You are on page 1of 318

Chemical Engineering and Processing

Copyright 2004 Elsevier B.V. All rights reserved




Volume 44, Issue 1, Pages 1-137 (January 2005)

1. A new method for producing anhydrous puffed borax
Pages 1-6
mer ahin, Nasrettin Genli and Mustafa zdemir

2. The influence of temperature and inlet velocity on cyclone pressure drop: a CFD study
Pages 7-12
J olius Gimbun, T. G. Chuah, A. Fakhrul-Razi and Thomas S. Y. Choong

3. A pollution reduction methodology in reactor design
Pages 13-21
Qishi Chen and Xiao Feng

4. Effects of hydraulic residence time on metal uptake by activated sludge
Pages 23-32
Tlay A. zbelge, H. nder zbelge and Murat Tursun

5. Porous catalyst intraparticle status of parallel, equilibrium-restrained reactions
Pages 33-39
Guangsuo Yu, Fuchen Wang, Zhenghua Dai and Zunhong Yu

6. PID controller tuning using mathematical programming
Pages 41-49
George Syrcos and Ioannis K. Kookos

7. Separation of acetic acidwater mixtures through acrylonitrile grafted poly(vinyl
alcohol) membranes by pervaporation
Pages 51-58
N. Alghezawi, O. anli, L. Aras and G. Asman

8. Countercurrent flow distribution in structured packing via computed tomography
Pages 59-69
Shaibal Roy, A. Kemoun, M. H. Al-Dahhan, M. P. Dudukovic, Thomas B. Skourlis and Frits M.
Dautzenberg

9. Effect of cycling operations on an adsorbed natural gas storage
Pages 71-79
O. Pupier, V. Goetz and R. Fiscal

10. Effect of internal on the hydrodynamics in external-loop airlift reactors
Pages 81-87
Tongwang Zhang, J infu Wang, Tiefeng Wang, J ing Lin and Yong J in

11. Multicriteria synthesis of flexible heat-exchanger networks with uncertain source-
stream temperatures
Pages 89-100
Cheng-Liang Chen and Ping-Sung Hung

12. On-line dynamic optimization and control strategy for improving the performance of
batch reactors
Pages 101-114
A. Arpornwichanop, P. Kittisupakorn and I. M. Mujtaba


13. Synthesis of nano-sized particles from metal carbonates by the method of reversed
mycelles
Pages 115-119
Christo Karagiozov and Dafina Momchilova

14. Mechanism of mass transfer from bubbles in dispersions: Part II: Mass transfer
coefficients in stirred gasliquid reactor and bubble column
Pages 121-130
V. Linek, M. Korda and T. Moucha

15. Separation of n-hexaneethyl acetate mixtures by azeotropic batch distillation with
heterogeneous entrainers
Pages 131-137
I. Rodriguez-Donis, J . Acosta-Esquijarosa, V. Gerbaud, E. Pardillo-Fondevila and X. J oulia

16. Inside front cover - Editorial Board EDITORIAL BOARD
Pages CO2-CO2



Volume 44, Issue 2, Pages 139-334 (February 2005)
Pneumatic Conveying and Handling of Particulate Solids
Edited by J anos Gyenis and Avi Levy

1. Special issue on conveying and handling of particulate solids
Pages 139-140
J anos Gyenis and Avi Levy

2. A review of the research work of Professor Predrag Marjanovi
Pages 141-151
David Mills

3. Influence of the particle diameter and density in the gas velocity in jet spouted beds
Pages 153-157
Mara J . San J os, Sonia Alvarez, Alvaro Ortiz de Salazar, Martn Olazar and J avier Bilbao

4. Taking-off model of particles with a wide size distribution
Pages 159-166
Isabelle Descamps, J ean-Luc Harion and Bernard Baudoin

5. Solids deposition in low-velocity slug flow pneumatic conveying
Pages 167-173
J . Li, C. Webb, S. S. Pandiella, G. M. Campbell, T. Dyakowski, A. Cowell and D. McGlinchey

6. Identification of material specific attrition mechanisms for polymers in dilute phase
pneumatic conveying
Pages 175-185
Lars Frye and Wolfgang Peukert

7. Two-dimensional numerical simulations of the pneumatic drying in vertical pipes
Pages 187-192
I. Skuratovsky, A. Levy and I. Borde

8. The formation of fine particles by salting-out precipitation
Pages 193-200
J udit Tth, Andrea Kardos-Fodor and Susan Halsz-Pterfi

9. Micronized cocoa butter particles produced by a supercritical process
Pages 201-207
J . -J . Letourneau, S. Vigneau, P. Gonus and J . Fages

10. Food powder handling and processing: Industry problems, knowledge barriers and
research opportunities
Pages 209-214
J ohn J . Fitzpatrick and Lilia Ahrn

11. Microencapsulation of particles using supercritical carbon dioxide
Pages 215-219
H. Krber and U. Teipel

12. Plasma spheroidization of ceramic particles
Pages 221-224
Z. Kroly and J . Szpvlgyi

13. Treatment of particulate metallurgical wastes in thermal plasmas
Pages 225-229
I. Mohai and J . Szpvlgyi
14. Defluidization modelling of pyrolysis of plastics in a conical spouted bed reactor
Pages 231-235
Roberto Aguado, Rubn Prieto, Mara J . San J os, Sonia Alvarez, Martn Olazar and J avier
Bilbao

15. The energy of bed processing during drum granulation
Pages 237-243
Tadeusz Gluba

16. Control of aggregation in production and handling of nanoparticles
Pages 245-252
Wolfgang Peukert, Hans-Christoph Schwarzer and Frank Stenger

17. Effect of different types of impact surface on coal degradation
Pages 253-261
R. K. Sahoo and D. Roach

18. Using statistical moments to describe grinding in a ball mill for industrial-scale
process
Pages 263-266
Andrzej Heim, Tomasz P. Olejnik and Agnieszka Pawlak

19. On possible instability of throughputs in complex milling circuits
Pages 267-272
V. Mizonov, V. Zhukov, A. Korovkin and H. Berthiaux

20. Optimising design of continuous grinding mill-classifier systems
Pages 273-277
P. B. Kis, Cs. Mihlyk and B. G. Lakatos

21. Air classification of solid particles: a review
Pages 279-285
M. Shapiro and V. Galperin

22. Application of a vertical venturi separator for improved recycling of automotive tires
Pages 287-291
W. McBride and S. Keys

23. Particle movement during granular intermingling in a pulsated bottom mixer
Pages 293-296
Mikls Nemnyi and Attila J . Kovcs

24. Estimating the homogenization efficiency of mammoth silos by process dynamics
and simulations: Comparing the results of process dynamics with the simulations
Pages 297-302
D. L. Schott, L. A. van Wijk and W. J . Vlasblom

25. Assessing the homogeneity of powder mixtures by on-line electrical capacitance
Pages 303-313
N. Ehrhardt, M. Montagne, H. Berthiaux, B. Dalloz-Dubrujeaud and C. Gatumel

26. Solid transport in a pyrolysis pilot-scale rotary kiln: preliminary resultsstationary
and dynamic results
Pages 315-321
N. Descoins, J . -L. Dirion and T. Howes

27. Controlling dust emissions and explosion hazards in powder handling plants
Pages 323-326
Peter Wypych, Dave Cook and Paul Cooper

28. Photocatalytic degradation of trichloroethylene (TCE) over TiO2/silica gel in a
circulating fluidized bed (CFB) photoreactor
Pages 327-334
Tak Hyoung Lim and Sang Done Kim

29. Inside front cover - Editorial Board EDITORIAL BOARD
Pages CO2-CO2
Chemical Engineering and Processing 44 (2005) 16
A new method for producing anhydrous puffed borax
mer Sahin

, Nasrettin Genli, Mustafa zdemir


Department of Chemistry, Harran University, S. Urfa, Turkey
Received 30 June 2003; received in revised form 29 July 2003; accepted 8 March 2004
Available online 24 April 2004
Abstract
This paper describes the production of anhydrous puffed borax from borax pentahydrate (BPH) in a batch calcinator. The calcination of
BPH is incomplete since agglomeration starts at 300

C. In order to avoid agglomeration at temperatures higher than 300

C, the surface of
BPH particles have been covered with a CaO layer in aqueous media having a higher melting point than pure BPH. To investigate the effect
of this CaO layer on the calcination of BPH, the samples were fed into the batch calcinator. During the calcination process, quantities such as
the bulk density value, particle size distribution, sodium borate and calcium content of anhydrous borax have been determined as a function
of temperature. It was found that the anhydrous borax can be obtained by calcination of BPH particles covered with CaO in 50, 20 and 5 min
time intervals at a temperature range of 300, 400 and 500

C, respectively. As a result, puffed anhydrous borax of 99% purity with bulk density
of 0.082 g cm
3
and containing 0.670% Ca
2+
has been produced by this method in 5 min interval at 500

C.
2004 Elsevier B.V. All rights reserved.
Keywords: Borax pentahydrate; Anhydrous borax; Calcium oxide; Coating
1. Introduction
Borax pentahydrate (BPH) is one of the most important
commercial boron compounds containing water of crystal-
lization. The structure formula of BPH can be best repre-
sented as Na
2
B
4
O
5
(OH)
4
2.67H
2
O which means that there
are 2 mol of water in the molecular structure, remaining be-
ing real crystal water [1]. BPH is used in many areas, such as
perborate and boric acid production and detergent formula-
tions. However, its water content is not appropriate in some
applications such as the manufacture of high quality glass,
frit production ceramic and the renement of precious met-
als. Thus, BPH should be dehydrated to anhydrous borax
(BA) state.
The production methods of BA from BPH or borax dec-
ahydrate may be categorized into three main groups. These
are: (a) azeotropic distillation, (b) melting and (c) dehydra-
tion in uidized bed.
The method of azeotropic distillation is not yet suitable
for industrial purposes because of low yield and high heat
input needed. In addition, BPH cannot be converted by BA
using this method [2].

Corresponding author. Fax: +90-414-315-1998.


E-mail address: osahin@harran.edu.tr (. Sahin).
With current technology, BA is produced from BPH or
borax decahydrate (BDH) by the melting method which is a
two-stage process: rst dehydration, and then fusion. BPH
or BDH is fused at a temperature higher than the melting
point of BA, ranging from 1000 to 1400

C at different parts
of rebox. The high temperature decreases the viscosity of
the molten mass, in order to produce proper uidization
conditions for ow out from the rebox. The molten borax
is highly corrosive which can only be prevented by forming
a layer of solid calcined borax on the refractory material of
the furnace, thus the melting of borax is the most critical
an the most expensive step, which cause some difculties in
operation and results in somewhat contaminated products.
Slow cooling gives crystal formation but immediate cooling
gives an amorphous glass [3]. By using this process, high
density of BA is produced, but there are many technical
difculties in the process, such as corrosion and handling
problems. The other disadvantage of this method is that the
product needs crushing, grinding and homogenization before
being used.
Production of BA without fusion is very attractive with
respect to energy consumption and corrosion. Thus, the at-
tempts have mostly been in dehydration borax pentahydrate
in a uidized bed via a stage-wise calcination [4,5]. The
main disadvantage of this process is the pufng of the par-
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.004
2 . Sahin et al. / Chemical Engineering and Processing 44 (2005) 16
ticles during removal of water. In a uidized bed, the calci-
nation of BPH cannot exceed 300

C at the earlier stage of


the process. At this temperature, it turns into an amorphous
form of BA and softens easily. Softening of the particles
cause agglomeration, which disturbs the uidization condi-
tion in the uidized bed calcinator [6].
BPH can be calcined to BA without pufng by at least
two-stage calcination which each needs at long time interval
in uidized bed [7]. As a result, the calcination of BPH in
uidized bed have some problemsuch as stage-wise, pufng,
agglomeration and needed long time.
Thus, a considerable number of attempts have been made
to develop a new method for production anhydrous borax
from BPH. As it is known the calcination of BPH cannot
be conducted above agglomeration temperature of 300

C in
uidized bed. At temperatures higher than 300

C, the amor-
phous BA particles which soften readily adhere strongly to
each other since BPH dissolves in their water. This prob-
lem has been solved in our system by covering the particles
with a material that has a higher melting point than the -
nal agglomeration temperature. For this propose, the surface
of BPH was covered with Ca(OH)
2
layer in aqueous me-
dia. The presence of Ca(OH)
2
prevents the agglomeration of
BPH particles up to 500

C, since it controls the acceleration


of water vapor on the surface of BPH particles. Using this
method BPH can be converted to puffed BA containing 99%
Na
2
B
4
O
7
in a 5-min interval at the temperature of 500

C.
The degree of pufng is very high, thus the calcination was
carried out in an oven.
The only disadvantage of this method is that the ob-
tained product (BA) has a lower bulk density. Such a puffed
product, with large quantities of liquids and gas, can be
loaded onto the expanded borax. For example, organic such
as trichloroethylene, cyclohexanone and pentachlorophenol
can be loaded at very high percentages onto the puffed bo-
rax, as can non-ionic and anionic surfactants, with the prod-
uct retaining its free-owing characteristics. This renders
the puffed borax very useful in such diverse compositions
as dairy cleaners, fabric softeners and bath additives [8].
In addition, this puffed BA having large surface area can
be used for solidgas reactions. Recently, one popular appli-
Fig. 1. Schematic diagram of the experimental procedure.
cation of this BA reaction has been the production of sodium
borohydride. Sodium borohydride which is used as a hydro-
gen storage medium can also be synthesized by the reaction
of BA, with MgH
2
through ball milling at room temperature
as shown the following reaction [9]:
8MgH
2
+Na
2
B
4
O
7
+Na
2
CO
3
4NaBH
4
+ 8MgO + CO
2
The obtained anhydrous borax is very fragile, thus it can be
easily converted to powder form by grinding. This operation
increases the bulk density of BA.
2. Experiments
The calcination of BPH to BA was carried out in a batch
calcinator (thermolyne 6000 furnace) which is heated to a
predetermined calcination temperature (Fig. 1). During con-
stant temperature experiments, particles were fed into the
calcinator, which was in thermal equilibrium at the experi-
mental temperature. At a predetermined time interval, sam-
ples were withdrawn using a vacuum sampling tube. Titri-
metric method was used to determine the Na
2
B
4
O
7
content
of sample [10]. The Ca
2+
content of samples was determined
by Jenway PFP 7 model ame photometer. Bulk density of
BA was measured by a standard method [11].
The particle size of BPH used in the experiments was
chosen in the range 630 + 450 m and bulk density was
determined as 0.91 g cm
3
. Technical grade BPH obtained
from ETIHOLDING (Kirka, Turkey) was used in the ex-
periments. Original Na
2
B
4
O
7
content of samples was deter-
mined as 69.95% by titrimetric method [10].
Each series of experiments carried out with 10 g of BPH
washing 20 ml Ca(OH)
2
solution with predetermined con-
centration during 1 min. This washed BPH particles were
then dried at 65

C in an oven for 5 h. In another series of


experiments, a mixture containing BPH and 2, 4, 5, 7, 10,
17 and 34 wt.% CaO were fed into the calcinator at the tem-
perature of 500

C for 10 min.
The other series of experiments, particles covering with
different Ca(OH)
2
fed into the calcinator to investigate the
thermal decomposition behavior in the calcinator at 300,
. Sahin et al. / Chemical Engineering and Processing 44 (2005) 16 3
400, 500

C. At each temperature samples was taken at 3,


5, 10, 20, 45 and 60 min intervals to determine the degree
of calcination.
3. Results and discussion
The main aim of this study is that the calcination of BPH
to BA is performed without agglomeration at higher temper-
ature than 300

C at which dehydration and decomposition


steps of it take place very fast. Normally, particles of BPH
dissolve their water and stick to each other to give agglom-
eration.
In this study, agglomeration was prevented by both cov-
ering the surface of BPH particles in Ca(OH)
2
aqueous so-
lution and BPHCaO mixture in solid state.
Table 1
The effect of different BPHCaO mixture on the calcination of BPH particles at 500

C at the end of 10 min


Particle size (m) %CaO
5 10 17
% fraction Ca
2+
% (g cm
3
) % fraction Ca
2+
% (g cm
3
) % fraction Ca
2+
% (g cm
3
)
+1250 40.517 0.51 0.0583 31.609 0.69 0.0612 39.931 1.40 0.0635
+1000 30.747 0.35 30.172 0.69 16.319 0.38
+800 6.897 0.12 10.632 0.35 6.597 0.12
+630 1.724 0.03 2.010 0.06
+560 0.287
+450
+315
315 20.115 4.05 25.287 5.21 18.942 10.16
Fig. 2. TG curves for BPH both in pure and containing 2.46 wt.% Ca
2+
.
In the rst group of experimental work, mixtures contain-
ing BPH and 2, 4, 5, 7, 10, 17 and 34 wt.% CaO were fed
into the batch calcinator for 10 min at 500

C. At the end
of this kind of calcination, the conversion of BPH to BA is
found as about 99.6%. But, the BPH samples containing 2
and 4 wt.% CaO agglomerated slightly. However, agglomer-
ation was not observed at the concentration of CaO higher
than 5 wt.%. Table 1 shows the values found at the end of
this group. As can be seen from this table, the maximum
content of Ca
2+
in BA particle was found in the particle size
smaller than 315 m. The particle size of BA is expanded
about twice higher than used for BPH particles at the begin-
ning of experiment by thermal shock. The result obtained
at the end of this step shows that the agglomeration can be
prevented in presence of CaO. However, the obtained prod-
uct (BA) at the end of this kind of operation contains high
4 . Sahin et al. / Chemical Engineering and Processing 44 (2005) 16
70
75
80
85
90
95
100
0 10 20 30 40
time(min.)
%
N
a
2
B
4
O
7
50
300 C
400 C
500 C
Fig. 3. Changes of calcination time with temperature for the BPH samples containing about 2.5 wt.% Ca
2+
.
Table 2
Properties of dehydrated BPH covering with 2.5 wt.% Ca
2+
layer
Particle size (m) 300

C 400

C 500

C
% fraction Ca
2+
% (g cm
3
) % fraction Ca
2+
% (g cm
3
) % fraction Ca
2+
% (g cm
3
)
+1250 96.407 1.01 0.0552 90.998 1.00 0.0312 81.344 1.00 0.0589
+1000 8.927
+800 0.638
+630
+560
+450
+315
315 3.593 1.88 9.002 1.361 9.009 1.50
values of CaO which is not suitable for industrial applica-
tions.
In the second group of experiment, the surface of BPH
particle were covered with Ca(OH)
2
in aqueous solution.
TG analysis of the BPH both in pure state and covered with
2.46 wt.% Ca
2+
were performed under nitrogen atmosphere
at 10

