Experiments and modelling of phase fraction in gas-liquid two-phase flow using a microwave resonant cavity sensor

Gas-liquid two-phase flow is prevalent in the natural gas industry, and accurate phase fraction measurement is crucial for enhancing productivity and energy efficiency in industrial processes. However, achieving high-precision, in-situ measurement remains challenging. To address this issue, this study proposes novel prediction models based on the microwave cylindrical resonant cavity (MCRC) sensor. Firstly, the MCRC sensor was implemented, and the experiments were conducted by incorporating a quick-closing valve calibration system into an existing gas-water reference system, capturing a multi-parameter dataset. The analysis indicated that a complex nonlinear relationship existed among phase fraction, relative frequency shift, pressure, and superficial gas velocity. Then, phase fraction prediction models, including void fraction and gas volume fraction (GVF) model, were developed using the empirical and machine learning modelling methods. The results revealed that empirical models without intermediate dielectric constant complex calculation achieved relative errors within ±5 %. Among the 5 machine learning models compared, the XGBoost model performed the best, with over 95 % of data points within ±2 %. Additionally, extended experiments were used to estimate the generalization ability of the GVF prediction models, demonstrating excellent performance. Finally, the comparative error analysis confirmed the superior accuracy of the proposed models. The findings suggest that the proposed models offer notable improvements in prediction accuracy and practical applicability, making them promising methods for phase fraction prediction in gas-liquid flow using the MCRC sensor in the natural gas industry.