Sei sulla pagina 1di 10

Chemical Engineering Science 56 (2001) 2435}2444

Barium sulphate precipitation in a pipe * an experimental study


and CFD modelling
Jerzy Ba"dyga*, Wojciech Orciuch
Department of Chemical and Process Engineering, Warsaw University of Technology, PL-00645 Warsaw, ul.Waryn& skiego 1, Poland

Abstract

Precipitation of barium sulphate in an unpremixed feed two-dimensional tubular precipitator is studied experimentally and
interpreted theoretically using a closure procedure proposed previously by the authors. The closure employs the presumed beta PDF
of the inert type composition variables formed with the local values of Ba> and SO\ concentrations and the turbulent mixer model.

The #ow "eld is computed using the k} model. The method enables the particle size and the particle size distribution to be predicted
and compared with experimental data.  2001 Elsevier Science Ltd. All rights reserved.

Keywords: CFD; Closure; Precipitation; Tubular reactor; Turbulence

1. Introduction involving instantaneous, homogeneous chemical reaction


and subsequent product crystallization in a system with
Precipitation of sparingly soluble materials is an im- inhomogeneous turbulent #ow. Prediction of the precipi-
portant unit operation in chemical industry, which is tation process requires a complete knowledge of the spa-
employed in the production of bulk and "ne chemicals. tial distribution of instantaneous, local concentrations of
The solid product has often a wide crystal size distribu- ions AL> and BL\, product C and other ions present in the
tion (CSD) which determines the "ltration, washing and system. Identi"cation of these concentrations requires
settling abilities of suspensions, as well as the product solving of a system of partial di!erential equations de-
quality (particle-size homogeneity, surface area etc.). Pre- scribing the momentum, mass, species and populations
diction of the crystal size and CSD depending on the balances in the system. To this end we describe the process
operation mode and process parameters is thus of key of precipitation using statistically averaged balance equa-
practical importance. E!ects of process parameters on tions governing mean quantities using the Eulerian or
precipitation can be predicted by a direct use of the "xed-frame perspective. The averaging procedure pro-
mechanistic micromixing models (Pohorecki & Ba"dyga, duces more unknowns than the number of balance equa-
1979; Marcant & David, 1991; Ba"dyga, PodgoH rska, tions and the set of di!erential equations becomes
& Pohorecki, 1995) or by linking the micromixing mod- unclosed. The `closure problema can be solved by intro-
els to a computational #uid dynamics (CFD) (Pipino, ducing some additional information, forming a `closure
Barresi, & Fox, 1994; Ba"dyga & Orciuch, 1997; Ba"dyga, hypothesisa. To this end a method employed by Ba"dyga
Henczka, & Orciuch, 1997; Wei & Garside, 1997; van and Orciuch (1997) for modelling of precipitation in
Leeuwen, Bruinsma, & van Rosmalen, 1998; Ba"dyga homogeneous turbulence is adapted to the case of in-
& Orciuch, 1999). This paper belongs to the second homogeneous turbulence. The closure procedure is based
group; we consider the process of precipitation of solid on the Beta probability density function (PDF) of the inert
products from two liquid ionic solutions A and B: type composition variable formed with the concentrations
of precipitating species, and the turbulent mixer model
AL>#BL\PC (1) implemented in the commercial k} CFD code (Fidap 8.5).
The same method was employed earlier for prediction of
* Corresponding author. Tel.: #1-48-22-6606376; fax: #1-48-22- the selectivity of complex reactions (Ba"dyga & Henczka,
8256037. 1997); preliminary results for the precipitation process
E-mail address: baldyga@ichip.pw.edu.pl (J. Ba"dyga). were also presented by Ba"dyga et al. (1997).

0009-2509/01/$ - see front matter  2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 9 - 2 5 0 9 ( 0 0 ) 0 0 4 4 9 - 8
2436 J. Ba!dyga, W. Orciuch / Chemical Engineering Science 56 (2001) 2435}2444

2. The population balance equation for turbulent 6ow moments can be expressed as



The closure is developed using the moment trans- m (x, t)" m (x, t, f )(x, t, f ) df
formation of the population balance of particles for one H H

internal coordinate ¸:


