Two-phase flow regime identification combining conductivity probe signals and artificial neural network

Important aspects of the hydrodynamics and thus, the correct identification of the flow regime could enhance safety and overall performance in multiphase flow systems. Several works on flow regime identification have been carried out in the past. Most of them consist, in a first stage, on measuring certain flow parameters that can be used as good flow regime indicators and, then, developing a flow regime map using these indicators. In this work, a vertical two-phase flow loop facility was used, whereby local conductivity signals were recorded and utilized for the development of an Artificial Neural Network (ANN) based method for the flow regime classification. The experimental database consists of a total number of 125 test cases covering a wide range of situations in the loop working area. Each experiment flow regime was identified by visual inspection, and classified into bubbly (13), cap-bubbly (CB), slug (S), churn turbulent (CT) or annular (A). The bubble chord length cumulative probability function (CPDF), calculated from the measured conductivity signals was selected as flow regime indicator. Different ANN configurations were designed, trained and optimized. The ANN types and the statistical parameters of the CPFD used as inputs were two of the four parameters varied in the ANN optimization process. The temporal length in the conductivity signal used to calculate the CPDF, was also modified during this study. The range of variation of the temporal signal was from I to 60 seconds. Together with this temporal variation, the number of training patterns (increasing with decreasing CPDF processing times or not) was also modified. The modification of all these variables allowed the identification of the ANN configuration that better fit the requirements of the specific study of flow regime classification.