Complex Networks From Multivariate Time Series For Characterizing Two-Phase Flow

Two-phase flow, as a complex nonlinear system, widely exists in many industrial applications, but so far, there has been no satisfactory understanding of the underlying dynamics in the way of the traditional theoretical analysis method. Therefore, a new theoretically method is st rongly required to detect and describe the d ifferent flow structures and t heir nonlinear dynamics from the modern information pro cessing perspective.First of al l, based on the existing knowledge about conductance sensor, we optimize and design a curve half-ring conductance sensor by the using finite element analysis. Then, we carry o ut the dynamical experiment in multiphase flow loop facility of our research team, and using the designed sensor measure the multivariate time series corresponding to bubble flow, slug flow and churn flow.Based on measured data corresponding to three typical gas-liquid two-phase flow patterns, we c onstruct weighted com plex n etworks, a nd de scribe th e t opological structures by using dendrogram. The results indicate that this method can identify slug flow from bubble flow and slug flow, but can't efficiently discriminate bubble flow from churn flow.In this regard, we construct t hree unweighted complex netwo rks, which ar e bubble a nd s lug flow, slug a nd c hurn f low, and bubble and c hurn flow co mplex networks. Through drawing the network community structures by the visualization software, we find that every network can be described as two communities, and the tightness of network corresponding to bubble flow and churn flow is less than that of slug flow. Although the tightness of network corresponding to bubble flow is similar with t hat of churn flow, the network still separates into two parts. In addition, we analyze the horizontal oil-wat er two-phase flow by t his method, and t he results indicate that our approach can well distinguish and characterize different flow patterns in the sense that different communities correspond to different flow patterns and the community structure can characterize the flow pattern dynamical properties.Finally, we co mpare the weighted complex networks to unweighted complex network in from the two-phase flow application perspective, and find that unweighted network works better than weighted network, not only because of omitting calculating weighted values, but also shows better identification of flow patterns and revelation of nonlinear dynamical properties. Thus, we provide a new perspective and a novel way for understanding the complex d ynamics u nderlying two-phase flow in terms o f complex network theory.