Masoumeh Akhfash. This thesis is presented for the degree of. Doctor of Philosophy. The University of Western Australia

Transcription

1 Hydrate Formation and Particle Distributions in Water-Dominated and Partially-Dispersed Systems by Masoumeh Akhfash This thesis is presented for the degree of Doctor of Philosophy of The University of Western Australia School of Mechanical and Chemical Engineering Fluid Science & Resources Division May 2017

2 Summary Most subsea oil and gas flowlines are naturally at risk of hydrate formation and subsequent blockage if expensive prevention measures are not taken. The risk of hydrate blockage, and the associated cost of avoidance, increases with field age as the amount of produced water grows significantly. However, while the conceptual mechanism by which hydrate blockages occur is wellestablished for multi-phase mixtures where oil is the continuous phase, only one previous study has proposed a description of plug formation in water-dominant systems, and no conceptual mechanism is available for partially-dispersed systems where separate continuous phases of water and oil exist. The primary aim of the research presented in this thesis was the development of more detailed conceptual mechanistic descriptions for gas hydrate plug formation in systems relevant to late field life. This was achieved by conducting an extensive set of high-pressure experiments in specialised autoclaves, which enabled new and direct insights into the relevant interactions between hydrate particles. A sapphire high-pressure autoclave, with a 1 inch internal diameter (ID) and the capacity for coincident measurements of resistance-to-flow and hydrate growth rate, was designed and used to validate a hypothesis in the literature about water-gas systems derived from flow-loop experiments: that the first increase in the measured resistance to flow of a water-continuous hydrate slurry corresponds to the transition from a homogenous to a heterogeneous distribution of particles at a critical hydrate volume fraction, φ transition. The sapphire cell enabled the first reported visual confirmation of the moving hydrate bed s formation at volume percent in the range 15 to 20 vol %, which corresponded with the first measurable increase in resistance-to-flow based on the autoclave s motor current. At medium shear rates, φ transition was also detectable from a coincident acceleration in growth rate as the water-gas interface was disrupted by the formation of the moving bed. The addition of monoethlyene glycol (MEG) to a concentration of 10 wt % was also found to increase the value of φ transition by about 10 vol % relative to the pure water case, suggesting the potential of an additional benefit to under-inhibited production strategies in water-dominant systems. Further experiments with the 1 inch sapphire autoclave investigated the formation of methane hydrate beds at the gas-water interface in continuous cooling/flow experiments, in shut-in/restart experiments, and as a function of Reynolds number. For both continuous cooling/flow and shutin/re-start experiments, the value of φ transition was observed to increase with the level of shear applied, although the initial hydrate growth-rates for shut-in/restart experiments were an order of magnitude larger than comparable constant cooling/flow experiments. While high shear was observed to increase hydrate growth rate, it delayed the onset of hydrate bed formation, suggesting the possibility of trading off these competing components to avoid a hydrate blockage in water- i

3 dominant systems. However, while the values of φ transition at which increases in resistance to flow were observed were consistent between autoclave and flow-loop experiments, the Reynolds number at which a given φ transition was achieved could not be used to upscale and compare autoclave and flow-loop results. This indicated that some other characteristic length-scale underlies the definition of φ transition. To improve the resolution with which the evolution of the moving hydrate bed could be studied, a 4 inch ID autoclave equipped with in-situ FBRM and PVM probes was used to study hydrate formation in gas-water mixtures. The FBRM distributions (particle size and particle number concentration) were used to estimate the onset of particle interaction and bed formation. These micron-scale observations reveal for the first time that hydrate particle interactions in a watercontinuous phase commence at concentrations of 2-9 vol % depending on shear rate, roughly when the average separation distance between hydrate particles becomes comparable with the average hydrate particle size. Furthermore FBRM data showed that the value of φ transition derived from resistance to flow measurements (at 16±2 vol %) occurs once the process of forming a hydrate bed (as inferred from FBRM measurements) is largely complete. Such results could facilitate development of a scaling relationship for predicting the onset of hydrate plugging behaviour in water-continuous systems. Finally, the 1 inch sapphire autoclave was used to study partially-dispersed mixtures with a free mineral oil phase on top of a free water phase. Models based on the conceptual mechanisms for either oil-dominant or water-dominant systems were unable to describe the observed growth rates and resistance to flow measurements. Accordingly, a new conceptual mechanism for plugging was proposed based on the previous model for oil-dominated systems, but with four additional stages: initial wall film growth; disruption of the stratified oil-water interface; full dispersion of the oil and water phases; and, rapid hydrate growth and wall deposition. Potentially, the wall film growth and deposition stages identified here could also occur in fully-dispersed oil-dominant flows. Additionally, some observations suggested the presence of a hydrate bed at the oil-water interface prior to its disruption. However, future higher resolution experiments are needed to explore these hypotheses. The work in this thesis provides valuable new insights into hydrate plug formation processes in water-dominant and partially-dispersed systems, which are particularly relevant to industrial production scenarios in late-field life. While further experiments with autoclaves, flow-loops and flowlines are still required, the results of this research lay the foundations for quantitative models of hydrate growth as well as resistance to flow as a function of shear rate and hydrate volume fraction. Such models are essential to the development of risk-based approaches to hydrate management, which are increasingly needed by industry in place of conservative but costprohibitive strategies of total hydrate avoidance. ii

4 Table of Contents Summary... i List of Figures... vi List of Tables... xii Acknowledgements... xiv 1 Introduction Outline of this thesis Publications arising from this thesis Hydrate formation and particle distributions in gas-water systems Introduction Experimental Method Apparatus Procedure Modelling Approach Results and Discussion Hydrate formation, particle distributions and flow resistance Variations in mass transport Effect of turbulence Under-inhibited systems Chapter Conclusions Cold restart Methane Hydrate Bed Formation in a Visual Autoclave: Cold Restart and Reynolds Number Dependence Introduction Experimental Methods Results and Discussion Evolution and shear rate dependence of hydrate transportability Evolution and shear rate dependence of hydrate growth Chapter Conclusions Microscale Detection of Hydrate Blockage Onset in High-pressure Gas-Water Systems iii

5 4.1 Introduction Materials and Methods Apparatus Experimental Procedure Results Resistance-to-flow measurements In Situ Images of Hydrate Formation FBRM Measurements Discussion Chapter Conclusions Gas Hydrate Plug Formation in Partially-Dispersed Water-Oil Systems Introduction Materials and Methods Apparatus Experimental Procedure Results and Discussion Visual Observations Hydrate Growth Rate Resistance-to-Flow Measurements Chapter Conclusions Supporting Information Conclusions and Future Work Conclusions Suggestions for future work Appendix A Hydrate Shell Growth Measured Using NMR A.1. Introduction A.2. Background A.2.1 Relevant NMR theory A.2.2 Relevant hydrate theory A.3. Methodology iv

6 A.3.1. Materials and methods A.4. Results and Discussion A.4.1. Hydrate Growth Kinetics A.5. Chapter Conclusions References v

7 List of Figures Figure 1. Conceptual hydrate plugging mechanism for high water content systems, adapted from Joshi et al. (2013b) Figure 2. Simplified schematic diagram (left) and picture (right) of the High Pressure Visual Autoclave (HPVA), adapted from Boxall and May (2011)... 9 Figure 3. Images of different hydrate formation stages in the HPVA for experiment 1. (a) Before hydrate formation; (b) 14 minutes after nucleation, at 1.4 vol % hydrate; (c) 72 minutes after nucleation, at 9.5 vol % hydrate; (d) 101 minutes after nucleation, at 19.6 vol % hydrate; (e) 105 minutes after nucleation, at 22 vol % hydrate; (f) 200 minutes after nucleation, at 53 vol % hydrate Figure 4. Cell pressure (left ordinate) and bath temperature (right ordinate) after initial hydrate formation in experiment 2, where experimental data (black) are compared with the pressure predicted using the mass-transfer limited formation model (grey) (eq. (2)). The hydrate equilibrium temperature (red) was calculated from the measured pressure and is always greater than the bath temperature indicating the experiment was not heat-transfer limited. The experiment s initial Reynolds number was (For interpretation of this color figure, refer to the web version of this thesis.) Figure 5. Hydrate volume % (left ordinate) and relative motor current (right ordinate) after initial hydrate formation in experiment 2, where experimental data (black) is compared with a masstransfer limited formation model (grey). The experiment s initial Reynolds number was Figure 6. Relative motor current as a function of hydrate volume % in experiments 1 (grey), 2 (black) and 3 (red). All experiments were performed at initial Reynolds numbers of approximately (For interpretation of this color figure, refer to the web version of this thesis.) Figure 7. Measured (black) and predicted (grey) hydrate volume % for experiment 4 (initial Re = 4930) and experiment 6 (initial Re = 310) Figure 8. Relative motor current as a function of hydrate volume % at different impeller speeds: experiment 6 (initial Re = 4930, black); experiment 7 (initial Re = 4930, grey); experiment 8 (initial Re = 310, black); and experiment 9 (initial Re = 310, grey) Figure 9. Images of different hydrate formation stages in the HPVA for experiment 10 (10 % MEG initially): (a) 87 minutes after nucleation, at 15.6 % hydrate volume showing a homogeneous distribution of hydrate particles; (b) 123 minutes after nucleation, at 25% hydrate volume and the vi

8 onset of (visual) φ transition ; (c) 137 minutes after nucleation, at 28 % hydrate volume shortly after (visual) φ transition ; and (d) 312 minutes after nucleation, at 47% hydrate volume near the end of experiment Figure 10. Methane consumption rate (left ordinate) for 0 % MEG (experiment 2, black curve), 10 wt% initial MEG concentration (experiment 8, grey curve), and MEG concentration for experiment 8 during hydrate formation (right ordinate). The MEG concentration was calculated based on a mass balance for the closed system, assuming no MEG was incorporated in the hydrate phase Figure 11. Relative motor current as a function of hydrate volume % in experiment 10 (10 wt % initial MEG concentration). This experiment was performed at 1140 Re Figure 12. Images of methane hydrate formation under (left) constant cooling at 50 RPM and (right) shut-in and restart at 50 RPM: (A) hydrate nucleation at the water-gas interface; (B) hydrate film growth at the wall near the water-gas interface; (C) hydrate bed formation; and (D) late-term hydrate growth and catastrophic plug formation Figure 13. Onset of hydrate bed formation (φ transition ) as a function of initial Reynolds number in the HPVA apparatus Figure 14. Top: Relative motor current (resistance-to-flow) as a function of time after nucleation for constant cooling/flow trials with initial Reynolds numbers of 1145 and 1450, with the maximum value observed in each trial circled. Bottom: Maximum relative motor current observed during constant cooling and shut-in/restart experiments, as a function of initial turbulence of mixing. The relative motor current is defined as the motor current measured during hydrate formation, normalized by the baseline motor current prior to hydrate formation Figure 15. Hydrate volume %, with respect to the liquid phase, as a function of time after nucleation for experiments 6 and 17 (250 and 800 RPM, respectively); the dashed curve represents predictions from the mass-transfer limited model (eq. (2)) for each experiment Figure 16. Maximum hydrate volume % (of original water volume) for constant cooling/flow (solid circles) and shut-in/restart (open circles) data, as a function of the initial turbulence of mixing; the cell is expected to be fully turbulent at Reynolds numbers above approximately 1000 (Naumann, 2008) Figure 17. Initial hydrate growth rate for continuous cooling/flow (filled data points) and shutin/restart (open data points) experiments listed in Table 5 as a function of initial turbulence of mixing in the cell. Growth rate was defined for the first 2 vol % of hydrate following nucleation or the restart of shear. The solid curve corresponds to the model from Skovborg and Rasmussen (1994), based on diffusion-limited hydrate growth in an aqueous phase vii

9 Figure 18. Simplified sketch (left) and picture (right) of the high-pressure autoclave cell Figure 19. (a) FBRM probe assembly; (b) the procedure of FBRM chord length measurements; (c) PVM probe assembly; (d) an example of PVM imaging: methane gas bubbles in water after hydrate dissociation, captured from one of the experiments performed in this work. Sketches (a) (c) are redrawn based on information from Mettler-Toledo-Autochem (2016) Figure 20. Relative motor current (resistance-to-flow) as a function of hydrate volume % for a gaswater system at 300 RPM (experiment 4). The insert figure contains the same data with a magnified ordinate range to schematically show how φ transition and φ + transition were calculated. The values reported in Table 7 were obtained based on the intersection of the fitted lines Figure 21. PVM images captured at different times after nucleation during the experiment carried out at 300 PM (experiment 8 in Table 7) Figure 22. Mean square weighted chord lengths (a-d) and chord counts (e-f) of gas bubbles and hydrate particles measured by FBRM as a function of time post nucleation for various impeller speeds. Prior to nucleation, the measured chord lengths and chord counts are indicative of the size and number of the of gas bubbles detected by the FBRM Figure 23. Chord counts-chord length distributions measured by FBRM, reported at different hydrate volume % for methane hydrate-in-water slurries. (Left panel): experiment 2 at 100 RPM and (right panel): experiment 6 at 700 RPM. Grey traces correspond to a sequence of distributions that increase in height with increasing hydrate volume %, while colored traces correspond to a sequence of distributions that decrease in height with increasing hydrate volume %. (For interpretation of the references to color in this figure legend, refer to the web version of this thesis.) Figure 24. Calculated hydrate surface area based on the FBRM chord lengths data as a function of hydrate volume % for six gas-water experiments at impeller speeds of 100 to 700 RPM Figure 25. Schematic representation of hydrate particle distribution evolution in water-continuous systems based on FBRM, pressure (this work) and visual measurements (Akhfash et al., 2013). Three regions of homogenous, heterogeneous and plugging conditions are considered; conceptual diagrams are shown for each stage to qualitatively represent the distribution of hydrate in the autoclave. In the autoclave schematics, white, green, and blue colors correspond to the hydrate, gas, and water phases, respectively. (For interpretation of the references to color in this figure legend, refer to the web version of this thesis.) viii

10 Figure 26. Calculated edge-to-edge distance between solid particles in the water phase based on a simplistic model for experiments 2 to 7 corresponding to mixing speeds of 100 to 700 RPM. (For interpretation of this color figure, refer to the web version of this thesis.) Figure 27. Conceptual picture of hydrate plug formation mechanism in oil-continuous, adapted from Turner et al. (2009b) Figure 28. Simplified schematic (left) and picture (right) of the high-pressure visual autoclave (HPVA) cell Figure 29. Images of mixing prior to hydrate formation captured in the sapphire autoclave (25.4 mm inner diameter) for systems at 500 RPM and different water cuts. (A) A fully dispersed system with 30% water cut. (B) A partially dispersed system with 70% water cut Figure 30. Images of the sapphire autoclave after hydrate formation at six stages throughout experiment 18 (70 % water cut, 500 RPM): (C) 68 minutes after nucleation with 3.1 vol % hydrate; (D) 100 minutes after nucleation at 5.5 vol % hydrate, beginning of the water-oil interface disruption; (E) 120 minutes after nucleation at 7.9 vol % hydrate; (F) 162 minutes after nucleation at 17.3 vol % hydrate; (G) 204 minutes after nucleation at 48 vol % hydrate; and (H) 310 minutes after nucleation with 63 vol % hydrate. (For interpretation of this color figure, refer to the web version of this thesis.) Figure 31. Initial gas consumption rate over the first 10 minutes after nucleation in each experiment for all experiments with repeat trials represented as an average value Figure 32. Experimental (x) and predicted (black curves) hydrate volume % as a function of time at 10% water cut for (a) experiment 1 at 300 RPM (PD) and (b) experiment 3 at 900 RPM (FD).. 79 Figure 33. Hydrate volume % (left axis) and motor torque (right axis) as a function of time at 70% water cut (experiments 17-20).(For interpretation of this color figure, refer to the web version of this thesis.) Figure 34. Torque (resistance-to-flow) as a function of hydrate volume % over 10-70% water cut at 300 RPM, where all experiments were classified as partially-dispersing. The severe torque fluctuations observed from vol % hydrate are associated with plugging-type behavior. (For interpretation of this color figure, refer to the web version of this thesis.) Figure 35. Torque (resistance-to-flow) as a function of hydrate volume % for 70 % water cut systems at 300 (black, PD), 500 (grey, PD, 2 repeats), and 900 (red, FD) RPM. The inset panel contains the same data with a magnified ordinate range, to enable better comparison of the torque ix

11 curves prior to the plugging-type fluctuations.(for interpretation of this color figure, refer to the web version of this thesis.) Figure 36. Torque as a function of hydrate volume % for % water cut at 900 RPM, at which speed all systems were fully dispersing Figure 37. Maximum torque achieved during hydrate formation as a function of initial water cut in the system, at the three different mixing velocities tested. All 300 RPM experiments and those with 50-70% water cut at 500 RPM were classified as PD Figure 38. Proposed conceptual mechanism for hydrate plug formation in partially-dispersing oil and water systems, based on the results presented in this work. This mechanism is adapted from that proposed by Turner et al. (2009b) with additional stages observed here for PD systems shown in red text. (For interpretation of this color figure, refer to the web version of this thesis.) Figure 39. Conceptual picture of hydrate formation in an oil dominant multiphase flow line (adapted with permission from Turner (2005)). Hydrate shell formation, indicated as white coloring, occurs around emulsified water droplets in the growth and agglomeration stages Figure 40. Modified SSE PFG pulse sequence which is able to differentiate between the water and the oil (cyclopentane) signal based on adequate differences in the T 1 relaxation parameter this was enabled by dissolving a Gd complex in the cyclopentane prior to use Figure 41. Results of the T 1 relaxation experiment. (a) T 1 distribution curves, showing the two sets of peaks for cyclopentane and water respectively; (b) water mole fraction obtained for two independent experiments showing the reduction in water mole fraction as a thicker hydrate shell is formed. (For interpretation of this color figure, refer to the web version of this thesis.) Figure 42. Results of the T 2 relaxation experiment for ice particles of nominal diameter µm. (a) T 2 distribution curves, showing the two separate group of peaks for cyclopentane and water respectively; (b) water mole fraction calculated by integrating the relaxation peaks for water and cyclopentane over time and adjusting for 1 H density for three different ice particle sizes. (For interpretation of this color figure, refer to the web version of this thesis.) Figure 43. (a) droplet size distribution (DSD) obtained with the PFG NMR diffusion experiments. A decrease in the droplet mean value is observed over 9 days, corresponding to the shrinking of the water core; (b) comparison between the mean droplet size obtained by self-diffusion and T 2 relaxation measurements respectively for two independent experiments. The solid lines are the predicted change in droplet size from day two onwards, based on the residual water content measured and reported in Figure 4(b), using the diameter, as measured using relaxation x

12 measurements, on day 2 as an anchor point. (For interpretation of this color figure, refer to the web version of this thesis.) xi

13 List of Tables Table 1. Summary of the experiments conducted Table 2. Summary of results for experiments with no MEG in the aqueous phase. The equilibrium temperature corresponding to the pressure at which nucleation occurred had an average value of (8.4 ± 0.1) C for these nine experiments. The equilibrium temperature corresponding to the final system pressure, T eq,final, was always above the final system temperature, indicating that the hydrate formation was not heat transfer limited. The duration of the experiments was typically 24 hours, with hydrate nucleation and significant growth occurring over a period of about 14 hours, starting about 8-10 hours after beginning the experiment Table 3. Three independent measures of the hydrate volume %, φ transition, at which the transition from a homogeneous to heterogeneous particle distribution occurred, identified from the motor current, pressure decrease, and visual observations. The reader is referred to Table 1 and 2 for other parameters relevant to each experiment number Table 4. Summary of results for experiments that started with 10 wt % MEG in the aqueous phase. The final MEG concentration reached in each experiment (based on the final hydrate volume % calculated from the measured decrease in gas pressure) is also listed. The equilibrium temperature corresponding to the pressure at which nucleation occurred had an average value of (7.90 ± 0.02) C for these four experiments. The equilibrium temperature corresponding to the final system pressure, T, was always above the final system temperature, indicating that the hydrate eq,final formation was not heat transfer limited Table 5. Summary of experiments conducted Table 6. Summary of experimental results from constant cooling/flow and shut-in/restart experiments, together with values calculated with an equilibrium model (Infochem, 2012) and a mass-transport limited model (Akhfash et al., 2013) Table 7. Summary of experiments conducted in the autoclave cell. In all experiments, the hydrate equilibrium temperature was 9.0 ± 0.1 ºC. The hydrate volume % in the liquid phase at the end of each experiment is denoted φ final. The relative motor current (R.M.C.) was calculated based on the ratio of the instantaneous motor current (post nucleation) to the average motor current measured prior to nucleation. The quantities φ transition and φ + transition represent the average and maximum hydrate volume % at which an increase in relative motor current above R.M.C. ave +u and R.M.C. ave +3u was xii

14 observed, respectively, where u is the standard deviation of the relative motor current measured prior to φ transition Table 8. Hydrate volume % at which the peak (φ 1 ) and the plateau (φ 2 ) were achieved in the plots shown in Figure 24. Also listed are the values of φ transition determined from motor current data in the same experiments as described in Section 3.1 and reported in Table Table 9. Approximate composition of the paraffin oil (at atmospheric pressure) used in HPVA experiments, from gas chromatography/mass spectrometry analysis Table 10. Summary of the experiments conducted on partially-dispersing (PD) and fully-dispersing (FD) systems of deionized water (W) and paraffin oil (O) Table 11. Summary of the results for all experiments. The final temperature of all experiments was 1.3 ± 0.2 ºC. The equilibrium temperature corresponding to the pressure at which nucleation occurred had an average of 8.0 ± 0.2 ºC. The quantity φ final represents the hydrate volume % in the liquid phase at the end of each experiment. The average number of moles of methane in the gas phase at nucleation was 0.23 ± moles Table 12. Hydrate volume % at which oil-water interface disruption, oil-water full entrainment and rapid growth started in the PD systems. The superscripts indicate whether the volume % was determined visually or, in the case of the rapid growth transition, determined independently from the measured pressure xiii

15 Acknowledgements Foremost, I would like to express my gratitude to my principle supervisor, Professor Eric May, for giving me the chance to experience life and study in Australia. I want to thank you for your excellent supervision, for being an inspiration in many ways, and for all of the support I was given during my PhD research. I have learned so much from you over the past few years, not only in the scientific area, but also on a personal level. My sincere thanks also goes to my co-supervisor, Associate Professor Zachary Aman. I appreciate your continuous support, positive attitude, and immense hydrate knowledge you maintained during my PhD. I am also thankful for all the time and effort you put into revising the publications that have arisen from this thesis; you helped me improve my writing skills. I appreciate your assistance in implementing the mass-transfer and kinetic models I used to quantify hydrate formation rate in gas-water and oil-water systems. I also would like to thank my other co-supervisor, Professor Michael Johns. At all levels of this research project, I have been appreciative of your wealth of knowledge and insightful comments. It was a privilege to work on NMR-Hydrate project under your supervision. I also wish to thank Dr. John Boxall for being my co-supervisor during the first couple of months of my PhD. The high-pressure autoclave apparatus he built was used to generate a significant amount of the results that went into this thesis. I would like to thank Professor Carolyn Koh, director of the Centre for Hydrate Research at Colorado School of Mines for giving me the opportunity to spend six months of my PhD with her group in Colorado. I am grateful for your knowledge, guidance and kindness throughout my stay at CSM. Also thanks to Dr. Jianwei Du, Dr. Litao Chen, Dr. Bo Ram Lee and Dr. Giovanny Grasso for their assistance with my experiments. A big thanks to Dr Paul Stanwix, Dr. Thomas Hughes and Dr. Brendan Graham for all their help in improving my experimental and troubleshooting skills in the laboratory. I would also like to acknowledge Dr. Agnes Haber for teaching me the NMR basics in an easy to understand way and her contribution to the NMR section of this thesis. I wish to thank Dr. Einar Fridjonsson for answering all my NMR-related questions during my PhD. I must also acknowledge Mr. David Amm, the senior technician in the mechanical workshop at UWA for his technical assistance in building the apparatus and tools I have used in the laboratory during these four years. Thanks also goes out to those at the UWA electrical workshop who provided me with technical support, in particular, Mr. Donald Allen. I would also like to thank Mr. Mike Stadick at CSM for assistance in maintaining the autoclave system. xiv

