You are on page 1of 15

CFD Simulations of Three-dimensional Wall Jets

in a Stirred Tank
Sujit Bhattacharya and Suzanne M. Kresta*

Department of Chemical and Materials Engineering, University of Alberta, Edmonton AB, T6G 2G6

C
omputational fluid dynamics (CFD) is increasingly being used for
the simulation of stirred tanks due to recent advances in The flow near the tank wall in a stirred tank driven
computer speed and efficiency of numerical schemes. While by a 45° pitched-blade turbine is simulated with
simulations involving both steady state (Kresta and Wood, 1991; Fokema Multiple Reference Frames, the k-e turbulence model
et al., 1994; Harvey et al., 1995; Ranade and Dommetti, 1996 and and standard wall functions. The results are compared
others) and time varying (Perng and Murthy, 1992; Derksen and Van den to the three-dimensional wall jet identified in a
Akker, 1999) methods have been reported in the literature, emphasis has previous paper. The self-similar velocity profiles in the
jet are predicted satisfactorily, but the decay of the
mainly been on the quantitative validation of flow close to the impeller.
local maximum velocity and jet expansion are
The main objective of this study is to extend the existing protocols for underpredicted. The underlying physical reasons for
simulating time-averaged velocity fields to flow near the tank wall and in this failure are investigated. The effect of impeller size
the bulk of the tank. and position on the impingement point of the
Studies with axial impellers (Bittorf and Kresta, 2001) show that wall impeller discharge and the jet core velocity are well
jets form in the region between the baffles and the tank wall and drive predicted by the simulations. The results provide a
the bulk flow, imposing a single characteristic velocity scale on both the benchmark for CFD/MRF in the bulk of a stirred tank,
upflow at the wall and the recirculation in the center of the tank. identifying where CFD over- or underpredicts
Accuracy in prediction of the mean flow characteristics of the wall jets performance.
and the simulation of the bulk flow in the tank are therefore intimately
L’écoulement engendré par une turbine à pales
linked. Moreover, Bittorf (2000) has shown that the velocities in the
inclinées à 45° près de la paroi d’un réservoir est
three-dimensional wall jets, in balance with the settling velocities of the simulé par la méthode des référentiels multiples
solids, determine the cloud height of suspended solids at high solids (MRF), le modèle de turbulence k-? et des lois de paroi
concentration. The cloud height of the suspended solids along with the standards. Les résultats sont comparés à ceux du jet
just suspended speed (Njs ) of the impeller determines the uniformity of de paroi tridimensionnel décrit dans un article
solids distribution in a stirred tank. While explicit relations between Njs antérieur. Les profils de vitesse auto-similaires du jet
and the fluid/particle properties and the impeller type are available, the sont prédits convenablement, mais la décroissance de
cloud height model proposed by Bittorf (2000) requires an accurate la vitesse locale maximale et l’expansion du jet sont
description of the effect of size, off-bottom clearance and speed of the insuffisamment prédites. Les raisons physiques
impeller on the core velocity or source velocity of the jet. This study aims sous-jacentes de cette lacune sont étudiées. L’effet de
la taille et de la position de la turbine sur le point
to address these issues by developing a low-cost CFD protocol which can
d’impact de l’écoulement de refoulement de la
be used to obtain geometry dependent parameters to the degree of turbine et la vitesse du noyau de jet sont bien prédits
accuracy required, without having to resort to scale model experiments. par les simulations. Les résultats fournissent un banc
While a detailed description of the wall jets is available in Bittorf and d’essai pour la CFD/MRF dans le cœur d’un réservoir
Kresta (2001), the key results are restated here to facilitate comparison agité, ce qui permet de déterminer les limites de
with CFD simulations in later sections. Figure 1a shows one of the wall performance de la CFD.
jets formed between the baffle and the tank wall. The expansion of the
jet and the decay of axial velocity in a vertical plane close to the baffle Keywords: stirred tank, PBT impeller, CFD,
are shown for different axial positions in the tank. At any axial location, turbulence, three-dimensional wall jets, mixing,
the axial velocity (U ) increases rapidly from the no slip condition at the multiple reference frames.
wall to its maximum value (Um), and then decreases with a smaller
gradient. At y ~ 1.7b, the axial velocity reverses direction due to recircula-
tion and asymptotically approaches the recirculating velocity (UR ). The characteristic length scale, b, is the half-width of the
jet and corresponds to the radial distance (y) from the
wall, where U = Um/2. Um is located at ym, a distance
of about 0.15b from the wall. The region 0 < y £ ym is
*Author to whom correspondence may be addressed. E-mail address: the inner region of the wall jet, where the flow
suzanne.kresta@ualberta.ca
characteristics are similar to those in a simple

The Canadian Journal of Chemical Engineering, Volume 80, August 2002 1


For a free jet with zero free stream velocity the similarity
profile follows the similarity solution of Goertler for a round jet
(Rajaratnam, 1976). By modifying the boundary conditions,
Kresta et al. (2001) extended the Goertler solution to include
the effect of the recirculating flow outside the jet region:
U
Um
[
= 1- B tanh2 F( h - d ) ] (1)

where

y
h=
b
y = y jet + d b
b = b jet + d b (2)

In Equation (1), B is used to capture the effect of the recirculating


velocity (UR) and is given by:

ÊU ˆ
B = 1- Á R ˜ (3)
Ë Um ¯
Figure 1a. Radial velocity profiles in the wall jet at different axial
locations in a stirred tank.
F is needed to force the solution through U/Um = 0.5 when h = 1.
The h – d term in Equation (1) accounts for the displacement of
the outer region of the jet by the inner boundary layer. The
boundary layer thickness grows linearly with the jet half-width
(boundary layer thickness = db and d ª 0.15 for wall jets). The
values of B, F and d for a three-dimensional wall jet are defined
using the velocity data of Bittorf and Kresta (2001). The resulting
similarity profile is:

U
Um
[
= 1- 1.58 tanh2 0.78( h - 0.15) ] (4)

This equation, in conjunction with expressions for the


maximum velocity decay and jet expansion, provides a complete
characterization of the vertical flow. Bittorf and Kresta (2001)
relate Um and b in two regions of the jet. In the characteristic
Figure 1b. Similarity profile of mean axial velocity in the three-
dimensional wall jet.
decay region a classical wall jet is still responding to the geometry
of the nozzle. In the stirred tank the wall jet is responding to the
combined effects of tangential flow and circulation.

-0.49
Um Ê zˆ
boundary layer. The region y > ym is the outer region, where the = 1.35Á ˜ (characteristic decay) (5a)
Ucore ËT ¯
flow characteristics are similar to a free jet. Interactions between
these two regions project the anisotropic influence of the wall
to the outer (free shear) region of the jet and make the flow field In radial decay, the jet is fully formed and has no recollection
somewhat more difficult to model. Thus, the accurate prediction of the initial effects of geometry. In cross-section it forms a quarter
of jet decay is intimately linked to the successful modeling of circle with expansion in the radial direction. The maximum
turbulence. In Figure 1a (D = T/3 and C/D ~ 1.0), the global velocity Um is inversely proportional to the distance traveled:
maximum velocity along the tank wall or the core velocity
(Ucore ) occurs at z/T ~ 0.2. As the jet moves upwards, it expands -1.15
Um Ê zˆ
in the y direction with a simultaneous decay in Um. The jet = 0.57Á ˜ (radial decay) (5b)
Ucore ËT ¯
finally collapses near the top of the tank (z/T ~ 0.7 for D = T/3
and C/D ~ 1.0). Between these two limits of formation and
disintegration, the three-dimensional wall jet is self-similar. This The jet half-width expands as:
means that the radial profiles of axial velocity collapse onto a
single similarity profile when U is scaled with Um and y is scaled b È z - zo ˘
with b (h = y/b), as shown in Figure 1b. = 0.38 Í (6)
T ˙
Î T ˚

