On the application of self-organizing neural networks in gas-liquid and gas-solid flow regime identification

Abstract

One of the main problems associated with the transport and manipulation of multiphase flow is the existence of flow regimes, which have a strong influence on important parameters of operation. An example of this occurs in gas-liquid chemical reactors in which maximum coefficients of reaction can be attained by keeping a dispersed-bubbly flow regime to maximize the total interfacial area. Another example is the pneumatic conveying of solids in which the regimes are associated with safety and energy consumption. Thus, the ability to identify flow regimes automatically is very important, specially to maintain multiphase systems operating according to design conditions. This work assesses the use of a self-organizing map (neural network) adapted to the problem of regime identification in horizontal two-phase flows. In order to achieve extensive results, two different types of two-phase flows were considered: gas-solid and gas-liquid. Tests were made to verify the performance of the neural network model, using data collected at the experimental facilities of the Thermal and Fluid Engineering Laboratory of the University of Sao Paulo at Sao Carlos. Results show that the neural network is capable of correctly identifying the regimes. The error percentage is bigger when analyzing the same regime with flow rates different from the one used as training data emphasizing the importance of training signals choice.