16 A big thank you to the University of Western Australia for providing me with my main scholarship, without which, I would have been unable to undertake this journey in Australia. Additional thanks to WA:ERA for their support through the Australia-China Natural Gas Partnership Fund. I wish to acknowledge my colleagues in the Fluid Science and Resources Division, for the time we were working together, and for all the fun we have had in the last four years: Song Ng, Dr. Vahab Honari, Dr. Tauqir Syed, Nathan Jensen, Dr Jerry Guo, Dr. Tom Saleman, Kumarini Seneviratne, Bruce Norris, Marco Zecca, Shane Morrissy, Paul Connolly, Nicholas Ling, Jordan Oakley, Yahua Qin, Narmada Rathnayake and Zhikao Li. I would also like to thank students in the hydrate group at CSM, in particular, Dr. Naveed Khan, Yue Hu, Davi Costa Salmin, Vishal Srivastava, Dr. Zachary Ward, Xianwei Zhang and Jose Dapena, who made my stay at CSM very enjoyable. A big thank you to my housemate at Mines Park in Golden, Megha Gandhi, for all the fun times we had together. You made my stay in Golden a happy one. I would also like to thank Dr. Cericia Martinez for showing me around Golden when I first arrived. A special thanks to my friends Dr. Leila Hiedarvand and Dr. Agnes Haber for their immeasurable support every time I needed it throughout my PhD. Finally, my deepest gratitude goes to my family. To my late father for teaching me strong moral values; to my hard-working mother for always supporting me in whatever path I choose; and to my sister Zari for being a constant source of encouragement and support. xv

17 If your tree has the fruit of wisdom, the universe shall be at your command xvi

18 1 Introduction 1.1 Outline of this thesis This thesis is organised as a series of five papers that have all been published and are covered in Chapters 2, 3, 4, 5 and Appendix A. While each chapter is an independent part of the work and preserves the content of the published paper, the combination of the chapters presents a coherent story. The relevant background literature and the motivation for the work are presented in each chapter; here an introduction to each chapter is given and the way they link together is described. Avoiding hydrate blockages is critical for the safe and economical production of deep-water fields as well as mature reservoirs nearing the end of their production life. As these assets produce large volumes of water, the use of thermodynamic hydrate inhibitors to prevent hydrate formation may not be economically feasible. Understanding plug formation mechanisms allows industry to better assess the plugging risk during operations. However, in contrast to the well-studied case of oildominated systems in which all the water is emulsified in the oil phase, in pipelines with a highwater cut, a free water phase might be present and the mechanism by which hydrate plugs form in such systems is not well understood. Recent flow-loop studies suggested that the transition to a heterogeneous particle distribution and the formation of a hydrate bed (a region where particles are highly-concentrated) represents a critical and irreversible stage in the formation of a hydrate blockage in water-dominated systems. However, this transition was inferred from in-direct measurements, and no visual observation of the change in hydrate particle distribution had been reported, and limited information about how this transition occurs was available. Another crucial knowledge gap exists for systems with moderate to high water cuts in the presence of appreciable amounts of light liquid hydrocarbons, such as gas condensates, which often form so-called partially-dispersed systems. In such systems, hydrate conceivably might form from restricted water droplets dispersed in the oil phase and/or from entrained gas bubbles in the water-continuous phase, and no unambiguous evidence was available about the plugging pathway such systems follow. To elucidate these important knowledge gaps the research in this PhD integrated a video stream with either macro- or microscopic resolution into an extensive set of autoclave experiments in gas-water systems and partially-dispersed oil-water conditions. First, as explained in Chapter 2, a 1 high-pressure sapphire autoclave apparatus was designed with the capacity to collect three independent and parallel types of measurement: hydrate growth rate (through pressure drop calculations), resistance-to-flow (through motor current readings) and visual observation of the spatial distribution of hydrate particles (through video records). The apparatus was used to study the relationships between these quantities upon hydrate volume fraction in 100 % water cut scenarios. These observations confirmed the importance of a critical 1

19 hydrate volume fraction, φ transition, at which the system exhibits a measurable increase in the resistance to flow, and which was observed to correspond with the formation of a moving hydrate bed at the gas-water interface. The measured hydrate growth rate in these water-dominated systems were compared with the predictions of a simplified model with no adjustable parameters. The differences between the model s predications and the experimental data were explainable in terms of mass-transfer limitations present in experiments, and our hypophysis was verified through conducting further tests at different shear rates. Finally, supplementary experiments were performed by adding 10 wt % mono ethylene glycol (MEG) to the aqueous phase to study whether MEG increased the risk of hydrate blockage in gas-water systems when complete hydrate inhibition was not achieved. The results help inform the possibility of using under-inhibition in gas-water systems as a risk management technique. In Chapter 3, the investigation started in Chapter 2 into the factors affecting φ transition in gas-water systems was extended to explore the dependence of φ transition on shear rate, and whether any differences exist between systems that undergo continuous mixing with those that are subject to shut-in/cold restart conditions. In shut-in/restart cases, mixing started when the system already contained a layer of hydrate at the gas-water interface, with the system having been within the hydrate formation region for at least one day. The purpose of these tests was to investigate if the process of cold restarts imposed any additional plugging risks on the system. Particular attention was paid to the differences in initial growth rate and the maximum resistance to flow as a function of shear rate. The dependence of φ transition on Reynolds number observed in these experiments was quantitatively inconsistent with that reported in the literature for flow-loops or larger autoclaves. This indicated that additional studies into the nature of φ transition were required to probe a range of length scales with higher resolution. Accordingly, Chapter 4 describes the use of a 4 stainless-steel autoclave equipped with two state-of-the-art particle size analysers, Focused Beam Reflectance Measurements (FBRM), and Particle Video Microscopy (PVM), to measure particle number concentrations and size distributions in the water phase and, consequently, the bed formation process on a microscale level. Through measurements of the number of hydrate particles (chord counts) and mean hydrate particle size (chord length) of solid particles present in the liquid phase, the FBRM provided information about the distribution of hydrate crystals (both size distribution and spatial distribution) and its evolution as a function of hydrate volume fraction. In a few experiments, the FBRM probe was used together with a PVM probe, providing visual information about the size and morphology of the hydrate crystals and aggregates. The effect of shear was again investigated and correlations with the resistance to flow explored. To the author s knowledge of the published literature, these experiments were the first time that an FBRM has been used to study the evolution of hydrate particle distributions (size and number density) and the onset of hydrate bed formation 2

20 (φ transition ) in high-pressure methane-water systems and, when combined with a simple model for the mean distance between hydrate particles, the resulting data elucidate the hydrate volume fraction at which particle interaction starts and the process by the moving bed forms. In Chapter 5, the next stage in the study of hydrate plug formation in systems representative of late field life was considered through the use of the original visual autoclave to study partiallydispersed water-oil systems, where both a free oil and free water phase are present prior to hydrate nucleation. The apparatus was upgraded to allow measurement of the torque applied by the motor to maintain a constant shear rate rather than monitoring the motor current, which provides a lower resolution measurement of the system s resistance to flow. The improved apparatus was then used to monitor the evolution of the system as a function of shear and hydrate volume fraction to establish the conceptual mechanism for plug formation that is specific to partially-dispersed systems. In Chapter 6 a summary of the conclusions derived from the research conducted as part of this thesis is presented and suggestions for future work are also detailed. Finally, in Appendix A, a demonstration of a low-field Nuclear Magnetic Resonance (NMR) approach capable of monitoring the evolution of hydrate particles (shells) is presented. The technique s development is based on the fact that NMR is frequently used as a non-invasive technique for characterization of emulsions as hydrate precursors in oil pipelines. In this work, micron-sized ice particles were used to form water-in-cyclopentane emulsions. Two independent NMR techniques including signal relaxation and restricted diffusion measurements were then used to probe the kinetics and shell thickness of cyclopentane hydrate around the water droplets, through monitoring the size of the unconverted water core over several weeks. With further refinements, the technique could be applied to monitor hydrate particle growth and morphology in high pressure systems, and provide complementary data to FBRM measurements. 1.2 Publications arising from this thesis Five publications have arisen from the research conducted as part of this thesis, four of which are already published: 1. Akhfash M., Boxall J.A., Aman Z.A., Johns, M.L., May E.F., Hydrate Formation and Particle Distributions in Gas Water Systems. Chemical Engineering Science 104, This manuscript appears as Chapter 2. I was responsible for conducting the experiments, data analysis and writing the Experimental and Results & Discussion parts of the manuscript. Dr. John Boxall contributed to the design and construction of the high-pressure autoclave apparatus. Associate Professor Zachary Aman contributed the use of the mass-transfer model to quantify the 3

21 hydrate formation rate in gas-water systems, and helped write the manuscript s introduction. Professors Eric May and Michael Johns provided project guidance and edited the manuscript. All of the co-authors contributed to the research motivation, experimental design, data analysis and review of the results. 2. Aman Z.A., Akhfash M., Johns M.L., May E.F., Methane Hydrate Bed Formation in a Visual Autoclave: Cold Restart and Reynolds Number Dependence. Journal of Chemical and Engineering Data 60, This manuscript appears as Chapter 3. I was responsible for the experimental design, conducting the experiments, data analysis and preparing the final graphs and tables for the manuscript. Associate Professor Zachary Aman contributed the use of the mass-transfer model to quantify the hydrate formation rate in gas-water systems and helped write the manuscript. Professors Eric May and Michael Johns provided project guidance and edited the manuscript. All of the co-authors contributed to the research motivation, experimental design, data analysis and review of the results. 3. Akhfash M., Aman Z.M., Du J., Johns M.L., Koh C.A., May E.F. Microscale Detection of Hydrate Blockage Onset in High-Pressure Gas-Water Systems. Published online 12 April 2017 doi: /acs.energyfuels.7b This manuscript appears as Chapter 4. I was responsible for the experimental design, conducting the experiments, data analysis and writing the manuscript. Dr. Jianwei Du helped with maintaining the apparatus and conducting some parts of the experiments. Associate Professor Zachary Aman and Professors Eric May, Carolyn Koh and Michael Johns provided project guidance and edited the manuscript. All of the co-authors contributed to the research motivation, experimental design, data analysis and review of the results. 4. Akhfash M., Aman Z.A., Ahn S.Y., Johns M.L., May E.F., Gas Hydrate Plug Formation in Partially-Dispersed Water Oil Systems. Chemical Engineering Science 140, This manuscript appears as Chapter 5. I was responsible for the experimental design, conducting the experiments, data analysis and writing the Experimental and Results & Discussion parts of the manuscript. Mr Sang Yoon Ahn helped with conducting some parts of the experiments. Associate Professor Zachary Aman contributed the use of the kinetic model to quantify the hydrate formation rate in oil-water systems and helped write the manuscript s Introduction. Professors Eric May, and Michael Johns provided project guidance and edited the manuscript. All of the co-authors contributed to the research motivation, experimental design, data analysis and review of the results. 4

22 5. Haber A., Akhfash M., Loh C.K., Aman Z.A., Fridjonsson E.O., May E.F., Johns M.L., Hydrate Shell Growth Measured Using NMR. Langmuir 31, This manuscript appears as Appendix A. I was responsible for the experimental design and conducted several of the experiments and assisted with the data analysis. Dr. Agnes Haber contributed to the conduct of the experiments, data analysis and writing the manuscript. Mr Charles Loh also with contributed to the measurements. Associate Professor Zachary Aman and Professors Eric May and Michael Johns provided project guidance and edited the manuscript. All of the co-authors contributed to the research motivation, experimental design, data analysis and review of the results. 5

23 2 Hydrate formation and particle distributions in gas-water systems 2.1 Introduction Clathrate hydrates are crystalline inclusion compounds, where molecular cages of water trap light hydrocarbon species, such as methane (Sloan and Koh, 2007). Hydrates are typically stable at high pressures and low temperatures, and play an important role in global energy systems. Naturallyoccurring gas hydrates exist in the region of the seafloor (Kvenvolden, 1994), where thermogenic and biogenic methane sources have generated substantial gas hydrate reserves (Paull et al., 2010). Gas hydrates also form in conventional oil and gas pipelines and pose operational and safety hazards, where high pressures and cool ambient temperatures may lead to rapid hydrate growth and complete blockage of the flowline (Davies et al., 2009). Hydrate prevention is most commonly managed by the injection of thermodynamic inhibitors (THIs), such as monoethylene glycol (MEG), which disrupt the hydrogen-bonded water network and shift the hydrate stability zone to lower temperatures (Koh et al., 2002). However, as the quantity of thermodynamic inhibitor required to fully inhibit hydrate formation scales linearly with the volume of water in the system (water cut), requiring up to $30 per barrel of water produced, the operational cost of THI injection may limit the economics of continued production from older reservoirs (Creek et al., 2011). Over the past two decades, significant progress has been made toward elucidating hydrate plug formation mechanisms. To help simplify the problem of hydrate formation in pipelines, plugging mechanisms may be categorized into oil-, gas- and water-dominated scenarios, based loosely on the most abundant phase (Zerpa et al., 2012). Systems with a dominant gas fraction are the least well-studied and understood; hydrate plug formation may largely be the product of hydrate-wall deposition (through hydrate film growth or deposition from a flowing slurry). Laboratory (Rao et al., 2013) and flow-loop (Nicholas et al., 2009b) experiments suggest hydrate film growth (Lingelem et al., 1994) directly on the pipeline wall is a dominant effect during the early stages of plug formation, with a maximum (initial) deposition rate of 1 mm per day (Rao et al., 2013). In oil-dominated systems, proposed four discrete stages of plug formation based on laboratory and field observations: i) water droplet dispersion in liquid hydrocarbon (Boxall et al., 2012); ii) hydrate nucleation and initial shell growth at the water-hydrocarbon interface (Taylor et al., 2007, Walsh et al., 2009); iii) hydrate particle agglomeration and wall deposition (Aman et al., 2011); and iv) plug formation, enhanced by jamming-type phenomena of large aggregates (Guariguata et al., 2012). In high water-cut cases (water-dominated systems), where the water-oil emulsion may invert (Moradpour et al., 2011) or the well is producing little-to-no oil, a combined suspension of gas bubbles, hydrate particles and oil droplets flow in an aqueous bulk phase. Hydrate may 6

24 nucleate and grow along the gas-water interface (Sun et al., 2007, Sakemoto et al., 2010), where continued growth is supported by dissolved methane in water. Aman et al. (2012) reported hydrate interparticle forces to be three to four times smaller in the water phase than in oil, primarily due to a change in cohesive mechanism (particles do not cohere through an aqueous liquid capillary bridge when suspended in the water phase). Consequently, the flowing hydrate particles are unlikely to aggregate unless the shear field is eliminated, after which the particles may start to cohere and begin sintering/growing together or to the wall (Aman et al., 2011). Joshi et al. (2013b) studied high water-cut systems by measuring pressure drops in a four-inch flow-loop over a variety of liquid loading and velocity conditions, observing three systematic regions in which pressure drop (i) appeared independent of hydrate fraction during the initial phase, (ii) increased directly with hydrate fraction during the second phase, and (iii) fluctuated significantly with hydrate fraction during the final phase. By drawing analogies with observed flow regimes in ice-water mixtures (Kauffeld et al., 2005), Joshi et al. (2013b) proposed that these regions corresponded to three distribution states of hydrate particles, as shown schematically in Figure 1 (Sum et al., 2012): i) homogeneous particle distribution; ii) heterogeneous particle distribution (i.e., variable particle concentration in the radial direction), leading to a moving bedtype accumulation (Hernandez, 2006); and iii) stationary bed formation and catastrophic deposition resulting in plug formation. Figure 1. Conceptual hydrate plugging mechanism for high water content systems, adapted from Joshi et al. (2013b). Joshi et al. (2013b) observed that the transition from the first region (homogeneous distribution) to the second region (heterogeneous distribution) occurred at a repeatable hydrate volume fraction, denoted φ transition that increased with the Reynolds number of the mixed flow. Furthermore, the transition was observed to be irreversible with respect to hydrate plugging behaviour; it was not possible to revert from region 2 back to region 1 by increasing the velocity of the flow. 7

25 In this paper, we report direct observations of hydrate formation and plug formation, and the associated distribution of hydrate particles between the gas and water phases, using a highpressure visual autoclave cell. In this study, the term hydrate particle distribution refers to the spatial distribution of the hydrate particle number density, principally along the autoclave s vertical axis. The distribution of hydrate particle size (including aggregate diameters) could be considered in future work, once the necessary improvements to the video capture system s resolution are made. Measurements of either type of particle distribution are typically very challenging in flow-loop experiments, often requiring flow to be halted prior during periods of visual observation. Joshi et al. (2013b) presented images of the homogenous hydrate distributions that occur prior to the transition. In this work we used a visual autoclave, with a geometry, shear field and other experimental systematics entirely different to those of a flow-loop, to obtain information about hydrate formation in gas-water systems from three independent sources: visual observation, hydrate formation rate (calculated from pressure decrease), and resistance to flow. The latter property, which is inferred from the observed pressure drop in flow-loop experiments, was measured here by monitoring the motor current required to maintain the visual autoclave s impeller at a constant rotational speed. These new results from bench-top scale experiments provide direct confirmation of the conceptual picture proposed by Joshi (2012) for high water content systems based on flow-loop scale experiments; as the hydrate fraction increases, a clear transition occurs between (i) a regime in which the particles are homogeneously distributed and the resistance to flow is insensitive to the amount of hydrate present, and (ii) a regime in which the particles are distributed heterogeneously and the resistance to flow increases directly with hydrate volume fraction. In addition, by comparing the observed hydrate formation rates with the predictions of a simple model containing no adjustable parameters, further insight was obtained into the mechanisms governing hydrate growth in high water content systems. We present hypotheses to explain the differences between the observed formation rates and those calculated using a mass transportlimited model of hydrate growth in gas-water systems (Joshi et al., 2013b). The final portion of this work explores the effect of under-inhibition on plug formation behaviour in the visual autoclave (where the aqueous phase contains a THI such as MEG, but not enough for full hydrate inhibition). 2.2 Experimental Method Apparatus A simplified process diagram of the high-pressure visual autoclave (HPVA) apparatus is shown in Figure 2. A LabVIEW data acquisition (DAQ) system was used to capture cell pressure, cell and 8

26 bath temperature, and the motor current required to maintain constant mixing velocity, at a fixed rate (set at one point every 30 seconds). A visual record of the contents of the HPVA as a function of time was captured using a digital video camera operating at 30 frames per second, mounted outside the bath. The duration of the experiments was typically 24 hours, with hydrate nucleation and significant growth occurring over a period of about 14 hours, starting about 8-10 hours after beginning the experiment. Figure 2. Simplified schematic diagram (left) and picture (right) of the High Pressure Visual Autoclave (HPVA), adapted from Boxall and May (2011). The HPVA consisted of a sapphire autoclave cell with a high-pressure, magnetically-coupled mixing impeller system (Dyna/Mag model #MM-T06). The autoclave was a DB Robinson type sapphire cell with 25.4 mm internal diameter, 150 mm height and 6.4 mm thickness (210 bara maximum pressure). Cell contents were mixed by a four-blade, vane-and-baffle geometry impeller, which was found to generate adequate mixing. The sapphire cell included fluid injection ports at both the top and bottom (axially), although in the present experiments the top port was used only to load the initial charge of gas into the cell (the bottom port was not used). The Dyna/Mag coupling, which had a maximum operating pressure of 210 bara, was incorporated to a Groschopp direct-drive DC motor (model #PM6015, 180V, <0.4A, 50W), equipped with a variable speed drive with a maximum of 1750 revolutions per minute (RPM). The motor speed was monitored with a resolution of ± 17.5 RPM using a DTA-108 Dynapar digital tachometer (Max Jr. Tach 1). The current required to maintain the motor speed at its set point, which was held constant over the duration of an experiment, was monitored using an Agilent digital multimeter (model #34401), with a relative uncertainty of ± 0.05 %. Changes in the measured motor current are correlated with the rheological properties of the mixture within the autoclave (Huo et al., 2001), although a quantitative mapping of current to either motor torque or effective viscosity has 9

27 not been established. We report herein the relative motor current, defined as the instantaneous motor current divided by steady state motor current measured before hydrate formation, as a measure of the relative increase in resistance to flow of the overall mixture. The sapphire cell was attached to a gas supply through a high-pressure manifold containing a pneumatic pressure control valve and several needle valves. Although the present experiments were conducted only in constant volume mode, the HPVA may be operated in constant pressure mode by controlling the pneumatic valve automatically using a digital control algorithm implemented in the DAQ software. The temperatures of the HPVA and pressure manifold were controlled by immersing them in a (transparent) glycol/water bath, which included a submersible pump to circulate the water/glycol mixture. Heat was continuously removed at a constant rate from the bath with a ThermoFisher immersion cooler (model IP-40 NC, 3-inch coil with a temperature range of -35 to 40 C), while an 1100 Watt electrical cartridge heater was immersed in the bath to provide intermittent heating. A proportional-integral control algorithm implemented in the DAQ software was used to maintain the bath temperature at its set point, by controlling the duration of the intermittent heating supplied by the cartridge heater. The HPVA was equipped with several pressure and temperature transducers with approximate locations as indicated in Figure 2. A pressure transducer (Omegadyne, strain-gauge on diaphragm) was connected to the top of the cell, with an uncertainty of ± bar and a full scale of 172 barg. A second pressure transducer (Omegadyne, barg, ± 0.1 bar) was used to monitor the gas reservoir and manifold outside the cell. Three 100 platinum resistance thermometers (PRTs) were used to monitor the temperatures of the cell and bath with an uncertainty of ± 0.2 K. One of the PRTs was inserted into the sapphire autoclave through the cell bottom while the other two PRTs monitored the bath temperature immediately outside the cell and at a point further from it. At steady state, the differences between the readings of the three PRTs were less than the stated uncertainty. However, only for those experiments discussed in this chapter (reported in Table 1), we report the temperature of the PRT in the bath closest to, but outside, the cell, as the reading of the PRT inside the cell exhibited more noise during the transient phases of the experiments. The issue within the PRT inside the cell was then resolved for the rest of experimental results presented in this thesis. Furthermore, this assumption was unlikely to have affected the results described in this chapter given that the size of the sapphire cell (95 ml) was small in comparison with the size of the bath (120 L), minimising any limitation on heat removal from the cell, particularly when heat was released during hydrate formation. This was confirmed by comparing the measured cell and bath temperatures during hydrate growth for those experiments repeated later when the cell temperature probe was fixed. 10

28 Cell pressure and temperature probes were calibrated with a reference pressure transducer (Paroscientific) and a standard platinum resistance thermometer calibrated to ITS-90, respectively. The exact volume of the apparatus including the tubing involved in the measurements (95.5 ml in total) was calculated by injecting a certain, known number of moles of gas (through an ISCO syringe pump) into the cell, while monitoring the cell pressure and temperature directly Procedure The following procedure was used to measure hydrate formation in water-gas systems using the HPVA apparatus: 1. The cell and casing were thoroughly cleaned with methanol and rinsed with deionized (DI) water. The cell was then dried completely with compressed air. 2. The cell was filled with a constant volume of DI water (15-20 vol % of the cell) and (in select experiments) monoethylene glycol, with a minimum volume (approx. 15 ml) maintained to completely cover the impeller blades. 3. The cell was flushed with ultra-high purity methane gas ( %), by pressurising it to approximately 20 bara and then slowly venting to 1 bara. This process was repeated three times to remove any residual air from the system. 4. The cell was then pressurized to the set-point pressure (e.g., 60 bara), and was sealed for minutes to test for leaks in the high-pressure tubing and equipment. 5. The mixing was initiated by setting the motor to a pre-determined, constant rotational speed. 6. In all experiments, the initial bath temperature was in the range (18 to 22) C. The bath was cooled to (1 to 3) C, by decreasing the set point temperature at a rate of 1 C/hr. The actual cooling rate was limited in most experiments to between (0.3 to 0.5) C/hr, as the immersion cooler could not remove the heat at a sufficient rate (ambient temperatures exceeded 39 C at some points). 7. In all experiments, hydrate formation was first identified from a sharp decrease in cell pressure. After reaching the final set-point, the bath temperature was maintained for the remainder of the experiment (10-20 hours). The cell pressure, bath temperature, and motor current were continuously recorded at 30-second intervals, which was found to be sufficient to constantly capture changes in the system given that duration of the experiments was typically 24 hours. 8. At the end of the hydrate formation process, the system was heated to the initial bath temperature (e.g. 22 C) to dissociate hydrate and to check for a mass balance closure. 11

