You are on page 1of 6

Troubleshooting refinery equipment

with multiphase CFD modelling


Computational fluid dynamics is a useful and increasingly practical tool for improving
the design of and increasing the understanding of common process equipment
Grant Niccum and Steve White Process Consulting Services

uch of the common process equipment in


refineries today was designed and built
according to traditional empirical design
methodologies that were developed decades ago.
Without an intimate understanding of the
complex flow patterns present within a given
system, designers had to rely on conservative
assumptions and trial and error to ensure that
equipment met design requirements. Modern
computational fluid dynamics (CFD) tools allow
designers to pull back the veil on complex internal flows, but their use has been limited by
available computing power. As computing power
continues to increase, CFD is becoming a practical tool for industrial scale problems. Through a
deeper understanding of standard process equipment, it is now possible to identify opportunities
to improve both the function and the capacity of
installed systems.
In many cases, small modifications can eliminate the need to design and fabricate new
equipment, resulting in significant cost savings
without compromising performance. This article
will discuss several cases demonstrating the
application of CFD to traditional process equipment. Each case presented will discuss the
motivation for the use of CFD, the assumptions
required to yield a practical and robust CFD
simulation, some details pertaining to the CFD
modelling itself, and most importantly the
practical outcome of the simulation exercise.
Central to the growing popularity of CFD for
industrial scale problems is an ability to simplify
a simulation. An extremely detailed simulation
incorporating all of the relevant physical minutiae is of little value if the results cannot be
interpreted and applied to solve a real world

www.digitalrefining.com/article/1000894

problem. Furthermore, the additional complexity


and computational expense required to perform
an extremely high fidelity simulation is often
unjustifiable or unattainable for many industrial
problems. In most cases, a simplified modelling
approach specifically developed to examine the
variable(s) of interest is the most efficient practice. By carefully considering all of the
independent and dependent variables relevant to
the design question at hand, the pain and
expense of a CFD project can be greatly reduced.
Just because an engineer can solve for every
possible variable throughout an entire domain
doesnt mean that he or she should. Unnecessary
physics complicates the setup of a simulation,
significantly increases the computational time
required, and may decrease the stability of the
simulation to the point where a converged solution is impossible.

Case 1: liquid knockout drum


After performing a detailed dynamic process
simulation study of a particular unit, it was
discovered that, given the right circumstances,
vapour/liquid rates could be far above the design
capacity of an existing liquid knockout drum. In
addition to incurring significant expense,
replacement of the new drum would have been
difficult due to space restrictions. It was hypothesised that internals could be added to the drum
to adequately increase the vapour-liquid separation. For verification, CFD could be used to
confirm the effectiveness of any design changes
that would see the drum operate while significantly under-sized according to traditional sizing
methods. Particle sizes of a certain critical diameter were considered the break point for effective

PTQ Q1 2014 1

operation of the separator. Therefore, the CFD


analysis was used to develop and test modifications to allow the drum to effectively trap
particles with larger than acceptable diameters
within the drum.
A brute force CFD approach to this design
problem would have been to model all of the
relevant physical phenomena at the same time:
multiphase vapour/liquid flow at the inlet,
breakup/coalescence of the liquid droplets,
formation of a liquid film on the walls of the
drum, collection/movement of free liquid in the
bottom of the drum, and so on. The modelling
task was greatly simplified, however, by carefully
considering the variables of interest. The functions of the liquid knockout drum are to separate
and collect liquid particles, and to prevent re-entrainment of the free liquid phase that has
collected in the bottom of the drum. The variables that needed to be evaluated to verify that
the drum would perform as required are the
fates of liquid particles entrained with the gas at
the inlet to the drum and the shape/size of the
stable liquid area in the bottom of the drum.
Variables such as liquid wall film thickness are
not significant to the overall function of the
drum and were not modelled, as their omission
did not significantly affect the variables of
interest.
Modelling was further simplified by segregating the variables of interest, as they are
independent of one another. Each design option
was evaluated using one model for particle
tracking and a second for monitoring the free
liquid phase. Although it may seem counterintuitive that two models would be more efficient
than one, this arrangement allowed the designer
to perform several design iterations using the
less computationally intense particle tracking
model before running the more complex gas/
liquid interface tracking model. Furthermore,
the separation allowed the two models to be set
up quite differently to give the best answers for
the variables that each was tasked with solving.
The first of the two simulations was used to
track particles entrained with the vapour at the
drum inlet. The discrete phase model (DPM)
was chosen in this situation for its ability to
track particles through the domain and because
the volume fraction of liquid entrained within
the vapour flow was low. Small, light particles
follow vapour flow streamlines more closely than

