Machinery Diagnosis Based on Wavelet Packets

Abstract

This paper proposes a machinery diagnosis method based on the wavelet packet theory to deal with the nonstationary contents of vibration signals generated in machinery owing to mechanical faults. The authors introduce a fault dependent wavelet packet basis to decompose and analyze the signals. The coefficients of the decomposed signals are exploited as pattern features for recognizing faults in machinery. Because of its fault dependent property, the basis does not change from record to record for a specific fault, and this greatly facilitates the effectiveness of pattern recognition. An application of the proposed approach for fault detection in ball bearings is presented. The results show that both the sensitivity and reliability of the proposed method are good.