C/min heating rate. Fig. 2 shows the results of TG


analysis. As can be seen from Fig. 2, the covered BPH par-
ticles with Ca(OH)
2
dehydrate slowly with respect to pure
state. This phenomena involves the simultaneous transfer of
heat to evaporate the liquid and transfer of vapor within the
solid and vapor from the surface into the hot carrier gas. In
the case of pure BPH at temperature higher than 300

C, all
Table 3
Properties of dehydrated BPH particles containing about 0.670 wt.% Ca
2+
Particle size (m) 300

C 400

C 500

C
% fraction Ca
2+
% (g cm
3
) % fraction Ca
2+
% (g cm
3
) % fraction Ca
2+
% (g cm
3
)
+1250 33.463 0.201 0.0727 61.318 0.170 0.044 28.940 0.210 0.0824
+1000 39.779 26.689 38.607
+800 18.739 5.574 22.708
+630 4.013 3.482
+560
+450
+315
315 4.006 0.467 6.419 0.498 6.263 0.471
crystal water and some structure water dehydrated quickly
caused agglomeration. Thus, the mass transfer in this step
looked as the constant rate period where moisture movement
within the solid is sufciently rapid to maintain a saturated
condition at the surface. In the case of dehydration of BPH
particles covered with Ca(OH)
2
layer to increase the mass
transfer resistance between the particles and hot carrier gas,
the calcination is mainly controlled by the Ca(OH)
2
layer.
Fig. 3 illustrates the Na
2
B
4
O
7
content of the samples
containing about 2.50 wt.% Ca
2+
taken from the batch cal-
cinator at different stages during isothermal decomposition
conducted at various temperatures. The total Na
2
B
4
O
7
con-
tent of samples increased with increasing calcination tem-
. Sahin et al. / Chemical Engineering and Processing 44 (2005) 16 5
peratures. The BA contains some CaO, hence the conver-
sion percentages of BA never attained 100% as seen from
Fig. 3. The conversion of BPH to BA takes place in two
stages namely dehydration and calcination [7]. But dehy-
dration stage cannot be observed in Fig. 3 because of rapid
dehydration of BPH. Fig. 3 illustrates that the calcination
of BPH to BA at 300

C completed in 50 min, whereas the


same degree of calcination realized in 5 min at the temper-
ature of 500

C. The Ca
2+
content, bulk density and sieve
analysis of BA obtained at the end of this group of exper-
iments are illustrated in Table 2. As can be seen from this
table, the Ca
2+
content of calcination realized at 300, 400
and 500

C is about 2.5 Ca
2+
wt.%. The chosen particle size
range of BPH before calcination was 630 + 450 m and
the particle size of almost nine to ten part of calcines was
increased to +1210 m shown in Table 2. Thus, it can be
said that pufng is more effective in cause size increment
with respect to shrinking and fragmentation of puffed parti-
cles. The operation of covering BPH particles with Ca(OH)
2
can not be attained at the temperature higher than 500

C
because of agglomeration. It is interesting to see that the
bulk density increased from 0.0317 to 0.0599 g cm
3
with
increasing the temperature from 400 to 500

C. This behav-
ior of BPH can be easily explained by increasing sintering
effect which depends on the increasing temperature. In the
light of above results, it can also be concluded that better
operation could be possible from the second group of ex-
periment by decreasing the Ca
2+
layer on the surface of
BPH particles. Thus, a third group of experiments were car-
ried out using very low Ca(OH)
2
aqueous solution to cover
the BPH particles surface. The changes of Na
2
B
4
O
7
con-
tent of samples containing about 0.60 wt.% Ca
2+
with time
in this group are given in Fig. 4 at various constant temper-
atures. The Na
2
B
4
O
7
content of the samples increases with
increasing calcination temperature. The calcination rate of
this group is higher than the previous group given in Fig. 3.
This behavior may be attributed to the fact that increas-
ing the percentages of additive caused the formation of a
70
75
80
85
90
95
100
0 10 20 30 40
time (min.)
%
N
a
2
B
4
O
7
50
300 C
400 C
500 C
Fig. 4. Effect of temperature on the calcination of BPH samples containing
about 0.67 wt.% Ca
2+
.
thick and homogenous cover on the surface of the anhy-
drous borax. The thickness of CaO layer controls the trans-
fer rate of liberated water from borax pentahydrate to air.
The sieve analysis, bulk density and Ca
2+
content of BA
obtained at the end of operation having different tempera-
Fig. 5. Microscopic photographs of borax pentahydrate: (a) pure, (b)
containing 0.67 wt.% Ca
2+
and (c) containing 2.5 wt.% Ca
2+
.
6 . Sahin et al. / Chemical Engineering and Processing 44 (2005) 16
tures are given in Table 3. As can be seen in this table, parti-
cle size shows a distribution which is attributed to the Ca
2+
content and temperature. In addition, the particles smaller
than 315 m contain twice higher Ca
2+
than all other parti-
cle sizes contained. Microscopic photographs of borax pen-
tahydrate in pure, containing 0.67 wt. and 2.5 wt.% Ca
2+
are given in Fig. 5, respectively. In pure state, the surface
of BPH particles are smooth whereas the quality of par-
ticles are decreased with increasing Ca
2+
quantity on the
surface.
4. Conclusion
In order to obtain anhydrous borax with low bulk den-
sity by calcination in batch calcinator, particular attention is
paid to cover BPH surface with Ca(OH)
2
layer. The follow-
ing conclusion can be withdrawn from the results obtained
during the calcination of covered BPH particles in a batch
calcinator:
(a) The calcination of BPH with single stage cannot be
carried out at the temperature higher than 300

C since
agglomeration is started. The calcination of BPH to BA
also takes longer at around 300

C.
(b) BPH can be transformed to BA without agglomeration
by mixing CaO with BPH in solid state at temperature
range 300500

C. In this case, the content of CaO in


BA must be higher than 7%.
(c) Covering the BPH particles with Ca(OH)
2
aqueous
solution also prevented the agglomeration in tempera-
ture range 300500

C. In this kind of operation, the


BPH particle surface was covered with a thin layer of
Ca(OH)
2
to control the transport rate of liberated water
from inside BPH particle to air. By this method, the
content of BPH can be decreased up to about 0.68 wt.%
Ca
2+
.
(d) Calcination of BPH covered with Ca(OH)
2
to BA re-
alized in 50, 20 and 5 min intervals at 300, 400 and
500

C, respectively. This result shows that at temper-


atures higher than 400

C, the puffed anhydrous borax


can be obtained by continuous operation. Also, the high
bulk density may be obtained by crashing puffed borax.
(e) The obtained low bulk density puffed borax is preferred
priority in some chemical process such as solid state
reaction, adsorption and high temperature application
of BA.
References
[1] R.P. Douglas, F.G. Donald, Renement of the structure of tincal-
conite, Acta Cryst. C47 (1991) 22792282.
[2] G. Nencetti, A. Pennacchi, Chim. Ind. Milan 46 (5) (1968) 518525.
[3] Kirk-Othmer, Enc. Chem. Thech., 4 ed., vol. 4, 1992, p. 388.
[4] O. Sahin, A.N. Bulutcu, Turk. J. Chem. 26 (2002) 8996.
[5] S. Kocakusak, K. Akcay, T. Ayok, H.J. Koroglu, O.T. Savasc, R.
Tolun, Ind. Eng. Chem. Res. 35 (1996) 14241428.
[6] O. Sahin, U.G. Beker, A.N. Bulutucu, Int. J. Storing Handl. Process.
Powder 7 (1995) 165167.
[7] O. Sahin, A.N. Bulutucu, Chem. Eng. Process. 41 (2002) 135141.
[8] T.E. Raymond, US Patent 4 412 978 (1983).
[9] Z.P. Li, N. Morigazaki, B.H. Liu, S. Suda, J. Alloys Compd. 349
(2003) 232236.
[10] D.F. Snell, C.L. Hilton, Encyclopedia of Industrial Chemical Anal-
ysis, vol. 7, Wiley, New York, 1968, pp. 373384.
[11] ISO 3424, Sodium perborate for industrial use-determination of bulk
density.
Chemical Engineering and Processing 44 (2005) 712
The inuence of temperature and inlet velocity on cyclone pressure drop:
a CFD study
Jolius Gimbun

, T.G. Chuah, A. Fakhrul-Razi, Thomas S.Y. Choong


Department of Chemical and Environmental Engineering, Faculty of Engineering, Universiti Putra Malaysia 43400 UPM Serdang, Selangor D. E., Malaysia
Received 16 February 2004; received in revised form 22 March 2004; accepted 22 March 2004
Available online 18 May 2004
Abstract
This work presents a computational uid dynamics (CFD) calculation to predict and to evaluate the effects of temperature and inlet velocity
on the pressure drop of gas cyclones. The numerical solutions were carried out using spreadsheet and commercial CFD code Fluent 6.1. This
paper also reviews four empirical models for the prediction of cyclone pressure drop, namely [Air pollution control: a design approach, in: C.
David Cooper, F.C. Alley (Eds.), Cyclones, second ed., Woveland Press Inc., Illinois, 1939, p. 127139] [Chem. Eng. (1983) 99] [Doctoral
Thesis, Havarad University, USA, 1988], and [Chem. Eng. Progress (1993) 51]. All the predictions proved to be satisfactory when compared
with the presented experimental data. The CFD simulations predict excellently the cyclone pressure drop under different temperature and inlet
velocity with a maximum deviation of 3% from the experimental data. Specically, results obtained from the computer modelling exercise
have demonstrated that CFD is a best method of modelling the cyclones operating pressure drop.
2004 Elsevier B.V. All rights reserved.
Keywords: Cyclone; CFD; Pressure drop; Temperature; Inlet velocity
1. Introduction
Cyclones are devices that employ a centrifugal force
generated by a spinning gas stream to separate particles
from the carrier gas. Their simple design, low capital cost
and nearly maintenance-free operation make them ideal for
use as pre-cleaners for more expensive nal control devices
such as baghouses or electrostatic precipitators. Cyclones
are particularly well suited for high temperature and pres-
sure conditions because of their rugged design and exible
components materials. Cyclone collection efciencies can
reach 99% for particles bigger than 5 m [12], and can be
operated at very high dust loading. Cyclones are used for
the removal of large particles for both air pollution control
and process use. Application in extreme condition includes
the removing of coal dust in power plant, and the use as a
spray dryer or gasication reactor.
Engineers are generally interested in two parameters in
order to carry out an assessment of the design and perfor-
mance of a cyclone. These parameters are the collection

Corresponding author. Tel.: +60-19-248-9101; fax: +60-38946-7120.


E-mail address: jolius21@yahoo.co.uk (J. Gimbun).
efciency of particle and pressure drop through the cy-
clone. An accurate prediction of cyclone pressure drop is
very important because it relates directly to operating costs.
Higher inlet velocities give higher collection efciencies for
a given cyclone, but this also increases the pressure drop
across the cyclone. Therefore, a trade off must be made
between higher collection efciency and low pressure drop
across the cyclone. Computational uid dynamics (CFD)
has a great potential to predict the ow eld characteristics
and particle trajectories inside the cyclone as well as the
pressure drop [8]. The complicated swirling turbulent ow
in a cyclone places great demands on the numerical tech-
niques and the turbulence models employed in the CFD
codes when modelling the cyclone pressure drop.
In this study, pressure drop calculations are performed us-
ing CFD and compared with four empirical model of Shep-
herd and Lapple [11], Casal and Martinez [3], Dirgo [5],
and Coker [4]. These four empirical models and CFD pre-
diction are compared with the experimental data presented
in the literature. In this study, the CFD calculations are
carried out using commercial nite volume code Fluent 6.1
and the empirical models are performed in Microsoft Excel
spreadsheet.
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.005
8 J. Gimbun et al. / Chemical Engineering and Processing 44 (2005) 712
D
b De
a S
h
H
B ELEVATION PLAN
Fig. 1. Tangential cyclone conguration.
2. Cyclone design
There are a number of different forms of cyclone but
the reverse ow cyclone represented in Fig. 1 is the most
common design used in the industry. The cyclone consists of
four main parts: the inlet, the separation chamber, the dust
chamber and the vortex nder. Tangential inlets are preferred
for the separation of solid particles from gases [1]. In this
study, the numerical simulation deals with the standard case
of reverse ow cyclone with a tangential rectangular inlet.
Cyclone dimension used in this simulation are as shown in
Table 1.
3. Computational uid dynamics approach
Fluent is a commercially available CFD code which
utilises the nite volume formulation to carry out coupled
or segregated calculations (with reference to the conserva-
tion of mass, momentum and energy equations). It is ideally
suited for incompressible to mildly compressible ows.
The conservation of mass, momentum and energy in uid
ows are expressed in terms of non-linear partial differen-
tial equations which defy solution by analytical means. The
solution of these equations has been made possible by the
advent of powerful workstations, opening avenues towards
the calculation of complicated ow elds with relative ease.
For the turbulent ow in cyclones, the key to the success
of CFD lies with the accurate description of the turbulent
behaviour of the ow [8]. To model the swirling turbulent
Table 1
Cyclone geometry used in this simulations
Geometry a/D b/D D
e
/D S/D h/D H/D B/D D
a
Stairmand high efciency 0.5 0.2 0.5 0.5 1.5 4 0.375 0.305
Bohnet [2] 0.533 0.133 0.333 0.733 0.693 2.58 0.333 0.15
a
Unit in meters.
ow in a cyclone separator, there are a number of turbulence
models available in Fluent. These range from the standard
k model to the more complicated Reynolds stress model
(RSM). The k model involves the solution of transport
equations for the kinetic energy of turbulence and its dis-
sipation rate and the calculation of a turbulent contribution
to the viscosity at each computational cell. The standard
k, RNG k and realizable k model was not optimized
for strongly swirling ows found for example in cyclones
[10,6]. Turbulence may be stabilised or destabilised in the
parts of ow domain where strong streamline curvature is
presence. However to reduce the computational effort the
RNG k model can be used with about 12% deviation on
experimental data [8]. The numerical studies carried out by
Fredriksson [7] reveal that the RNG k model under pre-
dicts the variation of the axial velocity prole across the
radial direction and also over predicts the magnitude of the
tangential velocity and the cyclone pressure drop.
The Reynolds stress model requires the solution of trans-
port equations for each of the Reynolds stress components
as well as for dissipation transport without the necessity to
calculate an isotropic turbulent viscosity eld. The Reynolds
stress turbulence model yield an accurate prediction on swirl
ow pattern, axial velocity, tangential velocity and pressure
drop on cyclone simulation [7,6,13,10].
The nite volume methods have been used to discretised
the partial differential equations of the model using the Sim-
ple method for pressurevelocity coupling and the second
order upwind scheme to interpolate the variables on the sur-
face of the control volume. The segregated solution algo-
rithm was selected. The Reynolds stress turbulence model
was used in this model due to the anisotropic nature of
the turbulence in cyclones. Standard uent wall functions
were applied and high order discretisation schemes were
also used.
Under the RSM second order upwind for discretisation
there is a difculty to reach the convergence in simulation.
The residuals may exhibit cyclic tendencies which mean
that the transient pattern occurs. In this instance, the solver
must be changed to a transient solver and makes the time
step something in the region of 0.025 s or a tiny fraction of
the residence time of the cyclone. The simulation is then
solved with a coupling of unsteady and steady state solver
in Fluent. For the simulation using RNG k model the
steady state solver is sufcient to reach the convergence. The
CFD simulation was performed with a Pentium IV 2.8 GHz
HP workstation XW8000 with 512 cache-memory, 1 GB
RAM-memory, and 110 GB hard disc memory.
J. Gimbun et al. / Chemical Engineering and Processing 44 (2005) 712 9
4. Pressure drop empirical models
The pressure drop across the cyclone is an important pa-
rameter in the evaluation of cyclone performance. It is a
measure of the amount of work that is required to operate the
cyclone at given conditions, which is important for opera-
tional and economical reasons. The total pressure drop over
a cyclone consists of losses at the inlet, outlet and within
the cyclone body. The main part of the pressure drop, i.e.
about 80%, is considered to be pressure losses inside the cy-
clone due to the energy dissipation by the viscous stress of
the turbulent rotational ow [9]. The remaining 20% of the
pressure drop are caused by the contraction of the uid ow
at the outlet, expansion at the inlet and by uid friction on
the cyclone wall surface.
In this study, four empirical models in the literature have
been chosen to predict the pressure drop over a cyclone,
namely Shepherd and Lapple [11], Casal and Martinez [3],
Dirgo [5], and Coker [4]. In these four models, the total
pressure drop in cyclone is either assumed equal to the static
pressure drop or as a function of cyclone dimension and
pressure drop coefcient. Generally cyclone pressure drop
is proportional to the velocity head and can be written in the
form of
P =

g
v
2
i
2
(1)
In the Shepherd and Lapple [11] model, is obtained by
assuming static pressure drop given as
= 16
ab
D
2
e
(2)
In Casal and Martinez [3], is derived from the statistical
analysis on experimental data given as
= 11.3

ab
D
2
e

2
+3.33 (3)
In Dirgo [5] model, is a function of cyclone dimension
given as
= 20

ab
D
2
e

S/D
(H/D)(h/D)(B/D)

1/3
(4)
In Coker [4], is given as
= 9.47
ab
D
2
e
(5)
5. Result and discussion
5.1. Pressure drop prediction under different inlet velocity
Measurement of the cyclone pressure drop was carried out
for inlet velocity ranging from 4.62 to 14.16 m/s by Bohnet
[2], and from 5.1 to 25 m/s by Grifths and Boysan [8].
The numerical calculation was made with a ne numerical
Fig. 2. CFD surface mesh for (A) Stairmand high efciency, and (B)
Bohnet [2] cyclone.
grid as shown in Fig. 2. Several empirical correlation from
literature, Shepherd and Lapple [11], Casal and Martinez [3],
Dirgo [5] and Coker [4], were also considered to compared
experimental data and numerical solution from Fluent code.
Figs. 3 and 5 present the comparison. The three-dimension
map of static pressure of Bohnet and Stairmand cyclones is
shown in Figs. 4 and 6, respectively.
5.2. Pressure drop prediction under different
operating temperature
Measurement of the cyclone pressure drop of different op-
erating temperature was carried out for temperature ranging
from 293 to 1123 K by Bohnet [2]. The comparison between
the Bohnet experiment, empirical model and CFD prediction
is shown in Figs. 7 and 8. Fig. 9 shows the three-dimension
map of static pressure for operating temperature of 950 K.
The calculated static pressure drop of cyclone between
inlet and outlet for the different numerical model is shown
in Figs. 3, 5, 7 and 8. It is shown that good agreement of
0
500
1000
1500
2000
2500
4 6 8 10 12 14
Inlet gasvelocity (m/s)
P
r
e
s
s
u
r
e