29 Hydrate formation was identified by a sharp decrease in cell pressure, and was confirmed visually in each experiment. The pressure, temperature and motor current data reported herein are relative to the time of hydrate formation. Experimental hydrate volume fractions were calculated from the measured pressure drop using compressibility factors for methane obtained from the Setzmann and Wagner (1991) equation of state, as implemented in the software REFPROP 9.0 (Lemmon et al., 2013) and using densities for the water and hydrate phases calculated using the software Multiflash 4.2 with the CPA model set (Infochem, 2012). The hydrate volume fractions reported hereafter were calculated based on the total volume of the hydrate V H, water V W, and MEG V MEG (where applicable) phases: i.e. = V H / (V H + V W + V MEG ). The specific conditions (pressure, temperature, impeller speed, and MEG concentration) of each experiment conducted are summarized in Table 1. Table 1 also contains an estimate of the mixture s initial Reynolds number, which was calculated from a standard definition for stirred tanks (Naumann, 2008): Re = NρD μ 2 (1) where N is the impeller rotational speed (s -1 ), D is the impeller diameter (m), and ρ and μ are, respectively, the density (kg/m 3 ) and viscosity (kg/m/s) of the aqueous phase mixture (affected by MEG concentration for experiments 10-13). Upon the formation of the hydrate slurry, the effective density and viscosity of the aqueous phase would vary with hydrate volume fraction, making estimation of the system s Reynolds number more difficult. The initial Reynolds number was calculated for the purposes of (i) estimating the extent of turbulence in the system, which would influence the efficiency of mass transfer, and (ii) enabling a more general interpretation of the results through the removal of apparatus-specific parameters. Sloan et al. (2011) recommend that, for a particular type of autoclave with a 300 ml volume and an impeller of diameter m, hydrate inhibitor testing should always be conducted at a rotational speed above 600 RPM. In stirred reactors, the flow is considered fully turbulent (Naumann, 2008) when Re>1000 and as shown in Table 1, this constraint was initially satisfied in all but experiments 8 and 9. Similarly, while it is difficult to calculate precisely the Reynolds numbers in the slightly larger autoclaves discussed by Sloan, Koh and Sum (Sloan et al., 2011), the recommendation to operate above 600RPM to avoid mass transfer limitations appears to be roughly consistent with turbulence threshold for stirred reactors and the results presented herein. 12

30 Table 1. Summary of the experiments conducted. Exp. P initial bara T initial C RPM Initial Reynolds Number Initial MEG Concentration 0 wt% wt% 2.3 Modelling Approach A simple model of hydrate formation rate was implemented to facilitate analysis of the experimental data. The objective of the modeling was to provide quantitative estimates of the hydrate formation rate, and to test whether established mass transfer limitations provided a reasonable description of the experimental HPVA results. Given this objective, and to ensure the modeling remained commensurately tractable, simplifying assumptions were made in the model s development as described below. Importantly, no adjustable parameters were introduced into the model to improve agreement between its predictions and the experimental observations. Skovborg and Rasmussen (1994) presented a model for hydrate growth rate that considers two kinetic contributions: the intrinsic reaction rate of hydrate formation for a given supersaturation (Kashchiev and Firoozabadi, 2002) and the rate of CH 4 transport from the gas phase into the aqueous phase, which determines the supersaturation in the vicinity of the hydrate-water interface. The supersaturation is given by the difference between the (time-dependent) concentration of CH 4 dissolved in the water, C (water) CH4 () t, and the concentration of dissolved CH 4 when the water is at equilibrium with the hydrate phase, * (wat-hyd) C CH4. The asterisk here denotes an equilibrium value, which is set by the system pressure, temperature and MEG concentration. At the shear rates achievable in the HPVA, the intrinsic reaction rate of hydrate formation (Vysniauskas and Bishnoi, 13

31 1983, 1985) is more than two orders of magnitude faster than the rate of CH 4 transport through the aqueous phase, and thus the model of Skovborg and Rasmussen (1994) may be simplified to: m hydrate k * (water) * (wat-hyd) MX AG / W CCH4 CCH4 t (2) where Δm hydrate is the mass of hydrate formed during time step Δt, k MX is the mass transfer coefficient for CH 4 from the gas phase to the aqueous phase (Aman et al., 2012) and A G/W is the total interfacial area between the gas and water phases. In deriving eq. (2), it was assumed that the rate of CH 4 transport through the aqueous phase is always sufficient to keep (water) CH4 () t at its equilibrium value (i.e., a constant supersaturation in the absence of hydrate) and that the product k MX A G/W is much smaller than the product of the intrinsic hydrate reaction rate with the area of the water-hydrate interface. To determine k MX and A G/W in eq. (2), average values of the hydrodynamic properties within the cell were estimated using empirical correlations. The mass transfer coefficient was estimated using a correlation by Hanratty (1991) as the product of the density and square of liquid phase velocity on the Schmidt number. The water-gas interfacial area was estimated as the sum of (i) the flat bulk interface and (ii) the area of entrained gas bubbles. The diameter of the gas bubbles was calculated by assuming a monodisperse distribution and using the correlation by Hesketh et al. (1987) which accounts for changes in fluid and gas density, viscosity and interfacial tension and has been validated experimentally over a bubble size range of 100 μm to 1 cm (Hesketh et al., 1987). C d D 1.38 We bubble crit c d c vc (3) where d bubble is the bubble diameter, σ is the water-gas surface tension, ρ c and ρ d are, respectively, the densities of the continuous and dispersed phases, μ c is the viscosity of the continuous phase and v c is the superficial velocity of the continuous phase (Hesketh et al., 1987). In this work, the critical Weber number, We crit, was estimated as 1.1 (Holmes, 1973). The total number of bubbles was calculated with the correlation by Gregory et al. (1978) for gas entrainment in a water slug under similar shear conditions, with the assumption that bubbles are entrained in the first 5% of the liquid phase from the gas-water interface. The values of the fluid velocity needed for eq. (3) were calculated by discretizing the radial dimension of the cell into 10 rectangles of equal width, each of which would sweep out a hollow concentric cylinder upon a volume of revolution. The fluid velocity was assumed to be distributed linearly along the cell s radius with a slope given by the impeller s angular velocity. 14

32 Four pieces of information from each experiment were used as inputs to the simplified model: i) initial liquid content; ii) initial gas pressure; iii) autoclave mixing speed and impeller diameter; and iv) the bath temperature as a function of time. Within the model, the amount of CH 4 consumed for each time step was calculated from m hydrate (via eq. (2)) by using a hydration number (i.e. number of water molecules per gas molecules) of 5.75 for structure I methane hydrate (Sloan and Koh, 2007). We note that other numbers also reported for methane hydration number, e.g (Infochem, 2012). However, such a change does not make a significant difference to amount of hydrate calculated in this work and, furthermore, is within the experimental uncertainty of the measurements presented. The corresponding pressure drop was then calculated using an iterative solution of the SRK equation of state (Soave, 1972). (The Setzmann and Wagner (1991) reference equation of state for pure methane was only used to convert experimental pressure drops to hydrate volumes; it was not used within the model because it is far more non-linear than a cubic equation of state and its iterative solution requires more computational time.) The bath temperature and calculated cell pressure were used in each time step to determine (i) hydrate equilibrium conditions and (ii) equilibrium solubility of methane in water with and without hydrate, respectively * (wat-hyd) C CH4 and * (water) C CH4, using Multiflash 4.2 with the RKS-advanced model set (Infochem, 2012). The total mass of hydrate formed in the cell per time step was calculated by summing the masses determined by applying eq. (2) to each of the 10 concentric cylinder volumes. Hydrate aggregation and slurry viscosification models were not considered for the present system, based on the low hydrate cohesive forces reported by Aman et al. (2012) for hydrates suspended in a bulk water phase. This simplified hydrate formation model does not attempt to describe the complex hydrodynamic nature of the flow within the autoclave. Rather, the purpose of the model is to enable an estimate of the hydrate formation rate without resorting to treating either the mass transfer coefficient or the gas-water interfacial area as adjustable parameters. 2.4 Results and Discussion Hydrate formation, particle distributions and flow resistance Figure 3 shows images of the hydrate formation in the HPVA cell at various points during experiment 1, demonstrating how the evolution of the hydrate particle distributions may be tracked visually from the recorded video footage. After initial hydrate formation, four sequential visual states were identifiable in most experiments: an initial homogeneous distribution of particles (Figure 3-c), identified by a uniform opacity of the cell; a heterogeneous distribution of particles (Figure 3-d), where particles begin concentrating in some sections of the cell; an apparent bedding of the hydrate particles atop the gas-liquid interface (Figure 3-e), with some stratification also 15

33 apparent below the interface; and a complete plug (Figure 3-f). These states are consistent with the conceptual picture presented by Joshi et al. (2013b). Identifying the time in the video record at which the transition from an image like that in Figure 3-c to that in Figure 3-d had occurred unambiguously, and correlating it with the volume of hydrate formed at that time based on the measured pressure drop, provided a visual measure 1 of the hydrate volume %, denoted (visual) φ transition, at which the transition from homogeneous to heterogeneous particle distribution occurred. In experiment 1, (visual) φ transition = 13.6 vol % hydrate, which is consistent with the observation of Hadsbjerg et al. (2011) who reported no plugging behaviour for less than 15 vol % hydrate in a gas-water system. a b c d e f Figure 3. Images of different hydrate formation stages in the HPVA for experiment 1. (a) Before hydrate formation; (b) 14 minutes after nucleation, at 1.4 vol % hydrate; (c) 72 minutes after nucleation, at 9.5 vol % hydrate; (d) 101 minutes after nucleation, at 19.6 vol % hydrate; (e) 105 minutes after nucleation, at 22 vol % hydrate; (f) 200 minutes after nucleation, at 53 vol % hydrate. 1 No attempt was made to estimate hydrate volume fraction from the number and average size of the hydrate particles in the recorded image because the video capture system s resolution was not adequate to do so accurately. 16

34 Figure 4 shows the measured system pressure and temperature as a function of time for experiment 2, together with the pressure calculated using the hydrate formation model. The hydrate equilibrium temperature predicted from the measured pressure using Multiflash 4.2 is also shown (red curve). Nucleation in experiment 2 occurred around 7 C and 60.6 bara, and as gas consumption caused the pressure to decrease, the corresponding hydrate equilibrium temperature also decreased. The difference between the equilibrium and bath temperatures (subcooling) was maintained at approximately 1.4 C through the majority of the formation process. The bath temperature (considered representative of the cell temperature in this particular measurement as discussed earlier) was consistently below the hydrate equilibrium temperature, indicating hydrate formation was not limited by heat removal from the cell. The bath temperature increase at approximately 700 minutes was caused by an increase in the laboratory air temperature (air conditioning shutdown). Figure 4. Cell pressure (left ordinate) and bath temperature (right ordinate) after initial hydrate formation in experiment 2, where experimental data (black) are compared with the pressure predicted using the mass-transfer limited formation model (grey) (eq. (2)). The hydrate equilibrium temperature (red) was calculated from the measured pressure and is always greater than the bath temperature indicating the experiment was not heat-transfer limited. The experiment s initial Reynolds number was (For interpretation of this color figure, refer to the web version of this thesis.) Table 2 contains a summary of the key results produced from the measured data and the hydrate formation model for the experiments listed in Table 1 containing no MEG in the water phase. All 17

35 five repeat trials at Re = 2460 (i.e., experiments 1-5 in Table 1) showed similar behaviour, where the final pressure, temperature and hydrate volume % (φ max ) are provided in Table 2. Experiments 2-5 formed vol % hydrate far beyond the liquid-phase hydrate packing limit (Camargo and Palermo, 2002, Sinquin et al., 2004) of 57 vol % by forcing hydrate deposits up the autoclave s walls into the gas phase region of the cell. Experiment 1 formed a hydrate plug that halted the impeller system, thus preventing hydrate growth beyond 53.7 vol % hydrate. The average and maximum relative differences between the predicted and measured pressures for all the experiments summarized in Table 2 were 9.7 and 19.8%, respectively. Table 2. Summary of results for experiments with no MEG in the aqueous phase. The equilibrium temperature corresponding to the pressure at which nucleation occurred had an average value of (8.4 ± 0.1) C for these nine experiments. The equilibrium temperature corresponding to the final system pressure, T eq,final, was always above the final system temperature, indicating that the hydrate formation was not heat transfer limited. The duration of the experiments was typically 24 hours, with hydrate nucleation and significant growth occurring over a period of about 14 hours, starting about 8-10 hours after beginning the experiment. Exp. Induction time hr T nuc C P nuc bara T final C T eq,final C P final bara (model) P final bara (exp) φ max vol% (model) φ max vol% Experiments 1-5 were performed with the same initial conditions of pressure, temperature and shear rate. While the expected stochastic variation in the induction time and nucleation conditions for these five experiments was observed, the formation behaviour following nucleation was similar in all five experiments; the similarities were particularly evident when the measured data were compared with the model predictions. Figure 4 shows a comparison between the experimental (black) and model (grey) pressure curves for experiment 2, where four distinct regions (labelled A- D) are apparent. All four regions were observed in each of the first five experiments. Each region (A-D) was characterized by a different hydrate formation rate (i.e., gas consumption rate) that may correspond to different mass transfer limitations encountered throughout the hydrate formation 18

36 process; a detailed discussion of these regions is provided in Section The pressure curve calculated with the mass transfer model of eq. (2) provides a reference rate of mass transfer based on the assumption that it is limited by methane transport through the aqueous phase and/or the area of the gas-liquid interface. Whilst the model contains several simplifying assumptions, the ability to make quantitative predictions of the formation rate allows discrepancies between the observed and predicted rates to be highlighted and analysed in further detail. This comparison may therefore shed light on the significance of additional mass transport processes. Further information about the mechanisms leading to the formation of a hydrate plug is available from the motor current data, which provide a measure of the mixture s resistance to flow. Figure 5 shows the hydrate volume % determined from the experimental pressure (black) and the model s predictions (grey) together with the relative motor current (normalized by the motor current value recorded prior to hydrate) as a function of time for experiment 2. The four shaded regions A-D identified exclusively from the pressure data in Figure 4 are also shown for comparison. The steady-state error between experimental (black) and predicted (grey) hydrate volume % curves in Figure 5 is 8.9 %, which reflects the difference in steady-state pressures shown in Figure 4. The motor current data in Figure 5 show little-to-no increase in regions A or B, followed by a mild increase in current (10 % max.) in region C and a substantial increase in relative motor current (70 % max.) in region D. Figure 5. Hydrate volume % (left ordinate) and relative motor current (right ordinate) after initial hydrate formation in experiment 2, where experimental data (black) is compared with a masstransfer limited formation model (grey). The experiment s initial Reynolds number was

37 The first increase in motor current (or resistance to flow) should correspond to the onset of a heterogeneous particle distribution, analogous to the pressure drop signature observed in flow-loop studies (Joshi et al., 2013b). The hydrate volume % at which the first significant increase in the experimental motor current occurs (labelled φ (current) transition measurement determined from the video data ( φ ). (visual) transition ) is compared in Table 3 to the independent Table 3. Three independent measures of the hydrate volume %, φ transition, at which the transition from a homogeneous to heterogeneous particle distribution occurred, identified from the motor current, pressure decrease, and visual observations. The reader is referred to Table 1 and 2 for other parameters relevant to each experiment number. Exp. (current) φ transition vol% (visual) φ transition vol% (pressure) φ transition vol% NR NR NR = no video imagery was recorded for the experiment; visual and pressure data did not yield conclusive transition volume % in 50 and 800 RPM experiments. Figure 6 shows the relative motor current as a function of hydrate volume % for experiments 1-3, providing an indication of the repeatability of the results. It should be kept in mind that the boundaries between each region (A-D) were identified exclusively from gas consumption data over these experiments (e.g., Figure 4). In each region, however, the motor current data from the three experiments clearly exhibit similar features: no increase from the pre-hydrate value in regions A and B, followed by a rise upon, or just after, the transition into region C, followed by substantial fluctuations in region D. Comparison of these substantial motor current fluctuations with the flow-loop pressure drop behavior reported by Joshi et al. (2013b) at high hydrate volume fractions may indicate a system with severe deposition and jamming-type behaviour. This is also consistent with the visual data record (e.g., Figure 3-f). 20

38 Figure 6. Relative motor current as a function of hydrate volume % in experiments 1 (grey), 2 (black) and 3 (red). All experiments were performed at initial Reynolds numbers of approximately (For interpretation of this color figure, refer to the web version of this thesis.) The correlation between the motor current data and the regions A-D indicates that a third independent metric for the transition between homogeneous and heterogeneous particle distributions, (pressure) φ transition, is potentially available from the measured pressure data. This pressure transition was observed for experiments 1-5 and (all 400 RPM), but was not observed in experiments 6-9 (at 50 or 800 RPM) for reasons discussed below. Table 3 lists the values of transition determined by these three measures for each of the experiments listed in Table 1 and shows that for a given experiment these three independent measures are in good agreement with each other. For example, the averages for experiments 1 to 3 are (pressure) φ transition = (13.9 ± 2.7) vol %, and (current) φ transition = (14.4 ± 3.4) vol %, (visual) φ transition = 13.6 vol %, where the uncertainty bounds denote the standard deviation of the three trials (unfortunately the video capture failed for experiments 2 and 3). In experiments 4 and 5, which were also conducted at 400 RPM and two months after experiments 1 3, the three independent measures were again consistent with each other, with (current) φ transition =(24.7 ± 1.6) vol %, (pressure) φ transition = (23.6 ± 1.1) vol %, and (visual) φ transition = (26.3 ± 1.3) vol %, where the uncertainty bounds represent (half) the difference in the values measured in the two trials. The measured transition points in experiments 4-5 were approximately 11 vol % larger than observed in experiments 1 to 3. The addition of MEG in experiments shifted the average value of φ transition by an even larger amount; furthermore, in those experiments only (current) φ transition and (pressure) φ transition were in good agreement, with the values inferred from the visual record being 6 to 9 vol % lower. 21

39 Joshi et al. (2013b) observed that φ transition depended on velocity, and proposed a linear correlation with Reynolds number based on flow-loop data. This correlation predicts φ transition to be essentially zero for all of the Reynolds number achieved in experiments 1-13, suggesting a heterogeneous distribution should be observed immediately upon hydrate formation. Our measurements at different shear rates also indicated a dependence of φ transition on Reynolds number, but the relation is clearly different from that reported from flow-loop observations (Joshi et al., 2013b). This comparison indicates that the Reynolds number-dependence of φ transition may also depend significantly on the system geometry. The variations in φ transition with shear and other experimental conditions, as well as the repeatability of the various means of measuring it, are important elements of future work currently underway Variations in mass transport Comparison between the measured hydrate volume fractions and those predicted using the formation model allows hypotheses on reasons for the differences to be formulated. The formation model discussed in Section 2.3 relies on the assumption that the water phase is fully saturated with methane. As heat transfer limitations were not encountered in these experiments, cases where the model over-predicts experimental hydrate formation rates may indicate conditions under which the assumption of complete methane saturation in water is not valid. In region A of Figure 4 and Figure 5, the agreement between the experiment and model is excellent, given that the model is entirely predictive; for experiment 2, the initial measured and predicted hydrate formation rates were 26.9 and 27.4 mg/min, respectively. In this thesis, the initial hydrate growth rate determined from the rate of gas consumption over the either first 10 minutes after nucleation or the first 2 vol % of hydrate formed, whichever was greater. The agreement between model and experiment in region A suggests that the estimated values of each of the quantities on the right hand side of eq. (2) are reasonable. In particular, as the water was mixed vigorously for many hours prior to nucleation, the assumption that concentration of CH 4 in the aqueous phase is at its equilibrium value ( * (water) C CH4 ) is appropriate and the area of the gas-water interface estimated using eq (3) is reasonable. Hence, the model s founding assumption that the hydrate formation rate is limited by the diffusion of methane through the (saturated) aqueous phase to the growing hydrate particle seems to be valid in Region A. In region B, however, the observed formation rate decreases below that predicted by the model. This disagreement suggests that one or more of the assumptions underlying eq. (2) is no longer valid, with a likely candidate being the concentration of methane dissolved in the water phase. The intrinsic reaction rate of hydrate formation is significantly faster than the transport of CH 4 through the aqueous phase; comparison of the value of k MX in eq. (2) with the models of Vysniauskas and Bishnoi (1983;1985) and Englezos et al. (1987) shows that kinetically-controlled hydrate 22

40 formation rates are at least a factor of about 500 larger than CH 4 transport rates at the liquid velocities achieved in autoclave or flow-loop experiments. This is consistent with the observations by Boxall et al. (2009), who required a fitting parameter of to describe kinetic hydrate formation in an oil-dominated flow-loop. Thus, if A G/L is not sufficiently large, the growth of the hydrate particle will lead to a depletion of the aqueous phase s CH 4 concentration in the vicinity of the hydrate-water interface. We hypothesize that in region B, the observed growth rate dropped below that predicted by the simple model because C () t C (water) CH4 * (water) CH4 during the period of time corresponding to this region. The time at which the transition from regions A to B occurs is set by the magnitude of A G/L, which in turn depends upon the degree of shear and entrainment of gas phase bubbles. The observed transition in hydrate particle distributions suggests a mechanism by which the hydrate grow rate accelerates in region C. At φ transition, buoyant hydrate particles may begin collecting at the methane-water interface (as per the conceptual Figure 1), disrupting the interface and generating additional interfacial area. The value of A G/W used in eq. (2) would then become a significant under-estimate of the physical interfacial area available because it was calculated from shear rate alone using eq.(3). In region C, the observed hydrate formation rate increases significantly such that the amount of hydrate present in the system recovers quickly from the reduced growth rate experienced in region B, and then overtakes the amount predicted using the model. This second hypothesis implies that, after the formation of a moving hydrate bed, additional formation would result in a positive feedback cycle between the volume fraction of hydrate present and the amount of gas-water interfacial area generated during flow. In region D, the observed growth rate returns to a value similar to or slightly below that predicted by the model and the hydrate volume fraction begins approaching the theoretical limit for particle packing (Camargo and Palermo, 2002, Sinquin et al., 2004). That is, hydrate formed in the aqueous phase must be displaced into the region occupied by the gas phase to allow continued growth in the aqueous slurry. The image in Figure 3-e, captured whilst the system was in region C at approximately 22 vol % hydrate, provides an illustration of the onset of hydrate displacement into the gas phase. Figure 3-f shows a nearly-plugged hydrate system in region D, with substantial wall deposition and a single demarcation to indicate the former liquid level. The slower growth rate observed during region D may be the result of stratification, where a dry (or water-poor) hydrate slurry collects near the gas phase and a wet (or water-rich) hydrate slurry collects near the base of the cell. This hypothesis represents an inversion of the fluidized and stationary hydrate beds picture presented by Hernandez (2006). That is, the dry hydrate bed may effectively limit the ability of methane to reach unreacted water lower in the cell. The formation of this bed is consistent with the significant motor current fluctuations observed only in region D, as 23

41 the particles may increase frictional energy losses in the liquid phase, along the gas-liquid interface and along the steel (impeller) interface, leading to intermittent mechanical jamming behaviour Effect of turbulence (water) The hypothesis that the reduced mass transfer rate in region B is due to a depletion in C t, CH4 () caused by an insufficient A G/W, implies that the transition from region A to B would vary with shear rate. To test this hypothesis, four additional experiments were performed, with similar initial pressures and temperatures to experiments 1-5 at impeller speeds of 50 and 800 RPM, corresponding to initial Reynolds numbers of 310 and 4930, respectively. The observations and model predictions for these new experiments reflected the changes in velocity, as demonstrated by Figure 7, which shows measured and calculated hydrate volume % for experiments 6 and 8. Figure 7. Measured (black) and predicted (grey) hydrate volume % for experiment 4 (initial Re = 4930) and experiment 6 (initial Re = 310). Experiments 6 and 7 (at initial Re = 4930) showed initial hydrate formation rates of 92.8 mg/min, approximately three and a half times greater than that for experiments 1-5 (initial Re = 2460). Experiments 8 and 9 (at initial Re = 310) showed initial hydrate formation rates of 0.62 mg/min, approximately 3 % of the value in experiments 1-5. At the highest shear rate, the experimental formation rate was remarkably consistent with the model predictions over the experiment s duration, indicating the appropriateness of considering only a single mass transport limitation to hydrate formation under such conditions. In contrast, the low shear experiments effectively 24

