Abstract: A noise reduction method is described. The method is based on the characteristic of singular spectrum of the orbit matrix, which denotes the reconstructed attractor of the vibration signal in reconstructed phase space. Its applications in the fault diagnosis of rolling bearing and gear box show that this method can reduce the noise contained in the original vibration signal effectively, and improve the fault characteristic included in the original vibration signal thus the fault in system can be diagnosed correctly from the processed time series of the vibration signal. This method can make for improving the correct rate of machine fault diagnosis.

Key words: Fault diagnosis, Gear box, Noise reduction method, Rolling bearing, Singular spectrum