In General, the frequency feature of cavitation was obtained by comparing it with normal signal. However, when there were a large number of samples, it was difficult to analyse the sensitive features of cavitation fault effectively. So, a neural network based method for sensitive frequency component analysis of cavitation fault was proposed, and Empirical Mode Decomposition(EMD) method, Fourier Transform and neural network were used. Firstly, the raw vibration signal was decomposed to 5 Intrinsic Mode Function(IMF) components and the frequency spectrum of each component were computed. So, the dataset of raw signal was divided into 5 datasets which contained different frequency components. And a neural network was built, trained and tested by the different datasets. By comparing the diagnosis accuracy of the neural network, the sensitivity of different IMF was analysed. And it is verified that the method can effectively analyse the sensitive frequency components of cavitation faults, reduce the size of the neural network.

No abstract

This paper proposes a method for diagnosing bolt looseness faults using the principle of PCA to extract time-domain features of monitoring data. First of all, five dimensionless factors of IMF are calculated after empirical mode decomposition (EMD) is performed on the original data. Then, principal component analysis (PCA) is applied to the data vectors, which are processed by dimensionality reduction and residual space projection, to calculate the prediction error of the data sample. At last, a fault judgment test of bolt loosening was carried out on the test bench of the intelligent water supply system. The test results show that the PCA model can effectively judge the bolt loosening fault.