" (x, t, ¸)¸H d¸ for j"0,1,2,2, (6)
m (u m ) 
H # NG H "0HR #jGm #B !D
t x , H\ D D where
G


for i"1,2,3, j"0,1,2,2 (2) 
m (x, t, f )" (x, t, ¸, f )¸H d¸ (7)
H
where u "(u , u , u ) is a vector of particle velocity 
N N N N
along external coordinates x"(x , x , x ) and the mo- is the moment within the #uid element identi"ed by f at
  
ments are de"ned as any position in time and space, and





m (x, t)" (x, t, ¸)¸H d¸ (3) (x, t, ¸)" (x, t, ¸, f )(x, t, f ) df. (8)
H 

Eq. (2) employs the local instantaneous values of the Eq. (5) should be solved together with the partial di!eren-
velocity and concentration. In turbulent #ow there are tial equations describing momentum, mass and species
#uctuations of the particle velocity, as well as #uctu- balances. For example, the local mean concentrations of
ations of species and particle concentrations. The appro- ions A and B are described by

 
priate generalisation of Eq. (2) for the case of turbulent c  c   c 
#ow can be obtained by making Reynolds averaging. The ? #u  ? " (D #D ) ?
t G x x ? 2 x
averaged form of the moment transformation of the G G G


population balance takes then the form k  
! ? N G( f )m ( f )( f ) df. (9)
2M 
m  u m  ! 
H # NG H
t x
G
3. Velocity and di4usivity of particles
"0HR #jGm #B !D 
, H\ D D
for i"1,2,3, j"0,1,2,2 (4) The local values of particle velocity u and particle
NG
turbulent di!usivity D appearing in Eq. (5) can signi"-
N2
Introducing a concept of gradient di!usion for particles cantly di!er from the local values of #uid velocity u and
G
and employing a mixture fraction f"c /c (f is a vol- turbulent di!usivity D . We consider spherical particles
  2
ume fraction of nonreacting #uid originating in A stream of diameter d smaller than the Kolmogorov microscale,
when A and B solutions are mixed) one can present Eq. d(, and apply the theory of particle motion by Tchen
(4) in the form in the form presented and developed by Hinze (1975).
The equation for motion of a spherical particle reads
m  m  u 
H #u  H #m  NG

t NG x H x dv v !v dv R (dv /dt)!dv /dt
G G NG " G NG #b G #c G NG dt, (10)
dt  dt (t!t
N R
 
 m 
" D H # 0HR  where
x N2 x ,
G G
[2( /)#1]d 3


  " N , b" ,
#j G( f )m ( f )( f ) df#B !D  (5) N 36 2( /)#1
H\ D D N


18 

where D is the local value of the particle turbulent c" .


N2 (2 #)d 
di!usivity coe$cient. N
The composition variable f enables identi"cation of The third term on right-hand side of Eq. (10) (the Basset
#uid elements; the statistics of #uid elements is then term) represents the e!ects of deviation in #ow pattern
represented using the probability density function ( f ). from steady state. As shown by Hinze (1975), this term
Notice that identi"cation of ( f ) enables calculation of can only be substantial when the particle is accelerated at
the local average value of any quantity provided that one a high rate by a strong external force, which is not a case
can express the local instantaneous value of this quantity in our study. Hinze (1975) has shown that for  /'1
N
in terms of f. For example, the averaged values of the the e!ect of the Basset term becomes of importance when
J. Ba!dyga, W. Orciuch / Chemical Engineering Science 56 (2001) 2435}2444 2437

