Fault prediction of fan bearing using time series data mining

Xingjie Chen and Wenfa Zhu

Abstract

The fault symptoms are regarded as a sort of temporal patterns hidden in a time series. A novel method based on time series data mining is proposed for the prediction of fan bearing fault. The time series, which is formed by large numbers of fan bearing vibration data, is embedded into a reconstructed phase space with time-delay. In this phase space, Genetic Algorithms are used to search for the optimal temporal pattern clusters which are the criteria to identify temporal patterns. The optimal collection of temporal pattern clusters is then used to test the other bearing vibration data of fan. Once the symptoms are detected, the fault is forecasted. The simulation results show the method is efficient