Fault Pattern Recognition Of Gear Based On Wavelet Neural Network

As the transmission part of mechanical equipment responding for the essential mission of transferring power and movement, the gear is extensively applied in the industrial production. However, the gear fault will not only terribly damage the gear itself and the normal work of the system, but also endanger the operators'life safety, bringing about heavy economic losses and wide-ranging social influence. Therefore, it is of great importance and practical engineering value to make researches on the gear fault pattern recognition and fault diagnosing. In recent years, with the development of computer technology, signal processing, and the artificial intelligence, the fault diagnosis method has been progressively promoted. Moreover, with the improvement of neural network technology, fault diagnosis method which was based on neural network has been widely studied and paid much attention to. Since one of the main segments of fault pattern recognition based on neural network is feature extraction, wavelet analysis, with characteristics of self-expanding and translation of its primary function, has been acted as an effective tool to process signals, and especially the combination of wavelet and neural network, so called the wavelet neural network, has become a focus in the gear fault diagnosis study and also accelerated the technology of gear fault diagnosis at the same time.Through the Wuhan university of science and technology′s gear fault laboratory bench applying, the author gathers several kinds of gear fault signals, carrys on an analysis of gear vibration signal in three modes: normal, wearing and circular pitch error, and applys wavelet-packet energy to extract characteristics as the input of neural network. According to the three different gear fault modes, the thesis employs Levenberg-Marquardt BP study algorithm and radial basis function network study algorithm to identify the fault pattern and train the network respectively. During the course, the opening problems of BP neural network, such as the determination of the number of input/output nodes, the number of hidden layers and its nodes, the speed rate of network and the transferring function, ect, were also analyzed and discussed. The result of the recognition indicated that the three different fault modes: normal, wearing and circular pitch error can be exactly identified by the two different network configuration prototypes.Finally, the identifying effectiveness against to the two different network was compared in the paper. The conclusion shows that the radial basis function network takes advantages over the BP network model in the aspects of network configuration, network performance and the network fault-tolerant performance, ect. 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 gear fault diagnosis.