Virtual Simulation Analysis and Experimental Study on Gear Fault Diagnosis Based on Wavelet Neural Network

Due to the incipient gear fault vibration signal are covered by heavy noisy, it is difficult to diagnose the gear faults just according to the time or frequency spectrum of the signals. The comparison results of the virtual prototype model simulation and the experimental test also prove that the traditional Fast Fourier Transform Algorithm (FFT) analysis is not appropriate for the gear fault detection and identification. The Wavelet Back-Propagation (BP) Neural Network therefore was applied to extract the feature sets of the gear fault vibration data and classify the faults. At the first step, the wavelet analysis was employed to decompose the vibration data, and for each sample its energy of each sub-band was calculated and then treated as the input feature vector for the BP network training. By means of this approach the gear defection can be detected and recognized. The experiment test results show that the method based on wavelet BP network is available and reliable for gear fault diagnosis, and the monitoring and identification of different gear conditions, including normal, wear, and tooth broken, are accomplished with high classification accuracy.