| Paper description |
Four main problems need to be solved in the fault diagnosis of rolling bearings when the speed and load both are changing: variable amplitude, fluctuating shocks interval, inconstant sample phase and signal noise pollution. To deal with these problems, this paper proposes a novel diagnosis method based on Teager Computed Order Spectrum (TCO Spectrum) and Stacking Auto-encoder (SAE). TCO-Spectrum, which is the pretreatment method for SAE, is the order spectrum of the signal after Teager energy operator demodulation. Two other pretreatment methods, Computed Order Spectrum and Hilbert Computed Order Spectrum, are also explored as comparisons to verify the effectiveness of TCO-Spectrum. In order to prove the powerful feature extraction ability of SAE, four traditional intelligent methods (Random Forest, Support Vector Machine, Hidden Markov Model and BP Network) are used as comparisons. The effectiveness of all methods is validated by the dataset composed of 56 training bearings and 8 testing bearings which involve 4 different health conditions. The speed range of each signal sample is controlled between 200 and 2000 rpm and the load is set from 7.7 kg to 48.1 kg. The diagnosis results demonstrate that the proposed method is capable to learn features adaptively from vibration signal of rolling bearings in spite of changing of speed and load. The proposed diagnosis method based on TCO spectrum and SAE achieve 95.72% recognition accuracy which is higher than other comparative methods. |
(创新港)
