Kenneth J. Fewel Jr.
Peerless Mfg. Co.
Dallas
James A. Kean
Atlantic Richfield Indonesia Inc.
Jakarta
Computational fluid dynamics (CFD) modeling is effective for designing retrofit internals of liquid/gas separators. This is practical because of recent advances in high-speed digital computers that permit full-scale computational modeling of fluid dynamics.
This technique is being used to design gas/liquid separators and pressure vessels, and it is important that engineers involved in process design understand the principles, powers, and limitations of CFD when applied to such problems.
LIQUID/GAS SEPARATION
Four common types of liquid/gas separating equipment are: gravity separators, mesh pads, centrifugal separators, and vane-type separators. Current designs employ more vane-type separators. These separators significantly reduce size, weight, and cost. Also, vane-type separators address safety and operational concerns.
GRAVITY SEPARATORS
In gravity separators, droplets are separated by gravitational force acting upon entrained droplets. When the gravitational force on a droplet exceeds the drag force imparted by the flowing gas stream, particle separation is achieved. When drag force equals gravitational force, the associated droplet velocity is called the terminal velocity.
Because of large vessel size, gravity separators are seldom used for separating droplets with diameters of less than 300 m. If fine droplet separation is not an operational requirement, these separators are often employed as vent scrubbers and slug catchers.
MESH PADS
As the gas stream flows through mesh pads (fibrous coalescers), droplet separation occurs by droplet interception and agglomeration mechanisms. Mesh pads are manufactured from porous blankets of knitted metal wire or plastic material with a void volume of up to 95%.1
Mesh pads (still frequently used in the natural gas industry) are used in the final separation stage that removes particles unaffected by the gravity separation region of the separator. When pads are properly designed, particles can be efficiently separated even if the droplets are less than 10 m in diameter.
It is still prudent, however, to consider alternate separation equipment such as vane-type separators, especially in high-capacity applications where size and weight can be reduced.
CENTRIFUGAL SEPARATORS
Centrifugal separators accomplish droplet separation by imparting centrifugal force to the gas stream and subsequently changing the stream's flow direction. These effects can be as large as 2,000 times the gravitational force.2
Droplets and particles are effectively separated at design conditions, and separator size can be significantly reduced. However, several operational drawbacks have been observed. These include increased pressure drop, limited turndown, and liquid creep into the outlet nozzle.
Centrifugal separators are used successfully in high-capacity, limited-turndown situations, and can achieve efficient separation of particles with diameters from 8 to 10 m.
VANE-TYPE SEPARATORS
Vane-type separators use the inertia of the droplets in the gas stream to separate the droplets into a film of liquid on the vane surface. The liquid surface then drains down the vane into a trough (Fig. 1a).
For many years, vanes have been used in a wide variety of mist extractor applications. These separators are particularly well-suited for applications in which fibrous pads were formerly used and for increasing gas capacity. Vane-type separators can also significantly reduce separator size.
Vane-type separators have evolved from bent, flat plates to today's configurations of meticulously designed pockets (or catchments). These pockets (Fig. 1b) greatly increase the separator's liquid capacity. Pockets come in two general forms: single pockets that extend into the gas flow stream and double (flush) pockets that do not obstruct the gas stream.
Single pockets, the earliest form, increase turbulence in the gas stream moving through the vane flow channel. Since the 1970s, double pockets have become popular because by not obstructing the gas flow channel, this design experiences less pressure loss than single-pocket vanes.
Double-pocket vane separators have generally higher capacities because the pocket areas are larger and better protected from the fast-moving gas stream.
Efficiency in double-pocket vanes is also improved because of improved gas dynamics at each bend. The resulting flow radius of curvature is tighter and thus more centrifugal force is created at each turn.
SEPARATION EFFICIENCY
Separator efficiency for various droplet sizes in vanes can be readily determined from Calvert's equation.3 The equation takes into account the inertial force created at each bend in the vane as well as gas flow drag on the droplets to be separated.
The equation does not allow for coalescing or pocket effects. To adjust for these effects, a diffusional factor (fd, found by experiment) multiplied by the number of baffles is used to modify the resulting Pt.
Pt=exp(-Ut 2nr0/U'gb') (1)
where:
Pt=Fractional penetration of a droplet size
Ut=Terminal settling velocity of a given droplet
U'g=Actual gas velocity between baffles
n=Number of baffles
r=Radius of curvature of flow streamlines
0=Angle of baffles
b'=Perpendicular distance between two consecutive baffles in the same row
PRESSURE DROP
Pressure drops across vane elements are usually measured at several different flow rates. A pressure drop-vs.-face velocity plot can be prepared.