2 PTQ Q1 2014

larger particles because they have less momentum relative to the drag caused by the bulk
vapour flow (lower Stokes number). With this
principle in mind, a conservative particle size
should be smaller than the maximum allowable
droplet size. These conservatively sized particles
were injected with the vapour at the separator
inlet and tracked throughout the domain as the
vapour travelled from the separator inlet to the
outlet. A simplifying assumption was that the
particles underwent partially elastic collisions
(some energy lost) when they encountered walls
within the vessel. Thus, wall collisions tended to
slow the liquid droplets down until they separated from the vapour flow and settled in the
bottom of the drum. In reality, some of these
collisions would have splashed to create multiple
smaller droplets, but this phenomenon was
ignored because escape of smaller droplets was
acceptable and therefore not consequential to
the design. The design was modified and simulated iteratively until no particles escaped
through the drum outlet.
The second simulation was designed to model
the stability of the free liquid phase in the bottom
of the drum. This simulation employed the
volume of fluid (VOF) model to track a well
defined vapour-liquid interface. Liquid volumetric flow rates were significantly lower than gas
volumetric flow rates, and liquid entering and
exiting the drum was not significant to the problem of maintaining a stable liquid layer. The
model was therefore built with no liquid flow in/
out, and a mass of liquid was manually placed
within the drum at the start of the simulation
and allowed to slosh around due to interaction
with the vapour flow. If the flow agitated the
liquid layer to the point where liquid mass
escaped through the outlet, the design was modified. Stabilising the free liquid layer with high
vapour flow rates proved more difficult than
trapping the initially entrained particles. The
high velocity vapour flow tended to re-entrain
significant quantities of liquid. However, a design
was developed that could satisfy both
requirements.
The use of these two complementary CFD simulations led to a robust design and confidence that
the knockout drum can perform adequately under
the given set of operating conditions. The use of
CFD in this case allowed many design iterations
to be evaluated within a matter of days to arrive

www.digitalrefining.com/article/1000894

at an optimal solution. For


Raw gasoline
comparison, a sizing exercise
volume fraction
using the traditional sizing meth1.00e+000
9.47e001
odologies was performed for a
8.95e001
8.42e001
new drum to handle the same
7.89e001
7.37e001
loads. According to traditional
6.64e001
6.32e001
sizing methods, a new drum for
5.79e001
5.26e001
the same service would be
4.74e001
4.21e001
approximately three times the
3.68e001
3.16e001
diameter of the existing drum
2.63e001
and would have a tangent-tan2.11e001
1.58e001
gent length roughly equal to the
1.05e001
5.26e002
current drum. The CFD opti0.00e+000
mised internals support the
continued use of the existing Figure 1 Contours of the volume fraction of liquid within the drum. The red
region is 100% liquid, and the blue region is 100% vapour. With the liquid
drum.

Case 2: fractionator overhead


receiver

level above the short inlet riser, liquid is entrained into the jet of vapour
entering the drum

In the case of an overhead receiver for an FCC


fractionator, CFD was used to predict the consequences of an existing design that appeared
intuitively flawed. The overhead receiver was
designed with the inlet at the bottom of the
drum, presumably to reduce the complexity of
the large diameter piping to the inlet. Although
there was a short inlet riser inside the drum, the
normal liquid level could easily exceed the height
of this riser. Intuitively, it was hypothesised that
an inlet jet agitating the fluid in the drum would
interrupt the separation process and lead to high
levels of water in the hydrocarbon outlet. CFD
was used to examine the effects of this inlet
design in detail.
Horizontal gravity separators are designed
based on cross-sectional flow velocities and fluid
residence time. Traditional methods assume plug
flow through the vessel, meaning that each phase
has a uniform cross-sectional velocity equal to
the volumetric flow rate divided by the cross-sectional area through which that fluid flows. In a
given fluid, a droplet of a certain size has a fixed
terminal velocity. The separator must be sized
with sufficient residence time for that droplet to
traverse into the proper phase at the outlet from
any starting position at the inlet. If the inlet
design leads to heavy agitation within the drum,
the most basic design assumption plug flow
through the cross-section of the drum fails to
hold. If the plug flow assumption is not met,
some fraction of the liquid in the separator may
experience residence times significantly below

