An enhanced technique for roller bearing defect detection using an impulse response wavelet based sparse code shrinkage de-noising algorithm

Detection of defects at early stage is crucial to fault prognostics. Periodic impulses indicate the occurrence of faults in roller bearings. However, it is difficult to detect the impulses of initiating defects because they are rather weak and are often immersed in heavy noise. Existing wavelet threshold de-noising methods are not efficient because they use orthogonal wavelets, which do not match correctly the impulse and do not utilize prior information on the impulses. Hence, a Sparse Code Shrinkage (SCS) method based on maximum likelihood estimation (MLE) for thresholding using an adapted wavelet is developed. Based on SCS de-noising, the present method gives an in-depth analysis of the inspected signal even at very low signal to noise ratio (SNR).