Professional Documents
Culture Documents
Flow Through Packed Beds PDF
Flow Through Packed Beds PDF
Flow Through Packed Beds PDF
Group 11
Yuan Jia, Yan Li, David Hlavka
April 29, 2009
Abstract
Our project studied the pressure drop across a packed bed and compared the data
to the analytical solution governed by Ergun’s equation (Ergun, 1952). Different types of
material and varying airflow rates were used to effect the range of our Reynolds number
and subsequent pressure drop. We found that there is a significant relationship between
pressure, frictional factor and Reynolds number with respect to flow rate and inlet
pressure. We used the change in pressure to determine two constants (k1 and k2) that
were then compared to Ergun’s equation. In the end we discovered that our experimental
1
Introduction
There are many engineering applications that utilize a packed bed. The forms
more commonly implemented today are incorporated with systems involving adsorption
of a solute, distillation, filtration and separation (e.g. water purification via waste
removal). In an ion exchange or catalytic reactor, a single fluid (liquid or gas) flows
desired pressure drop over the bed increases the diffusion efficiency of the reagents. In a
filtration system, pressure drop is crucial because fluid flows through the filter medium
by virtue of a pressure differential across the bed (Geankoplis, 2003). Studying the
pressure drop across the bed is therefore essential to determine filtration efficiency and
expected time span before excessive buildup of filtered material occurs. Packed beds can
be made in various forms that are designed in accordance with their application. For our
experiment, we used approximately 0.0085 m3 (0.3 ft3) of glass beads with a diameter 15
mm and copper-coated metal beads with a diameter of 4.5 mm. Other possible materials
like sand were considered, but due to time constraints not tested.
Our goal was to study the pressure drop of one fluid (air) flowing through a
variety of packed materials. This started with approximately 0.0085 m3 (0.3 ft3) of glass
determined the diameter of the beads to be too great to induce a readable pressure drop
with our current gage. The only way to remedy this issue was to either install more
accurate gages that would incur greater cost and take more time to order or use smaller
sized beads. We decided to use 4.5 mm copper-coated metal beads, for these were
2
Once we knew how and what we were going to test, we needed to find a way to
predict our results. Since pressure drop depends significantly on the Reynolds number,
we needed an equation that described flow behavior in both the laminar and turbulent
regions. We found that the Ergun equation (Ergun, 1952), which will be explained next,
Formulation:
the equivalent diameter of the particle Reynolds number vs. friction factor
Dp, the dimensionless void fraction , and , the density of the fluid (Fig. 1).
𝜌𝑣𝑠 𝐷𝑝
𝑅𝑒𝑝 = (1)
𝜇 1−𝜀
Ergun also defined that the friction factor, fp depends on νs, the pressure drop, p ,
∆𝑝 𝐷𝑝 𝜀3
𝑓𝑝 = ( ) (2)
𝐿 𝜌𝑣𝑠 2 1−𝜀
3
There are two independent equations that deal with flow behavior that have
different Reynolds numbers. When Rep < 10, the Blake –Kozeny equation suggests that
(1−𝜀)2 𝜇
𝑓𝑓 = 75 (3)
𝜀3 𝜌 𝑣𝑠 𝐷𝑝
When Rep > 1000, or in turbulent flows, the viscous force of the fluid is
insignificant. This means the Burke-Plummer equation describes the friction factor as
7 (1−𝜀)
𝑓𝑓 = 8 (4)
𝜀3
The Ergun equation (Ergun, 1952), which covers the entire range of flow rate, can
then be obtained by assuming that the viscous losses and the kinetic losses are additive,
(1−𝜀)2 𝜇 7 (1−𝜀)
𝑓𝑓 = 75 +8 (5)
𝜀3 𝜌𝑣𝑠 𝐷𝑝 𝜀3
After experiments with different packing material and different flow rate, Ergun
150
𝑓𝑝 = + 1.75 (6)
𝑅𝑒 𝑠
4
Apparatus Design and Calibration
Before we could begin testing, we needed to create a device that would accurately
able to control how much air left the apparatus. After the upper portion was completed
and sealed properly, we attached a flow meter and valve onto the base of the plenum to
accurately monitor the flow going into the column. The last modification we made to the
device was the addition of a pressure gage above our bed. Since there was already a port
5
made for a pressure gage below the bed, we merely mimicked this method and created a
second port approximately 30.48 cm (12 in) above the base of the column. The finished
calibrated the system via mercury filled U-tube manometer, because we did not know the
pressure gage sensitivity we would require. From our calculations we found our pressure
cm (4 in) in diameter.
