1 s2.0 S000925090500432X Main
1 s2.0 S000925090500432X Main
1 s2.0 S000925090500432X Main
www.elsevier.com/locate/ces
Abstract
This paper investigates the way modelling mixing phenomena occur in unsteady stirring conditions in agitated vessels. In particular, a
new model of torus reactor including a well-mixed zone and a transport zone is proposed. The originality of the arrangement of ideal
reactors developed here lies in the time-dependent location of the boundaries between the two zones. This concept is applied to model
the positive inuence of unsteady stirring conditions on homogenization process: the model avoids a mass balance discontinuity when the
transition from steady to unsteady stirring conditions is performed.
To ascertain the reliability of the model proposed, experimental runs with highly viscous uids have been carried out in an agitated
tank. The impeller used was a non-standard helical ribbon impeller, tted with an anchor at the bottom. The degree of homogeneity in
the tank was observed using a conductivity method after a tracer injection.
It is shown that for a given agitated uid and mixing system, model parameters are easy to estimate and that modelling results are in
close agreement with experimental ones. Moreover, it would appear that this model allows the easy derivation of a control law, which is
a great advantage when optimizing the dynamics of a mixing process.
2005 Elsevier Ltd. All rights reserved.
Keywords: Mixing; Modelling; Nonlinear dynamics; Parameter identication; Unsteady stirring; Torus model
1. Introduction
mixing:
The second way is to consider that, for a given mixing system, the ability of a ow to homogenize viscous products
can be signicantly enhanced with the help of unsteady
time-varying stirring approaches. Efcient mixing in laminar regime has been shown to be related to the amount
of stretching and folding generated within the tank by the
agitator (Ottino, 1989; De La Villeon et al., 1998; Alvarez-Hernndez et al., 2002). When stirring conditions
are steady, initially designated uid material will follow
closed streamlines in the vessel and consequently the mixing efciency will be rather poor since such regular ows
will induce a linear evolution of intermaterial area with
time (Niederkorn and Ottino, 1994). However, when a
suitable perturbation is superimposed on the steady velocity eld, ows reorientations will appear, uid elements
will be no longer trapped by closed steady streamlines
and will become free to wander throughout chaotic ow
domains. As the stretching rate is higher in these ow regions, the inter-material area will grow faster (Niederkorn
and Ottino, 1994; Alvarez-Hernndez et al., 2002) and
higher than average values of the efciency will be obtained.
However, there is a lack of systematic studies that provide
us with quantitative information about the conditions under
which these chaotic ows are produced within a stirred tank
and their actual benets on mixing efciency. Consequently,
the design of a sequence of ows which involves a reorientation of material elements (for instance, when periodic or
co-reverse rotation of the impeller is performed) has yet to
be clearly identied.
Moreover, most unsteady stirring approaches used to improve laminar mixing in batch reactors (Nomura et al., 1997;
Lamberto et al., 1996; Yao et al., 1998) deal with smalldiameter agitators which are usually devoted to work in turbulent regime and not suited for the batch mixing of viscous uids; their purpose being to prevent the formation of
isolated mixed regions (Metzner and Taylor, 1960) with coreverse or periodic rotational speed sequences. Such a work
has not been carried out for systems equipped with efcient
closed-clearance impellers.
1.2. Flow modelling in batch reactors
From this survey, it would appear clearly that there is
a strong need for rational studies which quantify the efciency of a stretching process for a given mixing system
under unsteady operating conditions. Numerical studies using computational uid dynamics (CFD) methods allow the
determination of the whole velocity eld for laminar mixing within the tank at steady and unsteady rotational speeds
and thus point out the well-mixed and stagnant zones (e.g.
