Chemical Engineering Science 66 (2011) 2935–2944

Contents lists available at ScienceDirect

Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces

Mixing times for process vessels with aspect ratios greater than one
T.L. Rodgers a,, L. Gangolf b, C. Vannier b, M. Parriaud b, M. Cooke a
a
School of Chemical Engineering and Analytical Science, University of Manchester, M60 1QD, UK
b
Ecole Nationale Supérieure des Industries Chimiques, BP 20451, 54001 Nancy Cedex, France

a r t i c l e i n f o abstract

Article history: Stirred tank reactors are one of the most important and common pieces of equipment used in speciality,
Received 2 March 2010 pharmaceutical, and agrichemical processes. It is also typical for these to be operated at large aspect
Received in revised form ratios; however, there is very little information in the open literature about mixing times in vessels
25 August 2010
with aspect ratios greater than one. This paper aims to provide new information in this area that will
Accepted 16 March 2011
enable the design of better reactors.
Available online 22 March 2011
Electrical resistance tomography is used to monitor the mixing time of systems with aspect ratios
Keywords: higher than one. The mixing time has been measured on vessels of 0.914 and 0.610 m diameter with
Homogenisation Cowles disc, Rushton turbine and mixed flow type impellers to aspect ratios of 2. The current
Large aspect ratio
correlation of choice, by Grenville and Nienow (2004), has been compared with the results and found
Mixing time
to under predict the mixing time at aspect ratios greater than one. The exponent on the H/T term has
Scale-up
Stirred vessel been explored and it has been found that this varies with agitator type, this information has never been
Tomography shown before.
The affect of adding a second impeller on the mixing time and flow pattern is also investigated.
Adding a second Rushton turbine creates zoning in the vessel which impedes the mixing; this can be
visualised using electrical resistance tomography.
& 2011 Elsevier Ltd. All rights reserved.

1. Introduction determining the mixing time have been used, e.g. dye addition
(Mann et al., 1987), pH shift (Singh et al., 1986), tracer monitored by
Stirred tank reactors are one of the most important and conductivity probes (Cooke, 1988; Khang and Levenspiel, 1976;
common pieces of equipment used in speciality, pharmaceutical, Ruszkowski, 1994), flow followers (Bryant and Sadeghzadeh, 1979),
and agrichemical processes (Paul et al., 2004); many of which are and electrical resistance tomography (Cooke et al., 2001).
operated at aspect ratios (the ratio of the liquid height to the It has been shown that the mixing time depends on the specific
vessel diameter) greater than one. The energy required for these power input ðP=rVÞ, the tank geometry and the agitator Reynolds
processes forms a significant part of the total energy used. The number (Re, this is the ratio of the inertial force supplied by the
process of mixing occurs as a result of the motion at three levels: agitator to the viscous force of the fluid). It has been common place
molecular, eddy, and bulk motion. in the literature to correlate the dimensionless mixing time ðy95 NÞ
A large amount of work has been carried out on the mixing of against other dimensionless values which represent the above
liquids within the turbulent regime for Newtonian fluids (Grenville variations, i.e. the agitator power number (Po), the impeller to tank
and Nienow, 2004), and a lot of work has gone into trying to predict diameter ratio (D/T), the liquid height to tank diameter ratio (aspect
the mixing time (the time taken for the system to become homo- ratio, H/T), and the agitator Reynolds number. For mixing in the
geneous on addition of material) from knowledge of the agitator and turbulent regime the dimensionless mixing time is independent of
the operating conditions; however, very little of this work has been the Reynolds number and has been correlated to Eq. (1) (Grenville
on aspect ratios greater than one. Currently there are no studies that and Nienow, 2004) which has a standard deviation of 10%.
have tried to correlate the mixing times for vessels with aspect  1=2
H
ratios much greater than one, or thoroughly studied different types y95 N ¼ 5:2Po1=3 ðD=TÞ2 ð1Þ
T
of agitator under these conditions. Many different methods of
It should be commented on here that the data for this correlation
was mainly collected on vessels with H=T r1 and it is not really
 Corresponding author. Tel.: þ 44 161 306 4358. applicable to large aspect ratios. In fact, most of the previous work
E-mail addresses: T.Rodgers@postgrad.manchester.ac.uk, has been predominantly carried out at liquid levels equal to or less
mjki3tr2@gmail.com (T.L. Rodgers). than the vessel diameter; of the 25 papers reviewed for mixing time

