Research On Multi-information Fusion Of Distributed Sensors In Two-phase Flow Based On Complex Network Theory

Two-phase flow is a common occurrence in various industrial processes such as petroleum,papermaking and aerospace industry.The industrial processes and efficiency can benefit from the research of the two-phase flow parameters.However,the flow patterns are complicated and the capture of local flow information regarding different flow patterns still remains a challenging problem which hinder the accurate measurement of the flow parameters.This challenge stimulates us to develop a new method to design a new four-sector distributed conductance sensor.Thereafter,oil-water and gas-water two-phase flow experiment are conducted respectively to obtain multivariate signals corresponding to different flow patterns.Then we use the multivariate pseudo Wigner distribution(MPWD)method to analyze the multivariate signals from the four-sector distributed sensor.The results show that MPWD enables to extract time-frequency features for probing the transient behavior of two-phase flow.The flow structure of two-phase flow is intricate and the transition mechanism of different flow patterns is still elusive.For solving this problem,we propose multivariate multiscale complex network(MMCN)to analyze the multivariate signals from the four-sector distributed sensor.The basic idea is as follows: We first perform a multivariate phase space reconstruction on each coarse-grained multivariate signals and then regard each phase space vector as a node and determine the edges in terms of their pairwise distance.We develop a new network measure,i.e.clustering coefficient entropy,to characterize the generated multiscale complex networks.The results suggest that the clustering coefficient entropy from the MMCN enables to probe the dynamical flow behavior and flow mechanism governing the transitions of two-phase flow.In addition,we design eight-electrodes cycle-motivation conductance sensor to capture more detailed flow information.Meanwhile,we develop a multivariate weighted recurrence network for realizing multi-information fusion,in which each sub-signal is deemed as the node and the weight of each edge is determined by the cross recurrence rate between pairwise sub-signals.In particular,we exploit graph energy and weighted clustering coefficient to characterize the derived complex networks.The results indicate that the network measures are very sensitive to the flow transitions and further allow uncovering local dynamical behaviors associated with water cut and flow velocity.These properties render our method particularly useful for analyzing multivariate time series.