A neural network algorithm for density measurement of multiphase flow

Measurement of the density of multiphase flow is crucial in multiphase flow meters. While radioactive-based measuring methods are known to give good results, end users prefer nonradioactive devices for obvious reasons. This paper shows the capability of an online neural network algorithm, that uses the differential pressure along a vertical pipe and across a venturi meter, in predicting the average density of the multiphase flow. A multiphase flow loop was constructed to conduct the training and evaluating experiments needed for the algorithm. Experimental results performed on the multiphase flow loop demonstrate that the density measurement can be achieved with good accuracy for liquid and gas velocities for up to 4 and 25 m/s (at 30°C and 7 bars), respectively, while covering different complex flow regimes including annular flow, slug flow, and dispersed flow. The neural network algorithm that was developed for this purpose gave very good results in measuring the flow density.