A NEW COMPUTER MODEL FOR GAS HYDRATES AND HYDRATE INHIBITION by B. Edmonds, R. A. S. Moorwood and R.

Szczepanski, Infochem Computer Services Ltd., 90 London Road, London SE1 6LN. Background and Objectives The theoretical understanding of gas hydrates stems from the work of van der Waals and Platteeuw (1) who proposed that gas hydrates are ice-like crystalline structures into which light gas molecules are adsorbed. Van der Waals and Platteeuw derived the statistical mechanics of gas hydrates modelling the interaction between the gas molecules and the hydrate lattice using the Lennard-Jones potential. Their model was refined by Parrish and Prausnitz (2) who used instead the Kihara potential and devised a computer method for solving the model to predict the hydrate dissociation point. The Parrish and Prausnitz model predicted the dissociation curves of simple hydrates of single gases well, but was not reliable for gas mixtures. An empirical modification was made to the Parrish and Prausnitz model by Ng and Robinson (3) which improved the results for the dissociation points of mixed hydrates. However, Ng and Robinsons modification is thermodynamically incorrect; despite this, it has been used as the basis of the hydrate predictions in some widely used commercial simulation programs (4). Other effects that influence the formation of hydrates, such as the presence of inhibitors or the effect of gas solubility, are usually handled by adding empirical corrections to the value of the activity of water in the thermodynamic expressions for the properties of the hydrates, i.e. the true influence of these effects on the fluid phases is ignored and a hidden assumption is made that a liquid water phase is present, which is not always correct. See (4) again, for example. The aim of this work, which was carried out in collaboration with Shell Research, was to produce a practical engineering model for gas hydrates and hydrate inhibitors that is thermodynamically consistent, that accurately represents all the published phase equilibrium data of relevance to the problem, and that is capable of solving all phase equilibrium problems associated with gas hydrates, not just determining the hydrate dissociation points. To achieve this, three complementary facilities had to be developed, namely, the mathematical techniques to solve the phase equilibrium problems involved, the thermodynamic model for the gas hydrates and the thermodynamic model for the fluid phases which include the hydrate inhibitors. These are described below. Mathematical Techniques The basis of all the calculations is a new computer algorithm, Multiflash, developed by the authors for solving multiphase equilibrium problems (5). Multiflash is able to perform all the normal engineering flash calculations, dew and bubble point searches etc. for problems involving any number and combinations of phases. The phases may have quite different properties

calculated from different thermodynamic models. The software automatically determines which phases will be present under given conditions and returns the relative amounts, compositions and properties of the phases. In this study, Multiflash was applied to problems involving up to six different phases: oil, gas, liquid water, ice, hydrate structure I and hydrate structure II, all phases being treated in an equivalent manner. The advantages of this approach are: The same methods and models can be used whatever combination of phases is present. For example, formation of hydrates from gases in the absence of a free water phase or from crude oils in the absence of a gas phase is predicted automatically. Because the thermodynamic models used are completely consistent, the predictions are as reliable as for the more normal situation where gas and liquid water are present. For example, Figure 1 illustrates the calculated values of the hypothetical vapour pressure of water over empty structure I hydrate compared with the values predicted by Sloan by analysis of experimental data (4). This result indicates that the program will reliably calculate the transition from gas to hydrate in the absence of free water as the underlying thermodynamic properties are accurately represented. The program can reveal the full complexity of the phase behaviour especially when soluble gases and inhibitors are present. Figures 2 and 3, for example, show the phase behaviour of a wet gas stream with methanol added. This example was studied as part of an investigation of a real process. Figure 2 shows the phase behaviour of the mixture when only the fluid phases are considered. Figure 3 in contrast also includes the hydrate and ice phases. Figure 3 shows a number of complex effects that are counter-intuitive and can only be calculated using a program that can handle the full phase behaviour of the system. The hydrate dissociation curve shows sharp discontinuities where it crosses the water dew-point line and the hydrocarbon phase envelope. This behaviour is reminiscent of the discontinuities that are known to occur at the hydrocarbon phase envelope for cases when large amounts of carbon dioxide are present (4). Another surprising feature is the conversion of hydrate back to liquid water as the temperature decreases. At low temperatures, the liquid water phase contains large amounts of methanol which forms because of the decreasing mutual solubility of methanol and hydrocarbons with decreasing temperature. Also notable is the prediction of a small ice region at low pressures. Finally but most importantly, a data regression program has been developed based on Multiflash that is able to fit model parameters simultaneously to many different sets of phase equilibrium data. Each data set may contain data for different combinations of components and phases. This regression program was essential for developing the new hydrate/hydrate inhibition model as it made possible the comparison of different types of data for consistency. It also ensured that the hydrate model represented the properties of structure I and structure II hydrates equally accurately.

