Chemical Engineering Science, Vol. 51, No. 10, pp.

1725-1734, 1996

Pergamon

S0009-2509(96)00031-0

Copyright 1996 Elsevier Science Ltd Printed in Great Britain. All rights reserved 0009-2509/96 $15.00 + 0.00

3-D HYDRODYNAMICS BY A DOUBLE-PROPELLER HAVING EVOLVING

IN A TANK STIRRED PROPERTIES

SYSTEM AND FILLED WITH A LIQUID RHEOLOGICAL

C. XUEREB and J. BERTRAND Laboratoire de G6nie Chimique URA CNRS 192, INP-ENSIGC - 18, chemin de la Loge 31078 Toulouse Cedex, France

A b s t r a c t - Some chemical reactions in the liquid phase can generate drastic changes of the rheological behavior of the liquid. This often happens in agitated tanks entirely baffled equipped with impellers which generate circulations which depend on the hydrodynamic regime - laminar or turbulent. Results issued from Computational Fluid Dynamics (CFD) are presented here concerning fields of velocity components and local energy dissipation, corresponding to a liquid initially with a low viscosity becoming highly viscous Newtonian and then pseudo-plastic. It clearly appears that the flow pauerns generated by the two propellers settled in the vessel are quite different, especially with large recirculations and dead zones in the laminar flow range. Depending on the rheological properties of the fluid, energy dissipation occurs in the vicinity of the propellers and/or near the walls.

INTRODUCTION Mixing operation is largely encountered at all process levels in chemical, food, pharmaceutical, nuclear industries, etc... Sometimes, problems in chemical reaction engineering are largely dependent of hydrodynamics phenomena. For instance, in the case of fast chemical reactions involving micro-mixing aspects, root-mean squared velocities (RMS) and energy dissipation are fundamental parameters for the results of the reaction (chemical orientation in the case of bifurcation, efficiency). On the other hand, in the case of slow chemical reactions, mean velocities are the most interesting hydrodynamic parameter because it is sufficient to ensure homogeneity in the reactor and to prevent dead zones formation. The aim of this paper is to present hydrodynamic results in the case of a reactor filled with a liquid having different rheological behaviors in the time: for instance, this case can represent various polymerisation operations. At the beginning of the reaction, the liquid has a low viscosity and behaves as water solution: the flow regime can be considered as turbulent Then, the viscosity is increasing due to the increase of molecular weight, and a second characteristical step is the case of a highly viscous Newtonian fluid: the flow regime can be considered as laminar. Then, the fluid begins to exhibit a pseudo-plastic behavior. The rheological properties can be represented by the socalled power-law model and the two following parameters: the consistency index (m) and the flow behavior index (n). As the polymerisation occurs, the consistency increases and the behavior index decreases from 1.0 to values of about 0.2. Designing stirred reactors requires a good knowledge of the liquid physical properties. Consequently, a given plant will be efficient in terms of homogenization only in a given operating range. The main condition for a correct operation consists in remaining in the hydrodynamic regime previously predicted when choosing the type of impellers, their dimensions and the way to set them in the reactor. In order to carry out the simulations, the commercial package Fluent is used as a solver. The algorithm is S.I.M.P.L.E.C. (Patankar, 1980), and, when necessary, the classical turbulence model k-e is used.

THEORETICAL ASPECTS The numerical method used in this paper is based on solving the mass conservation and the momentum conservation equations written in cylindrical coordinates. For mass conservation: L / ~ r V r ) + k ~ V 0 + ~Vz =0 r 3r r 30 3z 1725

1726 For momentum conservation:

C. XUEREB and J. BERTRAND

3Vr + v r 3 V r P ~ 3t 3r
tangential component

+ VO 3Vr r 30

VOz + v z 3 V r ] = r 3z /

3p _ br

(rz.) + r~r

+ pg~

p/3vo + vr3VO + v o 3 v o
3t br r 30

+v_~

+ vz3VO / = _ ~ o
3z / r 30

axial component

{3Vz+ vr3VZ P~ 3t ~r

+ V0 3Vz + VZ3~zZ } = _ 3p . ( l ~ _ ( r Z r z ) +13z0z + ~tzz)+ pg~ r 30 bz r 30 3z

Taking into account that the fluid is uncompressible, the components of the stress tensor can be expressed as follows:

Xr0='to~=-gt[r~(~} Xzo= xo~ = - t . t x~=z==-p[b~rZ

+ 13VrIr 30J + 1 30J + 3Vzr]

vzl

This can be written in a simplified way, using tensor notation: x = -gtA where gt is the liquid viscosity. In the case of nonNewtonian fluids, the concepts of generalized Newtonian fluid is applied. This means that a relationship can be written between the stress tensor and the shear rate tensor, in which P a p , the liquid apparent viscosity, is the proportionality factor. This is a scalar which depends on the shear rate tensor. Consequently, the apparent viscosity has to be defined in every point. In this case, the rheological behaviour of liquid can be described by the Ostwald-de Waele model, which correctly fits very numerous highly viscous fluids with a pseudo-plastic behavior. It is written as follows: x=-m A

level2

level 1

with I ( A : A ) = 2 where n and m are the behavior and the consistency indexes

i'Figure 1, Tank geometry.

3-D hydrodynamics in a tank stirred by a double-propeller system

1727

of the liquid, respectively, and Ov the viscous dissipation function which can be expressed in terms of shear rates: Ov = 2 [ ( ~ - } 2 + {l_3VO + ~_)2 + { 3 V z l 2 ] + [ r o _ _ ~ ( ~ ) + l O V r l 2 + [lOVz + 3V012+ [O~zr + 3Vz] 2 ~r 30 ~-z! J r O0 J Lr 30 Oz J Or J In the turbulent regime, time-averaging of the momentum equations makes terms of the Reynolds tensor (turbulent stresses) to appear. Among the different turbulence models developed in order to estimate these stresses, the k-e model is largely used in Process Engineering, and particularly in mixing operations (Bakker and Van den Akker, 1994, Dong et al., 1994, Fokema et al., 1994, Hutchings et al., 1989, Kresta et Wood, 1991, Luo et al., 1993, Murthy et al., 1994, Myers et al., 1994, Weetman, 1994). The k-e model is based on the concept of turbulent viscosity. Taking into account some hypothesis to simplify the problem (Tatterson, 1991), transport equations can be written for the turbulent kinetic energy k and for the eddy dissipation 8:
kinetic energy

Ok+ vrOk+ at Or

r~

az

Lr~-~

__ +

7a--~

az2J

+ 2vt/(OVrlZ+ F1 [0V0 + Vr)]'+ IG3Vzl2t + [(3VO 1 3Vzt2 (3 3Vr/2 + [1 3Vr r 3---[V0/]21 vt|tOz + r30! + ~rz + 3 z ! t r 30 + 3r~ r q J - ~

where Ok is a constant. The kinematic turbulent viscosity v t will be defined further. The turbulent kinetic energy is classically defined as follows: k = 1 (V-~ + V0"2 Vz'2) '2 2
e d d y dissipation

8~_~ + Vr3...8_e + VO3e + Vz38 = v_.3_t[l~_/ra8 ] + 1 3 + 8 _ 2_ 3t Or r 30 3z a~ [r 3r~ 3rl r 2 302

328] 3z2]

k
k

I~ 3r ]

l_ ~ (90 r
r D0 I +

I1

+ Vr

+ Cle~Vt {(O~zO + 10Vz/2

(~
+

t)~zr)2
+

[13Vr + r 3 (_V~)]} - c2e--k 2 82 --00

where Ok, oe, Cle et c2e are constants of the model. When k and e have been calculated by solving the previous equations, the value of the kinematic turbulent viscosity is deduced from the following relationship:

Vt -- eg(kll2)(k3/2) - cllk2 E E
It has been established by Rodi (1984) that the constants of the k-8 model have to take the following values: % = 0.09, clc = 1.44, czc = 1.92, Ok = 1.0, a~ = 1.3 This way, the k-8 model enables the user to close the system of time-averaged equations.
Power Consumption

In the case of a turbulent flow, the integration of the field of eddy dissipation on the entire volume leads to the value of the total power consumption in the vessel.

P=P IE dV

1728

C. XUeReBand J. BEgTRANO In the case of the laminar flow, the total power consumption is calculated from the general relationship:

P = fl~p
JV

(I)v dV

The dimensional analysis enables us to characterize power consumption in a stirred vessel through the power number Np defined as follows: Np P p N 3 D5

In the turbulent flow range, Np remains constant. In the laminar flow range, it is the product Np.Re which remain constant, where Re is the Reynolds number (Nagata, 1975).