0009-2509/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2011.03.036
2936 T.L. Rodgers et al. / Chemical Engineering Science 66 (2011) 2935–2944

correlations by Nere et al. (2003) only one looks at values of H/T  Increasing the fluid height will increase the mixing time
greater than one and their maximum value is 1.11. (so long as a is greater than 1/3).
Cooke (1988) has shown that for H=T b1 the exponent on H/T  The mixing time is independent of the fluid viscosity at a
can increase considerably due to zoning (zoning refers to the constant power input.
phenomenon where competing mixing loops hinder the exchange  At a constant power per unit mass the mixing time is slightly
of material from one loop to the other). The value of this exponent longer in a dished base tank compared to a flat bottom vessel.
was found to be 2.43 which is a lot larger than 0.5. However, it is  When scaling-up at constant power per unit mass and geo-
unclear if this dramatic increase in mixing time (over 10 times metry the mixing time will increase with the tank diameter.
longer for H¼3T compared to H¼T) was due to just the increase in
height or due to zoning effects as three Rushton turbines were used. The flow patterns generated by radial and axial impellers are
Zoning has been shown to significantly increase the mixing time in shown schematically in Fig. 1. For a radial impeller the flow travels
vessels, which is agreed with by Otomo et al. (1995) who have radially out from the impeller towards the vessel wall, where it
shown that by using a dual axial system, with much less zoning, splits and approximately half the flow moves up the vessel and the
they can reducing the mixing time at large aspect ratios by up to rest moves down the vessel. These flows then move around the
50% of that predicted by Cooke (1988). Vrábel et al. (2000) also show vessel back towards the agitator, Fig. 1(a). For an axial impeller, if
long mixing times, N y95  300, for four Rushton turbines and a down-pumping, the flow travels axially down the middle of the
significant reduction when replaced with axial impellers. vessel towards the vessel base, where it moves along the base and
Taking Eq. (1) but having an unknown exponent on H/T, a, back up the outside wall of the vessel. This flow then moves towards
gives the top of the vessel where it loops over back to the agitator,
 a Fig. 1(b). For an up-pumping impeller the flow pattern is the same
H
y95 N ¼ 5:2Po1=3 ðD=TÞ2 ð2Þ but just in the opposite direction. For a higher liquid height it has
T
been reported that there can be zoning and secondary flow loops
Eq. (2) can be rearranged using the definition of the power towards the top of the vessel both for radial impellers, Fig. 2(a), and
number to give Eq. (3), where r is the liquid density, V is the axial impellers, Fig. 2(b). This zoning can cause the mixing time to
liquid volume in the tank given by Eq. (4), where a is the ratio of a significantly increase, especially with a second impeller reinforcing
dished base height to the tank diameter for a half-ellipsoid base this zoning (Cooke, 1988).
(this is very similar to a torispherical base but simpler to consider, van’t Riet and Tramper (1991) use a bulk flow model analysis
0 for a flat base to 0.5 for a hemispherical base). based on turn over times in the vessel. A modification of this
 1=3  a   method can be given by Eq. (5), where the volume of the vessel is
P H H a 1=3
y95 ¼ 5:64 ðD=TÞ1=3 T 2=3  ð3Þ given by Eq. (6).
rV T T 3  
  H
p H a Ny95 pPo1=3 ðD=TÞ3 ð5Þ
V¼ T3  ð4Þ T
4 T 3
 
p H
Eq. (3), which only applies to the turbulent regime, reveals that: V¼ T3 ð6Þ
4 T
 Maintaining constant tank dimensions, agitators of the same This model produces the same dependence on power as Eq. (1)
diameter have the same mixing time if they have the same (Grenville and Nienow, 2004) but has different dependence on both
power input. D/T and H/T. The reason for this is most likely that the actual flow in
 Increasing the agitator diameter will decrease the mixing time the vessel is not like this ideal flow, and the constrained flow of the
at constant power input. vessel means that there is a smaller dependence on the agitator size.

Fig. 1. Schematic flow patterns for (a) radial type agitators and (b) axial type agitators.
 T.L. Rodgers et al. / Chemical Engineering Science 66 (2011) 2935–2944 2937

Fig. 2. Schematic of secondary flow loops for (a) radial type agitator and (b) axial type agitator.

Nienow (1997) uses a turbulence model approach based on

energy dissipation to examine the mixing time. This method Surface Feed
predicts Eq. (7) which agrees with the power dependence and T
D/T dependence of Eq. (1) (Grenville and Nienow, 2004).
 2
Ny95 pPo1=3 D=T ð7Þ