2 The Canadian Journal of Chemical Engineering, Volume 80, August 2002


zo in Equation (6) denotes the virtual origin of the jet. Bittorf development of the jet. This extension of CFD results represents
and Kresta (2001) modeled the three-dimensional jets driven by a new generation of investigations, moving beyond validation
axial impellers as originating from a point source located below and code development to a point where researchers can obtain
the bottom of the tank. The position of this virtual point source new information and insights from CFD results.
is described as the virtual origin. Equations (4), (5) and (6) are
valid for PBT, HE3 and A310 impellers (Bittorf and Kresta, 2001). Experimental Data
This point is important because it means that the vertical flow Experimental data from Bittorf and Kresta (2001) has been
field for all of these commonly used impellers can be character- extensively used in this study to illuminate and validate results of the
ized using these three equations. The reader will note that while CFD simulation. Since a detailed explanation of the experimental
Equations (4), (5) and (6) are universal for these axial impellers, setup is available elsewhere (Bittorf and Kresta, 2001; Bittorf,
the quantities zo and the core velocity (Ucore ) are dependent on 2000), only a brief outline is presented here. Of the various cases
the impeller and tank geometry and need to be determined studied in the original experiments, only the geometries and
separately. This provides an ideal opportunity for the use of CFD: the operating conditions used in the simulations are described.
geometry dependent parameters are predicted using simulations Axial velocities were measured using a Laser Doppler
and are then used in Equations (4) through (6) to obtain the Velocimeter (LDV). The important geometric variables for the
complete flow field. This work will show that CFD simulations stirred tank are shown in Figure 1a, b. The 0.24 m diameter tank
predict core velocities which are in reasonable agreement with had a flat bottom with an aspect ratio (H/T ) of 1 and was sealed
experimental data. However, simulations are not able to predict at the top surface. The impeller was a standard, 4-bladed, down-
zo because of its dependence on the decay of Um and expansion pumping, 45° pitched-blade turbine (PBT). Experiments used
of b, two aspects of jet simulation where CFD commonly fails. impeller diameters of T/3 and T/2. For the T/3 impeller,
The location of the wall jet in stirred tanks is determined by experimental data is available for two clearances (C/D = 0.4 and 1.0)
the point of impingement of the impeller discharge stream. The and two impeller speeds (N = 500 and 1000 rpm).
impingement point can be on the bottom of the tank or on
the tank wall, depending on the impeller type, diameter and CFD Simulations
off-bottom clearance. The impingement point has an effect on CFD simulations were carried out in two geometrically similar
the core velocity in the wall jet and the position of the jet in the stirred tanks with diameters of 0.24 and 1.00 m. The effect of
tank (Bittorf and Kresta, 2001). While experimental data impeller clearance on the velocity field was studied by varying the
typically lack the required resolution, detailed velocity profiles clearance from 0.4 < C/D < 1.9. For the smaller tank, simulations
obtained from CFD simulations are suitable for identifying the were carried out at impeller speeds of 500 and 1000 rpm
location of the impingement point and can be used to study (Re = 5.3 ¥ 104 and Re = 1.1 ¥ 105, respectively). The effect of
the effect of Reynolds number, off-bottom clearance and tank scale was tested using an impeller speed of 60 rpm in a T = 1.00 m
size on the formation and decay of the jet. In this paper, CFD tank. Water was used as the working fluid in all cases.
results are used to arrive at a qualitative understanding of the The fluid properties, tank geometry, impeller type and
relationship between the geometry of the vessel, the core dimensions (shaft diameter, hub size, blade width, and blade
velocity, and the impingement point. thickness), and the rotational speed, were set using MixSim‘
The simulations in this work have been carried out using the version 1.5, a pre-processing software specifically designed for
steady state, Multiple Reference Frames approach which CFD simulation of stirred tanks. PreBFC‘, a mesh generator,
requires that there be negligible interaction between the baffles uses the parameters defined in MixSim‘ to create a structured
and the impeller. If baffle-impeller interactions are large or if grid, and subsequently CFD calculations are launched in
large-scale transient structures (macroinstabilities) are present, FLUENT‘ version 4.5.2. Iterations were continued until the sum
unsteady state modeling using the sliding mesh approach of the normalized residuals of all variables (pressure, axial, radial
and/or large eddy simulation (LES) is advisable. However, the and tangential components of velocity) was less than 5 ¥ 10–4.
computational resources and post-processing required for steady Both the geometry of the impeller and velocity data at the
state simulations are only a fraction of those required for transient impeller are available for the PBT in the MixSim‘ library;
simulations, and jets are, by definition, time-averaged reduced however, Fokema et al. (1994) observed that the boundary
models of the flow. The MRF method has therefore been adopted conditions at the impeller are strongly affected by the off-bottom
in this study with a view to identifying: (i) the useful information clearance, and recommended extreme caution when selecting
that can be obtained from this modeling approach, and (ii) the data for this approach. In the present study, the geometry of the
conditions under which this approximation fails. pitched-blade turbine was defined in the grid and the velocity
The main objectives of this study can now be summarized as profile near the impeller was calculated directly in the simulation.
follows: This approach has been validated by Harvey and Rogers (1996).
• the development of protocols for MRF simulation of flow near Simulations were carried out using the Multiple Reference
the tank wall and in the bulk of the tank, and a determination Frames (MRF) scheme. In the MRF formulation, the grid is divided
of the limitations of this approach; into two regions: a rotating volume is associated with the impeller,
• the prediction of the key geometry dependant characteristics while a stationary volume is associated with the baffles and the
of the jet: virtual origin, core velocity and impingement tank wall, as shown in Figure 2. Although the grid in the impeller
point, for use in applications such as estimating the solids region is stationary, the rotation of the impeller is modeled directly
cloud height in solid suspensions; by solving the transport equations in a reference frame which
• the prediction of the decay of Um in the radial decay region rotates at the angular velocity of the impeller. The transport
of the jet. equations in the outer region are solved in the stationary reference
Once these objectives are met, the simulation results will be frame of the tank. Since the grid in the impeller region is kept
used to gain further insight on the flow field and on the stationary, it is fixed relative to the location of the baffles. The most

The Canadian Journal of Chemical Engineering, Volume 80, August 2002 3


important assumption made in the MRF model is that there is very
weak, if any, interaction between the impeller and the baffles. At
the boundary between the two reference frames, the terms in the
transport equations in one frame require values from the adjacent
reference frame. FLUENT‘ enforces continuity of the absolute
velocity across the boundary, to provide the correct ‘neighbor’
values for the region under consideration. Note that velocities are
not averaged over all tangential positions of the reference frame
boundary to obtain a full time-averaged solution (the ‘mixing-plane’
approach) but are passed point by point across it to give a
‘snapshot’ of the flow-field. The velocity in the rotating reference
frame, which has been calculated relative to the motion in this
region, is converted to the absolute inertial reference frame using
the following definitions for the position vector (r), etc.
r = x - xo (7)

the absolute velocity vector


v = vr + w ¥ r (8)

and the velocity gradient

— v = — v r + — (w ¥ r ) (9)

Since the velocity is now defined in terms of the stationary


reference frame, it can be readily used in the transport equations on
the stationary side of the boundary. For steady state calculations,
the tank is considered symmetrical, and hence, for the four-baffle
Figure 2. Isometric view of tank configuration and grid geometry. case, only one quarter of the tank is simulated.
Cylindrical co-ordinate equivalents to the FLUENT (i, j and R) co-
ordinates are q = i, r = j and z = k. Grid Resolution
Insufficient grid resolution can give rise to errors in simulation
even when there are no deficiencies in the physical models
applied to the system. Thus, an appropriate grid must be
defined before any further simulation is undertaken. After
reviewing numerous CFD studies on stirred tanks, Kresta (1996)

Figure 4. Effect of grid on maximum jet velocity: D = T/3, C/D = 0.4,


Figure 3. Grid refinement in a single i-plane. N = 1000 rpm, x = 3.6 mm, y ~ 5 mm.