Zalc et al., 2002; Arratia et al., 2004; Harvey and Rogers,
1996; Campolo et al., 2003). However, these nite element
methods require a long computation time (e.g. for a vessel
with close clearance impeller, see de la Villeon et al., 1998
5545
5546
Y (t)
to describe the response curve after a tracer injection, whatever the stirring approach adopted. Indeed, for steady approaches the network of ideal mixers used to simulate the
mixing process becomes similar to that of those used by
Khang and Levenspiel (1976) whose reliability have been
previously shown. Moreover, in the case of unsteady stirring, the model is also supposed to account for the experimental observation that an improvement in mixing occurs
when a positive variation in the rotational speed is enforced.
For example, in the case of a positive variation in impeller
rotational speed, the volume of the stirred tank reactor increases while that of the plug ow decreases. As the whole
volume of the torus loop is supposed to be unchanged, an
enhancement in mixing is expected.
Note that the model structure should not be confused with
real toroidal reactors (e.g. Benkhelifa et al., 2000).
Let us dene Vd+ (resp., Vd ) as the variation of volume Vd
due to the motion of S1 (resp., S2 ) in the torus, and let be
the residence time of the particle leaving the plug ow zone
at time t. Using notations previously introduced, the whole
system can be characterized by the following differential
equations (see Appendix A):
V = Vd (t) + Vp (t),
t
) d = V Vd (Q(t
))
Q(
t
Vp
Q (t)
Yd
S1
Y (t)
t
d[y(t)]
= (Q(t)
+ Vd+ )[y(t ) y(t)],
Vd (Q(t))
dt
Q(t)
= N (t).
(1)
2.1. Theorem
The mass balance in the species y(t) within the torus
reactor dened by Eq. (1) is respected (see proof in
Appendix B).
In our study, it is assumed that in the case of steady mixing
(constant rotational speed), the volume of the well-mixed
zone does not depend on the amplitude of the rotational
speed and has a constant value Vd1 . Note that integrating
Eq. (1), we obtain
t
t
+
Vd (t) =
Vd dt
Vd dt + Vd1 ,
(2)
0
S2
5547
will allow Vd+ and Vd to be obtained at each time. Consequently, Vd and Vp can also be computed.
The assumptions used in this paper concerning evolutions
of Vd+ and Vd with stirring conditions are the following,
which requires a third constant parameter k for the model:
dN
dN
dN
Vd+ = k
if
> 0, Vd+ = 0 if
< 0,
dt
dt
dt
dN
dN
dN
Vd = k
if
< 0, Vd = 0 if
> 0.
dt
dt
dt
(3)
Fig. 2. Picture and geometrical parameter of the mixing equipment investigated (other geometrical parameters of PARAVISC mixing system:
blade width, w = 0.032 m; impeller pitch, p = 0.560 m; impeller height,
L = 0.340 m; tank height, Hc = 0.443 m).
after a tracer injection. The signal was amplied by a converter (Type AT40, SFERE), and recorded with the help of
an I/O board (PCL-812 PG, ADVANTECH) plugged into a
PC. The sampling rate was 200 Hz.
The tracer pulse injected had the same physical properties as the uid in the tank (composition and temperature),
with an additional quantity of NaCl at a concentration of
100 g/l. The incorporation was performed with the help of a
pneumatic system with pistons (type DACO, PCM DOSYS)
equipped with a duct (DACC 48/40, DOSYS) which holds
the product at the end of the pipe. This device was able to
inject 72 ml (0.24% of the tank volume) of viscous tracer
into the tank with an accuracy of 2%. The injection duration is by a fraction of a second. It was checked, measuring
a sample of the injected uid before and after each injection, that the inuence of the addition of salt on density and
viscosity was negligible for a limited (40) number of successive trials. The volume of the tank was brought back to
30 103 m3 after each experiment. The conductivity probe
and the injection locations were kept unchanged throughout
the experiments (Fig. 2).
The I/O board allows the operating conditions to be accurately controlled, i.e., the injection time, the departure and
the magnitude of speed variations that were enforced on
the agitation system. The rotational speed and the conductivity signal were recorded throughout the mixing process.
Recording was activated 3 s before the tracer injection. Each
experiment (for one set of experimental conditions) was repeated four times to ensure repeatability.
