Study on the De-Noising Technology Based on SVD with Application to Automobile Driving-Axle Fault Diagnosis

Signal is polluted by broadband noise in mechanical fault diagnosis, which makes it difficult to extract the fault feature. A de-noising method based on Singular Value Decomposition (SVD) of attractor track matrix by time series is presented. We put emphasis on determination of time delay and embedding dimension of SVD appropriately, and the effectiveness of fault feature extraction before and after noise reduction is evaluated by wavelet packet decomposition. Experiment results show the vibration signal after noise reduction can enhance the fault characteristic, improve the availability and reliability of fault diagnosis.