4 The Canadian Journal of Chemical Engineering, Volume 80, August 2002


suggested that a minimum grid should consist of at least about It may be noted that the k-e model has several variants. Apart
20 cells in the tangential (i), 20-25 cells in the radial (j ) and 40 from its standard form (Jones and Launder, 1972), FLUENT‘
cells in the axial (k) direction. In this survey, cell sizes ranged 4.5.2 also supports the Renormalization Group (RNG) k-e
from 4.5 to 5 mm while the tank diameters varied from 0.15 to model, which is derived by the application of a rigorous statistical
0.44 m. Fokema et al. (1994) used a grid consisting of 43 axial, technique (RNG), to the instantaneous Navier-Stokes equation.
20 radial and 20 tangential cells for the simulation of flowfield The standard k-e model is based on the Boussinesq hypothesis
with a PBT. While simulating the flow for dual Rushton turbines, that the Reynolds stress is proportional to the mean velocity
Jaworski et al. (2000) recommended that a reasonable degree gradient, with the constant of proportionality being the
of grid independence could be obtained with a mesh density turbulent or eddy viscosity (mt ). The eddy viscosity is then
factor of at least 32 cells per tank diameter. defined in terms of two variables: the turbulent kinetic energy
In the present study, five grids were investigated to (k) and the rate of dissipation of turbulent kinetic energy per
determine the effect of grid resolution on the computed results. unit mass (e). The full k-e model is given in Table 1. Detailed
A structured grid was generated using the ‘fine grid’ (at least 60 discussions of the terms in the k-e model are available in the
cells per tank diameter) option in MixSim‘. A view of the grid literature (Launder and Spalding, 1972, 1974; Mohammadi and
in the r-z plane is shown isometrically in Figure 2 and is Pironneau, 1994). The k-e model has proven useful for many
reproduced as a plane in Figure 3. This mesh density produced turbulent flows (jet flow, duct flow etc); however, like other
a grid with 50 cells in the tangential, 32 cells in the radial and models based on Boussinesq viscosity approximations, the k-e
63 cells in the axial direction (50 ¥ 32 ¥ 63). It is the default grid model has a number of shortcomings. The effect of these
(Figure 3a). To improve grid resolution near the tank wall the last deficiencies on the simulation of complex flow fields has been
four cells in the radial direction were halved to increase the discussed in numerous articles (Launder and Rodi, 1983;
number of near-wall cells while maintaining a cell-aspect ratio Launder, 1995; Launder, 1999; Bradshaw, 1999). A short
of two (Figure 3b). This grid had 50 ¥ 36 ¥ 63 cells. A third grid summary of the problems associated with the eddy-viscosity
was obtained by further refining the near-wall cells in the assumption in simulating certain classes of jets is given below:
second grid by halving the last four cells in the radial direction 1. The k-e model does not predict the significantly larger lateral
to obtain 50 ¥ 40 ¥ 63 cells (Figure 3c). The fourth grid (Figure expansion in three-dimensional wall jets. This is primarily due
3d) was created by adding cells near the tank bottom, giving to its inability to account for the generation of streamwise
50 ¥ 40 ¥ 71 cells. Finally (Figure 3e) the last four cells near the vorticity by turbulent stresses. This occurs in addition to the
tank bottom in Figure 4e were halved to give a higher axial contributions by the mean velocity gradients.
resolution (50 ¥ 40 ¥ 75 cells). 2. The model is unable to simulate the correlation between
The effect of the grid resolution on an axial traverse of the pressure and stress transport. Thus, it cannot account for the
maximum streamwise velocity is shown in Figure 4. The axial damping of normal velocity fluctuations even in the ‘outer’
velocity is grid independent beyond 36 cells in the radial and free-shear region due to pressure reflections from the wall.
71 cells in the axial direction. Surprisingly, for z/T less than 0.2, the The overestimation of the rate of spread of plane wall jets
modified grids all show similar velocities. There is no effect of grid normal to the wall (Launder and Rodi, 1983) has been attrib-
resolution in this region after the initial refinement in the radial uted to this deficiency.
direction (changing from 32 to 36 cells). Similar results were 3. The k-e model is unable to account for the generation of
observed for traverses in the radial direction. A grid with 50 cells in additional turbulence due to the effect of streamwise
the tangential, 40 cells in the radial and 71 cells in the axial direction, curvature and is therefore very unreliable in simulating flows
using the near wall refinement shown in Figure 3d, was selected for with sharp curvature
all remaining simulations in this study. 4. The spread of an axisymmetric round jet is overestimated by
the standard k-e turbulence model. Various modelers have
Turbulence Modeling tried to adjust the transport equation for e to account for this
Although the most rigorous turbulence model is the instantaneous shortcoming, with varying degrees of success (Wood and
Navier-Stokes (NS) equation, direct numerical simulation (DNS) Chen, 1985).
of this equation requires computational resources of the order The RNG k-e model aims to develop a more general and
of the cube of the Reynolds number (Bradshaw, 1999). An fundamental model of turbulence. In this model formulation,
alternative method, Reynolds-averaged turbulence modeling, is renormalization group theory is first used to resolve the smallest
used extensively in engineering applications involving high eddies in the inertial range. Once these small-scale eddies have
Reynolds numbers. However, the averaged models suffer from been resolved, a small band of the smallest eddies is eliminated
a lack of universality, and successful numerical modeling of a by representing them in terms of the next smallest eddies. This
turbulent flow field strongly depends on the choice of an process of successively removing a band of the smallest modes
appropriate turbulence model. In the current study a preliminary is continued until a modified set of Navier-Stokes equations is
investigation was conducted to select a suitable turbulence obtained, which can then be solved using a relatively coarse grid.
model based on its ability to: The equations constituting the RNG k-e model are shown in Table 1,
1. accurately reproduce the self-similar axial velocity profile; and further details of the model are available elsewhere.
2. accurately reproduce the development and decay of the The full differential Reynolds Stress Model (RSM), also called
maximum local axial velocity (Um ); and the stress transport or second-moment model, involves solving
3. attain stable convergence. the transport equations for the six individual Reynolds stresses
The version of FLUENT‘ used in this study (4.5.2), supports two in the Reynolds-averaged Navier-Stokes equation (RANS). These
widely used and very successful turbulence models: transport equations can be obtained directly from the
1. two equation, k-e model; and momentum equation, and contain terms for the pressure-strain
2. Reynolds stress model (RSM) distribution, turbulent diffusion, and generation and dissipation

The Canadian Journal of Chemical Engineering, Volume 80, August 2002 5


Table 1. Simplified Equations for k-e and RSM turbulence models.

∂Ui ∂P ∂2Ui ∂Rij


Reynolds Averaged Navier-Stokes Equation (RANS) rUk =- + + (for steady, incompressible flow without body-force*)
∂xk ∂xi ∂x j ∂x j ∂x j
where the Reynolds stresses are Rij = -rui u j and viscous stresses are negligible.
(Mean values are represented by capital letters, and fluctuating quantities by lowercase letters).