The values of the rotational speed varied from 0.16 to
1.5 rev/s. Mixing and circulation times were determined
from the response signal recorded after tracer injection. The
mixing time is dened as the duration needed for the signal
to reach 95% of its nal value (Fig. 3). The circulation time
is dened as the signal period, when mixing at constant
impeller rotational speed (Fig. 3). When the conductivity
5548
tc
Signal (V)
+5%
-5%
tm
0
20
40
60
80
100
Time (s)
Fig. 3. A typical probe response curve.
and on the viscous media (which are maintained at a constant level in this study).
Parameters Vd1 and have been estimated from one
tracer experiment when mixing at constant impeller speed
(0.667 rev/strial number 1 in Table 1). Using the values of the parameters Vd1 and previously estimated, an
additional injection was performed with unsteady stirring
conditions (a speed pulsetrial number ve in Table 1) to
obtain the value of parameter k.
The set of model parameters were estimated using an
optimization algorithm (simplex method). The optimization
algorithm is based on the minimization of the mean absolute
error criterion dened in Eq. (4).
MAE =
M1
1
|(i.Te )|.
M
(4)
i=0
4. Results
4.1. Efciency of mixing using unsteady stirring conditions
The positive inuence of unsteady stirring condition on
mixing efciency is recalled in Table 2. It can be observed
that the mixing work required for unsteady stirring is less
signicant than those calculated for those mixing procedures
which would give identical mixing times at constant RPM.
These values of energy consumed and their determinations
have already been discussed (Dieulot et al., 2002) and are
not the key consideration of this work. Note simply that,
as presented in previous works, depending upon the type
of unsteady stirring conditions adopted, the energy savings
vary from 30% to 60% and justify the interest of introducing
time-dependent perturbations for a homogenization process.
5549
Table 1
Operating conditions (impeller rotational speed uctuations) adopted during the mixing process after tracer injection
Trial number
N1 (rev/s)
N2 (rev/s)
Time
parameters (s)
Steady
Stirring
Steady
stirring
0.667
0.833
Ramp
Speed
Ramp
Speed
0.667
1.333
RD
5
0.667
1.333
15
Pulse
Speed
Pulse
Speed
Pulse
Speed
0.667
1.333
PS
17
0.667
1.333
0.667
1.333
Step
speed
0.667
1.333
3
4
5
6
7
10
4
5550
Table 2
Experimental mixing performances of the helical mixing system studied using various stirring conditions (starting impeller rotational speed = 0.667 rev/s
except trial 2)
Trial 1
Steady
stirring
Trial 2
Steady
stirring
Trial 3
Ramp
(RD = 5 s)
Trial 4
Ramp
(RD = 15 s)
Trial 5
Pulse
(PS = 17 s)
Trial 6
Pulse
(PS = 10 s)
Trial 7
Pulse
(PS = 4 s)
Trial 8
Step
Experimental
mixing time (s)
80.7
72
33
38.5
60.7
58.5
65.1
53.3
Experimental
mixing work (J)
627.9
937.8
793.6
763.6
510.9
525.6
385
579.7
Values of mixing
work (J) for the
mixing process
which would give
same mixing time
at constant impeller
rotational speed12
627.9
937.8
1535.5
1214.3
740.4
814.6
1035.7
980.9
Energy savings
48.3
37.1
31.0
35.5
62.8
40.9
Table 3
Values of MAE and predicted values of mixing times obtained by the model for the helical mixing system studied using various stirring conditions
Operating conditions used for parameter identication
Trial 1
Trial 5
Steady stirring Pulse
N = 0.667 rev/s (PS = 17 s)
Trial 2
Trial 3
Trial 4
Trial 6
Trial 7
Trial 8
Steady stirring Ramp
Ramp
Pulse
Pulse
Step
N = 0.833 rev/s (RD = 5 s) (RD = 15 s) (PS = 10 s) (PS = 4 s)
Experimental
80.7
mixing time (s)
Predicted values
81.5
of mixing time (s)
Values of
0.24
criterion MAE (V)
60.7
72
33
38.5
58.5
65.1
53.3
66.7
65.6
31.5
37.4
66.7
66.5
47.5
0.26
0.27
0.24
0.28
0.20
0.18
0.24
Fig. 4. Predicted (-) and experimental (.) circulation curves for steady
speed at 40 rpm.