Hydrate Model The Parrish and Prausnitz model was investigated initially as, with the modification of Ng and Robinson, it has been widely used in oil and gas processing. Calculations showed that the values of the thermodynamic properties of the empty hydrate lattices proposed by Parrish and Prausnitz are not correct. Using this model, it is possible to obtain good predictions for simple hydrates but not for mixed hydrates. Figure 4, for example, shows the best fit that could be obtained for the methane hydrate dissociation curve when fitting to data for both hydrate structures I and II using the Parrish and Prausnitz empty lattice parameters. In contrast, Figure 5 shows the result when regressing the same data with revised values for the empty lattice parameters. Considering that the data ranges from 1 bar to 4000 bar over a temperature range of 125K the agreement in Figure 5 is very satisfactory. Some deviations are observed for pressures above 2000 bar which is due to the inability of cubic equations of state to model accurately the properties of the gas phase at such high pressures. From analysing nearly all the published data for simple and mixed hydrates, it was concluded that it is possible to set parameters in the van der WaalsPlatteeuw model to reproduce most of the measurements within 1K. The only exception to this is the data for n-butane rich hydrates at low temperatures. However, it is known that due to their size, n-butane molecules cause lattice strain of the hydrate, an effect that is neglected in the van der Waals-Platteeuw model. Bases in this analysis, there seems to be no need to introduce any empirical modifications to the van der Waals-Platteeuw model. The modification of Ng and Robinson appears only to be necessary to compensate for inaccurate parameter values in the Parrish and Prausnitz version of the model. The advantage of using a thermodynamically rigorous approach is that all properties of the hydrates can be calculated because they are consistent, and used in calculating the phase equilibria and associated physical properties of the system. Figures 6, 7 and 8 show a selection of typical results for the new model to illustrate the agreement with experiment. Because the Multiflash algorithm uses rigorous methods to calculate phase equilibria, it is always possible to check for given conditions whether a structure I or structure II hydrate will form. In the course of the work it was found that some of the published hydrate models erroneously predict structure changes. For example, the Parrish and Prausnitz model predicts that pure methane hydrate will change from structure I to structure II as the temperature falls. Computer calculations reveal a change to structure II on the water-hydrate-vapour line at 263.6K under a pressure of 10 bar. Independent algebraic analysis of the model confirmed this result. Experimentally, it is known that methane hydrate is structure I at all temperatures (4). Similar problems were encountered with other published models. For example, the model of Munck et al. (6) which uses the Parrish and Prausnitz values for the empty hydrate lattice parameters, erroneously predicts that pure methane hydrate changes to structure II at 252.2K under a pressure of 10 bar. The model developed in this work predicts the correct structure of methane hydrate.

The question arises as to why these incorrect predictions of structure changes have not been previously reported. It seems likely that many of the computer programs used to perform hydrate calculations do not use rigorous methods to determine which hydrate phase actually exists and as a result do not detect what their models are actually predicting. The main features of the new hydrates model are: It uses thermodynamically consistent models for all phases that may be present including the hydrates and the ice phase. The thermal properties (enthalpies and entropies) of the hydrates and ice are also returned permitting isenthalpic and isentropic flash calculations involving these phases. This makes it possible, for example, to predict if hydrates would form in a choke or a turbo-expander by performing the appropriate flash calculation. Calculations can be made for any possible combination of phases, including non-free water calculations, and the phases present at equilibrium are automatically found. The properties of the hydrates have been fixed by investigating data for both simple hydrates and mixed hydrates. For simple hydrates, all data from the ice-hydrate-vapour lines to the water-hydrate-liquid lines have been regressed going through both quadruple points where they exist. Simultaneously the data for mixed hydrates have been regressed to give accurate parameters for both structure I and structure II hydrates. Most hydrate data for both pure gases and mixtures are reproduced typically within 1K. The properties of the empty hydrate lattices have been investigated, and earlier values such as those proposed by Parrish and Prausnitz have been found not to be compatible with the data. Proper allowance has been made for the solubilities of gases in water when developing the model. If this is not done, the model parameters will be distorted as dissolved gases have an inhibiting effect on hydrate formation. This is particularly important for relatively soluble gases such as carbon dioxide and hydrogen sulphide. Rigorous thermodynamic calculations of the most stable hydrate structure have been made to ensure that the model agrees with the experimental evidence. These consistency checks have revealed that some published models erroneously predict hydrate structure changes and that many computer programs used for hydrate calculations are apparently unable to detect this.

