WHOC12-277
Experimental and numerical assessment of cold restart process of viscous oil pipeline A. TWERDA TNO;
[email protected]
Z. YANG Statoil;
[email protected] E. NENNIE TNO;
[email protected] H. VELTHUIS TNO;
[email protected] This paper has been selected for presentation and/or publication in the proceedings for the 2012 World Heavy Oil Congress [WHOC12]. The authors of this material have been cleared by all interested companies/employers/clients to authorize dmg events (Canada) inc., the congress producer, to make this material available to the attendees of WHOC12 and other relevant industry personnel.
and viscosity dependent, is also very well predicted. The simulations also show that there is a large dependency on temperature of the viscosity data. Therefore, it is very important to have proper data regarding the temperature dependency of the viscosity. Start-up simulations are more challenging, due to the non-uniform temperature (and viscosity) across the pipe section. As a result, 1D codes such as OLGA fail to accurately simulate the pressure gradient behaviour, as demonstrated by comparisons with experiments, limiting their range of applicability. Such limitations should be known when performing simulations to recommend operational procedures for viscous oil transport in pipelines.
Abstract Operating a flow line that transports so-called heavy oil can be cumbersome, especially at low temperatures. If a heavy oil flow has been stopped for awhile, the oil inside the flow line will cool, which significantly increases its viscosity. When re-starting the flow line, the pressure gradient necessary to transport the cooled oil has increased severely due to the increased viscosity. Furthermore, it is expected that 3D temperature effects play an important role in determining the restart pressure gradient behaviour, making it hard to accurately predict. Statoil has performed detailed experiments of heavy oil cool down and restart processes on their test rig in Porsgrunn. These experiments are simulated using two tools 1) the multiphase software tool OLGA. 2) the Computational Fluid Dynamics (CFD) code Fluent. The goal of the research is to identify the performance of each tool for the use in a cold start-up of offshore heavy oil production pipeline. Results demonstrate that both codes are very well capable of predicting the cool down of the heavy oil. The initial start-up pressure, which is fully temperature
Introduction As the resources of easy oil are decreasing the oil and gas industry is shifting its focus to the more difficult oils to produce. Offshore heavy oil is one of the candidates together with tar sands and oil from the fields in the artic regions. Heavy oil has received its name from the high viscosity of the fluid, especially at low temperatures, which makes it difficult to transport. Due to the high viscosities, high pressure drops are expected, especially after a cool down of the pipeline. The design of the
1
flow line, pumps and other equipment is complex with this type of fluid. Currently OLGA3 is used as the industrial standard for calculating these type of flows. OLGA uses onedimensional modeling concept for multi-phase flow. However, due to the –strong- temperature dependence of the viscosity the 1-dimensional assumption of the flow profile inside the pipe line can be significantly violated. This paper describes results of experiments and CFD modeling, in comparison with OLGA simulations to access the validity of this 1D assumption. The paper is set up as follows. First the experiments will be detailed. Next, the OLGA simulations are discussed followed by the CFD simulations. Finally some discussion and conclusions will be presented.
For material properties of the pipe line and insulation, typical values are used. For the heat transfer to the surrounding standard values in OLGA are used for air. The surrounding air velocity is taken equal to 2 m/s. This results in a heat transfer coefficient to the surrounding of 10.33 W/(m2K). Mass flows were taken from the flow loop measurements. The insulation value of the pipes is not known a priori, but is important to predict oil temperatures. The insulation value was estimated by matching simulation results with experiments, for a typical situation. This is not a recommended practice for validating simulations with measurements. It is recommended to determine the insulation value in an independent way.
Experiments
Figure 1 Schematic set up of flow loop of Statoil.
Figure 2 Temperature profile of the cool down period.