42 commence in the equivalent of Figure 5 s region B, after which the hydrate formation rate increased monotonically until hydrate growth stopped when the system plugged (as indicated by the plateau in Figure 7 for experiment 8). In both the low shear experiments 8 and 9, the system plugged and the motor ceased rotation at a hydrate volume % of less than 30 vol %. Plug formation, in turn, may have prevented further hydrate formation by substantially reducing the gas-water surface area. Motor current data for experiments 6-7 and 8-9 are shown in Figure 8 and exhibit excellent repeatability between the trials. The values of (current) φtransition exhibit a variation with shear, ranging from vol % in experiments 8 and 9, to vol % in experiments 6 and 7. The HPVA data in Figure 8 indicate that not only does the value of φ transition depend on the system s Reynolds number but that the sensitivity of the system to the hydrate volume fraction may also decrease with increasing turbulence; at very low shear rates the system plugs just after reaching φ transition, whereas at high shear rates the increase in flow resistance is significantly smaller. This sensitivity may physically correspond to the growth rate of the moving hydrate bed and, consequently, the system s propensity to form a complete hydrate plug. Figure 8. Relative motor current as a function of hydrate volume % at different impeller speeds: experiment 6 (initial Re = 4930, black); experiment 7 (initial Re = 4930, grey); experiment 8 (initial Re = 310, black); and experiment 9 (initial Re = 310, grey). 25

43 2.4.4 Under-inhibited systems Under-inhibited experiments were performed in the presence of 10 wt% monoethylene glycol (MEG), defined with respect to the total weight of the aqueous phase. The initial pressure in experiments was increased relative to the previous experiments, to provide the same initial driving force for hydrate formation in the presence of MEG. Key results, from both the experiments and the corresponding predictions of the formation model, are provided in Table 4 for the under-inhibited experiments. Across all four experiments, the average MEG concentration in water increased from 10.0 wt % to approximately 17.5 wt % (Table 4). Table 4. Summary of results for experiments that started with 10 wt % MEG in the aqueous phase. The final MEG concentration reached in each experiment (based on the final hydrate volume % calculated from the measured decrease in gas pressure) is also listed. The equilibrium temperature corresponding to the pressure at which nucleation occurred had an average value of (7.90 ± 0.02) C for these four experiments. The equilibrium temperature corresponding to the final system pressure, T, was always above the final system temperature, indicating that the hydrate eq,final formation was not heat transfer limited. Exp. Induction time hr T eq,nuc C P nuc bara T final C T eq,final C P final bara (model) P final bara (exp) φ max vol% (model) φ max vol% MEG final wt% Hydrate formation in the experiments with MEG (Figure 9) showed similar features to those observed in experiments without MEG, including (i) the transition of homogeneous to heterogeneous particle distribution, followed by (ii) the formation of a stationary hydrate bed at the gas-water interface. The presence of MEG led to a wet appearance in the hydrate, with droplets and wet particles on the HPVA s walls in the gas phase region being visible throughout the entire experiment, as shown in Figure 9. Figure 10 shows the result of one of the under-inhibited MEG trials at 400 RPM (initial Re = 1140), which formed about 4 vol % less hydrate without significantly altering the initial hydrate growth rate, when compared to the 0 wt % MEG trials at 400 RPM (initial Re = 2460) at a similar sub-cooling. This self-inhibition, which results from the effective increase in MEG concentration in the aqueous phase caused by hydrate formation, is consistent with the observations of others; Hadsbjerg et al. (2011), reported a decrease in achievable hydrate from 45 to 35 vol % with the 26

44 inclusion of 10 % MEG, and Hemmingsen et al. (2008) and Li et al. (2011) observed 100 % conversion for pure water and 70 % conversion in the presence of 10% MEG, for a flow wheel geometry. The average hydrate formation rate in the trials with MEG was 26.5 mg/min, compared to 26.6 mg/min for experiments 1-5. The hydrate formation rates predicted with the simple model presented in Section 3 were significantly larger than those observed and are not shown in Figure 10. The causes of model s poor predictions for the MEG experiments include the substantial * (water) C CH4 * (wat-hyd) C CH4 uncertainty in the requisite values of and as a function of (varying) MEG concentration. Addressing this issue will be the subject of future work as the model is extended to incorporate several of the mechanisms observed in this work. a b c d Figure 9. Images of different hydrate formation stages in the HPVA for experiment 10 (10 % MEG initially): (a) 87 minutes after nucleation, at 15.6 % hydrate volume showing a homogeneous distribution of hydrate particles; (b) 123 minutes after nucleation, at 25% hydrate volume and the onset of (visual) φ transition ; (c) 137 minutes after nucleation, at 28 % hydrate volume shortly after and (d) 312 minutes after nucleation, at 47% hydrate volume near the end of experiment. (visual) φ transition ; 27

45 Figure 10. Methane consumption rate (left ordinate) for 0 % MEG (experiment 2, black curve), 10 wt% initial MEG concentration (experiment 8, grey curve), and MEG concentration for experiment 8 during hydrate formation (right ordinate). The MEG concentration was calculated based on a mass balance for the closed system, assuming no MEG was incorporated in the hydrate phase. Figure 11. Relative motor current as a function of hydrate volume % in experiment 10 (10 wt % initial MEG concentration). This experiment was performed at 1140 Re. The relative motor current recorded in experiment 10 (with MEG), is shown in Figure 11, with the first significant increase in flow resistance occurring at 32.6 vol %. The average value of 28 (current) φ transition = (31.4 ± 1.1) vol %, for experiments 10-13, which is about 17 vol % larger than the average value of experiments 1-3 and 7 vol % larger than the average of experiments 4-5. The first maximum in

46 the motor current recorded for experiment 10 appeared at 38 vol % hydrate and was followed by a decrease to a nearly constant value through to the end of the experiment. Comparison of Figure 6 and Figure 11 suggests that, whereas in experiments 1-5 where the motor current increase was about %, the motor current increases for under-inhibited systems are modest (0-10%) even at high hydrate % (45-50 vol %). The values of φ transition determined from the motor current and pressure data across the four repeat MEG trials (experiments 10-13) were in excellent agreement with the average value of (pressure) φ transition =(31.3 ± 1.6) vol%. The visual determinations of the transition point were more scattered and less consistent with the other two measures with an average value of φ (visual) transition = (24 ± 2.2) vol%. Although further work is required to understand these variations between the different measures of φ transition, these initial experiments suggest that the visual- and pressure-based determinations have the poorest resolutions. The determination by visual means has an element of subjectivity while the determination from the pressure data requires the presence of clear, unambiguous changes in slope. The absence of a quantitative model that produces reasonable predictions of the average expected growth rate can make detecting slope variations in the experimental pressure curve even more difficult. Thus (current) φ transition is likely to be the most reliable of the three measures. Further work is also required to understand the mechanisms behind the observed shift in φ transition for the system with 10 wt % MEG. However, the initial concept may be highly important to high water cut field cases. If flowing production systems are designed with a tolerance for some volume fraction of hydrate, the under-dosage of a thermodynamic inhibitor may extend the range of viable production scenarios. 2.5 Chapter Conclusions A new high-pressure visual autoclave (HPVA) was used to study hydrate formation mechanics, which allowed pressure, temperature and motor current measurements to be combined with direct visual observations of the hydrate particles within the autoclave. Hydrate formation behaviour was studied for methane-water systems, over a range of shear conditions with initial Reynolds numbers ranging from 300 to A transition (at a hydrate volume fraction of φ transition ) from homogeneous to heterogeneous particle distribution, which was previously hypothesized based on analysis of flow-loop pressure drops, was observed directly in experiments with the HPVA at much lower Reynolds numbers. Three independent measures of φ transition were usually available from HPVA experiments by analysis of either the motor current (flow resistance), pressure drop (hydrate growth rate), or direct visual observation of the hydrate particle distributions; for a given experiment these three independent measures of φ transition were generally in good agreement with each other. As observed in flow-loop experiments, the value of φ transition increased with Reynolds 29

47 number, ranging from about 13 % at low shear (initial Re = 310) to about to 24 vol % at high shear rates (initial Re 4960). While the Reynolds number dependence of φ transition in the HPVA was difficult to measure precisely with the current version of the apparatus, the nature of that dependence was clearly different to that observed in flow-loop experiments at much higher shear rates. A simple hydrate formation model for gas-water systems with no adjustable parameters, in which methane diffusion through the aqueous phase to the growing hydrate particle was assumed to be the rate-controlling growth mechanism, was implemented to provide quantitative predictions of hydrate growth rates. At high shear rates, the predicted and observed growth rates were in excellent agreement. At lower shear rates, the nature of the deviations between the predicted and observed growth rates allowed new hypotheses to be formulated regarding two additional mechanisms, which give rise to mass transport variations. Prior to φ transition the growth rate was limited in low shear experiments by the rate of methane re-saturation in the water phase. Following φ transition the observed increase in growth rate was possibly caused by the formation of hydrate particle beds near the gas-liquid interface, which significantly increased the gas-liquid interfacial area. The inclusion of a thermodynamic hydrate inhibitor (10 wt % initial MEG concentration) in the aqueous phase, resulted in three important new conclusions about under-inhibited (and selfinhibiting) systems: i) no change was observed in initial hydrate formation rate; ii) the average transition volume % of hydrate φ transition ) was observed to increase from 18 vol % (experiments 1-5) to 31 vol % (experiments 10-13), at the same impeller speed; and iii) the relative motor current increase for the under-inhibited systems was only 20 % of the magnitude observed in the uninhibited case. These initial studies suggest that under-inhibition should be studied further as a potential management solution for hydrate formation in high water cut systems. Future work is needed to improve the accuracy of φ transition measurements as conditions in the autoclave vary, and investigate the relationship(s) between φ transition and shear rate and/or THI concentration. In addition, the hypotheses developed in this work relating to variations in mass transport limitations as a function of shear rate should be tested and extended to cover various hydrate inhibitor levels and across a broad range of flow conditions. 30

48 3 Cold restart Methane Hydrate Bed Formation in a Visual Autoclave: Cold Restart and Reynolds Number Dependence 3.1 Introduction Recently, Joshi et al. (2013a) proposed a conceptual mechanism (Figure 1) for hydrate plug formation in water-continuous systems, based on several flow-loop experiments performed over a wide range of liquid loading and velocity conditions. In these tests, the pressure drop measured across the flow-loop was observed to increase only after a finite amount of hydrate had formed in the system. This transition in the pressure drop behavior as a function of hydrate volume fraction was interpreted to correspond to the transition from a homogeneous distribution of hydrate particles to a heterogeneous distribution and the formation of a moving hydrate bed (Hernandez, 2006), where hydrate particles would collect at the gas-water interface due primarily to the density difference between the hydrate and external (water) phase. Joshi et al. (2013a) labeled the hydrate fraction at which the transition of the hydrate-in-water particle distribution from homogeneous to heterogeneous as φ transition, which is conceptually described in Figure 1. Initial hydrate formation will lead to a sparse and non-interacting distribution of hydrate particles in the water phase (labeled as a homogeneous dispersion). Continued hydrate growth will increase the hydrate volume fraction toward a critical point at which particle begin interacting, where a sufficient magnitude in particle collisions will enable buoyant collection at the water-gas interface. Akhfash et al. (2013) provided the first direct visual confirmation of the existence of this transition and the formation of a hydrate bed, confirming the hypothesis of Joshi et al. (2013a) through a series of experiments conducted in a high-pressure visual autoclave, which is an experimental geometry very different to that of a flow-loop. In these autoclave experiments, φ transition could be measured by up to three independent methods: (i) direct visual observation of the spatial distribution of hydrate particles in the sapphire autoclave; (ii) an increase in the rate of gas consumption to form hydrate; and (iii) an increase in the motor current required to maintain constant mixing velocity. The quantity utilized in this third method, which was the most reliable for determining φ transition, is an equivalent measure of the hydrate slurry s resistance to flow and is an analogue of the pressure drop measured in flow-loop experiments. The present study extends the work initiated by Akhfash et al. (2013) to quantify the effect of turbulence on φ transition in a methane-water system using the visual autoclave. In addition, this work studies the effect of transient operations (simulated shut-in and restart) on φ transition, during which subsea oil and gas operations are at the highest risk of hydrate plug formation (Sloan et al., 2011). 31

49 Joshi et al. (2013a) proposed a linear relationship between φ transition and mixture velocity; over six flow-loop experiments, φ transition was estimated at 10, 18, and 28 vol % for mixture velocities of 1.0, 1.75 and 2.5 m/s, respectively. Furthermore, they conducted experiments where the mixture velocity was changed from 1 to 2.5 m/s before and after 10 vol% hydrate had formed (the value of φ transition for 1 m/s). In both experiments, the system s resistance to flow evolved along the pathway observed for flow velocities of 1 m/s. These experiments indicated that the transition to a heterogeneous particle distribution was both irreversible and path-dependent. Joshi et al. (2013a) suggested that hydrate agglomeration or particle deposition on the wall may explain this behavior. The present work quantifies the hydrate bed formation point over 20 different turbulence conditions, and reports an observation of deposition and film growth that with further research might help identify the cause of the irreversibility and path dependence of φ transition. 3.2 Experimental Methods A high-pressure visual autoclave (HPVA) apparatus was deployed in the present study, and is discussed in detail in section on page 8. Prior to each experiment, the cell was rinsed for 60 seconds with toluene, ethanol, and acetone, and allowed to air dry. The cell was then filled with approximately 18 vol% deionized water, at which point the mixing impeller was fully submerged without creating an unstirred volume of water above the impeller level. While the bath remained at 20 C, the cell was pressurized with 20 bara methane (ultra-high purity, %) and vented; this flushing process was repeated three times to remove any residual air from the vapor space. The cell was then pressurized to the target initial condition (typically 60 bara) and sealed for 60 minutes to test for leaks within the system. In constant cooling/flow experiments, the mixing system was then engaged at constant velocity and the cooling ramp was initiated (1 C/hr to a target temperature of 1 C). Hydrate formation was detected by a sharp decrease in pressure, usually 0.5 to 1 hours after the cell temperature decreased below the hydrate equilibrium temperature. In shut-in and restart experiments, the system was cooled at 1 C/hr to a target temperature of 1 C without mixing. A thin film of hydrate was observed to form at the water-gas interface (approximately 0.5 mm thick, corresponding to about 1 vol% of the liquid in the cell), after which the mixing system was initiated at a constant speed. The minimum mixing speed (50 RPM) was set by the motor capability, and the restart mixing speed was varied over several experiments up to 800 RPM where the cell was previously established to be fully turbulent by Akhfash et al. (2013). After reaching the hydrate formation point, experiments typically required approximately 20 hours to reach a steady-state pressure signal. At the end of each experiment, the cell was re-heated to 20 C while being stirred at constant velocity to confirm complete recovery of the pressure signal and close the mass balance. 32

50 After hydrate nucleation was confirmed visually in each experiment, hydrate growth was estimated from a decrease in the pressure signal using methane compressibility factors from the Setzmann and Wagner (1991) equation of state implemented in REFPROP 9.0.(Lemmon et al., 2013). In the results presented below, the hydrate volume fraction is defined as the ratio of hydrate volume to the sum of hydrate and water volume, where the amount of hydrate formed is based on the measured decrease in the gas pressure. Water and hydrate phase densities were calculated using Multiflash 4.2 with the CPA model set (Infochem, 2012). For each mixing speed, the initial Reynolds number for the water-gas system was calculated according to the standard definition from Naumann (2008) for a stirred tank, as discussed in eq. (1). The volume of hydrate was compared to predictions made with a mass transport-limited growth model discussed in eq. (2), which was originally introduced by Skovborg and Rasmussen (1994). A detailed discussion on the parameterization of this model, including the relationships employed to calculate k MX and A G/W, is given by Akhfash et al. (2013). 3.3 Results and Discussion Evolution and shear rate dependence of hydrate transportability Akhfash et al. (2013) reported that after the initial hydrate formation in constant cooling/flow experiments four sequential visual states were identifiable in most experiments: a homogeneous distribution of particles, a heterogeneous distribution of particles, an apparent bedding and stratification of particles near the gas-liquid interface, and a complete plug. These same stages were observed in this work; however, as a result of conducting a large number of both shutin/restart experiments and continuous cooling/flow trials over a wide range of shear rates, we identified an additional stage that could potentially explain the irreversibility of φ transition with Reynold s number. In all cases (including those described by Akhfash et al. (2013)), hydrate nucleation was first observed at the water-gas interface (Figure 12-A), resulting in the growth of a thin hydrate film on the cell wall near the water-gas interface (Figure 12-B) and above the initially homogeneous distribution of hydrate particles. This hydrate film was observed to grow along the wall and the transition from homogeneous to heterogeneous distribution appeared to emanate from the level of the film, propagating in a downwards direction. Continued hydrate growth after the film had formed resulted in a hydrate bed with stratification near the gas-water interface (Figure 12-C). On the time-scale of the experiment (20 hours), the time required for the particle distribution to become heterogeneous and a hydrate bed form was small (few minutes), and thus the change in hydrate volume fraction present between the formation of the heterogeneous distribution and the stratified bed was negligible. The onset of the spatially heterogeneous particle distribution and bed formation coincided with an increase in the motor current needed to sustain the impeller speed; this measurement of an 33

51 increased resistance to flow was used to identify the value of φ transition. As reported by Akhfash et al. (2013), the formation of a hydrate bed increased the water-gas surface area, resulting in rapid hydrate growth in systems where the growth rate had been retarded by mass transfer limitations. In the final stage of the experiment, severe fluctuations in motor current were observed as the amount of hydrate reached the limit of hydrate plugging. These late-term growth stages and their impact on resistance to flow are discussed in detail by Joshi et al. (2013a) and Akhfash et al. (2013) for flow-loop and autoclave geometries, respectively. Figure 12. Images of methane hydrate formation under (left) constant cooling at 50 RPM and (right) shut-in and restart at 50 RPM: (A) hydrate nucleation at the water-gas interface; (B) hydrate film growth at the wall near the water-gas interface; (C) hydrate bed formation; and (D) late-term hydrate growth and catastrophic plug formation. Direct comparison between continuous cooling and shut-in/restart trials in Figure 12 illustrates that both experimental procedures trace the same late-term stages of hydrate plug formation, while the 34

52 initial nucleation behavior is likely to vary between continuous cooling (i.e. nucleation in the aqueous phase) and shut-in/restart (nucleation at the interface). The plug formation stages observed in both experiments are consistent with the conceptual picture shown in Figure 1, with the addition of hydrate film growth discussed above. The repeated observation of hydrate film growth (Figure 12-B) suggested its importance as a precursor to bed formation; no other phenomenon was consistently observed in the cell between initial hydrate nucleation and the onset of a heterogeneous particle distribution. This phenomenon may partially explain the irreversibility and path dependence of φ transition observed by Joshi (2012) in flow-loop experiments. Table 5. Summary of experiments conducted Exp. P initial bara T initial C φ transition RPM Re initial Re equivalent vol% Constant Cooling and Flow Operating Mode Shut-in and Restart Operating Mode

53 Figure 13. Onset of hydrate bed formation (φ transition ) as a function of initial Reynolds number in the HPVA apparatus. Eighteen constant cooling experiments were performed in the HPVA apparatus (Table 5), with a range of initial pressures of bara and temperatures of C; the systems were mixed at constant shear rates of RPM, corresponding to initial Reynolds numbers of In addition, twelve experiments were performed under shut-in and restart conditions, which were mixed at RPM after 1 vol % hydrate was detected in the cell. The Reynolds numbers for shut-in and restart trials were estimated based on the applied shear rate for a system without any hydrate growth (Re equivalent ) and ranged from For both continuous cooling and shutin/restart experiments, the Reynolds number was not recalculated as hydrate grew in the cell and only the Reynolds number prior to hydrate formation is reported; per eq.(1), there is no currently established viscosity relationship for hydrate-in-water slurries. Table 5 also lists the φ transition values for each of the 30 experiments performed determined from the increase in motor current, which range from vol % hydrate in constant cooling/flow experiments and 6-33 vol % in shut-in and restart trials. The calculated φ transition from motor current data are shown in Figure 13 as a function of initial Reynolds number in the cell for all 30 experiments. Two primary observations are apparent from the data in Figure 13. First, no significant difference in φ transition was observed between constant cooling/flow and shut-in/restart experiments. The average deviation in φ transition between constant cooling/flow and shut-in/restart experiments conducted at the same shear rate was 4.5 vol %, which is within the repeatability of the measurement. For shut-in and restart operations in production pipelines where the bathymetry does not contain low spots, this result suggests that the 36

54 process of restarting a cold system may not engender additional risk from a consideration of hydrate transportability. However, the process of restarting a cold system will naturally increase the hydrate growth rate (as discussed below), which may allow the system to readily surpass φ transition and enter a hydrate plugging region. Second, φ transition increases directly with Reynolds number. That is, the number of hydrate particles required to stabilize a bed increases with the system s degree of turbulence. This observation qualitatively agrees with flow-loop observations by Joshi et al. (2013a), but it is not quantitatively consistent with the values for the Reynolds number dependence of φ transition indicated by Sum et al. (2012) from those flow loop data. In addition, Joshi (2012) reported the results of three experiments made using a 4-inch internal diameter autoclave, where φ transition increased from about 12 to 19 vol % over corresponding stirred cell Reynolds numbers of 35,000 to 60,000 ( RPM). Comparing these ranges from Sum et al. (2012) and Joshi (2012) to the data reported in Figure 13 indicates that φ transition values are quantitatively similar for the same mixing speeds (RPM) in cylindrical geometries, but are inconsistent on the basis of Reynolds number. It is therefore unclear whether Reynolds number is the correct basis on which to scale φ transition values; further studies are required to determine the appropriate scaling relationship for φ transition under across flow-loop and autoclave geometries with varying internal diameters. The values of φ transition obtained for continuous cooling/flow experiments performed at 800 RPM ( Re initial ) were lower than those of the experiments performed at 700 RPM ( Re initial ) by about 13 vol %, which is larger than the reproducibility observed between repeat trials at lower shear rates. The timescale of φ transition in the present experiments is not addressed explicitly, as φ transition is observed to depend primarily on hydrate volume fraction and is coupled explicitly in time by hydrate growth rate; further studies by Akhfash et al. (2013) provide an experimental validation of a growth rate model, which may be deployed for a variety of mixing geometries to estimate the time dependence of φ transition. We have included the two data points at 800 RPM for completeness, because there is currently insufficient evidence to establish whether this is observation is indicative of a physical local maximum in φ transition (due, for example, to the mixing geometry deployed here) or whether these values simply reflect a poorer experimental reproducibility at high velocities. Further tests are clearly needed to resolve this issue, but should be designed in the context of answering the broader question relating to how a scaling relationship across system geometry might be established. For water-dominant systems, both Joshi et al. (2013a) and Akhfash et al. (2013) reported that the maximum resistance to flow occurred after φ transition, upon entering the later stages of hydrate and plug formation. For the autoclave experiments conducted here and previously (Akhfash et al., 37

55 2013), the maximum resistance to flow corresponded to the maximum motor current necessary to maintain the experimental shear rate. Akhfash et al. (2013) noted that, at the lowest shear rates, the maximum motor current was larger and was reached at a much lower hydrate volume fraction than at the highest shear rates; moreover the resistance to flow was sufficiently large in the low shear experiments that it stopped the impeller from rotating. In this work, the relationship between maximum motor current and shear was investigated more extensively. Figure 14 shows the way in which maximum relative motor current (i.e. normalized to its value before hydrate nucleation) was identified in each experiment, and how that peak value varied with the initial (equivalent) Reynolds number of each experiment. Systems in laminar or transition flow regimes (Naumann, 2008) exhibited a monotonic increase in the maximum resistance-to-flow with increasing initial Reynolds number. For systems that were in the fully turbulent region (Re initial >1000), the maximum resistance-to-flow decreased with increasing Reynolds number, with an abrupt transition being apparent from the laminar flow value. This latter observation may be the result of increased shear forces and turbulent eddies acting to reduce the degree of hydrate deposition and annealing to the walls (Aspenes et al., 2010, Aman et al., 2013a). An alternative mechanism is required to explain the reverse observation in the laminar and transition flow regimes. One possibility is that, at these low shear rates, the adhesive strength of the hydrate deposit exceeds the shear stress from the bulk fluid. The growing hydrate deposit provides a surface for continued hydrate growth, occluding the flow channel and interacting directly with the mixing blades. As is demonstrated in the following section, hydrate growth rate increases directly with the turbulence of mixing. Consequently, the observed peak in relative motor current at Re init = 1145 (Figure 14, bottom panel) represents a worst-case mixing velocity, resulting in the highest possible growth rate that does not enable sufficient sloughing of hydrate deposited on the wall. The trend observed for the shut-in/restart experiments is consistent with this picture; as discussed below, the initial formation rates in these experiments were less sensitive to shear rate, and were significantly higher than those observed at low mixing velocities in the continuous experiments. 38

