A team of scientists took inspiration from the ancient Chinese board game Go to train an AI designed to provide optimal cooling strategies.
The research team set out to analyze and predict the most effective methods of spray cooling to keep power grids and data centers operating during surges in demand.
The scientists also claim their research could lead to more effective ways to prevent engines, individual computers and turbines from overheating.
Inspired by AlphaGo
One of the scientists behind the new study, Virginia Tech associate professor Jiangtao Chen, has been playing Go since high school. Invented over 2,500 years ago, the two-player strategy board game requires the winner to “control” most of the area on the board.
Since 2014, Google's AlphaGo has allowed human players to play against AI-powered opponents. Using machine learning, AlphaGo was able to refine its approach as it played, and was able to beat professional human players within a year. Since then, he has played against and defeated the world's top players.
According to a press release, Chen decided to take on AlphaGo himself. He kept losing, but surprisingly, the experience gave him the idea to develop a strategy to cool hot machines.
“Just as a spray cooling system is a network of interacting parameters, Go is a game of interconnected dynamics,” Chen said. “Whether winning a game or optimizing a system, success requires a holistic understanding of the network and careful management of its interactions. This task can be greatly enhanced using AI to analyze complex patterns, predict outcomes, and guide optimal strategies.”
Analyzing water droplets using AI
The team published the research results as a paper in an academic journal. artificial intelligence reviewaimed to present a comprehensive analysis of the effectiveness of spray cooling.
The key is water droplets. When they hit the surface of a hot object, the tiny droplets evaporate and pick up some of the heat, regulating the surface temperature and helping to cool the object.
“When water is exposed to heat, the way it changes is different than in water droplets,” said Lori, lead author of the study. “Because droplets boil and evaporate very quickly, they pick up heat and carry it away faster. This fast turnaround cycle allows droplets to enable a more effective approach to temperature control.”
However, analyzing individual droplets is a difficult task and raises many questions. For example, what is the optimal water droplet size to effectively deal with heat? What type of spray nozzle is most effective in producing droplets of these sizes? Should water alternatives (solvents, lubricants, or artificial mixtures) be considered?
Inspired by AlphaGo, the researchers used machine learning to analyze publicly shared data from 25 existing studies on water droplets. This allowed us to evaluate the fundamental properties of liquids, how they form droplets and how they absorb heat.
“Even though AI always wins on the Go board, I never felt frustrated and learned how to leverage AI to address real-world challenges and dilemmas, such as thermal management of high-power density electronics,” Cheng said.
“By bridging thermofluid science and AI, we do more than just improve spray cooling,” he continued. “We are actually redefining how we understand and design future thermal systems.”
