Development of a Visual Digital Twin Dashboard for Two-Phase and Three-Phase Flow Regime Classification in Industrial Systems

Abstract

Accurate classification of multiphase flow regimes is critical for efficient operation in energy and process industries. While significant progress has been made in two-phase flow classification, extending these techniques to three-phase systems introduces additional complexity due to interactions among phases and variations in flow geometry. In this study, we develop a visual digital-twin-based dashboard for real-time classification and visualization of flow regimes in both two-phase and three-phase flows. Experimental data were generated using the multiphase-flow facility at Hamad Bin Khalifa University, whose principal horizontal test section is approximately 6 m long with a 50.8-mm (2-in.) internal diameter and transparent acrylic sections for video observation. The two-phase water-air dataset contains 65 no-leak experiments spanning liquid superficial velocity from 1.0711 to 3.2134 m/s and gas superficial velocity from 0.6252 to 3.4041 m/s. The three-phase oil-water-air dataset contains 36 experiments spanning water superficial velocity from 1.4059 to 3.1427 m/s and air superficial velocity from 0.063 to 0.320 m/s, including 12 no-leak, 12 single-leak, and 12 dual-leak experiments. Flow dynamics were characterized using high-response dynamic pressure sensors to capture time-series pressure signals representative of multiphase flow behavior. In addition, high-speed imaging (up to 2000 frames per second) was employed to capture the visual characteristics of the flow regimes. Supporting measurements included thermocouples, differential pressure sensors, and electrical process tomography to enhance flow characterization. Visual inspection of the three-phase records reveals bubbly and dispersed structures that are more heterogeneous than those in the two-phase water-air reference data, motivating an analysis workflow that preserves the relationship among sensor signals, video, and operating conditions. These synchronized experimental records are organized as digital threads and streamed to the cloud-hosted digital twin software-as-a-service (SaaS) system. As pressure data arrive, deployed leak-detection models analyze the signals, and the visual twin dashboard presents the associated time series, pipeline video, operating state, and mapped detection or localization outputs for operator review. The MGSB two-phase model achieved a mean held-out detection F1 score of 0.930 +/- 0.082 and an F1 score of 0.783 +/- 0.110 under the prescribed input-corruption test. The MGSB three-phase model achieved an F1 score of 0.984, ROC AUC of 0.9987, false-alarm rate of 0.09%, and mean detection delay of 0.03 +/- 0.01 s on an eleven-experiment dual-orifice evaluation matrix. The novelty of this work lies in integrating experimental multiphase flow measurements with AI-based classification and a visual digital twin interface capable of identifying both two-phase and three-phase flow regimes across multiple geometries in real time. This framework provides a practical decision-support tool for industry applications, including pipeline monitoring, flow assurance, separator performance evaluation, and multiphase transport optimization in oil, gas, and broader process industries. By enabling rapid and reliable flow regime identification using minimal sensor inputs, the proposed system has the potential to significantly improve operational efficiency, reduce uncertainty in multiphase transport, and support next-generation digital oilfield applications.