HEFAT2007

5th International Conference on Heat Transfer, Fluid Mechanics and Thermodynamics

Sun City, South Africa
Paper number: AT1

CONVECTION HEAT TRANSFER IN BAFFLED MIXING TANK

Tuomo Aho and Reijo Karvinen
Tampere University of Technology
Energy and Process Engineering
P.O.Box 589, FIN-33101 Tampere, Finland
Tuomo.Aho@tut.fi

ABSTRAC INTRODUCTION
Single -phase flow field and temperature distribution Impeller-stirred reactors are widely used in the
in a baffled tank stirred by a three- bladed impeller was chemical industry to provide effective mixing of
investigated both computationally and experimentally. chemical reactants to form desired products. The type
The computational model employed a sliding mesh of impeller determines flow patterns and therefore the
technique in fully three-dimensional grids for a small- efficiency of the mixing process. Knowledge of factors
scale unit, in which also velocity and heat transfer such as mixing efficiency, heat transfer rate, residence
measurements were made with water. Turbulence time and concentration levels are critical to the
effects were simulated using the standard k-ε model. successful operation of an impeller-stirred tank. These
Two different boundary conditions, namely, constant key operating factors are typically investigated by
heat flux and constant temperature on the wall were conducting measurements using a small pilot- or bench-
used for heat transfer simulations. Mean velocity and scale equipment. Accurate measurements in
turbulence were measured using LDA. In temperature commercial scale units are difficult and often
measurements thermocouples were used. By impossible to carry out. The smaller scale testing
comparing experimental and modelled results non- process is usually scaled up to commercial scale
dimensional variables of velocity were found, which operating conditions. Since the scaling procedure is also
gave very similar results to the pilot unit and full- size very complex, scale-up models for mixing tanks have
reactor of 12 m3, for which some modelled heat transfer been limited. Computational fluid dynamics provides a
results are given. useful method to simulate the performance of both
bench-scale and full-size units. Much work on stirred
NOMENCLATURE tank computations in three dimensions has been carried
out in the last few years using different types of
D [m] tank diameter methods. In the first stage, the impeller was replaced by
d [m] impeller diameter
H [m] tank height a jet and turbulence was handled with the two-equation
k [m2/sec3] turbulence kinetic energy models of turbulence [1]. Later on, also the impeller
LDA Laser-Doppler anemometry could be included in predictions employing a rotation
L1, L2, L3 [m] baffle distances mesh around the impeller [2]. In order to couple
N [rpm] rotational speed
NRe impeller Reynolds number together the impeller and the remainder of the tank
PIV Particle Image velocimetry there are many possibilities: steady- state, quasi-static
q [W/m2] heat flux and transient approaches [3]. As a matter of fact, the
R [m] tank radius methods are same as in the modelling of impeller
R1, R2 [m] bottom radius
Rbl [m] impeller blade radius pumps [4]. If use is made of two-equation models
r [m] radial coordinate details of turbulence cannot be found. If extended
T [°] mean temperature computational resources are available, more
U [m/s] axial mean velocity sophisticated approaches like LES can be adopted [5].
Wtip [m/s] impeller tip speed
W [m/s] tangential mean velocity
There also exist plenty of experimental papers in the
x [m/s] axial coordinate literature. Mean velocities [6] and power consumption
have been measured [7] and dimensionleess numbers
Special characters have been tried to found to descript the performance
ε [m2/sec3] turbulence dissipation
[8]. During recent years the structure of turbulence in a
μ [kg/s/m] dynamic viscosity
ρ [kg/m3] density mixing tank has also been measured with PIV and LDA
τ [sec] time [9,10].
φ [°] tangential coordinate Published data for the case of a stirred tank with heat
transfer are very limited and there exist only some
Superscripts
* dimensionless parameter
papers [11]. The objective of this investigation is to
model the velocity field, to map the temperature
distributions and to model heat transfer in a stirred tank. The calculations were performed using a non-
To obtain this goal, an impeller-stirred tank is modeled uniform multiblock grid generation technique. Four
using computational fluid dynamics simulations with a grid densities were used for checking the solution
sliding grid technique, which has been applied dependency on the grid density. Especially, at the near
successfully for the stirred tanks in earlier studies wall region, where heat transfer takes place, the grid
[2,3,12]. Simulations are carried out with two different density must be large. The fine grid consisted of
boundary conditions for heat transfer. Data from LDA 115200 cells. The surface grid for the coarse case and a
measurements of mean velocities in a bench-scale unit side cross-section for the fine grid are illustrated in
are used to validate calculated velocities. Measurements Figure 2. The solved velocity field was used for heat
of actual temperature distribution are provided as a transfer calculations as an initial guess. The fluid was
basis for checking model reliability. water, for which density and specific heat can be
assumed constant, but the effect of temperature on
STIRRED VESSEL CONFIGURATION viscosity was taken into acconunt. The no-slip
condition is applied at the surface of the vessel, baffles,
X*, U * impeller blades and shaft. The free surface is treated as
D a plane of symmetry. Heat flux through the free surface
0.875
and tank bottom was assumed to be zero.
0.75 L1
φ, W*
0.625
R
0.50 H
R1
0.375
R bl
0.25 d
R2 L2