Boundary conditions
On the walls, bottom and baffles of the vessel, no-slip conditions have been imposed, as it is commonly assumed for inelastic fluids. The surface of the liquid is always considered as plane, that is to say that the vortex phenomenon is considered to be absent. This is warranted when the tank is fully baffled because Coriolis forces are reduced due to a decrease of the tangential motion. The liquid surface is free, which corresponds to the following conditions: Vz = 0 ; ~Vr = 0 az ; ~V__~0 0 = az

The propellers which are settled in the vessel are classically characterized by three profiled paddles. When using a non-sliding mesh, Poly0 in the case of a baffled tank, it is necessary to represent the rotating agitator by a set of boundary conditions experimentally obtained at the Polyl 2.8 1.0 immediate vicinity of the propellers. As propellers are essentially axial Poly2 4.8 0.8 impellers, it is convenient to take into account the boundary conditions Poly3 24.0 0.2 corresponding to the inferior plane of the propeller, which is the pumping plane. In order to obtain better convergence, it is not suitable Table 1. Consistency and to assign boundary conditions on the entire cylinder described by the behavior index of liquid solution rotating impeller for reasons of mass conservation. at four stages of the reaction The velocity components on the inferior plane of the propellers are measured by Laser Doppler Anemometry. A single component 4W argon LDA (Dante, c/Spectra Physics, model Stabilite 2017), with a green beam (1=514.5nm) is used in back scatter mode. A photomultiplier converts the light signal into an electrical one which is then processed by a flow-velocity-analyzer. Then, data are sent to a computer, analyzed and stored. Two sets of experiments have been carried out in order to obtain the boundary conditions corresponding to the propellers used in the industrial vessel. They correspond to different operating conditions: turbulent and transition regime, depending on the rheological properties of the liquid and of the speed of rotation. Boundary conditions in turbulent regime are used for the vessel filled with water. They consist in the three velocity components and in the turbulence characteristics. Boundary conditions which are used in the laminar flow range describe the vessel filled with a CMC 2% solution. m 0.001 n 1.0

G E O M E T R I C A L C H A R A C T E R I S T I C S AND O P E R A T I N G C O N D I T I O N S The cylindrical vessel considered in this work is equipped with a curved bottom the radius of which is equal to the tank diameter T, and with four vertical baffles (0.1 T). Figure 1 shows the geometry of the stirred tank. The liquid height is twice the tank diameter. The impellers are two MIXEL-TT three-blades downwards pumping propellers, the diameter D is equal to half the tank diameter. The clearance C between the lower propeller and the bottom is classically equal to T/3, and the spacing between the two agitators AC is equal to T. The computations concern an industrial 10 cubic meters cylindrical tank, the diameter T of which is equal to 1.85 m. The speed of rotation is equal to 1.03 s -1, in order that the tip velocity of the propeller be equal to 3 m.s -1, which is a usual industrial value. As experiments of Laser Doppler Anemometry have to be carried out, a laboratory plant is used. It consists of a cylindrical vessel of 190 mm in diameter mounted in a cubical tank, both made of transparent methylmetacrylate. Moreover, this vessel is provided with a torquemeter mounted on the shaft. When filled with water or with a liquid whose refraction index is of the same order of magnitude as the one of the

3-D hydrodynamics in a tank stirred by a double-propeller fq'---~ ....... T\

system

1729

t
I

....
.........

t-'i
I '

i .........
i
.... i .......

~.1
l,i t,i
~ l l i i I

i,j
(a)

..........

..it1

lr tl
-

.......

+,!

......
....

t,!

<'>

.....

,d+l

,,,i,l

..........
'" "''~\~llllill

,l,i

,,,, '"' lfr~ I I i~

"\\\\\tll

:'~\~ttt
~ ! ! i

'l r r t,.l['./~\\~.'?;yJ !,;

\\.t~.~l
. :-

,tit
rt/'l tiff,

"\\\t.l 1 i I li~ ~\ x~))!j7, s!'. # ,

~,~,~,ki',. +, i

"

t,.... . .
1/11

, "If-if .......

.
''

,rrf,I f'!
i.i

'

,, , ,

- ........

.........