acquired an appropriate gage from 0 to 41.4 kPa (0 to 6 psi). Next we calibrated the new
pressure gage via LabVIEW and a tank draining operation. We used a similar set up and
procedure as the tank draining experiment previously conducted in the semester. Our VI
can be seen in Fig. 3. Finally in order to ensure that our seals were generally working to
attached our pressure gage and filled the apparatus with air. We then let it sit for
approximately ten minutes and observed any pressure changes. Since there were no large
6
observable deviations, we were safe to assume that the apparatus was properly sealed and
functioning.
Experimental Procedure/Data
Before we could begin testing our pressure gradients, we had to calibrate our new
pressure gage. To do this, we followed the same procedure as seen in the tank draining
experiment done earlier in the semester. We filled a column with 500 mL of water until
Once we filled the tank to our desired height, we allowed the tank to drain out the base
with an inlet pressure of 137895 Pa (20 psi). Once five seconds elapsed, we found no
results. After we tried multiple pressure ranges ([137895 Pa (20 psi) – 344738 Pa (50
psi)] at 68948 Pa (10 psi) increments), we determined that the current configuration was
7
inadequate to extrapolate any usable data. A smaller diameter bead was necessary to
Fig. 6 Friction Factor vs. Re Number Fig. 7 Pressure Drop vs. Velocity
we knew our configuration was appropriate for proper data collection. We moved the VI
collection time up to 10 seconds and, starting at 137895 Pa (20 psi), moved up to 344738
Pa (50 psi) at 34474 Pa (5 psi) increments. This gave us enough data to see a significant
8
relationship between inlet velocity and pressure change and frictional factor and
When analyzing the first compiled data, we found a large error to be present (in
excess of 1000%). Upon further inspection of our apparatus, we discovered that a flow
meter intended for water was being used instead of a meter intended for air. We replaced
the flow meter, but due to time constraints, could not get one that was able to read the
ranges we were operating in. Because of this, we were forced to come up with a means
of predicting the values of our flow rate at higher-pressure ranges. Since there was a
predict where
higher pressure (pressures exceeding 172369Pa (25 psi)). When we had the proper flow
meter installed, we recalibrated our gage and found our conversion equation again (Fig.
8). Our RMS value for the calibration was 164, so our deviation was not too large. Since
9
we were operating at room temperature with standard humidity, we converted our flow
rate from scfm to cfm via a one to one relationship using the standard conversion
equation (8).
Anything above that was too great for Fig. 9 Friction Factor vs. Re Number
Fig. 10 Friction Factor vs. 1/Re Fig. 11 Pressure Drop vs. Velocity
10
our system to handle, so we were unable to generate any higher pressures. After we
gathered the data, we analyzed it and came to conclusions from there (Fig. 9, 10 and 11).
Discussion
When looking at the graphs, one can see that there is deviation between our data
and the theoretical. Since we were working with such small changes in pressure (no
greater than 689.5 Pa (0.1 psi), what appears to be large on the graph is not very large at
all. Each error bar on the graph has a maximum value of 20%, so our error does not
exceed 20% when comparing Reynolds numbers. Looking at the inverse Reynolds graph
and taking the trend line gives us an equation from which we can get our k values (k1 =
650.61 and k2 = 1.1232). Comparing the experimental with the theoretical, we see that
our values, while not exactly the same, are within an acceptable range that is around 20%.
