Вы находитесь на странице: 1из 7

0263±8762/99/$10.00+0.

00
€ Institution of Chemical Engineers
Trans IChemE, Vol 77, Part A, June 1999

MIXING HYDRODYNAMICS IN A DOUBLE


PLANETARY MIXER
P. A. TANGUY, F. THIBAULT, C. DUBOIS* and A. AIÈT-KADI**
URPEI, Department of Chemical Engineering, Ecole Polytechnique, Montreal, Canada
*DREV, Val-Belair, Canada
**Department of Chemical Engineering, Laval University, Quebec, Canada

T
he mixing hydrodynamics in a double planetary mixer is investigated numerically and
experimentally over the course of a cross-linking reaction. Using various visualization
techniques, it is shown that this mixer provides good radial dispersion capabilities but
poor axial (top-to-bottom) pumping, irrespective of the viscosity level. The power drawn by
the mixer evolves dramatically from about 180 W m ± 3 at 40% conversion up to approximately
4 kW m ± 3 at 85% conversion. Overall, the numerical predictions and the experimental results
exhibit good agreement although at 85% conversion, the numerical model is not accurate
enough to predict adequately the power consumption due to physical phenomena not
considered in the computations.
Keywords: planetary; mixing; kneading; dispersion; ¯ ow simulation; viscous ¯ uid; solid
propellant; cross-linking

INTRODUCTION (Bertrand et al., 19942 , Tanguy et al., 19953 , Tanguy et al.,


The mixing of very ® ne solids at high loading rates in a 19964 ), no work has been published on double armed
viscous polymeric matrix is one major step in the produc- planetary mixers. In related ® elds, however, Jenson and
tion of solid propellant for the aerospace industries. The Talton5 , Hall and Godfrey6 , and Kappel7 highlighted the
mixing technologies capable of preparing solid propellants interest of multiple impellers for the processing of very
ef® ciently are fairly limited and costly, and fabrication viscous liquids and pastes, slurry blending and condensation
cycles are time consuming. When designing a manufactur- polymerization, and proposed some design guidelines.
ing line, two major processing dif® culties are encountered, The objective of this study is to investigate the hydro-
namely: dynamics performance of the double arm planetary mixer
in the case of solid rocket propellants, focusing on the
(a) The product is extremely viscous and its ¯ owing limit evaluation of the dispersive and distributive mechanisms
is quickly reached. generated by the rotating arms. For this purpose, ¯ ow
(b) Formulations contain some ingredients in very tiny pro® les, tracer trajectories, the distribution pattern of
quantity (few ppm) which must be thoroughly dispersed particles with time and power consumption will be
and blended to impart the right mechanical and combustion considered.
properties. The work will be carried out using three-dimensional
The make-down operation can be carried out as a batch numerical simulations and experimental work on a labora-
or in a continuous fashion in various types of mixers tory scale kneader with model ¯ uids (inert propellants).
(Fluke, 19681 ), the selection of which is usually made from
economical considerations, and from its capability of
providing time-consistent products. In practice, four EQUIPMENT DESCRIPTION
mixing systems have been widely used namely: vertical Figure 1 shows a picture of the Double Planetary Mixer. It
planetary kneaders, horizontal multiple impeller kneaders, consists of a jacketed steel vessel of 10 US gallons (ca. 38 l),
helical ribbon impellers and more recently twin-screw and two scraping arms (anchor impeller) mounted on a
extruders. rotating carousel in an off-centred fashion. As the com-
In the present work, one fairly common mixer is bination of the impellers and the carousel motion recalls
considered, the Double Planetary Mixer (Charles Ross & the movement of planets, the term `planetary mixer’ is
Son Company, Hauppauge, NY). This type of apparatus given to this technology. This kneader can be operated
which has been used for propellant mixing since the mid at atmospheric pressure or under partial vacuum with the
60s ® nds many applications in the adhesive, food and help of an ancillary vacuum pump.
cosmetics industry. The two impellers rotate at the same speed and their
The scienti® c literature on planetary kneaders is speed can be set between 0 and 60 rpm. The speed of the
relatively scarce. Apart from the work of Fluke1 and carousel varies between 8 and 35 rpm. The speed ratio
the literature cited therein, and recent contributions on the between the impellers and the carousel is ® xed by design
modelling of the mixing ¯ ows with intermeshing impellers and it is equal to 2.
318
MIXING HYDRODYNAMICS IN A DOUBLE PLANETARY MIXER 319