This method predicts the exponent on D/T by comparison with

empirical relations such as Eq. (1), so it is not surprising that it H
agrees. The basis of this theory is that the mixing time is inversely D
proportional to the turbulent diffusion, which means that it is
proportional to the local energy dissipation to the power of minus
a third, ype1=3 . The flow far from the agitator determines the
mixing time, so the local energy dissipation rate at the vessel
walls is the important factor. This value can be given by a
proportional value between it, eR , and the average energy dis- T/2
sipation rate, e ¼ ð4=pÞPoN 3 ðD=TÞ3 D2 , such that ype 1=3 eR . It is
1=3
a
assumed by Nienow (1997) that eR pðD=TÞ which allows it to
give the final equation with (D/T)  2 (gives a ¼ 1). This could be
extended to assume that eR pðD=TÞa ðH=TÞb which would give a T/10
variation with H/T; however, there is not enough information to Fig. 3. Schematic representation of the stirred tank reactor geometry for the
find the value of b. Studies of energy dissipation in vessels with experiments.
methods such as LDA (laser Doppler anemometry) (Ducci and
Yianneskis, 2005) have shown that there is variation of energy
dissipation with height, and could calculate values of b, but not baffles of width T/10. A dished base has been used in this study as
enough information has yet been collected. they are commonly used in industrial processes. The cylindrical
This paper investigates the variation of the mixing time with section is 1.5 m high and is fitted inside a square jacket through
change in the liquid height. This is important to aid in the design which water can be circulated for temperature control. The square
of reactors which often operate at these higher heights. This paper jacket provides distortion free viewing windows for flow visualisa-
also compares the mixing times of systems with multiple agita- tion. The agitator rotational speeds are monitored using a Ferro-
tors as multiple agitators have previously been seen to effect the magnetic proximity sensor coupled to a COMPACT MICRO 48
mixing time in surprising ways. tachometer. The shaft is continuous and fits into a PTFE bearing
located in the centre of the dished base. The agitator power is
measured using a dual torque strain gauge system bonded to the
2. Material and methods shaft, designed to cancel out bending moments. The strain gauge is
wired to an ASTECH Strain Gauge Bridge fitted with a telemetry
2.1. Experimental equipment and procedure readout, which is connected to a LABVIEW data acquisition unit. The
temperature of the vessel contents is monitored using a Pico PT100
Two vessels were used in the experiments, a 0.914 m (3 ft) and a temperature probe inserted through an analysis block. The analysis
0.610 m (2 ft) diameter vessel. The 0.914 m vessel is a Perspex block runs all the way up the back of the vessel allowing probes
cylindrical vessel with a shallow dished base and four standard to be inserted through 3/4 in BSP (British standard pipe) tappings.
2938 T.L. Rodgers et al. / Chemical Engineering Science 66 (2011) 2935–2944

Fig. 4. Agitators used: (a) Cowles disc, (b) six blade Rushton turbine and (c) four blade mixed flow agitator.

The vessel is also fitted with eight rings of 16 equally spaced EIT Table 1
electrodes in a baffle cage configuration which are connected to an ITS P2000 data acquisition settings.
ITS P2000 tomography measurement system. Fig. 3 shows a sche-
Setting Value
matic of this vessel. The 0.610 m vessel is a geometrically scaled
down version of the 0.914 m vessel; however, the cylindrical section Excitation frequency 9.6 kHz
is still 1.5 m high. Sampling time interval 40 ms
Three different types of agitator are used in the experiments The Samples per frame 1
Frames per download Equal to the maximum number of frames
first is a Cowles disc agitator (CD), Fig. 4(a), with 32 angled teeth Delay cycles 3
(16 up and 16 down) of diameter DCD ¼T/3 and a clearance of c¼T/2. Injection current 50 mA
The agitator was driven clockwise for the experiments, angled tooth Sampling strategy Normal adjacent
edge first. The second is a Rushton turbine (RT), Fig. 4(b), with six
blades of diameter DRT ¼ T/2 and T/3, and a clearance of c¼T/2. The
third is a mixed flow agitator in either an up-pumping modality
(MFU), Fig. 4(c), or a down-pumping modality (MFD) with four electrodes. The electrodes are mounted on a baffle cage, constructed
blades of diameter DRT ¼T/2 and a clearance of c¼T/2. of 16 thin wall-flush plastic ducts, each with eight electrodes. The
These agitators have been used in this study as they are typically bottom plane of electrodes sits 0.5 cm above the join between the
used in industrial mixing processes. Rushton turbines and four blade cylindrical part of the vessel and the dished base.
mixed flow agitators are commonly used for low viscosity blending; Electrical impedance tomography (EIT) is commonly used to
however, Cowles discs are generally used for high viscosity in low monitor processes which require good temporal and spacial
viscosity bulk blending. The mixing times still need to be studied for resolution. It has been used to measure mixing times in industrial
Cowles discs as this is still an important factor for processing. All the single phase and multi-phase systems (York, 2001). The ITS P2000
results were taken in the turbulent regime for all the agitators. was chosen for the experiments presented in this paper as it is the
This was examined visually and also was shown by generation of best performing EIT instrument, available to us, for experiments
constant values of the dimensionless mixing time with variation in requiring high temporal resolution and is capable of successfully
agitation rate, which is only true for the turbulent regime. monitoring homogeneity (Stephenson et al., 2007). Table 1 sum-
Experiments for the mixing times were carried out by adding a marises the data acquisition settings employed during the experi-
brine solution (50 g table salt dissolved per litre of tap water) to the ments. The signal-to-noise ratio (SNR) was checked to ensure that
surface of the liquid. For visualisation runs the tracer was also dyed no voltage measurements were saturating the analogue to digital
with 50 g per litre water soluble Nigrosine dye (Fisher). The brine converter. If this occurs the recorded voltages read a constant
solution was added to the vessel in 100 ml aliquots using a funnel value for the given reading thus making the image reconstruction
placed less than 1 cm from the surface in a fixed position 2/3 of the unreliable for calculating the conductivities inside the vessel.
radius from the agitator shaft and between two baffles. The energy Therefore, a SNR over 100 dB should be avoided. The optimum
input of the addition was small compared to that given by the injection current was found to be 50 mA resulting in a SNR ratio
agitator. The mixing times were calculated over a range of agitation of approximately 52 dB for water at the start of the experiment.
rates (50–300 rpm) and several repeats were taken for each speed. The SNR is the ratio of the mean of one voltage reading over a
The funnel position was tested and no variation in the mixing time number of frames, n, to the standard deviation of the reading,
was discovered, this was also true for injection into the agitator. averaged over the number of readings in a frame,
The turbulent power number was calculated using the cali-   " P #
n
brated dual strain gauge system mentioned above for the Cowles Mean V2
SNR ¼ 20log ¼ 10log n P i ¼ 1 i2 1
disc agitator. The power of an agitator can be calculated from the Standard deviation n
i ¼ 1 Vi
definition of the power number or from the torque. Rearranging ð9Þ
these definitions for the torque gives Eq. (8). In the turbulent
regime, the power number is a constant which means that the An accurate geometric model of the vessel, including the exact
torque is proportional to the agitation speed squared, as shown in electrode positions, was developed using constructive solid geo-
the following equation: metry for all the liquid heights (Fig. 5(a) gives and example for
H¼T). The baffles and the agitator were also modelled as they
Porb D5 2
G¼ N ð8Þ strongly affect the measured voltage signals (Stephenson, 2008)
2p
and so are requisite for the finite element model accuracy
(Rodgers and Kowalski, 2010). The geometric model was meshed
2.2. Tomography settings and reconstruction (using an advancing front surface mesh and Delaunay techniques)
to give tetrahedral elements using the Netgen mesh generator
The vessel is equipped with an array of 128 steel electrodes, 5 cm (Schöberl, 1997) (Fig. 5(b) gives an example for the Cowles disc at
high and 3 cm wide, set in eight rings of 16 equally spaced H¼T and has 8198 elements).
 T.L. Rodgers et al. / Chemical Engineering Science 66 (2011) 2935–2944 2939

