Premcor Port Arthur Refinery, part of the Premcor Refining Group has been expanding the capacity of a vacuum tower processing almost one million pounds per hour of heavy hydrocarbon feed. The feed is deficient in lighter, more volatile components and is extremely viscous at room temperature. The process is intended to squeeze as much useful fuel as practical out of the oil feed. During the past 5 years, CFD has become noticeably more widespread in solving single-phase flow problems, but progress in solving multiphase flows has been much slower. There are at least three primary solution methods currently available to solve a dispersed multiphase flow problem. The contract with Premcor called for the use of a Eulerian method. Later, as an in-house test, Flow Simulations studied the model of the tower using two other methods.
The Premcor port Arthur Refinery, part of the Premcor Refining Group, was expanding the capacity of a vacuum tower processing almost one million pounds per hour of heavy hydrocarbon feed. The feed is deficient in lighter, more volatile components and is extremely viscous at room temperature. The process is intended to squeeze as much useful fuel as practical out of the oil feed.
The feed, heated to 735.8°F, consists of liquid droplets suspended in a vapor. The stream enters the tower under vacuum, 0.483 psia. Liquid droplets are deflected downward by a baffle plate and removed in the bottom or “boot” portion of the tower, while vapor moves upward through a series of trays and demisting pads. The pads prevent entrained droplets of liquid from flying up into the rest of the tower. Trays collect the lighter product after it condenses in the upper sections of tower.
The bottom liquid product becomes delayed coker feed, which eventually yields a solid product. The vapor products undergo further processing to yield gasoline.
As part of an upgrade to increase the refinery’s output, the vacuum tower feed rate was to be almost doubled. A critical consideration in the tower was that liquid droplets would be entrained upward into the product draw-off tray. Because the droplets contain heavy-metal contaminants, they must be controlled to keep the product on spec.
The original liquid flow rate was approximately one-half of the vapor flow rate on a mass basis. However, on a volumetric basis, the tower inlet feed concentrations were 1.7 x 1(H volume fraction for the original operation and 3.7 × 10-4 volume fraction for the planned feed increase.
John Krawczyk is president of Flow Simulations Inc. in Miramar, Fla.
A primary concern with the increased feed rate was that there would be a disproportionate increase in the liquid entrainment. To ascertain whether this would be the case, Premcor commissioned Flow Simulations Inc. of Miramar, Fla., to perform a computational fluid dynamics model of the bottom region of the tower. If the entrainment increased proportionately to the liquid feed increase, or less, then no modification of the tower internals would be necessary.
During the past five years, CFD has become noticeably more widespread in solving single-phase flow problems, but progress in solving multiphase flows has been much slower. There are at least three primary solution methods currently available to solve a dispersed multiphase flow problem. The contract with Premcor called for the use of a Eulerian method. Later, as an in-house test, Flow Simulations studied the model of the tower using two other methods.
In all three studies, several simplifying assumptions were made to reduce the complexity of the physical problem to one that could be modeled economically with CFD.
Vaporization and condensation were neglected. That is, in the model it was assumed that mass transfer between the two phases had essentially ceased after the material entered the lower section of the tower. This information was not provided by the refinery, nor was it deemed critical to the simulation. The tower is 110 feet high and 39 feet in diameter. The feed inlets are at an elevation of 15 feet; at 30 feet is the first of several product draw-offs, followed at an elevation of 37.5 feet by a metal mesh demister.
Only the lower third of the column was modeled. There were more product draw-off trays and other features above this, but the entrainment was critical only in the lower column section. No significant penetration of droplets is expected above the first demister pad. The research assumed that virtually all the entrained liguid droplets would leave the tower in the first product draw-off.
All liquid droplets were modeled at the same size. While not a requirement for all of the dispersed phase models, this is a practical requirement for some of them. Two different operating modes of the tower were modeled. These were the original operation, which is referred to as the Basis Case. Here, the CFD models were validated and adjusted since the actual tower entrainment was measurable from product analysis. The second operating model was the planned feed increase, referred to as Case 2.
