Acadlore takes over the publication of IJCMEM from 2025 Vol. 13, No. 3. The preceding volumes were published under a CC BY 4.0 license by the previous owner, and displayed here as agreed between Acadlore and the previous owner. ✯ : This issue/volume is not published by Acadlore.
Open AccessFlow Regime Detection Using Gamma-Ray-Based Multiphase Flowmeter: A Machine Learning Approach
,,,,,
View Full Article|
Download PDFAbstract:
The presence of intermittent flow regime such as slug flow could cause issues to oil-gas well pipe- line-riser structures due to large fluctuations in pressure, leading to the production rate reduction and damage in the pipe structure. Monitoring multiphase flow regimes in production pipe systems is thus important. There are nowadays increasing use of multiphase flowmeter (MPFM) for well production flowrate metering. The associated phase fraction and flowrate measurement sensors in MPFMs could be potentially employed for multiphase-flow regime detection with no additional component required. In this study, a machine learning model is proposed to infer multiphase-flow regime from the measure- ments of a vertically installed gamma-ray and Venturi-based MPFM. Flow loop tests have been carried out at Singapore Well Testing Center with flows of various flow regimes observed at the horizontal inlet pipe section, such as dispersed bubbly, stratified, intermittent (slug) and annular flow regimes. The flow regime has been determined by visualization from a side glass in the flow loop pipe section and from real-time images reconstructed by an electrical-capacitance tomography system. The MPFM real-time measurements and derived or calculated data (such as Venturi differential pressure and gamma-ray mixture density) are then used as machine learning training data, with the flow regimes to be the train- ing target. Various machine learning methods have been experimented, such as convolutional neural network (CNN), long short-term memory (LSTM) and CNN-LSTM. It has been found that LSTM method with regularization, balancing and logarithmic normalization of the calculated parameters can achieve the highest accuracy on flow regime prediction (99.6%). This study is the first attempt to pre- dict flow regime at horizontal entrance section upstream of an MPFM with measurements made at a vertical Venturi throat section. The study also proves that flow regime could be accurately predicted by a gamma-ray and Venturi-based MPFM.