Turbulence Models*

i) k-e model Boussinesq hypothesis È ∂U


2 ∂U j ˘
(isotropic stresses) Rij = -rui u j = -r kdij + mt Í i + ˙
3 ÍÎ ∂x j ∂xi ˙˚

k2
Turbulent viscosity mt = rC m
e
Cm = 0.09
ui ui ∂u Ê ∂u ∂u j ˆ
where k = and e = n i Á
i +
˜
2 ∂x j Ë j ∂xi ¯
∂x

A) Standard Form Transport equation for Ê ∂U j ∂U ˆ ∂U j


∂k ∂ Ê mt ∂k ˆ
turbulent kinetic energy (k) rUi = mt Á + i˜ + Á s ∂x ˜ - re
∂xi Ë ∂xi ∂x j ¯ ∂xi ∂xi Ë k i¯
sk = 1.0
Convection Production Diffusion Dissipation

Transport equation for rate


∂e Ê e ˆ Ê ∂U j ∂Ui ˆ ∂U j ∂ Ê mt ∂e ˆ Ê e2 ˆ
of dissipations of k(e) rUi = C1e Á ˜ mt Á + ˜ + Á ˜ - C 2erÁ ˜
∂xi Ë k ¯ Ë ∂xi ∂x j ¯ ∂xi ∂xi Ë s e ∂xi ¯ Ë k ¯
C1e = 1.44
C2e = 1.92 Convection Production Diffusion Dissipation
se = 1.3

∂k ∂ Ê ∂k ˆ
rUi = mt S 2 + ak meff - re
B) RNG k-e Transport equation for ∂xi ∂xi ÁË ∂xi ˜¯
turbulent kinetic energy (k)
Convection Production Diffusion Dissipation
where,
1 Ê ∂U j ∂Ui ˆ
S ∫ 2Sij Sij , Sij ∫ Á + ˜ and meff = m + mt
2 Ë ∂xi ∂x j ¯

∂e Ê eˆ ∂ Ê ∂e ˆ Ê e2 ˆ
Transport equation for rate of rUi = C1e Á ˜ mt S 2 + a e meff - C 2erÁ ˜ - R
∂xi Ëk¯ ∂xi ÁË ∂xi ˜¯ Ë k ¯
dissipations of k(e)
Convection Production Diffusion Dissipation Other terms related
to mean strain and
turbulence quantities
ii) RSM

∂ui u j ∂Jijk
Reynolds stress transport equations: rUk = Pij + Fij - eij + , where the terms can be expanded as follows:
∂xk ∂xk

Ê ∂U j ∂Ui ˆ
∑ Production: Pij ∫ -rÁ ui uk + u j uk (calculated)
Ë ∂xk ∂xk ˜¯

Ê ∂u ∂u j ˆ
∑ Pressure - strain distribution: Fij ∫ p Á i + ˜ (modeled)
Ë ∂x j ∂xi ¯

∂ui ∂u j
∑ Dissipation: eij ∫ 2m (calculated from e )
∂xk ∂xk

∑ Turbulent diffusion: Jijk = -Ê rui u j uk + p d jkui + diku j ˆ


( ) (modeled)
Ë ¯

6 The Canadian Journal of Chemical Engineering, Volume 80, August 2002


of Reynolds stresses. The formulation of the RSM used in this substantial to justify the increase in computational requirements
work is also given in Table 1. More detailed discussion of the due to the larger set of highly coupled differential equations.
individual terms and assumptions in this formulation is given by Integrating the turbulence models (either k-e model or RSM)
Launder et al. (1975) and Launder (1989). The RSM avoids the through the near-wall region and applying the no-slip condition
isotropic eddy-viscosity assumption of the k-e models, and at the wall requires some additional consideration, since the
therefore has very good potential for accurately predicting turbulent quantities in the near-wall region are significantly
complex flows with streamwise curvature, swirl, rotation and affected by the proximity of the wall. Turbulence is heavily
high strain rates. However, these models require wall-reflection damped very close to the wall; however, in the outer part of the
terms which may need to be redesigned for a particular near-wall region, it is rapidly augmented by the production of
geometry. The improvement in prediction accuracy must be kinetic energy, due to Reynolds stresses and the large gradient
in the mean velocity. To overcome these difficulties, two
approaches can be used:
• Application of wall-functions to ‘bridge’ the viscosity-
affected region between the wall and the fully turbulent
region of the flow.
• Resolving the near-wall viscosity affected region (including
the viscous sub-layer) with extra-fine mesh all the way to the
wall. This near-wall modeling approach is implemented in
FLUENT‘ using a two-layer zonal model. This method does
not rely on the semi-empirical log-law, but requires much
higher CPU time.
Since the grid resolution appropriate for the latter approach
requires substantial computational resources, wall-functions
were used in all simulations in this work.
Various combinations of the three turbulence models
described above and the two wall-functions were tested. The most
important mean characteristic of the wall jet is the self-similar axial
velocity profile. In Figure 5a, the computational results for a
single traverse location (z = 13.6 mm above the bottom of the tank,
x = 3.6 mm from the baffle) are compared with the experimental
data at the same location and the universal similarity profile for three-
dimensional wall jets in a stirred tank. The standard wall-function and
the pressure-sensitized non-equilibrium wall-function gave very
similar results and in fact overlap. Since adverse pressure
Figure 5a. Comparison of turbulence models at z/T ~ 0.56: T = 0.24 m, gradients are not important in this flow field, this result is not
D = T/3, C/D = 0.4, N = 1000 rpm, x = 3.6 mm, z ~ 13.6 mm. surprising. Both the k-e model and the RSM with the standard
wall-function are able to reproduce the similarity profile of the
wall jet. The RSM gives exact agreement with experimental
results at this location. The time required for convergence with
the RSM was about nine times that for the k-e model. The RNG
k-e model with the standard wall-function was not able to
reproduce the similarity profile, overestimating the size of the
inner region by a factor of four, underpredicting the point of
flow reversal, y0 , and overpredicting UR.
A more demanding test of turbulence models is the development
and decay of the maximum velocity in the jet, Um , shown in
Figure 5b. All velocities were measured at a distance of 5 mm
from the tank wall and 3.6 mm from the baffle. The k-e model
with the standard wall-function and the RSM with the standard
wall-function come closest to reproducing the magnitude of
Um. The RNG k-e model does not give the correct magnitude of Um,
but the growth, location of maximum velocity, and the decay
with distance match the experimental decay almost perfectly.
The k-e model is able to capture the similarity profile and match
the magnitude of Um , while being computationally the least
demanding. This model is used for all remaining simulations.

Results and Discussion


Figure 5b. Effect of turbulence models on dimensionless axial velocity In the first part of this section, the similarity profile and the
profile: T = 0.24, C/D = 0.4, N = 1000 rpm, x = 3.6 mm, y = 5 mm. Standard overall flow field obtained from CFD simulations are compared
wall functions were used for all cases. The results for the non-standard and with experimental results from previous studies. This part simply
standard wall functions overlap for the test case with the k-e model. confirms that CFD is able to generate circulation patterns