Because pressure losses through vanes are almost pure inertial losses, pressure losses conform to a second-order power curve. The viscous loss is negligible. Hence, it is convenient to express the pressure loss characteristics as K, a dimensionless inertial loss coefficient.
_P=Kxd.p. (2)
where:
_P=Pressure loss
K=Inertial loss coefficient
d.p.=pV2/2gc=Dynamic pressure across inlet of vane separator
p=Fluid density
V=Fluid velocity
gc=Gravitational constant
Although K may change with the amount of turbulence present, the change is usually insignificant for most vane applications.
If an agglomerator vane or fibrous filter is used, then a viscosity term must be added to the above equation. This equation (Equation 3) can be determined through regression using a second-order parabolic fit.
Carefully measured pressure drop data almost always fits this type of curve.
CAPACITY
Capacity is usually the single most important variable in gas/liquid separator design. Capacity is frequently overstated, however, because of errors in scale-up of experimental results.
All designers must use a safety margin, the magnitude of which depends upon individual engineering expertise.
Using dynamic pressure (pV2) is a simple, practical way of predicting the allowable vapor rate for vanes. This value represents the energy level of vapor flow necessary to re-entrain thin films from the vane surface or drainage channel.
If the films are sufficiently thin, this energy dominates the breakthrough condition regardless of surface tension. However, if the surface tension, viscosity, and density of the tested liquid differ significantly from those of water, the allowable pV2 is altered proportionately.
Dynamic pressure allowables are practical guides to designing vane separators. Their values are frequently modified by experience with the application, simulation analysis, or experiment.
Dynamic pressure allowables are best determined by field tests using actual gas flow conditions. However to arrive at the design for a prototype, an air/water scale test is often used to determine maximum capacity.
SCALE-UP
Scale-up from laboratory testing using pV2 has been utilized for years to predict field application capacity. Unfortunately, many problems with scale-up can occur. Its accuracy is impaired by the differences in physical properties of the gases and liquids to be separated.
Density, viscosity, and surface tension will be much different in a field application. These variables affect such important attributes as droplet distribution in the inlet pipe, flow patterns of the gas, re-entrainment of separated liquids, and efficiency of separation-vs.-droplet size.
These challenges are not insurmountable. Experimental testing with differing flow properties has led to empirical guidelines which can reasonably predict the changes encountered in a typical field application.
Field experience is added to the equation to supplement the scale-up technique, and most recently, computational fluid dynamics (CFD) has provided a new and powerful procedure for scaling-up of prototypes.
CFD THEORY
CFD is two or three-dimensional simulation of fluid flow using finite elements or finite difference mathematical analysis. This analysis is based upon solving nonlinear differential equations that can completely determine fluid flow.
Although the differential equations are represented by simplified algebraic equations, the resolution of the network, or grids, used to define the flow geometry can be made fine enough to accurately predict flow.
When utilizing CFD, a massive number of calculations is necessary to arrive at an accurate and converged solution. Modern workstations equipped with flow simulation software have largely overcome this obstacle.
Programs such as Fluent use a finite difference representation with conservation of mass, momentum, and turbulence to yield solutions to real viscous flows.
MODEL BUILDING
Fig. 2a illustrates a typical computational grid that defines the flow domain, or volume available, for flow modeling. Every domain must have inlet cells and exit cells with prescribed boundary conditions such as velocity, density, and turbulence.
Within the prescribed domain, "live" (or empty) cells are available for flow. Wall cells can be set anywhere to define the boundaries of the flow.
Porous cells can also be designated to act as surrogates for fibrous filters, packings, or separation devices such as vanes.
Notice that in certain regions, the grid is denser than in others. This density fluctuation is sometimes necessary to generate accurate flow predictions in regions of high-velocity gradients.
Because any finite difference scheme relies upon approximating the differences between cells, a coarse grid assumes that small differences are present between cells.
Consequently, the accuracy cannot be determined without creating a finer grid and recomputing the flow. This process must continue until the solution is observed to change very little between grid refinements.
An experienced CFD engineer can usually create a fine enough grid in one trial to yield a flow prediction of appropriate accuracy.
Fig. 4a also depicts porous cells used to model devices which cause viscous (or inertial) resistance to flow. These must be used properly for accurate flow predictions.
Porous cells in Fluent are created with a modified Brinkman equation.4 This equation combines viscous and inertial terms with coefficients determined from flow testing.
