Identification of flow regimes using conductivity probe signals and neural networks for counter-current gas–liquid two-phase flow

• Published in: Chemical engineering science Vol. 84; pp. 417 - 436
• Main Authors: Ghosh, S., Pratihar, D.K., Maiti, B., Das, P.K.
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 24.12.2012; Elsevier
• Subjects: air; Algorithms; Applied sciences; Back propagation; Chemical engineering; Descriptions; Electric potential; Exact sciences and technology; Flow regimes; Fluid mechanics; Hydrodynamics of contact apparatus; Multilayers; Multiphase flow; Neural networks; Optimization; Radial basis function; Simulation; Time series; time series analysis; Fluid mechanics; Multiphase flow; Simulation; Flow regimes; Neural networks; Optimization; Statistical analysis; Two phase flow; Time series; Entropy; Neural network; Countercurrent flow; Wire; Flow field; Vertical tube; Design; Radial basis function; Flow regime; Statistical parameter; Genetic algorithm; Thermodynamic properties
• ISSN: 0009-2509; 1873-4405
• DOI: 10.1016/j.ces.2012.08.042

A rigorous experimental investigation has been carried out on counter-current air–water two-phase flow in vertical tube with circular cross-section over a wide range of phase flow rates of air and water. It is a novel attempt to capture, identify and predict the individual flow regimes for different combinations of air and water superficial velocities based on some of its objective descriptions. Two different conductivity probes (parallel wire and ring probes) have been designed to collect objective descriptions of the flow regime in terms of the time series voltage signals. Statistical analyses have been carried out to characterize those time series voltage signals in terms of some statistical parameters. Three methodologies have been developed using a genetic algorithm (GA)-tuned multi-layer feed-forward neural network with back-propagation algorithm (GA-MLFFNN), a GA-tuned multi-layer radial basis function neural network with entropy-based clustering and back-propagation algorithm (GA-MLRBFNN–EC) and lastly, a GA-tuned multi-layer radial basis function neural network with fuzzy C-means clustering and back-propagation algorithm (GA-MLRBFNN–FCMC) to predict the different flow regimes after knowing the objective descriptions of the flow pattern in terms of the statistical parameters. Parallel wire conductivity probe is found to be better to capture the objective signature of the individual flow regime. GA-MLRBFNN–EC is found to be the best to predict flow regimes successfully. ▸ Experimental investigation on vertical gas–liquid counter-current two-phase flow. ▸ Complex nonlinear relationships between the developed flow regimes and input parameters. ▸ Automatic classification and prediction of flow regimes based on conductive probe voltage signals. ▸ Parallel probe is found to be the most effective one. ▸ GA-tuned radial basis function neural network with entropy-based clustering is found to be the most effective approach.