Fig. 5. Predicted (-) and experimental (.) circulation curves for speed
pulse from 40 to 80 rpm, starting at 17 s, duration 5 s (trial 5 in Table 2).
Fig. 6. Predicted (-) and experimental (.) circulation curves for speed
ramp from 40 to 80 rpm, starting at 17 s, ramp duration RD = 5 s (trial 4
in Table 2).
5551
Fig. 9. Predicted (-) and experimental (.) circulation curves for speed
pulse from 40 to 80 rpm, starting at 4 s, duration 5 s. PS = 4 s (trial 7 in
Table 2).
ds = (Q(t)
+ Vd+ ) dt,
Fig. 7. Predicted (-) and experimental (.) circulation curves for speed
ramp from 40 to 80 rpm, starting at 17 s, ramp duration RD = 15 s. (trial
3 in Table 2).
(5)
and
Vd (U (s))
= Y (s ) Y (s),
Fig. 8. Predicted (-) and experimental (.) circulation curves for speed
pulse from 40 to 80 rpm, starting at 10 s, duration 5 s. PS = 10 s (trial 6
in Table 2).
dY
(s)
ds
(6)
5552
plug ow zone will move counter-clockwise and will overlap an area which was previously in the well-mixed zone.
The effect of u < 0 is thus more limited than in the case
where u > 0.
Finally, dening by W =Vd (U ) as the control parameter
and using volume balance in the torus, it can be written
dY
(s) = Y (s ) Y (s)
ds
+ Vd (U (s )) = V .
W
and
(7)
By construction W [0, V ] and a positive solution for always exists when W is a continuous function of s. When the
equation has several roots, should be chosen as the smallest. Consequently, using the torus model, an optimal solution for the control should be quite simple to obtain using
algebraic methods. Preliminary results have been obtained
(Dieulot and Richard, 2001), which will be extended in future work.
5. Conclusion
A torus model has been developed to describe a mixing
process at unsteady rotational speeds. The combination of
ideal reactors proposed includes a well-mixed and a plug
ow zone contained in a torus volume. The boundaries between the two zones vary with the ow rate (proportional to
impeller rotational speed) and are supposed to represent the
enhancement of mixing efciency, experimentally observed
when using unsteady stirring conditions. Only the knowledge of three constant parameters Vd1 , , k is required for
the model proposed. Moreover, only two trials are necessary to estimate the three xed parameters (one at constant
impeller speed Vd1 , , and one at unsteady rotational speed
k). Finally, the model proposed gives a close agreement between predicted and experimental circulation curves and allows us to estimate the mixing times, for any kind of timedependent rotational impeller speed tested.
Of course, the model proposed fails to demonstrate that
the use of dynamic ow perturbations (time-dependent revolution per minute) contributes to generate a more global
chaotic ow which reduces segregated regions and enhances
mixing as Tanguy et al. (1998) and Lamberto et al. (2001)
have done with CFD applications. The model proposed does
not allow to obtain the time-dependent map of the segregated regions. However, our model is quite complementary
to CFD applications and very useful since according to us,
so far, there was no way to predict by an arrangement of
ideal reactors the enhancement of mixing when using timedependent stirring conditions. In this sense the model proposed succeeds in quantifying quickly the gain in mixing
time and energy provided by applying time-dependent RPM.
Moreover, this study has been conducted with a helical
ribbon impeller but the approach proposed is not limited to
this kind of agitators and can be extended to other mixing
systems. It would be even possible to propose a new classi-
cation of mixing systems based on their homogenization performances during unsteady stirring and would at last allow
to propose new mixers that have an appropriate behaviour
when mixing under such operating conditions.
