Feature Extraction And Condition Recognition Of Hydraulic Units Based On HHT And SVM

With the development of design and manufacturing level of hydraulic turbine, more and more high-capacity units have been put into use. The proportion of hydraulic power in electric system is becoming more and more bigger. The structure and automation level of hydraulic units is also getting more complex and higher respectively. In view of this, the reliability, security and stability of hydraulic units'operation are getting more important now.Compared with the Fourier-based linear and stationary spectral analysis, Hilbert-Huang transform(HHT) based on empirical mode decomposition (EMD) is considered more effective in dealing with non-linear and non-stable signal. This paper attempts to use HHT method in signal analysis of hydraulic units. After signal analysis of turbine guide bearing, this paper found that Hilbert spectrum has higher resolution in time-frequency distribution compared to Wavelet distribution and Wigner-Ville distribution. Besides, the time-frequency distribution has more physical meaning. In order to use HHT method in practical application, the class of THHT is designed by Pascal programming language which includes the algorithms of EMD and HT (Hilbert Transform) and a diagnostic module is fulfilled based on the class of THHT.Support vector machine(SVM) is a rising star in artificial Intelligence . SVM is new learning machine based on statistical learning theory(SLT) and structural risk minimization(SRM) principle. In order to apply SVM to condition recognition of hydraulic units, this paper constructed the multi-class SVM classifier based on Decision directed acyclic graph(DDAG). When talking about the feature extraction method of hydraulic units, we adopted the energy distribution based on the IMF(Intrinsic mode function) component of signal. With this method, we get the train samples and test samples of 5 conditions signal of turbine guide bearing. This paper tested the DDAGSVM multi-class classifier's performance by the test samples. We can draw the conclude that DDAGSVM can reasonable construct the multi-class classifier of hydraulic units and the energy distribution based on IMF can effective and comprehensive extract the feature of units.