(1/d)( / )0.6, where is the angular frequency; sub-


stituting for the maximum frequency
0.495(/ )
B
(Ba"dyga & Bourne, 1999) we see that the Basset term can
be neglected when d(. Estimating now dv /dt+v / ,
G ) )
where v and  are the Kolmogorov velocity and time
) )
scale, respectively, we see that the second term on the
r.h.s. can be dropped when   , or
N )
 ( /). (11)
N
Substituting from Eq. (10) for  we get again the condi-
N
tion d(.
Note that when the relaxation time  is much smaller
N
than the Kolmogorov time microscale  , most of the
)
time v +v within the Kolmogorov eddy; clearly one
NG G
can assume then that Fig. 1. Dependence of the di!usion-coe$cient ratio between discrete
particles and #uid on time for various particle diameters and the rate of
u "u . (12) energy dissipation values.
NG G
Tchen's theory yields the following relations between the
particle and #uid di!usivity coe$cients:
where  is the time scale of turbulence
2
E for short di!usion times, k
 " 2 "1.5c (18)
D v  1#b /¹ 2 k I
N2 " NG " N *, (13) 
D v 1# /¹
2 G N * and Sc and Sc
2 N2
are turbulent #uid and particle
Schmidt numbers. For BaSO particles D +D ; in the
where the Lagrangian integral time scale ¹ can be
*  N2 2
estimated as extreme case of "40 m s\ and for d"10 m we
observe (D /D )'0.93.
N2 2
The occurrence of (D /D )'1 may be observed for
u 0.8 k N2 2
¹ +0.4 G " ; (14) relatively large particles and can be attributed (Hinze,
*  3 
1975) to the fact that particles respond only to turbulent
E for long di!usion times, eddies of a scale larger than the particle radius ( "ltering
e!ect). For particles smaller than the Kolmogorov scale
D this e!ect should not be thus observed. Note also that the
N2 "1; (15) divergence of the particle velocity u /x [see Eq. (5)]
D
2 NG G
is in general not equal to zero. Only for small values of
E for intermediate t one has the relaxation time ( /¹ 1) this term can be neglected
N *
and this is a case for BaSO particles of d(10 m.
D 1!b e\RON !e\R2* 
N2 "1# . (16)
D (¹ / )!1 1!e\R2*
2 * N
4. Experimental
Fig. 1 illustrates this phenomenon and shows that
for d(10 m D +D for any di!usion time. Experi- Experiments for BaSO precipitation were carried
N2 2 
mental observations for ( /)'1 show however, out at ¹"298 K in a tubular reactor with a i.d. of
N
that both relations (D /D )(1 and (D /D )'1 D"0.0320 m equipped with a concentrically located
N2 2 N2 2
are possible (Hinze, 1975; McComb, 1990). The case tube with an i.d. of d "0.0018 m and o.d. of 0.0025 m as
GL
(D /D )(1 may result from relaxation of the particle shown in Fig. 2. The entrance length to the reactor before
N2 2
velocity in the inhomogeneous turbulence. To account the position of injection from the small feed pipe was
for this e!ect a simple approach relying on empiricism equal to 2 m (i.e. 62.5 times the main tube diameter) to
can be used. We consider here as an example of such an eliminate the entrance e!ect (25}40 pipe diameters).
approach the method developed by Melville and Bray A concentrically located feed pipe was constructed su$-
(1979) yielding ciently long (0.24 m, or 96 times the o.d. of the feed nozzle
pipe) to eliminate e!ects of the #ow disturbance by

 
D Sc  \ the radially oriented part of the nozzle. An aqueous solu-
N2 " 2 1# N , (17)
D Sc  tion of Na SO (A) was introduced over the entire
2 N2 2  
2438 J. Ba!dyga, W. Orciuch / Chemical Engineering Science 56 (2001) 2435}2444

mean velocity, turbulent kinetic energy and turbulent-


energy-dissipation-rate "elds at steady state. Calcu-
lations of the #ow "elds requires integration of the set of
model equations represented by Eq. (19) and Table 1:

1 
(ru )# (u )
r r P x V
Fig. 2. Sketch of the reactor head.

   
1   
"  r #  #S . (19)
r r  r x  x
cross-sectional area of the reactor, whereas a solution of
BaCl (B) was fed through the concentrically located In Table 1 we have
"
#
"
#0.09k/ and
  2
injector. Analytical-grade chemicals and doubly distilled p "p#k. The coe$cients which appear in Eq. (19)

water were used to prepare reactant solutions. The reac- are assumed to be universal and thus constant
tor was cleaned before each experiment with concen- ( "1, "1.3, c "1.44 and c "1.92) although
I C C C
trated sulphuric acid to remove the barium sulphate from they can vary slightly from one #ow to another. As
the reactor walls. a steady-state process is considered we employ here the
During experiments the pipe Reynolds number varied time averages denoted by the overbar instead of the
from 20,000 to 70,000, the feed concentration of Na SO ensemble averages. Integration of Eq. (19) requires em-
 
was equal to c "0.015 mol dm\ and the feed concen- ploying of the boundary conditions; the inlet conditions

tration of BaCl was equal to c "0.100 mol dm\ or are found by computing separately the #ow "elds in the
 
c "1.500 mol dm\. The velocity ratio Ru (equal to central tube and in the annular #ow at z"!8D and

