| Gas-liquid two-phase flow widely exists in modern industry process. The two-phase flow and heat transfer character are extremely influenced by the flow regimes, meanwhile the accurate measurement of other parameters and the performance of two-phase flow are influenced too. Therefor, the identification of different flow regimes has long been cansidered as a signification topic in the parameter measurement of two-phase system. The accurate identification of flow regimes is important for the operation and design of interrelated instruments. Based on a large amount of experimental data, wavelet transform, chaos theory, neural network and date fusion are used in flow regime identification. Intelligent regime-identification method using neural network is discussed systematically from the aspects of theory and experiment. In view of the low rate of identification, a method for identifying flow regimes on the basis of neutral network is proposed using the D-S (Dempster-Shafer)evidence theory. Firstly, the pressure-difference fluctuation signals are analyzed by utilizing wavelet packet decomposition and utilizing correlation principle identifies the noise signals. The pressure-difference fluctuation signals whose frequency is above 64Hz are noise signals. The signals whose frequency is below 64Hz are denoised by the wavelet. The influence on denoising effect is compared by using different mother wavelet and wavelet threshold rules. For the test data of this thesis, the denoising effect adopting the mother wavelet 'db4'and the lightened threshold rule 'Heursure'is the best. Secondly, through the time-frequency two-dimension analysis of the different flow regimes'signals utilizing the Wigner spectrum, the strong instability is identified. And statistic theory, wavelets transform and chaos theorys are used in the characteristic extraction of the flow regimes. The denoised pressure-difference fluctuation signals are analyzed by statistical theory. Four statistic parameters, i.e. average, standard deviation, skewness, PSD (Power Spectrum Density) energy ratio are calculated. The change rules of different flow regimes are analyzed. On the basis of discussion of state space reconstruction technique and calculating methods about chaos parameters, three chaos parameters of the pressure-difference fluctuation signals, including correlation dimension, Kolmogorov entropy and Hurst exponent, are calculated. At the same time, the change laws of flow regime along with the change of gas superficial velocity are discussed. So the eight feature parameters can be regarded as one of the feature vectors of flow regimes. Another two feature vectors can be obtained by extracting wavelet packet energy and information entropy feature of 16 frequency bands signals, which are obtained by utilizing wavelet packet decomposition. The above three feature vectors as the feature vector are inputted separately to BP neutral network, RBF (Radial Basic Function) neutral network, Kohonen neutral network and SVM (Support Vector Machine). By training network with train samples, the last identifying model is regarded as pattern recognition network. The simulation result shows that the combination of wavelet packet information entropy and RBF neutral network is the best modelin these models, but the difference of identifying rate is not distinct. Aimed at the low rate of single characteristic identification, a method of flow regime identification based on multi-character fusion is proposed using D-S evidence theory. Taking the initial identification result that various local network obtained as the evidence bodies of D-S evidence theory, the final identification result is obtained according to the D-S evidence fusion rules. Compared with the single feature identification method, its identification rate of flow is improved largely. Fusion identification provides a new way to identify the gas-liquid two-phase flow regimes from the aspects of theory and technology.
|
PDF Full Text Request