Improved WNN To Rotating Machinery Fault Diagnosis

In the production of modernization, machinery and equipment fault diagnosis technology has been gaining more importance, if a malfunctioning piece of equipment failed to detect and remove, the result will not only lead to damage to the equipment itself, and may even cause machine crash serious consequences . In the system of continuous production in the enterprise, if a critical equipment failure and not due to the continued operation, often involving the entire business operation of production systems equipment, causing huge economic losses. Therefore, continuous production systems, such as Turbo-generator power system, metallurgical and chemical process equipment, such as the key to the process, fault diagnosis of great importance.Wavelet network is the perfect combination of the theory of wavelet analysis and the theory of artificial neural network, it is compatible with the superiority of the wavelet and neural networks. On the one hand, it makes full use of time-frequency localized properties of wavelet transform; On the other hand, it puts the self-learning ability of neural network into full play so it has a strong ability to tolerant mistakes and close. Because of its superior characteristics, wavelet networks are widely used in many aspects such as signal processing, function fitting, data prediction, system identification, fault diagnosis and automatic control. Of course, there are deficiencies in wavelet network: more complicated structure, compared with BP network, the wavelet network computing complexity increased; and mapping high-dimensional wavelet networks prone to learning "dimension disaster" problem.In this paper, the compact structure of wavelet neural network are formed from the theory of wavelet analysis, and the wavelet neural network based on the traditional BP algorithm was improved. And also it provides the initial parameter settings of wavelet neural network of the combination of types of wavelet, wavelet time-frequency parameters and the study sample. This method is different from random assignment of the the initial weight of traditional network and it increases network stability and convergence precision. It introduces the improved wavelet neural network based on the BP algorithm and apply it to the examples of rotating machinery fault diagnosis in order to avoid the low efficiency of traditional algorithm of network structure, and improve the performance of the network learning. It Effectively overcome some shortcomings of wavelet neural networks based on the BP algorithm, such as improper initialized parameters of wavelet network will result in non-convergence of the learning process of the network and it is easy to make the entire network into a local minimum. The data of Modeling for de-noising and normalization. Finally, we will use improved wavelet neural network and traditional BP neural network to make diagnosis of rotating machinery fault and show a contrast of the effect of diagnosis and make an analysis of the advantages of the algorithm.