Table 2
Power numbers of the agitators used in the experiments.

Agitator Po

Cowles disc T¼0.914 0.52

Cowles disc T¼0.601 0.60
4MFU 1.09
4MFD 1.33
Rushton turbine D/T ¼1/2 4.66
Rushton turbine D/T ¼1/3 5.11

produce an analysis of the mixing time,

" 2 #
N 
1 1X st,i s0,i
ln½sRMS  ¼ ln 1 ð10Þ
2 N i ¼ 1 sinf,i s0,i

Fig. 5. Representation of the stirred tank for the tomography reconstruction: In Eq. (10), sRMS is the root mean squared conductivity, N is the
(a) electrode configuration for H¼T liquid level (the top planes of electrodes are number of elements, st,i is the conductivity of element i at the
not shown as they are above the water level, in this case) and (b) finite element
given time step, s0,i is the initial mean conductivity for element i
model for H¼T liquid level (only surface elements visible).
and sinf,i is the final mean conductivity for element i. This
approach gives a good calculation of the mixing time as it weights
The collected voltage data was reconstructed using a general- the answer towards the elements that take the longest to mix.
ised singular value decomposition (gsvd) algorithm (Hansen, The vessel can be considered 90% mixed if the value of lnðsRMS Þ
1989) based on the finite element model. This approach decom- drops below  2.3, which is equivalent to ð1=2Þln½ð0:91Þ2 . The
poses the image into individual spatial frequency components 90% mixing time is then the time at which the value never
and affords the ability to control the number of generalised becomes greater than 2.3 again. When this is the case the
singular values incorporated into the solution. The inclusion of a variation from the end homogeneous concentration is less than
low number of singular values in the solution yields an image 10%. The 90% mixing time is used as there is about 8% noise on the
with lower spatial resolution but which is robust to measurement final mean conductivity.
noise. Conversely, the inclusion of a high number of singular As the whole vessel is in the turbulent regime then this time
values yields an image with potentially higher spatial resolution can be scaled to a 95% mixing time for comparison with literature
which is less robust to measurement noise. The regularisation correlations by use of the following equation:
parameter in the algorithm was chosen frame-by-frame to be the
value obtained by an analysis of the discrete Picard condition ln½10:95
(Hansen, 1990). The discrete Picard condition compares the
y95 ¼ y90 ð11Þ
ln½10:9
generalised singular values (representing the change in the data)
with the Picard coefficients (representing the noise in the data), This methodology was checked by testing the scaling from y80 to
and gives the value where they are equal. Values of the general- y85 and to y90 for all the runs.
ised singular values greater than the Picard coefficients contain Fig. 6 shows the global average conductivity trace for a single
recoverable data and should be utilised, which occurs if the experiment with the Cowles disc at 100 rpm with H¼T. The start
regularisation parameter is set to this equality (Rodgers and of the trace shows the initial water concentration of 0.01 S m  1.
Kowalski, 2010). The regularisation parameter varies with the At 3 s the tracer is added and the global conductivity increases
frame number; during the mixing the regularisation parameter steeply to it’s peak value of about 0.01055 S m  1. The global
increases as the sharp changes in conductivity create more noise conductivity then falls off as the brine tracer is distributed around
in the data. The automatic identification of optimum algorithm the vessel. It is difficult to tell from this graph how long the
parameters, based on measured data, is a novel development and process takes to mix. This is the reason that the root mean
contrasts strongly with common practice where parameters are squared conductivity method is used. Fig. 7 shows the root mean
selected based on best fit with known or anticipated solutions. squared conductivity trace for the same experiment as Fig. 6.
The pre-addition value in Fig. 7 can be seen to around 0 with
small variation, this corresponds to the initial conductivity of
