Predicting Coolant Leaks in Data Center Cooling Systems Using Machine Learning

Timely detection of coolant leaks is an important task in the cooling systems of modern data centers, due to the fact that a potential leak can cause serious environmental pollution, financial losses and equipment inaccessibility. However, detecting this problem at an early stage is a difficult task, and usually data center operators learn about the leak at a late stage, when a large amount of liquid has already leaked, and the equipment has begun to fail. Detecting leaks using sensors is not always an effective method, as leaks can occur at any point along the refrigerant path. The study explores the possibility of using graph neural networks to solve the problem since they have the ability to study inter-series correlations based on data. In this paper, we consider the prediction of future coolant leaks using machine and deep learning methods, and also anomaly detection. The study is carried out on a dataset collected from a real container data center.