D
r
o
p

(
P
a
)
16
Dirgo
Shepherd & Lapple
Coker
Casal & Martinez
CFD RSM
CFD RNGk-
Fig. 3. Evolution of pressure drop with inlet velocity. Comparison between
data presented by Bohnet [2], the predictions of CFD and four empirical
models (P = 1 bar, T = 293 K, D = 150 mm, geometry Bohnet [2]).
10 J. Gimbun et al. / Chemical Engineering and Processing 44 (2005) 712
Fig. 4. Evolution of pressure drop with inlet velocity. Comparison between
data presented by Grafths and Boysan [8], the predictions of CFD and
four empirical models (P = 1 bar, T = 293 K, D = 0.305 m, geometry
Stairmand high efciency.
0
500
1000
1500
2000
2500
5 10 15 20
Velocity (m/s)
P
r
e
s
s
u
r
e

D
r
o
p

(
P
a
)
25
Shepherd & Lapple
Dirgo
Coker
Casal & Martinez
CFD RSM
CFD RNG k-
Fig. 5. 2D and 3D map of static pressure of Bohnet [2] cyclone for inlet
velocity of 4.62 m/s and temperature 293 K.
Fig. 6. 3D map of static pressure of Stairmand cyclone for inlet velocity
of 20 m/s and temperature 293 K.
0
500
1000
1500
2000
2500
0 200 400 600 800 1000 1200
Temperature (K)
P
r
e
s
s
u
r
e

D
r
o
p

(
P
a
)
Dirgo
Coker
Casal & Martinez
CFD RSM
Shepherd & Lapple
CFD RNGk-
Fig. 7. Evolution of pressure drop with operating temperature. Compar-
ison between data presented by Bohnet [2], the predictions of CFD and
four empirical models (Q = 100 m
3
/h, T = 2931123 K, D = 150 mm,
geometry Bohnet [2].
0
200
400
600
800
1000
1200
1400
1600
0 200 400 600 800 1000 1200
Temperature (K)
P
r
e
s
s
u
r
e

D
r
o
p

(
P
a
)
Dirgo
Coker
Casal & Martinez
CFD RSM
Shepherd & Lapple
CFD RNGk-
Fig. 8. Evolution of pressure drop with operating temperature. Compar-
ison between data presented by Bohnet [2], the predictions of CFD and
four empirical models (Q = 80 m
3
/h, T = 2931123 K, D = 150 mm,
geometry Bohnet [2].
Fig. 9. 3D Map of static pressure of Bohnet [2] cyclone for inlet velocity
of 11.48 m/s and temperature 850 K.
J. Gimbun et al. / Chemical Engineering and Processing 44 (2005) 712 11
the CFD numerical calculation when compared with experi-
mental data, and predictions from empirical correlation. The
results show that the CFD prediction by using the Fluent
code can be used for pressure drop evaluation in cyclone
design. This low-pressure centre can be responsible for the
ow reversion and deviation of the axial velocity peak to
the wall of the vortex nder pipe as showed in Figs. 4, 6
and 9.
The Fluent code with the RSM turbulence model, predict
very well the pressure drop in cyclones and can be used in
cyclone design for any operational conditions (Figs. 3, 5,
7 and 8). In the CFD numerical calculations a very small
pressure drop deviation were observed, with less than 3%
of deviation at different inlet velocity which probably in
the same magnitude of the experimental error. The CFD
simulations with RNG k turbulence model still yield a
reasonably good prediction (Figs. 3, 5, 7 and 8) with the
deviation about 1420% of an experimental data. It consid-
erably tolerable since the RNG k model is much less on
computational time required compared to the complicated
RSM turbulence model. In all cases of the simulation the
RNG k model considerably underestimates the cyclone
pressure drop as revealed by Grifths and Boysan [8].
However under extreme temperature (>850 K) there is no
signicant difference between RNG k and RSM model
prediction.
The cyclone pressure drop can be rewritten as a func-
tion of inlet velocity head. The empirical model used for
the prediction of pressure drop is much depends on the cy-
clone operating condition. Shepheard and Lapple [11] and
Dirgo (1990) model show a good prediction on cyclone pres-
sure drop under different operational inlet velocity (Figs. 3
and 4), the prediction within 620% of the measured value.
However, Dirgos model does not take into account temper-
ature in its model: its predictions are, therefore, not reliable
under different operating temperature (Figs. 7 and 8). Under
high temperature Dirgos model considerably overestimates
the cyclone pressure drop with relative error of more than
90%.
The pressure drop decreases signicantly with rising
temperature. This effect is mainly due to the decrease of the
density and the increase of the viscosity of the gas. Accord-
ing to Figs. 7 and 8, the models of Shepheard and Lapple
give quite a good approximation of the pressure drop with
an error in the prediction of about 37%. The model of Casal
and Martinez, and Coker were under predicts the cyclone
pressure drop under different operating temperature with
relative error of 72 and 52%, respectively. Since Casal and
Martinez, and Coker models consistently underestimate the
cyclone pressure drop in all the conditions studied, they are
therefore not particularly useful for design purposes. It is
always more practical to design for a larger pressure drop
than for a smaller one. In overall, the cyclone pressure drop
is somewhat closer to Coker model for the operating tem-
perature more than 500 K and close to Shepherd and Lapple
model for the operating temperature lower than 500 K.
6. Conclusions
The CFD code FLUENT with the RSM turbulence model,
predict very well the pressure drop in cyclones and can be
used in cyclone design for any operating conditions. In the
CFD numerical calculations a very small pressure drop de-
viation were observed, with about 3% of deviation, probably
in the same magnitude of the experimental error. However
behind the accuracy of the complicated RSM model it does
require much expensive computational effort compared to
the RNG k model. CFD with RNG k turbulence model
still yield a reasonably good prediction on cyclone pressure
drop with deviation of 1418% on measured value.
The cyclone pressure drop can be rewritten as a function
of inlet velocity head. The model used for the prediction of
pressure drop depends on the cyclone operating condition.
Both Shepherd and Lapple, and Dirgo models show a good
prediction on cyclone pressure drop under different opera-
tional inlet velocity. However, Dirgos model is unable to
predict accurately the pressure drop under different oper-
ating temperature. For the various temperature conditions,
Shepherd and Lapples pressure drop model prediction is
the best. We therefore, conclude that the Shepherd and Lap-
ple model should be used for estimation of pressure drop
in cyclone design.
Acknowledgements
The authors would like to thank Dr. Tom Fraser, Fluent
India and Fluent Europe UK for their guidance and support.
The authors are grateful to the referees for their useful com-
ments.
Appendix A. Nomenclature
a cyclone inlet height (m)
b cyclone inlet width (m)
B cyclone dust outlet diameter (m)
D cyclone body diameter (m)
D
e
cyclone gas outlet diameter (m)
h cyclone cylinder height (m)
H cyclone height (m)
P cyclone pressure drop (Pa)
S cyclone gas outlet duct length (m)
v
i
inlet velocity (m/s)
Greek letters
velocity head, pressure drop coefcient (m)

g
gas density (kg/m
3
)
References
[1] S. Altmeyer, V. Mathieu, S. Jullemier, P. Contal, N. Midoux, S. Rode,
J.-P. Leclerc, Comparison of different models of cyclone prediction
12 J. Gimbun et al. / Chemical Engineering and Processing 44 (2005) 712
performance for various operating conditions using a general soft-
ware, Chem. Eng. Process. 43 (2004) 511522.
[2] M. Bohnet, Inuence of the gas temperature on the separation ef-
ciency of aerocyclones, Chem. Eng. Process. 34 (1995) 151156.
[3] J. Casal, J.M. Martinez, A better way to calculate cyclone pressure
drop, Chem. Eng. 90 (1983) 99.
[4] A.K. Coker, Understand cyclone design, Chem. Eng. Progr. 28 (1993)
5155.
[5] J. Dirgo, Relationships between cyclone dimensions and performance.
Doctoral Thesis, Havarad University, USA, 1988.
[6] T. Fraser, personal communication, 2003. fraserta1@cf.ac.uk, http://
www.cfd-online.com.
[7] C. Fredriksson, Exploratory experimental and theoretical studies of
cyclone gasication of wood powder. Doctoral Thesis, Lulea Uni-
versity of Technology, Sweden, 2003.
[8] W.D. Grifths, F. Boysan, Computational uid dynamics (CFD) and
empirical modelling of the performance of a number of cyclone
samplers, J. Aerosol Sci. 27 (1996) 281304.
[9] A. Ogawa, Separation of particles from air and gasses, vols. 1 and
2, CRC Press, Boca Raton, Florida, USA, 1984.
[10] M. Reddy, Fluent India, personal communication, 2003, mxr@
uent.co.in.
[11] C.B. Shepherd, C.E. Lapple, Air pollution control: a design approach.
In: C.D Cooper, F.C. Alley (Eds.), Cyclones, second ed., Woveland
Press Inc., Illinois, 1939, pp. 127139.
[12] P.D. Silva, C. Briens, A. Bernis, Development of a new rapid method
to measure erosion rates in laboratory and pilot plant cyclones,
Powder Technol. 131 (2003) 111119.
[13] M. Slack, Cyclonic separator, QNETCFD application Challenge,
http://www.qnet-cfd.net, Accessed on 8th August 2003.
Chemical Engineering and Processing 44 (2005) 1321
A pollution reduction methodology in reactor design
Qishi Chen, Xiao Feng

Department of Chemical Engineering, State Key Laboratory of Multi-Phase Flow in Power Engineering, Xian Jiaotong University, Xian 710049, China
Received 20 January 2003; received in revised form 18 March 2004; accepted 26 March 2004
Available online 18 May 2004
Abstract
An algorithm for waste and pollutants reduction in reactor design is presented in this paper. This algorithm use potential environmental
impact balance (PEI) and PEI rate-law expression to track the generated PEI throughout reaction process within a reactor; and study how
reaction conditions (temperature, pressure, concentration, etc.) and various engineering factors (heat and mass transfer, and back-mixing and
so on) affect process environmental performance. The form of PEI rate-law expression should be consistent with the method for calculating
the overall PEI of mixtures of chemicals inside reactors. The algorithm can be used as a tool to aid in designing chemical reactors with
environmentally friendliness processes. The use of the methodology is illustrated with the reaction system of allyl chloride production.
2004 Elsevier B.V. All rights reserved.
Keywords: Reactor design; Potential environmental impact; Chemical reaction engineering; Impact analysis
1. Introduction
There is currently a great deal of interest in the devel-
opment of methods that can be used to prevent or at least
reduce the generation of pollution, because it may result in
lower operating costs due to better utilization of raw mate-
rials and energy and reduced waste treatment and disposal
costs. Researches have shown that the most opportunities
of environmental impact minimization exist in the process
research and development and design stage of chemical pro-
cesses [1,2]. Waste and pollutant generation in many chemi-
cal manufacturing process can be traced back to the reaction
system, where reaction pathways that produce high yields
may also generate toxic byproducts or use environmentally
unfriendly raw materials, solvents, and catalystswhich
must be dealt with downstream at a signicant cost [3];
and the size, conguration, and composition of the reactor
are very important factors controlling the greenness of a
process [4]. Therefore, the design of new reactor is a very
important element in preventing pollution at the source.
Abbreviations: CSTR, continuous stirred-tank reactor; EIM, environ-
mental impact minimization; NP, non-products; PEI, potential environ-
mental impact; PFR, plug-ow reactor

Corresponding author. Tel.: +86-29-82668980;


fax: +86-29-83237910.
E-mail address: xfeng@mail.xjtu.edu.cn (X. Feng).
To address the ideal including environmental impact con-
siderations into process design, Cabezas et al. [5] introduced
a potential environmental impact (PEI) balance as an amend-
ment of the Waste Reduction Algorithm [6]. However, this
algorithm is simply a tool to be used to aid in evaluating the
environmental friendliness of a process [7].
Making clear the effects of reaction conditions such
as temperature, pressure and concentration as well as en-
gineering factors involving back-mixing, heat and mass
transfer and so forth, on the performances of reaction pro-
cess is one of critical bases for benign reactor design. In
this paper, potential environmental impact balance and PEI
rate-law expression are proposed as a tool to be used to
effectively discover these effects on environmental perfor-
mance of a reaction process. With this methodology, the
PEI of waste and pollutants could be tracked throughout
reaction process within reactors, and appropriate reactor
type and reaction conditions, which could lead to envi-
ronmentally benign reaction processes, could be selected
expediently.
2. The potential environmental impact of a
chemical process
In the algorithm [59], the potential environmental im-
pact of a certain quantity of a certain material and energy
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.006
14 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321
is dened as the effect that this material and energy would
have on the environment if they were to be emitted into the
environment. For a steady-state process, the balance equa-
tion is as follows:

I
in


I
out
+

I
gen
= 0 (1)
where

I
in
and

I
out
are the input and output rates of PEI,
respectively, and

I
gen
denotes the PEI generation rate inside
the processes.

I
in
=

j
m
(in)
j

k
x
kj

k
+ (2)

I
out
=

j
m
(out)
j

k
x
kj

k
+ (3)
where m
(in)
j
is the mass ow rate of input stream j, m
(out)
j
the mass ow rate of output stream j, x
kj
the mass fraction
of chemical k in stream j, and
k
the overall potential envi-
ronmental impact (PEI) of chemical k.

k
=

s
kl
(4)
where
s
kl
is the specic potential environmental impact of
chemical k for environmental impact category l, in units of
PEI/kg of chemical k. Details of the method for calculat-
ing
s
kl
appeared in reference [7]. The impact categories in-
volved in this theory fall into two general area concerning
with four categories in each area:
(1) The global atmospheric categories:
Global warming potential (GWP).
Ozone depletion potential (ODP).
Acidication and acid-rain potential (AP).
Photochemical oxidation or smog formation potential
(PCOP).
(2) The local toxicological impact categories:
Human toxicity potential by ingestion (HTPI).
Human toxicity potential by either inhalation or der-
mal exposure (HTPE).
Aquatic toxicity potential (ATP).
Terrestrial toxicity potential (TTP).
In Eq. (4),
l
is a relative weighting factor for impact type
l independent of chemical k, and represent the value that
society places on particular types of environmental impact
[8]. This factor allows Eq. (4) to be customized to specic
or local conditions. The suggested procedure is to initially
set all
l
to a value of one, and allow users to vary indi-
vidual
l
from 1 to 10 according to local needs and polices
[5,8].
3. The algorithm
In order to perform waste reduction and pollution pre-
vention in reactor design, studying the effects of reaction
conditions and engineering factors and the like on the gen-
eration of waste or pollutant is very important. However,
studying each of wastes and pollutants within reactor is
time-consuming and cost-consuming and the results may be
conicting or competing, that is to say that reducing one pol-
lutant or waste may increasing the others. In addition, this
research method does not address the impact of the pollu-
tion generated within a reactor. For example, reactor A may
produce 50 kg/h of pollutants while reactor B may produce
100 kg/h. However, the pollutants generated within reactor
A may be much more environmentally unfriendly than those
generated within reactor B. Thus, the strategies for reduc-
ing pollutants and wastes generated from reaction process
should be studied through their PEI rather than their mass
or concentrate, the results would be, then, scientic and be-
yond conicting.
Traditionally, the foundation of reactor design is the equa-
tions such as reaction rate equation, mass, energy and mo-
mentum balances as well as the model describing the uid
ow within reactors. With these equations, reactor designers
can track compositions of chemicals, pressure and tempera-
ture throughout reaction process and discover the effects of
reaction conditions as well as engineering factors on these
parameters, and then, an appropriate reactor with protable
processes can be designed. Nevertheless, this kind of design
method does not consider environmental perspective of pro-
cesses effectively.
If the PEI generated within reactors could be tracked
throughout the process, the inuence of reaction conditions
and engineering factors on the generated PEI could also
be claried, and then reactors that generate the minimum
amount of PEI would be easily obtained. The above thinking
can be implemented by using the PEI balance and the PEI
rate-law expression in reactor design. In this way, the design
method is similar to the traditional reactor design method
based on balance equation and reaction rate equation.
3.1. PEI rate-law expression
The PEI transformation rate of chemical k can be dened
as:
r
PEI,k
=
dI
k
V dt
(5)
where I
k
represents PEI of chemical k, I
k
= m
k

k
, m
k
is the mass of substance k,
k
the overall PEI per unit
mass of chemical k in units of PEI/kg, and V the reaction
volume. According to Eq. (2), the PEI transformation rate
of a reaction system r
PEI
, in units of PEI/(Ls), is:
r
PEI
=

k
r
PEI,k
=

k
r
k

k
M
k
10
3
(6)
where M
k
denotes molecular weight of chemical k in units
of g/mol, and r
k
the rate of appearance of chemical k in units
of mol/(Ls).
Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321 15
Such an expression, which shows how the rate depends on
the concentrations of reactants, is called PEI rate-lawexpres-
sion. Note, PEI rate-lawexpression should be consistent with
the method for calculating the overall PEI in Eqs. (2) and
(3), which ignore the combinational impacts that could be
associate with mixtures of chemicals which accounts for the
additional terms not included into those equations. It should
be realized that if the method for calculating the overall PEI
of mixtures in reactors improve or adopt other method, the
form of the PEI rate-law expression should be adjusted ac-
cordingly.
Obviously, r
PEI
can be expressed in terms of reaction tem-
perature and concentration or partial pressure of chemicals
and so on. Therefore, the PEI rate-law expression can be
helpful to analyze the effect of concentrations and temper-
ature on the PEI transformation rate, r
PEI
, and to study the
inuence of back-mixing on process environmental perfor-
mance. At least, the PEI rate-law expression can reveal the
factors that control the transformation rate of PEI, so as to
provide guidance for the selection of reactor type and opera-
tion conditions, and the inner structure of the reactor, which
produce desired products while creating minimum undesired
potential environmental impact.
3.2. PEI balance
Since desired products are the substance we desire and the
objective of this paper is to present a methodology of waste
reduction, that is the primary concern is reducing the impact
and the amount of the non-products, the
k
of the desired
products is not considered in PEI balance. This insures that
the user or producer is not directly penalized for producing
a chemical that has a high PEI value. Auxiliary materials
(catalyst, solvent and so on) used in reaction processes must
leave the production process completely as waste or emis-
sion [10], so the PEI of the auxiliary materials should be
considered. Raw materials that if possible shall be entirely
processed into the desired product are not as a rule com-
pletely converted into the product. The losses that arise are
the cause of generated waste and emission [10], thus their
potential environmental impact should be taken into account.
To sum up, the PEI balance of reaction system under steady
state is:

I
NP
gen
=

I
NP
out


I
NP
in
(7)
where the superscript NP represents materials except de-
sired products, that is, non-products.