......... ......... ......... ......... "',',',',', \ ,, l

ii i,!

i,~
II Il

I,i

, t i,~lt:lttT)~)t.-, \\\\~i

' :~l]-l.~\\\tl/[i

II I~ ..... ~l~,i IP ql ...... liii,f 11 11. . . . . ~lili~! : :11" 1"""ill i,f

ib

II. . . . . . . . .

~,i

i+

! Ii

c-\t\kkk~ft~, i (c)

tilt.

i
. . . . . . . . . .

.....

f
i~ i,

@
............... .................
, i IIl~__...~\\\lll

....
. . . . . . . . . . . . . .

t! i! ii ii I I
II

(e)

I
,#1]~

,,lr,,.

t lll,,,~'~'\\\U~:ii

~l l/+,
rl l+tr
""--~ ...... ............. .............
. . . . . . . . . . .

; I ', !
II "1! !!
+~tl

.............

ii
!i Ii ii i ii
ij ,,i i i

........ .......
. . . . . .

F i g u r e 2. F l o w p a t t e r n s in a v e r t i c a l p l a n e b e t w e e n two baffles (a) P o l y 0 ( V m a x = 1 . 2 2 m . s - l ) , (b) P o l y l ( V m a x = 1 . 0 6 m . s 1), (c) P o l y 2 ( V m a x = 1 . 0 6 r e . s - I ) (d) P o l y 3 ( V m a x = 1 . 0 6 m . s - 1 ) , (e) P o l y 3 w h e n b a f f l e s are r e m o v e d (Vmax = 1.06re.s-I).

.....

. . . . . . . . . . . . ............ ............ ,,
ill/.

1~+.

, t l ~'.

',:,Z----'-',,,,, ' I 117 ,- ~-"+""\\\} 11 ' i

,T_i~
~

~-X,5 ~l'i

1730

C. XUEREB and J. BERTRAND

..... .......

". . . . i . . I '"i,.i

i
i

. . . . . . . . "'"+,i
.11,ii ..... ""',i~ i .,ll,,.".,,"qlil,i ., J " , " / A ~ , i , r ~,.., q,,!ll/h, i

liquid inside the cylindrical vessel, the cubical vessel essentially eliminates the distortion of laser beams and no additional correction is needed for refraction effects, The simulations are carried out at four different stages of the polymerisation reaction. Table 1 reports the rheological properties of the fluid at these stages. Fluid density is considered to be equal to 1000kg.m3. RESULTS

The knowledge of velocity fields enables the user to quickly estimate i ~il)J',. "u~.///~ i whether the vessel is correctly designed in order to ensure a good homogeneity of a given liquid. Actually, in order to achieve this task, large circulation loops are needed, the partition of the vessel in different zones i ., ! must be avoided as well as development of dead zones. i I1! ;~..... i,'~ .... At the beginning of the reaction, the flow regime is turbulent and the II:::::: :::::r fluid is called Poly0. The whole volume of the reactor is agitated (Figure 2). Nevertheless, two zones are clearly identified around the two propellers. The liquid passes down through the volume swept by the propeller, goes up near the wall and is pumped in again by the same propeller. This characterizes a self-feeding of these impellers due to the large distance between the two propellers. These circulations are not ideal but are acceptable in the case of the turbulent flow regime, because of the very important turbulent diffusion phenomena which occur and which enable liquid to pass from one circulation loop to the other one. In the laminar flow range corresponding to the three next stages (Figures 2b, 2c, 2d), flow patterns appear to be quite different for the same operating conditions (tip velocity of the paddles equal to 3re.s-l). Firstly, it has to be noticed that the boundary conditions under the Figure 3. Flow patterns in a vertical propellers are not identical to the ones experimentally obtained in the plane just before the baffle. Case of turbulent flow range. The first difference consists in a flow more radial and Polyl (Vmax = 1.06m.s-1). less axial in the case of the highly viscous fluid. The second one lies in the fact that the characteristic pumping number Nq of the propellers is lower in the case of the laminar flow regime (0.45 in the laminar flow range and 0.65 in the turbulent flow range). The pumping number corresponds to the undimensional value of the liquid flow rate passing through the volume swept by the propeller: Nq QP N D3

',H,,

Velocity

fields

Secondly, there are many circulation loops surrounding the two main ones found around the propellers. These secondary loops which are turning in the opposite direction are generated by viscous forces that create

!,'~~-.~,
.' / "I

(a)

,-"m! / \

"

",

" "'

\
L--..'X'Y
.i--.'...'.". ',"~-...'-

'
, .

