Multiphase Flow Metering with Multimodal Sensor Suite for Identification of Flow Regimes, and Estimation of Phase Fractions and Velocities
Abstract
In this Master Thesis, data-driven multiphase flow metering models are developed to classify flow regimes and to estimate phase fractions and velocities of two-phase air/water flow after collecting data from the horizontal flow rig situated at University of South-Eastern Norway (USN).
Two types of multimodal sensors are used in this thesis namely Electrical Capacitance Tomography (ECT) and Ultrasonic Sensors (US). Experiments are performed on the flow rig at USN with ECT to collect capacitance data of the flow. Conventional measurements of pressure and flow rate are also collected during the experiments. Ultrasonic transit time data was already available through historical experiments.
Exploratory data analysis is performed on ECT and US data for feature engineering. ECT and US features are used to train and test flow regime classification models. Machine learning algorithms including Decision Tree, K-Nearest Neighbors, Artificial Neural Networks and Support Vector Machines are mainly employed to train flow classification and phase fractions estimation models. For flow velocity estimation, cross correlation technique is employed on 2-planes ECT data. Lastly, comparison of flow visualization with ECT and US data is performed. The images produced from ECT and US data are also compared for interface detection in multiphase air/water flow.
The flow regime classification models using ECT achieved an accuracy of more than 96%. Sensor fusion models of flow regime classification achieved accuracy of more than 97%. Flow velocity was accurately estimated using cross-correlation for slug regime. The phase fraction estimation Neural Network model achieved an R value of more than 0.95.