I
NP
out
and

I
NP
in
are input
and output rates of PEI caused by non-products, and

I
NP
gen
the rate at which impact is generated in a reactor.
For a differential ow reactor, dV, we write a PEI balance
equation over the reactor:
d

I
gen
= r
PEI
dV (8)
With PEI balance Eq. (7), combining with other equations,
designers can reveal how PEI change with reaction time or
axial and radial direction inside reactors, i.e. can track the
changing of PEI throughout reactors; using Eq. (8) combin-
ing with other rate equations, the inuence of operation pa-
rameters such as temperature, concentration, pressure and
reaction time, etc. on environmental metrics of a chemical
process could be analyzed facilely, and then, reactors gen-
erating a minimum of PEI could be designed easily.
4. Case study
In this case study, we use the environmental impact indices
below, which can be derived from Eq. (1), to characterize
environmental performance of reaction processes [7,8]:
(1) Potential environmental impact generation rate by non-
products

I
NP
gen
.
(2) Specic potential environmental impact due to non-
products
PEI
.

PEI
=

I
NP
gen

p
p
p
(9)
where p
p
is the mass ow rate of product p, and
PEI
has
units of potential environmental impact per mass of prod-
ucts. Obviously, the smaller the values of

I
NP
gen
and
PEI
,
the more environmental friendly the process, and all others
factors such as economics being equal, the more desirable.

I
NP
gen
would be useful in comparing different designs on an
absolute basis, while
PEI
would be useful in comparing
different designs independently of plant size [8].
These indices characterize some aspects of the genera-
tion of PEI within a reaction process. They are very useful
in addressing questions related to the ability of the reactor
to produce desired products while creating a minimum of
undesired potential environmental impact.
To illustrate the use of the algorithm, a case of allyl chlo-
ride manufacturing was used. Allyl chloride is manufactured
by means of non-catalytic chlorination of propylene. Con-
sider the following reaction system.
The principle reaction is:
Cl
2
+ CH
2
=CHCH
3
CH
2
=CHCH
2
Cl + HCl (10)
The main secondary reaction is further chlorination of allyl
chloride to 1,3-dichloropropene:
Cl
2
+CH
2
=CHCH
2
Cl ClCH=CHCH
2
Cl +HCl (11)
Another important side-reaction is the additive chlorination
of propene to 1,2-dichloropropane:
Cl
2
+ CH
2
=CHCH
3
CH
2
ClCHClCH
3
(12)
This reaction system is a kind of complex van de vusse
reaction, a typical reaction process involving consecutive
and parallel reactions. It is, therefore, sufciently complex
to illustrate the algorithm. Due to the limit to the space of
the article, we only select temperature, concentration and
back-mixing to be the object of research to illustrate the use
of the algorithm.
16 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321
The following calculations and curves drawing can be ac-
complished by mathematical software such as Mathcad2000
or Matlab.
4.1. Preconditions
The kinetics of the reaction system [11] are:
r
1
= k
1
C
A
C
B
(13)
r
2
= k
2
C
A
C
C
(14)
r
3
= k
3
C
A
C
B
(15)
where r
1
, r
2
, and r
3
represent the rates of the reactions
in reaction Eqs. (10)(12), respectively, expressed in
terms of mol/(Ls), and k
1
= 1.5 10
6
e
66271/RT
, k
2
=
4.4 10
8
e
99410/RT
and k
3
= 100 e
33140/RT
. Letters AF
represent Cl
2
, propylene, allyl chloride, hydrogen chloride,
1,3-dichloropropene and 1,2-dichloropropane, respectively.
Since the purpose of this case study is only to illustrate
the use of PEI balance and PEI rate-law expression, not
a practical reactor design, we might as well suppose the
reaction process is isovolumetric.
For all cases in this case study, the inlet concentrations
of all chemicals are C
A,0
= C
B,0
= 0.024 mol/L, C
C,0
=
C
D,0
= C
E,0
= C
F,0
= 0 mol/L. The volumetric ow rate
of the reaction feed at the reactor inlet is v
0
= 325 L/s.
The specic potential environmental impact values for
each impact category for the chemicals involved in the pro-
cess are shown in Table 1. Details of the method for calcu-
lating these data appear in reference [7]. Using these data,
the overall PEI of each chemical can be obtained according
to Eq. (4).
4.2. Mathematical models of ideal reactors
4.2.1. Plug-ow reactors (PFR)
The differential mass balances of PFR reactors are:
v
0
dC
A
= r
A
dV (16)
v
0
dC
B
= r
B
dV (17)
v
0
dC
C
= r
C
dV (18)
where r
A
= r
1
+r
2
+r
3
, r
B
= r
1
+r
3
, and r
C
= r
1
r
2
.
r
A
, r
B
and r
C
are the rate of appearance of A, B and C,
Table 1
The specic potential environmental impact values for each category for the chemicals used in this case study [9]
Compound HTPI HTPE ATP TTP GWP PCOP AP
A 0 5.4 10
4
22 0 0 0 0
B 0 0 3.1 10
2
0 0 2.1 0
C 0.51 5.4 10
4
0.10 0.51 0 0 0
D 0.78 2.3 10
4
4.6 10
4
0.78 0 0 0.86
E 1.1 0 0.014 1.1 0 0 0
F 0.61 0 0.0059 0.61 0 0 0
Values are in PEI/kg.
respectively.The differential PEI balance of PFR reactors
is:
d

I
gen
= r
PEI
dV (19)
Dividing Eq. (18) by Eq. (17), then integrate, we obtain:
(C
C
)
PFR
=
b
a 1
(C
B
C
1a
B,0
C
a
B
),
a =
k
2
k
1
+ k
3
, b =
k
1
k
1
+ k
3
(20)
Dividing Eq. (19) by Eq. (17), and then integrate this, we
obtain:
(

I
gen
)
PFR
= v
0

C
B
C
B,0
r
PEI
r
B
dC
B
(21)
4.2.2. Continuous stirred-tank reactors (CSTR)
For CSTR reactors, the mass and PEI balances are:
v
0
C
C
= r
C
V (22)
v
0
(C
B,0
C
B
) = r
B
V (23)

I
gen
= r
PEI
V (24)
From Eqs. (23) and (24), we obtain:

I
gen
=
r
PEI
r
B
(C
B
C
B,0
)v
0
(25)
From Eqs. (22) and (23), we obtain:

r
C
r
B

CSTR
=
C
C
C
B
C
B,0
= a

C
C
C
B

b
Solving this equation, we can get:
(C
C
)
CSTR
=
bC
B
(C
B,0
C
B
)
a(C
B,0
C
B
) + C
B
(26)
4.3. The effect of back-mixing on environmental
performance
It is well known that PFR represents reactors without
back-mixing, CSTR represents reactors with the maximum
back-mixing, we can therefore study effect of back-mixing
on process environmental aspect by studying these two kinds
of reactor.
Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321 17
4.3.1. The effect on

I
gen
4.3.1.1.
l
= 1. When
l
= 1, namely, the importance of
the above eight environmental impact categories is deemed
the same, and the PEI of each non-product is considered,
the PEI rate-law expression due to non-products is:
r
PEI
= 1.6k
1
C
A
C
B
1.5k
3
C
A
C
B
1.2k
2
C
A
C
C
(27)
r
PEI
has units of PEI/(Ls). The value of r
PEI
is negative be-
cause the overall PEI of the reactant A is far greater than that
of each reaction products, in addition, the PEI of product C
is not considered. Suppose reaction temperature T = 530 K,
and at the outlet of reactor, C
B,out
= 0.007 mol/L. Substi-
tute Eqs. (20) and (26) into r
PEI
/r
B
, respectively, and then
graph them in Fig. 1, respectively.
r
PEI
r
B
=
1.6k
1
+ 1.5k
3
+ 1.2k
2
(C
C
/C
B
)
k
1
+ k
3
(28)
According to Eq. (25),

I
gen
of CSTR is:

I
gen
= [A
ABCD
+ 1.5(0.024 0.007)]v
0
where A
ABCD
is the area of rectangle ABCD in Fig. 1. Ac-
cording to Eq. (21),

I
gen
of PFR is:

I
gen
=

A
A

BCD
+ 1.5(0.024 0.007)

v
0
where A
A

BCD
is the area of curved trapezoid A

BCD

(the
shaded region) in Fig. 1. Obviously, when the outlet concen-
tration C
B,out
is identical for both reactors, A
ABCD
is greater
than A
A

BCD
, thus, lower

I
gen
will be achieved by using
CSTR. The calculation results show that, when C
B,out
=
0.007 mol/L, and v
0
= 325 L/s,

I
gen
of CSTR is 3.7
10
4
PEI/h, and

I
gen
of PFR is 3.310
4
PEI/h. Apparently,

I
gen
of CSTR is about 12% lower than that of PFR.
Fig. 2. Curves of r
PEI
/r
B
vs. C
B
of PFR ignoring the overall PEI of A. r
PEI
/r
B
in units of PEI/mol. The numbers 17 denote the reaction temperatures
of 450, 470, 490, 500, 510, 520 and 530 K, respectively.
Fig. 1. Curves of r
PEI
/r
B
vs. C
B
, r
PEI
/r
B
in units of PEI/mol. The solid
curve is for PFR, the dashed for CSTR.
4.3.1.2. Not considering PEI of A. To illustrate the use of
the algorithm effectively, suppose
A
= 0 PEI/kg, and the
weighting factor for each impact type is likewise 1, and the
PEI of each non-product except Cl
2
is also considered. Note
that this case is probably impractical in real manufacturing
processes. The PEI rate-law expression, r
PEI
, in units of
PEI/(Ls), is:
r
PEI
=1.1 10
3
k
1
C
A
C
B
+ 0.049k
3
C
A
C
B
+0.33k
2
C
A
C
C
(29)
r
PEI
r
B
=
1.1 10
3
k
1
0.049k
3
0.33k
2
(C
C
/C
B
)
k
1
+ k
3
(30)
Using the same method as mentioned previously, we can
obtain the curves describing the changing of r
PEI
/r
B
of
PFR and CSTR with C
B
at various reaction temperature, as
shown in Figs. 2 and 3, respectively.
According to Eq. (21), if the inlet and outlet concentration
of B are 0.024 and 0.007 mol/L, respectively,

I
gen
of PFR
equals the area of the region below the curves in the range
of C
B
= 0.0070.024 mol/L in Fig. 2 multiplied by v
0
. For
example, when reaction temperature T = 450 K,

I
gen
of
18 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321
Fig. 3. Curves of r
PEI
/r
B
vs. C
B
of CSTR ignoring the overall PEI of A. r
PEI
/r
B
in units of PEI/mol. The numbers 17 denote the reaction temperatures
of 450, 470, 490, 500, 510, 520 and 530 K, respectively.
PFR equals the area of the shaded region multiplied by v
0
.
From Eq. (25),

I
gen
of CSTR is the product of v
0
and the
area of the corresponding rectangle in Fig. 3. For example,
when reaction temperature is 450 K,

I
gen
of CSTR equals
the area of the rectangle in Fig. 3 multiplied by v
0
. In such
case, the value of

I
gen
is positive number. Comparing Fig. 2
with Fig. 3, it can be found that lower

I
gen
will be obtained if
we adopt PFR and lower reaction temperature. For example,
when reaction temperature is 500 K, and v
0
is 325 L/s,

I
gen
of PFR is 4.8 10
1
PEI/h, while for CSTR,

I
gen
= 1.1
10
3
PEI/h, which is 2.3 times that of PFR.
4.3.2. The effect on
PEI
From Eqs. (9), (21) and (25), we obtain the formulas for
calculating
PEI
. Suppose v
0
is 325 L/s, then for PFR:

PEI
=

I
gen
v
0
C
C,out
M
C
10
3
=
13.0
C
C,out

C
B
C
B,0
r
PEI
r
B
dC
B
(31)
for CSTR:

PEI
=

I
gen
v
0
C
C,out
M
C
10
3
=
13.0
C
C,out
r
PEI
r
B
(C
B
C
B,0
)
(32)
where C
C,out
is the outlet concentration of C.
According to Eqs. (31) and (32), using the curves about
r
PEI
/r
B
versus C
B
and corresponding area, the superior-
ity in environmental index
PEI
about PFR and CSTR
can be discovered. When
l
= 1, the outlet concentration
C
C,out
= 0.011 mol/L and reaction temperature T = 500 K,
using CSTR will obtain lower
PEI
than using PFR. In
this case, for CSTR,
PEI
= 31 PEI/kg; for PFR,
PEI
=
26 PEI/kg. Nevertheless, if not considering
A
, for the
same C
C,out
and T, using PFR will obtain lower
PEI
than using CSTR, for PFR,
PEI
= 0.30 PEI/kg, while for
CSTR,
PEI
= 0.80 PEI/kg, which is 2.6 times that of PFR.
4.4. The effect of temperature on environmental
performance
4.4.1. The effect on

I
gen
When
l
= 1, from Eq. (27), it can be seen easily that the
value of r
PEI
decreases (its absolute value increase) as the
reaction temperature rises, that is,

I
gen
can be lowered by
increasing reaction temperature. For example, when reaction
temperature increases from 500 to 600 K, and the outlet
concentration C
B,out
= 0.007 mol/K, and v
0
is also 325 L/s,

I
gen
of PFR and CSTR would decrease from 3.210
4
and
3.510
4
to 3.710
4
and 4.210
4
PEI/h, respectively.
The effect of reaction temperature on

I
gen
can also be found
from Figs. 2 and 3.
When not considering the PEI of Cl
2
, from the PEI
rate-law expression, expressed in Eq. (29), we can nd that
in the initial stage of the reaction, the determining part
in the right side of Eq. (29) is 1.1 10
3
k
1
C
A
C
B
+
0.049k
3
C
A
C
B
, due to the concentration of C is very low
while the concentrations of A and B are relatively high.
Comparing the magnitudes of the two activation energies
in the two terms, we conclude that the value of this part
decreases as the reaction temperature increases. Thus, in
the early stage of this reaction process, to lower the value
of r
PEI
, the reactor should operate at relatively high tem-
perature. As the reaction proceeds, the concentration of C
is increasing while C
A
and C
B
is decreasing, so that the
inuences of the third term, 0.33k
2
C
A
C
C
, is becoming sig-
nicant. Since the magnitude of activation energy in this
term is the maximum, operating at relatively low temper-
ature would result in a relatively low value of r
PEI
in this
stage. Accordingly, for the reaction system in this case, a
PFR operating at higher temperature rst and then lower
temperature must generate lower value of

I
gen
than that sim-
ply operating at a corresponding uniform temperature. The
Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321 19
curves in different temperature in Fig. 2 intersect each other,
this fact also indicate this result. For example, when the inlet
and outlet concentrations of B are 0.024 and 0.007 mol/L,
respectively, and v
0
is 325 L/s, for PFR operating at reaction
temperature of 450K,

I
gen
= 3.9 10
1
PEI/h, and for PFR
at 530 K,

I
gen
= 6.7 10
1
PEI/h. While for a two-stage
PFR, in which the rst stage operates at 530 K, and the
second stage operates at 450 K, the inlet concentration of B
of the second stage is 0.018 mol/L,

I
gen
= 3.4 10
1
PEI/h.
Obviously, it is 13% lower than that of the PFR simply
operating at reaction temperature of 450 K, and 49% lower
than that of the PFR simply operating at 530 K.
4.4.2. The effect on
PEI
Known from Eqs. (31) and (32), r
PEI
/r
B
can be used to
study the effect of the reaction temperature. When
l
= 1,
it is not difcult to nd that increasing reaction temperature
would decrease the values of r
PEI
/r
B
, accordingly the value
of
PEI
also decreases. For example, for CSTR operating at
500 K, if its outlet concentration C
C,out
= 0.011 mol/L, and
v
0
is 325 L/s, the value of
PEI
is 31 PEI/kg, while for
CSTR operating at 530 K and with the same C
C,out
, the value
of
PEI
is 36 PEI/kg, that is, this CSTR reduced the value
of
PEI
by 16%. Similar to this example, for the case of not
considering the potential environmental impact of Cl
2
, one
can also conclude the effect of temperature on
PEI
.
4.5. The effect of concentration on environmental
performance
We here only study the effect of concentration on
PEI
.
Similar to previous method in Section 4.4.2, one can also use
r
PEI
/r
B
to study the effect of concentration. From Eqs. (31)
and (32), it can be found that for a particular outlet concen-
tration of C the value of
PEI
would be decreased by in-
creasing the value of r
PEI
/r
B
. When
l
= 1, from Eq. (28),
the value of r
PEI
/r
B
can be increased by augmenting C
C
and
decreasing C
B
within a reactor. Since C is a product whereas
B is a reactant, using CSTR will lead to higher C
C
and lower
C
B
within a reactor than using PFR. Therefore, in this case,
using CSTR would result in lower value of
PEI
than using
PFR operating at the same reaction conditions. When not
considering the potential environmental impact of Cl
2
, from
Eq. (30), decreasing C
C
and increasing C
B
within a reactor
will increase the value of r
PEI
/r
B
, accordingly, using PFR
would result in lower
PEI
than using CSTR. These results
are also concluded in Section 4.3.2.
5. Discussion
In the case study, the indexes involved are

I
gen
and
PEI
.
However, these two indexes only express the situation of
PEI generation within processes. The environmental perfor-
mance of a process also relates to its emission of PEI. There-
fore, in the WAR algorithm, the output rate of PEI,

I
out
, and
the total PEI per mass of product streams leaving the system,

I
out
, are another category of indices to characterize some
aspects of emission of PEI from a manufacturing process.
According to Eq. (1):

I
out
=

I
in
+

I
gen
(33)

I
out
is obtained by dividing

I
out
by the output of products to
give:

I
out
=

I
out

p
p
=

I
in
+

I
gen

p
p
(34)
Substituting Eq. (9) into Eq. (34), we obtain:

I
out
=

I
in

p
p
+
PEI
(35)
For the case study, it is not difcult to obtain that

I
in
=
4.6 10
4
PEI/h, i.e.

