Gas-Liquid Two-Phase Flow Metering Using Coriolis Flowmeters

This thesis describes a novel methodology for the gas-liquid two-phase flow measurement using Coriolis flowmeters incorporating a semi-empirical physical model. A review of methodologies and techniques for the gas-liquid two-phase flow measurement is given, together with the advantages and limitations of using Coriolis flowmeters for such measurement. The proposed methodology can be implemented in the KROHNE OPTIMASS 6400 Coriolis flowmeters with an external input of GVF (Gas Volume Fraction). Detailed developments and evaluation of this analytical model and the comparison with other state-of-the-art solutions are reported. The parametric dependency between the behaviours of the Coriolis flowmeters and test conditions is discussed. Existing analytical models describing the interactions of gas and liquid phases in the Coriolis flowmeters are reviewed, evaluated, and improved according to the experimental data. Experimental results confirm that, with the corrections from the proposed physical model, 94.2% of the mass flowrate readings from the Coriolis flowmeters achieved a relative error of less than 10% while 97.8% of the GVF predictions achieved an absolute error of no greater than 5% within GVF of 0% - 45%.

Extensive experimental tests were conducted on two air-water two-phase test rigs with one-inch and two-inch bore test sections, respectively and a gas-liquid CO2 two-phase test rig with half-inch bore test section. A stability control algorithm is designed and implemented for this work, resulting in significant improvements on the performance of the rigs. Comprehensive tests under different test setups and conditions were carried out. These experimental results provide foundations for the development of the semi-empirical model in this work and for the future development of soft computing models for multiphase flow metering.

Same experimental data is used to compare the performance of the improved analytical model with that using soft computing models. The SVM (Support Vector Machine) model, which is believed to be the best among soft computing models, is able to allow 99.4% of experimental data to achieve less than 10% relative error and 5% absolute errors for mass flowrate and GVF, respectively.