Abstract: In order to improve the accuracy rate of state diagnosis of rolling bearing based on time domain signals, a novel diagnosis system which combined significance sequence of vibration signals and machine learning was proposed. Firstly, the residual spectral of the signal was obtained by subtracting the logarithmic amplitude spectrum from the mean logarithmic amplitude spectrum which was acquired by analyzing the normalized vibration signals in frequency domain. Then, the spectral residual of the vibration signals was mapped back to the time domain by inverse Fourier transform to obtain the significance sequence. Finally, the state diagnosis model was used to classify the significance sequence to achieve fault detection of the bearing state. The experiment results indicate that the significant sequence can effectively improve the classification accuracy rate comparing with the original vibration signal, especially for the vibration signal mixed with the different signal noise ratio (SNR) of Gaussian white noise, such as mixed with white Gaussian noise of -6 dB, using support vector machine(SVM) or convolution neural network(CNN) as the state diagnosis model respectively, the significance sequence can respectively improve the accuracy rate of state diagnosis by 9% and 10.75%. Whats more, this method can also effectively shorten the training time of the network model.

Key words:  vibration signal; rolling bearing status diagnosis; significance sequence; spectral residual; machine learning

LIU Zhixiang, ZHU Ming, FU Ming, et al. Rolling bearing state diagnosis method based on vibration signal significance sequence［J］.Journal of Mechanical & Electrical Engineering, 2021,38(8):944-951.