3. Results and discussion 0.01 S m  1. When the trace is less than  2.3 this corresponds to
the final conductivity of about 0.0102 S m  1. It is clear from Fig. 7
3.1. Agitator power numbers that the mixing seems to occur as an exponent (i.e. is a linear
trend on the log scale), as would be expected. Using Fig. 7 the
The power numbers for the agitators were measured using the mixing time can be calculated. The arrow shows the time at
method outlined in Section 2.1 and the results are given in which the system can be considered 90% mixed and the vertical
Table 2, the power number of the agitators did not appreciably line shows the time of addition, this means the mixing time is the
change with liquid height or speed. As the mixing time is related difference.
to the power number, the Rushton turbine will mix the fastest, The variation in the mixing time, y95 , can be seen in Fig. 8 for
followed by the mixed flow, and then the Cowles disc. the Cowles disc with a liquid height of H¼T for a range of
agitation speeds, N. With plotting the mixing time against the
reciprocal of the agitation speed means that the dimensionless
3.2. Agitator mixing times mixing time, y95 N, is given by the gradient of the line. This figure
also shows the error of the results, as it shows the variation in the
The values of the conductivity for the finite elements were collected data. The variation in the agitation rate is less than 1%
combined together using the root mean squared conductivity to throughout the experiment and the variation in power is less than
2940 T.L. Rodgers et al. / Chemical Engineering Science 66 (2011) 2935–2944

0.0106 400
CDT = 0.914
CDT = 0.601
350
Global average conductivity / S m-1

0.0105 6RTD/T = 1/3

6RTD/T = 1/2
4MFD
300 4MFU
0.0104
250
0.0103

Nθ95
200
0.0102
150
0.0101
100
0.01 50

0.0099 0
0 10 20 30 40 50 0.5 0.75 1 1.25 1.5 1.75 2 2.25
Time / s H/T
Fig. 6. Global average conductivity trace for the Cowles disc at 100 rpm with H¼T. Fig. 9. Variation of the dimensionless mixing time, N y, with change in the aspect
ratio, H/T.

1
10
CD T = 0.914
0 CD T = 0.601
6RT D/T = 1/3
6RT D/T = 1/2
4MFD
Nθ95/(5.2Po-1/3 (D/T)-2)
-1 4MFU
ln [σRMS]

-2 θ90
1
Grenville and Nienow (2004)
-3

-4

-5
0 10 20 30 40 50 0.1
Time / s 0.5 1 1.5 2
H/T
Fig. 7. Root mean squared conductivity trace for the Cowles disc at 100 rpm with
H¼ T. Fig. 10. Variation of y95 N=ð5:2Po1=3 ðD=TÞ2 Þ with H/T and comparison with
Eq. (1) (Grenville and Nienow, 2004).
80
are one standard deviation in the dimensionless mixing time
70 values.
θ95 = 63.01/N As the aspect ratio increases, the dimensionless mixing time
60
increases. This is most likely due to the increased volume of the
50 liquid needed to be mixed and the increase in circulation time.
Further analysis can be made of the results if they are put onto
θ95 / s

40 the same basis, this can be done by rearranging Eqs. (2)–(12). This
means that plotting the logarithm of both sides of the equation
30 will produce a straight line with a gradient equal to a.

20  a
y95 N H
  ¼
1=3 D 2 T
ð12Þ
10 5:2Po T

0 Fig. 10 shows the plot of y95 N=ð5:2Po1=3 ðD=TÞ2 Þ against H/T on a

0 0.25 0.5 0.75 1 1.25 log-log scale. It can be seen that the variation of this value on the
1/N / s plot is a straight line. However, the different styles of agitator do
not all lie on the same line. The gradient of the lines, i.e. the
Fig. 8. Variation of the mixing time, y95 , with 1/N for the Cowles disc with H/T ¼1.
exponent on H/T are given in Table 3.
The gradient produces the value of the dimensionless mixing time N y.
The Cowles disc has the largest exponent, with the mixed flow
agitators showing little difference between each other, and the
5%. These amounts are less than the variation in the collected Rushton turbine with the lowest exponent. The lower the expo-
mixing times. nent the less the mixing time is affected by the height change.
The variation of this dimensionless mixing time with the According to the theory of flow rates (Eq. (5)) the mixing time
change in aspect ratio, H/T is seen in Fig. 9. The error bars given should be proportional to H/T. As the exponent increases away
 T.L. Rodgers et al. / Chemical Engineering Science 66 (2011) 2935–2944 2941

