This research proposes an adaptive time-frequency decomposition scheme for fault diagnosis of gearbox with non-stationary input shaft slew rate. Due to the fact that in run-up and cost-down stages the vibration signal has a non-stationary dynamic signature, traditional spectral analysis and state-of-the-art wavelet transform does not lead to precise fault detection. On the contrary, pioneering signal processing techniques could effectively analyse such vibrations. In this paper, we utilise Adaptive Gaussian Chirplet (AGC) decomposition along with Discriminant Analysis (DA) to alleviate the problem of gearbox fault diagnosis in run-up stage. AGC estimation, which has not drawn researchers' attention yet in the area of gearbox fault diagnosis, is applied as a novel approach for fault feature extraction in a varying-speed gearbox. AGC estimates the short-time durations of dominant slew orders in varying-speed phase. The parameters of decomposed adaptive chirplets such as chirp amplitude, centre time, centre frequency, and chirp rate are selected as sensitive features to gear impact faults. Since impact faults are present in short durations of vibration which posses higher levels of energy, the proposed method is highly effective for detection of local faults. In addition, wear defects in gearbox change the amplitude, phase and chirp duration properties of the estimated chirplets. Consequently, the combination of these features is used for gearbox fault classification. Besides, the proposed method offers a sparse representation of gearbox vibration which provides low dimension features which could be used prudently for gearbox fault diagnosis. Simulation and experimental evaluation proved the presented method to be applicable for discriminating between both impact and wear gear defects with different intensities. Eventually, Sub-space DA is applied successfully on the set of extracted features from experimental data for means of gear fault detection and condition classification.
展开▼