the mean velocity in the injector u over the mean velo- z"!3D for the annular and central #ow, respectively.
A
city in the reactor u ) varied from 1.0 to 6.7. Each experi- The wall boundary conditions for turbulence employ
ment was repeated 3}4 times. Samples of the product y> (dimensionless wall distance) values equal to 30; for
stream were collected at the reactor outlet and the crystal y>(30 the van Driest's mixing length approach was
size distribution was measured after the experiment using used (standard in FIDAP 8.5). A grid consisting of 9200
a Coulter-Counter. Particles of BaSO were observed to 30,000 nodes was used in computations with grid mesh
 dense near the feeding point; such number of nodes was
and photographed with the use of scanning microscope.
Only experiments with no aggregation observed are pre- checked to make the results independent of the grid
sented and interpreted in this paper. resolution.
To calculate distributions of concentrations variances
of f characterising the inertial}convective  , viscous}

5. Procedure of computations convective  and viscous}di!usive  subranges of tur-
 
bulent spectrum the turbulent mixer model (Ba"dyga,
The commercial CFD package FIDAP 8.5 employing 1989) is used. The model interprets the mixing process as
the k} model of turbulence is employed to compute the the process of convection, dispersion and dissipation of

Table 1
Meaning of terms of Eq. (19)

  S

1 0 0

   
u  1 u  u p
V
r P #
V !
r r x x x x

   
u  1 u  u 2
u p
P
r P #
V ! P!
r r r x r r r

         
k u  u  u  u u 

2 P # P # V # P# V !
2 r r x x r
I

           
  u  u  u  u u 
C )
2 P # P # V # P# V !C 
k C 2 r r x x r C
C
J. Ba!dyga, W. Orciuch / Chemical Engineering Science 56 (2001) 2435}2444 2439

the concentration variance : and c(x, f ), c(x, f ) can be easily calculated from
1 
"[ f (x, t)!f (x)]"  (x)#  (x)#  (x). c(x, f )!c(x, f )#c
1   
(20) f"   (30)
c #c
The mixing calculations are performed by integrating the  
following set of equations: as the A and B ions cannot coexist (i.e. when c'0 then

c"0 and vice versa)
fM fM The average concentrations are given by
u #u
V x P r



c(x)" c(x, f )(x, f ) df (31)
   
1 fM  fM ? ?
" (D #D )r
K 2 r
# (D #D )
K 2 x
, (21) 
r r x
for instantaneous precipitation, whereas for very slow
    precipitation c (x)"c fM (x) and c (x)"c [1!fM (x)].
u L #u L   
V x P r Once the PDF for the mixture fraction f is identi"ed
one can use the relations between the ion concentrations

   
1     
" (D #D )r L # (D #D ) L and the mixture fraction to calculate:
r r K 2 r x K 2 x
#R !R for n"1,2,3, (22) E the local mean concentrations of A and B ions
.L "L


where R and R stand for production and dissipation 
.L "L c (x)"
?
c (x, f )(x, f ) df,
?
(32)
terms: 
E the local mean nucleation rate
    
fM fM  fM 
R "!2 u f  "2D # , (23)
. G x 2 x r

G 
R (x)" R (x, f )(x, f ) df. (33)
  , ,
R "R " 
 R , R"1.86, (24) 
" .   k In Eq. (33) R is given by the sum of homogeneous
1 ,
and heterogeneous nucleation rates:

 
R "R "E  , E"0.058 , (25)
" . 
 
R (x, f )
ln ,
R
  
17,050
R "G  , G"E 0.303# . (26)
" 
   
Sc c (x, f )c (x, f )  \
"!A ln   (x, f ) (34)
K 
Having fM and  one can identify the probability density 1?
1
function of f, ( f ), which is approximated using the Beta where K is thermodynamic solubility product
1?
function (K "1.1;10\ kmol m\) and  is the mean
1? 
activity coe$cient for the salt (Bromley, 1973). The


(x, f )"fT\(1!f )U\ yT\(1!y)U\ dy (27) nucleation kinetics was described by "tting Eq. (34) to
 experimental data by Nielsen (1969) (A"44.60,
where v"fM ( fM (1!fM )/  !1), w"v(1!fM )/fM . This distri- R "1.06;10 m\ s\ and A"3020, R "
 
