An artificial neural network applied to multiphase flow metering

This paper aims to show the feasibility of applying a multilayer feed forward (MLF) neural network to predict the liquid and gas flow rates of a multiphase mixture. Pressure drops in the converging and diverging parts of a multiphase Venturi meter and current values of void fraction and water cut, are used as input data for training and prediction. Non-linear processing capability of Artificial Neural Networks is used to investigate the complexity of multiphase flowing mixtures: non-linear nature of multiphase flows represents the main obstacle to obtain good information on flow rate parameters by using conventional signal processing techniques. Numerical modeling also requires the specification of inner rules which govern the flow evolution and this can hardly be done with a satisfying level of generality and accuracy. A comparison with the conventional signal processing techniques based on flow modeling has been realized.