I
in
is a constant. Therefore, when the
feed condition of a reaction process is xed, back-mixing
and reaction temperature affect

I
gen
and

I
out
in the same
way. Since

p
p = v
0
C
C,out
M
C
10
3
, it is not difcult
to nd that, when the feed condition and C
C,out
are spec-
ied, back-mixing and reaction temperature inuence

I
out
and
PEI
in the same way, too.
Thus, in Section 4.3, when
l
= 1 and the reaction pro-
cess is operated in the specied conditions, the results that

I
gen
and
PEI
of CSTR are less than those of PFR indicate
that using CSTR can obtain lower

I
out
and

I
out
than us-
ing PFR. That means at the given conditions, back-mixing
make the emission of potential environmental impact de-
crease. By contrast, when not considering the PEI of A, the
results are converse, i.e. using PFR would gain lower PEI
generation rate than using CSTR; and the value of
PEI
of
CSTR that operate at reaction temperature 500 K and out-
let concentration C
C,out
= 0.011 mol/L is 2.6 times that of
PFR. The results indicate that at the given reaction condi-
tions, back-mixing will increase the emission of potential
environmental impact of the reaction processes, accordingly,
reactor designers should try to avoid it in such case.
In Section 4.4, when
l
= 1, the results show that a higher
reaction temperature is favorable for the decrease of

I
gen
and
PEI
in the giving conditions. Therefore, increasing the
reaction temperature will lead to the decrease of

I
out
and

I
out
. When not considering the PEI of Cl
2
, PFR operating
at a high temperature in the rst section and low tempera-
ture in the remainder section can result in lower generation
and emission of PEI than that operating at a corresponding
uniform temperature.
To study the effects of concentration on environmental
performance of reaction processes, two analyses are intro-
duced in Section 4.5. The results show that when
l
= 1,
operating at relatively low concentration of B and high
concentration of C is favorable for decreasing the poten-
tial environmental impact created in manufacturing a unit
mass of the product C. This also is true for total PEI leav-
ing the system per mass of product streams according to
20 Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321
the relations about these two categories of indexes. By
contrast, when neglecting the PEI of Cl
2
, the results are
converse.
In the cases study, we have only studied the effects of
back-mixing on the environmental performance. Whereas
the phenomena of back-mixing is only one of engineering
factors, which still involve pre-mixing, heat and mass trans-
fer, thermostability and so on. However, the effects of those
engineering factors not involved in the case study on envi-
ronmental performance can also be simulated and therefore
well studied through a combination of reactor mathematical
model, which relates to corresponding engineering factors,
with PEI balances. As for the effects of other reaction con-
ditions not involved in the case study, such as space time,
pressure, compositions of feeds and so on, which may af-
fect the environmental performance of reaction processes
signicantly in some cases, can also be studied with the
help of corresponding PEI balances and the PEI rate-law
expressions, because we can ultimately simulate the change
of the environmental indexes with these reaction conditions
through PEI balance and PEI rate-law expression.
The reaction system involved in the case studied is a kind
of relatively complex van de vusse reaction, nevertheless the
reaction system in real manufacturing process may involve
more reaction types and, therefore, is more complex than
that one. One can, however, simulate the change of envi-
ronmental indexes within a reactor by combining traditional
reactor mathematical model with the PEI balance, and may
also discover the effects of reaction conditions and engineer-
ing factors on environmental performance by PEI rate-law
expression and/or combinations it with other reaction rate
equations as well as other related equations in reactor math-
ematical models.
Note that assuming
l
= 1 and not considering the PEI of
Cl
2
are just for illustrating the algorithm, these assumptions
may be inappropriate for a real reactor design. Other users of
this methodology may choose to weight effects differently
to reect local needs and polices where the plants locate [8].
In the case study, the quantitative environmental impact
assessment and the PEI indexes are cited from the WAR al-
gorithm, it is worth noting that, however, quantitative impact
assessment and environmental indexes are likely to improve
with time, and that these improvements can be incorporate
Appendix A. Nomenclature
AF chlorine, propylene, allyl chloride, hydrogen chloride, 1,3-dichloropropene and
1,2-dichloropropane, respectively
C
A
, C
B
, etc. concentration of chemicals A, B, etc. (mol/L)
C
A,0
, C
B,0
, etc. initial concentration of chemicals A, B, etc. (mol/L)
C
B,out
, C
C,out
, etc. outlet concentration of chemicals B and C, etc. (mol/L)
dV, dC, d

I
gen
differential V, C and

I
gen

I
gen
potential environmental impact generation rate (PEI/h)

I
NP
gen
potential environmental impact generation rate of non-products (PEI/h)
into this methodology as they become available. It is suggest
that either the WAR algorithms assessment method and the
PEI indexes, another similar method and indexes, or even a
more sophisticated and comprehensive method and indexes
be used to this pollution reduction methodology.
In the case study, the calculations of the impact indexes
should be accurate to no more than one signicant gure
[12]. But two signicant gures are used in order to allow
readers to reproduce the calculations if necessary. It is also
worth noting that, even being accurate to one signicant
gure, one can still discern signicant differences in the
index between different cases.
6. Conclusions
To perform pollution prevention in the process of reac-
tor design, studying the effects of reaction conditions such
as temperature, pressure and concentration as well as vari-
ous engineering factors on environmental performance of a
reaction process is very important. In this paper, potential
environmental impact balance and PEI rate-law expression
are proposed as a tool to be used to carry out this work. The
form of PEI rate-law expression should be consistent with
the method for calculating the overall PEI of mixtures of
chemicals inside reactors. The effectiveness of the method
has been demonstrated through the reaction of allyl chlo-
ride production. From the processes of the analyses in the
case study, analyzing with PEI balance and PEI rate-law ex-
pression combining with other corresponding equations and
some mathematic knowledge can give important insights
into how reaction conditions as well as engineering factors
affect the environmental performance of reaction processes.
This kind of study is one of the bases for developing chem-
ical reactors with environmental benign processes.
Acknowledgements
Financial support provided by the National Natural Sci-
ence Foundation of China under Grant No. 20176405 and
the Major State Basic Research Development Program un-
der Grant No. G2000026307 is gratefully acknowledged.
Q. Chen, X. Feng / Chemical Engineering and Processing 44 (2005) 1321 21

I
in
potential environmental impact input rate (PEI/h)

I
NP
in
potential environmental impact input rate of non-products (PEI/h)
I
k
potential environmental impact of chemical k (PEI)

I
out
potential environmental impact output rate (PEI/h)

I
NP
out
potential environmental impact output rate of non-products (PEI/h)
m
(in)
j
mass ow rate of stream j into a process (kg/h)
m
(out)
j
mass ow rate of stream j out of a process (kg/h)
m
k
mass of chemical k
m
k,0
mass of chemical k at initial state of reaction
m
k, out
mass of chemical k at reactor exit or at the moment of reaction termination
m
p
mass ow rate of desired product p (kg/h)
M
k
molecular weight of chemical k (g/mol)
r
A
, r
B
, etc. the rate of appearance of chemicals A, B, etc. (mol/(Ls))
r
PEI
total PEI transformation rate (PEI/(Ls))
r
PEI,k
PEI transformation rate of chemical k (PEI/(Ls))
S area of a certain gure
T reaction temperature (K)
v
0
volume ow rate entering a reactor (L/s)
x
kj
mass fraction of chemical k in stream j
Greek letters

l
weighting factor for impact category l

PEI
generated PEI per unit mass of all desired products (PEI/kg)

s
kl
the specic PEI of chemical k for environmental impact category l (PEI/kg of chemical k)

k
overall PEI of chemical k (PEI/kg of chemical k)
References
[1] H.T. Kohlbrand, From waste treatment to pollution prevention and
beyond: opportunities for the next 20 years, AIChE Symp. Ser.
94 (320) (1998) 117121.
[2] A.P. Rossister, Waste Minimization Through Design, McGraw-Hill,
New York, 1995.
[3] J.A. Dyer, K.L. Mulholland, Prevent pollution via better reactor
design and operation, Chem. Eng. Prog. 94 (2) (1998) 6166.
[4] A.D. Curzons, D.J.C. Constable, D.N. Mortimer, So you think your
process is green, how do you known? Using principles of sustain-
ability to determine what is greena corporate perspective, Green
Chem. 3 (2001) 16.
[5] H. Cabezas, J.C. Bare, S.K. Mallick, Pollution prevention with chem-
ical process simulator: the generalized waste reduction (WAR) algo-
rithm, Comput. Chem. Eng. 21 (1997) S305S310.
[6] A.K. Hilaly, S.K. Sikdar, Pollution balance: a new methodology for
minimizing waste production in manufacturing process, J. Air Waste
Manage. Assoc. 44 (1994) 13031308.
[7] D.M. Young, H. Cabezas, Designing sustainable process with simu-
lation: the waste reduction (WAR) algorithm, Comput. Chem. Eng.
23 (1999) 14771491.
[8] H. Cabezas, J.C. Bare, S.K. Mallick, Pollution prevention with
chemical process simulator: the generalized waste reduction (WAR)
algorithmfull version, Comput. Chem. Eng. 23 (1999) 623
634.
[9] D.M. Young, R. Scharp, H. Cabezas, The waste reduction (WAR)
algorithm: environmental impacts, energy consumption, and engi-
neering economics, Waste Manage. 20 (2000) 605615.
[10] W. Schramm, New ndings on the generation of waste and emissions,
and a modied cleaner production assessment approachillustrated
by leather production, J. Cleaner Prod. 5 (4) (1997) 291300.
[11] B. Pahor, Z. Kravanja, N.I. Bedenik, Synthesis of reactor networks
in overall process owsheets within the multilevel MINLP approach,
Comput. Chem. Eng. 25 (2001) 765774.
[12] S. Mallick, H. Cabezas, J.C. Bare, et al., A pollution reduction
methodology for chemical process simulator, Ind. Eng. Chem. Res.
35 (11) (1996) 41284138.
Chemical Engineering and Processing 44 (2005) 2332
Effects of hydraulic residence time on metal uptake by activated sludge
Tlay A. zbelge

, H. nder zbelge, Murat Tursun


Chemical Engineering Department, Middle East Technical University, 06531 Ankara, Turkey
Received 28 April 2003; received in revised form 12 April 2004; accepted 13 April 2004
Available online 17 June 2004
Abstract
The combined uptake of Cu
2+
and Zn
2+
by activated sludge (biomass) was investigated at steady state in an activated sludge process (ASP)
without recycle (namely, once-thro ASP), for different values of inuent metal concentrations and hydraulic residence time in the range of
2.540 h. The experiments were performed at a constant pH of 7 and temperature of 25

C. The results showed that the percentage removal of


both copper and zinc by activated sludge increased with the increasing residence time; moreover, the percent metal uptakes were higher for
both metals at the low level of inuent metal concentrations [(Cu
2+
) = 1.5 mg/L and (Zn
2+
) = 9 mg/L], than those at the higher level of
concentrations [(Cu
2+
) = 4.5 mg/L and (Zn
2+
) = 27 mg/L]. The least represented forms of both Cu
2+
and Zn
2+
in activated sludge were
their suldes, and the dominant metal forms were found to be organically bound for both of the metals, Zn
2+
and Cu
2+
.
2004 Elsevier B.V. All rights reserved.
Keywords: Biosorption; Activated sludge process; Removal of heavy metals; Sequential extraction of metals; Metal uptake
1. Introduction
Industrial discharges containing heavy metals (HMs) from
mining, metal rening and other industries are hazardous for
public health unless the HMs are removed at the site of the
waste production by appropriate methods [1]. One of these
techniques for removing the HMs from wastewaters is to
use the well-known activated sludge process (ASP) [2]. The
studies on the treatment of these wastes biologically are not
complete yet, since some metals are toxic to microorganisms
(MOs) even at very low concentrations and their inhibition
effects are not exactly known.
In giving a literature survey on this subject, the following
are the important points to be noted: (i) the form of HM in
ASP affects the degree of its toxicity, the soluble forms be-
ing the most toxic [2]; (ii) different oxidation states of the
same metal can have different binding mechanisms to the
sludge [3]; and (iii) it is known that the dissolved oxygen
concentration in the medium has an important effect on the
oxidation states of the metals present in the system [4]. The
other factors affecting the forms of HMs are the solution pH
and the presence of complexing agencies in the medium [5].

Corresponding author. Tel.: +90 312 210 2621;


fax: +90 312 210 1264.
E-mail address: tozbelge@metu.edu.tr (T.A. zbelge).
It was observed that the toxic effects of HMs decreased as
the concentrations of complexing agencies increased [5,6],
which was explained with the decrease in free metal ions
due to their reactions with complexing agents. Toxicity char-
acteristics of a certain metal can be greatly modied by
the other HMs because they compete for available organic
ligands in sludge and wastewater [7]. On the other hand,
non-interactive effects among some metals may also occur
in ASP [8]. The diversity of MOs in ASPs may vary from
plant to plant. Individual species of MOs present may even
change within the same system from time to time. There-
fore, responses of different systems, or even the same sys-
tem at different times and operating conditions, to HMs are
unlikely to be identical, and the response mainly depends
on the composition of activated sludge [4].
Complexation, adsorption and precipitation reactions in-
volving HMs are all strongly inuenced by hydrogen ion ac-
tivity. The effects of pH value on the degree of dissociation
of potential ligands, competition with metal ions for uptake
and metal precipitation have been investigated [3,9]. Low
pH values tend to reduce the degree of dissociation, which
resulted in decreased adsorption and complexation. At pH
= 7, ligands such as carboxyl and phosphate were favored
for cation uptake, because pK
a
values were considerably less
than 7. When pHwas increased to a level at which other com-
plexing anions could no longer successfully compete with
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.004
24 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332
the hydroxide ions present, metals precipitated as hydrox-
ides. Furthermore, increase in metal inhibition with decrease
in pH was attributed to the increase in free metal ions in the
medium [10,11]. Toxicities of HMs are also affected by cell
residence time in ASP. Nelson et al. [12] reported that toxic
effects of cadmium and zinc decreased considerably by in-
creasing cell residence time from 1 to 15 days in ASP with
recycle. Uloth and Mavinic [13] explained this decrease with
the increasing production of bacterial extracellular polymer
substances (EPS) at high mean cell residence times. Some
investigators [9,14,15] reported that an increase in substrate
and mixed liquor suspended solids (MLSS) concentrations
led to increase in metal uptake and decrease in adverse ef-
fects of HMs to ASP; they attributed this decrease to the
decreasing proportion of active cellular components, which
were tied up as metalligand complexes [15]. Combined ef-
fects of copper, zinc and nickel on the treatment efciency
of ASP and the combined uptakes of these metals by acti-
vated sludge have been reported [16], as have the effects of
copper and zinc on activated sludge bacteria/growth kinet-
ics [17,18]. However, combined uptakes of only copper and
zinc by sludge, and the effect of hydraulic residence time in
this process have not been reported, although Vian et al. [16]
discussed the effects of long sludge age (kept almost con-
stant at 9 days) on the combined uptakes of copper, nickel
and zinc in an ASP with recycle. In another study [19],
transfer of organic matter from wastewater (WW) to acti-
vated sludge was investigated and observed that the transfer
of soluble fraction always occurred from wastewater to ac-
tivated sludge, whereas bidirectional transfer occurred for
the colloidal fraction. The biosorption capacity of activated
sludge was found to be at a ratio of chemical oxygen de-
mand (COD)/TSS of about 10400 mg/L, and biosorbable
fraction of wastewater was almost 45% of non-settleable
(colloidal + soluble) fraction [19]. A single metal (Zn
2+
)
study was reported [20], where zinc biosorption by activated
sludge was more effective than by digested sludge in remov-
ing zinc from metal plating efuent. The objective of this
work is to investigate the effect of hydraulic residence time
on combined uptake of Cu
2+
and Zn
2+
by activated sludge
in once-thro ASP (OTASP) without recycle, which is not
encountered in the available literature.
2. Experimental
2.1. Experimental set-up
A modular continuously stirred glass tank reactor (from
Cole-Parmer, Inc., USA) with an effective volume of 2 L was
used in the OTASP. Peristaltic pumps were used at the inlet
and at the exit of the reactor for feed/harvest, respectively.
Aeration and efcient mixing was provided in the reactor;
temperature and pHwere controlled at 251