Figure 1 shows the layout of the flow loop where the experiments are performed. The pipe diameter was 2” and surrounded by 80 mm of insulation. The flow loop is especially designed to handle viscous oils, together with water and or gas. In this paper only single fluid (oil) experiments are considered. At several locations, denoted by the red circles, the temperature and pressure (drops) are measured. The shut-in of the flowline was achieved by quickly closing the valves at both end of 2 inch flowline. At the same time the by-passing line which links the both end of the test flowline was open such that both feeding pumps for oil or water can still be running. After the shut-in of the flowline, the fluid in the pipe was cooled via natural convection in air. The restart of the flowline was conducted by injecting oil or water flow into the 2 inch pipeline after a period of natural convection cooling (no flow in the 2 inch pipeline) or with pre-defined temperature. During the period of shut-in in of 2inch’s test flowline, the flow simply by-passes the 2-inch’s flowline, and is routed the outlet of the test flowline. The shut-in of the flowline is achieved by closing the valves at both ends of 2-inch’s flowline, and by opening the valve on the line directly linking ESP and the first-stage separator. The restart process is just a reverse process of the shut-in.
Figure 3 Pressure drops over several sections of the flow loop after restart calculated with OLGA compared with the measurements. Figure 2 shows the temperature behavior during cool down. As can be seen OLGA predicts it very well. The differences in measured values can be explained by local insulation differences at the flow loop and the ondulation can be related to daily variations of the ambient temperature. Figure 3 and Figure 4, show the pressure drop and temperatures respectively during restart. Both OLGA results and measured values are plotted. The initial pressure drop is well predicted by OLGA. However prediction of the transient behaviour is less good. Prediction of the hot fluid arrival time by OLGA is significantly delayed compared to the measurements. Unexpectedly the temperature evolution is quite well predicted.. The OLGA predicted temperatures correspond of course a uniform section-wise temperature.
Olga simulations For the OLGA simulations version 6.1.0. is used. The material properties are computed using PVTsim, with the exception of the viscosity which is recomputed using the exprimental data.
2
Figure 6 Temperatures at several locations of the flow loop after restart calculated with CFD. Figure 4 Temperatures at several locations of the flow loop after restart calculated with OLGA.
The pressure drop during restart is shown in Figure 5 and compared with experiments. One can see from this comparison that the initial peak in pressure drop is predicted nicely. For the last two measurement sections we however see that CFD predicts the time of arrival of the temperature front too early. This contradicts the prediction of the temperature on the outside wall in Figure 6, which is clearly under predicted. Figure 7 and Figure 8 show the temperature and velocity profile at several times after the restart. Here it can be seen that the hot core moves fast through the pipe, while the fluid close to the wall heats up slowly and remains viscous. Looking at the previous results we see that the radial temperature increase is under-predicted by the CFD simulation. One of the reason could be that the thermal conductivity (λ) has not the right value coming from the PVT data. The result of increasing the heat conductivity with a factor of 10 is shown in Figure 9 where the wall temperature during restart is plotted. As can be seen the temperatures are now much better predicted. For the pressures the differences are small and are therefore not given.
CFD simulations Next to the OLGA simulations CFD simulations were also performed. The CFD code used is the commercially available ANSYS Fluent4 version 13. The CFD solves the equations of conservation of mass, momentum and energy in more than one dimension on a predefined grid1. As the flow is presumed to be laminar and single phase, CFD is expected to give accurate solutions.
Case setup The geometry was modelled as an axis-symmetric set-up. The total axial length was 200m. The diameter and thickness of the pipe, wall and insulation are identical to the ones used in the OLGA simulations. The grid consisted of hexahedral cells with the following distribution. The number of cells in radial direction is: ten (10) cells inside the pipe, three (3) cells inside the steel wall and nine (9) cells inside the insulation. The grid is locally refined in the radial direction near the steel wall, because the largest gradients are expected in this region. The cell size in the axial direction is approximately 4cm. This resulted in the total mesh size to be around 100,000 cells.
