Extraction of weak fault using combined dual-tree wavelet and improved MCA for rolling bearings

The successful diagnosis of the faulty signal in rolling element bearings hinges on the accurate detection of the early fault present within the components of bearings. Because the fault signature is heavily covered by the system noise and resonance of the components, the early diagnosis of the fault frequency of bearings is not easy to execute. The kurtosis and root mean square values of the vibration signal are usually used as indicators for fault. However, these indicators could result in inaccurate diagnostic results because of the stochastic nature of the vibration signal of bearings. This paper presents a method using the dual-tree wavelet transform (DTWT) combined with an optimized morphological component analysis (MCA) to extract the weak fault in the early degradation stage of bearings. The DTWT decomposes the signal into multiple layers based on the fundamental frequencies of bearings. The MCA takes the decomposed signal and separates the signal into two components as output. Alternating parameter selection is implemented to improve the result of the MCA. The weak fault signature of the bearing is extracted from the separated components of the MCA. Simulated and experimental data are used to validate this method. The proposed optimization method is compared with an unscented Kalman filter parameter optimization process. The proposed diagnostic model demonstrates the superior capability of the early detection of the faulty signal within bearings in comparison with the traditionally used wavelet decomposition method and the unscented Kalman filter.