Application of DWT-Sigmoid Entropy Feature Parameter Extraction in Gear Fault Diagnosis

After analyzing the non-stationarity and nonlinear characteristics of gear vibration signals and the problem of fault signal feature extraction, we propose a gear fault classification method based on DWT-sigmoid entropy and BP neural network. The method firstly uses discrete wavelet transform (DWT) to decompose and denoise the vibration signals of four kinds of gear faults and extracts high-frequency and low-frequency coefficients. Then the energy features and singular value features of the high-frequency and low-frequency coefficients are calculated respectively. Secondly, the signal is reconstructed according to the high-frequency coefficients and the low-frequency coefficients. Then the sigmoid entropy feature of the reconstructed signal is calculated. Finally, the five features are fused and input to the BP neural network to classify different faults of gears. Experiments show that the method can effectively perform gear fault classification with an accuracy of up to 100%.