A data-driven approach to simultaneous fault detection and diagnosis in data centers

Abstract

The failure of cooling systems in data centers (DCs) leads to higher indoor temperatures, causing crucial electronic devices to fail, and produces a significant economic loss. To circumvent this issue, fault detection and diagnosis (FDD) algorithms and associated control strategies can be applied to detect, diagnose, and isolate faults. Existing methods that apply FDD to DC cooling systems are designed to successfully overcome individually occurring faults but have difficulty in handling simultaneous faults. These methods either require expensive measurements or those made over a wide range of conditions to develop training models, which can be time-consuming and costly. We develop a rapid and accurate, single and multiple FDD strategy for a DC with a row-based cooling system using data-driven fault classifiers informed by a gray-box temperature prediction model. The gray-box model provides thermal maps of the DC airspace for single as well as a few simultaneous failure conditions, which are used as inputs for two different data-driven classifiers, CNN and RNN, to rapidly predict multiple simultaneous failures. The model is validated with testing data from an experimental DC. Also, the effect of adding Gaussian white noise to training data is discussed and observed that even with low noisy environment, the FDD strategy can diagnose multiple faults with accuracy as high as 100% while requiring relatively few simultaneous fault training data samples. Finally, the different classifiers are compared in terms of accuracy, confusion matrix, precision, recall and F1-score.