An investigation on early bearing fault diagnosis based on wavelet transform and sparse component analysis

Rolling bearings, as important machinery components, strongly affect the operation of machines. Early bearing fault diagnosis methods commonly take time–frequency analysis as the fundamental basis, therein searching for characteristic fault frequencies based on bearing kinematics to identify fault locations. However, due to mode mixing, the characteristic frequencies are usually masked by normal frequencies and thus are difficult to extract. After time–frequency decomposition, the impact signal frequency can be distributed among multiple separation functions according to the mode mixing caused by the impact signal; therefore, it is possible to search for the shared frequency peak value in these separation functions to diagnose bearing faults. Using the wavelet transform, time–frequency analysis and blind source separation theory, this article presents a new method of determining shared frequencies, followed by identifying the faulty parts of bearings. Compared to fast independent component analysis, the sparse component analysis was better able to extract fault characteristics. The numerical simulation and the practical application test in this article obtained satisfactory results when combining the wavelet transform, intrinsic time-scale decomposition and linear clustering sparse component analysis, thereby proving the validity of this method.