from one the flow pattern in the vertical direction can be viewed 3.3. Comparison of single and double impeller arrangements
as less ideal. The Rushton turbine has the value closest to one
which means that its flow pattern at large aspect ratios is the As previously noted, Cooke (1988) attributed long mixing
most similar to that in Fig. 1(a) and most likely has the least times with multiple agitators to zoning within the vessel caused
zoning. However, the Cowles disc has an exponent much larger by poor interchange between the agitator loops. From Eq. (12)
than one which means that the flow pattern at large aspect ratios and the appropriate value from Table 3, the predicted dimension-
is more similar to Fig. 2(a) with two, or maybe even more, zones. less mixing time with two D/T¼ 1/2 Rushton turbines (total
The reason for this is most likely due to the flow generated by power number of 9.32) with a liquid height of H/T¼1/2 is 22.7.
each agitator; the Rushton turbine has a flow number (ratio of the However, experimentally measuring the mixing time for this
volumetric liquid flow rate to the agitator pumping rate, ND3) of two Rushton turbine combination gives a dimensionless mixing
about 0.73 (Costes and Couderc, 1988), where as the Cowles disc time of 32.5, which is around 43% higher. Visually examining the
has a flow number of about 0.04 (Grenville et al., 2008). The high mixing reveals that with the two Rushton turbines the zone
flow number of the Rushton turbine means that it produces more around the top agitator is mixed before material exchanges into
flow and a more stable flow loop at high aspect ratios unlike the the bottom agitator zone, Fig. 11. However, this is not seen with
Cowles disc, this is why it has a more ideal flow loop. one Rushton turbine as the material is drawn all the way down
The fitted values of exponent on the H/T term fit well for the vessel to the agitator (in the same position as the lower
aspect ratios larger than one; however, they do not fit well for agitator in the two Rushton turbine arrangement), Fig. 12.
aspect ratios less than one. This is most likely due to different With two Rushton turbines after 21 s, Fig. 11(e), the top half of
mechanisms of the flow loops. At large aspect ratios the upper the vessel seems to be well mixed; however, it is clear that very
flow loop expands and secondary zones can form leading to little material has penetrated into the bottom half of the vessel.
exponents greater than one on H/T. At aspect ratios less than Whereas with only one Rushton turbine the vessel is almost fully
one the top loop is constrained and the mixing time will mainly mixed at 21 s, Fig. 11(g). This shows the large affect of the zoning
depend on the size of the loop below the agitator, as this is the with the vessel and helps to clearly show why Cooke (1988) found
biggest. This logic dictates that the mixing time will not change as such large mixing times for three Rushton turbines.
the bottom loop size will not change, however, as the top loop is As there is this clear divide between the bottom and top half of
constrained some of the flow will be forced into the bottom loop the vessel for the two Rushton turbines combination and the
enhancing the mixing, leading to a slight reduction in mixing time electrical resistance tomography takes measurements over the
with decreasing aspect ratio. In fact, it can be seen from Fig. 10 whole vessel, then the measurements can be divided into these
that Eq. (1) (Grenville and Nienow, 2004) is a better fit for aspect two regions as well, producing Fig. 13.
ratios less than one, i.e. those that it was correlated with. Fig. 13 clearly demonstrates that the bottom of the vessel
receives very little of the tracer until a while into the mixing
Table 3
process. In fact, Fig. 14 shows that the top area of the vessel can
Values of a for Eq. (2).
be considered mixed almost 12 s before the bottom of the vessel.
Agitator a Material only enters the bottom area of the vessel slowly from the
top area, through the almost barrier-like zone boundary.
Cowles disc 2.72 The value of the exponent in Eq. (2) for the two Rushton
MFU/MFD 1.67
Rushton turbine 1.20
turbines is calculated to be 1.72. Values of this same parameter
can be calculated from data by Cooke (1988), for three turbines,

Fig. 11. Visualisation of the vessel for a tracer addition with two Rushton turbines—top row and the tomographic reconstruction—bottom row. (a) Start, (b) 8 s, (c) 10 s
(d) 13 s (e) 21 s (f) 26 s (g) 25 s and (h) 45 s.
2942 T.L. Rodgers et al. / Chemical Engineering Science 66 (2011) 2935–2944

Fig. 12. Visualisation of the vessel for a tracer addition with one Rushton turbine—top row and the tomographic reconstruction—bottom row. (a) Start, (b) 3 s, (c) 6 s
(d) 9 s (e) 11 s (f) 14 s (g) 21 s and (h) 40 s.

