Identification And Characterization Of Gas-liquid Two-phase Flow Patterns Based On HMM

Gas-liquid two-phase flow widely exists in various industrial processes andapplications, such as chemical industry, nuclear energy and electric power industry. Atpresent, issues on gas-liquid two-phase flow mainly involve measurement andclassification of flow patterns, gauge of flow rate and phase fraction and modeling offlow systems. Meanwhile, heat and mass transfer rates, momentum loss, and pressuregradient greatly depend on two-phase flow patterns and their dynamical behaviors. Inthis regard, accurate identification of two-phase flow patterns and uncovering theirevolution and dynamical behaviors are significant to industrial safety and automation,design and optimization of pipelines and development of two-phase or multi-phaseflow measurement. In this study the classifier method based on multi-scale signalprocessing was employed to identify several typical kinds of gas-liquid two-phase flowpatterns, while entropy measure method which can be used to uncover the complexityof systems was combined with multi-scale analysis to characterize flow dynamics ofgas-liquid two-phase flows. In this way, innovative achievements were presented inthe thesis.In view of the non-linearity and non-stability of two-phase flow pattern signals,an adaptive analysis method called ensemble empirical mode decomposition (EMD)and local mean decomposition(LMD) were applied to process flow pattern signals.Thus, feature vectors made of intrinsic mode function components and productfunction on different scales could be obtained as input features of the classifier. To dealwith HMM is not easy to convergence to the global optimum problem, the particleswarm optimization (PSO) to optimize the HMM makes it to be a good pattern classifier. This method can optimize HMM training model, HMM can quickly andaccurately find the global optimum. The feature vector input to stream HMM classifiertype to identify the flow pattern. And experiments show, HMM classifier has smallconvergence error, and can successfully identify the flow pattern. And through dataanalysis, comparative experiments with the classical HMM, with significanteffectiveness of the method.In order to characterize the dynamic and evolution of gas-liquid two-phase flowpatterns, information entropy and statistical complexity measure were involved in thestudy. Firstly, information entropy of IMF components two-phase flow patterns ondifferent scales was computed. The distributions of entropy values in different flowpatterns greatly differ with each other, and present evolutionary characteristics. butalso has some features of evolution. Then the power spectral entropy and sampleentropy to calculate the complexity of two-phase flow, flow pattern can be explainedby the evolution of the characteristics according to its entropy and complexity ofchange.