www.digitalrefining.com/article/1000894

the design value, and good separation will not be


achieved.
A CFD simulation was designed to test the
validity of the plug flow assumption with the
bottom inlet design. The volume of fluid (VOF)
model was chosen for this multiphase problem.
The CFD model contained a set liquid volume
inside the drum, and the vapour load was
applied at the inlet of the drum, escaping
through the gas outlet. Since the design velocities within the liquid phase of the drum are
extremely low, inclusion of liquid mass entering
and exiting the drum would not have affected
the variables of interest (anomalous cross-sectional velocities), so this was omitted from the
simulation. Any regions with significant movement can be assumed to be operating away
from the plug flow design assumption.
Significant recirculation or shortcutting within
the liquid region of the drum should be viewed
as an indicator of poor separation that will
result in significant amounts of water in the
hydrocarbon outlet.
The results of the CFD simulation explained
the problems that this FCC unit main fractionator had been experiencing. The CFD showed that
the vapour jet leaving the riser entrained a
significant amount of liquid. Figure 1 shows the
intense agitation of the liquid within the drum.
This created a large recirculation current within
the liquid portion of the drum, completely violating the plug flow assumption used in the
separator sizing process. Based on this result, it

PTQ Q1 2014 3

was confidently predicted that


more
uniform
distribution
Bottom inlet
the liquid-liquid separation was
across the tower according to
agitates liquid
poor, and there was likely a
the placement of the pipes. By
in drum
large amount of water in the
fully understanding the mechahydrocarbon outlet, which feeds
nisms that allow this inlet
the fractionator reflux pumps.
device to achieve good results,
This water contains ionic
better design decisions can be
compounds that will deposit in
made in future installations.
Water in reflux
the tower as salts as the water
The vapour was modelled as
carries salts
evaporates (see Figure 2). The
incompressible due to the relaSalt deposition
tower was, indeed, experiencing
tively low pressure drop in the
plugs trays
issues related to salt in the overdomain. Detailed simulations
head system, and the circulation Figure 2 Poor separation in the
were run with and without the
of large quantities of water in overhead receiver leaves water
inlet distributor, and the results
the reflux is a contributing in the reflux, which carries ionic
were compared in an effort to
factor. Lab samples have compounds that will deposit in
find meaningful differences that
confirmed that the weight the tower as salts when the water
would explain why the distribuevaporates. These salts can cause
percentage of water in the reflux corrosion and plugging of tower
tor had such a noticeable effect
stream is very high.
when installed. The discrete
internals
The next application of CFD
phase model (DPM) was used to
to this drum will be to correct
simulate the fate of liquid dropthe deficiencies in the inlet design. By incorpo- lets entrained with the feed. With no inlet
rating CFD into the revamp process, distributor, the CFD model predicted that the
modifications can be tested and an iterative inlet vapour would form a well defined jet before
design developed to ensure that the recirculation impinging on the opposite wall of the tower.
zones are eliminated and that the flow in the Above the jet, the flow upward through the
drum more closely resembles the plug flow tower was far from uniformly distributed.
assumed in the initial vessel sizing calculations. Furthermore, the stability of the jet suggested
Correction of this problem will eliminate the that the impingement point on the wall opposite
salting problems occurring in the tower.
the inlet was unlikely to move over time. This
created a stagnation region at the impingement
Case 3: FCC fractionator feed inlet device
point with essentially zero flow velocity and no
A common feed inlet device for FCC fractiona- wall shear stress.
tors is a series of pipes intended to break up and
While the impingement point saw little vapour
distribute the flow within the
flow velocity, the rate at which it
column
(see
Figure
3).
was impacted by any liquid
Experience has shown that,
particles in the feed was high. If
when installed properly, this
these particles were sufficiently
design improves tower perforlarge (high Stokes number), the
Inlet device
mance and prevents coking on Reactor
particles tended to detach from
the tower wall opposite the effluent
the vapour flow and follow a
inlet. A CFD simulation was
trajectory dictated by their
Quench
performed to understand the
momentum. Thus, as the vapour
mechanisms by which this
streamlines turned abruptly at
device works and to look for
the stagnation point, the liquid
simple improvements to the
continued along the path of the
design that could increase
jet and impacted the wall in the
performance or reduce complex- Figure 3 Field experience has shown stagnation region. The high
ity and cost. The intuitive theory that installing a feed inlet device
liquid impact rate combined
going into the study was that similar to the one shown improves
with near zero vapour velocities,
the flow was redirected upward performance and reduces coking on and low wall shear stress
by the pipes, thus creating a the rear wall of the tower
created favourable conditions

