Research On The Gearbox Fault Diagnosis Technology Based On Wavelet Analysis

With the development of technology, as a GM parts gearboxes are widely applied in modern machinery and equipment. Due to the complex structure and harsh operating conditions, gearboxes are easy to cause mechanical malfunction. In order to reduce economic losses, fault diagnosis technology is now becoming one of the key research topic. Fault diagnosis technology mainly has three parts:acquisition and pre-processing of vibration signal,the fault feature extraction and operating status recognition. As an advanced signal processing tools, wavelet analysis has multi-resolution features. Wavelet can amplify the signal and clearly characterize the local signal information, in time-frequency two domains. So it is widely applied to fault diagnosis technology.The mainly research work of this thesis includes the following aspects:The vibration rationale of gearbox is analyzed. For the signal of lots noise, wavelet transform is introduced. The mandatory thresholds,default thresholds and other methods are used to do high frequency wavelet de-noising. Besides, the wavelet packet decomposition can do multilayer band division, various wavelet packet de-noising methods are compared. The simulation shows the wavelet theory and wavelet packet decomposition are useful in de-noising preprocessing of gearbox vibration signal.The wavelet transform can decompose signal finely and extract features accurately, so it is widely used as the feature extractor. For the complex signal and wavelet selection problem, adaptive wavelet theory is introduced. The traditional adaptive wavelet construction algorithm ignores the scaling function. Duo to this drawback, an improved algorithm is proposed. The simulation results show that the improved adaptive wavelet function has greater flexibility and more precise in signal decomposition. It can use a small amount of data represent richer information.This thesis focuses on the energy spectrum and envelope spectrum features. Do adaptive wavelet packet decomposition of vibration signal, combined it with spectral energy distribution and elope spectrum to extract the useful features. Experiments show that these features can characterize the signal characteristics of the gearbox effectively.The minimum error probability based on Bayesian is used to determine the state. While, the Bayesian decision is defect in the case of feature vectors interweave. Based on this, an improved algorithm based on the interweave samples is proposed. It treat F-measure as comprehensive evaluation criteria and put misclassification probability density into criterion function. Gradient descent iterative calculation is used to obtain the discrimination function. In order to ensure the operation of state recognition accuracy, the classifier combined of Bayesian classifier and classifier based on interweave samples is used to do fault identification in this thesis.