Study On Two-phase Flow Regimes Identification Based On The Feature Extraction To Cross Section Measurement Information

Along with the rapid development of science and technology, the investigation on two-phase flow is taking more places in national economy and human life. The detecting techniques on two-phase flow parameters are widely used in the field like petroleum, metallurgy, chemical engineering, pharmacy and so on. The accurate measurement of two-phase flow parameters is always the urgent issue that needs to be solved in engineering and science.In this paper, the flow parameter detecting techniques, especially the flow regime identification techniques of two-phase flow are concluded, and the application of process tomography technique, particularly of the Electrical Resistance Tomography (ERT), on identification of two-phase flow regime is analyzed primarily. Based on the above, the characteristic vector database of different flow regime is established on the feature extraction of measured data from ERT system by means of mathematical statistic analysis and discrete wavelet transform method. Furthermore, the Support Vector Machine (SVM) of Statistical Learning Theory is discussed and introduced into the flow regime identification of gas-liquid two-phase flow, and good results are achieved.The main achievements in this paper are as follows:Firstly, field experiments were brought out by ERT system in the oil/gas/liquid three-phase flow laboratory of Tianjin University Measuring and Testing Techniques and Instruments school, and different flow regime data of gas/liquid two-phase flow in vertical upward pipe are obtained.Secondly, to extract character from measured data of ERT system, first of all, the characteristic vector that consists of three feature values is obtained from different flow regime in horizontal pipe by mathematic statistic analysis; and then, for the feature extraction of flow regime in vertical pipe, one-dimensional and two-dimensional wavelet transform are adopted respectively.By comparing these two methods, it is better to excavate more regime information in measured data itself and to improve the dimensionally reduction methods from the previously only...