Finally, the mathematical equations of the system are indeed easily tractable which allows to dene an optimal control strategy for the torus model. This will be tackled in a
future work. The optimal control would be a compromise between the additional energy required to damp down quickly
the degree of homogeneity and additional energy required
to create dynamic ow perturbations (unsteady rotational
speed).
Notation
D
Hc
HL
k,
L
N
p
Q
S1 , S2
t
tm
T
Te
V
Vd
Vp
w
Wm
y(t)
impeller diameter, m
tank height, m
liquid height, m
model parameters (see units in text)
impeller height, m
impeller rotational speed, rev/s
helical ribbon pitch, m
uid ow rate, m3 /s
moving boundaries for the torus volume, m2
time, s
mixing time, s
tank diameter, m
sampling period used for estimation, s
vessel or torus reactor volume, m3
volume of the well-mixed zone for the torus
volume, m3
volume of the plug ow zone for the torus
volume, m3
blade width, m
mixing work, J
tracer concentration, kg/m3
Greek letters
proportionality constant, m3
time-varying delay, s
viscosity of Newtonian uid, Pa s
uid density, kg/m
d[Vd (Q(t))]
= Vd+ Vd .
dt
(8)
5553
d[Vd (Q(t))]
dy(t)
d[Vd (Q(t))y(t)]
= Vd (Q(t))
+ y(t)
,
dt
dt
dt
(10)
(z, t))
Q(t)
+ Vd (t) = (Q(t
(j(t (z, t), z))
+ Vd (t (z, t))) 1
,
jt
j(z, t)
(Q(t (z, t)) + Vd (t (z, t))) = S.
jt
d[Vd (Q(t))y(t)]
= (Q(t)
+ Vd+ (t))y(t )
dt
(Q(t)
+ Vd (t))y(t),
Vd (Q(t))
d[y(t)]
= (Q(t)
+ Vd+ )[y(t ) y(t)].
dt
(9)
(11)
t
t(z,t)
t(z,t)
First we calculate
d
y(t (z, t))Sdz
dt plug f low
Vp /S
y(t
(z, t))
= Vp (t)y(t ) +
0
(j(t (z, t), z))
1
S dz,
jt
which becomes, using previous equations,
d
y(t (z, t))S dz
dt plug f low
Vp /S
= Vp (t)y(t ) +
y(t
(z, t))
0
d
dt
t
t
Vp () d
t
=
Vd () d
t
t
=
(Vd Vd+ ) d,
t
t
Q(t)
+ Vd (t)
S dz
(z, t)) + V (t (z, t)))
(Q(t
d
plug f low
= Vp (t)y(t ) + (Q(t)
+ Vd (t))
Vp /S
j(z, t)
y(t
(z, t))
S dz,
jz
0
and, integrating the last equation
d
y(t (z, t))S dz
dt
plug f low
= Vp (t)y(t ) + (Q(t)
5554
+ (Q(t)
+ Vd (t))(y(t) y(t (t))),
= (V + V + Vp )y(t ) = 0,
A(t)
d
d
which completes the proof.
References
Alvarez-Hernndez, M.M., Shinbrot, T., Zalc, J., Muzzio, F.J., 2002.
Practical chaotic mixing. Chemical Engineering Science 57, 3749
3753.
Arratia, P.E., Lacombe, J.P., Shinbrot, T., Muzzio, F.J., 2004. Segregated
regions in continuous laminar stirred tank reactors. Chemical
Engineering Science 59, 14811490.
Benkhelifa, H., Legrand, J., Legentilhomme, P., Montillet, A., 2000. Study
of the hydrodynamic behaviour of the batch and continuous torus
reactor in laminar and turbulent ow regimes by means of tracer
methods. Chemical Engineering Science 55, 18711882.