L3

R *, V *

Figure 1 Gross-section and plan view of bench-scale

stirred tank and blade impeller.

Figure 1 shows a schematic diagram of the

experimental stirred tank and blade impeller. The
cylindrical tank with an inner diameter D of 246 mm Figure 2 Coarse surface grid and side cross-section for
was filled with water at room temperature to a height of fine grid.
329 mm from tank bottom. The mixing vessel had a
dished bottom,the curvature of which is given by R1 and EXPERIMENTAL TECHNIQUES
R2 equal to 197 mm and 38 mm, respectively. The tank Velocity Measurements
was equipped with two internal baffles each with a Laser-Doppler anemometry (LDA) system is shown
width L1 of 18.5 mm. Both baffles were mounted at a in Figure 3. The vessel was placed in a square
distance L2 of 18.5 mm (0.075D) from the tank wall. plexiglass tank filled with water in order to minimize
The clearance between each baffle and the tank bottom laser beam refraction through the curved tranparent
L3 was 61.5 mm. The curved blade impeller diameter plastic surface of the cylindrical mixing vessel. The
d and width were 138 mm and 25 mm, respectively. tank bottom was manufactured from opaque, white
The radius of the blade curvature, Rbl, was 61.5 mm, plastic which extended to a height of 60 mm from the
and the blades were tilted 10 degrees past vertical (- tank bottom surface. This opaque region could not be
10 degrees rake). penetrated by the laser beam, so measurements and
data collection in this region was not possible. A
NUMERICAL MODELLING transparent acrylic lid was fitted on the top of the
mixing tank at height H to prevent the entrainment of
The sliding grid method was used for three-
air bubbles from the free surface into the flow.
dimensional flow velocity field modeling around the
An Ar-ion laser generated a multicolour laser beam,
complex impeller . Turbulence was modeled using the
which was then split into two green and two blue beams
standard k-ε model in conjunction with logarithmic
in a transmitter. These beams were transferred through
wall functions. The commercially available
fiber optical cables to transmitting and receiving optics,
computational package CFX was used in modelling. At
which converged the beams to intersect and form a
the beginning of the calculation, a time-step size was
small elliptical control volume. Water in the stirred tank
chosen such that a rotation of the inner grid zone was
was seeded with titanium oxide particles of an
one half the angle increment between the impeller
approximately mean diameter of 10 μm, which
blades, i.e. 60°. The final steady-cyclic operating
scattered light as they travelled through the control
conditions were then achieved by starting from the
volume. The scattered laser light was then collected by
calculated velocity field as an initial condition and then
receiving optics and the resulting Doppler shift was
choosing a time-step so that the grid rotates through one
measured with photodetectors. A signal processor
azimuthal cell at the each time step.
converted electric signals from the photodetectors into
digital information for computer processing. On the Heat Transfer Experiments
computer, measurements were used to produce both The apparatus shown in Figure 4 was used to
average and fluctuating velocities on-line. experimentally obtain the temperature field in the
stirred tank filled with water in a a room temperature.
The geometry of the tank was the same as in velocity
Square
Tank
Signal PC experiments. However, in order to make wall heat
Processor
transfer resistance negligible, copper instead of plastic
was used as the wall material of the cylindrical tank.
Transmitter and
Receiver Optics Photodetectors The dished bottom, similar to the one used in the flow
Transmitter
measurements, was thermally insulated. A cylindrical
Traverse heating element encircled the copper wall and provided
Controller a constant heat flux of 5000 W/m2. The outer surface of
Ar- ion Laser
heating element was insulated to minimize ambient heat
loss.
Thermocouple
Figure 3 Laser-Doppler velocimetry system. probe
Thermocouples
X*
The computer-driven traverse controller was used to
move the laser probe in a predetermined grid. Data for 0.875 Copper
wall
axial and tangential velocity components were 0.75
measured at one vertical plane, ie. φ = 0°. A grid with 0.625
φ
six different axial heights in Figure 1(X* equal to 0.25,
0.50
0.375, 0.50, 0.625, 0.75 and 0.875) and 36 radial
0.375
measuring points at each height was set up to obtain a
0.25
total of 216 measuring locations in the tank. Due to the
Cylindrical
shaft symmetry, only one half of the tank was mapped. heating element
Since there were two pairs of laser beams and the
traverse controller could move in two directions, it was R*