1
bution was veri"ed and recommended by many authors 1.50;10 m\s\ for heterogeneous and homogene-
(Ba"dyga, 1989; Rhodes, 1975; Li & Toor, 1986). ous nucleation, respectively). Following the suggestion
The local instantaneous concentrations of ions A and by Nielsen (1969) we have neglected here ion pair
B are estimated using the method described by Ba"dyga formation; considering this e!ect should lead to other
and Orciuch (1997). The method is based on linear inter- values of the nucleation kinetics constants but to sim-
polation of c (x, f ) for "A, B, between the values for ilar values of the rate of nucleation.
?
instantaneous precipitation c(x, f ) and for very slow E The local mean crystal growth rate
?
precipitation c (x, f ):

? 
GM (x)" G(x, f )(x, f ) df (35)
c (x)!c(x) 
c (x, f )"c(x, f )#[c (x, f )!c(x, f )] ? ?
? ? ? ? c (x)!c(x) with G(x, f ) calculated from the two-step model
? ?

  
for "A, B, (28) c (x, f )c (x, f )  
G(x, f )"k  1 1  (x, f )!1
E K  1
where 1?
"k [c (x, f )!c (x, f )]
c ( f )
 "f"1!
c ( f ) "   1
(29)
c c "k [c (x, f )!c (x, f )], (36)
  " 1
2440 J. Ba!dyga, W. Orciuch / Chemical Engineering Science 56 (2001) 2435}2444

where k "4.0;10\ m s\ (Wei & Garside, 1997)


E
and k "1;10\ (m s\) (m kmol\) (Ba"dyga
"
& Orciuch, 1997).
E The local, instantaneous values of a solid-phase con-
centration in the suspension, c ( f ), can easily by cal-
!
culated from the local balance
c (x, f )"c (x, f )!c (x, f ). (37)
!  

Assuming that the moments are proportional to the


solid-phase concentration
m (x, f ) c (x, f )
H " ! (38)
m (x) c (x)
H ! Fig. 3. Mean crystal size vs. Reynolds number. c "0.015 kmol m\,

and neglecting breakage and aggregation one can present c "1.500 kmol m\, Ru"1.

the population balance in the form
m m
u H #u H
NV x NP r

   
1 m  m
" D r H # D H #0HR
r r N2 r x N2 x ,


m 
#j H\ G(x, f )c (x, f )(x, f ) df for "0,1,2,2
c !
! 
(39)
In Eq. (39) we have neglected divergence of the particle
velocity and substituted u , u and D by u , u and
NV NP N2 V P
D , respectively, which is justi"ed for the small BaSO
2 
particles of d(10 m. It should be pointed out that the
model presents now a closed set of equations and any
Fig. 4. Distribution of particle number concentration N
"tting or adjusting of the model parameters or constant (number of particles/m) along the tubular reactor for Re"30,000,
is not necessary. c "0.015 kmol m\, c "1.500 kmol m\, Ru"1.
 
Integration of the balance equations (9), (19), (21), (22)
and (39) gives the distributions of average concentrations
of all species c and the distributions of the moments (c "0.015 mol dm\, c "1.500 mol dm\) and
?  
m . The surface shape factor k "8.17, the volume shape compares experimental data with model predictions. In
H ?
factor k "1.36 and related sphericity  "(/k )" the "gure the mean crystal size, d , is expressed as
  ? 
0.727 are used for the low concentration experiments
(c "0.015 mol dm\, c "0.100 mol dm\), whereas m 1
  d "  . (40)
k "348.00, k "58.00,  "(/k )"0.208 are used  m 
?   ?
for interpretation of the high concentration experimental  J
data (c "0.015 mol dm\, c "1.500 mol dm\). There is a good agreement between the model predic-
  tions and experimental data; in both cases increase of Re
The shape factors used in this paper were determined
earlier by PodgoH rska (1993); see also Ba"dyga et al. decreases the crystal size. E!ect of Re results from mi-
(1995); Ba"dyga and Orciuch (1997), and Philips et al. cromixing and is related to signi"cant decrease of con-
(1999). centration #uctuations with increase of Re. Neglecting
the concentration #uctuations ( "0), equivalent to ap-
1
plying the PDF in the form
6. Results and discussion (x, f )"( f!fM (x)) (41)
In what follows the model predictions are presented results in the crystal size slightly increasing with Re. This
and compared with the experimental results. Fig. 3 e!ect of micromixing is illustrated in Figs. 4 and 5; Fig.
shows the variation of the mean crystal size with 4 shows the number concentration pro"le along the reac-
the Reynolds number at higher concentrations tor and Fig. 5 present evolution of the mean particle
J. Ba!dyga, W. Orciuch / Chemical Engineering Science 56 (2001) 2435}2444 2441