Campolo, M., Sbrizzai, F., Soldati, A., 2003. Time-dependent ow
structures and Lagrangian mixing in Rushton-impeller bafed-tank
reactor. Chemical Engineering Science 58, 16151629.
Delaplace, G., Leuliet, J.-C., Relandeau, V., 2000a. Circulation and
mixing times for helical ribbon impellers. Review and experiments.
Experiments in Fluids 28 (2), 170182.
Delaplace, G., Torrez, C., Andr, C., Belaubre, N., Loisel, P., 2000b.
Numerical simulation of ow of Newtonian uids in an agitated vessel
equipped with a non standard helical ribbon impeller, Proceedings
of 10th European Conference on Mixing, (Delft 25 july), (Elsevier
Amsterdam): 289296.
De La Villeon, J., Bertrand, F., Tanguy, P.A., Labrie, R., Bousquet, J.,
Lebouvier, D., 1998. Numerical investigation of mixing efciency of
helical ribbons. A.I.Ch.E. Journal 44 (4), 972977.
Dieulot, J.-Y., Richard, J.-P., Tracking control of a nonlinear system with
input-dependent delay. IEEE Int. Conf. Decision and Control. CDC
01, Orlando, 2001.
Dieulot, J.-Y., Delaplace, G., Gurin, R., Brienne, J-P., Leuliet, J.-C.,
2002. Laminar mixing performances of a stirred tank equipped with
helical ribbon agitator subjected to steady and unsteady rotational
speed. Transactions of Institution of Chemical Engineers 80 (Part A),
335344.
Harvey, A.D., Rogers, S.E., 1996. Steady and unsteady computation of
impeller stirred tank reactors. A.I.Ch.E. Journal 42, 27012712.
Khang, S.J., Levenspiel, O., 1976. New scale-up and design method for
stirrer agitated batch mixing vessels. Chemical Engineering Science
31, 569577.
Lamberto, D.J., Muzzio, F.J., Swanson, P.D., Tonkovich, A.L., 1996. Using
time-dependent RPM to enhance mixing in stirred vessels. Chemical
Engineering Science 51 (5), 733741.
Lamberto, D.J., Alvarez, M.M., Muzzio, F.J., 2001. Computational analysis
of regular and chaotic mixing in a stirred tank reactor. Chemical
Engineering Science 56 (16), 48874899.
Metzner, A.B., Taylor, J.S., 1960. Flow patterns in agitated vessels.
A.I.Ch.E. Journal 6 (1), 109114.
Niederkorn, T.C., Ottino, J.M., 1994. Chaotic mixing of shear-thinning
uids. A.I.Ch.E. Journal 40 (11), 17821793.
Nomura, T., Uchida, T., Takahashi, K., 1997. Enhancement of mixing
by unsteady agitation of an impeller in an agitated vessel. Journal of
Chemical Engineering of Japan 30 (5), 875879.
Ottino, J.M., 1989. The Kinematics of Mixing. Stretching, Chaos and
Transport. Cambridge University Press, Cambridge.
Tanguy, P.A., Thibault, F., Brito-De La Fuente, Espinosa Solares, T.,
Tecante, A., 1998. Mixing performance induced by coaxial at
blade-helical ribbon impellers rotating at different speeds. Chemical
Engineering Science 52 (11), 17331741.
Tatterson, G.B., 1994. Scale Up and Design of Industrial Mixing Processes.
McGraw-Hill, New York.
Yao, W.G., Sato, H., Takahashi, K., Koyama, K., 1998. Mixing
performance experiments in impeller stirred tanks subjected to unsteady
rotational speeds. Chemical Engineering Science 53 (17), 30313043.
Zalc, J.M., Szalai, E.S., Alvarez, M.M., Muzzio, F.J., 2002. Using CFD
To Understand Chaotic Mixing in Laminar Stirred Tanks. A.I.Ch.E.
Journal 48, 21242134.
Zenger, K., Ylinen, R., 1994. Simulation of variable delays in material
transport models. Mathematics and Computers in Simulation 37, 57
72.