0.0106 1
Top half Top half
Bottom half Bottom half
Global average conductivity / S m-1

0.0105 Total vessel 0 Total vessel

-1
0.0104
θ90
ln [σRMS]

-2
0.0103
-3
0.0102
-4
0.0101
-5
0.01
-6
0 10 20 30 40 50 60
0 10 20 30 40 50 60
Time / s
Time / s
Fig. 13. Global average conductivity trace for two Rushton turbines at a liquid
Fig. 14. Root mean squared conductivity trace for two Rushton turbines at a liquid
height of H/T¼2 for the top half of the vessel only, the bottom half of the vessel
height of H/T¼ 2 for the top half of the vessel only, the bottom half of the vessel
only, and the total vessel.
only, and the total vessel.

and Vrábel et al. (2000), for four turbines. This data is presented in (2005) with three agitators shows slight zoning in their images,
Fig. 15; where it can be seen that there is an approximate linear but they do not develop this further.
relationship between them. The values calculated from data from Knowledge of the effect of multiple agitators is very important
Cooke (1988) and Vrabel et al. (2000) seem to fit well. when designing reactors as a reaction could occur in the region of
Values recorded by Kukuková et al. (2005) and Moštěk et al. one agitator before even passing into the rest of the vessel. This
(2005) using computational fluid dynamics agree well with the could be severely detrimental, especially with competing reactions
data shown in Fig. 15. The value of the exponent for two Rushton or bio-reactions where oxygen concentration is very important.
turbines is slightly greater than in this study, but the value However, certain processes could make use of this fact by almost
calculated from data from Kukuková et al. (2005) is only based keeping reactions separated in zones of the vessel while they react.
on their limited data. These authors also analyse the mixing time
of multiple pitched blade turbines (six blades instead of this
studies four blades), they also predict that the exponent increases 4. Conclusions
up to 2.1 for four agitators. This is less than for four Rushton
turbines which is expected as the pitched blade turbines should  Increasing the aspect ratio increases the mixing time. Different
not form as many zones as the Rushton turbines. However, they agitators have a different affect on this increase. This is
only present limited data on this, one speed per arrangement. The important to know when designing reactors and has never
computational fluid dynamics simulations of Montante et al. been discussed before.
 T.L. Rodgers et al. / Chemical Engineering Science 66 (2011) 2935–2944 2943

3 gsvd generalised singular value decomposition

MFD mixed flow agitator down-pumping
MFU mixed flow agitator up-pumping
2.5
RT Rushton turbine

1.5
a

Acknowledgements

1 Tom Rodgers would like to thank The University of Manche-

ster’s EPSRC CTA (Collaborative Training Account) and Unilever
0.5 for financial support during his PhD. The authors would like to
thank the SCEAS workshop staff who helped with equipment
modifications and construction.
0
0 1 2 3 4
No. of RTs References

