A new approach to predict fan failures with fan speed correlation

When aging fans in server chassis wear out, they slow down and circulate less air, eventually causing overheating condition and unexpected thermal shutdown. We propose a novel approach to predictive analysis of fan failures before the significant wear-out or slow-down by modeling the correlation of the speed measurements of multiple fans. We present the empirical result that the speed measurements of multiple fans in a server are highly correlated and exploit the observation to devise the use of data-driven anomaly detection methods in predictive fan failure analysis. A comparative study on several variants of the approach is performed on the simulated data sets demonstrating how certain heuristics substantially improve the performance.