The only area where there is a greater than 20% variance is in our inlet vs.
velocity graph. Since we did not have the most accurate flow meter, this can explain the
large deviation of our graph. Even though the deviation exists, our graph mimics the
orientation of the theoretical curve, but is merely lower than expected. If we had enough
time to acquire the appropriate flow meter, there is a greater possibility that the deviance
Overall our test was a success, but there were many areas that contributed to error
in our results. Most importantly was maintaining a full seal throughout the device. Since
we had an apparatus that was previously used, the seals were not as strong and had the
potential to leak or rupture due to degradation over time. During our initial
11
testing/calibration, the seal around the plenum ruptured due to a large internal pressure
that rapidly accumulated from too much inlet pressure. Once the integrity of the seal was
compromised, we had to reseal the entire apparatus to ensure no other seals burst during
actual testing. As mentioned above, we attempted to check the seals to ensure accuracy
in our results.
A second error was pipe leakage from the inlet. This discovery was made when
we observed a large variance between our experimental and theoretical results. After
use data variance as our check. We reduced the overall leakage by using Teflon tape on
the connection threading and ensured that the fittings were tight. Even with these
precautions, there were still minor leaks that could not be avoided.
A third significant source of error was found in our flow meter. As mentioned
above, we were using the incorrect flow meter initially and this caused us to have a large
error (approximately 1000%). Once we had the appropriate flow meter, it was too small
for the flow ranges we were operating in. Since we did not have enough time to acquire a
better meter, we came up with a linear correlation factor of 1.2 and used this to find our
flow rates. We did not have enough time to obtain a better flow meter that could retrieve
more accurate results so we had to assume our method was accurate enough. This error
is most noticeable when correlating inlet velocity and pressure drop as seen in Fig. 11.
A final source of error was the derivation of Ergun’s equation. Ergun derived his
equation from experimental data that he acquired from repeated testing of various
12
materials in a packed bed fashion (Ergun, 1952). He did not account for all the variables,
but made a general equation from his macro level observations. For one thing, he did not
that surface roughness does in fact alter the pressure, but not by a significant amount at
the macro level (Kececioglu, 1994). Because of the derivation being from empirical data
only, there is a level of error between our data and what Ergun derived.
In the end we came to the conclusion that Ergun’s equation is generally true on
the macro level, for our data. We did, however, find that there are a few inconsistencies
present within the equation that in turn caused variances between our theoretical and
actual findings. Further analysis could attempt to find just how inaccurate the equation
can be.
Further Study
Due to time constraints, we were unable to complete all the testing proposed at
the beginning of the semester. Because of this, there are many areas of further research
that can be done to advance the understanding of flow over packed beds. Listed below
are a few ideas that we thought would be interesting for further research.
As mentioned above, the first area of study could be trying different aggregates of
varying size, shape and surface roughness to establish the inaccuracies present in Ergun’s
would behave compared to ours. Also studying finer aggregate (i.e. sand) would be
another area of research. We attempted to study sand, but ran into difficulty establishing
13
an ideal packed bed and actual void fractions present. We ran out of time before we
could come to any significant conclusions, but thought a vibratory rig would be
A second area that one can study is how a more viscous fluid behaves over a
packed bed. These fluids can be as simple as water or as complex as heavy oils, but there
would be some interesting results found when taking our testing procedure and applying
it to various fluids. One could also try the varying fluids over different sizes of packed
A final area of study would be to test various bed heights and widths at a
consistent inlet velocity. Our findings seem to point to the fact that the taller the bed, the
larger the pressure change and vice versa. We didn’t, however, have the time to
empirically prove this true through testing, so any further testing in this area could have
interesting conclusions.
14
References
Carpinlioglu, M.O., Ozahi, E. “A simplified correlation for fixed bed pressure drop”
Powder Technology 187 94-101 (2008).
Ergun, S., Orning, A.A. "Fluid Flow Through Packed Columns", Chemical
Engineering Progress. 48 89-94 (1952).