Inhibition Model The properties of gas hydrates are of interest to engineers as it is necessary to know when they may form. The properties of hydrate inhibitors are possibly of even greater interest as the objective with hydrates in gas processing is to suppress them. The inhibitors studied in this work were methanol, monoethylene glycol (MEG), diethylene glycol (DEG) and triethylene glycol (TEG). The mechanism of hydrate inhibition is that the inhibitor mainly dissolves in the free water phase thereby reducing the fugacity of water. As water is the principal component of gas hydrates, reducing the fugacity of water also reduces the tendency of hydrates to form. It is important to realise that inhibitors modify the properties of the fluid phases, not the hydrate phase, and as a result they have a number of other effects. The presence of inhibitors raises the solubility of gases and hydrocarbons in the water phase thereby inhibiting the formation of hydrates still further. The inhibitor also depresses the ice point. It reduces the vapour pressure of water, i.e. it has a dehydrating effect on the vapour and hydrocarbon liquid phases. The inhibitor itself vaporises and also dissolves in the hydrocarbon liquid, i.e. it partitions. The conventional method of simulating hydrate inhibition is to modify the thermodynamic expressions for the formation of hydrate. This does not model the actual mechanism of inhibition and does not therefore give a complete picture of what happens. In this study, the effect of inhibitors was modelled by adopting new mixing rules for the equation of state. The preferred equation of state is the RKS equation which is known to give better values for the fugacities of the components of natural gases than the Peng-Robinson equation (7). However, the Peng-Robinson equation can be used if required. The mixing rule is related to the Huron-Vidal mixing rule (8). It matches the excess Gibbs energy of a fluid mixture using an expression derived from the NRTL equation, but it reduces to the conventional van der Waals mixing rule if water or inhibitor is not present, i.e. for oil and gas mixtures it gives identical results to the conventional equation of state. All the open literature data for the effect of methanol, MEG, DEG and TEG on hydrate formation have been studied. In order to build up a comprehensive model, all relevant types of phase equilibrium data were considered including freezing point depression curves, vapour-liquid equilibrium data for waterinhibitor mixtures, vapour-liquid, liquid-liquid and solubility data for light gases, hydrocarbons and inhibitors. The interaction parameters have all been correlated so that they can be automatically generated for any oil or gas mixture with water and inhibitors. When the dissociation curve of a simple hydrate is reported, the meaning of the reported data is clear. The dissociation curve is a three phase equilibrium point for a mixture of three components. From the phase rule it follows that for a given pressure, there is a unique dissociation temperature and vice versa. A difficulty of assessing published data for hydrate inhibition is the fact that the hydrate dissociation point for a gas mixture in the presence of inhibitor is not a unique point but is dependent on the relative amounts of the phases present. Figure 9 illustrates this for a gas mixture where the ratio of water phase to hydrocarbon phases has been varied. In this example the hydrate point changes by over 20K. Unfortunately virtually none of the published data

for the effect of inhibitors on hydrate dissociation curves gives the relative amounts of the phases present so it cannot be properly interpreted. The practical consequences of this are: 1. The available experimental data only gives an approximate indication of the effect of inhibitors. 2. To calculate the effect of inhibitors in a process accurately, it is necessary to know the amounts of all phases present and it is necessary to use a computer model that is able to take these effects into account. Figures 10, 11 and 12 give a selection of results for the inhibitors studied. The calculations reported here were done with the arbitrary assumption of equimolar mixtures of aqueous phase and hydrocarbon phases. Inhibitor partitioning is an important factor as it affects the design of recovery systems. It is difficult to develop a comprehensive model for partitioning as there are few data available; also, the reliability of some of these data is open to question. Table 1 shows an example of the partitioning of methanol calculated from the model compared with data reported by Ng and Robinson (9). The calculated methanol mole percent in the vapour at 0.6C shows the largest deviation, but comparison of the experimental values with the Ng and Robinsons reported feed compositions shows that the methanol compositions do not obey mass balance, i.e. the experimental values cannot be accurate. The results are also somewhat deviant for carbon dioxide. The chemistry of aqueous solutions of carbon dioxide is complex and it is very difficult to represent the solubility of carbon dioxide in water accurately with a simple engineering model. Apart from this, the model gives good predictions. In summary, the inhibitor model has the following features: The model is a thermodynamically consistent model for all fluid phases. It can explicitly represent all the effects of inhibitors including hydrate suppression, depression of freezing point, reduction of water vapour pressure and inhibitor partitioning. The model has been developed using extensive phase equilibrium and hydrate data for methanol, MEG, DEG and TEG. The model can show the effect of different relative amounts of the phases present on the hydrate inhibition. The model can be extended to include other hydrate inhibitors as the methods used are quite general. Future developments The new model has been developed to the point where it can be used reliably to calculate hydrate problems both as a stand-alone utility and as a subroutine library that can be called from a larger program. New developments are being planned to enhance the model. The model for the inhibitors needs to be complemented by a model for the inhibiting effect of salinity of the water phase. A practical and consistent method for accounting for salinity is under development that can be used in conjunction with the existing model but does not require a full electrochemical treatment of aqueous electrolyte solutions.