hydroxy-terminated polybutadiene (HTPB) are the most


common elastomer matrices. The blend is cross-linked using
a curing agent like hexamethylene diisocyanate (HMDI).
The typical cycle time is about 2 hours, but it can be extended
to about 6 hours by reducing the amount of catalyst used.
This makes this polymerization reaction particularly slow.
In the present study, the mixing experiments were not
carried out with real solid propellants for safety reasons.
Rather, propellant inerts were prepared based on partially
reacted PPG (ARCO P3025) with HMDI (Desmodur H,
Bayer). Three inerts were used in the study, corresponding
to three levels of conversion, namely 40%, 70% and 85%.
Their rheological properties were characterized, with a
rheometer RDS II from Rheometrics. It was found that
these inerts behave like Newtonian ¯ uids, a slight shear-
thinning tendency being however observed for the product
with the higher conversion. The values of the viscosity
were 2.3 Pa s at 40% conversion, 22.5 Pa s at 70% conver-
sion and 143 Pa s at 85% conversion.

EXPERIMENTAL WORK
During the reactive mixing experiments, the time
evolution of the torque was monitored against the carousel
speed. The power consumption in the mixer bulk was then
determined using the basic relation:
P 2pNT 1
where N is the carousel rotating speed (revolution per
second) and T the torque (N m). For this study, the carousel
speed was set at 10 rpm and the mixer was operated near
its maximum volumetric capacity.
In order to estimate the dispersing capability of the
kneader as well as its radial and axial pumping ef® ciency,
particle counting experiments were performed at 40%, 70%
Figure 1. Double Planetary Mixer with blade details.
and 85% conversion. In this experiment, the vessel volume
was divided in 6 compartments. Each compartment was a
wedge formed by an angular sector of 120 degrees (1/3 of
The industrial unit used for this study was modi® ed so a cross-section) and a height corresponding to half that
that it was possible to measure the torque acting on the of the vessel. The experiment consisted of counting the
carousel and monitor the exact revolution speed, making number of particles in each compartment with time. At
it possible to establish the power curve of the mixing initial time, 100 clustered particles were injected close to
system. For this purpose, the drive unit was raised by the center at the free surface. As for the particles, red pellets
approximately 0.75 m. The space thus created was used of polyphenylene oxide/polyphenylene ether blend were
to install a Himmelstein non-contact torquemeter model used. These particles are neutrally buoyant in PPG.
MCRT 3904X. The measurement capabilities of this In addition, mixing ¯ ow visualization was carried out
torquemeter range from 0 to 2000 lbf.in (0±225 Nm) with using video equipment and image analysis. This allowed
a full scale linearity of 0.1%. Although the manufacturer the observation of the cluster dispersing mechanism
provides a calibration chart for the torquemeter, it was imparted by the complex arms motion.
considered relevant to check this calibration in dynamic
mode at regular intervals during the experiments. For this NUMERICAL MODELLING
purpose, a close-clearance cylinder was carefully mounted
Consider the ¯ ow of an incompressible viscous ¯ uid
on the carousel instead of the double arms so that a
in the vessel. If t, cÇ , v and p represent the stress tensor,
Couette ¯ ow could be reproduced. Torque vs speed relations
the rate-of-strain tensor, the velocity and the pressure
were determined with viscous ¯ uids of known viscosity
respectively, the ¯ ow can be described using the classical
(based on corn syrup), which then allowed veri® cation of
Navier-Stokes equations, namely:
the calibration using elementary ¯ uid mechanics principles.
¶v
r v grad v grad p div 2g Çc 2
MODEL FLUIDS ¶t
div v 0 3
Solid rocket propellants are thermoset composite
materials consisting of a ground oxidizer dispersed into where r is the ¯ uid density, g the viscosity and Çc 1/2.
a rubbery matrix. Polypropylene glycol (PGG) and [grad v (grad v)T ].