Cand 70.01,
respectively, by electronic controllers. A synthetic wastew-
ater with known inlet substrate concentration (in terms of
Table 1
Composition of synthetic wastewater
Constituents (mg/L)
Proteose-peptone 1221.7
NaCl 407.4
Na
2
SO
4
44.6
K
2
HPO
4
44.6
MgCl
2
.6H
2
O 3.7
FeCl
2
.2H
2
O 3.7
CaCl
2
.2H
2
O 3.7
MnSO
4
0.057
H
2
MoO
4
0.031
NaOH 0.008
ZnSO
4
0.046
CoSO
4
0.049
CuSO
4
0.076
chemical oxygen demand) of a COD of S
o
= 1300 mg/L
was fed, in which Proteose-peptone (1221.7 mg/L) was the
sole carbon and nitrogen source. Other minor nutrients were
also added in sufcient quantities to make carbon the limit-
ing nutrient (Table 1). Solutions of CuSO
4
and ZnSO
4
were
added to the wastewater to provide Cu
2+
and Zn
2+
ions at
the desired concentrations. Further details are given else-
where [17,2123].
2.2. Experimental method
Biomass obtained from the wastewater treatment plant
of Middle East Technical University (METU) was used as
seed in the experiments. In the acclimation of MOs to HMs,
maximum concentrations of Cu
2+
and Zn
2+
reached dur-
ing acclimation were decided from previous work [17,21].
The literature showed that when these metals were present
individually in the system, ASP failed to operate at steady
state values of 15 mg/L for Cu
2+
and 90 mg/L for Zn
2+
, re-
spectively [5]. Therefore, one-third of threshold concentra-
tions of these metals (that is, 5 mg/L for Cu
2+
, and 30 mg/L
for Zn
2+
) were used to constitute the combined threshold
concentration for mixtures of copper and zinc ions. Accli-
mation experiments were performed by gradually increasing
the concentrations of Cu
2+
and Zn
2+
to two, three and four
times their combined threshold concentration. The optimum
condition for a successful acclimation providing the growth
of healthy MOs, as observed from the increase in turbidity
values, was obtained at the concentrations corresponding to
twice their combined threshold concentration [21].
Experiments without HMs were conducted using syn-
thetic WW inoculated with 10% (v/v) acclimated seed MOs
to twice their combined threshold concentration (that is,
10 mg/L Cu
2+
, and 60 mg/L Zn
2+
). The reactor was op-
erated as a batch system for 48 h with the growth of new
biomass. Then, continuous feeding of synthetic WW to the
reactor at a desired volumetric ow rate (or desired resi-
dence time) was started. The system reaches steady state
in about 1244 days depending on the inuent metal con-
centrations and the hydraulic residence time (HRT); efuent
T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332 25
samples from the reactor were analyzed for substrate, S, and
biomass, X, concentrations. MLSS concentration was used
as the primary indicator for steady state conditions. At each
residence time, steady state was reached when the efuent
S (COD) and X (MLSS) concentrations remained constant
with respect to time.
Experiments with HMs were performed at two concentra-
tion levels of each metal (low levels: (Cu
2+
) = 1.5 mg/L
and (Zn
2+
) = 9 mg/L; high levels: (Cu
2+
) = 4.5 mg/L
and (Zn
2+
) = 27 mg/L, for different values of HRT in the
range of 240 h using the same synthetic WW having the
composition given in Table 1. In these experiments, 1.8 L
of synthetic WW was inoculated with 0.2 L of acclimated
seed MOs in the activated sludge reactor, and the system
was operated in batch mode for almost 72 h to ensure the
growth of biomass by observing the increase in measured
turbidities of the samples from the reactor. Then, the con-
tinuous operation was started by feeding the reactor with
synthetic WW containing copper and zinc ions at the de-
sired inlet concentrations. Inuent ow rate was adjusted
in order to provide a required HRT. Samples were taken
from the efuent stream at different times in order to deter-
mine the efuent S and X concentrations until steady state
was reached. Then, a sample of 1.5 L from efuent stream
was collected for the metal analyses, and the operation was
stopped.
2.3. Analytical methods
Substrate concentrations were determined in centrifuged
sample aliquots by standard COD analysis, and gravimet-
ric method was used to measure the biomass concentrations
[24]. Analyses were repeated at least twice with the sam-
ples under the same experimental conditions and the average
values were obtained.
2.3.1. Distribution of metals among the constituents of
activated sludge mixed liquor
Since metals were distributed between the two phases of
mixed liquor in the reactor, namely solution and biomass, the
amounts of copper and zinc associated with these phases and
the forms of these metals in the biomass were determined
at steady state condition for each HRT.
Firstly, the concentrations of HMs in efuent were de-
termined. At least three samples of 40 mL were acidied
with 10 mL of concentrated HNO
3
to decompose MOs.
These samples were diluted to the sensitivity range of the
atomic absorption spectrophotometer (AAS) (Model: Philips
9200X-AAS) (diluted 5 times for Cu
2+
analysis, and 25
times for Zn
2+
analysis) and then analyzed. The average of
these measurements yielded the total HM concentration in
the efuent, C
T
(mg/L), for that specic experimental run.
To convert (mg metal/L) to (mg metal/kg sludge), the mea-
sured metal concentration (as, mg metal/L) at each HRT
was divided by the steady state MLSS concentration (as, kg
TSS/L).
Secondly, concentrations of the HMs associated with the
solution, and solid phases of the efuent were determined.
Two of 50 mL aliquots were collected from the efuent of
the reactor, centrifuged at 1300 rpm for 15 min, by using
a RC5C type Dupont centrifuge, to separate the solution
and solid phases, and the supernatants transferred to sam-
ple tubes. Then, supernatants and the residual solid phase
were acidied by adding 10 mL concentrated HNO
3
to each
sample tube. Samples were then analyzed by AAS after per-
forming the necessary dilutions to the sensitivity range of
the instrument as explained above. The average of the results
from the two samples were recorded as C
S
(metal concen-
tration in the solution phase), and C
A
(metal concentration
in the solid phase) values for each HM.
The distribution of different metal forms in the sludge
was determined by using the sequential extraction pro-
cedure [25]. Activated sludge mixed liquor of 1.5 L,
corresponding to 0.2500.100 g of sludge, was collected
from the outlet of the reactor at the steady state. In order
to separate the sludge and supernatant, centrifugation at
1300 rpm for 15 min was carried out using a RC5C type
Dupont centrifuge. Then the extraction process was started
by adding the reagents of KNO
3
, KF, Na
4
P
2
O
7
, EDTA
(ethylene-diamine-tetraacetic-acid) and HNO
3
on the con-
centrated sludge (order of addition given in Table 2), which
are capable of extracting exchangeable, adsorbed, organi-
cally bound, carbonates, and suldes forms of the metals
in the sludge sample, respectively. To provide complete
extraction, samples were shaken on a platform shaker, then
centrifuged at 1300 rpm for 15 min, and supernatants were
decanted into separate tubes. Between each extraction, the
sludge residue was washed with 50 mL of de-ionized water
by shaking for 30 min, and the samples centrifuged again
under the same conditions. Both supernatants from the ex-
traction and washing were combined and acidied with two
drops of 5 M HNO
3
before metal analysis, in order to pre-
serve the solutions. All metal concentrations were measured
as before and conversions were made to obtain the concen-
trations of each form of each metal as mg metal/ kg sludge,
for which the metal concentration as mg/L was divided by
the dried MLSS concentration (kg TSS/L), determined at
the steady state. The same procedure was followed for the
other reagents, the only exception being that double extrac-
tion was performed with EDTA and supernatants pooled
before the metal analyses. The remaining solid residue in
Table 2
Details about sequential extraction process parameters [25]
Reagents Molarity
(M)
pH Extraction
time (h)
(Solution/solids)
ratio
KNO
3
1 24 50/1
KF 0.5 6.5 24 80/1
Na
4
P
2
O
7
0.1 24 80/1
EDTA 0.1 6.5 224 80/1
HNO
3
1 24 50/1
26 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332
the sample tube, at the end of the extraction procedure, was
decomposed with 5 mL of concentrated HNO
3
in order to
determine the levels of metals not extracted by the protocol.
3. Results and discussion
Two phases exist in the activated sludge reactor; the liq-
uid phase consists of water, soluble ions and soluble sub-
strate while the solid phase (biomass or sludge) has settleable
waste materials, dead and alive MOs in it. Metals associated
with these phases were investigated in this study. The ma-
terial balance for the metals at the steady state is dened as
follows:
C
T
= C
A
+C
S
(1)
where C
T
is the total HM concentration in efuent (mg/L),
C
A
is the HM concentration taken up by the sludge (mg/L)
and C
S
is the heavy metal concentration in the solution
(mg/L). This balance was applied to each run using HMs. In
Tables 36, material balance results for copper and zinc are
presented for two sets of experiments (in the rst set, inu-
ent HM concentrations being (Cu
2+
) = 1.5 mg/L, (Zn
2+
)
= 9 mg/L, and in the second set, inuent HM concentrations
Table 3
Material balance on copper at the inuent concentration of (Cu
2+
)
= 1.5 mg/L

h
(h) C
T
C
S
C
A
[(C
S
+ C
A
) C
T
]
100/C
T
17.18 1.47 1.38 0.15 +1.04
14.68 1.61 1.42 0.12 4.35
9.22 1.63 1.60 0.09 +3.68
5.12 1.63 1.58 0.06 +0.61
3.75 1.43 1.43 0.02 +1.40
Table 4
Material balance on zinc at the inuent concentration of (Zn
2+
) = 9 mg/L

h
(h) C
T
C
S
C
A
[(C
S
+ C
A
) C
T
]
100/C
T
17.18 11.13 10.32 0.54 2.43
14.68 10.57 10.70 0.55 +6.43
9.22 10.68 10.70 0.44 +4.31
5.12 9.88 9.09 0.39 4.05
3.75 9.79 9.83 0.29 +3.37
Table 5
Material balance on copper at the inuent concentration of (Cu
2+
)
= 4.5 mg/L

h
(h) C
T
C
S
C
A
[(C
S
+ C
A
) C
T
]
100/C
T
40 5.24 4.88 0.22 2.67
20 4.68 4.53 0.18 +0.64
10 4.27 4.23 0.13 +1.02
5 4.20 4.19 0.08 +1.67
2.5 4.58 4.57 0.04 +0.66
Table 6
Material balance on zinc at the inuent concentration of (Zn
2+
)
= 27 mg/L

h
(h) C
T
C
S
C
A
[(C
S
+ C
A
) C
T
]
100/C
T
40 29.20 29.14 0.72 +2.26
20 27.13 26.23 0.65 0.92
10 29.69 28.75 0.63 +1.04
5 28.38 27.68 0.52 0.63
2.5 28.88 28.52 0.44 +0.28
being (Cu
2+
) = 4.5 mg/L, (Zn
2+
) = 27 mg/L. It is seen
in these tables that the total metal concentration in the ef-
uent is greater than that in the inuent to the reactor at the
steady state due to the experimental error occurred in the
determination of these concentrations in very small amounts
of sludges produced. In spite of this, the obtained results are
consistent as indicated by the small differences between the
two sides of the material balance at each run, the maximum
percent experimental error being within 5%.
Fig. 1 shows the percent metal removals. Small percentage
removals (up to 10%) are because only small amounts of
biomass were produced in the systemunder the experimental
conditions used. This is expected in a system where toxic
0
2
4
6
8
10
12
0 10 20 30 40
residence time (h)
p
e
r
c
e
n
t

m
e
t
a
l

r
e
m
o
v
a
l
Cu: 1.5 mg/L
Cu: 4.5 mg/L
Zn: 9 mg/L
Zn: 27 mg/L
Fig. 1. Percent removal of each metal [(C
A
/C
T
) 100] in the ex-
periments of Set-1 (inuent concentrations: (Cu
2+
) = 1.5 mg/L,
(Zn
2+
) = 9 mg/L) and Set-2 (inuent concentrations:
(Cu
2+
) = 4.5 mg/L, (Zn
2+
) = 27 mg/L).
T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332 27
metals inhibit the microbial community to yield low MLSS
values at the steady state and thus low removals of HMs
[9,1215]. Since the highest value of HRT is 40 h (less than
2 days) in this work, which is equal to the sludge age in
the present OTASP, therefore, the biomass concentrations
in the efuent stream were low as expected. However, in
an ASP with recycle, it may be possible to obtain higher
percent removals of copper and zinc by increasing the sludge
age up to 812 days [1216] due to the higher biomass
concentrations in the reactor. As seen in Fig. 1, the percent
copper and zinc removals by biomass increased with the
increasing HRT or the decreasing dilution rate (D = 1/
h
),
keeping the inlet HM concentrations constant. At the low
level of inuent HM concentrations ((Cu
2+
) = 1.5 mg/L,
(Zn
2+
) = 9 mg/L), the percent metal uptakes were higher
than those at the higher level, ((Cu
2+
) = 4.5 mg/L, (Zn
2+
)
= 27 mg/L), for both of the metals at each residence time,
due to the lower HM toxicities to biomass at the low level
of the HM concentrations. The higher percent removals of
zinc compared to those of copper at the short residence
times of
h
<5 h can be attributed to the lower toxicity of
zinc in spite of the lower biomass concentrations at these
residence times. Although the toxicity of copper is greater
than that of zinc at the same level of concentration, the
percent copper removals exceeded those of zinc at the longer
residence times of
h
>5 h; the reasons for this nding is
because of the lower inuent copper concentrations studied
in this experimental work compared to those of zinc, and the
occurrence of higher biomass concentrations at the longer
residence times which has an effect in decreasing the HM
toxicities. This may also be due to the greater contribution
of the precipitated form of copper to the sludge produced,
as reported by Vian et al. [16].
When the amount of HM removed per kg of activated
sludge produced (mg metal/kg sludge) for each metal is ob-
served in Fig. 2 that these values are higher for Zn
2+
be-
cause of the higher inuent concentrations of Zn
2+
in the
experiments; (mg metal/kg sludge) values for zinc decrease
with the increasing residence time due to the increase in the
amount of sludge produced at the higher HRTs. This de-
creasing effect on the ratio of masses will be more signi-
cant for zinc than that of copper, since the percent removals
are lower for zinc than those of copper as seen in Fig. 1; but,
still these values are above those of copper due to the higher
inuent concentrations of zinc ions studied at both the low
and high concentration levels of the HMs in the experiments
compared to those of copper.
The increase in the amount of copper and decrease in the
amount of zinc removed by activated sludge, as the residence
time is increased, can be explained by transport processes,
adsorption onto surfaces or the possible competition between
copper and zinc ions in binding to available sites on the
enzymes. It may be that the interaction between copper and
biomass increases as the residence time increases eventually
leading to an increase in the amount of copper taken up
by activated sludge. As a result, the afnity of zinc for the
0
1000
2000
3000
4000
5000
6000
7000
0 10 20 30 40 50
residence time (h)
a
m
o
u
n
t

o
f

m
e
t
a
l

r
e
m
o
v
e
d

(
m
g

m
e
t
a
l

/

k
g

s
l
u
d
g
e
)

Cu: 1.5 mg/L
Cu: 4.5 mg/L
Zn: 9 mg/L
Zn: 27 mg/L
Fig. 2. Amount of metal bound to the sludge as (mg metal bound/kg
sludge produced) for the experiments of Set-1 and Set-2.
sludge decreases, since the available sites on the enzymes
to be occupied by the metals are limited. It was reported
earlier that both nickel and zinc uptakes decreased when the
inlet concentration of copper in the mixture was increased, at
constant inlet concentrations of nickel and zinc [16]. It was
also reported by Vian et al. [16] that the percentage of zinc
removal was higher than that of copper due to the increasing
percentage of zinc removed by precipitation although the
percent of zinc uptaken by sludge was lower than that of
copper for the higher toxicity effect of zinc at its higher
concentrations. Although these data [16] seem to support
the above hypothesis, they did not discuss the effect of HRT
on metal uptake by activated sludge in an ASP with recycle
where the sludge age was kept almost constant. Therefore,
it is necessary to investigate this hypothesis further.
3.1. Distributions of different metal forms in the sludge
According to the sequential extraction procedure [25] out-
lined earlier, the forms of HMs in the sludge are dened as
exchangeable (extracted by KNO
3
), adsorbed (by KF), or-
ganically bound (by Na
4
P
2
O
7
), carbonates (by EDTA), and
suldes (by HNO
3
). The results of these experiments are
given in Figs. 36. Rankings of the different forms of copper
and zinc in the sludge for the two sets of experiments (rst
set: inuent (Cu
2+
) = 1.5 mg/L, (Zn
2+
) = 9 mg/L; sec-
ond set: inuent (Cu
2+
) = 4.5 mg/L, (Zn
2+
) = 27 mg/L
suggest that,
28 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332
0
100
200
300
400
500
600
700
800
2 4 6 8 10 12 14 16 18
residence time (h)
m
g

m
e
t
a
l

r
e
m
o
v
e
d

/

k
g

s
l
u
d
g
e
Exchangeable
Adsorbed
Organically bound
Carbonates
Sulfides
Fig. 3. Distribution of Zn
2+
among its forms in the sludge using the data
of Set-1 experiments.
0
1000
2000
3000
4000
0 10 20 30 40
residence time (h)
m
g

m
e
t
a
l

r
e
m
o
v
e
d

/

k
g

s
l
u
d
g
e
Exchangeable
Adsorbed
Organically bound
Carbonates
Sulfides
Fig. 4. Distribution of Zn
2+
among its forms in the sludge using the data
of Set-2 experiments.
0
20
40
60
80
100
120
140
160
180
2 4 6 8 10 12 14 16 18
residence time (h)
m
g

m
e
t
a
l

r
e
m
o
v
e
d

/

k
g

s
l
u
d
g
e
Exchangeable
Adsorbed
Organically bound
Carbonates
Sulfides
Fig. 5. Distribution of Cu
2+
among its forms in the sludge using the data
of Set-1 experiments.
in Fig. 3, for (Zn
2+
) = 9 mg/L: organically bound >
carbonates > exchangeable > adsorbed > suldes,
in Fig. 4, for (Zn
2+
) = 27 mg/L: organically bound
= exchangeable = adsorbed > carbonates > suldes,
in Fig. 5, for (Cu
2+
) = 1.5 mg/L: carbonates > organi-
cally bound > exchangeable > adsorbed > suldes,
in Fig. 6, for (Cu
2+
) = 4.5 mg/L: organically bound >
exchangeable > carbonates > adsorbed > suldes.
As shown in Figs. 36, at low concentration levels of both
metals, namely for (Cu
2+
) = 1.5 mg/L and for (Zn
2+
)
= 9 mg/L, the rankings of metal forms remained unchanged
as the HRT increased. On the other hand, at the high
concentration levels of these metals, namely for (Cu
2+
)
= 4.5 mg/L and for (Zn
2+
) = 27 mg/L, the rankings of
the different forms of these metals changed with the HRT;
the least represented forms for both are the suldes, and the
dominant metal forms are the organically bound forms for
both zinc and copper.
The rankings of the different forms are more marked at
lowinuent metal concentrations than those at the high level.
However, the dominant and the least represented forms are
the same. It has been reported that the dominant forms of
HMs could vary for different sludges [25]. Therefore, the
rankings of the metal forms for copper and zinc, should not
be assumed to be the same for all kinds of sludges.
To determine the efciency of the metal extraction pro-
cedure, the sample remaining at the end of the extraction
was acidied with HNO
3
to determine the amounts of resid-
ual metals not extracted by the reagents. Ideally, the total
T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332 29
0
100
200
300
400
0 10 20 30
residence time (h)
m
g