The di!erence is clearly observed; predicted and


measured distributions agree better for the full model.
The e!ect of mixing should be better observed at still
larger concentrations, but then we have observed signi"-
cant particle aggregation. Aggregation e!ect is neglected
in present modelling and will be included in our next
publication.
At smaller concentrations (c "0.015 mol dm\,

c "0.100 mol dm\) both models (i.e. the one includ-

ing concentration #uctuations and the one neglecting
them) yield the same d value that is moreover not

sensitive to Re. At Re"40,000, c "0.015 mol dm\,

c "0.100 mol dm\ and Ru"5 one has d "
 
Fig. 5. Evolution of average volume of crystals ( "k m /m ) along 3.79 m when considering concentration #uctuations
A J  
the tubular reactor for Re"30,000, c "0.015 kmol m\, c " and d "3.82 m when neglecting them. Both models
  
1.500 kmol m\, Ru"1. yield almost identical particle size distributions as shown
in Fig. 7a and b; agreement of the predicted and mea-
sured particle size distribution is good also at low con-
centration. Fig. 8 shows that at low concentration the
crystal size is still sensitive to the velocity ratio Ru. The
e!ect of Ru on the size of particles is well predicted by
both models that again yield identical results. Hence, in
this case there is no e!ect of micromixing but e!ects of
turbulent di!usion and convection. The observation that
micromixing e!ects are present only at high concentra-
tions (so for low values of the time constants for precipi-
tation comparing to time constants for mixing) agree well
with considerations presented by Ba"dyga and Bourne
(1999) (see Chapter 14) in relation to choosing an ad-
equate model of mixing for modelling of precipitation
processes. The time constant related to the dissipation of
the concentration #uctuations can be estimated as
 +(k/R)#1/E#1/G; the smallest values of  ob-
K K
served in the precipitation zone are 0.03 s at Re"20,000
and 0.008 s at Re"60,000. The time constant for turbu-
Fig. 6. Predicted and measured (averaged) crystal size distributions for lent di!usion decreases from 0.3 s at Re"20,000 to 0.1 s
Re"30,000, c "0.015 kmol m\, c "1.500 kmol m\, Ru"1. at Re"60,000. The e!ect of concentration #uctuations
 
on the CSD is expected when the characteristic time
constant for nucleation,  "N/R , or the characteristic
, ,
time for the crystal growth,  "(c M )/( Gk m )
% ! ! N ? 
volume. Neglecting the concentration #uctuations results are smaller than the time constant for mixing,  .
K
in faster precipitation and yields in this case slightly We observe  +10\ s at higher concentrations
,
di!erent "nal results. The crystal size distribution M (¸) (c "0.015 mol dm\, c "1.500 mol dm\), which
 
can be recovered from the moments (we use the "rst six means that a competition between micromixing and nu-
moments) by using the statistically most likely probabil- cleation phenomena is possible in this case. Of course,
ity distribution (Pope, 1979) turbulent di!usion can a!ect precipitation in this case as
well. Decreasing of the concentration of component B to

 
H c "0.100 mol dm\ increases  by about one order
 (¸)"B exp  A ¸L . (42)  ,
LH  L of magnitude, which yields only competition between the
L turbulent di!usion and nucleation. The characteristic
The constant B and the #1 coe$cients A in Eq. (42) time for the crystal growth is larger than time constant
 L
are determined from the  moments and from the condi- for mixing for both concentration levels. Notice, how-
tion that  (¸) integrates to unity. ever, that for still lower concentrations the only mecha-
LH
Fig. 6 gives an example of the crystal size distribution, nism to decrease supersaturation becomes the crystal
measured and calculated using both: the full model growths and then the CSD results from competition
and the model that neglects concentration #uctuations. between nucleation and growth.
2442 J. Ba!dyga, W. Orciuch / Chemical Engineering Science 56 (2001) 2435}2444

Fig. 7. Predicted and measured (averaged) crystal size distributions for Re"40,000, c "0.015 kmol m\, c "0.100 kmol m\; (a) Ru"3, (b)
 
Ru"6.

Ru from Ru"1 to 7 at Re"40,000 increases Re in the


H
zone of precipitation up to 110. This shows that a good
agreement between the model and experiment is, as ex-
pected, at Re '40.
H
Generally, one can conclude that a combination of the
mixing-precipitation model with CFD enables e!ective
interpretation of the precipitation process.