The Canadian Journal of Chemical Engineering, Volume 80, August 2002 7


observed in experiments and that the similarity profiles discharge at the tank wall and the vertical distance in the tank over
obtained from CFD conform to Equation (4). As discussed which the jet exists are estimated from CFD simulations. These
earlier, the three-dimensional wall jet equations (4, 5 and 6) can quantities cannot be readily obtained from experiments because
be used to completely characterize the vertical flow field for all of the requirement of high data resolution. The results deepen and
commonly used axial impellers if the geometry dependent confirm our physical understanding of the flow field, providing
parameters, Ucore and zo , are known. The second part of this insights which are not available from experiments alone.
section considers the ability of CFD to adequately estimate
these parameters. The reader may note that zo is obtained by Circulation Pattern and Similarity Profile
extrapolating the loci of b backwards until the resulting line Figures 6a to 6d show the predicted circulation patterns in an
intersects the plane of the wall. Accurate prediction of zo is r-z plane for two impeller sizes (D = T/3 and T/2), and two off-bottom
therefore dependent on the rate of decay of Um and the clearances. The discharge from the impeller flows downwards
expansion of b. Finally, the impingement point of the impeller and impinges either on the bottom or on the wall of the tank.
If the impeller discharge stream impinges on the bottom of the
tank (Figures 6a and 6c) it forms a single, upwards circulation
loop at the wall. If, however, the discharge impinges on the wall
(Figures 6b and 6d) it splits in two. Most of the flow moves
upwards into the wall jet but below the point of impingement,
a portion of the stream flows downwards into the weaker
secondary circulation. Comparison of Figures 6b and 6d shows that
the secondary circulation is much stronger for the larger (D = T/2)
impeller. There is good qualitative agreement between these
results and flow patterns observed in the flow visualization and LDV
experiments of Kresta and Wood (1993), and Bakker et al. (1996).
The MRF formulation correctly predicts the off-bottom
clearance where the flow shifts from a single upwards circulation
to wall impingement. This point is at C/D = 0.67 for the D = T/2
impeller (Kresta and Wood, 1993).
As shown in Figure 7, there is also good agreement between
the similarity profiles obtained from CFD simulations and
Equation (4). Beyond h = 1.7, the velocity profile is affected by
recirculation and similarity is lost at larger h. For h < 1.7 the
similarity profile is immune to changes in C/D, N (not shown in
the figure) and scale of the vessel. These results support the
experimental observations of Bittorf and Kresta (2001), which
show that the similarity profile is unaffected by tank geometry.

Figure 6. Velocity vector fields showing the interacting effects of impeller Figure 7. Effect of tank size and off-bottom clearance on similarity
diamter and off-bottom clearance: D = T/3, T = 1.00 m, N = 60 rpm, i = 47. profiles: D = T/3, Re = 105, x = 3.6 mm, z ~ 136 mm for T = 0.24 m
Note the impingement on the tank wall at higher off-bottom clearances. and 558 mm for T = 1.00 m.

8 The Canadian Journal of Chemical Engineering, Volume 80, August 2002


Geometry-Dependent Parameters – Ucore and zo Table 2. Comparison of Ucore from CFD simulations with data from
Before starting a discussion of the geometry-dependent Bittorf and Kresta (2001). (D = T/3, T = 0.24 m)
parameters, the important characteristics of the flow near the
tank wall are identified using the development and decay of Um Turbulence C/D Ucore/Vtip Ucore/Vtip %
along the tank wall at different impeller clearances for T = 0.24 m model (Experimental) (CFD) Difference
and D = T/3, as shown in Figure 8. Starting from the bottom k-e 0.4 0.32 0.3003 6.2
of the tank, the reader should note the regions of reverse flow, k-e 0.8 0.28 0.2576 8.0
the point of impingement, and Ucore labeled in the figure. Note k-e 1.0 0.27 0.2315 14.3
that the small region of reverse flow near the tank bottom at k-e 1.5 0.17 0.1868 -0.1
low impeller clearances (C/D = 1.2 and 1.4) is not due to wall RNG 0.4 0.32 0.27 15.6
impingement but to a small eddy which forms in the corner RSM 0.4 0.32 0.248 22.5
between the baffle, tank bottom and the wall (observe the
bottom left corner in Figures 6a and 6c). A region of mild
reverse (downward) flow is also seen in the top 20% of the
tank. This is present at all clearances. The core velocity, Ucore , is data (Bittorf and Kresta, 2001) more closely, with smaller mean
identified as the global maximum of Um , and is one of two deviations, as given in Table 2.
parameters needed to characterize the vertical flow in a new For both impeller sizes, the core velocity initially drops with
geometry. The relationship between Ucore and the tank an increase in C/D and then starts increasing. Thus, for the T/3
geometry is discussed in the next section. impeller in Figure 9a, Ucore/Vtip reduces with increasing
clearance for C/D less than 1.4. At higher clearances there is a
Core Velocity slight increase in the core velocity due to compression of flow
Reynolds scaling dictates that the velocity field in a stirred tank towards the tank wall by the secondary circulation loop. For the
scales with Vtip under fully turbulent conditions as long as T/2 impeller in Figure 9b, Ucore /Vtip decreases rapidly when
geometric similarity is maintained. Hence, Ucore is scaled with the clearance is increased from 0.33 to 0.6. Thereafter, the core
Vtip in Figures 9a and 9b to compare values obtained for different velocity increases with an increase in C/D due to the influence
tank sizes (T = 0.24 and T = 1.00 m for D = T/3 impeller). A of the strong secondary flow.
number of important observations can be made from the CFD simulations using the k-e turbulence model provide
figures. First, Ucore/Vtip is unaffected by tank size, confirming estimates of Ucore which are in sufficient agreement with
that Ucore scales with the tip speed according to Reynolds experimental data to be used as estimates for Equation (5). This
similarity. Secondly, the Ucore obtained from simulations is is especially true for the smaller (T/3) impeller. Since only two
within 14% of the experimental data for the T/3 impeller in experimental data points are available for the T/2 impeller, a
Figure 9a. Figure 9b shows that the deviations are higher for conclusive judgement about the accuracy of the results for this
the T/2, impeller but the experimental trends are followed by impeller size cannot be made, but the predicted values match
the simulations. Thirdly, the core velocities predicted using the the experimental trends.
RSM and RNG k-e models underestimate the experimental results
by 16% and 23%, respectively, in Figure 9a. Predictions based on Decay of Maximum Velocity, Expansion of Jet Half-width
the k-e model follow the trend defined by the experimental and Virtual Origin
As explained earlier, the virtual origin, zo , is obtained by
extrapolation of the loci of the jet half-width, b, back to the
point where they intersect the tank wall. The expansion of b is
related to the decay of the maximum velocity by the
momentum integral constraint, which requires the momentum
in the jet to be conserved (Rajaratnam, 1976). Hence, the decay
of Um is closely tied to both the expansion of b and the location
of zo. Using C/D = 0.4 and 1.0, Bittorf and Kresta (2001) found
that Um/Ucore varies with z/T–0.59 in the characteristic decay
region and with z/T–1.15 in the radial decay region. Their results
are compared with the Um/Ucore values obtained from CFD
simulations in Figure 10. Three regions can be identified in the
figure, corresponding to the potential core, characteristic decay
and radial decay regions in a three-dimensional wall jet. The
simulated values of Um are closer to the experimental values in
the characteristic decay region, but the slope is much smaller
than the experimental results in both regions. Consequently,
the jet half-width obtained from simulations expands more
slowly than the experimental results.
The flow field along the wall of the tank provides further insight
into the development and growth of the wall jet along the baffles.
Figure 11a shows the contours of Um/2 at different axial locations
in the 0.24 m tank, with C/D = 0.4 and D = T/3. The filled regions
represent the locations on horizontal slices (r-q plane), where the
Figure 8. Variation of maximum axial velocity (Um) with axial position: axial velocity is greater than Um/2. The impeller discharge
T = 0.24 m, D = T/3, N = 1000 rpm, x = 3.6 mm, y = 0.56 mm.

