Development of a real-time objective gas-liquid flow regime identifier using kernel methods

Currently, flow regime identification for closed channels have mainly been direct subjective methods. This presents a challenge when dealing with opaque test sections of the pipe or at gas-liquid flow rates where unclear regime transitions occur. In this paper, we develop a novel real-time objective flow regime identification tool using conductance data and kernel methods. Our experiments involve a flush mounted conductance probe that collects voltage signals across a closed channel. The channel geometry is a horizontal annulus, which is commonly found in many industries. Eight distinct flow regimes were observed at selected gas-liquid flow rate settings. An objective flow regime identifier was then trained by learning a mapping between the probability density function (PDF) of the voltage signals and the observed flow regimes via kernel principal components analysis (KPCA) and multi-class Support Vector Machine (SVM). The objective identifier was then applied in real-time by processing a moving time-window of voltage signals. Our approach has: (a) achieved more than 90% accuracy against visual observations by an expert for static test data; (b) successfully visualized conductance data in 2-dimensional space using virtual flow regime maps, which are useful for tracking flow regime transitions; and, (c) introduced an efficient real-time automatic flow regime identifier, with only conductance data as inputs