Research And Application Of Wavelet Transform And Fuzzy Neural Network In Bearing Fault Diagnosis

Turbo generator sets can ran for a long time when their vibration in a certain range, but the bigger vibration can directly threaten to the sets' safe. And long-time vibration can result in the damages to the foundation and surrounding buildings. The noise of the vibration can also take huge disadvantages to operators' physiological and mental conditions. What's worse, the bearing faults can accelerate the vibration of the turbo generator sets. Therefore, it is necessary to diagnose the bearing faults of the turbo generator sets.Wavelet transform has the characteristics of multi-resolution and time-frequency localization, especially suited to the analysis of non-stationary signals; and fuzzy LMBP neural network not only has the highly nonlinear mapping capability, but also has a strong advantage in the fuzziness of class boundaries.This paper firstly used wavelet analysis to extract the features of the fault signals, and used the extracted features as the input of fuzzy neural network, forming so-called wavelet fuzzy neural network to put fault diagnosis on. This paper used Matlab software to simulate and use the wavelet fuzzy neural network to diagnosis on, with the data of the fault bearing which come from the vibration testig simulating sets about the steam turbine, which proved that it can greatly improve the adaptive ability of diagnosis system greatly. This paper consists of two parts as follow:1. Extracting the features of fault signals Using wavelet analysis to extract the features of fault signals. First, wavelet decomposition should be done about the original fault signals by the way of the characteristics of multi-resolution and time-frequency localization. Second, used the feature vectors which consist of the first high frequency of the wavelet can stand for the feature of the original fault signals as the input of the fuzzy neural network.2. Pattern recognition for fault bearing signalsThe pattern recognition of the signal of the fault bearing can be developed by way of three methods as fellow(1) Using LMBP neural network to take pattern recognition. Comparing with the Elman neural network and the typical BP neural network, the LMBP neural network has an advantage over them on the error and the performance of real time.(2) Using wavelet neural network to take pattern recognition. Although this method can be used for diagnosing on the fault signals and satisfy the request of the real time in some degree, there is some disadvantage on precision.(3) Using wavelet fuzzy neural network to take pattern recognition. This method can diagnose the fault signals efficiently and satisfy the request of the real time.Comparing with the simulated result of the three methods on Matlab software, the best result is which use the method of the wavelet fuzzy neural network to find the pattern recognition. This method can diagnose the fault signals efficiently and satisfy the request of the real time.