56 Figure 14. Top: Relative motor current (resistance-to-flow) as a function of time after nucleation for constant cooling/flow trials with initial Reynolds numbers of 1145 and 1450, with the maximum value observed in each trial circled. Bottom: Maximum relative motor current observed during constant cooling and shut-in/restart experiments, as a function of initial turbulence of mixing. The relative motor current is defined as the motor current measured during hydrate formation, normalized by the baseline motor current prior to hydrate formation. In water-dominant systems, it is clear that the increase in the effective viscosity of the hydrate slurry occurs by quite a different mechanism to that reported for oil-dominant systems. Aman et al. (2012) demonstrated that the cohesive force between cyclopentane hydrate particles suspended in an aqueous bulk phase was minimal, and it is clear from the results shown here as well as by Joshi et al. (2013a) and Akhfash et al. (2013) that there is no effective viscosification of the watercontinuous hydrate slurry as long as the particle distribution remains homogenous. However, the 39

57 mechanism by which the resistance to flow increases upon the transition to a heterogeneous particle distribution remains unclear. We hypothesize that one mechanism may be the effective viscosification of localized volume elements within the heterogeneous distribution, in which the particle number density approaches the limit of particle jamming. Even though the cohesive forces between hydrate particles may be minimal, the relative viscosity of the localized volume element could become very large, particularly if, as proposed by Mills (1985), it diverges to infinity as the local particle concentration approaches about 57 vol%. Since the number of localized volume elements with a large effective viscosity will increase with the thickness of the hydrate bed, their overall contribution may come to dominate the flow resistance of the entire system. This hypothesis suggests that avoiding the formation of a hydrate bed will be crucial to the management of plugging risk in a water-dominant flow. Anti-agglomerant type strategies, which aim to reduce the effective viscosity of the slurry by lowering inter-particle forces between hydrate, are unlikely to be effective in water-dominant systems because (i) the inter-particle forces are already very low, and (ii) regardless of how small they are, the slurry s effective viscosity will increase when localized regions within the bed reach very high particle densities Evolution and shear rate dependence of hydrate growth The hydrate formation rates measured during the constant cooling/flow and shut-in/restart experiments were compared to the mass transport-limited hydrate growth model introduced above. Experimental values of temperature, pressure and hydrate volume % for all experiments are listed in Table 6 and compared with values calculated using either the MultiFlash equilibrium model (Infochem, 2012) or the mass transport-limited hydrate growth model discussed by Akhfash et al. (2013). The measured cell and calculated equilibrium temperatures indicate that, in all constant cooling/flow and shut-in/restart cases, a positive subcooling (T eq T) was maintained for the duration of the experiment, indicating the absence of heat transport limitations between the cell and bath. In all experiments, the cell temperature was always maintained below the hydrate equilibrium temperature corresponding to the cell pressure, confirming the appropriate application of a mass transport-limited growth model. At low mixing speeds (e.g. 250 RPM in Figure 15), the hydrate volume % was observed to plateau at an intermediate value, while the system remained in the hydrate formation region (Table 6). Under highly turbulent conditions (e.g. 800 RPM in Figure 15), the initial hydrate growth rate was two orders of magnitude larger and growth was sustained for a longer period, which resulted in a volume % well above 50 % at steady-state. Figure 16 shows that the maximum amount of hydrate formed in each continuous cooling/flow experiment increased with initial Reynolds number over the range , but remained constant at (71 ± 7) vol % for turbulent shear rates 40

58 (1000 < Re initial 4515). The largest hydrate volume % achieved was 84 % at a shear rate of 550 RPM (Experiment 11), where the final subcooling was 0.1 C. The experiments presented herein behaved similarly to those reported by Akhfash et al. (2013) with the water phase not being completely converted even after 24 hours of mixing. This was likely the consequence of forming a stationary hydrate bed at and above the water-gas interface, which prevented the convective transport of methane to unreacted water at the bottom of the cell. Table 6. Summary of experimental results from constant cooling/flow and shut-in/restart experiments, together with values calculated with an equilibrium model (Infochem, 2012) and a mass-transport limited model (Akhfash et al., 2013). Exp. T nuc P nuc T eq,nuc T final T eq,final P final (model) P final (exp) φ max (model) φ max vol% C bara C C C bara bara vol%

59 Figure 15. Hydrate volume %, with respect to the liquid phase, as a function of time after nucleation for experiments 6 and 17 (250 and 800 RPM, respectively); the dashed curve represents predictions from the mass-transfer limited model (eq. (2)) for each experiment. Figure 16. Maximum hydrate volume % (of original water volume) for constant cooling/flow (solid circles) and shut-in/restart (open circles) data, as a function of the initial turbulence of mixing; the cell is expected to be fully turbulent at Reynolds numbers above approximately 1000 (Naumann, 2008). The results for the shut-in/restart experiments shown in Figure 16 are somewhat different to those for the continuous cooling/flow experiments. For Re equiv <3000, the final hydrate volume % 42

60 achieved was quite consistent and remained below 50 vol %, which is nearly a factor of 2 lower than that produced in the continuous cooling flow experiments with Re init >1000. This indicates that unless the shear was very high, the break-up of initial hydrate film upon restart ultimately led to a stationary bed that was more effective at preventing the conversion of water to hydrate than was the case in the continuous cooling/flow experiments. However, the opposite was true when one considers the hydrate formation rates achieved at the beginning of each experiment. Figure 17 shows the initial hydrate growth rate achieved for both the constant cooling/flow and shut-in/restart experiments, determined from the rate of gas consumption for the first 2 vol % of hydrate formed after nucleation or after re-start, respectively. For the continuous cooling/flow experiments, the initial hydrate growth rate increased exponentially with initial Reynolds number. A similar pattern was observed for the shut-in/restart experiments with two notable changes. First, the scatter between replicate data points was consistently larger, which may be the result of surface area variance during the growth of an initial hydrate film at the gas-water interface. Second, the minimum growth rate attained (at 50 RPM) was approximately one order of magnitude larger than that for the constant cooling experiments at the same shear rate. This result is expected as the initial hydrate formation period for shut-in and restart experiments begins with a substantially larger hydrate-water surface area. Figure 17. Initial hydrate growth rate for continuous cooling/flow (filled data points) and shutin/restart (open data points) experiments listed in Table 5 as a function of initial turbulence of mixing in the cell. Growth rate was defined for the first 2 vol % of hydrate following nucleation or the restart of shear. The solid curve corresponds to the model from Skovborg and Rasmussen (1994), based on diffusion-limited hydrate growth in an aqueous phase. 43

61 The kinetic-type relationship discussed by Vysniauskas and Bishnoi (1985) indicates that the rate of hydrate growth will depend linearly on the existing hydrate crystal surface area. In contrast, when considering mass-transport limited growth, Skovborg and Rasmussen (1994) identified the gas-water interfacial area and transport through the aqueous phase to be the quantities that limit the formation of the hydrate phase. The initial growth rate predictions of the model by Skovborg and Rasmussen (1994) are shown in Figure 17, which are calculated using eq.(2). These model calculations use the geometric formalism described by Akhfash et al. (2013) for a stirred autoclave cell are in reasonable agreement with the initial rates measured for constant cooling/flow experiments, particularly if the shear rate was sufficiently high. The substantially higher growth rates observed in the first stages of the shut-in/restart experiments indicate that, at least initially, the formation was kinetically limited rather than mass-transport limited. It is very difficult to estimate the hydrate-water surface area involved with this kinetic reaction and between repeat experiments this surface area would very likely have varied significantly. However, the data indicate that this kinetically-limited initial growth rate increased with Re equivalent, which could reflect that higher rates of initial shear caused the film to be broken into smaller pieces with a correspondingly larger hydrate-water surface area. 3.4 Chapter Conclusions Methane hydrate growth, bed formation, and resistance-to-flow were studied in a high-pressure visual autoclave. Hydrate plugs were formed through both constant cooling/flow and shutin/restart procedures, leading to four primary conclusions: 1. The hydrate volume fraction required to produce a moving hydrate bed at the water-gas interface (at a volume fraction φ transition ) increases directly with the degree of turbulence in the system. 2. Forming hydrate plugs from a cold restart did not affect the relationship between φ transition and turbulence. 3. For constant cooling/flow experiments with systems under laminar or transitional flow regimes, the maximum resistance-to-flow observed (following φ transition ) increased with turbulence. For systems in the fully turbulent regime, the maximum resistance-to-flow decreased with shear. 4. Initial hydrate growth rates for cold restart experiments were approximately one order of magnitude larger than comparable constant cooling experiments. The third conclusion, coupled with the observations of the system evolution possible in the visual autoclave, suggest that hydrate film growth and particle deposition at the wall represent critical 44

62 steps in the formation of a hydrate plug, and possibly explain the path dependence and irreversibility of φ transition with Reynolds number. Quantitative comparisons of the shear dependence of φ transition reported here and for different flow geometries by Sum et al. (2012) and (2013a) indicate that a universal relation is yet to be found; additional comparative tests between autoclave and flow-loop geometries are required to determine whether Reynolds number, as currently used, is the most appropriate relation through which to upscale autoclave results to pipeline-type geometries. Finally, the results reported here indicate that for water-dominant systems, the more turbulence applied during a re-start process, the greater the transportability of the resulting hydrate slurry, and the lower the risk of plug formation. 45

63 4 Microscale Detection of Hydrate Blockage Onset in Highpressure Gas-Water Systems 4.1 Introduction Investigations of hydrate formation in gas-water systems using flow-loops and autoclaves (Joshi et al., 2013a, Akhfash et al., 2013, Aman et al., 2015a, Akhfash et al., 2016) have suggested that a critical stage in the formation of blockages is the buoyant collection of hydrate particles to form a moving bed near the gas-water interface. Joshi et al. (2013a) performed several flow-loop studies of high water cut systems (including 100% water cut) at different fluid velocities and liquid loadings. The behaviour of the pressure drop across the flow-loop during hydrate growth resulted in a three-stage conceptual model for hydrate blockage formation (Figure 1). During the initial stages of hydrate formation in these flow-loop tests no increase in pressure drop was observed, which was assumed to correspond with a homogeneous distribution of solid particles in the aqueous phase. Continued hydrate growth resulted in an increased pressure drop, consistent with the formation of a moving hydrate bed near the gas-water interface; in this condition, the particles are heterogeneously distributed thorough the cross-section of the aqueous phase. Joshi et al. (2013a) labelled the hydrate volume fraction at which this heterogeneous behaviour was first observed as φ transition. This transition from a homogeneous to a heterogeneous hydrate-in-water particle distribution preceded the ability for hydrate particles to form a stationary bed and deposit on the flow-loop wall, as the final stage in this mechanism. In this latter stage, the frictional pressure drop in the flow-loop increased by at least one order of magnitude. Akhfash et al. (2013) and Aman et al. (2015a) observed similar patterns to Joshi et al. (2013a) s flow-loop data by monitoring resistance-to-flow (torque or motor current) in a sapphire visual autoclave, where resistance-to-flow was observed to increase significantly only above a critical hydrate volume fraction (φ transition ). These measurements allowed direct visual observation of hydrate bed formation in the autoclave geometry, and provided visual confirmation of the onset of a heterogeneous particle distribution prior to severe hydrate blockage formation. Experiments conducted in both flow-loops (Joshi et al., 2013a) and autoclaves (Aman et al., 2015a) showed that an increase in shear rate (flow velocity or impeller speed, respectively) delayed the onset of hydrate bed formation to higher values of φ transition. Further, Joshi (2012) suggested that the formation of a moving hydrate bed was irreversible, which may have been the consequence of initial wall film growth visually observed by Aman et al. (2015a). However, while qualitative agreement in the shear-dependence of φ transition was observed between experiments in different autoclaves and flow-loops, significant inconsistencies are found when any dependence of φ transition on Reynolds number is considered, where the latter dimensionless group might be 46

64 expected to provide a quantitative measure of the shear force that is independent of flow geometry. As discussed by Aman et al. (2015a), this suggests that the various length scales used in the calculation of Reynolds number within different geometries are not suitable for the generalised prediction of hydrate bed formation. To build such a relation, more information about the relevant length scales at which hydrate particles begin to interact in water-continuous systems is needed. In assessing a system s propensity for forming a moving hydrate bed, the dispersed particle size is a critical parameter. Focused Beam Reflectance Measurements (FBRM) are commonly used to study this property; an FBRM probe measures the number and size of chord lengths at a particular point in space, enabling the generation of a chord length distribution (CLD). Examples of FBRM experiments include characterizing emulsions and droplet size distributions in oil (Greaves et al., 2008, Boxall et al., 2010, 2012, Less and Vilagines, 2013, Melchuna et al., 2016), describing hydrate nucleation and growth behaviour (Barrett and Glennon, 2002, Greaves et al., 2008, Turner et al., 2009a), analysing the performance of anti-agglomerant chemical additives (Leba et al., 2010, Cameirao et al., 2012, Lv et al., 2014, Chen et al., 2014, Chen et al., 2015), and modelling the rate of hydrate formation and dissociation in gas-water systems (Clarke and Bishnoi, 2004, 2005). An FBRM probe enables in situ measurements of particle size in opaque or dark surroundings and in suspensions at high solid concentrations. It does not require an assumption of particle shape to calculate the CLD, although refinements can be made using additional information about the particle size distribution (PSD) that may be available as discussed by several authors (Ruf et al., 2000, Wynn, 2003, Worlitschek et al., 2005, Li and Wilkinson, 2005, Li et al., 2005, Kail et al., 2009). Further to this, Heath et al. (2002) highlighted the use of different weighting functions (no weight, length weight, square weight and cube weight) that can be applied to FBRM measurements, which may be used to correct for measurement bias when reporting statistical parameters of the distribution. Heath et al. (2002) reported FBRM measurements for well-known particle size distributions between 50 and 400 microns, and concluded that the mean square weighted chord length provided the best estimate of the actual mean particle size. This square weighted function was therefore applied to all the experimental results discussed in this paper. However, while the measured CLDs are assumed to be close representations of the hydrate PSD, the objective of this work was to study the evolution of those distributions rather than their particular numerical values. A review of the existing literature indicates numerous uses of the FBRM technique in oilcontinuous systems, such as works from Greaves et al. (2008), Boxall et al. (2010;2012), Less and Vilagines (2013) and Turner et al. (2009a). For instance, Turner et al. (2009a) utilized an FBRM probe in an autoclave to monitor the changes in the size distribution of water droplets in the oil as 47

65 hydrate grew in the system. Apart from a minor increase in the population density of large particle sizes during hydrate nucleation, Turner et al. observed no change in CLD until the end of the experiment, validating the hypothesis that water droplets gradually convert to hydrate. In other oil studies, Leba et al. (2010) and Cameirao et al. (2012) combined in situ FBRM measurements with pressure drop and temperature signals from flow-loop experiments, to study agglomeration processes in water-in-oil emulsions at low water cuts. In both studies no increase in the mean chord length signal (un-weighted or square-weighted) was observed during early stages of hydrate formation, indicating no change in the size of water droplets/hydrate particles. Approximately 50 minutes after hydrate nucleation, the FBRM chord length measurements suddenly increased, which was considered to indicate the onset of particle agglomeration. This shift in the FBRM CLD was also consistent with an increase in the measured pressure drop across the flow-loop. A model for hydrate aggregation based on a monodispersed population of water droplets in water-in-oil emulsions was then proposed (Leba et al., 2010), which was subsequently improved by Cameirao et al. (2012) by considering a polydispersed log-normal distribution of water droplets (Rajagopal, 1959, Boxall et al., 2010). Simulated CLDs were then compared to experimental measurements of CLD obtained from the FBRM at different levels of agglomeration, to adjust the fractal dimension and number of hydrate primary particles contained within the agglomerates (Cameirao et al., 2012). However, few studies are available on the use of FBRM probes to characterize water-continuous systems. Clarke and Bishnoi (2004;2005) used FBRM size measurements to define the moments of the particle size distributions, in order to model the rate of gas hydrate formation and dissociation in a gas-water system. They assumed that mixing in the stirred tank was sufficiently high to consider the FBRM readings representative of the whole system, and that the FBRM probe scanned all the particles in the field of view. Reasonable agreement was obtained between the predicted and experimental formation/dissociation rates based on their FBRM measurements. In this work, a stainless steel autoclave equipped with an FBRM probe was used to monitor the changes in the distribution of hydrate particles in gas-water systems over a range of mixing speeds (50 to 700 RPM). The FBRM data were coupled with resistance-to-flow (motor current) measurements to explore the relationship between hydrate particle distribution and bed formation. From our knowledge of the published literature, this study represents the first use of FBRM measurements to investigate the onset of hydrate bed formation (φ transition ). Further, in one experiment, the FBRM probe was used together with a Particle Video Microscopy (PVM) probe, which provided some visual information about the size and morphology of the hydrate crystals and aggregates. 48

66 4.2 Materials and Methods Apparatus A simplified schematic diagram of the high-pressure autoclave apparatus is shown in Figure 18. The apparatus used in this study was essentially identical to that used by Greaves (2007) and Boxall (2009), so only the differences to the original apparatus and a summary of its key features are described here. The autoclave was constructed by Challenger Manufacturing using 316 stainless steel, with a 10.2 cm internal diameter (four inches) and a 22.9 cm height (nine-inches). The internal volume of the cell was 1775 ml, and the cell was rated to a maximum operating pressure of 83 bar (1200 psi). The cell contents were agitated using an anchor-type impeller with a 6.5 cm diameter. The autoclave apparatus consisted of a magnetic drive shaft with maximum torque of 15 kg.cm, connected to an AC drive motor with a maximum rotational speed up to 2000 RPM, provided by Sejin Young Tech, Korea. The cell was connected to a high-pressure gas reservoir (2286 ml) through a pressure control valve. All experiments in this work were performed under isobaric conditions, using the gas injection port at the top of the cell; the amount of gas consumed as hydrates formed in the autoclave cell was determined from the change in pressure of a constant volume reservoir that supplied the gas. The autoclave cell was submerged in a waterglycol bath, where the bath fluid was continuously circulated through two Fisher Scientific chillers to control cell temperature and remove heat during the exothermic process of hydrate formation. To help maintain isothermal conditions, injection gas from the high-pressure reservoir passed through 10 meters of 1/8 inch stainless steel tubing located inside the glycol bath before injection into the autoclave. The temperatures of the autoclave cell and bath were measured using two 1/8 inch T-Type thermocouples from Omega Engineering, where the cell thermocouple was positioned about 2 cm inside the liquid phase. Two Omega transducers (0-207 bar) with a resolution of ± 0.1 bar monitored the cell and reservoir pressures. A LabVIEW data acquisition system (DAQ) was used to record cell and bath temperatures, cell and reservoir pressures, and the motor current required to maintain a constant mixing velocity within the cell (e.g. 300 RPM) at 5 s interval over the entire period of the experiments. 49

67 P, T, and motor current to DAQ system Motor and Magnetic Coupling Cold Fluid Circulation From External Gas Source Gas Reservoir Gas Injection line PVM Probe P, T FBRM Probe P, T Recirculation Pump Autoclave Cell Water/Glycol Bath T Figure 18. Simplified sketch (left) and picture (right) of the high-pressure autoclave cell. The FBRM probe used in this study was a Mettler-Toledo Lasentec D600X particle size analyser. The FBRM probe was directly in contact with the fluids throughout the experiment. The FBRM functions by emitting a class I, 3 mw laser beam with wavelength nm through the probe tip at the sapphire window (Figure 19). The optical system rotated at a speed of 2123 RPM, corresponding to a linear velocity of 2 m/s (by considering a tip (scanning) diameter of 18 mm), allowing the laser to scan across the surface of particles in front of the sapphire window; the laser was reflected when it scanned the surface of a particle. The scanned chord length was calculated based on the measured reflectance time and the known rotational speed. Generally, thousands of particles were scanned per second, generating a detailed chord length distribution in real time. The measured data were analysed to determine the mean chord length and number of chord counts as a function of time. Chord length distributions were analysed for the range of 1 to 1000 μm on a logarithmic basis. The PVM probe used in this work was a Mettler-Toledo Lasentec model V800S. The PVM used a high-resolution camera and six lasers to illuminate the material in front of the probe window, to obtain high-resolution images of particle suspensions or emulsions (Figure 19). The PVM field-ofview was 860 by 645 μm, with a minimum resolution of 20 μm. The FBRM and PVM probes were inserted at angles of 45 o into the cell to maximize the flow across their windows and to allow representative particle sampling (Mettler-Toledo-Autochem, 2016). The position of the probes in the autoclave was determined based on the simulation of the flow behaviour inside the autoclave, where the details are discussed by Greaves (2007). 50

68 Figure 19. (a) FBRM probe assembly; (b) the procedure of FBRM chord length measurements; (c) PVM probe assembly; (d) an example of PVM imaging: methane gas bubbles in water after hydrate dissociation, captured from one of the experiments performed in this work. Sketches (a) (c) are redrawn based on information from Mettler-Toledo-Autochem (2016) Experimental Procedure Gas hydrate was formed in the high-pressure autoclave apparatus according to the following procedure: (1) The cell was thoroughly cleaned with acetone, rinsed with deionized water (DI), and dried completely with compressed air. (2) To calibrate FBRM and PVM readings prior to each experiment, the probes were cleaned with DI water and acetone, then dried carefully until the total chord counts in the FBRM reading was 51

69 reduced below 100 (over 10 s) and no particle was visible in the PVM window. To decrease the hydrate deposition/accumulation on the sapphire windows of the FBRM and PVM, a silane-based hydrophobic surface treatment was applied to the surface of the probes, as discussed by Greaves (2007) and Boxall (2009). (3) The cell was loaded with 900 ml deionized water to completely cover the impeller blade, after which the cell was submerged in the water/glycol bath. (5) The cell was flushed with high purity methane gas (99.999%) at 20 bar to remove any remaining air from the system. (6) The bath temperature was maintained at 20 and the cell was pressurized to 65 bar and leaktested for at least two hours. (7) The motor speed was set at a pre-defined rotational speed (e.g. 300 RPM); all valves between the cell and gas reservoir were set to the required configurations for constant-pressure operation, and the system was left overnight to ensure complete saturation of water occurred prior to hydrate formation. (8) The bath temperature was cooled to 1 to allow for hydrate formation, with the chillers working at their maximum capacity to deliver a cooling rate of 8 /h. Hydrate nucleated within minutes after the system crossed over the equilibrium point for Experiments 2-8 (listed in Table 7). However, for Experiment 1 at 50 RPM, nucleation occurred after 13 h. Experiments typically required 7 to 24 h to reach a steady state pressure signal. (9) After the hydrate formation process ceased, the system was heated to 20 in a stepwise manner to dissociate hydrate in a constant-pressure mode. Finally, a mass balance was performed by calculating the number of moles of gas in the dissociation reservoir with respect to the number of moles consumed during hydrate formation. In the present study, experiments were initiated at ambient temperature (22 ± 1 ºC) and with a constant cell pressure of 65.3 ± 0.5 bar. Hydrate formation was identified from a slope change in the measured pressure signal with time, due to gas consumption. The amount of hydrate formed in the cell was calculated through the measured pressure decrease of the reservoir. Multiflash 4.2 using the cubic plus association (CPA) model set was used to calculate water and hydrate phase densities and the methane compressibility factor (Infochem, 2012), while the methane hydration number was assumed to be a constant The initial Reynolds number (Re initial ) of the water phase was calculated at 21 and 65 bar (a common initial state) for each mixing condition, using the standard definition for stirred tanks (Nauman, 2008), as discussed in eq. (2). The hydrate volume fractions reported hereafter were calculated based on the total volume of the hydrate, V H, 52

70 and water, V W, phases: i.e. = V H / (V H + V W ). The reproducibility of the results obtained at various shear rates was checked by conducting two repeat tests. 4.3 Results Resistance-to-flow measurements Eight experiments were performed in the autoclave cell (Eight experiments were performed in the autoclave cell (listed in Table 7) at different motor speeds ranging from 50 to 700 RPM; the initial Reynolds numbers ranged from 3500 to Hydrate nucleated at temperatures ranging from 1.5 to 6.2, equivalent to a subcooling range of 3 to 7.5 for pure methane gas, where the hydrate equilibrium temperatures (reported in Table 7) were predicted from the measured pressure using the Cubic Plus Association model set implemented in Multiflash 4.2. While the final set point for the bath temperature was fixed at 1, the temperature observed during the latter stages of each experiment ranged between 6.5 and 8.7. This indicates that once hydrate formed the cooler s power was insufficient for the thermal mass of the system and the heat transfer from the environment for it to reach the bath temperature within the experimental timeframe. However, this is unlikely to have any significant impact on the results observed here, given that sufficient hydrate growth always occurred over the experiment s duration to achieve hydrate volume % much larger than transition. Table 7 also shows maximum amount of hydrate formed in each experiment. As discussed above, the measured increase in motor current provides a resistance-to-flow metric analogous to pressure drop in a flow-loop or pipeline, and is generally a function of hydrate volume fraction in the system. As explained in detail by Joshi et al. (2013a), Akhfash et al. (2013) and Aman et al. (2015a), hydrate particles impose a negligible amount of resistance-to-flow on the system as long as they are distributed uniformly in the aqueous phase and at low concentration. Eventually, particle interaction and bed formation result in a heterogeneous particle distribution throughout the system and the measured resistance-to-flow increases. Figure 20 shows an example of relative motor current as a function of hydrate volume %, as observed for experiment 4 listed in Table 7 (300 RPM). The relative motor current was calculated based on the ratio of the instantaneous motor current in the presence of hydrates to the average motor current prior to nucleation. In Figure 20, the initial increase in relative motor current above the baseline (R.M.C. ave line in Figure 20) was used to estimate the value of φ transition. This was done by calculating the standard deviation of the baseline relative motor current u(r.m.c. ave ) (i.e. measured well before any consistent increase in current occurred). A threshold value of three times u(r.m.c. ave ) was taken to be the point at which an unambiguous change in the relative motor current had occurred. By fitting a line to the relative motor current data measured above (and in the vicinity of) this threshold, φ transition was estimated to be the point of intersection between R.M.C. ave +u(r.m.c. ave ) 53