The Canadian Journal of Chemical Engineering, Volume 80, August 2002 9


Figure 9a. Effect of off-bottom clearance and tank size on dimensionless Figure 9b. Effect of off-bottom clearance and tank size on dimensionless
core velocity: D = T/3. core velocity: D = T/2.

impinges on the bottom of the tank and then swirls outwards the baffle and provide further support for the flow pattern
towards the tank wall. On reaching the wall, the fluid is described earlier. The tangential movement along the curved
deflected upwards and the radial swirling flow is transformed walls of the tank compresses the fluid towards the baffles and
into an upward flowing column of fluid with a strong tangential affects the growth of the three-dimensional wall jet in front of
movement towards the baffle. Evidence of the tangential motion it. Thus, the prediction of the wall jet characteristics along the
is seen in Figure 11a, in the shrinkage of the contours at baffle depends on the ability of the numerical scheme and
locations far from the baffle and the simultaneous radial growth the turbulence model to accurately simulate flow along the
at locations close to the baffle. Note also the velocity field (inset curved tank surface.
in Figure 11a). Velocity vectors show the movement towards If the impeller-baffle interactions are significant, then the
snapshot approach used in the MRF formulation will itself lead to
inaccurate results. In the presence of impeller-baffle interactions,
the impeller will influence the tangential component of the
velocity near the wall throughout the 90∞ sweep, while the MRF
formulation will not be able to capture this effect. Figure 11b
shows the axial velocity profiles in the tangential direction at
different axial positions, as obtained from CFD simulations and
from the experimental data of Bittorf and Kresta (2001). At low
axial positions the flow is reproduced fairly well by CFD.
However, higher up along the tank wall, CFD simulations show
only a slow decrease in velocity as the angular distance from the
baffle is increased, while a much sharper reduction in velocity is
apparent in the experimental data. The larger axial velocities in
the CFD results imply that the tangential flow towards the baffle
is smaller in the simulations than in the experiments because of
continuity and the two-dimensional nature of the near-wall flow
field. The smaller tangential flow is due to the inability of the
MRF formulation to account for the impeller-baffle interaction
and results in less compression of the fluid onto the baffle.
Figure 12 is a log-log plot of the dimensionless maximum
axial velocity (Um/Ucore) versus dimensionless axial distance
(z/T ) for various impeller clearances. The slopes of these curves
give the decay of the maximum velocity (Um ) with distance.
Figure 10. Regions in the wall-jet: T = 0.24 m, D = T/3, C/D = 0.4,
Two regions can be defined for all clearances. These correspond
N = 1000 rpm. The simulated slopes for the decay of Um/Ucore and
to the characteristic and radial decay regions first shown in
expansion of b are smaller than the experimental results of Bittorf and
Kresta (2001). Figure 10. For C/D < 1.4, decay of Um in both regions is much

10 The Canadian Journal of Chemical Engineering, Volume 80, August 2002


smaller than the experimental results as detailed in Table 3. The
slower decay in Um also leads to a slower expansion of b and
larger zo , as shown in Tables 4. An increase in tank size from
0.24 m to 1.00 m results in a very small improvement in
performance, which is similar to the improvement obtained by
using the RSM turbulence model. The RNG k-e model is able to
capture the slope of the velocity decay very accurately but the
values are underestimated by more than 20%. Similar performance
is observed for high off-bottom clearance (C/D ~ 1.6), but in
this case the maximum velocity is overestimated.
Since CFD is unable to accurately predict the decay of Um ,
reliable estimates of zo cannot be obtained. The failure of CFD
calculations to properly predict the decay of Um can be linked
to a failure to properly predict the tangential flow along the
curved tank surface. The marginally higher velocity decay for
the larger tank indicates a small effect of tank curvature, so part
of the failure in predicting the flow along the tank wall lies in
the inability of the standard k-e model to properly simulate flow
along curved streamlines. Another important factor contributing
to the failure in predicting the wall jet velocity decay is the
assumption of no impeller-baffle interaction in the MRF
formulation. Averaging of the outwards flow over the tangential
plane may provide a more accurate representation of the time-
averaged flow field. Simulations using transient methods, like
the large eddy simulation, should also be explored.

Impingement Point (zI ) and Axial Distance over


which Jet Extends
When the impeller discharge stream impinges on the tank wall,
the axial position at which this impingement occurs is known as
Figure 11a. Contours of axial velocity (Um/2) at different axial the impingement point, zI. The three-dimensional wall jet starts
locations showing the transition from characteristic decay to radial to form between the tank wall and the baffle a short distance
decay: D = T/3, C/D = 0.4, T = 0.24 m, N = 1000 rpm. Insert shows from the impingement point; hence zI influences not only the
velocity vectors at the wall of the tank where the flow is convected position of the virtual origin, zo , but also the location and
towards the baffle. magnitude of Ucore. Moreover, zI determines the axial position

Figure 12. Decay of maximum velocity: T = 0.24 m, D = T/3, N = 1000 rpm,


Figure 11b. Tangential profile of axial velocity at the tank wall: T = 0.24 m, x = 3.6 mm (T = 0.24 m) and 11.5 mm (T = 1.00 m), y = 0.56 mm. Unless
D = T/3, N = 1000 rpm, C/D = 0.4, y = 5 mm. stated otherwise all data refer to T = 0.24 m.

The Canadian Journal of Chemical Engineering, Volume 80, August 2002 11


Table 3. Decay of maximum velocity in the characteristic and radial Table 4. Rate of expansion of jet and virtual origin: T = 0.24 m and
decay regions. D = T/3.

T C/D Slope in Slope in C/D Slope Virtual origin Experiment*


(m) characteristic radial decay (mm)
decay region region 0.4 0.1327 –75
(FLUENT) (FLUENT) 0.8 0.137 –78
0.24 0.4 –0.2664 –0.5164 1 0.1265 –108
0.8 –0.3278 –0.6097 1.2 0.1424 –97 Slope = 0.38
1 –0.3707 –0.7147 1.4 0.1792 –39 Virtual origin at
1.2 –0.2391 –0.7732 20 1.5 0.188 –21 –20 mm
1.4 –0.6235 –1.1104 1.6 0.1947 –7
1.5 –0.5973 –1.1552 1.7 0.2021 9
1.6 –0.6819 –1.3102 1.9 0.1444 –27
1.7 0.7474 –1.3852
1.9 –1.02 –1.9243 * Based on C/D of 0.4 and 1.0.
Averages from FLUENT –0.3755 –1.0555
(for all C/D given above) (s = 0.49) (s = 0.45)
Averages from FLUENT
(for C/D of 0.4 and 1) –0.3186 –0.6156
Averages from experiments
(for C/D of 0.4 and 1) –0.59 –1.13
1 0.4 –0.3104 –0.5618
0.8 –0.5954 –0.9482
0.9 –0.4421 –0.6062
1 –0.457 –0.7598
1.1 –0.4838 –0.7787
1.4 –0.6153 –0.9369
1.5 –0.7878 –1.1529
1.7 –1.1043 –1.8316
1.9 –1.4384 –0.999
Averages from FLUENT –0.3837 –0.6608
(for all C/D given above. (s = 0.36) (s = 0.38)
Experimental data not available.)