4 PTQ Q1 2014

www.digitalrefining.com/article/1000894

for coke formation in the stagVelocity, ft/s


nation region.
1.07e+002
Contrary to expectations, the
1.00e+002
9.33e+001
addition of the inlet distributor
8.67e+001
8.00e+001
to the model did not greatly
7.33e+001
6.67e+001
improve tower cross-sectional
6.00e+001
5.33e+001
flow distribution (see Figure 4).
4.67e+001
4.00e+001
The
simulation
nonetheless
3.33e+001
2.67e+001
provided evidence to explain the
2.00e+001
1.33e+001
decreased coke formation on the
6.67e+000
0.00e+000
wall opposite the inlet. The inlet
device caused the inlet jet to
widen more in the horizontal
direction, but had little effect on Figure 4 Velocity profile on a vertical plane within the tower. The case with
the vertical size of the jet the inlet device (left) is similar to the base case (right). Significant quantisignificant vapour quantities ties of vapour are not directed upward by the inlet device
were not redirected upwards into
the tower with each impact. This was due to each impact velocities and high wall shear forces
pipe only blocking a small part of the flow. Any around the pipes caused the droplets to shatter
vapour that was deflected near a pipe was easily and break up into many smaller droplets. These
corrected back into the mean jet flow by the smaller droplets had lower Stokes numbers,
momentum of the surrounding jet that had not meaning that their momentum was less significant in relation to the drag caused by the vapour,
directly impacted one of the distributor pipes.
If the pipes did not effectively redistribute the and the droplets were more likely to follow the
feed within the tower, how did they decrease path of the bulk vapour flow rather than impactcoke formation on the opposite wall of the ing the rear wall of the drum. By impacting and
tower (the stagnation region) as seen in the shattering the droplets on the pipes, where wall
field? The CFD shows that the pipes acted to shear is at its maximum, rather than in the stagscatter the jet on the wall of the tower rather nation region at the point of jet impingement,
than to distribute the feed. As the jet flowed coke formation is minimised. Armed with an
around the pipes, it became less defined, mean- understanding of the fundamental mechanisms
ing that the stagnation region was not as behind this style of inlet device, future designs
apparent as it was without the pipes. The stag- can be adjusted to maximise the positive funcnation region was broken up over the back wall tions while minimising negative consequences.
of the tower rather than remaining stably
focused on a single point. The disruption of the Conclusion
stagnation region greatly reduced coke forma- CFD has been applied to these case studies to
tion for two reasons: 1) the liquid impacting the gain a better understanding of the phenomena
back of the tower did so over a larger area, that govern the performance of examples of
concentrating less material on one spot; and 2) common process equipment. The designer can
the area immediately surrounding the impinge- take advantage of this better understanding of
ment point had the highest wall shear stresses the underlying physical phenomena to improve
of anywhere on the tower wall. As the impinge- performance. For example, in the case of the
ment region was less defined, the shear stresses FCC unit fractionator inlet distributor, underwere more evenly distributed, contributing to standing that droplet breakup and jet scattering
the shearing of any coke.
are the two key principles that allow this device
Droplet breakup is another factor that may to prevent excessive coke formation will lead to
contribute to a reduction in coke formation with future designs that specifically maximise these
the inlet device installed. As the vapour flowed phenomena rather than designs that attempt to
around the pipes, larger droplets with high improve the device through misguided attempts
momentum could not change direction fast to affect the distribution of the flow within the
enough to avoid a collision with the pipes. High tower.

www.digitalrefining.com/article/1000894

PTQ Q1 2014 5

CFD is an extremely useful and increasingly


practical tool for improving the design of and
increasing the understanding of common process
equipment. By carefully defining a problem and
the variables of interest, simulations can be built
that give good information where necessary
without incorporating unnecessary complexity.
Finally, it is worth mentioning here that CFD
inputs, assumptions and results should be carefully scrutinised. Commercial CFD software will
work dutifully on any problem that it is given,
and will often reach a converged answer, but
the software has no notion of correctness.
Incorrect inputs or the failure to include important physical aspects of a simulation will lead to
incorrect answers. Part of the engineers responsibility in running a CFD simulation is to
consider all of the trade-offs between complexity
and accuracy so that an adequate solution can be
developed with the minimum level of computational expense. The ultimate usefulness of a
simulation lies in an engineers ability to use the

6 PTQ Q1 2014

information to draw concrete conclusions and


develop solutions that improve performance.
Grant Niccum is a Process Engineer with Process Consulting
Services, Inc. in Houston, Texas. Process Consulting Services
provides grassroots and revamp front-end process engineering to
the refinery industry worldwide.
Steve White is a Chemical Engineer with Process Consulting
Services. He has more than 37 years of process design experience
in refinery revamps and grassroots units including crude/
vacuum, FCC, hydrotreater, alky, butamer, reformers and others.
He previously worked for Jacobs Engineering, UOP and ARCO.

LINKS
More articles from: Process Consulting Services
More articles from the following categories:
Process Modelling & Simulation

www.digitalrefining.com/article/1000894

You might also like