Experimental Study Of Gas-liquid Two-phase Flow Measurement Based On Differential Pressure Feature Fusion

Gas-liquid two-phase flow has important research significance in many fields,such as pipeline transportation and industrial production.Flow pattern and gas fraction are important parameters in gas-liquid two-phase flow.Flow pattern affects the accurate measurement of other parameters in two-phase flow.Gas fraction can provide the basis for measuring flowrate,pressure gradient and analyzing the flow condition.The main works and innovations of this dissertation are as follows:(1)This dissertation uses wavelet packet to denoise the differential pressure signals.The amplitude-frequency diagrams and time-frequency diagrams are analyzed.Considering noise signals are irrelevance,correlativity analysis of signals after reconstruction using wavelet packets is performed to realize noise identification and denoising within different frequency ranges.Comparing the amplitude-frequency diagrams and time-frequency diagrams of the differential pressure signals before and after denoising,the main frequency range and energy distribution of the signals are almost coincident,except for some high-frequency micro energy.(2)Differential pressure signals have different amplitude-frequency and time-frequency features on top and bottom sections of Venturi tube.Deformation occurs when two phase flow through the throat of Venturi tube.Gas that originally existed on the top of the pipeline is formed a large number of bubbles and spreads around in the diffusion section.By analyzing amplitude-frequency and time-frequency features,it is found that the energy of differential pressure signal is mainly concentrated on 3~10Hz with a single energy peak,and the remaining energy is scattered on the top section.The energy of differential pressure signal is concentrated between 3~8Hz and 18~32Hz with a double energy peaks,and little energy scatters at the bottom section.(3)In this dissertation,variance,kurtosis and skewness are selected as the first category features.They distributed in different regions for bubble flow and slug flow.Information entropy related to turbulence feature and the fluctuation feature are selected as the second category features.The second category features for bubble flow has a linear relationship with gas fraction.The second category features for slug flow has a non-linear relationship with gas fraction.The generalized Hurst index related to chaotic feature is selected as the third category feature.The generalized Hurst index in the experimental conditions is less regularity.The third category features for bubble flow has a roughly linear relationship with gas fraction.The third category features for slug flow has a non-linear relationship with gas fraction.(4)The principal component of the feature parameters is extracted.Echo state network is used to identify flow patterns and estimate gas fraction.Principal component analysis method is used to extract the principal component of the feature parameters of the differential pressure signals.Three principal components are extracted as new parameters,and their cumulative contribution rate of variance reaches 95%.The principal components are sent to echo state network to identify flow patterns and estimate gas fraction.The root mean square error of gas fraction estimation is 6.45% and 5.10% based on the differential pressures on the top and bottom sections of Venturi tube,respectively.