The simplified Brinkman equation is:
_P=(mV/k)+(KpV2/2gc) (3)
where:
m=Fluid viscosity
k=Permeability of the media
The other variables are the same as in Equation 2.
Equation 3 can be used to predict the resistance to flow for a variety of separation elements. For example, vanes, mesh pads, packings, and filters can all be simulated with this equation.
FLOW VISUALIZATION
Once computed, the flow field can be visualized by graphics of velocity vectors, contours of velocity, pressure, turbulence, or other flow properties. A velocity vector plot is most useful for three-dimensional flow visualization.
For example, Fig. 2b illustrates the velocity vectors for a horizontal separator. Note how this graphic reveals the evenness of flow through the vane separators (represented by porous cells).
Velocity magnitude contours can be used to see any areas of the vane where allowable velocity limits have been breached. These high velocity areas can be flattened out by increasing the number of vanes, adding perforated plates, or changing the baffling devices used.
Contours can also depict velocity levels over free liquid surfaces where entrainment can start. If the velocity levels are higher than the allowable limits, then design changes can be made to prevent entrainment.
Turbulence properties also can be viewed with contours. Detrimental levels can be reduced through design changes.
TWO-PHASE FLOW
Droplet trajectories are computed based upon the computed single-phase flow field. Droplets can be introduced at any of the inlet cells and their trajectories computed using the equations for drag and inertia that govern motion.
Thus, even vane geometries can be modeled and checked for their separation efficiency for any gas and liquid properties (Fig. 2c).
Fluent uses the laws of settling (or drag) to compute accurate trajectories. The random effects of turbulence can also be included to simulate their influence upon droplet trajectories.
Two-phase flow analysis enables a separator designer to provide for the necessary separation elements. Knowledge of droplet sizes is also important in accurate CFD modeling.
As an auxiliary to CFD, empirical-based analysis of two-phase flow in the inlet nozzle is helpful. Estimates of droplet and mass distributions are valuable to both CFD modeling and any retrofit design.
A number of methods exist for estimating droplet sizes in two-phase flow. Most rely upon air and water experiments. Some use other fluids and pressures. Central to most formulations, however, is the Weber number:5
We=(pg Vg2d)/s (4)
where:
pg=Gas density
Vg=Slip velocity
d=Droplet diameter
s=Liquid surface tension
For inviscid fluids, the critical value of the Weber number above which droplets will shatter is 10-20. For viscous fluids, the Weber number can be adjusted to reflect the increased stability of viscosity.5
The mean droplet sizes in two-phase annular flow are generally thought to be proportional to the inverse of the square root of the Weber number. However, because of coalescence which results from high liquid-to-gas ratios, a second term is necessary. This term is proportional to the liquid-to-gas ratio, with coefficients determined from experiments.6
CFD models can be used to simulate two-phase flow in the primary inlet chamber of a vessel. Such simulations can reveal problems with baffles or other primary separation devices.
The walls and baffle surfaces can be made to trap (or separate) droplets which impinge. Although droplet coalescence is not simulated in CFD models, this analysis provides a conservative method for estimating inlet chamber separation. The analysis also furnishes a means to compare separation performance of different inlet chambers.
Because two-phase flow analysis also provides an estimate of the percentage of entrainment, optimum sizing and determination of the proper liquid capacity and drainage system for a given unit is also made possible.
The percent of entrainment can be calculated from the Wallis correlation that is based upon a dimensionless number:7
p1=Vg(m/s)(pg/p1)0.5 (5)
where:
m=Gas viscosity
p1=Liquid density
The other variables have been defined previously.
The correlation takes the form of an S-curve as shown in Fig. 2d.
VESSEL RETROFIT
As gas production increases in a typical oil and gas processing application, the gas capacity of the gas/liquid separators is eventually breached. At this point, reentrainment of separated liquids begins. The reentrainment increases exponentially as the gas flow rate is increased.
Thus, many producers are forced to add gas capacity either through additional vessels or revamped internals. The new internals use the existing shell, but are capable of higher capacity than the previous designs. In many cases, this alternative is more practical and economical than a complete vessel replacement.
The engineering of separator internals is continually evolving into designs of higher capacity and efficiency. These improvements lead to smaller shell diameters. Therefore, yesterday's capacity limitations based upon shell diameters are usually understated.
Often, a separator of older design can accommodate new high-capacity internals, thus increasing the separator's capacity by as much as 50% or more.
PRELIMINARY DESIGN
The retrofit process is best carried out by a specialist in gas/liquid separation. This process involves a combination of analytical, experimental, and field application experience to be successful and yield the greatest increase in capacity for the dollar spent.
