Investigation of methane hydrate formation kinetics using machine learning and computational fluid dynamics tools

Zare, Marziyeh (2022) Investigation of methane hydrate formation kinetics using machine learning and computational fluid dynamics tools. Doctoral (PhD) thesis, Memorial University of Newfoundland.

[img] [English] PDF - Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.

Download (4MB)


Hydrate blockage in oil and gas facilities can cause a significant economic impact in terms of deferred production and remediation costs, particularly in harsh conditions (e.g., deep water). Financial considerations and safety concerns have motivated most operating companies to apply the hydrate management approach rather than the hydrate avoidance strategy. The hydrate management strategy requires a detailed understanding of how hydrates form, accumulate, deposit, and jam in pipeline systems. Despite all efforts that have been accomplished to find out the best method to manage and control hydrate formation in oil facilities, hydrate formation remains a challenge for the industry, and more research is needed to find reliable/effective methods for hydrate management. This research thesis starts with an extensive literature review, and the first series of simulation runs are performed to study methane hydrate formation in a jumper using a computational fluid dynamics (CFD) software, named Star CCM+. The numerical model for the simulation phase is developed through considering transport phenomena equations, including conservation of mass, momentum, and energy in which mass transfer, hydrate reaction kinetics model, and heat of hydrate formation are incorporated in the multiphase flow equations in the form of source terms in the CFD software. An extensive sensitivity analysis is performed to study the influences of changes in the inlet fluid velocity, gas volume fraction, inlet temperature, and subcooling on the hydrate formation in the jumper. The results indicate that the developed CFD model can simulate methane hydrate formation in the jumper with high precision. The amount of hydrate decreases when the value of the liquid inlet velocity and gas inlet temperature parameters increases. In contrast, an increase in subcooling and gas volume fraction leads to more hydrate formation in the jumper. More hydrate can be observed close to the wall, where the temperature is low and subcooling has high values. In the next phase, the induction time for the methane hydrate formation in the presence of Luvicap 55W (a kinetic hydrate inhibitor - KHI) solutions is determined using artificial intelligence models, including least squares support vector machine (LSSVM), adaptive network-based fuzzy inference system (ANFIS), and gene expression programming (GEP). For these models, 440 experimental data taken from the literature are employed, where 85% of data is utilized for the training step and 15% for the testing step. Induction time is considered as a target and the molecular weight of solution, mass fraction of KHI, temperature, pressure, and subcooling are the input parameters for these deterministic models. The performance of the smart models for the training and testing steps is evaluated using average relative error percent (ARE %), average absolute relative error (AARE %), and coefficient of determination (R²). Moreover, the Pearson correlation coefficient is calculated for the input parameters based on the ANFIS model to identify the influence of the input parameters on the induction time. The outcome shows that among LSSVM, ANFIS, and GEP models, the GEP technique has an excellent performance in predicting induction time. For instance, the values of the coefficient of determination (R²) for the developed GEP model are 0.9582 and 0.9726 in the training and testing steps, respectively. Also, the results reveal that the most influential parameters are the system pressure and temperature. Other input parameters, including the molecular weight of the solution, mass fraction of the Luvicap 55 W, and subcooling, have an indirect relationship with the induction time. In the next phase of this thesis, methane hydrate formation is simulated in an agitated reactor using Star CCM+ with a stirring rate of 300 RPM, a volume fraction of 0.04, and a pressure of 5,500 kPa. Then, the results are validated using the experimental data adapted from literature where an overall absolute average deviation (AAD%) of 15.6% is obtained. The effect of various parameters, including stirring rate, methane volume fraction, pressure, and subcooling is investigated on the hydrate formation in the stirred reactor. It is found that hydrate is formed more close to the wall and the impeller blades when the wall temperature is 274.15 K. Moreover, an increase in the parameters of stirring rate, methane volume fraction, pressure, and subcooling increases the amount of methane hydrate formation in the reactor. This CFD model can simulate hydrate formation in the stirred reactor with an acceptable accuracy. This model can be extended to other geometries of oil and gas facilities; it can be useful for corresponding industries to predict hydrate formation in transportation and processing facilities.

Item Type: Thesis (Doctoral (PhD))
Item ID: 15813
Additional Information: Includes bibliographical references
Keywords: hydrate CFD, jumper, stirred reactor, machine learning models
Department(s): Engineering and Applied Science, Faculty of
Date: October 2022
Date Type: Submission
Library of Congress Subject Heading: Machine learning; Computational fluid dynamics; Natural gas--Hydrates

Actions (login required)

View Item View Item


Downloads per month over the past year

View more statistics