Looking at the Reynolds number (Re) of the flow during start-up it reaches values close to the transition value from laminar to turbulence for pipe flow, which is Re=2300. This also explains the spikey behaviour of the pressure drop in the measurements in Figure 5 at the end of the experiment. Therefore, we also investigated the influence of turbulence modelling on the results. Because turbulence enhances mixing and is usually modelled as an effective heat conductivity it could explain the difference between the modelled and measured temperature values. The turbulence model which is applied is a transitional k-ω2. In this model an additional (next to k and ω) transport equation for the laminar kinetic energy (kL) is solved. This variable describes the pretransitional fluctuations in the flow. The transition process to turbulence is represented in this model by the transfer of energy from the laminar kinetic energy to the turbulent kinetic energy (k). This model is regarded to be well suited for predicting the laminar turbulent transition combined with heat exchange to the wall. As inlet boundary condition for a turbulent intensity of 0.1 was chosen with the avlue for kL of 1e-6. As can be seen in the temperature profile of Figure 11, the increase of temperature is now over predicted. The CFD model clearly predicts a turbulent flow, which enhances the
Results
Figure 5 Pressure drops over several sections of the flow loop after restart calculated with CFD compared with the measurements.
3
radial temperature conductivity. The pressure drop, shown in Figure 10, is also less well predicted, with a pressure front which is significantly delayed. However, the sharp pressure gradient at the pressure front is much better resolved(circled in the figure).
Figure 10 Pressure drops over several sections of the flow loop after restart calculated with CFD using the k-w turbulence model compared with the measurements.
Figure 7 Temperature profile at a measurement location at several times computed with CFD.
Figure 11 Temperatures at several locations of the flow loop after restart calculated with CFD using the k-w turbulence model compared with the measurements.
Comparison with OLGA Now we will compare some CFD results with OLGA simulations and the measurements. In Figure 12 the pressure drop at one of the measurement locations is plotted. We see that CFD gives an improvement to the OLGA results, however some differences with the measurements remain. For the temperature, which is plotted in Figure 13 we see that OLGA predicts the increase in temperature too fast and CFD too slow. As already mentioned after an increase of the heat conductivity, the simulated results match nicely.
Figure 8 Velocity profile at a measurement location at several times computed with CFD.
Figure 12 Comparison of the simulations of OLGA, CFD and experiments with respect to the pressure drop.
Figure 9 Temperatures at several locations of the flow loop after restart calculated with CFD, using the higher value for λ, compared with the measurements.
4
[2] Walters, K. W., Cokljat, D., (2008), "A Three-Equation Eddy-Viscosity Model Reynolds-Averaged Navier-Stokes Simulation of Transitional Flow", J. of Fluids Engineering 130-121401. [3] OLGA, www.sptgroup.com/Products/olga/ [4] ANSYS Fluent, www.ansys.com
Figure 13 Comparison of the simulations of OLGA, CFD and experiments with respect to the temperature.
Discussion & Conclusion The prediction of the pressure drop at restart and the period after restart, boils down to the correct prediction of fluid temperature in the flow loop in the various periods, given a known initial condition and known boundary conditions. In general OLGA is very well capable of predicting the cool down behaviour of the heavy oil in the flow loop. For the trends in temperature and pressure drop after startup we conclude that OLGA has some difficulties to predict the correct values. Quantifying OLGA’s uncertainties cannot be easily evaluated since this is dependent on the specific properties of the oil and of the flowing conditions. For the CFD results and looking at the experiments we see that the initial and transient pressure drop is well predicted. The measured temperatures during startup were more difficult. An increase of the heat conductivity (λ) with a factor of 10(!) gives the best of the data. Several additional simulations were performed to examine this. Interpreting these we state that the reason(s) for increased (effective) heat conductivity could be: • Turbulence • Errors in heat conductivity taken from PVTsim • 3D effects, such as bends or locally degraded insulation
NOMENCLATURE Re = Reynolds number λ = heat conductivity CFD = Computational Fluid Dynamics
REFERENCES [1] Wesseling, P., (2000) Principles of Computational Fluid Dynamics, Springer-Verlag New York, Inc. Secaucus, NJ, USA
5