Trans IChemE, Vol 77, Part A, June 1999


320 TANGUY et al.

The ¯ ow modelling in the mixer is considered as a ¯ ow


problem in an enclosure with moving internal boundaries,
where the location of the boundaries is known with time.
Such a problem constitutes a formidable challenge with
classical numerical methods for partial differential equa-
tions (® nite difference, ® nite volume or ® nite element
methods). The application of an alternative frame of
reference sometimes makes possible the solution of a
steady state problem with a single mesh. For example, the
case of a rotating impeller mounted in a central fashion
in the tank. The frame of reference can be attached to the
impeller (Lagrangian viewpoint), simplifying drastically
the simulation tasks. In the case considered here, no
simplifying frame of reference could be used, and one
would be practically compelled to generate a numerical grid
at each time step. Figure 2. Finite element mesh (surface mesh only).
To overcome this problem, a new ® nite element approach
has recently been introduced called the virtual ® nite element
method (Bertrand et al., 19978 ). With this method, the
moving parts in the domain are taken into account with error. In this work, a variable time-step fourth order Runge-
constrained optimization techniques. The moving bound- Kutta method was used. The analysis of the ¯ ow paths
aries are discretized with a set of control points, on which allowed the numerical determination of the dispersive
kinematic constraints (the impeller velocity in the present mechanism induced during the complex planetary motion
case) are imposed and integrated into the governing set of of the kneading arms.
equations using Lagrange multipliers and a penalty method From a practical standpoint, the simulation work was
(Uzawa algorithm). carried out with POLY3DTm software (Rheotek, Montreal),
The practical implementation of this method in a standard whose latest release includes an implementation of the
Galerkine ® nite element code is simple, as the treatment above virtual ® nite element technology.
of the additional constraints is made explicitly. Details In Figure 2 the ® nite element mesh of the mixer is shown.
pertaining to this approach can be found in Bertrand9 . This mesh was generated using I-DEAS Tm Master Series 2.1
Flow numerical simulation yields directly the time (SDRC, Milton). Only the surface mesh has been made
evolution of the velocity and pressure ® elds in the vessel. visible in order to avoid cluttering the ® gure. The numerical
Velocity pro® les at various vertical and horizontal cross- grid contains 10,747 tetrahedral elements P1 P0 . The
sections allow the appraisal of the radial and axial pumping impellers have been depicted with 2247 velocity point-
capabilities of the kneading arms and the determination wise constraints (virtual ® nite element method). The mesh
of the overall circulation ef® ciency in the vessel for a set of was suf® ciently re® ned so that no signi® cant mesh effect
operating conditions and ¯ uid properties. was noticed on the macromixing pattern.
From the knowledge of the ¯ ow ® eld in the kneader, it
is also possible to generate very useful information from
a design perspective like the power constant (or better the
kneading power curve) and the map of dissipated energy.
Below are basic de® nitions of these parameters.
P
Power number Np 4
rN 3 D 5
rND 2
Reynolds number Re 5
g
Power constant Kp Np Re 6
where N and D are the characteristic rotating speed and
diameter respectively. The power drawn by the impellers
in the medium, P, can be obtained either from a macro-
scopic energy balance, or using equation (1).
In addition, ¯ ow paths (neutrally buoyant particle trajec-
tories) can be determined by integration of the ¯ ow with
time:
t Dt
t Dt t
X X vdt 7
t

Care must be taken in the computation of the integral due Figure 3. Radial dispersion. (a) Initial position of clustered particles;
to the exponential nature of the numerical integration (b) 45% conversion; (c) 70% conversion; (d) 85% conversion.

Trans IChemE, Vol 77, Part A, June 1999


MIXING HYDRODYNAMICS IN A DOUBLE PLANETARY MIXER 321

Table 1. Variation of mixer power with conversion.

Conversion Power measured Power computed

40% 3,6 W 3,6 W


70% 24 W 32 W
85% 85 W 201 W

In this work, the mixing ¯ ow was simulated at the three


levels of conversion considered in the experimental work.
The ¯ uid surface was considered as ¯ at.

