小波降噪, 权值, Adaboost算法, 故障诊断

In order to solve the problem of identifying incipient faults of a hydraulic pump, a novel method of fault diagnosis based on gradual asymptotic weight selection of wavelet and Adaboost ensemble was proposed. Aiming at the features of incipient faults not abstracted effectively, based on optimization theory, selection of gradual asymptotic weight by using the signals from traditional wavelet was to get the higher SNR factor of denoised signals. The denoised signals were used to select the optimal features. Then, aiming at the problem of neural network’s over-learning and under-learning, the optimal features were trained with Adaboost algorithm to identify the different fault cases. Testing results show, compared with traditional wavelet, gradual asymptotic weight selection of wavelet can denoise, improve SNR factor, and abstract the optimal fault features effectively. Adaboost algorithm has a higher classification success rate than the BP neural network.

wavelet, weight, Adaboost algorithm, fault diagnosis