71 and the fitted line. The uncertainty in the determination of φ transition by this method was estimated from the intersection of the fitted line with the threshold R.M.C. ave +3u(R.M.C. ave ). For the particular experiment at 300 RPM shown in Figure 20, φ transition was estimated to be about 15.8 vol% hydrate with φ + transition estimated to be 16.7 %. The observed increase in resistance-to-flow continued until about 35 vol % hydrate formed in the cell (see Figure 20). At these later stages of hydrate formation, rapid fluctuations in motor current were interpreted as being a result of particle jamming and stationary bed formation (or particle deposition) throughout the experimental cell. Table 7. Summary of experiments conducted in the autoclave cell. In all experiments, the hydrate equilibrium temperature was 9.0 ± 0.1 ºC. The hydrate volume % in the liquid phase at the end of each experiment is denoted φ final. The relative motor current (R.M.C.) was calculated based on the ratio of the instantaneous motor current (post nucleation) to the average motor current measured prior to nucleation. The quantities φ transition and φ + transition represent the average and maximum hydrate volume % at which an increase in relative motor current above R.M.C. ave +u and R.M.C. ave +3u was observed, respectively, where u is the standard deviation of the relative motor current measured prior to φ transition. Exp. RPM Re initial T nucleation ( C) T final ( C) P consumed (bar) φ final (vol%) R.M.C. ave before φ transition u(r.m.c. ave ) before φ transition φ transition + φtransition Maximum R.M.C * * * * ** ** ** ** ** ** 4.1 *: Not able to be specified due to significant fluctuations in motor current observed in Experiment 1 at low hydrate volume %. **: Not able to be specified, as the amount of water in the cell was not measured accurately. The result is reported because in Experiment 8, both the FBRM and PVM probes were in service. Two main observations may be derived from the resistance-to-flow and φ transition results presented in Table 7. First, there was no statistically significant difference in the values of φ transition obtained for experiments performed at different shear rates (16±2 vol % hydrate), within the confidence assigned to the motor current measurements. Although some previous observations (Joshi, 2012, Akhfash et al., 2013, Aman et al., 2015a) showed a direct dependence of φ transition with the degree of turbulence in the system, this trend was not clear within the range of shear rates applied in this 54

72 study, potentially because of the limited resolution of the motor current measurements. As discussed below, φ transition 16 vol % may correspond to the upper threshold for hydrate particle transport in water, if the system is to maintain minimal interparticle interactions. Second, the maximum relative resistance-to-flow achieved in the system (during the later stages of plug formation) decreased as shear rate increased, despite the conversion of additional water to hydrate, as was observed and discussed in detail by Aman et al. (2015a). Table 7 shows that the maximum resistance-to-flow reduced by a factor of 8 as the mixing speed was increased from 100 to 700 RPM. At low shear rates (e.g. experiment 2, 100 RPM), 40 vol % hydrate in the aqueous phase was sufficient to stop the impeller rotation, but at higher shear rates (e.g. 300 and 700 RPM), between 56 and 78 vol % hydrate was required to achieve a plugging condition. R.M.C. ave +3u.R.M.C. ave +u.r.m.c. ave φ transition Figure 20. Relative motor current (resistance-to-flow) as a function of hydrate volume % for a gas-water system at 300 RPM (experiment 4). The insert figure contains the same data with a magnified ordinate range to schematically show how φ transition and φ + transition were calculated. The values reported in Table 7 were obtained based on the intersection of the fitted lines. 55

73 The values of φ transition derived from these motor current measurements agree with observations from the literature for water-gas systems. Joshi (2012) reported φ transition in the range of vol% for three methane-water experiments performed in a 4-inch autoclave cell with a vane-blade impeller geometry over the range of to Reynolds number (calculated using eq. 1), which corresponded to a shear rate of RPM. The present observations are also in qualitative agreement with the observations of Akhfash et al. (2013) and Aman et al. (2015a) in a 1 inch sapphire autoclave cell, where φ transition was found to be in the range of vol % for shear rates of RPM. However, because of the smaller ID, these shear rates correspond to a Reynold s number range of only , which are an order of magnitude below those observed in the larger autoclave and flow loop experiments. This suggests that while shear rate affects the value of φ transition, the Reynolds number as defined by eq (1) and/or the length scale therein are not generally relevant to the onset of hydrate bed formation In Situ Images of Hydrate Formation In addition to resistance-to-flow observations (Section 4.3.1), the morphology and diameter of hydrate crystals were monitored in one experiment with a PVM probe inserted in the autoclave cell. Figure 21 illustrates sample PVM images captured at different times after nucleation for experiment 8 (300 RPM). The PVM images did not clearly show solid hydrate particles until five minutes after nucleation, presumably because the microscope s objective lens limited the PVM s ability to resolve particles smaller than 20 micron. In Figure 21, images are only shown up to 40 minutes after nucleation because at that point a large mass of hydrate adhered to the PVM objective lens, preventing any further meaningful measurements. 56

74 6 min 5 min 9 min 200μm 10 min 200μm 17 min 11 min 200μm 18 min 200μm 200μm 22 min 26 min 200μm 200μm 200μm 40 min 31 min 29 min 200μm 200μm 200μm 200μm Figure 21. PVM images captured at different times after nucleation during the experiment carried out at 300 PM (experiment 8 in Table 7). While the PVM provided an approximate range of 50 to 700 microns for the size of hydrate particles, it did not provide any quantitative information on their spatial distribution in the aqueous phase nor any evidence for the formation of a hydrate bed. In contrast, the FBRM was able to identify particle sizes as small as 0.5 microns, providing a 40-fold improvement from the minimum detectable size by the PVM. The FBRM was used to measure particle number concentrations and size distributions in the water phase as a function of time. 57

75 4.3.3 FBRM Measurements The distribution of gas bubbles and hydrate particles in the liquid phase on a micron length scale was quantified through FBRM online measurements. The PVM and FBRM results were broadly consistent in terms of the size ranges of solid particles in the region where PVM was able to measure particle size. However, in addition to the wider range of measurable particle sizes, the FBRM was also successful in detecting hydrate nucleation through a sharp increase in the chord length and number of chord counts measured which corresponded directly to the decrease in the cell pressure caused by gas consumption upon hydrate formation. The FBRM measurements provided valuable information on the size of entrained gas bubbles in water prior to hydrate formation. Figure 22 shows the mean square weighted chord length as a function of time after nucleation, where chord length can be taken as proportional to the size of gas bubbles and hydrate particles. Two important conclusions can be drawn from the data in Figure 22. First, the chord length of the entrained gas bubbles was measured in the range of 1 to 400 microns with no measurable dependence on shear rate prior to hydrate nucleation. This result may provide guidance as to a reasonable approximation of initial gas-water surface area estimates, which are central to hydrate growth rate calculations. Second, Figure 22 illustrates that upon initial hydrate nucleation, the diameter of the dispersed particles observable with the FBRM increased within a matter of minutes to a maximum value, and then decreased over a sustained period (tens of minutes). Collectively, the data in Figure 22 suggest that the initial formation of hydrate in aqueous systems may include appreciable numbers of hydrate-encrusted gas bubbles (Sloan and Koh, 2007), and that following initial hydrate formation, the particles may experience shear-induced breakup, particularly since interparticle cohesive forces (which might balance the disruptive shear forces) are negligible in water-continuous systems (Aman et al., 2012). Figure 22e and f show that, prior to hydrate formation, the total FBRM signal was small because of limited reflection of light by gas bubbles. The FBRM measurements are sensitive to the optical and surface properties of the particles being scanned, and the refractive index of gas bubbles is significantly lower than the refractive index of hydrate particles. Consequently, while the FBRM could detect a range of bubble sizes, the total number of counts observed before nucleation was much smaller than occurred afterwards. Following nucleation, the rapid increase in the size of detectable particles may have corresponded to the conversion of large, hard-to-detect bubbles to hydrate-encrusted bubbles, which can be easily observed with the FBRM. Figure 22d and f indicate that with increased shear (700 RPM), a wider range of bubble sizes were detectable as the number of counts increased several fold over the number and range of bubble sizes observed at low shear (100 RPM). 58

76 (a) (b) (c) (d) (e) (f) Figure 22. Mean square weighted chord lengths (a-d) and chord counts (e-f) of gas bubbles and hydrate particles measured by FBRM as a function of time post nucleation for various impeller speeds. Prior to nucleation, the measured chord lengths and chord counts are indicative of the size and number of the of gas bubbles detected by the FBRM. 59

77 As a particle size analyzer, the FBRM was also able to measure the chord length distribution during the hydrate growth phase. Figure 23 shows chord length distributions of the hydrate slurry at different hydrate volume % for both a low shear rate (experiment 2 at 100 RPM is shown in the left panel) and a high shear rate (experiment 6 at 700 RPM is shown in the right panel). The FBRM measurements were collected at 10 s intervals. The distributions shown in Figure 23 have been selected to show the trend in particle size distribution measured throughout the hydrate growth phase. In both the low and high speed mixing experiments, once hydrate nucleation was detected the measured chord counts immediately increased indicating the FBRM was successful at detecting hydrate growth. However, after about 3 vol % hydrate at 100 RPM, or 8 vol % hydrate at 700 RPM, an unexpected decrease in the number of counts was observed: as hydrate continued to form (based on the measured pressure drop) the number of counts decreased until a limiting hydrate volume % was reached (10 to 20 vol %). Above these limits, the FBRM was no longer sensitive to the amount of hydrate in the system, as the CLD no longer varied with time. Figure 23. Chord counts-chord length distributions measured by FBRM, reported at different hydrate volume % for methane hydrate-in-water slurries. (Left panel): experiment 2 at 100 RPM and (right panel): experiment 6 at 700 RPM. Grey traces correspond to a sequence of distributions that increase in height with increasing hydrate volume %, while colored traces correspond to a sequence of distributions that decrease in height with increasing hydrate volume %. (For interpretation of the references to color in this figure legend, refer to the web version of this thesis.) On first consideration, it might be expected that the number of particles measured by the FBRM should have increased as hydrate continued to grow in the cell, and as larger particles were broken up due to shear. Further, one might attribute the drop in the number of particle counts to the 60

78 agglomeration/coalescence process, as larger particles require a longer measurement time and the FBRM can only scan a limited number of particles per measurement for each time step. However, this is inconsistent with the continuous decrease observed in chord length measurements. As discussed earlier, cohesive forces between hydrate particles are negligible in water-continuous systems (Aman et al., 2012). Therefore, the observation of a consistent decrease in the number of chord counts at a relatively early stage of hydrate formation suggests that while particles still existed in the water phase they were not all detected by the FBRM probe. 4.4 Discussion The FBRM CLDs were used to estimate the total hydrate-water surface area, assuming spherically shaped hydrate particles. The measured mean chord lengths were employed to estimate the surface area of an average particle, and then chord counts distributions were integrated numerically to obtain the total number of particles and, ultimately, the total hydrate surface area. This estimate was not intended to be rigorous but rather to help illustrate and interpret the acquired FBRM data and particularly the evolution of distributions such as those plotted in Figure 23. We note that the relation between FBRM chord length measurements and actual sizes of hydrate particles in oil systems has been discussed by Boxall (2009). However, as this work mainly discusses the qualitative evolution of particle distributions (both in number and size) without focusing on the absolute values, those corrections have not been used in this work. Furthermore, additional calibrations would be required for those correlations to be applicable to the present measurements. In this context, Figure 24 is presented to summarise the evolution of hydrate surface area as a function of equivalent hydrate volume % for all experiments performed (100 to 700 RPM). A consistent behaviour was observed for each experiment in terms of the surface area increasing to a maximum followed by a decrease to a plateau: the hydrate volume % (φ) at which the maximum point and plateau were achieved are labelled φ 1 and φ 2, respectively. Table 8 compares φ 1 and φ 2 with the value of φ transition estimated based on the resistance-to-flow data. While φ transition ranged from about 14 to 18 vol %, with any dependence on shear rate partially obscured by the measurement s noise, φ 1 varied from approximately 1.6 to 9.1 vol % for 100 to 700 RPM, respectively, while φ 2, was about 10 vol % for experiments at RPM, and about 20 % for experiments at RPM. 61

79 φ 1 φ 1 φ 2 φ 2 φ 1 φ 1 φ 1 φ 2 φ 2 Figure 24. Calculated hydrate surface area based on the FBRM chord lengths data as a function of hydrate volume % for six gas-water experiments at impeller speeds of 100 to 700 RPM. Table 8. Hydrate volume % at which the peak (φ 1 ) and the plateau (φ 2 ) were achieved in the plots shown in Figure 24. Also listed are the values of φ transition determined from motor current data in the same experiments as described in Section 3.1 and reported in Table 7. Exp. RPM φ 1 φ 2 φ transition bounds Estimated Based on Motor Current Data

80 The observed evolution in the CLDs measured with the FBRM is graphically summarized in Figure 25 and could be interpreted as follows: During the early stages of hydrate formation, particles were distributed homogenously in the water phase, so the FBRM measurements provided a representative sampling of the aqueous phase. With continued hydrate growth, the number of solid particles in the liquid phase became sufficient to enable particle interaction and this coincided with the maximum identified in Figure 25 at φ 1. We hypothesise that this turning point in surface area represents the actual onset to a heterogeneous particle distribution in the aqueous phase, albeit at a level that has negligible effect on the macroscopic resistance to flow. At the highest shear rate tested (700 RPM) a homogeneous hydrate slurry was maintained at hydrate percent up to 9 vol %, suggesting that the shear force exerted by the continuous water phase influences the point at which interparticle interaction becomes sufficient to affect the spatial distribution of particles. Above this threshold, particles are essentially removed from the FBRM s field of view because, for example, they are buoyant and collect at the gas-water interface, and/or begin depositing on the cell wall or impeller blades. A general comparison between the value of φ 2 and φ transition suggests that the point at which the FBRM-measured surface area of hydrate particles stopped decreasing and levelled off roughly corresponds to the point where the heterogeneous particle distribution begins to affect the slurry s resistance-to-flow. At lower shear rates ( RPM), the value of φ 2 determined from measurements at the micron length scale appears to occur slightly before the value of transition derived from macroscopic measurements, whereas at higher shear rates ( RPM) φ 2 occurred slightly after transition. The mechanism discussed in Figure 25 is based on, and consistent with, the behaviour observed by Joshi et al. (2013a), Akhfash et al. (2013) and Aman et al. (2015a), which were all based on macroscopic measurements of pressure drop in a flow-loop or motor current/torque in a visual autoclave. The FBRM data reveal that hydrate bed formation occurs as a process rather than having an onset at a point: the improved resolution indicates that hydrate bed formation is initiated at lower hydrate volume percent (2-9 vol %) than previously considered. Potentially, resistance-toflow measurements may only be able to resolve the point at which a moving hydrate bed has fully formed in the system. The results suggest that risk management strategies for high water cut systems may need to include an additional level of conservatism in estimating the maximum tolerable hydrate volume fraction, as particle interaction and bed formation may be initiated at lower hydrate fractions than detectable from (increases in) flowline pressure drop. 63

81 φ 1 φ 2 Figure 25. Schematic representation of hydrate particle distribution evolution in water-continuous systems based on FBRM, pressure (this work) and visual measurements (Akhfash et al., 2013). Three regions of homogenous, heterogeneous and plugging conditions are considered; conceptual diagrams are shown for each stage to qualitatively represent the distribution of hydrate in the autoclave. In the autoclave schematics, white, green, and blue colors correspond to the hydrate, gas, and water phases, respectively. (For interpretation of the references to color in this figure legend, refer to the web version of this thesis.) After each FBRM measurement (every 10 s) prior to φ transition (as determined from the motor current data), the mean separation distance between hydrate particles was estimated using a simplified approach that combined the FBRM data (mean chord lengths) with the measured hydrate volume %. By assuming that the particles were spherical and distributed homogeneously in the aqueous phase, the chord length data acquired at each time step was used to calculate the average volume for each particle. The total number of particles, N, was estimated by dividing the total volume of hydrate by this average particle volume. Finally, the total volume of the system (hydrate + water) was divided to N-volume elements. In this idealisation, each volume element was occupied by one hydrate particle at the centre, enabling estimation of the average distance 64

82 separating two hydrate particles. This analysis was used to compare the calculated average separation distance between particles with the reported values of φ 1 in Table 8. That is, this estimation of average particle separation distance was used to assess whether the interpretation of φ 1 as being the onset of particle interaction was reasonable. Figure 26 shows the results for the estimated edge-to-edge distance between particles. The hydrate volume % at which the chord counts decreased (the values of φ 1 reported in Table 8) correspond to hydrate particle separations in Figure 26 of 380, 395, 295, 460, 210 and 150 microns for experiments 2 to 7, respectively. The corresponding CLD-based average particle sizes at those hydrate volume % (φ 1 ) were measured to be 184, 240, 190, 260, 195 and 180 microns for experiments 2-7, respectively. The ratios of these values ranges from 2 at the lowest shear rate of 100 rpm, to 1.6 at rpm, to 0.9 at the highest shear rate of 700 rpm. This qualitatively supports the interpretation that the onset of particle interaction starts to occur when the length scales of particle size and separation become comparable. φ 1 φ 1 φ 1 φ 1 Figure 26. Calculated edge-to-edge distance between solid particles in the water phase based on a simplistic model for experiments 2 to 7 corresponding to mixing speeds of 100 to 700 RPM. (For interpretation of this color figure, refer to the web version of this thesis.) Our observation of a particle number concentration limit for the homogeneous state of a solidliquid suspension is also consistent with the relevant literature. Studies of Einstein s relation for dilute solid particle suspensions indicate that particle interaction starts occurring at around 5 vol%; particle collision and aggregation, together with shear rate should be considered for concentrated and non-dilute suspensions beyond this point (Einstein, 1906, Roscoe, 1952, Brinkman, 1952, Simha, 1952). The shear rate above which a homogenous suspension is achieved for a given concentration of non-interacting solid particles can be estimated using the Zwietering correlation 65

83 (Zwietering, 1958). Several studies with non-interacting solids have shown experimentally that for particle sizes, stirred tanks and mixing speeds comparable to those considered in this work homogenous suspensions are achieved for concentrations less than 10vol % (Mak, 1992, Stevenson et al., 2010). In this work, we have extended such studies of solid suspensions by considering the evolution of hydrate particles, which are both interacting and less dense than the liquid phase. 4.5 Chapter Conclusions In this study, we present a new method of characterizing the onset to hydrate bed formation in a water-continuous system, through FBRM-based chord length distributions captured in a highpressure autoclave cell over a range of mixing speeds ( RPM). The results from the FBRM data provided a range of 1 to 400 microns for the chord length of entrained gas bubbles in the aqueous phase prior to nucleation. This is valuable as it can help guide first estimates of the initial gas-water surface area in water-continuous systems, and might help improve the predictions of hydrate growth rate models. The FBRM was effective in identifying hydrate nucleation, evidenced by clear changes in the measured number and size of the particles concurrent with the decrease in cell pressure. From the FBRM chord length and chord count distributions, it was possible to estimate the total hydrate surface area in the centre of the aqueous phase during the measurement. An increase in the total surface area was observed after hydrate nucleation; this surface area continued to increase until a maximum was reached between 2 and 9 vol % hydrate for 100 to 700 RPM, after which the total surface area decreased rapidly, despite the fact that hydrate continued to grow in the cell. We hypothesise that this reduction in surface area measured by FBRM may correspond to the onset of particle interactions and evolution of a moving hydrate bed, which may not be detected in macroscopic (resistance-to-flow) measurements until hydrate volume percent in excess of 14 vol% are reached. An increase in the measured motor current was detected at hydrate volume percent above 16 2 vol %, which has previously (Joshi et al., 2013a, Akhfash et al., 2013, Aman et al., 2015a) been interpreted to correspond to the onset of spatially heterogeneous hydrate distribution and hence hydrate bed formation. Around these hydrate concentrations, the FBRM chord length signal reached a plateau at 10 vol % for low shear rates ( RPM) and 20 vol % for high shear rates ( RPM), which may correspond to the primary growth of a moving hydrate bed with minimal particle entrainment in the bulk liquid phase. In general, the FBRM micron-scale measurements suggest that the onset of a moving hydrate bed may occur at much lower hydrate concentrations than those inferred from macroscopic measurement techniques (i.e. resistance-toflow measurements). 66

84 A simplified model was used to estimate the mean particle distance between hydrate particles in the system by integrating the FBRM chord length distributions with the measured hydrate volume %. The results indicate the onset of a moving hydrate bed detected by FBRM corresponded to the point at which the average particle separation distance was comparable with the average particle size, with the ratio varying from 2 to 1 as the shear rate increased. 67

85 5 Gas Hydrate Plug Formation in Partially-Dispersed Water- Oil Systems 5.1 Introduction Significant effort has been expended over the past decade to identify the critical path of hydrate plug formation in systems where crude oil is the dominant phase. Turner et al. (2009b) presented a four-step mechanism (Figure 27) to describe hydrate plug formation in oil-continuous systems: (i) water droplets are emulsified in the oil phase; (ii) hydrate nucleation at the interface between emulsified water droplets and oil results in the formation of a hydrate shell with an interior water core; (iii) the cohesive interaction between hydrate shells/particles results in the formation of multi-particle aggregates; and (iv) the resultant increase in apparent viscosity decreases the velocity of flow, enabling localized particle build-up and jamming-type failure. Figure 27. Conceptual picture of hydrate plug formation mechanism in oil-continuous, adapted from Turner et al. (2009b). The required emulsification of water droplets in a liquid hydrocarbon phase was studied extensively by Boxall et al. (2012), who developed a predictive model for mean droplet size based on flow-loop and autoclave measurements. The model balances the shear stress exerted by the continuous phase on the droplet against interfacial restorative forces (Janssen et al., 1994) with consideration for two distinct regimes: (i) inertial breakup, where droplet diameters exceed the length scale of the smallest turbulent eddies in the system; and (ii) viscous breakup, where droplets are smaller than those turbulent eddies and breakup is governed by sub-eddy scale viscous stresses exerted by the continuous phase on the interface. The mean droplet size predicted by both breakup modes is directly proportional to continuous phase density and inversely proportional to the wateroil interfacial tension (Boxall et al., 2012). When compared to typical crude oil properties 68