RESULTS AND DISCUSSION


Figure 3 shows the radial pumping mechanism obtained
by tracking the trajectories of tracers (neutrally buoyant
particles) over 5 seconds. The tracers were initially
clustered near the right blade. Trajectories are seen from
the top for the three levels of conversion (counter-clockwise
direction). It can be noticed that the radial dispersion is
adequate and only weakly affected by the viscosity level.
Figure 4. Axial dispersion near the free surface. (a) Initial position of A similar result is shown in Figures 4 and 5 for the
clustered particles; (b) 45% conversion; (c) 70% conversion; (d) 85%
conversion. axial pumping. The trajectories of 50 tracers were tracked
during 60 seconds. In Figure 4, the tracers were initially
clustered near the surface, while in Figure 5, they were
A main dif® culty of the present work is the post- located at the bottom. Results are represented as a
processing of the ¯ ow results. Indeed, as the geometry and perspective view. It can be seen that the axial dispersion
the velocity ® eld change with time, all the nodal variables is not very effective at the surface although it becomes
at every time step must be stored. The computation of fairly good at the bottom due to the action of the horizontal
trajectories must be carefully carried out using the velocity
® eld at the correct time value, using, if needed, time
interpolation. In order to speed up the whole post-
processing, it was decided to express the time evolution
of each velocity component by a time Fourier series and
store only the coef® cients. These coef® cients were obtained
by using a fast Fourier transform (FFT) algorithm.

Figure 5. Axial dispersion at the bottom. (a) Initial position of clustered Figure 6. Numerical (top) and experimental (bottom)Ð stretching-folding
particles; (b) 45% conversion; (c) 70% conversion; (d) 85% conversion. mechanism.

Trans IChemE, Vol 77, Part A, June 1999


322 TANGUY et al.

Figure 7. Particle counting experiment (40% conversion). Experimental Figure 9. Particle counting experiment (85% conversion). Experimental
results (top); numerical results (bottom). results (top); numerical results (bottom).

arm. These observations are valid irrespective of the for an anchor. Following Tanguy et al.4 , the power constant
conversion level. It is important to note that, in the case of of an anchor impeller is around 200. The present results
the 85% conversion, the assumption of a ¯ at free surface are then fairly consistent, the power consumption being
is not very realistic. The chaotic nature of the surface between 2.5 and 4 times higher than that of a single
contributes more signi® cantly to the axial dispersion impeller.
mechanism. According to Ottino1 0 , the laminar mixing mechanism
In order to assess the simulation accuracy in this problem obeys a `stretching-folding-breaking’ sequence. Figure 6
and at the same time determine the power dissipated in illustrates the folding phase of a stretched ® lament of tracers
the vessel, the power consumption has been computed by located at one arm tip close to the wall. This `Vee’ shape
performing a macroscopic energy balance and the results can be observed numerically and experimentally in a
compared with the experimental data. Results are summar- remarkably similar fashion. This proves, if necessary, the
ized in Table 1. reliability of the computation of the mixing pattern in
Considering the experimental uncertainty and the the kneader that can pick up details as ® ne as this
numerical precision both estimated at 20%, the agreement orientation phenomenon which occurs during the break-up
between the predictions and the experiments is fairly good of agglomerates.
up to a conversion level of 70%. Beyond this value, the Figures 7 to 9 present the results of particle counting
signi® cant difference observed is explained by the inaccu- experiments. The particles were all injected at the surface.
racy of the computer model. Three reasons may be invoked: Overall, the numerical and experimental sets of results
(a) the shear-thinning character of the cross-linked polymer give fairly similar trends. In terms of top-to-bottom
matrix not taken into account in the model; pumping, it can be seen that a signi® cant time is needed
(b) the chaotic nature of the free surface whose numerical for the ® rst particles to cross the upper half of the volume
description is presently out-of-reach; and vessel. The radial dispersion appears ef® cient and rapid.
(c) the viscous dissipation that would make the ¯ uid less After 60 s, there is already a good amount of particles
viscous. Overall, the power per unit volume varies from in each compartment (in the upper part of the vessel).
180 W m ± 3 to 4 kW m ± 3 over the course of the reaction, Finally, Figures 10 to 12 show the instantaneous density
which is fairly typical with polymerization reactions. of energy dissipation (in W m ± 3 ) at three time steps,
corresponding to three different positions of the kneading
From the above results, it is possible to estimate the blades. Three different views are given in each ® gure. It
value of the power constant Kp using equation (6). As for can be observed that most of the energy is dissipated in
the characteristic parameters, two options can be envisaged: the clearance areas between the arms, and the arms and
the carousel speed along with the vessel diameter, or the walls. These regions correspond to the zones with the
alternatively, the arm rotating speed and the diameter. maximum rate of deformation and therefore the most
Based on the former set of characteristic dimensions and ef® cient mixing. In the bulk itself, there is almost no energy
the power consumption results, the Kp value ranges from dissipation.
500 to 830. As the kneading arm of the Double Planetary
Mixer is fairly similar to an anchor impeller, it is interesting
to compare this range of Kp values with the typical values CONCLUDING REMARKS
The mixing hydrodynamics in a double planetary mixer
has been investigated over the course of a cross-linking
reaction. At 10 rpm (speed of the carousel), it is shown
numerically and experimentally that this mixer provides
good radial dispersion capabilities but poor axial (top-to-
bottom) pumping, irrespective of the viscosity level. The
power drawn by the mixer evolves dramatically from
about 180 W m 3 at 40% conversion up to approximately
4 kW m ± 3 at 85% conversion.
The numerical predictions and the experimental results
exhibit good agreement although at 85% conversion, the
Figure 8. Particle counting experiment (70% conversion). Experimental numerical model is not accurate enough to predict ade-
results (top); numerical results (bottom). quately the power consumption. It is believed that this is