Geankoplis, Christie J. 2003 Transport Process and Unit Operations. 4th ed chapter 3
Prentice Hall: New Jersey.
Kececioglu, I., Jiang, Y “Flow Through Porous Media of Packed Spheres Saturated With
Water ” Transactions of the ASME 116 164-170 (1994).
15
Appendix A
𝑠𝑝
𝑎𝑣 = 𝑣 (2)
𝑝
where Sp is the surface area of a particle in m2 and vp the volume of a particle in m3. For a
spherical particle,
6
𝑎𝑣 = 𝐷 (3)
𝑝
6
𝐷𝑝 = 𝑎 (4)
𝑣
6
𝑎 = 𝑎𝑣 1 − 𝜀 = 1−𝜀 (5)
𝐷𝑝
where 𝛼 is the ratio of total surface area in the bed to total volume of bed (void volume
The average interstitial velocity in the bed is v m/s and it is related to the superficial
1
Geankoplis, Christie J. 2003 Transport Process and Unit Operations. 4th ed chapter 3 Prentice Hall: New
Jersey.
16
𝑣𝑠 = 𝜀𝑣 (6)
The hydraulic radius rH for the flow is defined as
𝜀
𝑟𝐻 = 6(1−𝜀) 𝐷𝑝 (8)
Since the equivalent diameter F for a channel is 𝐷 = 4𝑟𝐻 , the Reynolds number for a
4𝑟𝐻 𝑣𝑝 4𝜀 𝑣𝑠 𝜌 4 𝐷𝑝 𝑣𝑠 𝜌
𝑅𝑒𝑝 = = 6(1−𝜀) 𝐷𝑝 = 6(1−𝜀) (9)
𝜇 𝜀 𝜇 𝜇
For packed beds Ergun defined the Reynolds number as above but without the 4/6 term:
𝐷𝑝 𝑣𝑠 𝜌
𝑅𝑒𝑝 = 𝜇 (1−𝜀) (10)
velocity in tube, m/s; D is inside diameter, m; and (L2-L1) is length of straight tube.
17
32𝜇𝑣 ∆𝐿 32𝜇 (𝑣𝑠 𝜀)∆𝐿 72𝜇 𝑣𝑠 ∆𝐿(1−𝜀)2
∆𝑝 = = = (12)
𝐷2 (4𝑟𝐻 )2 𝜀 3 𝐷𝑝 2
The true ∆𝐿 is large because of the tortuous path and use of the hydraulic radius predicts
too large a vs. which gives the Blake-Kozeny equation for laminar flow, void fractions
less that 0.5, effective particle diameter Dp, and Reynolds number < 10.
150𝜇 𝑣𝑠 ∆𝐿 (1−𝜀)2
∆𝑝 = (13)
𝐷𝑝 2 𝜀3
For turbulent flow in packed beds, the same procedure was used. First the pressure drop
is defined as
∆𝐿 𝑣 2
∆𝑝 = 4𝑓𝜌 (14)
𝐷 2
Then combining equation (6), (8) and (14)
3𝑓𝜌 𝑣𝑠 2 ∆𝐿 1−𝜀
∆𝑝 = (15)
𝐷𝑝 𝜀3
For highly turbulent flow the friction factor should approach a constant value. Also, it is
assumed that all packed beds should have the same relative roughness. Experimental data
indicated that 3f = 1.75. Hence, the final equation for turbulent flow for Reynolds
1.75𝜌 𝑣𝑠 2 ∆𝐿 1−𝜀
∆𝑝 = (16)
𝐷𝑝 𝜀3
Adding equation (13) for laminar flow and equation (16) for turbulent flow, Ergun
proposed the following general equation for low, intermediate, and high Reynolds
18
150𝜇 𝑣𝑠 ∆𝐿 (1−𝜀)2 1.75𝜌 𝑣𝑠 2 ∆𝐿 1−𝜀
∆𝑝 = + (17)
𝐷𝑝 2 𝜀3 𝐷𝑝 𝜀3
∆𝑝𝜌 𝐷𝑝 𝜀 3 150
= 𝑅𝑒 + 1.75 (18)
(𝑣𝑠 𝜌)2 ∆𝐿 1−𝜀 𝑝
19
Appendix B
An open circuit blower type air flow test set up consisting of horizontal and vertical rigid
pipes of inner diameter D of 103 mm (Fig.B1) was designed and constructed Air flow
was generated by means of a fan in combination with an AC motor speed control unit
(Siemens Micromaster Type MM 110). The packed bed location on the vertical pipe line
was at Xv = 1050 mm from the 90° elbow. Packed beds were made up of uniformly sized
spherical and non-spherical particles of irregular shape covering a range of Φ (spherocity,
where 1 is a perfect sphere) where 0.55 ≤ Φ ≤ 1.00 with the form of random loose