m
e
t
a
l

r
e
m
o
v
e
d

/

k
g

s
l
u
d
g
e
40
Exchangeable
Adsorbed
Organically bound
Carbonates
Sulfides
Fig. 6. Distribution of Cu
2+
among its forms in the sludge using the data
of Set-2 experiments.
amount of the metals determined by the extraction proce-
dure should be equal to the amount of metal bound to the
sludge, the discrepancy between the two being denoted by
C
H
. Thus, the material balance for extraction can be shown
as:
C
A
= C
E
+C
H
(2)
where C
A
is the concentration of metal bound to the sludge
(mg/L, or mg metal/kg sludge), C
E
is the concentration
of metal extracted in the analysis (mg/L or mg metal/kg
sludge) and C
H
is the concentration of metal not extracted
by the used reagents in the extraction (mg/L or mg metal/kg
sludge). Thus, the extraction procedure used is capable of
extracting most of the metals bound to the biomass, varying
between 93.85 and 98.98% of the total amount of the HMs
in the sludge. Also, the total concentrations of the metals
extracted are mostly consistent with the C
A
values. The ex-
perimental error in the sequential extraction analysis falls
within 10%.
It should be noted that these data were obtained by ex-
tracting the relatively small amounts of biomass, which can
exagerate the experimental error. The reasons for working
with small amounts of biomass are two-fold; rstly, the sam-
ples were collected here at the steady state directly from the
efuent of the system at one day intervals. Samples can de-
generate due to the continuing interaction between metals
and biomass before the analysis, which can cause changes
in the relative amounts of the metal forms in sludge over
the time. Secondly, biomass concentrations in the efuent of
the reactor are quite low anyhow due to metal inhibition ef-
fects, a problem which can be overcome if a recycle is used
in the system. Therefore, further studies are planned to be
performed in an ASP with recycle.
In the literature, equilibrium biosorption of Zn
2+
and
Cu
2+
by nonviable activated sludge was investigated in a
packed column adsorber [26], where metabolically induced
interactions between HMs and MOs were eliminated; thus,
the problems, such as toxicity of metals to MOs, mainte-
nance requirements of MOs in controlling the pH within a
range suitable for the MOs were avoided. It was observed
by Utgikar et al. [26] that equilibrium metal uptakes from
solutions containing single metal ion were 2.5 and 3.4 mg/g
dried biomass for Zn
2+
, and 1.9 and 5.9 mg/g dried biomass
for Cu
2+
at pH of 3.0 and 3.8, respectively. Equilibrium up-
take of metals was modeled using the Langmuir adsorption
isotherm. Equilibrium uptakes from binary mixtures were
30% lower than single component solution uptakes for both
metals, indicating some competition between the two metals.
No hysteresis was detected between adsorption and desorp-
tion equilibria. The simultaneous sorption of metal cation
and sulfate anion was probably occurring at pH of 3.0, while
proton exchange predominated at pH of 3.8. They pointed
out the usefulness of non-viable activated sludge as a biosor-
bent for removing metal ions from WWs.
On the other hand, the removal of metal ions in an ASP
was affected adversely due to the toxicity of HMs to MOs.
It was reported that the presence of Cu
2+
ions in wastewater
inhibited the net specic growth rate of autotrophic biomass
to a larger extent than Zn
2+
ions [27]. Therefore, in an ASP
treating such a WW, it is expected that the concentration
and the composition of EPS (consisting of protein, polysac-
charide and DNA) produced by microbial cells will vary
[28], since it is known that the EPS formation rate is pro-
portional to the rate of substrate utilization, thus to the rate
of biomass production [29]. When EPS are not involved in
the process, MOs may not easily display high occulation
and metal uptake capacity, as the oc forming bacteria can
interact with each other [30]. However, the precise role of
EPS in biosolidliquid separation is not well understood al-
though EPS are thought to be of considerable importance in
bioocculation and settling of activated sludge [28]. For ex-
ample, the EPS content has been shown to be directly [31],
or conversely [32] related to sludge volume index (SVI).
Besides, the low protein levels were obtained at high car-
bohydrate concentrations depending on microbial consortia
due to the intracellular storage of excess carbon substrate and
lower sludge retention time, particularly, in the pulppaper
and textile industries lab-scale activated sludge systems.
The low protein content was probably due to the inhibition
of EPS production in the aforementioned industies as re-
ported by Dignac et al. [33]. It was noted that the wastewater
composition rather than operating conditions had a strong
inuence on EPS composition and the physicochemical
properties of ocs in an ASP. For example, if the ASP was
operated at high F/M ratios, both carbohydrate and DNA
content of EPS increased as the protein content decreased;
30 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332
as a result of this, a more hydrophobic (larger contact angle)
and less negatively charged oc surface was observed [28].
The present OTASP was operated at high F/M ratios [23];
although no data were obtained about the concentration and
the composition of EPS produced in this work, it might be
predicted from the available literature that the protein con-
tent of the produced EPS in the ocs was low; thus the set-
tling ability of the resulting sludge, most probably with a
high value of SVI was not good. Actually, low settlability of
the produced sludge was observed in the experiments. The
effect of operational conditions on EPS composition will be
the subject of a future study in which it will be investigated
in an ASP with recycle to compare the effects of sludge
residence time (SRT), HRT, EPS and F/M ratio on the set-
tlability of produced sludge and metal uptake. In this study,
the change in bacterial prole of the MLSS with metal con-
centration was not followed, but it was reported in the lit-
erature [34] that many different species of bacteria isolated
from activated sludge were shown to produce EPS. These
polymers might be in the form of loose slime or capsules
adhered to the cell wall. In activated sludge, slime polymers
remained in the dissolved and colloidal phases of an efu-
ent, while capsular polymers remained attached to the ocs
and hence settled with the sludges. Some activated sludge
bacteria did not produce capsules or extracellular material.
In these bacteria, metals might be accumulated either in the
cytoplasm or by adsorption on to the cell wall. Some re-
searchers [35] concluded that only the half of the observed
metal uptake could be attributed to EPS and that cell itself
played a role in accumulating metals. In another work [36],
EPS were extracted, cells were broken up using ultrasound
and cell envelopes were sedimented by centrifugation. Only
small quantities of copper were found in the cytoplasm, but
substantial amounts of copper were bound to cell envelopes
and EPS. The metabolism of elements occurred in many
species of bacteria. It was also reported that four major types
of metal transformation in organisms occurred [37], these
being chelate formation by the binding of metals to organic
ligands, shifts in metal valencies, substitution of one metal
for another and biomethylation of metals to microorganisms.
When metals were chelated to substrate molecules, the ad-
sorption of these molecules might represent a means of accu-
mulating metals within a cell. These metabolic transforma-
tions of metal ions by bacteria were different for each metal
and for different combinations of metals in the medium.
Therefore, in such a dynamic system as is ASP, it is not
plausible to use metal uptake as a measure of EPS content.
As it is seen in Fig. 2, the copper uptake increases with the
HRT up to 20 h then it shows a slight decrease between 20
and 40 h. This can be explained with the change in the given
ranking in Fig. 6. The chemical speciation and distribution
of copper in an activated sludge slurry can be quantied
using an equlibrium constant approach including complex-
ation with inorganic and organic ligands and adsorption by
activated sludge solids [38]. It can be observed in Fig. 6 that
the amount of copper carbonate per kg of sludge is higher
than the amount of organically removed copper at a HRT of
40 h, while it was lower than that of organically bound cop-
per at a HRT of 20 h. Therefore, it is shown in Fig. 2 that
the copper uptake at a HRT of 40 h is less than that at 20 h.
The copper uptake is probably less at a HRT of 40 h due to
the effect of changing metabolic activity of microbial ocs
and the solubility constant of copper carbonate complex in
the medium at the prevailing conditions. The desorption of
copper into the solution might have occurred due to the sol-
ubility constant of this complex.
4. Conclusions
The effect of hydraulic residence time (in the range of
2.540 h) on the removal of Zn
2+
and Cu
2+
ions by acti-
vated sludge was investigated in a once-thro activated sludge
process (OTASP) at a constant pH of 7 and temperature of
25

C. It was found that the percentage removals of both cop-


per and zinc by activated sludge increased with the increas-
ing residence time due to the increasing amounts of sludge
produced with time. Higher percent removal of zinc at the
low level of inuent concentrations of both the copper and
zinc, (Cu
2+
) = 1.5 mg/L and for (Zn
2+
) = 9 mg/L, due
to the higher inlet concentration of zinc over that of copper
was replaced by the higher percent removal of copper at the
high level of inuent concentrations of each metal, (Cu
2+
)
= 4.5 mg/L and for (Zn
2+
) = 27 mg/L, although copper is
known to be more toxic to the microorganisms than zinc.
This suggested that the lower toxicity of zinc at its low con-
centration level is compensated by another factor such as
the removal of both metals by precipitation also contribut-
ing to the removal process. However, it was not possible
to differentiate between the percent removals of each metal
by biosorption or by precipitation in OTASP. The previous
study [16] cannot render this information due to the presence
of a third metal ion such as Ni
2+
in the system. Therefore, a
further study is required in an activated sludge system with
recycle which will differentiate the percent removals of each
metal by precipitation and biosorption clearly. In this study,
at short residence times (a range of 2.55 h), the percent-
age of zinc bound to the biomass is higher than for copper
due to the dominating effect of lower toxicity of zinc; this
is also shown by the sequential extraction results yielding
the dominant form of both Zn
2+
and Cu
2+
as organically
bound indicating that both types of ions are removed by
biosorption rather than by precipitation due to the lower de-
gree of inhibition of MOs at small sludge ages or HRTs in
OTASP. This point can be claried by investigation in an
activated sludge process with recycle. It is also found that
the least represented forms of both the copper and zinc ions
in the sludge are their suldes. A further study is under-
way to investigate the effect of operational conditions on the
composition of extracellular polymer substances in an ASP
with recycle to compare the effects of sludge residence time,
hydraulic residence time, EPS and food to microorganism
T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332 31
(F/M) ratio on the settlability of produced sludge and the
mechanism of metal uptake.
Acknowledgements
The nancial support by TB

ITAK (Scientic and Tech-


nical Research Council of Turkey) and DPT (State Plan-
ning Organization) for the Project (No. KTCAG-DPT-8) is
greatly appreciated.
The invaluable comments of the referees are highly ac-
knowledged in improving the quality of this contribution
and for encouraging the authors to continue a more detailed
experimental study, which will clarify the role of extracel-
lular polymer substances in the mechanism of metal uptake
by activated sludge process.
Appendix A. Nomenclature
C
A
concentration of metal bound to the sludge
(mg/L)
C
E
concentration of metal extracted in the
analysis (mg/L)
C
H
concentration of metal unaccounted for by
extract (mg/L)
C
S
metal concentration in the solution phase (mg/L)
C
T
total metal concentration in the efuent (mg/L)
COD chemical oxygen demand (mg O
2
/L)
D dilution rate (h
1
)
EPS extracellular polymer substances
HM heavy metal (plural: HMs)
HRT hydraulic residence time (h)
K
a
solubility product constant
MLSS mixed liquor suspended solids (mg/L)
MO microorganism (plural: MOs)
OTASP once-thro activated sludge process
(ASP without recycle)
S
o
substrate concentration in the inuent
(mg COD/L)
S substrate concentration in the efuent
(mg COD/L)
SVI sludge volume index
TSS total suspended solid concentration
(mg/L or kg/L)
WW wastewater (plural: WWs)
X biomass concentration in the efuent (mg/L)
Greek letters

h
hydraulic residence time (h or days)

c
mean cell residence time or sludge age (days)
References
[1] R.D. Neufeld, J. Gutierrrez, R.A. Novak, A kinetic model and equi-
librium relationship for heavy metal accumulation, J. WPCF 49
(1977) 489498.
[2] S. Sujarittanonta, J.H. Sherrard, Activated sludge nickel toxicity
studies, J. WPCF 53 (1981) 13141322.
[3] W. Stumm, H. Bilinski, Trace metals in natural waters: difculties
of interpretation arising from our ignorance on their speciation,
Proceedings of the Sixth International Conference, Advances in Water
Pollution Research, Jerusalem, 1972, pp. 3949.
[4] G. Tchobanoglous, F.L. Burton, Wastewater Engineering: Treatment,
Disposal, Reuse, third ed., Metcalf and Eddy, McGraw-Hill, New
York, 1991.
[5] C.P.C. Poon, K.H. Bhayani, Metal toxicity to sewage organisms, J.
Sanitary Eng. Div., ASCE 97 (1971) 161169.
[6] J. Gardiner, Complexation of trace metals by EDTA in natural waters,
Water Res. 10 (1976) 507517.
[7] Y.P. Ting, F. Lawson, I.G. Prince, Uptake of cadmium and zinc by
the Alga Chlorella vulgaris: multi-ion situation, Biotechnol. Bioeng.
39 (1991) 445455.
[8] E.F. Barth, M.B. Ettinger, B.V. Salotto, G.N. McDermott, Summary
report on the effects of heavy metals on the biological treatment
processes, J. WPCF 37 (1965) 8696.
[9] M.H. Cheng, J.W. Patterson, R.A. Minear, Heavy metal uptake by
activated sludge, J. WPCF 47 (1975) 362368.
[10] J.E. Loveless, H.A. Painter, The inuence of metal concentrations
and pH value on the growth of Nitrosomonas strain isolated from
activated sludge, J. Gen. Microbiol. 114 (1968) 144.
[11] F. Braam, A. Klapwijk, Effects of copper on nitrication in activated
sludge, Water Res. 15 (1981) 10931098.
[12] P.O. Nelson, A.K. Chung, M.C. Hudson, Factors affecting the fate
of heavy metals in the activated sludge process, J. WPCF 53 (1981)
13231333.
[13] V.C. Uloth, D.S. Mavinic, Aerobic biotreatment of a high strength
leachate, J. Environ. Eng. Div., ASCE 105 (1979) 647661.
[14] M.M. Bagby, J.H. Sherrard, Combined effects of cadmium and nickel
on the activated sludge process, J. WPCF 53 (1981) 16091619.
[15] R.D. Tyagi, D. Couillard, F.T. Tran, Analysis of nal settling tank in
relation to control of metal inhibition in the activated sludge process,
Can. J. Chem. Eng. 69 (1991) 534543.
[16] A. Vian, F. Mirada, J.J. Rodriguez, Copper (2+), zinc (2+) and
nickel (2+) uptake by activated sludge, Sep. Sci. Technol. 20 (1985)
587597.
[17] N.Y. Beyenal, T.A. zbelge (Ba ser), H.. zbelge, Combined effects
of Cu
2+
and Zn
2+
on activated sludge process, Water Res. 31 (1997)
699704.
[18] A. Cabrero, S. Fernandez, F. Mirada, J. Garcia, Effects of copper
and zinc on the activated sludge bacteria growth kinetics, Water Res.
32 (1998) 13551362.
[19] A. Guellil, F. Thomas, J.C. Block, Transfer of organic matter between
wastewater and activated sludge ocs, Water Res. 35 (2001) 143150.
[20] F. Bux, B. Atkinson, H.C. Kasan, Zinc biosorption by waste activated
and digested sludge, Water Sci. Technol. 32 (1999) 127130.
[21] N. Yurt, Combined effects of heavy metals on activated sludge
process, M.S. Thesis, Department of Chemical Engineering, Middle
East Technical University, Ankara, 1994.
[22] P. Altinten, Combined effects of copper and zinc on the activated
sludge process, M.S. Thesis, Department of Chemical Engineering,
Middle East Technical University, Ankara, 1997.
[23] M. Tursun, Effects of heavy metals on activated sludge process, M.S.
Thesis, Department of Chemical Engineering, Middle East Technical
University, Ankara, 1997.
[24] L.S. Clesceri, A.E. Greenberg, A.D. Eaton (Eds.) Standard Meth-
ods for Examination of Water and Wastewater, APHA, WPCF, and
AWWA, 20th ed., New York, 1998.
[25] R.C. Stover, L.E. Sommers, D.J. Silviera, Evaluation of metals in
wastewater sludge, J. WPCF 48 (1976) 21652175.
[26] V. Utgikar, B.Y. Chen, H.H. Tabak, D.F. Bishop, R. Govind, Treat-
ment of acid mine drainage: I. Equilibrium biosorption of zinc and
copper on non-viable activated sludge, Int. Biodeterior. Biodegrad.
46 (2000) 1928.
32 T.A. zbelge et al. / Chemical Engineering and Processing 44 (2005) 2332
[27] S.R. Juliastuti, J. Baeyens, C. Creemers, D. Bixio, E. Lodewyckx,
The inhibitory effects of heavy metals and organic compounds on
the net maximum specic growth rate of the autotrophic biomass in
activated sludge, J. Hazard. Mater. B100 (2003) 271283.
[28] D.T. Sponza, Investigation of extracellular polymer substances (EPS)
and physicochemical properties of different activated sludge ocs
under steady-state conditions, Enzyme Microb. Technol. 32 (2003)
375385.
[29] C.S. Laspidou, E.R. Bruce, A unied theory for extracellular
polymeric substances, soluble microbial products, active and inert
biomass, Water Res. 11 (2002) 27112720.
[30] J.A. Scatt, J. Karanjkar, Immobilized biolms on granular activated
carbon for removal and accumulation of heavy metals from contam-
inated streams, Water Sci. Technol. 38 (1998) 97204.
[31] V. Urbain, J.C. Block, J. Manem, Bioocculation in activated sludge:
an analytical approach, Water Res. 27 (1993) 829838.
[32] J.A.S. Goodwin, C.F. Forster, A further examination into the com-
position of activated sludge surfaces in relation to their settlement
characteristics, Water Res. 19 (1985) 527533.
[33] M.F. Dignac, V. Urbain, D. Rybacki, A. Bruchet, D. Snidaro, P.
Scribe, Chemical description of extracellular polymers: implications
on activated sludge oc structure, Water Sci. Technol. 38 (1998)
4553.
[34] M.J. Brown, J.N. Lester, Role of bacterial extracellular polymers in
metal uptake in pure bacterial culture and activated sludge, Water
Res. 16 (1982) 15491560.
[35] B.A. Friedman, P.R. Dugan, Concentration and accumulation of
metallic ions by the bacterium Zoogloea, Dev. Ind. Microbiol. 9
(1968a) 381388.
[36] W.A. Corpe, Metal-binding properties of surface materials from ma-
rine bacteria, Dev. Ind. Microbiol. 16 (1975) 249255.
[37] A. Jernelv, A.L. Martin, Ecological Implications Of Metal
Metabolism By Micro-Organisms, Swedish Water and Air Pollution
Research Laboratory, Stockholm, 1975, p. 32.
[38] B.R. Fristoe, P.O. Nelson, Equilibrium chemical modelling of
heavy metals in activated sludge, Water Res. 17 (1983) 771
778.
Chemical Engineering and Processing 44 (2005) 3339
Porous catalyst intraparticle status of parallel,
equilibrium-restrained reactions
Guangsuo Yu

, Fuchen Wang, Zhenghua Dai, Zunhong Yu


Institute of Clean Coal Technology, East China University of Science and Technology, P.O. Box 272, 130 Meilong Road,
Shanghai 200237, Peoples Republic of China
Received 17 February 2004; received in revised form 11 March 2004; accepted 7 April 2004
Available online 1 June 2004
Abstract
The catalyst intraparticle reaction-diffusion process of parallel, equilibrium-restrained reactions for the methanation systemwas studied. The
non-isothermal one-dimensional and two-dimensional reaction-diffusion models for the key components have been established, and solved
using an orthogonal collocation method. The simulation values of the effectiveness factors for methanation reaction
CH
4
and shift reaction
CO
2
are fairly in agreement with the experimental values.
CH
4
is large, while
CO
2
is very small. The shift reaction takes place as direct and reverse
reaction inside the catalyst pellet because of the interaction of methanation and shift reaction. For parallel, equilibrium-restrained reactions,
effectiveness factors are not able to predict the catalyst internal-surface utilization accurately. Therefore, the intraparticle distributions of the
temperature, the concentrations of species and so on should be taken into account.
2004 Elsevier B.V. All rights reserved.
Keywords: Catalyst; Parallel reactions; Equilibrium-restrained; Reaction-diffusion model
1. Introduction
Gas-phase mass transport in porous catalysts is an im-
portant factor in predicting observed reaction rates for
heterogeneous catalytic systems. Mathematical models of
diffusion and reaction in porous catalysts play a fundamen-
tal role in the design of catalysts and catalytic reactors,
and in the interpretation of experimental rate data. The
phenomenon of simultaneous transport and reaction in-
side a catalyst particle has been widely studied over the
years, but mainly for single reactions and simple kinetics.
The classical diffusion-reaction model for a porous cata-
lyst describes the inuence of intraparticle concentration
gradients on the observed rate through the effectiveness
factor, which is dened as the ratio of the volume-averaged
reaction rate over the reaction rate at the external sur-
face. For single reactions, the effectiveness factor can
describe the utilization degree of catalyst internal surface
accurately.