Fig. 15. Variation of a in N y95 ¼ 5:2Po1=3 ðD=TÞ2 ðH=TÞa against the number of
Bryant, J., Sadeghzadeh, S., 1979. Circulation rates in stirred and aerated tanks.
Rushton turbines. Value for 3 from Cooke (1988), value for 4 from Vrábel et al. In: Procedures of the Third European Conference on Mixing, York, UK,
(2000). pp. 325–336.
Cooke, M., 1988. Processing of Solid–liquid Suspensions. Butterworth-Heinemann,
UK, pp. 192–221.
 The smaller the flow from the agitator the larger the exponent Cooke, M., Bolton, G., Jones, D.H., Housley, D., 2001. Demonstration of a novel
on the H/T term seems to be. retrofit tomography baffle cage for gas–liquid mixing studies under intense
 Eq. (2) with the values from Table 3 predicts the mixing time operating conditions. In: Second World Congress on Industrial Process Tomo-
well for aspect ratios greater than one. graphy. Hannover, Germany, pp. 652–660.
Costes, J., Couderc, J.P., 1988. Study by laser Doppler anemometry of the turbulent
 Eq. (1) (Grenville and Nienow, 2004) gives a good prediction of flow induced by a Rushton turbine in a stirred tank: influence of the size of the
the mixing time for aspect ratios less than one, i.e. the values units—1. Mean flow and turbulence. Chemical Engineering Science 43,
that it was correlated against. 2751–2764.
Ducci, A., Yianneskis, M., 2005. Direct determination of energy dissipation in
 Using multiple agitators with competing flow loops can cause stirred vessels with two-point LDA. AIChE Journal 51, 2133–2149.
the vessel to divide into zones which is detrimental to mixing Grenville, R.K., Nienow, A.W., 2004. Handbook of Industrial Mixing: Science and
across the whole vessel. Practice. John Wiley & Sons, Hoboken, New Jersey, pp. 507–542.
Grenville, R.K., Tilton, J.N., Simpson, T.A., Brown, D.A.R., Padron, G.A., Etchells III,
 Mixing with multiple agitators can be very complicated and
A.W., 2008. Flow and shear in agitated vessels. In: AICHE Annual Meeting,
the exact flow patterns should be known before they are used Philadelphia PA.
in a system. Hansen, P.C., 1989. Regularization, GSVD and truncated GSVD. BIT Numerical
Mathematics 29, 491–504.
Hansen, P.C., 1990. The discrete Picard condition for discrete ill-posed problems.
BIT Numerical Mathematics 30, 658–672.
Khang, S.J., Levenspiel, O., 1976. New scale-up and design methods for stirrer
Nomenclature agitated batch mixing vessels. Chemical Engineering Science 31, 569–577.
Kukuková, A., Moštěk, M., Jahoda, M., Machoň, V., 2005. CFD prediction of flow and
homogenization in a stirred vessel: part I. Vessel with one and two impellers.
c agitator clearance, m Chemical Engineering and Technology 28, 1125–1133.
D agitator diameter, m Mann, R., Knysh, P.E.A., Rasekoala, E.A., Didari, M., 1987. Mixing of a tracer in a
closed stirred vessel: a network-of-zones analysis of mixing curves acquired
H liquid height, m
by fire-optic photometry. In: Fluid Mixing III, IChemE Symposium Series,
N agitation speed, s  1 University of Bradford, UK, no. 108, pp. 49–62.
N number of readings, – Montante, G., Moštěk, M., Jahoda, M., Magelli, F., 2005. CFD simulations and
n number of frames experimental validation of homogenisation curves and mixing time in stirred
newtonian and pseudoplastic liquids. Chemical Engineering Science 60,
P power, W 2427–2437.
T vessel diameter, m Moštěk, M., Kukuková, A., Jahoda, M., Machoň, V., 2005. CFD prediction of flow and
t time, s homogenization in a stirred vessel: part II. Vessel with three and four
impellers. Chemical Engineering and Technology 28, 1134–1143.
V voltage, V Nere, N.K., Patwardhan, A.W., Joshi, J.B., 2003. Liquid-phase mixing in stirred
V volume, m3 vessels: turbulent flow regime. Industrial & Engineering Chemistry Research
w baffle width, m 42, 2661–2698.
Nienow, A.W., 1997. On impeller circulation and mixing effectiveness in the
a ratio of dished base height to tank diameter, – turbulent flow regime. Chemical Engineering Science 52, 2557–2565.
e local energy dissipation rate, m2 s  1 Otomo, N., Bujalski, W., Nienow, A.W., 1995. An application of a compartment
e mean energy dissipation rate, m2 s  1 model to a vessel stirred with either dual radial or dual axial flow impellers.
In: Procedures of the 1995 Institution of Chemical Engineers Research Event.
eR ratio of the energy dissipation rate at the wall to the
Edinburgh, UK, pp. 829–831.
mean, – Paul, E.L., Atiemo-Obeng, V.A., Kresta, S.M. (Eds.), 2004. Handbook of Industrial
G torque, N m Mixing: Science and Practice. John Wiley & Sons Inc., New Jersey, USA.
Rodgers, T.L., Kowalski, A., 2010. An electrical resistance tomography method for
r density, kg m  3
determining mixing in batch addition with a level change. Chemical Engineer-
s conductivity, S m  1 ing Research and Design 88, 204–212.
sRMS root mean squared conductivity, S m  1 Ruszkowski, S., 1994. A rational method for measuring blending performance, and
y95 95% mixing time, s comparison of different impeller types. In: Proceedings of the Eighth European
Conference on Mixing, Cambridge, UK, pp. 283–292.
Po agitator power number, ¼ P=rN 3 D5 Schöberl, J., 1997. NETGEN: an advancing front 2D/3D-mesh generator based on
Re agitator Reynolds number, ¼ rND2 =m abstract rules. Computing and Visualization in Science 1, 41–52.
Singh, V., Hessler, W., Fuchs, R., Constantinides, A., 1986. On-line determination of
SNR signal-to-noise ratio, dB mixing parameters in fermenters using pH transients. In: Procedures of the
CD Cowles disc agitator First International Conference of Bioreactors. Cambridge, UK, pp. 231–256.
2944 T.L. Rodgers et al. / Chemical Engineering Science 66 (2011) 2935–2944

Stephenson, D.R., 2008. Choices and implications in three-dimensional electrical Vrábel, P., van der Lans, R.G.J.M., Luyben, K.C.A.M., Boon, L., Nienow, A.W., 2000.
impedence tomography. Engd. Thesis, University of Manchester. Mixing in large-scale vessels stirred with multiple radial or radial and axial
Stephenson, D.R., York, T.A., Mann, R., 2007. Performance and requirements of up-pumping impellers: modelling and measurements. Chemical Engineering
process ERT instruments. In: Fifth World Congress on Industrial Process Science 55, 5881–5896.
Tomography, WCIPT5, Bergen, Norway. York, T.A., 2001. Status of electrical tomography in industrial applications. Journal
van’t Riet, K., Tramper, J. (Eds.), 1991. Basic Bioreactor Design. Marcel Dekker. of Electronic Imaging 10, 608–619.