~, I s '
, fl
. _

\
I
I

~ J ' l . _ z ' . _ . ~ .

". " ,. i

...... 1

Figure 4. Flow patterns in a horizontal plane at level 1 (see figure 1). (a) Polyl (Vmax = 0.55re.s-I), (b) Poly3 (Vmax = 0.47m.sl).

3-D hydrodynamics in a tank stirred by a double-propeller system

i
_

i
-

(a)

~
/ 1

1731

(b)

.~.

-... ~

~....~-__-_ -

;;.-.---?-;-. \
" . -'" "

--

Figure 5. Flow patterns in a horizontal plane at level 2 (see figure 1). (a) Polyl (Vmax = 0.27m.s-1), (b) Poly3 (Vmax = 0.25m.s-1). entrainment of fluid elements at the vicinity of the main flows. Let us remark in particular the motion of fluid in the bottom of the vessel: in this zone, the replacement of fluid elements is very difficult, what constitutes a real problem in order to achieve a satisfying homogenization. The flow patterns exactly correspond to the observations carried out in the laboratory plant by illuminating a plane by a sheet of laser light when the vessel is filled with a solution of CMC 2%. While the fluid is becoming more viscous and non-Newtonian, the structure of the flow patterns remains of the same type, but dead zones are more and more important. These zones are constituted by almost stagnant liquid. Circulation loops around the propellers are smaller and smaller. These phenomena are the result of a poor momentum transport occuring when the behavior index of the fluid is far lower than 1.0. It has to be noticed that in the case of Poly2, the behavior index 0.8 is not far from 1.0, and the flow patterns are quite similar. In the case of Poly3, baffles have been removed (Figure 2e) in order to verify whether baffles are responsible of the large dead zones developed. It can be concluded that the presence of baffles does not modify the flow significantly in this geometry. The flow pattern just before the baffle is represented in figure 3 for Polyl and this figure clearly points out the 3-D character of the flow. A recirculation of the fluid appears near the surface, and the velocity vectors decrease in the vicinity of the baffle. The secondary circulation loop below the upper propeller is displaced towards the center. The influence of baffles on the flow patterns also appears in figures 4 and 5, representing horizontal sections of the vessel. At level 1 (defined in Figure 1), below the inferior propeller, the non-Newtonian character of the liquid does not generate important differences, except near the shaft where the radial components go to the center in the case of Poly3. At level 2, between the two propellers, let us remark in particular the large circulations which occur in opposite sense between two baffles in the case of Polyl. This phenomenon is replaced by an almost stagnant zone with the highly non-Newtonian fluid Poly3. At this level, it can be pointed out that the motion is less tangential in the case of the non-Newtonian fluid.

Energy dissipation
The vertical field of eddy dissipation corresponding to the case of the turbulent regime is given in the figure 6a. It can be noticed that the proximity between the lower propeller and the bottom generates a zone characterized by a rather high dissipation of energy. The major part of energy dissipation takes place in the vicinity of the propellers. When the liquid becomes highly viscous and the laminar flow regime (figure 6b and 6c) the maxima of energy dissipation occur at the level of the propellers, but nearer to the center of the vessel. The power consumption is also important near the walls, due to high shear rates in this area. This phenomenon tends to decrease when the fluid becomes more pseudo-plastic. The influence of baffles is illustrated in figure 7 corresponding to the power consumption per unit mass in a horizontal plane at level 2. In the case of the turbulent flow, the energy dissipation essentially takes place just before the baffle, in a large zone near the walls. In the case of the laminar flow, the power consumption takes place at the tip of the baffle, and the tangential aspect of the dissipation zones disappears, due to very low values of the tangential velocity.

1732 (a)

C. XUEREB and J. BERTRAND

(b)
f

(c)

I
I I I

1 58E-01

2.25E ol

9.38E 02
1.31E-02 'E-O1

.84E -01

2.G2E-OI

Figure 6. Field of energy dissipation in a vertical plane between two baffles. (a) Poly0, (b) Polyl, (c) Poly3.

(b)
,, ,

e_.3_

_ _/ . . . . .

........