packing. Crushed solid particles of natural zeolite, chick-pea and spherical particles of
glass beads were used as packing materials. Packed beds of four different L providing a
range of 0.24 ≤ L / D ≤ 1.46 were constructed. Bed limiters were made up of synthetic
cloth with an open area ratio, β = 0.92, wire diameter dw = 0.39 mm and mesh size
M = 1.8 mm. The utilized packed bed codes, measured bed porosities and the range of the
conducted test cases in terms of Rep are given in Table 2.
2
Carpinlioglu, M.O., Ozahi, E. 2008 “A simplified correlation for fixed bed pressure drop”
Powder Technology 187(1) 94-101.
20
Table 1: Packed bed parameters and range of test cases
Packed bed Packing material ρm L D / Dp Φ ε Rep of test
code (kg/m3) (mm) cases
Z0640 Zeolite Dp = 6 mm 640 40 17.16 0.82 0.37 1104, 1142,
1187
Z0660 60 0.36 687, 950,
1075, 1125
Z06150 150 0.36 675, 837,
912, 1009
C0925 Chickpea 970 25 11.44 0.55 0.44 1307, 2464,
Dp = 9 mm 2592, 2721
C0940 40 0.40 1200, 1850,
2180, 2300
C0960 60 0.40 1200, 1850,
2150, 2280
C09150 150 0.38 1596, 1761,
1838
G1625 Glass bead 2675 25 6.43 1.00 0.44 2380, 4647,
Dp = 16 mm 4761, 5047
G1640 40 0.40 2168, 4177,
4408, 4462
G1660 60 0.39 2115
Z1825 Zeolite Dp = 18 mm 640 25 5.72 0.76 0.56 3954, 6818,
6981, 7772
Z1840 40 0.48 3000, 5630,
5815, 6046
Z1860 60 0.45 2792, 4887,
5192, 5541
Z18150 150 0.40 2400
21
Table 2: The measured mean exit velocity and pressure drop data for all test cases
22
Appendix C
Our first task was to build an apparatus that could hold our sand. The tank holding
our sand was built of acrylic glass, since it was necessary to build the tank out of a clear
substance in order to observe the height and behavior of the sand. Acrylic glass was
readily available in the lab. Our tank was chosen with dimensions similar to those about
which we had read. It is cylindrical with a 10 cm diameter and it is over a meter tall. This
seems large, but the height is necessary to contain splashes from bubbling sand.
Next, we needed to properly aerate the sand. Since our air comes from a
compressor via small lines, we have a plenum designed to uniformly (roughly) push air
into our tank, rather than shoot a jet of high velocity air directly through the center of the
sand. This plenum is also made of cylindrical acrylic glass, and had to be partially
machined, making it the second most expensive part of our experiment. We also had to
create an effective seal between the plenum and the tank to keep the sand from falling
into the plenum. The seal consists of a sheet of fiberglass supported by two perforated
metal sheets and air tightened by two gaskets. Fiberglass is an ideal filter as it allows easy
flow of air but still blocks our tiny sand particles. The tank is flanged, allowing it to be
3
Tully.A, Karasek.M, & Doraiswamy.S 20 February 2009
http://www.me.rochester.edu/courses/ME241/11-Sand.pdf
23