The Identification Method Of Gas-liquid Two-phase Flow Regime Based On Independent Component Analysis

The phenomenon of two-phase flow widely exists in every aspect of industrialproduction，it has closely relationship with human life and production. Because thechanges of flow not only affect the flow characteristics of gas-liquid two-phase flowand heat transfer characteristics greatly, but also influence the exact measurement offlow parameters and the motion characteristics of two-phase flow system, so theidentification of gas-liquid two-phase flow is an important research directions fortwo-phase flow parameters measurement, also provide a reliable technical choice forrelated production equipment design and safe and economic operation..The traditional two-phase flow pattern identification method mainly includestwo categories: the first category is the use of flow chart or similarity criteria foridentification of flow regimes; the second category is based on the differentialpressure fluctuation signal feature extraction, and input the feature to thecorresponding neural network, so as to achieve the purpose of pattern recognition.This paper is based on the past research. Experimental studies are implementedon the laboratory bench of the gas-liquid two-phase and get the pressure wave signalsof the air-water two-phase flow of the horizontal tube. Firstly, apply the method ofICA to extract the characteristic which was independent and has minimal mutualinformation. At the same time, used the four methods of Singular ValueDecomposition, Wavelet Packet Decomposition, Chaos Theory and ICA to extract thecharacteristic parameters, and put them into and into RBF neural network in order toidentify flow pattern. By compared the running time, convergence and identificationrate accuracy of the above four methods in order to choice which is the best method.Experiments show that: in the above four methods, the ICA-WPD has the advantagessuch as: running time short, convergence fast, identification rate accuracy, so thismethod can be used as the first choice of flow feature extraction., it also provides anew, efficient method for on-line flow pattern recognition.