possible to measure two velocity components Figure 4 Schematic diagram of heat transfer
simultaneously. The axial and tangential velocity measurement.
components were chosen for measurement, since the
radial velocity component is insignificant in the region To measure temperature distribution on the wall
adjacent to wall where heat transfer takes place. The surface, six copper-constant thermocouples of 0.3 mm
criterion for the duration of data acquisition was set at in diameter were affixed to the wall with a 41 mm
10 000 validated samples or 3 minutes time period. spacing between the heating element and the copper
However, at the bottom part of the tank, there was wall at two tangential measuring locations, namely φ =
distinct discrepancy between the measured and 0° and φ = 90° (refer to Figure 4). The temperature field
calculated tangential velocities in region near the wall. inside in the tank was measured using a probe
Due to this discrepancy, the velocity measurements composed of three thermocouples spaced 40 mm apart
were repeated at the three lowest measuring heights. from each other. Fluid temperature measurements with
The new criterion for data acquisition duration was the probe were repeated at six different axial heights
increased to 20 000 validated samples or a 5 minutes (X* equal to 0.25, 0.375, 0.50, 0.625, 0.75 and 0.875, as
time period. shown in Figure 4).
The impeller Reynolds number NRe based on
rotational speed Ns and diameter d is defined as: RESULTS AND DISCUSSION
ρN s d 2
Ν Re = Velocity Field
μ
Three measured rotation speeds with respective The number of nodes was increased as computation
impeller tip speeds and Reynolds numbers are given in progresses to check that calculated results did not
Table 1. More details can be found in reference [13]. change too much with different grid sizes. The grid
refinement testing was carried out using four non-
Table 1 Measured rotation speeds. uniform grid systems as mentioned ealier. The test
showed that the axial velocity is insensitive to the grid
Experiment Ns (r/s) NRe Wtip (m/s) size, the greatest changes in axial velocity occurring
expt1 6.52 1.23*104 2.83 near the impeller shaft. Since heat transfer from the
expt2 9.75 1.85*104 4.23 wall is considered, the most important velocity
expt3 11.38 2.16*104 4.94 components are the axial and tangential components,
while the radial velocity component clearly becomes
zero at the wall. When comparing the predictions made
with the different meshes, very similar qualitative The axial velocity profiles in Figure 7 agree well with
trends, especially in the near wall area, were obtained. each other. The comparison of tangential velocities
Figure 5 and 6 show quantitative comparisons of the shows quantitatively similar trends although some
measured data and predicted normalized axial and differences exist. The reason for differences in
tangential mean velocity profiles at some axial heights tangential velocity component profiles is due to the
between two baffles using dimensionless presentation measuring method. The LDA measurements are more
x r U W k sensitive to disturbances in this case as rotation speed is
X ∗ = , R* = , U * = ,W* = , k* =
H R Wtip Wtip 0.5Wtip2 increased. It was found that the flow is characterized by
one large top-to-bottom main circulation loop in both
It can be observed that the predicted velocities show vertical planes. Furthermore, it was noticed that the
good agreement with the LDA data. In the measured center of the vertical loop lies close to the wall at the
apparatus, there was a lid at a height H. However, bottom part of the tank. The flow patterns are
during simulations, the symmetry boundary condition dominated by the axial velocity component due to the
was used to model the free surface. The lid decreases shape of the impeller and the tank bottom, with a strong
the velocity near the wall at the upper part of the tank. axial jet flowing upwards at the near wall region and
In addition, a hole in the lid at tank centerline caused a downwards near the shaft. However, at the top of the
downward flow jet. It can be concluded that the tank, there is a zone that is nearly stagnant at the top of
comparison between experimental data and numerical the tank when examining velocities in the cross-
predictions match both quantitatively very well. Thus, sectional planes.
the computation gives a sound velocity field prediction.
0.4
0.3
U* *
*
X = 0.25
U*
X = 0.75 expt 1
computation 0.2 expt 2
experiment expt 3
0.1 small-scale unit
0.1 full-size unit
0.0
0.0
-0.1
-0.1
-0.2