These skills are typically not found outside the realm of a specialist with a lot of experience.
The retrofit of a gas production vessel can be used as a good example. This large, 7,600 x 1,677 mm (25 x 5.5 ft) ID seam-to-seam, high-pressure horizontal gas scrubber was handling only half its ultimate capacity.
Mesh pads were used as the final separation internals and, since no primary baffle was present, the principal separation mechanism was gravity (Fig. 3a).
The V-bank design is a classic vane separator arrangement which maximizes the capacity of horizontal shells in many applications.
As shown in Fig. 4, the flow usually enters one end of the V-bank and must turn 90 to enter the vanes on both sides of the V-bank. After passing through the vanes, the flow contracts to exit through the nozzle (or nozzles).
The design is sensitive to flow momentum at the entrance and special consideration must be given to the flow rate pV2 at this point.
Allowables are defined at these points based on experimental testing. The vane face velocity also has experimentally allowable pV2 limits, and these limits act as design guides in the preliminary design phase of a retrofit study.
Based upon pV2 allowables, the V-bank vane internals shown in Fig. 4 could be projected to provide a 100% increase in handling capacity.
Because the horizontal shell of the vessel was long enough to allow the installation of the V-bank internals, the vessel was a perfect candidate for a vane retrofit.
FEED PIPE ANALYSIS
The droplet size distribution (Fig. 5), determined from two-phase flow analysis, is important to the final separation stage. In this case, the analysis revealed a mass median of 9 m and a most-frequent size of 5 m.
This size distribution and frequency were outside the separation capability of a standard vane-type separator, and thus special internals such as fibrous coalescers were required.
CFD ANALYSIS
After determining the preliminary retrofit design, the design was further evaluated by computational fluid dynamics (CFD) analysis. The CFD analysis checked how the gas and liquid flows were affected throughout the vessel by the revised internal design.
The CFD analysis of separator internals is in many respects similar to a physical test. Indeed, laboratory tests performed on various arrangements match CFD results remarkably well.
Typically, velocity vectors match physical measurements to within 5%. Because the computer analysis closely matches the results of physical testing, CFD studies can make accurate gas capacity predictions.
Because improper grid design often leads to erroneous computational results, significant expertise in CFD separator modeling is required to construct these models.
The grid for this case study (Fig. 6a) required 24,000 nodes and approximately 1,800 iterations to obtain results. In this case, the vessel shell and internal features were modeled mathematically using a grid network of computational nodes.
Fig. 6b illustrates the steady-state solution of the flow field. Contour plots of velocity are useful for checking the allowable deviation in the vane separator.
In this V-bank separator, a plot of the contours of velocity magnitude (Fig. 6c) revealed that the maximum velocity did not breach the allowable level determined during physical testing. Similarly, plots of turbulence (Fig. 6d) and pressure (Fig. 6e) confirmed the quality of the design.
No shattering turbulence or pressure areas were detected.
FINAL DESIGN
Replacement elements were double V-bank mist extractors with wire mesh agglomerators. One V-bank unit was fitted beneath each of the two top outlet nozzles at opposite ends of the scrubbers (Fig. 3b). Individual vane assemblies were fitted into the housings and bolted into place.
When the retrofit was performance-tested, full gas capacity of 500 MMscfd was achieved. No trace of carryover of hydrocarbon liquid was seen in the downstream glycol contactors.
The retrofit has been in service for a number of years, and performance of the vessel and separation equipment has been exceptional.
Results such as these attest to the reliability of computational analysis combined with experimental design data. The technique yields the most complete assurance of separator design quality.
REFERENCES
- Fewel, K.J, and Kean, J.A., "Vane Separators in Gas/Liquid Separation," ASME Energy Sources Technology Conference, January 1992.
- Talavera, P.G., "Selecting Gas/Liquid Separators," Hydrocarbon Processing, June 1990, p. 83.
- Calvert, et al., Entrainment Separators for Scrubbers, Fifth Edition, APT Inc., October 1974, p. 44.
- Fluent Users' Manual, Vol. 3.03, p. 213, Mar. 6, 1990.
- Wallis, G. B., One-Dimensional Two-Phase Flow, McGraw-Hill, New York, 1969, pp. 376-77.
- Azzopardi, B.J., "Drop Sizes in Annular Two-Phase Flow," Experiments in Fluids, 1985, p. 53.
- Steen, D.A., and Wallis, G.B., AEC Report NYC-3114-2, 1964, p. 69.
Copyright 1992 Oil & Gas Journal. All Rights Reserved.