All simulations were performed using structured and unstructured CFD solvers from Fluent Inc. of Lebanon, N.H. Calculations were performed on a dual-processor Pentium-Pro computer running under Windows NT. The model solution times ranged from a day to a week or more for the Case 2 time-dependent simulations.
The Eulerian multiphase method, sometimes also called the multifluid model, is characterized by the solution of an individual set of momentum equations for each phase, liquid and vapor. These equations are coupled by an interphase drag term. A continuity equation also conserves mass across all phases. It is possible to add additional phases, but in practice, this becomes very expensive computationally.
All models of this type must be run in a time-dependent fashion for numerical stability. This, in turn, requires that models run for a sufficiently long time to achieve a steady state value. As a consequence, the geometry and computational meshes for Eulerian models must be greatly simplified.
Graphical CFD results for the vacuum tower after 10,000 time steps showed contours of liquid volume fraction at various elevations or slices within the tower. Liquid accumulating above the inlet baffle plate could be clearly seen as well as the entrainment of liquid up through the front slot in the baffle plate. Another region of fluid accumulation was directly underneath the demister pad, which was modeled as a porous medium.
More illustrative than the contour plots were several graphs of results.
The graphs show that entrained liquid leaving the top of the column was only 23 percent greater for Case 2 over the Basis Case. The liquid leaving the bottom of the column was 86 percent greater for Case 2 over the Basis Case. From this graph it was seen that Case 2 did not have proportionately greater liquid holdup in the column and most of the additional liquid exited the column bottom.
A Lagrangian, or particle tracking, multiphase model injects a finite number of particles with predefined properties into the flow. Unlike the Eulerian modeling method, the Lagrangian model can handle particles of differing sizes. Modeled particles can undergo size changes and mass transfer to the primary phase during their residence in the flow. Particles can undergo partially elastic collisions with walls, and can be trapped or retained by certain surfaces. Furthermore, these models do not have to be run in a time-dependent fashion.
For all these apparent advantages there are drawbacks to the particle tracking models. One drawback is that the particles actually have zero volume and thus are unaffected by their own concentrations, except for momentum effects. Simply put, these models deviate from reality in areas where the particles would have a significant volumetric fraction.
A rule of thumb used for the applicability of this modeling method is to use it only where the secondary phase volume fraction is about 10 percent or less. Since there were regions inside the column where these values were greatly exceeded, the Langrangian results may not be definitive for calculating mass loading of the different phases throughout the tower. Nonetheless, the particles do an excellent job of illustrating where the entrained liquids with their higher momentum will go inside a complex geometry.
Results from the particle tracking model were graphically comparable to the previous Eulerian results and illustrated which areas inside the column may permit high entrainment. These were the slot at the front of the baffle plate and the center free space inside the baffle plate.
The algebraic slip model solves for the momentum of the primary phase and calculates the momentum of the secondary phase using a constant slip factor, or about how much faster one phase moves than the other. Thus, factors such as droplet entrainment forces and gravity forces are modeled. One of the basic assumptions of the model is that the droplets cannot accelerate independently of the primary phase. This simplification is more accurate for smaller liquid droplet sizes.
Results from this model show the entrained droplets coming up through the front slot in the baffle plate as well as through the center free space. Some liquid can be seen accumulating on the top surface of the baffle plate, which agrees with the contour plots from the Eulerian model.
As the results of these CFD models show, Premcor Refining did not have to introduce additional baffling to its vacuum tower when increasing the liquid fraction in the feed. The CFD model showed areas in the tower baffling that could be improved to reduce liquid entrainment while not adding significantly to the tower pressure drop. In practice, tower internals are usually left alone unless there is a compelling reason to change them.
Approximately two years after the CFD model was commissioned, the tower went into operation under the Case 2 operating conditions. Analysis of the product indicated that the liquid droplet entrainment was proportionate to the liquid feed increase or less, validating that the Eulerian multiphase CFD model was correct in its prediction.