-0.2 -0.3

-0.4
-0.3 0.0 0.2 0.4 0.6 R* 1.0
0.0 0.2 0.4 0.6 R* 1.0
0.40
Figure 5 Comparison of computed and measured axial
velocities at the high X* = 0.75 (φ = 0°). W* *
X = 0.25
expt 1
0.30 0.30 expt 2
*
expt 3
X = 0.75 0.25 small-scale unit
W* computation
experiment 0.20
0.20 0.15

0.15 0.10

0.05
0.10
0.0
0.05 0.0 0.2 0.4 0.6 R* 1.0

0.0 *
Figure 7 Comparison of small and full size computed
0.0 0.2 0.4 0.6 R 1.0 axial and tangential velocity profiles to the three
measured data at the high X* = 0.25 (φ = 0°).
Figure 6 Comparison of computed and measured
tangential velocities at the high X *= 0.75 (φ = 0°). At the different heights, the maximum axial velocity
components upwards and downwards have about the
The experimentally determined axial and tangential same values. The axial velocity profile shows that the
velocities at three rotation speeds and the predicted flow direction is upwards, when R* is between 0.8 and
profiles for the small-scale and commercial size-unit 1.0, but downwards otherwise. The radial component
are reported in Figure 7. These profiles at the bottom was about one-tenth in magnitude when compared to
part of the tank prove that, if the results are scaled using the axial and tangential components. Axial and
the chosen dimensionless variables, they could be tangential components approximately equal in
applied to any rotation speed and any size of the tank. magnitude and dominate the velocity field. The axial
velocity component dominates at the tank bottom and
naturally decreases with increasing tank height. At the 1.0
1000 sec 200 sec
top part of the tank, the tangential velocity becomes X*
more significant. Figure 6 shows the effect of baffles on 0.8
the tangential velocity components between the baffles. 0.7
The presence of a baffle is seen to cause a decrease in 0.6
tangential velocity in the region behind the baffles 0.5 τ
between baffle and wall. This can be clearly seen 0.4
between the baffles especially at the tank top, where the 0.3
axial velocity component is quite small compared to 0.2
tangential velocity. The profiles in Figure 6 showed 0.1
that the actual tangential velocity is highest in a region 0.0 o