Notation

A, B, C chemical species
A constant in Eq. (34), dimensionless
Fig. 8. Mean crystal size vs. velocity ratio Ru for Re"40,000, b, c variables in Eq. (10), dimensionless
c "0.015 kmol m\, c "0.100 kmol m\. B birth function, mH m\ s\
  D
c concentration of substance , kmol m\
?
c concentration of non-precipitating tracer,
?
kmol m\
Figs. 3, 7a, b and 8 shows that agreement between the c concentration due to instantaneous precipita-
?
model predictions and experimental data improves with tion, kmol m\
increasing both: the Reynolds number, Re, and the velo- c concentration of  in its feed stream,
?
city ratio, Ru. A possible explanation of this e!ect is as kmol m\
follows. The models of turbulence and the models of c concentration of "A or B at the crystal
? 1
turbulent mixing that are used together with CFD codes surface, kmol m\
(including the ones employed in this paper) are usually d nozzle diameter, m
GL
based on the assumption of developed turbulence, i.e. d mass weighted particle diameter, m, m

they assume that the Taylor microscale Reynolds num- d particle diameter, m
ber, Re , takes high enough values to make the model D tubular reactor diameter, m
H
constants independent of the Reynolds number. As D particle turbulent di!ussivity, m s\
N2
shown by Sreenivasan (1995, 1996) the Kolmogorov con- D turbulent di!usivity, m s\
2
stant and the Obukhov}Corrsin constant are hardly de- D molecular di!usivity, m s\
K
pendent on Re for Re '40. Results of computation D death function, mH m\ s\
H H D
obtained with the k} model enable estimation of the E engulfment parameter, s\
local values of the Taylor microscale Reynolds number, f mixture fraction, dimensionless
Re "2.58k/( . For Ru"1, the Re values in the G linear growth rate of crystals, m s\ or mo-
H H
precipitation zone are equal to 29, 38 and 46 for lecular di!usion parameter, s\
Re"20,000, 40,000 and 60,000 respectively. Increasing k kinetic energy of turbulence, m s\
J. Ba!dyga, W. Orciuch / Chemical Engineering Science 56 (2001) 2435}2444 2443

k surface shape factor, dimensionless ( f ) probability density function, dimensionless


?
k mass transfer coe$cient, (m s\)(m kmol\) (¸) particle distribution function, m\
"
k rate constant for surface reaction, m s\  (¸) most likely particle distribution function
E LH
K thermodynamic solubility product, (normalised), m\
1?
kmol m\ angular frequency, rad s\
k volume shape factor, dimensionless
J
¸ crystal size, m Others
M molar mass of precipitated substance,  ensemble average
!
kg kmol\ * time average
m jth moment, mH m\
H
p pressure, Pa
R constant in Eq. (24), dimensionless
References
R dissipation terms (n"1,2,3), s\
"L
R constant in Eq. (34), m\ s\
 Ba"dyga, J. (1989). Turbulent mixer model with application to homo-
R nucleation rate, m\ s\
, geneous, instantaneous chemical reactions. Chemical Engineering
R production terms (n"1,2,3), s\ Science, 44, 1175}1182.
.L Ba"dyga, J., & Bourne, J. R. (1999). Turbulent mixing and chemical
Re Reynolds number, dimensionless
Re Taylor microscale Reynolds number, dimen- reactions. New York: Wiley.
H Ba"dyga, J., & Henczka, M. (1997). Turbulent mixing and parallel
sionless chemical reactions in a pipe * application of a closure model.
Ru velocity ratio, dimensionless Proceedings of the Nineth European Conference on Mixing Paris,
Sc Schmidt number, dimensionless France (pp. 341}348).
Sc turbulent Schmidt number, dimensionless Ba"dyga, J., Henczka, M., & Orciuch, W. (1997). Closure problem for
2 mixing and precipitation in inhomogeneous turbulent #ow. Acta
Sc turbulent particle Schmidt number, dimen-
N2 Polytechnica Scandinavica, Ch244, 41}43.
sionless Ba"dyga, J., & Orciuch, W. (1997). Closure problem for precipitation.
u Eulerian velocity component, m s\
G Transactions of the Institution of Chemical Engineers, 75A, 160}170.
¹ Lagrangian integral time scale, s Ba"dyga, J., & Orciuch, W. (1999). Closure method for precipitation in
* inhomogeneous turbulence. Proceedings of the 14th international
v variable in Eq. (27), dimensionless
v Lagrangian velocity component, m s\ symposium on industrial crystallization, Cambridge, England (p. 86).
G Ba"dyga, J., PodgoH rska, W., & Pohorecki, R. (1995). Mixing-precipita-
v Kolmogorov velocity microscale, m s\
) tion model with application to double feed semibatch precipitation.
v Lagrangian particle velocity component,
NG Chemical Engineering Science, 50, 1281}1300.
m s\ Bromley, L. A. (1973). Thermodynamic properties of strong electrolytes
w variable in Eq. (27), dimensionless in aqueous solutions. A.I.Ch.E. Journal, 19, 313}320.
x Cartesian coordinates, m Hinze, J. O. (1975). Turbulence. New York: McGraw-Hill.
G van Leeuwen, M. L. J., Bruinsma, O. S. L., & van Rosmalen, G. M.
(1998). Computational #uid dynamics approach to precipitation
Greek letters reactions: The importance of subgrid-scale #uctuations. Proceeding
 mean activity coe$cient for salt, dimension- of international conference on mixing and crystallization, Tioman
 Island, Malaysia (p. 37).
less
 rate of kinetic energy dissipation, m s\ Li, K. T., & Toor, H. L. (1986). Chemical indications as mixing probes.
A possible way to measure micromixing simply. Industrial and
 Kolmogorov microscale, m Engineering Chemistry, Fundamentals, 25, 719}723.