86 (Sjöblom et al., 2010), condensate-type hydrocarbon liquids have lower densities, lower viscosities and higher interfacial tensions with water. When compared to crude oil, these condensate properties result in the generation of small water droplet sizes during mixing, with a greater amount of total energy required to maintain the dispersion compared to a crude oil. For these reasons, the maintenance of emulsified water droplets in a condensate fluid generally requires substantial and continuous kinetic energy input (i.e. highly turbulent conditions). Under lower turbulence conditions, or at moderate-to-high water cuts, condensate fluids tend to exhibit partially-dispersing behavior through the formation of a separate, free water phase. The oil-continuous conceptual model for hydrate formation (Figure 27) assumes gas hydrate will nucleate at the interface of the emulsified water droplets in the oil, where both water and gas molecules exist in substantial quantities (Walsh et al., 2009). After nucleation, hydrate will spread across and form a shell around the water droplet surface at a rate that depends on guest size (Aman et al., 2011), due to the availability at this interface of both water and dissolved hydrocarbon species. The thickness of the initial hydrate shell was reported by Taylor et al. (2007) to be on the order of 50 microns, which has recently been confirmed via magnetic resonance measurements by Haber et al. (2015). Continued growth of the hydrate shell is limited by the diffusion of either guest or water molecules across the crystal lattice (Davies et al., 2010). However, in a partiallydispersing system, it is currently unclear whether hydrate will nucleate and grow primarily by forming shells around the limited water dispersed in the oil phase, or instead within the continuous water phase where the availability of guest species is limited by the solubility and diffusion of light gases in the aqueous phase (Skovborg and Rasmussen, 1994). Answering this question is one of the primary motivations for the present investigation. In water-continuous systems, Joshi et al. (2013b) suggested the primary mechanism of hydrate plug formation was related to the formation of a moving hydrate bed at the water-gas interface. This bed functions to retard the local fluid velocity, resulting in the buildup of hydrate particles and eventual formation of a stationary bed that simultaneously restricts the flow channel and enables wavy- or slugging-type hydrodynamic behavior (Hernandez, 2006). Akhfash et al. (2013) and Aman et al. (2015a) confirmed this interpretation using a high-pressure sapphire autoclave, which allowed direct observation of the transition in hydrate particle distribution throughout the aqueous phase from homogeneous to heterogeneous, whereupon a hydrate bed formed. These visual observations regarding the change in spatial distribution of the hydrate particles occurred concurrently with the first significant increase in the motor torque required to continue mixing the hydrate slurry at constant rotational speed above its baseline value. Small quantities of hydrate particles (e.g. < 3 vol % of the liquid phase) have been demonstrated to have a negligible impact on the apparent viscosity of either the oil (Sinquin et al., 2004) or water 69

87 (Joshi et al., 2013b, Akhfash et al., 2013). However, as hydrate continues to grow, interparticle and particle-wall interactions respectively enable the formation of large agglomerates and wall deposits. Aman et al. (2011) described a model for cohesive interactions between cyclopentane hydrate particles in a continuous oil phase, which includes contributions from both capillary bridge cohesion (enabled by the presence of unconverted water) and interparticle growth/sintering. Further studies revealed that the cohesive force between cyclopentane hydrate particles in a continuous aqueous phase was approximately a factor of three less than the cohesive force in a continuous oil phase (Aman et al., 2012); this result was invoked by Joshi et al. (2013b) to interpret the results of large-scale flow-loop experiments. Although hydrate-wall adhesive forces have not been measured in systems with a continuous water phase, studies by Aspenes et al. (2010) of oil-continuous systems demonstrated the important role of capillary bridges to the adhesive force mechanism. Hydrate-wall sintering was subsequently observed by Aman et al. (2013a) with calcite and quartz surfaces in liquid cyclopentane, suggesting a secondary mechanism by which hydrate particles may deposit on the flowline wall. By analogy with the observed cohesive interactions, it is likely that hydrate-wall adhesion forces may be significantly reduced by the presence of an aqueous phase. In partially-dispersing systems, the primary phase in which hydrate formation occurs will inform whether particle agglomeration/deposition or moving bed formation mechanisms control resistance-to-flow behavior. Sohn et al. (2015) reported autoclave torque (resistance-to-flow) measurements of mixed gas hydrate slurries in mineral oil at % water cut at the same rotational velocity; the study reported the most severe torque increases were associated with systems containing water cuts of 40 and 60 %, where stratified oil and water phases were likely to be stable in the blind autoclave. Laboratory experiments capable of decoupling these steps are required for the development of a generalized conceptual model that describes both fullydispersing and partially-dispersing systems. Such laboratory-based studies would help improve the capacity of risk-based hydrate assessment in partially-dispersing systems, calculations of which currently assume a behavior identical to fully-dispersing systems. This work thus seeks to address two knowledge gaps. First, do partially-dispersing systems follow the same pathway to hydrate plug formation as oil-continuous systems, or are different phenomena observed? Second, can hydrate growth in partially-dispersing systems be reliably estimated with a kinetic-type equation (Davies et al., 2009), such as is used for fully-dispersed systems? To help achieve these objectives a sapphire visual autoclave was used to identify the correct order in which the stages of hydrate plugging occur in partially-dispersing systems, and to monitor hydrate growth rate as well as the resulting resistance to flow of the system during those stages. 70

88 5.2 Materials and Methods Apparatus A simplified sketch of the high-pressure visual autoclave (HPVA) apparatus is shown in Figure 28. The basic experimental setup was explained in detail in Section on page 8. The HPVA apparatus consisted of a 25.4 mm internal diameter, 150 mm long sapphire cell, containing an MRK Mini magnetically-coupled mixing shaft connected to a ViscoPakt Rheo-57 motor. This magnetic coupling and motor constitute upgrades to the system described in Section The ViscoPakt motor allowed the amount of torque required to maintain constant mixing velocity in the cell to be measured. The motor was capable of producing mixing speeds up to 1800 RPM, with a tolerance in practice of ± 1 RPM. The cell s contents were mixed by a four-blade vane-andbaffle geometry impeller, which was shown by Akhfash et al. (2013) to provide adequate mixing while maintaining a stratified gas-liquid interface. Cell and bath temperatures were measured with 100 Ω platinum resistance thermometers (PRT) with a resolution of ± 0.2 C, while cell pressure was monitored by an Omegadyne transducer (0-345 barg) with a resolution of ± 0.1 bar. Signals from the PRTs, pressure transducer, and torque sensor (up to 57 N cm with 0.04 N cm resolution) were recorded by a LabView data acquisition system at 30 second intervals over the course of the experiment. The autoclave cell and gas manifold (120 bara rating) were submerged in a waterglycol bath (1:1 by volume), which contained a ThermoFisher immersion cooler that operated continuously to remove heat. The bath was fitted with an 1100 W electrical cartridge heater, which was activated intermittently by a LabView control algorithm to maintain cell temperature within a tolerance of 0.2 C. The amount and particularly the location of hydrates within the cell, as well as their evolution with time, were recorded by a digital video camera mounted outside the bath. Figure 28. Simplified schematic (left) and picture (right) of the high-pressure visual autoclave (HPVA) cell. 71

89 5.2.2 Experimental Procedure Paraffin oil (specific gravity 0.85, viscosity 7.5 cp, Sigma-Aldrich) was used as the partiallydispersing oil. Simple bottle stability tests (Schramm, 1992) were conducted by following the emulsification procedure of Sjöblom et al. (2010) to establish that the paraffin oil did not form a stable emulsion with deionized water at pressures above the methane saturation point at room temperature. The composition of the paraffin oil is provided in Table 9, indicating an alkane distribution from C 14 to C 21 with a peak at C 17. Table 9. Approximate composition of the paraffin oil (at atmospheric pressure) used in HPVA experiments, from gas chromatography/mass spectrometry analysis. Mineral oil component Mass fraction nc nc nc nc Pristane (C 19 alkane isomer) 0.02 nc Phytane (C 20 alkane isomer) 0.06 nc nc nc Hydrate was formed in the HPVA apparatus according to the following procedure: (1) The cell was rinsed using the solvents toluene, ethanol and acetone in sequence for 60 seconds each, and dried with compressed air. (2) The cell was then loaded to approximately 17 vol % with the liquids to be studied (deionized water and paraffin oil at a pre-determined ratio). This liquid level was chosen to fully submerge the impeller blades, but not lead to the creation of an unmixed liquid head. (3) The cell was filled and vented thrice with high-purity methane gas ( %) at 20 bara to remove any oxygen and nitrogen from the system. (4) The cell was then pressurized to the initial set-point (e.g. 65 bara) and mixing was started at a pre-defined rotational speed (e.g. 300 RPM) outside the hydrate equilibrium region. The bath was maintained above the hydrate equilibrium temperature to ensure complete saturation of the paraffin oil occurred prior to cooling and hydrate formation. 72

90 (5) Experiments began with initial cell pressures of bara and temperatures of C, after which the system was cooled to 1.3 ± 0.3 C at an intended rate of approximately 2.5 C/hr. The cooling was always sufficiently slow that the liquid phase remained saturated with gas before any hydrate formation started. This was consistent in all experiments presented here. (6) After reaching the target temperature, experiments typically required 12 hours to reach complete conversion and/or steady state. At the conclusion of the experiment, the cell was rapidly heated (4 C/hr) to the initial temperature to confirm pressure recovery and enable closure of the mass balance. Hydrate nucleation was typically observed within 30 to 200 minutes of induction time, and was identified by a slope change in the pressure signal; hydrate formation was independently confirmed by visual observations of the cell. The hydrate volume fraction present was calculated from the change in cell pressure with time assuming a constant hydration number of 5.75 (Sloan and Koh, 2007). Phase densities for deionized water, paraffin oil (treated as a single pseudo component) and hydrate were calculated using Multiflash 4.2 using the CPA model set (Infochem, 2012). A summary of the experimental conditions is provided in Table 10. Table 10. Summary of the experiments conducted on partially-dispersing (PD) and fullydispersing (FD) systems of deionized water (W) and paraffin oil (O). Exp. P initial bara T initial C RPM Dispersion Status Initial Reynolds number PD W=2338, O= FD FD PD W=2338, O= FD FD PD W=2338, O= FD FD PD W=2338, O= PD W=3896, O= FD PD W=2338, O= PD W=3896, O= FD 217 Watercut vol%

91 The experiments listed in Table 10 were classified as either partially-dispersing (PD) or fulldispersing (FD), based on visual observation of the phase distributions in the sapphire cell; if a stratified phase boundary with a length scale of the cell s diameter was observed between the mineral oil and water at the point of hydrate nucleation, the system was characterized as partiallydispersed. To aid in this visual identification, a red, oil-soluble Sudan III dye was added to the paraffin oil. This classification system was applied to all of the experiments at each rotational speed for which a total of five water cuts (10-70 vol % of the liquid phase) were studied. In PD systems, separate water and oil phases remained after up to 15 hours of mixing in the autoclave cell. All of the experiments conducted at 300 and 900 RPM were found to be PD and FD, respectively, independent of water cut. Consequently, the experiments performed at 500 RPM provided particularly valuable information, as at this speed water cuts of 10-30% were FD while water cuts of 50-70% were PD. Table 10 also contains estimates of the initial Reynolds number (Re) of the fluids calculated at 20 and 63 bara (a common initial state) for each mixing condition. For FD cases, Re was calculated by assuming the mixture was a homogenous water-in-oil emulsion, and was thus dominated by the density and viscosity of the oil phase. In PD cases, separate Re calculations were made for both the water and oil phases, as listed in Table 10. The value of Re was calculated using the expression for stirred tanks (Naumann, 2008), as discussed in eq. (1). In FD systems, oil and water were assumed to form a homogeneous liquid mixture with an average density based on the volumetric ratio of each component. The viscosity of the FD mixtures was calculated based on the Pahn-Thien and Pham viscosity model for emulsions (Phan-Thien and Pham, 1997), where the oil phase was assumed to be continuous. The water cuts reported in Table 10 were calculated at ambient pressure and room temperature, when the cell was loaded, and correspond to the volume fraction of water relative to the total volume of liquid (oil + water). Once hydrate formed, the hydrate volume fractions (reported below) were also calculated based on the total volume of condensed phases (hydrate + oil + water) present. 5.3 Results and Discussion Visual Observations Figure 29A provides an example of the homogeneous water-in-oil dispersion present prior to hydrate formation, for experiment 10 (30 % water cut, 500 RPM). Some limited water pooling is apparent around the baffles in Figure 29A, while the liquid in the remainder of the cell is visibly homogeneous and, therefore, this mixture was classified as a FD system. Figure 29B shows that experiment 18 (70 % water cut, 500 RPM) was a PD system prior to hydrate formation, in which water and oil phases were clearly distinct. 74

92 Figure 29. Images of mixing prior to hydrate formation captured in the sapphire autoclave (25.4 mm inner diameter) for systems at 500 RPM and different water cuts. (A) A fully dispersed system with 30% water cut. (B) A partially dispersed system with 70% water cut. Images taken throughout the formation of hydrate in experiment 18 are shown in panels C-H of Figure 30, and are representative of the observations acquired for each of the PD experiments. In addition, the full video record of experiment 18 is provided as Supplementary Information because the higher-resolution and dynamic nature of the video makes it easier to identify several of the features discussed below. Hydrate growth was first detected in the oil phase by visual observation of white particles therein (Figure 30C). Initial growth in the oil phase was observed consistently for hydrate volume percent of vol % over the entire range of PD experimental conditions listed in Table 10. Continued hydrate growth resulted in the formation of a hydrate film at the oilsapphire interface (Figure 30D), which grew downward to the water-sapphire interface (Figure 30D & E). The hydrate film is identifiable as being on the sapphire interface, because stagnant white crystalline particles can be observed in the video footage. Eventually, the amount of hydrate formed at this stage (in the range of 2-6 vol % hydrate over the entire range of PD experimental conditions listed in Table 10) was sufficient to disrupt the oil-water interface and initiate dispersion of oil into the water, as evidenced by the increasing pink hue of the water shown in Figure 30D & E. Throughout these stages, however, a distinct interface between the water and oil phase was observed to remain in place. Some degree of agglomeration between hydrate particles was also observed in the oil phase during stage E, which became more significant with continued hydrate growth. 75

93 In the latter stages of hydrate growth, the water-oil interface also moved downward (Figure 30F) toward a transition point at which the remaining liquid water became fully entrained in the oil phase. This full entrainment was initiated between 6 and 12 vol % hydrate over the entire range of PD experimental conditions listed in Table 10. The elimination of the water-oil interface and formation of a fully dispersed system eventually enabled rapid hydrate growth (Figure 30G), which led to a second and more catastrophic deposition stage (Figure 30H), with large solid white hydrate deposits appearing on the sapphire wall. The white appearance of these wall deposits suggests their structure had a relatively low porosity given that the oil phase contained a red dye. In most experiments, the observation of this much particle deposition on the wall (Figure 30H) was the final stage before the resistance-to-flow reached its maximum level that was occasionally sufficient to stop the impeller from rotating. Hydrate film growth at oilsapphire interface C D E White particles within the circle indicate initial hydrate growth in the oil phase System is still partially dispersed. Oil phase volume has not changed. Hydrate film formation at water-sapphire interface in D and E. Water getting pink due to the beginning of oil/water dispersion. F G H Further hydrate growth resulted in displacement of the oil-water interface and full entrainment of water in oil. Particle deposition on sapphire wall Rapid hydrate growth in the whole cell following the full dispersion of water in oil Figure 30. Images of the sapphire autoclave after hydrate formation at six stages throughout experiment 18 (70 % water cut, 500 RPM): (C) 68 minutes after nucleation with 3.1 vol % hydrate; (D) 100 minutes after nucleation at 5.5 vol % hydrate, beginning of the water-oil interface disruption; (E) 120 minutes after nucleation at 7.9 vol % hydrate; (F) 162 minutes after nucleation at 17.3 vol % hydrate; (G) 204 minutes after nucleation at 48 vol % hydrate; and (H) 310 minutes after nucleation with 63 vol % hydrate. (For interpretation of this color figure, refer to the web version of this thesis.) 76

94 In the context of the first knowledge gap listed above for PD systems, the visual observations shown in Figure 30 illustrate three additional phenomena that have previously not been incorporated in the conceptual framework of hydrate plug formation: (i) initial growth of particles acts to disrupt the water-oil interface; (ii) this disruption leads to migration of the water-oil interface until the PD system becomes FD; and (iii) the deposition of hydrate particles on the wall late in the experiment leads to severe fluctuations in the observed resistance-to-flow (motor torque) that are characteristic of plugging Hydrate Growth Rate The key results relating to hydrate formation and plugging for all experiments are reported in Table 11. Table 11. Summary of the results for all experiments. The final temperature of all experiments was 1.3 ± 0.2 ºC. The equilibrium temperature corresponding to the pressure at which nucleation occurred had an average of 8.0 ± 0.2 ºC. The quantity φ final represents the hydrate volume % in the liquid phase at the end of each experiment. The average number of moles of methane in the gas phase at nucleation was 0.23 ± moles. Exp. P nuc bara T nuc C P final bara T eq,final C φ final vol% Final Water Conversion % Initial Gas Consumption mmol/min Maximum Maximum Torque (N.cm) Maximum Relative Torque

95 A distribution in the pressure and temperature of hydrate nucleation was observed throughout the experiments; nucleation pressures ranged from 56.6 to 60.1 bara at temperatures from 1.2 to 6.9 C, corresponding to nucleation subcoolings between 1.2 and 6.5 C across the experiments. The average steady-state temperature achieved during the experiments was 1.3 C. For 10-30% water cut experiments, the average final water conversion was approximately 95 % with the remaining liquid water possibly having been sequestered inside hydrate shells (Turner et al., 2009b). At water cuts of 50 and 70 %, the average final water conversion decreased to 86 %; this effect may be the consequence of a hydrate plug forming at the gas-liquid interface, limiting the dissolution of methane in the liquid phase during the late growth stages. The average initial rate of gas consumption in each experiment was used as a proxy for the initial hydrate growth rate over the first 10 minutes after nucleation, with the results shown in Figure 31 for each rotational speed as a function of water cut. For systems at % water cut, there was a geometric increase in initial growth rate with mixing speed, while the growth rate in high water cut systems depended first on the mixing speed and second on the PD/FD state. These data indicate that PD systems have much lower initial growth rates in comparison with FD systems at similar shear rates. Figure 31. Initial gas consumption rate over the first 10 minutes after nucleation in each experiment for all experiments with repeat trials represented as an average value. The hydrate growth rate profiles for experiments 1 and 3 respectively 300 and 900 RPM for 10% water cut were compared to predictions of a hydrate kinetic model, with the results shown in 78

96 Figure 32. In short, this model assumes that all water is emulsified in oil at a monodispersed droplet size dictated by the correlation of Boxall et al. (2012), and hydrate growth rate is directly proportional to subcooling as explained by Davies et al. (2009). Aman et al. (2015b), Aman et al. (2015a) provide a comprehensive explanation of both the droplet size and kinetic growth component models used in these calculations, which took standard values for the kinetic and growth rate coefficients. The results demonstrate that the kinetic model agreed reasonably well with the FD experimental data at 900 RPM, and performed poorly for the PD experimental data at 300 RPM. At both shear rates, the model does a reasonable job at predicting the observed initial growth rates, which indicates that the calculated values of the number and size of the water droplets entrained in the oil phase initially are reasonable. However, at 300 RPM, the slope change in the (pressure) data that occurs approximately 30 minutes after hydrate nucleation corresponds to the point in the experiment at which complete water emulsification in the oil phase has occurred. Consequently, the model accuracy decreases after 30 minutes in Figure 32a, because the description of water-oil interfacial area (Boxall et al., 2012) does not account for the surface area increase caused by this complete emulsification; the model instead assumes that the surface area remains constant at the value set by the original (low) shear rate. Accordingly, it may be possible to improve the prediction of interfacial area by introducing a more complex description of the droplet size that accounts for the presence of a solid phase. Current growth rate models are unable to account for this phase inversion point in PD systems. Future work to improve growth rate predictions might consider coupling phase inversion predictions (Ling et al., 2014) which incorporate interfacial tension, viscosity, density, and shear rate with the droplet size model in use. Figure 32. Experimental (x) and predicted (black curves) hydrate volume % as a function of time at 10% water cut for (a) experiment 1 at 300 RPM (PD) and (b) experiment 3 at 900 RPM (FD). In contrast, at 900 RPM (Figure 32b) where the mixture starts and remains completely emulsified throughout the experiment, the observed growth rate 15 minutes after hydrate nucleation temporarily drops below that predicted by the model. This drop in the physical growth rate is 79

97 likely caused by an increase in mass transfer resistance and/or a decrease in the interfacial wateroil surface area because of the viscosification of the liquid mixture due to the formation of hydrate particle aggregates. Subsequently, around 90 minutes after nucleation, the observed growth accelerates to a rate faster than that predicted by the model, which can be attributed to the generation of additional surface area between the reacting components. This acceleration in growth is similar to that observed by Akhfash et al. (2013) in water-gas systems where the formation of a sufficient amount of hydrate led to the disruption of the interface between the water and gas phases. By analogy, the present data suggest that for FD oil-dominant mixtures, a sufficient amount of hydrate can cause disruption of the liquid-gas interface and an increase in surface area available to the reactants involved in hydrate growth. It is important to note the differences between the interfacial breakdowns that occur in PD and FD systems. In the partially dispersed mixture, initial hydrate growth at the microscopic interfaces between the oil and the limited amount of entrained water eventually disrupts the macroscopic, stratified oil-water interface, which then ultimately leads to the complete entrainment of water droplets in the oil phase producing an FD system. In systems that were originally FD, no stratified oil-water interface exists, and only much later does growth accelerate following disruption of the liquid-gas interface. From the perspective of hydrate growth in oil and gas production systems, the impact of this secondary liquid-gas disruption may be mitigated by the much greater turbulence in industrial flowlines that would supress mass transport limitations. In contrast, for production systems in which the water cut is sufficient to result in a PD system, the disruption and elimination of the macroscopic oil-water interface by hydrate particles as observed in these autoclave experiments may represent a controlling step in accurately estimating hydrate growth rate. Table 12 lists the hydrate volume % at which (i) the oil-water interface was initially disrupted ( φ ( φ (visual) disruption (visual) full-dispersion ) based on visual observation, (ii) the water became fully dispersed in the oil phase ) based on visual observation, and (iii) rapid hydrate growth commenced. The volume % at which rapid growth started ( φ rapid growth the change in the rate of pressure decrease with time, hydrate growth in cell, ) was determined by two independent measurements: 80 (press) φ, and visual observation of rapid rapid growth (visual) φ based on the time at which hydrate disruption of the liquid-gas rapid growth interface occurred unambiguously. The two independent measures are in reasonable agreement given the limitations associated with the visual observation method. In most cases, the visual observation-based measurement comes after the pressure signal but the delay is always less than 6 vol% hydrate. The hydrate volume % at which rapid growth begins based on the pressure signal is a more reliable measurement and appears to be independent of shear rate. It does exhibit a dependence on water cut, increasing from about 3 vol % hydrate at 11 % water cut to around 25 vol % hydrate at 70 % water cut. This observation suggests that at higher water cuts in PD

98 systems, a larger volume of water can be emulsified and converted into hydrate before rapid growth is detected in the pressure signal. Regardless of the mechanism by which φ rapid growth is set, the time series data presented in Figure 33 show that while the amount of time taken to reach the onset of rapid growth can vary, the value of φ rapid growth is consistent across three separate experiments at the same water cut. Table 12. Hydrate volume % at which oil-water interface disruption, oil-water full entrainment and rapid growth started in the PD systems. The superscripts indicate whether the volume % was determined visually or, in the case of the rapid growth transition, determined independently from the measured pressure. Exp. Watercut % Shear Rate RPM (visual) φ disruption vol% (visual) φ full-dispersion vol% (press) φ rapid growth vol% (visual) φ rapid growth vol% * * 2.8 * * * * * * = Experiment s video record had inadequate resolution to determine this parameter reliably. In the context of the second knowledge gap for PD systems, these results indicate that the current kinetic model for FD systems cannot be used reliably to predict hydrate growth rate in PD systems, which is initially much lower than in comparable FD systems but can rapidly increase according to the extent of hydrate growth Resistance-to-Flow Measurements In addition to visual observations (Section 5.3.1) and measurements of gas consumption (Section 5.3.2), the evolution towards and severity of hydrate plug formation may be quantified from the motor torque data collected in each experiment. Functionally, the motor torque represents a resistance-to-flow in the autoclave cell and may be considered as an analogue to pipeline pressure drop. These torque data are most useful when paired with hydrate volume fraction data, and allow for an assessment of pseudo-mechanical equilibrium through the cell as the hydrate plug forms. Figure 33 shows the evolution after nucleation of both hydrate volume % (left axis) and motor torque (right axis). It is apparent that in PD systems rapid growth precedes, and presumably leads 81

99 to, increased resistance-to-flow. However, if a sufficient volume fraction of hydrate is produced (i.e. water cut is large enough), the system s resistance-to-flow can vary significantly depending on the shear rate and whether the system was partially dispersed. In the most severe cases, large spikes and fluctuations in the measured torque are observed and signal the formation of a hydrate plug, sometimes sufficient to stop the flow altogether. Joshi et al. (2013b) and Grasso et al. (2014) conducted experiments with a 4 inch diameter recirculating flow-loop and interpreted the occurrence of rapid oscillations in the resistance-to-flow (pressure drop) data as indicative of hydrate plug formation. Figure 33. Hydrate volume % (left axis) and motor torque (right axis) as a function of time at 70% water cut (experiments 17-20).(For interpretation of this color figure, refer to the web version of this thesis.) In Figure 34, the temporal abscissa is eliminated and torque is plotted as a function of hydrate volume % for each of the partially-dispersing systems studied at 300 RPM. The measured torque showed a minimal increase for hydrate percent of up to 30 vol %. At the lower water cuts, this amount of hydrate represented near full conversion of the water phase and no hydrate plug was formed. In contrast, as the 50 and 70% water cut systems approached 40 vol % hydrate, substantial increases in torque were observed, followed by rapid oscillations and plugging-type behavior. The repeatability in the torque readings obtained for all five water cuts below 30 vol % hydrate suggest that systems below this threshold have a reduced plugging risk. 82

