Hydrate film growth and risk management in oil/gas pipelines using experiments, simulations, and machine learning

Abstract

The formation of gas hydrates in oil and gas flowlines is considered a major flow assurance (FA) issue due to hydrate blockage forming within minutes to hours. Traditionally, hydrate formation was prevented in oil and gas flowlines via the injection of Thermodynamic Hydrate Inhibitors (THIs), such as methanol or mono-ethylene glycol (MEG). THIs operate by shifting the hydrate phase equilibrium to higher pressure and lower temperature. However, this method can become uneconomic as large quantities of THIs (40 vol.%, with respect to the produced water) are often required. More recently, there has been a paradigm shift from “hydrate prevention” to “hydrate management”, in an effort to reduce operating costs. The latter method can involve allowing gas hydrates to form in the flowline, and controlling the hydrate formation and agglomeration by injecting Low Dosage Hydrate Inhibitors (LDHIs). The hydrate kinetic model CSMHyK was coupled to a transient multiphase flow simulator (OLGA) as a user-customized plug-in in this work. CSMHyK-OLGA was used to assess the level of hydrate resistance to flow in a black-oil field, in comparison with field observations. CSMHyK simulations can provide information on the hydrate volume fraction, the location of the hydrate resistance/plug, and the transportability of the fluid containing a hydrate slurry. Based on the field simulation study, it was concluded that besides water cut (WC) and gas-oil ratio (GOR), FA engineers should also take into account the hydrate cohesive force and the water-oil emulsion stability in delivering hydrate management strategies. In order to develop more advanced and cost-effective hydrate management strategies, hydrate plugging mechanisms, such as film growth and deposition at various flow conditions need to be investigated. Previous hydrate deposition studies focused predominantly on the gas-dominated systems where water can condense, be entrained, and settle on the upper part of the pipe wall. However, the deposition mechanism in the liquid phase is not well studied. A high-pressure lab-scale deposition loop, capable of handling gas-liquid flow, was used to study hydrate deposition on the pipe wall in this study. The parameters which affect hydrate film growth such as liquid holdup, liquid and gas phase velocity, water cut, and subcooling were varied. Visual ports of the high-pressure system were used to observe hydrate film growth and deposition. Based on the observed mechanism and the film growth rate, it was concluded that the hydrate film growth depends primarily on the accessibility of hydrate-forming components at the wall, and the driving force for hydrate formation. This experimental study supported the evidence that hydrate film growth in an oil-dominated system is controlled by water droplets wetting the pipe surface. In addition, after a hydrate deposit layer is established, there is a potential risk of sloughing and blockage. In the sloughing case study, it was concluded that hydrate sloughing occurs as a result of increased shear stress and warm bulk flow heating. So far there is no mathematical model that is able to describe the complex overall hydrate plugging mechanism. Standard industry predictive analytic tools like OLGA require certain a level of knowledge in fluid chemistry and multiphase flow to set up well-defined case studies. Usually, a numerical simulation study is also quite time-consuming. On the other hand, there exists a large amount of laboratory data that cannot be easily analyzed using normal statistical techniques. Machine learning, especially deep learning can serve as a universal approximator, to determine the mapping functions between operation conditions and the plugging risk and process phenomena (hydrate growth and relative pressure drop across a pipe). Machine learning models can achieve high prediction accuracy and calculation efficiency, given a sufficient number of data points and a good computer algorithm. In this thesis, we demonstrated that based on the data from the pilot-scale flowloop hydrate tests, machine learning models can predict hydrate plugging and hydrate volume fraction in the flowloop studies, as well as hydrate risks in the field pipelines.