Figure 7. Field of energy dissipation in an horizontal plane at level 2. (a) Poly0, (b) Polyl.

The total power consumption has been calculated from the fields of local energy dissipation. turbulent flow (Poly0): Experimental Np = 1.25 ; Calculated Np = 1.21 ; laminar flow (Polyl): Experimental Np.Re = 40 ; Calculated Np.Re = 42 ; The agreement between CFD and measurements is quite satisfying. CONCLUSION Four cases of rheological behaviors have been taken into account with a possible change of geometry (removal of baffles). It has been shown that the flow patterns are quite different in the turbulent and in the laminar flow regimes. In this last case, recirculation loops develop, and dead zones appear, which become more important

3-D hydrodynamics in a tank stirred by a double-propeller system

1733

as the fluid becomes more non-Newtonian. Power consumption obtained from the calculations has been succesfully compared with experimental results. These results can be a good aid for the design and the operation of a stirred vessel, and they enable us to point out the real need of well adapted geometry of a vessel according to the physical properties of the fluid, even though this often turns out to be difficult when it is the chemical reaction which generates the change of rheological properties. It is clear that a given geometry cannot perfectly adapt both to laminar and to turbulent flows. The study of agitation of non-Newtonian fluids is a great field of interest. The commercial simulation tools enable us nowadays to take into account the pseudo-plastic behavior of many fluids, but it remains impossible to take into account thixotropy or elasticity, rheological characteristics which are often encountered during the synthesis of polymers. The improvement of the capacity of the computers will certainly allow us in the future to enter into the software such complex models of rheology.

Acknowledgements
Authors are grateful towards H61~ne Barru6 and Isabelle Naude, both graduate students, for their active contribution to this work, especially in using the Fluent package.

NOTATION cl.t, Cle, c2E = constants in the k-e model

D g k m n N Np p
r

= propeller diameter = gravity = turbulent kinetic energy = consistency index = behavior index = speed of rotation = power number = pressure
= radial coordinate

Re t T Vr V0 Vz Vi Vi' z

= = = = = =

Reynolds number (= ND2p/la) time tank diameter radial velocity tangential velocity axial velocity

= mean velocity component = fluctuating velocity component = axial coordinate

Greek letters
e A
"C

= energy dissipation = shear rate tensor

}.t gap
Vt

= viscosity = apparent viscosity

turbulent kinematic viscosity

p 0 Ok, ere (I)v

= fluid density = tangential coordinate = constants in the k-e model = viscous dissipation function

REFERENCES
Bakker, A. and Van den Akker, H. E. A., 1994, T.I.Ch.E. Part A 72, 583. Dong, L., Johansen, S.T. and Engh, T. A., 1994, Chem. Engng Sc. 49-20, 3511 Fokema, M. D., Kresta, S. M. and Wood, P.E., 1994, Can. J. Chem. Eng. 72, 177.

1734

C. XUEREBand J. BERTRAND

Hutchings, B. J., Weetman, R. J. and Patel, B. R., 1989, presented at ASME Winter Annual Meeting, San Francisco, USA Kresta, S., and Wood, P. E., 1991, AIChe Journal 37-3, 448. Luo, J. Y., Gosman, A. D., Issa, R. I., Middleton, J. C. and Fitzgerald, M. K., 1993, TJ.Ch.E. Part A 71,342. Mishra, V. P. and Joshi, J. B., 1994, T.I.Ch.E. Part A 72, 657. Murthy, J. Y., Mathur, S. R. and Choudhury, D., 1994, I. Chem. E. Symp. Ser. 136,341. Myers, K. J., Fasano, J. B. and Bakker, A., 1994, I. Chem. E. Symp. Ser. 136, 65. Nagata, S., 1975, Mixing: Principles and Applications, ed. John Wiley & Sons, New York. Patankar, S. V., 1980, Numerical Heat Transfer and Fluid Flow, Hemisphere Publishing Company, Washington, U.S.A. Rodi, W., 1984, Turbulence models and their application in hydraulics, 2nd ed., Intern. Ass. Hydraulic Res., Rotterdamseweg 185, 2660 MH Delft, the Netherlands. Tatterson, G. B., 1991, Fluid Mixing and Gas Dispersion en Agitated Tanks, ed. Mc Graw Hill, New York. Weetman, R. J., 1994, I. Chem. E. Symp. Ser. 136, 49.