viscosity, Pa s Marcant, B., & David, R. (1991). Experimental evidence for and predic-

turbulent viscosity, Pa s tion of micromixing e!ects in precipitation. A.I.Ch.E. Journal, 37,


2
kinematic viscosity, m s\ 1698}1710.
turbulent kinematic viscosity, m s\ McComb, W. D. (1990). The physics of yuid turbulence. Oxford: Claren-
2 don Press.
 density, kg m\ Melville, E. K., & Bray, K. N. C. (1979). A model of the two-phase
 density of crystals or particles, kg m\
N turbulent jet. Journal of Heat and Mass Transfer, 22, 647.
,  variance of mixture fraction, n"1,2,3, di- Nielsen, A. E. (1969). Nucleation and growth of crystals at high super-
1 L saturation. Kristall und Technik, 4, 17}38.
mensionless
 characteristic time constant for crystal Philips, R., Rohani, S., & Ba"dyga, J. (1999). Mixing in a single-feed
% semi-batch precipitation process. A.I.Ch.E. Journal, 45, 82}92.
growth, s Pipino, M., Barresi, A. A., & Fox, R. O. (1994). A PDF approach to the
 Kolmogorov time microscale, s
) description of homogeneous nucleation. Proceeding of quarto con-
 time constant for dissipation of concentration vegno internazionale yuidodinamica multifase nell'impiantistica indus-
K trialles, Ancona, Italy.
#uctuations, s
 characteristic time constant for nucleation, s PodgoH rska, W. (1993). Inyuence of micromixing on precipitation. Ph.D.
, thesis, Warsaw University of Technology, Poland.
 particle relaxation time, s
N Pohorecki, R., & Ba"dyga, J. (1979). The in#uence of intensity of mixing
 time scale of turbulence de"ned by Eq. (18), s
2 on the rate of precipitation. Proceedings of Industrial Crystalliza-
 sphericity, dimensionless tion '78 (pp. 249}258). Amsterdam: North Holland.
J
2444 J. Ba!dyga, W. Orciuch / Chemical Engineering Science 56 (2001) 2435}2444

Pope, S. B. (1979). Probability distribution of scalars in turbulent Sreenivasan, K. R. (1995). On the universality of the Kolmogorov
shear #ow. In ¹urbulent shear -ows 2 (pp. 7}16). Berlin: constant. The Physics of Fluids, 7, 2778}2783.
Springer. Sreenivasan, K. R. (1996). The passive scalar spectrum and the Obuk-
Rhodes, R. P. (1975). A probability distribution function for hov}Corrsin constant. The Physics of Fluids, 8, 189}196.
turbulent #ow. In S. N. B. Murthy (Ed.), Turbulent mixing Wei, H., & Garside, J. (1997). Application of CFD modelling to precipi-
in nonreactive and reactive yows (pp. 235}241). New York: Plenum tation systems. Transactions of the Institution of Chemical Engineers,
Press. 75A, 219}227.

Potrebbero piacerti anche