Research On Bearing Fault Diagnosis Based On Wavelet Analysis And Neural Network

With the development of complexity and automatic level of the machine equipment, machine equipments fault diagnosis is more and more important. Besides, it is important to choose appropriate method for fault diagnosis for precision of result. On the research of intelligence diagnosis technology, the wavelet analysis and the neural network technology is the hotspot, also is the research edge. Combination of Wavelet analysis and neural network is also an appealing topic.The wavelet analysis theory was studied in this paper. On the basis of the characteristic that wavelet packet could decompose the signals to different frequency segments in accordance with any time-frequency resolution, the method for features extraction based on energy of wavelet packet. and simulation analyzing confirmed that the method is accurate and effective.The modeling principle and training algorithm of BP neural network, RBF neural network and Fuzzy ART network were Summarized. Aiming at the shortage of BP algorithm, the Levenberg-Marquardt optimization algorithm was introduced. Wavelet analysis possesses excellent characteristic of time-frequency localization and is suitable for analyzing the time-varying or transient signals. However, neural network is successful in recognizing non-linear system and classifying pattern. Neural network intelligence diagnostic model has been established with the above principle, and it has been used in rolling bearing's failure diagnosis. According to the vibration signal features of frequency-domain, energy eigenvector was established by means of wavelet packet. Then recognition of fault pattern of rolling bearing was presented using neural network. The experimental result shows that the system can not only detect the fault of bearing but also can recognize fault pattern correctly.When the BP neural network using the Levenberg-Marquardt optimization algorithm, the convergence rate is quick, and has the high accuracy rate of failure diagnosis, but still had the disadvantage, for example, convergence is unstable. The radial basis function neural network has advantages over the BP network model in the aspects of network configuration, network performance and the network fault-tolerant performance etc. Based on the wavelet analysis and the radial basis function network, the fault pattern recognition method is much more effectively and accurately,which should be better applied in the fault diagnosis of mechanical equipments. The Fuzzy ART network model had quick and stable identification ability for those features which had been learned, but the fault diagnosis accuracy rate lower than the RBF neural network and BP neural network.