Radar Telegram alerts Open Radar

The artificial intelligence algorithm will increase the reliability of data center cooling systems, the press service of the Higher School of Economics of the National Research University reported. The development of Russian researchers based on graphical neural networks makes it possible to quickly identify cooling leaks and contamination of heat exchangers.

תרגם לעברית

ЭНЕРГОПОЛЕ · · Source

The artificial intelligence algorithm will increase the reliability of data center cooling systems, the press service of the Higher School of Economics of the National Research University reported. The development of Russian researchers based on graphical neural networks makes it possible to quickly identify cooling leaks and contamination of heat exchangers.

In data center cooling systems, there are not only obvious critical failures, but also obscure problems that develop gradually. These include small coolant leaks and gradual contamination of the heat exchanger, which over time dramatically impair heat removal and cause large-scale accidents. Existing algorithms and safety systems are designed to detect sudden changes in the operation of the cooling circuit, so it is difficult for them to catch such changes.

Russian researchers solved this problem using a graphical neural network that monitors abnormalities not only in the operation of individual sensors, but also in the entire cooling circuit. The development takes into account the engineering design of the entire system and is based on information about the connections between the compressor, the evaporator, the condenser and the other elements. The model is trained on data from work equipment and predicts the expected values ​​of indicators; If the actual measurements begin to deviate consistently from the prediction, the algorithm detects an anomaly and alerts people.

The leading researcher at the Higher School of Economics of the National Research University, Mikhail Goshchin, explained: "To diagnose complex technical equipment, it is not enough to simply analyze many sensor readings. It is important to consider how these indicators are related to each other in a real physical process. If you put such knowledge into an artificial intelligence model, it will be able to distinguish not only individual abnormal violations of the system, but also abnormal violations of the system."

Tests showed that the artificial intelligence detected leaks with a success rate of 87% and found them on average 200 minutes after the fault occurred, which is significantly better than the results of algorithms already used. In the future, this will significantly increase the reliability of data centers, the scientists concluded.

More on Radar

Want to see every report in real time?

Hundreds of sources worldwide, translated into your language, with alerts for what matters to you.

Sign up free
Share on WhatsApp Get alerts on Telegram