Machine Learning Approach for Single Phase and Multiphase Flow Leak Detection and Monitoring in Onshore/Offshore Pipelines and Subsurface Sequestration Sites.

Leaks may occur in the existing pipelines although designed with quality construction and appropriate regulations. The economic impact of oil spills and natural gas dispersion from leaks can be huge. Failure to detect pipeline leaks promptly will have an adverse impact on life, the economy, the environment and corporate reputation. Therefore, early detection of leaks, their location, and size with a high degree of sensitivity and reliability is important for efficient hydrocarbon transportation through a pipeline, both in onshore and offshore applications. The objective of the proposed project is to develop a multiphase flow leak detection model and visualization tool that is ready to be used by the industry to integrate machine learning and digital twin technique. The proposed novel digital twin for leak detection will be used by industry for accurate prediction of the location, size, number, and orientation of both small chronic leak and larger leak with improved accuracy, sensitivity, reliability, and robustness. The ultimate goal is to take preventive action by artificial intelligence without requiring human interference. Our analysis will be based on experiments of single-phase (natural gas with impurities or crude oil), two-phase (oil/water or gas/water) mixture, three-phase (gas/water/solid) mixture, four-phase (oil/water/gas/solids) flow and its relation to the leak identification and dispersion prediction. For multiphase flow and leak characterization, we will use electrical resistance tomography, electrical capacitance tomography, high-speed visualization, pressure point, temperature point, differential pressure, dynamic pressure sensor, and flowmeter as the internal leak detection methods. We will also deploy several external leak detection methods such as laser diagnostics (particle image velocimetry), high-speed visualization, and hydrophone. We will monitor the real-time transient response of these internal and external methods concurrently and we confirm the leak event, avoiding any false alarm. The critical question is to differentiate the sensor response due to this leak for ocean wave or flow variation or external atmospheric flow direction or any other natural disturbances. The implementation of a numerical technique using Computational Fluid Dynamics (CFD) validating the base case experimental and mechanistic model data will enable us to investigate the effects of real industrial pipeline leak scenarios whilst reducing the amount of lab and field research, costly experiments, and ambiguity in understanding the fundamentals of leak detection in harsh environments. The proposed project will also help in reducing the carbon footprint by early detecting the gas/liquid leak and taking preventive actions. Hydrocarbon is extracted from the onshore or offshore production manifold, transported through a long pipeline, separated in the refinery where carbon dioxide is removed from the feed gas. Next, it is transported back to the injection well through an onshore/offshore pipeline and finally injected into the subsurface sequestration sites. In this respect, we will perform a leak detection study for the CO2 injection wellbore in a subsurface sequestration site. For computational modeling, we will use ANSYS Fluent, Olga, Matlab, PVTsim, and open-source IDAES platforms. From our previous experience of drift-flux modeling and multiple regression analysis of two-phase gas/liquid flow, solid/liquid flow, and gas/solid/liquid flow, we will develop a mechanistic model with leak and no-leak conditions. We will develop several non-dimensional numbers for scaling up the model and experimental data using the Buckingham-Pi theorem and using the multiphase mixture properties. We will propose a Nomograph for a better prediction of the fate of the release from a leak. From the proposed project (experimental and modeling), we will extract statistically significant data points and we will develop a multidimensional high-level statis

Hamad Bin Khalifa University (HBKU)