of the wall jet and hence the location of the region which is
actively involved in the mean circulation within the tank. This
region was described as the active zone by Bittorf and Kresta Figure 13. Impingement point of impeller discharge stream on the tank wall:
(2000). Figure 13 shows the impingement points obtained from Re = 1 ¥ 105. At high off-bottom clearances, the point of impingement of
CFD simulations for different tank and impeller geometries. For the impeller discharge stream is deflected upwards by the secondary
the smaller (D = T/3) impeller and both tank sizes, the discharge circulation. This effect is much larger for the T/2 impeller.
stream impinges on the tank bottom when C/D < 1.2, while
impingement is on the tank wall when C/D > 1.5. It may be
noted that Bittorf and Kresta (2000) found bottom impingement
of the impeller discharge stream for C/D < 1.5. Since clearances The point of impingement of the impeller discharge on the
with C/D > 1 are highly unusual, it may be concluded that for tank wall may also be estimated from purely geometrical
all practical purposes impingement is on the tank bottom for a considerations, assuming that the surrounding flow field does
T/3 impeller and the active volume extends only to the lower not cause any deflection of the impeller discharge stream. If the
two-thirds of the tank (Bittorf and Kresta, 2000). discharge leaves the impeller at an angle of 45°, the axial
For D = T/2 the impeller discharge stream impinges on the distance between the lower surface of the impeller and the
tank wall for C/D > 0.55 (extrapolated value, Figure 13). This impingement point will be equal to the radial distance between
agrees with the observation of Bittorf and Kresta (2000) that the the tank wall and the point of discharge. Since the peak
limit of C/D below which the impeller discharge impinges on velocity in the stream is at r = 0.4D from the axis of rotation, the
the tank bottom is 0.5. Since the impeller discharge impinges impingement point (measured from the bottom of the tank) is
on the tank wall for C/D > 0.55, the active zone is in the middle estimated as:
of the tank with the zones of least activity distributed between
the bottom and the top sections. The relationship between the Ê zI ˆ C Ê Dˆ
Á ˜ = - Á 0.5 - 0.4 ˜ (10)
point of impingement and the location of the active zone is ËT ¯ T Ë T¯
made clearer in Figure 14, which will be discussed shortly.

12 The Canadian Journal of Chemical Engineering, Volume 80, August 2002


In Figure 13, the impingement points calculated from CFD Conclusions
are compared with geometric estimates from Equation (10). For Bittorf and Kresta (2001) showed that the flow field close to the
the T/3 impeller, the deflection of CFD data points from tank walls can be completely characterized by the wall jet
Equation (10) increases at high clearances (C/D ≥ 1.4), indicating Equations (Equations 4, 5 and 6), if the geometry dependent
that the secondary flow must deflect the impeller stream more variables, Ucore and zo , are known. In this study, a CFD
and more as the impeller is lifted up. For the T/2 impeller, benchmark is reported for simulation of the flow at the wall in
deviations from Equation (10) are larger as compared to the T/3 a stirred tank. The simulation, which uses the k-e turbulence
case because of the stronger secondary flow. For C/D > 0.7, the model with standard wall functions, a grid refined at locations
CFD data points are parallel with Equation (10), indicating a close to the tank walls, and a steady state MRF approach, builds
well-developed secondary loop at higher clearances. This on previous studies that focused on the flow near the impeller.
agrees with Kresta and Wood (1993) who reported strong The CFD simulations were compared with the experimental
secondary circulation at C/D > 0.667. For both impeller sizes, data of Bittorf and Kresta (2000) to validate the results and to
impingement point predictions match physical understanding highlight areas where the simulations failed.
well and shed more light on limited experimental data. The main conclusions are as follows:
The active zone in the stirred tank can be directly related to (i) The circulation patterns and the similarity profiles of the
the position of the three-dimensional wall jets driven by the three-dimensional wall jets (Equation 4) extracted from CFD
impeller. Figure 14 shows the axial distance over which the simulations match the experimental data well. This confirms
three-dimensional wall jet is self-similar for a range of impeller the ability of steady state simulations to capture the overall
clearances. As expected, Ucore forms at a higher axial location as characteristics of the near-wall flow field.
the impeller clearance is increased. Moreover, at higher (ii) Accurately predicting the geometry dependent parameters,
clearances this location increases faster with an increase in C/D, Ucore and zo , is a more demanding test. Steady state simulations
due to the deflection of the impeller discharge towards the wall with the k-e model were able to predict Ucore with good
by the secondary circulation loop. The point at which the jet accuracy, giving results which were within 14% of the
disintegrates is close to z/T = 0.7 for all C/D less than 1.4. This experimental values for D = T/3 and within 20% for the D = T/2
result is in good agreement with the experimental results of impeller. The effect of the impeller off-bottom clearance on
Bittorf and Kresta (2000), who observed that the three- Ucore was also captured accurately by the simulations.
dimensional jet ended its upward climb at a height equivalent (iii) The present simulations failed to predict the rate of decay of
to two-thirds of the tank diameter (T ). Moreover, for C/D > 1.5, Um and the expansion of the jet half-width, b, recorded in
when the impeller discharge stream impinges on the tank wall, experiments. Since the virtual origin, zo , is dependent on
the jet region is found to exist only in the middle of the tank, as these quantities, the steady state, snapshot simulations also
was also observed by Bittorf and Kresta (2000). At small failed to predict zo.
clearances, the impeller is located near the lower extremity of The simulations were also used to determine two quantities
the jet, while at higher C/D ratios its location is well within the which provide important information about the flow field, but
jet decay regions. Figures 13 and 14 represent a significant which are not readily available from experiments:
extension of our understanding of the active zone in the tank. (i) the impingement point of the impeller discharge on the tank
wall; and
(ii) the axial distance over which the three-dimensional wall jet
extends.
Both these quantities need high resolution for accurate
determination and are therefore resistant to experimental
approaches. The results presented here show that CFD provides an
efficient method of obtaining consistent values for the impingement
point and the extent of the jet, quantities which deepen our
understanding of the complex flow field in a stirred tank.

Acknowledgements
The initial simulations and analysis were carried out by Mike Serink
(Serink, 1999; Kresta and Serink, 1999; and Kresta et al., 1999). The
financial support provided by NSERC and the software provided by
FLUENT are gratefully acknowledged.

Nomenclature
b radial distance to Um/2, (m)
bjet distance from edge of boundary layer to Um /2, (m)
B constant in Equation (3) which is a function of UR
(1–B = UR/Um)
C impeller off-bottom clearance, (m)
Cm constant used in defining mt
C1e, C2e constants in the k-e model (Table 1)
D impeller diameter, (m)
H tank height, (m)
Figure 14. Axial distance over which jet extends: T = 0.24 m, D = T/3, N
= 1000 rpm. Jijk turbulent diffusion term in RSM equation (Table 1), (kg/m·s3)