Trans IChemE, Vol 77, Part A, June 1999


MIXING HYDRODYNAMICS IN A DOUBLE PLANETARY MIXER 323

Figure 10. Instantaneous density of energy dissipation after 3rd time step in three different planes. Scale: low (light grey); high (dark grey).

Figure 11. Instantaneous density of energy dissipation after 9th time step in three different planes. Scale: low (lightgrey); high (dark grey).

Trans IChemE, Vol 77, Part A, June 1999


324 TANGUY et al.

Figure 12. Instantaneous density of energy dissipation after 13th time step in three different planes. Scale: low (light grey); high (dark grey).

due to the shear-thinning character of the ¯ uid that has not 5. Jenson, W. P. and Talton, R. T., 1965, AIChE Symp Series, 10: 82.
been considered in the computations and possibly to the 6. Hall, K. R. and Godfrey, J. C., 1968, Trans IChemE, 46: T205.
7. Kappel, M., 1979, Int Chem Eng, 19: 571.
very irregular shape of the free surface in the vessel that was 8. Bertrand, F., Tanguy, P. A. and Thibault, F., 1997, Int J Num Meth
ignored in the model. Fluids, 25: 719.
This mixer is an interesting device for tough mixing 9. Bertrand, F., 1995, TheÁse de doctorat, (Institut National Polytechnique
applications and has a good potential for improvements, de Lorraine, France).
10. Ottino, J., 1989, The Kinematics of Mixing, (Cambridge University
especially the shape of the mixing arms. Press).

REFERENCES
1. Fluke, M., 1968, ACS Chemistry Series No. 88: 176. ADDRESS
2. Bertrand, F., Thibault, F., Tanguy, P. A. and Choplin, L., 1994, Process Correspondence concerning this paper should be addressed to Professor
MixingÐ Chemical and Biochemical Applications, G. Tatterson and P. A. Tanguy, URPEI, Department of Chemical Engineering, Ecole
R. Calabrese (Eds) AIChE Symp Series No. 299: 106. Polytechnique, PO Box 6079, Station Centre-Ville, Montreal H3C 3A7,
3. Tanguy, P. A., Bertrand, F., Labrie, R., Brito-De La Fuente, E. and Canada. E-mail: tanguy@urpei.polymtl.ca
Fages, M. H., 1995, Recents Progres en Genie des Procedes, (Lavoisier
TecDoc) 38, 87.
4. Tanguy, P. A., Bertrand, F., Labrie, R. and Brito-De La Fuente, E., The manuscript was received 16 November 1998 and accepted for
1996, Chem Eng Res Des, 74: 499. publication 18 January 1999.

Trans IChemE, Vol 77, Part A, June 1999

Вам также может понравиться