Composite fault location of gearbox based on wavepack and BP neural network ensemble

A neural network ensemble classifier based on Bagging and back propagation neural networks was proposed. This classifier has a strong classification ability, and overcomes the local optimum defect of back propagation neural network. It was applied to fault location of gearbox in composite fault condition. Gearbox vibration signals were decomposed by wavelet packet transform and then frequency band energy features were extracted. The principal component analysis was used to compress features. The neural network ensemble classifier was trained by feature datasets. Results towards the testing data show that the proposed neural network ensemble classifier obtains better generalization ability and gives a better success rate than single back propagation neural network reaching 97.5%.