led the new work, which was published in Applied Physics Letters on June 23 "superior. In previous research, his team spent nearly five years developing a technology that can simplify related image processing by machine learning. In the experimental installations of these two projects, a 2 cm wide transparent heater is located under the water bath. The infrared camera is located under the heater, points upwards and records at a rate of 2,500 frames per second, with a resolution of approximately 0.1 mm. In the past, people studying videos had to manually calculate bubbles and measure their characteristics, but Bucci trained a neural network to complete this tedious task, which would shorten the three-week process to about 5 seconds. "Then we said,'Let's see if we can learn something from artificial intelligence in addition to processing data,'" Butch said.
The goal is to estimate how close the water is to the boiling crisis. The system looked at 17 factors provided by the image processing AI: "nucleation site density" (the number of sites that regularly grow bubbles on the heating surface per unit area), and for each video frame, the average infrared of these sites Radiation and other 15 statistics about the radiation distribution around these sites, including their changes over time. Manually finding a formula that correctly weighs all these factors will be a daunting challenge. But "artificial intelligence is not limited by the speed of our brains or data processing capabilities," Butch said. In addition, our preconceived assumptions about boiling are "unbiased."
To collect data, they boiled water on the surface of indium tin oxide, alone or with one of the following three coatings: copper oxide nano-leaf, zinc oxide nanowire, or silicon dioxide nano-particle layer. They trained a neural network with 85% of the data from the first three surfaces, and then tested it on 15% of these conditions plus the data from the fourth surface to see how it affects the new conditions. Generalization. According to one indicator, its accuracy rate is 96%, even if it has not been trained on all surfaces. "Our model is more than just remembering features," Butch said. "This is a typical problem in machine learning. We can extrapolate predictions to different surfaces."
The team also found that all 17 factors make a significant contribution to forecast accuracy (although some factors have more factors than others). In addition, instead of treating the model as a black box using 17 factors in an unknown way, they identified three intermediate factors that explain this phenomenon: nucleation point density, bubble size (calculated by eight out of 17 factors). Out) and the product growth time and bubble departure frequency (calculated based on 12 of the 17 factors). Bucci said that the models in the literature usually only use one factor, but this work shows that we need to consider many factors and their interactions. "this is a big problem."
"This is great," said Rishi Raj, an associate professor at the Indian Institute of Technology in Patna, who was not involved in the work. "Boiling has such complex physics." It involves at least two phases of matter and many factors that lead to chaotic systems. "Despite extensive research on the subject for at least 50 years, it is almost impossible to develop a predictive model," Raj said. "The new tools of machine learning are meaningful to us."
Researchers have been arguing about the mechanism behind the boiling crisis. Is it caused entirely by the phenomenon of heating the surface, or is it from distant fluid dynamics? This work shows that surface phenomena are sufficient to predict the event.
Predicting a crisis close to boiling will not only increase security. It also improves efficiency. By monitoring conditions in real time, the system can push the chip or reactor to its limit without throttling it or building unnecessary cooling hardware. It's like Ferrari on the track, Bucci said: "You want to release the power of the engine."
At the same time, Bucci hopes to integrate his diagnostic system into a feedback loop that can control heat transfer, thereby automating future experiments, enabling the system to test hypotheses and collect new data. "The idea is actually to press the button and return to the laboratory after the experiment is completed." Is he worried about losing his job because of the machine? "We will only spend more time thinking instead of doing things that can be automated," he said. In any case: "This is to raise the bar. Not to lose work." Further exploration of new understanding of boiling water heat transfer can improve the efficiency of power plants. More information: Madhumitha Ravichandran et al., data measured by high-resolution infrared temperature measurement Drive to explore and decrypt the boiling crisis, Applied Physics Letters (2021). DOI: 10.1063/5.0048391 Journal Information: Applied Physics Letters
Provided by MIT
This story was reposted by MIT News (web.mit.edu/newsoffice/), a popular website covering MIT research, innovation, and teaching news.
More from other physics topics
If you encounter spelling errors, inaccuracies, or want to send an edit request for the content of this page, please use this form. For general inquiries, please use our contact form. For general feedback, please use the public comments section below (please follow the guidelines).
Please select the most suitable category to facilitate the processing of your request
Thank you for taking the time to provide feedback to the editor.
Your feedback is very important to us. However, due to the large volume of messages, we do not guarantee a separate reply.
Your email address is only used to let the recipient know who sent the email. Neither your address nor the recipient's address will be used for any other purpose. The information you enter will appear in your email, and Phys.org will not keep it in any form.
Send weekly and/or daily updates to your inbox. You can unsubscribe at any time, and we will never share your details with third parties.
Medical research progress and health news
The latest engineering, electronic and technological advancements
The most comprehensive technology news report on the Internet
This website uses cookies to assist in navigation, analyze your use of our services, collect data for advertising personalization, and provide content from third parties. By using our website, you acknowledge that you have read and understood our privacy policy and terms of use.