Two Kinds of Neural Network Fusion of Aero-engine Rotor Vibration Signal Fault Diagnosis

Abstract

In order to improve the accuracy of fault diagnosis, this paper puts forward a method of wavelet packet combining neural network fusion. By wavelet packet decomposition and reconstruction of normal and fault vibration signals of aero-engine rotor, the feature vector from the vibration signal can be extracted. Then put the feature vectors which are the input vector of neural network into the BP (Back Propagation) neural network and PNN(Probabilistic Neural Network),and the paper puts forward an algorithm to fuse the results of BP and PNN. It turns out that the method can recognize the fault patterns well and improve the accuracy of diagnosis. In this paper, wavelet packet analysis is used to extract the fault feature of the rotor system of the aero engine, two kinds of network-BP and probabilistic neural networks are used for pattern identification. At the end, the fusion algorithm is used to fuse the results of two kinds of neural networks. It turns out that, two kinds of neural networks can realize the identification and classification of the rotor fault pattern. And the fusion result is better than both of the two neural networks independently.