Ensemble learning in the estimation of flow types and velocities of individual phases in multiphase flow using non-intrusive accelerometers' and process pressure data

in oil and gas industries. Accurately identifying flow types and estimating flow velocities of the individual phases are crucial for different purposes, such as observing the process status and providing inputs to control systems. This paper presents a solution for identifying flow contents and estimating flow rates in single-phase or each phase in multiphase flows by using pressure measurements and pipe vibrations caused by the flows. The necessary experiments were performed using the multiphase flow rig with three-inch diameter pipelines transporting natural gas, water, and crude oil in a closed loop with a separator tank as source and sink. A series of tree-based ensemble machine learning models have been developed and tested with the data collected from accelerometers, differential pressure transmitters, and upstream- and downstream pressure transmitters. With these inputs, the developed models can identify volume ratios of individual phases (such as water cut) and can estimate the flow velocity of each phase in the flow loop, including the open/close status of the choke valve. After describing briefly, the P&ID diagram of the multiphase flow rig, the paper focuses on exploratory data analysis of the data from three accelerometers and three pressure sensors using three submodels cascaded to perform ensemble learning.