halfway up the tank. 20 25 30 35 40 45 50 55 60 T (C) 70 75

Heat Transfer
Figure 8 shows the measured wall temperature
profiles at 200 seconds time intervals. The predicted
wall temperature field after 1000 seconds heating time
is also given in Figure 8. It corresponds to the thick
profile line on the graph. Figure 8 shows a clearly
defined temperature distribution in the axial direction.
At the tank bottom, the axial and tangential velocities
are highest, therefore the rate of heat transfer is highest,
resulting in lowest wall temperatures in this region. The
lowest heat transfer coefficient observed was at the top
of the tank between the baffles, where there is a
relatively stagnant area in the velocity field. Figure 8. The axial wall temperature profiles at 200 sec
Measurements and calculations also showed that there time intervals (φ = 90°) and predicted wall temperature
was a temperature difference between tangential field after 1000 sec heating.
locations, thus the heat transfer coefficients vary with
angular position with respect to baffle location and the
wall temperatures were 2-3 degrees higher between the
baffles than near the baffle location. The wall
temperatures were lower at the baffle location since
flow velocity was highest there, resulting in a higher
convective heat transfer rate.
Figure 9 shows a typical predicted temperature field
contour on a vertical plane slice located midway
between the two internal tank baffles, i.e. φ = 0°, for
the constant heat flux boundary condition at heating
time equal to 1200 seconds. Figure 9 shows how
60
uniform the temperature field is with a constant heat
flux boundary condition. The temperature profile o
T (C ) Measured
Predicted
develops quickly very near the hot wall, while Heat balance
40
elsewhere there is adequate mixing to keep the
temperature of liquid almost uniform and it varies only 30
by a few degrees in the tank interior. However, the
20
temperature variation along the wall in the axial
direction can be found in this countour also. 10
The average temperature of the fluid as a function of
0
time is illustrated in Figure 9. The fluid temperature 0 300 600 900 τ 1500
was measured with an instrument made up of three
thermocouples, which gaged the same temperatures at Figure 9 Contour of temperature between two baffles
each of three radial distances. The uniform average (φ = 0° and τ = 1200 sec) on the top. Measured,
temperature variated linearly with respect to time and predicted and heat balance average temperature as
the measured values were the same as derived from function of time, when heat flux on wall is constant.
heat balance. The comparison of predicted and
measured average liquid temperature predictions agree
well, given confidence to predicted results when using
the constant wall temperatures as well.
 o
T (C )
AKNOWLEDGEMENT
100
The authors gratefully acknowledge the support of
90
Tavg Kemira Fine Chemicals Oy and discussions with Dr. P.
80
Tw Oinas.
70
60
50
REFERENCES
40 [1] Kresta, S.M. and.Wood, P.E.,Prediction of the Three-
30 Dimensional Turbulent Flow in Stirred Tanks, AIChE
20
J.1991;37:448-460.
0 100 200 300 τ(sec) 500 [2] Ng K., Fentiman N., Lee K. and Yianneskis M.,
Assessment of Sliding Mesh CFD Predictions and LDA
Figure 10 Predicted average temperature as function of Measurements of the Flow in a Tank Stirred by a Rushton
time, when Tw = constant in commercial size reactor. Impeller. Trans IChemE. 1998; 76:737-747.
[3] Micale.G., Brucato,A., and Grsafi,F.,Prediction of Flow
After scaling up the model to the commercial size Fields in a Dual-Impeller Stirred Vessel.AIChE J.
1999,.45:445-464.
unit, the constant wall temperature boundary condition [4] Koivikko, M., Modelling the Effect of Geometric
was used to model the heat transfer in a commercial Parameters on the Performance and Efficiency of
unit, which volume was 12 m3. Figure 10 shows the Centrifugal Pump Impellers, PhD Thesis. Tampere
development of computed mean temperature for a University of Technology, Tampere 2006.
constant wall temperature boundary condition. [5] Derksen,J., and Van den Akker,H.E.A., Large Eddy
Simulations on the Flow Driven by a Ruston Turbine.
CONCLUSIONS AIChE J. 1999;45:209-221.
In the study, the numerical predictions of the flow [6] Komori, S.,and Murakami, Y., Turbulent Mixing in
field and temperature distribution of a baffled stirred Baffled Stirred Tanks with Vertical-Blate Impellers. AIChE
J.1988;34:932-937.
tank are presented. Grid dependency test showed that [7] King, R.L., Hiller, R.A. and Taterson, G.B., Power
the tangential and radial velocities are more sensitive to Consumption in a Mixer. AIChE J.1988;34:506-509.
grid size than the axial component. The results indicate [8] Delaplace,G., guern, R., and Leuliet, J.C., Dimensional
that the flow was dominated by the axial velocity Analysis for Planetary Mixer: Modified Power and
component, resulting in a major top-to-bottom Reynolds Number.AIChE J.2005;51:3094-3100.
recirculation loop inside the tank. Uniform wall heating [9] Ducci, A., and Yianneskis, M., Direct Determination of
boundary condition showed that the wall temperature Energy Dissipation in Stirred Vessels with Two-Point
and the local heat transfer coefficient strongly depended LDA. AIChE J.2005;51:2133-2149.
on location. However, the temperature distribution of [10] Kilander,J., and Rasmuson, A., Energy Dissipation
and Instabilities in a Stirred square tank Investigated Using
the liquid was almost uniform and the average an LEPIV Approach and LDA measurements.Chemical
temperature of the liquid varied linearly with respect to Engineering Science.2005;60:6844-6856.
time. The fluid temperature distribution was clearly [11] Haam, S., Brodkey, R. and Fasano, J., Local Heat
defined in the case of an isothermal wall boundary Transfer in a Mixing Vessel Using Heat Flux Sensors. Ind.
condition. The liquid was warmest at the upper part of Eng. Chem. Res., Vol. 31, No. 5, 1992, 1384-1391.
the tank and a warm core flowed downwards near the [12] Harvey III, A., Lee, C., and Rogers, S., Steady-State
shaft. The variation of the average liquid temperature as Modeling and Experimental Measurement of a Baffled
a function of time was exponential. In general, the Impeller Stirred Tank, AIChE journal, Vol. 41, No. 10,
experimental data and the model predictions agree very October 1995, 2177-218.
[13] Aho, T., Fluid Dynamics and Heat Tranfer in a
well. The present study has shown that computational Impeller-Stirred Tank, M.Sc Thesis Tampere University of
prediction of good accuracy can be obtained across the Technology. Tampere 1999.
flow field and temperature distribution in a stirred tank.
Thus, a quick assessment of the influence of rotation
speed and geometrical variables, such as other types of
impellers, can be further explored using only numerical
simulations. It is important to note that the two-
equation model of turbulence is enough for the
modelling of velocity field and heat transfer.
Since the need for more accurate boundary conditions
on the jacket side is important, the modelling of the
jacket should be scrutinized in future work. Future
work should also be directed towards obtaining a
detailed technique for rapid heat transfer measurements.
A local Nusselt number correlation for a stirred tank as
a function of the Reynolds and Prandtl number should
be the ultimate goal.