The Canadian Journal of Chemical Engineering, Volume 80, August 2002 13


uu
k turbulent kinetic energy ( k = i i ), (m2/s2) Stirred Tanks Agitated with Axial Impellers”, Chem. Eng. Sci., 55,
2
N impeller speed, (rps) 1325–1335 (2000).
Njs just suspended speed of impeller for solids suspension, (rps) Bittorf, K.J. and S.M. Kresta, “3-D Wall Jets: Axial Flow in a Stirred Tank”,
p fluctuating component of pressure, (N/m2) AIChE J., 47, 1277–1284 (2001).
P mean pressure, (N/m2) Bittorf, K.J., “The Application of Wall Jets in Stirred Tanks with Solids
Pij production term in RSM equation (Table 1), (kg/m·s3) Distribution”, PhD thesis, University of Alberta, Edmonton, AB (2000).
r position vector relative to the origin of rotation given in Bradshaw, P., “The Best Turbulence Models for Engineers”, in
Equation (7), (m) “Modeling Complex Turbulent Flows”, M. Salas, J.N. Hefner and
r,q,z cylindrical co-ordinates; also denoted as FLUENT L.Sakell, Eds., ICAS/LaRC Interdisciplinary Series in Science and
co-ordinates j, i and k (Figure 2) where r = j, q = i and k = z Engineering, Kluwer Academic Pub., The Netherlands (1999), pp. 9–28.
Re Reynolds number (ND2/n) Derksen, J., and H.E.A. Van den Akker, “Large Eddy Simulations on the
Rij Reynolds stress, (Pa) Flow Driven by a Rushton Turbine”, AIChE J. 45, 209–221 (1999).
s standard deviation based on sample Fokema, M.D., S.M. Kresta and P.E. Wood, “Importance of Using the
S modulus of mean rate of strain, S ∫ 2Sij Sij , (s–1) Correct Impeller Boundary Conditions for CFD Simulations of Stirred
Sij mean rate of strain tensor, (s–1) Tanks”, Can. J. Chem. Eng. 72, 177–183 (1994).
t time, (s) Harvey III, A.D., C.K. Lee and S.E. Rogers, “Steady-state Modeling and
T tank diameter, (m) Experimental Measurement of a Baffled Pitched-Blade Impeller
ui ith component of fluctuating velocity, (m/s) Stirred Tank”, AIChE J. 41, 2177–2186 (1995).
Ucore jet core velocity, (m/s) Harvey III, A.D. and S.E. Rogers, “Steady and Unsteady Computations
U axial component of velocity, (m/s) of Impeller-Stirred Reactors”, AIChE J, 42, 2701–2712 (1996).
Ui ith component of the mean velocity, (m/s) Jaworski, Z., W. Bujalski, N. Otomo and A.W. Nienow, “CFD Study of
Um local maximum velocity in the jet, (m/s) Homogenization with Dual Rushton Turbines: Comparison with
UR recirculating velocity, (m/s) Experimental Results. Part I: Initial Studies”, Trans. IChemE, Part A
v absolute velocity vector, (m/s) 78, 327–333 (2000).
vr relative velocity vector, (m/s) Jones, W.P. and B.E. Launder, “The Prediction of Laminarization with a
Vtip impeller tip speed, (m/s) Two-equation Model of Turbulence”, Int. J. Heat Mass Transfer 15,
x position vector in Cartesian coordinates, (m/s) 301–314 (1972).
xo origin vector of the rotating frame, (m/s) Kirkpatrick, A.T. and A.E. Kenyon, “Flow Characteristics of Three-
xi Cartesian coordinates, (m) dimensional Wall Jets”, ASHRAE Trans. 104, Part 1B, ASHRAE, GA,
y radial distance from tank wall, (m) 1755–1762 (1998).
yjet radial distance in outer region of jet, (m) Kresta, S.M. and P.E. Wood, “Prediction of the Three-Dimensional Flow
ym radial distance to Um , (m) in Stirred Tanks”, AIChE J. 37, 448–460 (1991).
y0 radial distance at which U = 0, (m) Kresta, S.M. and P.E. Wood, “The Mean Flow Field Produced by a 45°
z axial distance from tank bottom, (m) Pitched Blade Turbine: Changes in Circulation Pattern Due to Off
zo virtual origin, (m) Bottom Clearance”, Can. J. Chem. Eng. 71, 42–53 (1993).
zI impingement point of impeller discharge on tank wall, (m) Kresta, S.M., “Boundary Conditions Required for the CFD Simulations
of Flows in Stirred Tanks”, in “Multiphase Reactor and
Greek Symbols Polymerization System Hydrodynamics, Advances in Engineering
ae, ak constants in diffusion terms of RNG k-e transport equations Fluid Mechanics Series”, N.P. Cheremisinoff, Ed., Gulf Pub. Co.,
for k and e (Table 1) Houston, TX (1996), pp. 297–316.
d proportionality constant relating boundary layer Kresta, S.M. and M. Serink, “Use of Computational Fluid Dynamics
thickness to b (CFD) in Conjunction with Experiments to Determine the Jet
dij Kronecker delta function, dij = 1 when i = j; d = 0 when i π j Impingement Point in a Stirred Tank”, AIChE Annual Meeting,
e dissipation of turbulent kinetic energy (k) given by terms in Dallas, TX, Oct. 31-Nov. 5 (1999).
Table 1, (m2/s3) Kresta, S.M., M. Serink and K. Bittorf, “Use of Computational Fluid
eij dissipation term in RSM equation (Table 1), (kg/m·s3) Dynamics (CFD) in Conjunction with Experiments to Determine the
h dimensionless radial distance given as y/b Jet Impingement Point in a Stirred Tank”, 49th Canadian Chemical
m dynamic (molecular) viscosity, (Pa·s) Engineering Conf., Oct. 3–6, 1999, Saskatoon, SK (1999).
mt turbulent or eddy viscosity (Table 1), (Pa·s) Kresta, S.M., K.J. Bittorf and D.J. Wilson, “Internal Annular Wall Jets:
meff effective viscosity (meff = m + mt), (Pa·s) Radial Flow in a Stirred Tank”, AIChE J. 47, 2390–2401 (2001).
n kinematic viscosity (m/r), (m2/s) Launder, B.E. and D.B. Spalding, “Lectures in Mathematical Models of
r density, (kg/m3) Turbulence”, Academic Press, London, UK (1972).
sk, se constants in k-e model Launder, B.E. and D.B. Spalding, “The Numerical Computation of
F similarity constant in Equation (1) Turbulent Flows”, Computer Methods Applied Mech. Eng. 3,
Fij pressure-strain distribution in RSM equation (Table 1), (kg/m·s3) 269–289 (1974).
w angular velocity vector of reference frame, (s–1) Launder, B.E., G.J. Reece and W. Rodi, “Progress in the Development of
a Reynolds-Stress Turbulence Closure”, J. Fluid Mech. Part 3 68,
References 537–566 (1975).
Bakker, A., K.J. Myers, R.W. Ward and C.K. Lee, “The Laminar and Launder, B.E. and W. Rodi, “The Turbulent Wall Jet: Measurements and
Turbulent Flow Pattern of a Pitched Blade Turbine”, Trans. IChemE, , Modeling”, Ann. Rev. Fluid Mech. 15, 429–459 (1983).
Part A 74, 485–491 (1996). Launder, B.E., “Second-moment Closure: Present…and Future?”,
Bittorf, K.J. and S.M. Kresta, “Active Volume of Mean Circulation for Interm. J. Heat Fluid Flow 10, 282–300 (1989).

14 The Canadian Journal of Chemical Engineering, Volume 80, August 2002


Launder, B.E., “Modelling the Formation and Dispersal of Streamwise Ranade, V.V. and S.M.S. Dommeti, “Computational Snapshot of Flow
Vortices in Turbulent Flow”, 35th Lanchester Lecture, Aeronaut. J. 99, Generated by Axial Impellers in Baffled Stirred Vessels”, Trans
(990), 419–431 (1995). IChemE 74, Part A, 476–484 (1996).
Launder, B.E., “The Modeling of Turbulent Flows With Serink, M., “The Impingement Point of the Discharge of a Pitched Blade
Significant Curvature or Rotation”, in “Modeling Complex Turbine in a Stirred Tank”, Senior Research Project Report, University
Turbulent Flows”, M. Salas, J.N. Hefner and L. Sakell, Eds., ICAS/LaRC of Alberta, Edmonton, AB (April 1999).
Interdisciplinary Series in Science and Engineering, Kluwer Academic Wood, P.E. and C.P. Chen, “Turbulence Model Predictions of Radial Jet
Pub., The Netherlands (1999), pp. 29–51. – A Comparison of k-e Models”, Can. J. Chem. Eng. 63, 177–182
Mohammadi, B. and O. Pironneau, “Analysis of k-epsilon Turbulence (1985).
Model”, John Wiley and Sons, Paris, France (1994).
Perng, C.-Y. and J.Y. Murthy, “A Moving Mesh Technique for Simulation
of Flow in Mixing Tanks”, AIChE Meeting, Miami Beach, FL (1992).
Rajaratnam, N., “Developments in Water Science: Turbulent Jets”, Manuscript received August 7, 2001; revised manuscript received
Elsevier Scientific Publishing Co., New York, NY (1976). March 20, 2002; accepted for publication May 8, 2002.

The Canadian Journal of Chemical Engineering, Volume 80, August 2002 15

You might also like