Research On Fault Diagnosis Method Based On Chaotic Neural Network

With the development of hydraulic technology, the Hydraulic systems are widelyused in many important fields. Its functions enhance continually, at the same time, thestructure becomes more and more complex, which also increases the possibility ofhydraulic system failure. As the source of power for the entire hydraulic system, hydraulicpump works in very poor condition and its complex structure, it makes the pump to failureeasily. The pump working conditions will be the key to the normal work of the entirehydraulic system and the whole device, thus the hydraulic pump condition monitoring andfault diagnosis is particularly important. In recent years, the hydraulic pump faultdiagnosis technology becomes a hot topic and develops toward automation, intelligentdirection. So this paper uses a method of combining chaos theory and neural networks tocomplete the hydraulic pump failurediagnostic process.In nonlinear science, the study of chaos theory is very active, combining chaos theoryand neural network to constitute a more superior performance of the chaotic neuralnetwork becomes one of the hot. In this paper, based on the feed-forward chaotic neuralnetwork, one type of feed-forward chaotic neural networks based on Logistic mapping areestablished, and its training algorithm is studied. Through the application of chaosmechanism, the chaotic network can effectively prevent the neural network easy to fallinto the shortcomings of local minima in the training process, and has a better recognitionof the tiny difference, and has good ability of search, generalization and patternrecognition.In order to verify the validity of the method, in this paper, take the swashplate axialpiston pump as research object, monitor the pump’s status and collect vibration signal ofthe pump under different working conditions in its end cap at the hydraulic pump. Makeperpendicular to the end cap of the vibration signal as information, using short-termmaximum entropy spectral analysis method obtained the resonance frequency range of thevarious fault conditions, which provides the basis for the wavelet packet band-passfiltering. The signal processing is completed by using wavelet packet theory and envelope demodulation algorithm based on Hilbert transform, then complete power spectrumanalysis. The amplitude domain feature vectors can be extracted from the envelope signal,then these feature vectors will input the chaotic neural network. Make simulation throughthe MATLAB software, which proves that the feed-forward chaotic neural network isfeasible in the hydraulic pump fault diagnosis. Compared with the widely used BP neuralnetwork diagnosis, the feed-forward chaotic neural network has faster convergence speedand the higher diagnostic accuracy than the L-M BP neural network, which reflects thesuperiority of chaotic neural network in the hydraulic pump fault diagnosis.