Bearing fault detection for drivetrains using adaptive filters based wavelet transform

Predicting a localized defect on a rolling bearing during the degradation process before a complete failure is crucial to prevent system failures, unscheduled downtimes and substantial loss of productivity. During this process, impulses associated with the fault are weak, nonstationary or time-frequency varying, and contaminated by noises, which render the problem of extracting these impulses very difficult. This work investigates the effectiveness of common signal processing techniques on predicting incipient faults, e.g. Fast Fourier transform, Short-Time Fourier transform, Wavelet transform. It was found that an adaptive filter is required to enhance and reconstruct the signals during the degradation process, and a combination of adaptive filter and Morlet wavelet transform is necessary in order to effectively detect a localized defect on rolling element bearings during degradation. The proposed method was applied to analyze vibration signals collected from a run-to-failure test of drivetrain. The analysis shows that the frequency associated with a bearing defect can be well identified in the early stage or during degradation.