The recent discovery of structure H hydrates is an obvious extension as the Multiflash algorithm can determine which hydrate structure will exist under given conditions. All that is required is a new model option to return the properties of structure H. Structure H hydrates are of importance as they may exist at higher temperatures than structures I and II under certain conditions (10). Models for other solid depositions such as waxes and asphaltenes are another obvious extension, again because the numerical methods are in place to determine when a phase will exist. Acknowledgement The authors would like to acknowledge many useful discussions with Dr. Eric Hendriks of the Royal Dutch Shell Laboratory, Amsterdam, in the course of the work. References (1) J. H. van der Waals, J. C. Platteeuw: Adv. Chem. Phys., 2, 1, (1959). (2) W. R. Parrish, J. M. Prausnitz: Ind. Eng. Chem. Proc. Des. Dev., 11, 26, (1972). (3) H.-J. Ng, D. B. Robinson: Ind. Eng. Chem. Fundam., 15, 293, (1976). (4) E. D. Sloan: Clathrate Hydrates of Natural Gases, Dekker, New York, (1990). (5) J. F. Counsell, R. A. S. Moorwood, R. Szczepanski: Calculating Multiphase Equilibria, VLE 90 Conference, Aston University, (1990). (6) J. Munck, S. Skjold-Joergensen, P. Rasmussen: Chem. Eng. Sci., 43, 2661, (1988). (7) L. Lundgaard, J. M. Mollerup: Fluid Phase Equil., 70, 199, (1991). (8) M. J. Huron, J. Vidal: Fluid Phase Equil., 3, 255, (1979). (9) H.-J. Ng, D. B. Robinson: GPA Report RR-66, (1983). (10) J. A. Ripmeester, C. I. Ratcliffe, G. E. McLaurin: The Role of Heavier Hydrocarbons in Hydrate Formation, Session on Hydrates in the Gas Industry, AIChE Meeting, Houston, (April 1991).

Table 1
Temp Pressure wt % Methanol 0.6 C 1.48 MPa 10% Methanol Mole Percent Phase Vapour Component Water Methanol Nitrogen Methane Ethane Propane n-Butane n-Pentane CO2 Water Methanol Nitrogen Methane Ethane Propane n-Butane n-Pentane CO2 Water Methanol Nitrogen Methane Ethane Propane n-Butane n-Pentane CO2 Water Methanol Nitrogen Methane Ethane Propane n-Butane n-Pentane CO2 Water Methanol Nitrogen Methane Ethane Propane n-Butane n-Pentane CO2 Water Methanol Nitrogen Methane Ethane Propane n-Butane n-Pentane CO2 Water Methanol Nitrogen Methane Ethane Propane n-Butane n-Pentane CO2 Water Methanol Experiment 0.045 0.058 6.378 74.640 5.241 2.191 0.886 0.925 9.636 94.006 5.771 0.005 0.059 0.005 0.002 0.152 0.028 0.077 6.219 74.194 5.196 2.174 0.872 0.950 10.290 93.337 5.830 0.020 0.280 0.010 0.002 0.521 0.032 0.059 6.357 74.828 4.908 2.022 0.982 0.823 9.989 87.527 12.080 0.009 0.107 0.010 0.003 0.264 0.029 0.115 6.242 73.245 5.251 2.197 0.804 0.588 11.529 86.899 11.989 Calculated 0.047 0.029 5.926 71.20 4.696 1.931 0.780 0.779 14.611 93.794 5.840 0.002 0.072 0.007 0.004 0.002 0.003 0.281 0.044 0.070 5.958 71.415 4.726 1.947 0.788 0.788 14.264 93.704 5.749 0.016 0.410 0.015 0.004 0.000 0.000 0.731 0.023 0.040 5.994 71.525 4.638 1.700 0.760 0.771 14.553 86.181 13.116 0.003 0.156 0.024 0.002 0.001 0.002 0.470 0.031 0.131 6.008 71.482 4.662 1.844 0.789 0.795 14.258 85.76 12.644

Aqueous liquid

Vapour 15.7 C 16.71 MPa 10% Methanol

Aqueous liquid

Vapour -3.9 C 2.11 MPa 20% Methanol

Aqueous liquid

Vapour 9.4 C 16.78 MPa 20% Methanol

Aqueous liquid

Nitrogen Methane Ethane Propane n-Butane n-Pentane CO2

0.028 0.355 0.014 0.002 0.713

0.018 0.651 0.034 0.009 0.003 0.003 0.878