100 Figure 34. Torque (resistance-to-flow) as a function of hydrate volume % over 10-70% water cut at 300 RPM, where all experiments were classified as partially-dispersing. The severe torque fluctuations observed from vol % hydrate are associated with plugging-type behavior. (For interpretation of this color figure, refer to the web version of this thesis.) The limited increase in torque associated with experiments below 30 % water cut supports the hypothesis of Sjöblom et al. (2010) regarding the associated minimal hydrate plugging risk for such systems whether they are FD or PD; this was also apparent from the autoclave experiments reported recently by Sohn et al. (2015). Figure 34 demonstrates, however, that the onset of severe hydrate plugging behavior occurs in PD systems with amounts of water sufficient to generate large volume fractions of hydrate. The onset of a significant increase in torque (from the baseline) was observed around 20 vol % hydrate in the 50 and 70% water cut experiments and coincided with the value of φ rapid growth derived from both pressure data and visual observations. The coincidence of this initial increase in resistance-to-flow with φ rapid growth suggests an analogue to the bedding-type transition behavior observed by Joshi et al. (2013b) and Akhfash et al. (2013) in 100 % water cut systems. The subsequent sharp increase in torque observed above 40 vol % hydrate visually corresponded to hydrate build-up in the bulk gas as the result of that catastrophic growth (Figure 30G), followed by particle deposition (Figure 30H) and plugging-type fluctuations. As illustrated by Figure 30D & E, hydrate deposition on the sapphire wall in the vicinity of the oilwater interface were crucial to its disruption and the subsequent rapid growth phase. The effect of increasing shear rate in experiments with 70 % water cut (Figure 35), where 300 and 500 RPM experiments were PD and the 900 RPM experiment was FD, indicates that such deposition also 83

101 plays a critically important role in contributing to the increase in the measured resistance-to-flow behavior. The final hydrate volume % in these three experiments ranged from %; however the 900 RPM FD system did not exhibit an initial increase in torque as large as the PD systems, nor did it exhibit plugging-type fluctuations. As all three experiments were performed at the same water cut, an increase in the mixing speed functionally increases the shear stress being applied to the wall during hydrate growth. In the 900 RPM experiment, the severity of wall film growth and deposition was visibly reduced, suggesting the shear stress was sufficient to prevent particle buildup at the wall. Figure 35. Torque (resistance-to-flow) as a function of hydrate volume % for 70 % water cut systems at 300 (black, PD), 500 (grey, PD, 2 repeats), and 900 (red, FD) RPM. The inset panel contains the same data with a magnified ordinate range, to enable better comparison of the torque curves prior to the plugging-type fluctuations.(for interpretation of this color figure, refer to the web version of this thesis.) At 900 RPM, all experiments were classified as FD over % water cut. Although the maximum resistance-to-flow (torque) was up to a factor of 3 times larger at hydrate volume % above 30 vol % than at nucleation, the high shear stress present at the wall was sufficient to prevent substantial deposition, and thus large torque fluctuations (Figure 36). While some fluctuations in torque were observed at hydrate volume concentrations above 40 %, the maximum torque obtained in each 900 RPM experiment was approximately 10% of the maximum torque obtained in the corresponding experiments at lower mixing speeds. These data further support the 84

102 role of shear stress exerted by the flow in minimizing the extent and severity of hydrate film growth and deposition on the pipeline wall. Of course, this insight must also be paired with the heuristic of Sloan and Koh (2007), which notes that an increase in fluid velocity will cause a competing increase in the hydrate growth rate; that is, faster flow does not always decrease hydrate plugging risk. The optimal fluid velocity in subsea flowlines from the point of view of mitigating hydrate risk when inside the equilibrium boundary may be considered as a balance between hydrate growth rate in the continuous phase (Davies et al., 2009) and decreased risk of timedependent hydrate build-up at the wall (Lorenzo et al., 2014). Figure 36. Torque as a function of hydrate volume % for % water cut at 900 RPM, at which speed all systems were fully dispersing. The maximum torque values obtained for all experiments are shown in Figure 37 as a function of water cut. This result further supports the hypothesis of Sjöblom et al. (2010) that systems below 30 % water cut represent a low risk of hydrate plug formation. Furthermore, the data also suggest that there is little difference between PD and FD systems for low water cut conditions. However, for PD systems at 50 and 70 % water cut, the maximum torque observed in each experiment was at least a factor of 10 greater than the baseline value due to hydrate formation, and several times larger than the maximum torque values observed at higher shear rates. Clearly, if a sufficient amount of hydrate is able to form in a PD system, the risk of hydrate blockage is increased at high water cuts relative to FD systems. 85

103 Figure 37. Maximum torque achieved during hydrate formation as a function of initial water cut in the system, at the three different mixing velocities tested. All 300 RPM experiments and those with 50-70% water cut at 500 RPM were classified as PD. 5.4 Chapter Conclusions The severity of, and mechanisms by which, hydrate plug formation occurs in partially-dispersing oil and water systems was investigated using a high-pressure sapphire visual autoclave, at water cuts of % and mixing velocities between RPM. The results demonstrate that partially-dispersing systems do not increase hydrate growth rate or blockage severity at water cuts below 30%, and increase the severity of hydrate blockage by an order of magnitude at water cuts of 50-70%. Together with the visual observations summarized by Figure 30, the hydrate growth and resistance-to-flow (torque) data suggest that a revised conceptual mechanism is needed to describe hydrate plug formation in partially-dispersing oil systems. Our current hypothesis of the mechanism by which plugs form in high water cut PD systems is shown in Figure 38, and builds upon the hypothesis for oil-continuous systems developed by Turner et al. (2009b) in collaboration with J. Abrahamson. Figure 38 includes three new primary contributions that are supported by the experimental data and visual observations presented here: (i) in PD systems, initial hydrate particle growth leads to the disruption of the water-oil interface; (ii) continued disruption of the interface leads to full dispersion of the water phase and rapid hydrate growth; and (iii) particle deposition on the walls following the full dispersion leads to plugging. 86

104 Figure 38. Proposed conceptual mechanism for hydrate plug formation in partially-dispersing oil and water systems, based on the results presented in this work. This mechanism is adapted from that proposed by Turner et al. (2009b) with additional stages observed here for PD systems shown in red text. (For interpretation of this color figure, refer to the web version of this thesis.) Two of the component elements of this extended conceptual mechanism wall film growth and particle deposition on the wall are likely to also apply in fully dispersed oil-continuous systems. However, to our knowledge this has not been previously observed. In this respect, the visual autoclave is a particularly powerful tool, because it provides additional information to help interpret the growth and resistance-to-flow data. Nevertheless, there are still limits to the insight achievable currently; for example, it is difficult to identify in the PD system how hydrate particles disrupt of the oil-water interface. Some observations suggest that it may be the result of a bed of hydrate particles accumulating at the oil-water interface acting in concert with the growing wall film. Similarly, following full dispersion we speculate that a moving bed of hydrate aggregates might be the precursor to the observed particle deposition. However, further investigation is clearly required to confirm the existence or otherwise of this moving bed. 5.5 Supporting Information The video record of Experiment 18, from which the snapshots shown in Figure 30 are taken, is available online as Supporting Information. 87

105 6 Conclusions and Future Work 6.1. Conclusions The main objective of this PhD research was to develop a detailed description of conceptual plugging mechanisms in both water dominated and partially-dispersed systems, by systematically measuring hydrate particle distributions (both spatial distribution and size distribution), resistance to flow, pressure, and temperature in specialized autoclaves. The review of the existing literature describing mechanisms of hydrate plug formation in multi-phase systems revealed a wellstablished plug formation trajectory for oil-continuous systems. However, in late-field life production, where the volume of produced water is high, the continuous phase is frequently water. In the published open literature, only one study has addressed plug formation behaviour in water dominated systems (Joshi et al., 2013a) and none have proposed a picture for plugging in partiallydispersed systems, where both oil and water continuous phases are present. This thesis has delivered several valuable and new insights into hydrate inter-particle interactions and plug formation in such systems. To obtain the data necessary to deliver such insights, a sapphire visual autoclave with 1 inch internal diameter (ID) and the ability to measure both resistance-to-flow and hydrate growth rate was designed and used initially to study hydrate formation in methane-water systems. Visual observations described in this thesis provided direct confirmation of the conceptual picture previously hypothesized in the literature based on flow-loop studies (Joshi et al., 2013a): that plug formation behaviour in gas-water systems is dominated by the formation of a moving hydrate bed and a transition from a homogeneous to a heterogeneous particle distribution, which appears to occur at a specific hydrate volume fraction, φ transition. Hydrate particles had almost no effect on the mixture s resistance-to-flow as long as they were uniformly distributed in the liquid phase, confirmed by virtually no increase in motor current data. Upon the onset of a heterogeneous distribution of particles, where they began concentrating in some sections of the cell, the first measurable increase in motor current was observed. The value of φ transition, where this occurred was found to be in the range of 15 to 20 vol %. This increase then was followed by severe fluctuations in motor current as additional hydrates formed in the cell, indicating jamming- and plugging-type behavior. At moderate shear rates, the observed transition in hydrate particle distributions was also corresponded to an obvious increase in hydrate growth rate. At φ transition, buoyant hydrate particles began accumulating at the gas-water interface and forming a moving hydrate bed (as per the conceptual picture in Figure 1 on page 7), interrupting the interface and producing additional surface area. Resistance-to-flow patterns in the autoclave were consistent with those observed in flow-loops, where motor current may be considered as analogous to the pressure drop signature. 88

106 Adding 10 wt% of a thermodynamic hydrate inhibitor (MEG) to the aqueous phase revealed three important observations about under-inhibited scenarios, compared to pure water cases at same flow conditions. First, while no noticeable effect on the initial hydrate growth rate was observed for such MEG concentrations, the amount of hydrate formed at steady state was decreased as the result of self-inhibition. Second, MEG delayed the onset of hydrate bed formation to higher volume % by increasing the value of φ transition by about 10 vol %. Third, MEG resulted in slushtype hydrates and consequently decreased the maximum resistance-to-flow achieved in the system to about 20 % of the magnitude observed in un-inhibited cases. These results suggest further study is required for under-inhibition strategies as they could be advantageous to production and transportation in high water cut systems. To further study the moving hydrate bed that forms at the methane-water interface and to quantify the φ transition dependence on Reynolds number, additional experiments were conducted in the 1 inch visual autoclave, in both continuous cooling/flow and shut-in/restart operating procedures, and under different turbulence conditions. The same sequential stages for hydrate plug formation were observed visually as those mentioned above. However, as the result of conducting a large number of trials, an additional stage of hydrate wall film growth prior to bed formation was identified in most experiments. The formation and growth of this wall film might explain the irreversibility of φ transition with Reynolds number, a phenomenon which has been observed in flow-loop studies (Joshi, 2012). In both constant cooling/flow and shut-in/restart experiments, higher mixing rates shifted the values of φ transition to higher hydrate volume %. No major difference in φ transition was observed between the two different operating modes when the shear rate used was the same, which indicates that both modes undertook similar paths to hydrate plug formation. However, initial hydrate growth rates, as determined for the first 2 vol% of the hydrate formed following nucleation or the restart of shear, were higher by an order of magnitude for cold restart cases, due to the presence of a greater hydrate-water surface area. In both types of the experiments, while sufficiently high shear rates decreased the maximum resistance-to-flow achieved in the system and delayed the onset of hydrate bed formation, it promoted hydrate growth rate. Trading-off between these competing aspects can be critical in the development of operation strategies and the risk management of hydrate plugs in water-dominated systems. For the experiments conducted with the 1 inch sapphire autoclave, values of 6.6 to 40 vol% hydrate were observed for φ transition over a range of initial Reynolds numbers of 240 to 4500 (corresponding to RPM). These values of φ transition, which were identified from an increase in the measured resistance to flow and supported by visual observations about the spatial distribution of the particles, quantitatively agree with other autoclave and flow-loop observations (Sum et al., 2012, Joshi, 2012); however they were inconsistent with regards to the Reynolds number dependence of φ transition as the Reynolds numbers in the 1 inch sapphire autoclave are 89

107 significantly smaller.. This reveals the need for another method by which to scale φ transition thresholds between various flow geometries and within similar geometries with different internal diameters. To this end, and to better understand the evolution of the moving hydrate bed in gas-water systems, hydrate plugs were formed in a 4 inch ID, high-pressure autoclave cell, equipped with insitu FBRM probe, at mixing speeds between 50 and 700 RPM ( Reynolds number). In addition to the macroscopic measurements of pressure drop (growth rate) and motor current (resistance to flow), the online FBRM distribution measurements enabled the successive study of size and number density of the hydrate particles in the liquid phase. Ultimately, this was used to detect the onset of particle interaction and bed formation on a micron length scale. This substantially increased resolution established that in gas-water systems, the process of bed formation and particle interaction was initiated at significantly lower particle concentrations (2-9 vol % depending on shear rate) rather than those inferred from an increase in resistance-to-flow (16±2 vol %). A simplified approach was employed to assess FBRM-based calculation of particle interaction onset by combining the FBRM mean chord lengths with the measured hydrate volume fractions. It was estimated that the onset of a moving hydrate bed detected by FBRM corresponded to the point at which the mean separation distance between hydrate particles was comparable with the average size of particles, with the ratio varying from 2 to 1 as the shear rate increased, enabling inter-particle interaction and thereby affecting the spatial distribution of particles. More significantly, the higher-resolution (FBRM) results suggested that in water-continuous systems the slurry s resistance flow begins to increase appreciably when the number of hydrate particles in the aqueous phase becomes sufficient to form a moving hydrate bed at the gas-water interface. That is, φ transition, seemed to be correlated with the point where particles in the bulk aqueous phase started to migrate out of the FBRM s field of view and towards the gas-water interface. Further, the FBRM measurements suggested a range of 1 to 400 microns as the size of entrained gas bubbles in the aqueous phase prior to hydrate nucleation. This result may provide guidance to estimates of gaswater interfacial area in water-continuous systems, and help refine the predictions of hydrate growth rate models. Finally, a comprehensive set of experiments was conducted in the 1 inch sapphire visual autoclave using 10 to 70 vol % water in mineral oil, to evaluate plugging mechanisms in partially-dispersed mixtures. Our results suggested a low risk of hydrate plug formation at low water cuts (10-30 %), similar to fully-dispersed conditions. Furthermore, for the first time, wall film growth and deposition stages also observed in fully-dispersed oil-continuous systems. However, at moderate to high water cuts (50-70 %), the severity of hydrate blockage (as measured by the maximum resistance to flow observed) was increased by a factor of 10 in partially-dispersed systems. The hydrate growth rate profiles were compared to predictions of a well-established model for oil- 90

108 dominant systems, which relies on the assumption that all the water is emulsified in the oil phase. The significant inconsistency between the model and experimental results demonstrated the inability of current models for oil-dominant systems to reliably predict the risk of plug formation in partially-dispersed scenarios. Accordingly, a new conceptual model was proposed to explain plugging behaviour in partially-dispersed systems, which was developed based on visual observations in conjunction with resistance to flow measurements and hydrate growth rate data. This new model was based on the current conceptual mechanism for oil-continuous systems, but with the addition of four critical stages: (i) Hydrate wall film growth in both the oil and water phases, (ii) Hydrate-induced oil/water interface disruption; only small amount of hydrate (2-6 vol%) was needed to initiate oil/water dispersion. (iii) Accelerated hydrate growth and full dispersion of the remaining oil and water phases; (iv) Rapid hydrate growth and particle/wall deposition, leading to plugging. Additional high-resolution visual studies are, however, needed to further clarify these stages. For example, some observations suggested the formation of a hydrate bed at the stratified oil-water interface, which might be responsible for the interface interruption. This work has confirmed particle distribution and bedding as critical factors associated with hydrate plugging in water-dominated systems, and has developed a comprehensive description of plugging in partially-dispersed oil-water scenarios. These results could help industry extend the production life of existing developments by reducing the cost of inhibition in high water cut fields. High-quality hydrate formation data such as those obtained during the course of this research allow quantitative improvements to be made to current models for hydrate growth rate as well as resistance to flow behavior as a function of shear rate and hydrate volume fraction. The data may allow correlations for the dependence of φ transition on turbulence to be developed based on estimates of particle sizing and spacing. Once validated through supplementary autoclave/flow-loop/flowline tests, such correlations could be readily used within multiphase flow simulators, to better manage the risk of hydrate plug formation in real-world operations Suggestions for future work During the course of this research, significant experimental work was performed to study hydrate plug formation in water-continuous, partially dispersed oil-water, and under-inhibited systems. The suggestions for future work below are grouped according to these classifications: Water-Gas Systems: The flow-loop data reported in the literature (Sum et al., 2012, Joshi, 2012) reported a dependence of φ transition on shear rate and proposed a linear correlation with Reynolds number. Our numerous measurements at different impeller speeds using a 1 inch ID sapphire autoclave cell also indicated a dependence of φ transition on Reynolds number - but the relation is clearly different from that derived from flow-loop observations. For the results obtained with the 4 inch ID stainless steel autoclave for the same range of impeller rotational velocities, the 91

109 dependence of φ transition on shear rate was also different from that observed for either the 1 inch ID autoclave or the flow-loop. Although the comparison of autoclave and flow-loop data [this work and others (Sum et al., 2012, Joshi, 2012)] indicates that values of φ transition qualitatively agree within both different geometries and similar geometries with various internal sizes, the differing velocity-dependence of φ transition indicates there may be a dependence on system geometry. Perhaps more significantly the results suggest Reynolds number might not be the correct basis with which to scale φ transition values between different systems. The FBRM results obtained in this work suggest that estimates of the inter-particle spacing may be a more appropriate length scale with which to translate results obtained with different apparatus. Additionally PVM images might be used along with FBRM data to provide further insights regarding the separation distance between hydrate particles. Further flow-loop/flowline/autoclave tests, preferably instrumented with FBRM and PVM, need to be designed in the context of answering the broad question of how a scaling relationship across different geometries and/or varying size systems can be established. This scaling relationship should not be limited to the dependence of φ transition on shear rate, but might also be extended to other hydrate slurry properties such as the magnitude of the flow resistance associated with a given hydrate volume fraction in a water-dominant system. Further, a mass transfer-limited model containing no adjustable parameters was employed to quantitatively predict hydrate formation rates in water-continuous systems and it was successful in describing sufficiently turbulent systems. In laminar/transition flow regimes, based on the discrepancies between the model s prediction and experimental data, it was hypothesized that before φ transition, hydrate growth is limited by the rate at which the aqueous phase can be resaturated with gas, as a result of dissolution limitations present in experiments. The model also did not consider the increased gas-water interfacial area caused by hydrate bed formation after φ transition. Future work should further test and extend the descriptions of hydrate growth in gas-water systems developed in this work across a broad range of flow conditions. In addition, the FBRM-measured sizes of entrained gas bubbles in the aqueous phase prior to hydrate formation in water-continuous systems could now be implemented into current growth models to improve their estimates of the gas-water interfacial area. It would be also valuable to obtain FBRM bubble and particle size distribution measurements in flow-loop experiments to establish the extent to which the results obtained in this work with the four inch autoclave translate to larger scales more representative of industrial production. Partially-dispersed systems: This work has identified the sequential stages of the hydrate plugging process in partially-dispersed systems. The current kinetic model did an acceptable job at estimating the observed hydrate growth rates in fully-dispersed oil systems, which indicates that the calculated values of the number and size of the water droplets entrained in the oil phase were 92

110 reasonable. However, the model was unsuccessful in estimating the proper oil-water surface area after the dispersion inversion occurred in partially-dispersed systems, when complete water emulsification in the oil phase occurred following the initial hydrate growth. Thus, future work should focus on integrating the newly developed conceptual mechanism into growth rate models with improved estimates of the oil/water interfacial area both prior to nucleation and in the presence of a solid hydrate phase. Such models could couple phase inversion predictions available in literature (Ling et al., 2014) - which integrate interfacial tension, viscosity, density, and shear rate - with the current model describing droplet size as a function of shear. Future work could also help clarify the limits of the results obtained in this work. For instance, it is still uncertain how the presence of a solid hydrate phase destabilizes the stratified oil-water interface in the autoclave. The video records obtained in this work suggest particle bedding at the oil-water interface along with initial wall film growth as possible mechanisms. Future work could aim to elucidate the role of those two factors, possibly by improving the video capture system s resolution to obtain sufficiently detailed images. Under-inhibition: Few laboratory data are currently available to describe the effect of THIs on hydrate formation and transportability in under-inhibited conditions. This work has showed an increase in φ transition by about 10 vol % could be achieved by adding 10 wt % MEG to the aqueous phase. Future work could focus on understanding the physical basis for this increase in φ transition. It would also be valuable to perform a combination of visual observations (the extent of wall film growth, bed formation and deposition), measurement of hydrate growth rate, φ transition and the severity of hydrate blockage (resistance-to-flow) in under-inhibited water-dominated and oil-water systems at different THIs fractions, water cuts, and shear rates. This would help to provide a fundamental basis for identifying high- and low-risk operating regions for under-inhibited pipelines with respect to flow parameters such as THI concentrations, degree of sub-cooling, water cut and mixture velocity. It would be worthwhile examining whether any adverse effects arise from the use of THIs at comparatively low concentrations, such as accelerated formation rate (Austvik et al., 1995, Yousif, 1996, Yousif et al., 1996), which might be explained through an increase in gas solubility in the aqueous phase due to the presence of THIs. For oil-dominant systems, the intensified agglomeration and deposition reported at the lower THIs fractions (Hemmingsen et al., 2008, Li et al., 2011) might be explained through strengthened capillary forces and sintering effects due to the presence of unconverted water. Conducting such a comprehensive set of experiments could also help to develop a robust hydrate growth model for under-inhibited systems, and a establish a more reliable pressure drop model capable of considering the effect of both hydrate agglomeration/bedding in the continuous phase, and hydrate wall film growth and deposition on the pipeline wall. 93

111 Appendix A Hydrate Shell Growth Measured Using NMR A.1. Introduction In conventional oil transportation, where generally all three phases are present (water, oil and (natural) gas), hydrates may form and agglomerate according to the conceptual model (Figure 39) proposed by Turner et al. in collaboration with J. Abrahamson (Turner, 2005, Turner et al., 2009b). Initially, water droplets emulsify into the continuous oil phase because of shear in the pipeline, often assisted by other chemical agents in the oil phase (Sjoblom, 2012). Hydrates will nucleate and grow at the water interface on the droplet surface, where there is also a supply of guest molecules (typically methane) dissolved in the oil, to form a hydrate shell around the droplet water core. Over time, the hydrate particles will agglomerate and can eventually jam and plug the pipeline (Sloan and Koh, 2007, Koh et al., 2011, Aman et al., 2013b). Flow assurance engineers seek to avoid this situation by either preventing hydrate formation conditions from occurring in the pipeline, or when this is unavoidable, by actions based on a quantitative understanding of the mechanisms by which hydrates grow and form plugs (Koh et al., 2011). Water entrainment Hydrate growth Agglomeration Plugging Gas Oil Water Figure 39. Conceptual picture of hydrate formation in an oil dominant multiphase flow line (adapted with permission from Turner (2005)). Hydrate shell formation, indicated as white coloring, occurs around emulsified water droplets in the growth and agglomeration stages. Generally, industry aims to completely prevent hydrate formation by injecting sufficient quantities of inhibitor chemicals such as monoethylene glycol (MEG) or methanol; however, such an approach becomes increasingly expensive the deeper and further offshore the hydrocarbon exploration and transportation occurs (Cowie et al., 2005). Accordingly, industry is now moving from complete hydrate avoidance to a risk management approach (that may involve partial inhibition) (Sloan, 2005). However, there are still significant knowledge gaps relating to how hydrates form, agglomerate and plug pipelines, as well as with respect to the exact mechanisms by which various forms of hydrate inhibition, that target these phenomena, operate (Kelland, 2006, Ribeiro Jr and Lage, 2008). Studying these phenomena at a microscopic level, such as measuring adhesion/repulsion forces between hydrate particles and understanding the chemistry between 94