Combined VMD-Morlet Wavelet Filter Based Signal De-noising Approach and Its Applications in Bearing Fault Diagnosis

ObjectiveRolling element bearings are an essential part of rotating machinery. Sudden bearing failure may lead to catastrophic machine failure. Early bearing fault detection is essential to avoid machine failure. Vibration data received from bearings typically contain impulsive fault information. The characteristics acquired from the vibration signals generated by bearings are primarily used to identify bearing defects. The derived features might not be able to accurately pinpoint the failure's timing due to background noise in the observed vibration signal. External noise reduction from the vibration signal is essential for extracting important features for effective fault diagnosis. A helpful de-noising method at present is variational mode decomposition (VMD). However, the VMD method alone may not eliminate the noise from the vibration data.MethodsThe present work proposes a methodology for noise reduction combining VMD and an optimized Morlet filter. Initially, the signal is split using the VMD approach into various intrinsic mode functions (IMF), and the most efficient IMF is chosen using the maximum kurtosis criterion. Next, the golden ratio optimization method (GROM) based Morlet wavelet filter is applied to the effective IMF for reducing unwanted noise. The convolutional neural network (CNN) technique is then employed to identify the bearing defects.ConclusionThe proposed approach is tested upon bearing simulation datasets, bearing experimental datasets, gearbox experimental datasets, and sound datasets to validate its efficiency. The validation of the proposed algorithm using gear and sound datasets indicates its broad applicability.