Corresponding author.
E-mail address: gsyu@ecust.edu.cn (G. Yu).
Several quantitative analyses of the effect of intraparticle
heat and mass transport have been carried out for parallel, ir-
reversible reactions [1]. Roberts and Lamb [2] have worked
on the effect of reversibility on the selectivity of parallel
reactions in a porous catalyst. The reaction selectivity of a
kinetic model of two parallel, rst order, irreversible reac-
tions with a second order inhibition kinetic term in one of
them has also been investigated [3].
The reaction-diffusion model inside a catalyst pellet is
a differential equations set of boundary-value problem, the
standard solving methods for which are nite difference,
numerical integration shooting, weighting residual, singular
perturbation [4], Green function [5], false transient [6], or-
thogonal collocation [7], moving grid nite element [8], etc.
Based on the investigation of reaction, heat transfer and
mass transfer of the KD306-type sulfur-resisting methana-
tion catalyst [911], the non-isothermal one-dimensional
and two-dimensional reaction-diffusion models for the key
components have been established, and solved using an or-
thogonal collocation method in this paper. The scope is to
study the catalyst intraparticle reaction-diffusion processes
that involve parallel, non-rst order, equilibrium-restrained
reactions.
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.003
34 G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339
2. Reaction system and fundamental data
With the reducing supply of low sulfur petroleum, the
demand for residual oil and coal with high-sulfur con-
tent is increasing. The development and application of
sulfur-resisting catalysts have received great attention. The
KD306-type sulfur-resisting methanation catalyst, a sulde
catalyst such as Mo and W of group VIB as the primary
species and Fe, Co, Ni of group VIII as the indispens-
able promoters, can be used in methanation reaction with
sulfur-containing syngas as the reactants at high reactivity
and selectivity. The cylindrical catalyst pellets (diameter
= 5.422 mm, length = 4.327 mm) were prepared by com-
pressing particles.
Methanation reaction and watergas shift reaction
(WGSR), which are chosen as the examples for presenta-
tion, are expressed as follows:
CO +3H
2
= CH
4
+H
2
O, H

298
= 206 kJ/mol
CO +H
2
O = CO
2
+H
2
, H

298
= 41 kJ/mol
The former is a volume-decreasing reaction, while the
latter is not. Both reactions are exothermic. Methanation
is a deep hydrogenation reaction for carbon monoxide and
WGSR is a complete oxidation reaction in which carbon
monoxide is oxidized into carbon dioxide and water is re-
duced with the formation of hydrogen. As in the preparation
of methane, other hydrocarbons, low alcohols and particu-
larly, carbon dioxide and water are formed. Because of the
presence of water, WGSR always occurs in the methana-
tion process, which reduces the selectivity and yield of the
desired product.
Fig. 1 shows reaction equilibrium constants of the metha-
nation reaction and the WGSR [12]. The mole composition
of town gas is: CO 20%, H
2
40%, CO
2
10%, H
2
O
0%, N
2
the rest. Methanation reaction is almost irreversible
673 723 773 823 873
1
10
100
1000
10000
0
4
8
12
16
K
p
,
s
h
i
f
t
K
p
,
m
e
t
h
a
n
a
n
t
i
o
n
,
a
t
m
-
2
T, K
Fig. 1. Reaction equilibrium constants of methanation reaction and shift
reaction.
under the reaction conditions at 400600

C. The equilib-
rium constant of the WGSR is small and varies insignif-
icantly with temperature. Due to the water formation and
the concentration variations in carbon monoxide and hydro-
gen, which are consequences of the methanation reaction,
the overall WGSR occurs both as the direct and reverse re-
action at internal surface of the catalyst. So, the fundamental
features of this reaction system are equilibrium-restrained,
namely reversible, parallel reactions.
2.1. Reaction kinetics [9]
The intrinsic kinetics was measured in an isothermal in-
tegrated reactor and the reaction rate equations in terms of
power function have been established
(r
CH
4
)
w
= 3.8333 10
6
exp

46 944
RT

P
0.58
CO
P
0.52
H
2
mol s
1
kg
1
(r
CO
2
)
w
= 4.1089 10
5
exp

58 349
RT

P
0.58
CO
P
0.98
H
2
O
(1 ) mol s
1
kg
1
where = J
P
/K
P
.
The macrokinetics were measured in a gradientless reac-
tor.
2.2. Effective diffusivity D
effA
[10]
Mass transport inside the catalyst has been usually de-
scribed by applying the Fick equation, by means of an effec-
tive diffusivity D
effA
. Based on properties of the interface
and neglecting the composition effect, composite diffusivity
of the multi-component gas mixture is calculated through
the simplied Wilke model [13]. The effect of pore-radius
distribution on Knudsen diffusivity is taken into account.
The effective diffusivity D
effA
is given by
D
effA
=

r
max
r
min

1
1 y
A
N

j=A
y
j
D
Aj
+
1
D
Ak
(r)

1
f(r) dr
where is the tortuosity factor of the KD306 catalyst, which
was measured using single pellet string reactor technique.
2.3. Thermal conductivity
eff
[11,14]
Based on the Woodside stochastic model, thermal con-
ductivity
eff
can be calculated from gas composition and
temperature of the catalyst exterior surface by

eff

s
=

fm

where
s
is given by

s
=19.90005 +0.12333T 2.38169(
1
100
T)
2
+0.151983(
1
100
T)
3
G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339 35
3. Theory [15,16]
The assumptions made for the reaction-diffusion models
are as follows:
(1) parallel, equilibrium-restrained reactions occur (namely,
methanation reaction is irreversible, while WGSR is re-
versible);
(2) non-rst order reactions;
(3) non-isothermal;
(4) quasi-homogeneous;
(5) only key components are considered (methane and car-
bon dioxide).
3.1. One-dimensional model
Commonly [17], when the length-to-diameter ratio of a
cylindrical catalyst is close to 1, the cylindrical catalyst can
be simplied as a sphere, the radius of which, R
p
, is calcu-
lated by 3V
p
/S
p
. The one-dimensional, key-component based
reaction-diffusion models of methanation system are as fol-
lows:
d
2
C
CH
4
dx
2
+
2
x
dC
CH
4
dx
=
R
2
p
D
eff,CH
4
(r
CH
4
)
w

p
d
2
C
CO
2
dx
2
+
2
x
dC
CO
2
dx
=
R
2
p
D
eff,CO
2
(r
CO
2
)
w

p
d
2
T
dx
2
+
2
x
dT
dx
=
R
2
p

eff

p
[(r
CH
4
)
w
(H
CH
4
)
+(r
CO
2
)
w
(H
CO
2
)]
Boundary conditions
x =
r
R
p
= 1, C
CH
4
= (C
CH
4
)
s
, C
CO
2
= (C
CO
2
)
s
,
T = T
s
x = 0,
dC
CH
4
dx
= 0,
dC
CO
2
dx
= 0,
dT
dx
= 0
Applying orthogonal collocation method, the above equa-
tions become
N+1

k=1

B
jk
+
2
x
A
jk

C
CH
4
,k
=
R
2
p
D
eff,CH
4
(r
CH
4
)
w,j

p
N+1

k=1

B
jk
+
2
x
A
jk

C
CO
2
,k
=
R
2
p
D
eff,CO
2
(r
CO
2
)
w,j

p
N+1

k=1

B
jk
+
2
x
A
jk

T
k
=
R
2
p

eff

p
[(r
CH
4
)
w,j
(H
CH
4
)
+(r
CO
2
)
w,j
(H
CO
2
)]
The solutions of the above non-linear equations can be ob-
tained using the Broyden quasi-Newton method [18]. Based
on the mass balance, the concentrations of other component,
such as CO, H
2
, H
2
O and so on, can be calculated from the
key components [9].
According to the denitions

CO
2
=
4R
2
p
D
eff,CO
2
(dC
CO
2
/dR)
s
(4/3)R
3
p

p
(r
CO
2
)
w,s
the effectiveness factors for the catalyst pellet can be calcu-
lated from

CO
2
=
3

N+1
j=1
A
N+1,j
C
CO
2
,j
(R
2
p
/D
eff,CO
2
)
p
(r
CO
2
)
w,s
,

CH
4
=
3

N+1
j=1
A
N+1,j
C
CH
4
,j
(R
2
p
/D
eff,CH
4
)
p
(r
CH
4
)
w,s
3.2. Two-dimensional model
R
P
and H
P
are the radius and semi-length of cylindri-
cal catalyst, respectively. Assume a cylindrical ring element
with the length dH and the radial thickness dR at the axial
position H and the radial position R in the 2D cylindrical
system. Introducing the dimensionless variables
x =
R
R
p
, z =
H
H
p
The species (methane and carbon dioxide) mass balance
and energy conservation in the cylindrical ring element at
steady state lead to the following 2Dpartial differential equa-
tions:

2
C
CH
4
x
2
+
1
x
C
CH
4
x
+

R
p
H
p

2
C
CH
4
z
2
=
R
2
p
D
eff,CH
4

p
(r
CH
4
)
w

2
C
CO
2
x
2
+
1
x
C
CO
2
x
+

R
p
H
p

2
C
CO
2
z
2
=
R
2
p
D
eff,CO
2

p
(r
CO
2
)
w

2
T
x
2
+
1
x
T
x
+

R
p
H
p

2
T
z
2
=
R
2
p

eff

p
[(H
CH
4
)(r
CH
4
)
w
+(H
CO
2
)(r
CO
2
)
w
]
with the associated boundary conditions
x = 0,
C
CO
2
x
=
C
CH
4
x
=
T
x
= 0
z = 0,
C
CO
2
z
=
C
CH
4
z
=
T
z
= 0
x = 1 or z = 1, C
CO
2
= C
CO
2
,s
, C
CH
4
= C
CH
4
,s
, T = T
s
Each item of the model at collocation point (I, J) may be
expressed as
36 G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339
F = C
CO
2
F = C
CH
4
F = T
(F)
I,J
N+1

j=1
N+1

i=1
a
ij
x
2i2
I
z
2j2
J
N+1

j=1
N+1

i=1
b
ij
x
2i2
I
z
2j2
J
N+1

j=1
N+1

i=1
c
ij
x
2i2
I
z
2j2
J

F
x

I,J
N+1

j=1
N+1

i=1
(2i 2)a
ij
x
2i3
I
z
2j2
J
N+1

j=1
N+1

i=1
(2i 2)b
ij
x
2i3
I
z
2j2
J
N+1

j=1
N+1

i=1
(2i 2)c
ij
x
2i3
I
z
2j2
J

2
F
x
2

I,J
N+1

j=1
N+1

i=1
(2i 2)(2i 3)a
ij
x
2i4
I
z
2j2
J
N+1

j=1
N+1

i=1
(2i 2)(2i 3)b
ij
x
2i4
I
z
2j2
J
N+1

j=1
N+1

i=1
(2i 2)(2i 3)c
ij
x
2i4
I
z
2j2
J

2
F
z
2

I,J
N+1

j=1
N+1

i=1
(2j 2)(2j 3)a
ij
x
2i2
I
z
2j4
J
N+1

j=1
N+1

i=1
(2j 2)(2j 3)b
ij
x
2i2
I
z
2j4
J
N+1

j=1
N+1

i=1
(2j 2)(2j 3)c
ij
x
2i2
I
z
2j4
J
where, I = 1, 2, . . . , N
R
+1, J = 1, 2, . . . , N
A
+1.
We obtain
N
X
+1

k=1
BX
Ik
C
CH
4
,kJ
+
1
x
I
N
X
+1

k=1
AX
Ik
C
CH
4
,kJ
+

R
p
H
p

2
N
Z
+1

k=1
BZ
Jk
C
CH
4
,kI
=
R
2
p

p
D
eff,CH
4
(r
CH
4
)
w,IJ
N
X
+1

k=1
BX
Ik
C
CO
2
,kJ
+
1
x
I
N
X
+1

k=1
AX
Ik
C
CO
2
,kJ
+

R
p
H
p

2
N
Z
+1

k=1
BZ
Jk
C
CO
2
,kI
=
R
2
p

p
D
eff,CO
2
(r
CO
2
)
w,IJ
N
X
+1

k=1
BX
Ik
T
kJ
+
1
x
I
N
X
+1

k=1
AX
Ik
T
kJ
+

R
p
H
p

2
N
Z
+1

k=1
BZ
Jk
T
kI
=
R
2
p

eff
[(H
CH
4
)
IJ
(r
CH
4
)
w,IJ
+(H
CO
2
)
IJ
(r
CO
2
)
w,IJ
]
The concentration and temperature proles are calculated
from the above non-linear equations using the Broyden
quasi-Newton method. The effectiveness factors for the
catalyst pellet may be expressed as

CO
2
=

N
R
+1
I=1

N
A
+1
J=1
(r
CO
2
,w
)
I,J
W
R,I
W
A,J
(r
CO
2
,w
)
s

N
R
+1
I=1
W
R,I

N
A
+1
J=1
W
A,J
,

CH
4
=

N
R
+1
I=1

N
A
+1
J=1
(r
CH
4
,w
)
I,J
W
R,I
W
A,J
(r
CH
4
,w
)
s

N
R
+1
I=1
W
R,I

N
A
+1
J=1
W
A,J
where W
R
, W
A
are weight coefcients of the Gauss integral
formulae, dened as

1
0
x
2i2
x
1
dx =
N+1

j=1
W
j
x
2i2
j
=
1
2i 2 +
,
i = 1, 2, . . . , N
4. Results and discussion
4.1. Model testing
The experimentally-determined effectiveness factor is
determined as the ratio of the experimental macro reaction
rate to the intrinsic reaction rate under the same interface
(bulk) composition and temperature. Based on the ex-
perimental conditions of the macrokinetics, the predicted
effectiveness factors of the methanation reaction and the
WGSR are obtained by solving the above non-isothermal
one-dimensional and two-dimensional reaction-diffusion
models for the key components. Table 1 shows the calcu-
lated effectiveness factors and the experimental values. By
taking into account the fact that there are errors in measuring
intrinsic kinetics, macrokinetics, effective diffusivity, and
thermal conductivity, the calculated effectiveness factors
are in reasonable agreement with the experimental values.
From Table 1, it can be concluded that the simulation results
of the one-dimensional model are close to the results of the
two-dimensional model; also, the methanation reaction ef-
fectiveness factors of the two-dimensional model are closer
to the experimental data than those of the one-dimensional
model. With these models, the concentration proles and
temperature prole in the catalyst pellet can be predicted.
4.2. Results
Predicted parameter distributions within catalyst pellet for
some experimental conditions are presented in Tables 25.
As shown in Tables 2 and 3, the deeper inside the pellet,
the higher the concentration of carbon dioxide and methane
are, and the lower the concentration of carbon monoxide is.
The effectiveness factor
CH
4
of the methanation reaction
is large, while the effectiveness factor
CO
2
of the WGSR
is very low and nearly equal to zero. Because the equilib-
rium constant of WGSR is small and varies only slightly
with temperature, and the methanation reaction changes the
gas composition, there are zones where the direct reaction
and the reverse reaction of WGSR dominate at internal sur-
face of catalyst, which is shown in Tables 2 and 4. The
low mole-fraction of water results in the reverse reaction of
WGSRwithin the region beneath the exterior surface. Due to
reaction consumption of carbon monoxide, the mole-fraction
of carbon monoxide within the core region is low, which
leads to the reverse WGSR. The reaction rate of reverse
WGSR is large, especially at the exterior surface and the
G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339 37
Table 1
Simulation results of one-dimensional and two-dimensional models by orthogonal collocation method
Number T (K) P (atm) Interface composition
CH
4
,e

CO
2
,e

CH
4
,cone

CO
2
,cone

CH
4
,ctwo

CO
2
,ctwo
y
CH
4
y
CO
1 703.45 7.91 0.1774 0.1773 0.9277 0.0440 0.6699 0.0090 0.7192 0.0699
2 702.65 14.22 0.2508 0.1672 0.6992 0.0084 0.6128 0.0098 0.6696 0.0926
3 704.35 10.28 0.2739 0.1499 0.6679 0.0255 0.6490 0.0183 0.7022 0.0824
4 703.15 11.86 0.2816 0.1446 0.8273 0.0198 0.6351 0.0163 0.6901 0.0816
5 738.35 12.05 0.2836 0.1560 0.5774 0.0151 0.5558 0.0085 0.5832 0.0428
6 738.55 8.30 0.2330 0.1730 0.6665 0.0082 0.6081 0.0134 0.6513 0.0727
Table 2
One-dimensional parameter distribution within catalyst pellet for Number 1 (T = 703.45 K, P = 7.91 atm)
x y
CO
2
y
CH
4
y
CO
T (K) r
CH
4
10
3
mol s
1
kg
1
r
CO
2
10
3
mol s
1
kg
1
0.0000 0.1944 0.3443 0.0427 703.79 0.2620 5.4989
0.2154 0.1902 0.3392 0.0486 703.78 0.2862 0.1375
0.4206 0.1846 0.3236 0.0590 703.74 0.3391 0.1907
0.6063 0.1736 0.2977 0.0781 703.69 0.4319 0.3417
0.7635 0.1558 0.2634 0.1066 703.62 0.5625 0.4908
0.8851 0.1330 0.2256 0.1411 703.55 0.7124 0.7889
0.9653 0.1137 0.1935 0.1704 703.48 0.8398 2.3307
1.0000 0.1118 0.1774 0.1773 703.45 0.8905 18.9060
core center. The direct WGSR dominates between the men-
tioned above regions but its rate is low. The region of direct
reaction is big, while the region of reverse reaction is small.
Internal diffusion is not the major factor leading to low
CO
2
.
The existence of regions of the direct and reverse WGSR
contributes to a greater extent to low
CO
2
, which in turn
causes the low net reaction rate. This low reaction rate does
Table 3
Two-dimensional parameter distribution within catalyst pellet for No.
3-y
CH
4
(T = 704.35 K, P = 10.28 atm)
z x
0.0000 0.2390 0.5262 0.7639 0.9275 1.0000
0.0000 0.4220 0.4183 0.4002 0.3615 0.3085 0.2739
0.2390 0.4175 0.4141 0.3969 0.3597 0.3081 0.2739
0.5262 0.3967 0.3942 0.3809 0.3504 0.3053 0.2739
0.7639 0.3561 0.3548 0.3474 0.3289 0.2981 0.2739
0.9275 0.3052 0.3051 0.3027 0.2966 0.2852 0.2739
1.0000 0.2739 0.2739 0.2739 0.2739 0.2739 0.2739
Table 4
Two-dimensional parameter distribution within catalyst pellet for No.
3-r
CO
2
10
3
(T = 704.35 K, P = 10.28 atm)
z x
0.0000 0.2390 0.5262 0.7639 0.9275 1.0000
0.0000 18.909 9.619 8.904 8.136 9.467 52.009
0.2390 10.049 0.155 0.206 0.557 0.812 52.009
0.5262 9.140 0.306 0.333 0.641 0.767 52.009
0.7639 8.375 0.687 0.678 0.905 0.604 52.009
0.9275 10.533 16.161 1.648 1.513 3.006 52.009
1.0000 52.009 52.009 52.009 52.009 52.009 52.009
not imply that WGSR hardly occurs in the catalyst pellet due
to internal diffusion. WGSR occurs remarkably at all sites of
catalyst pellet, and the reaction rate of each site is not equal
to zero. As a result of the existence of direct and reverse
reaction of WGSR, and also the volume-decreasing metha-
nation reaction, there is small variation of carbon dioxide
concentration due to the site change in the pellet.
It is shown in Tables 2 and 5 that the methanation reac-
tion takes place as a direct reaction throughout the whole
internal-surface of the pellet. Carbon monoxide, which is
the product of the reverse WGSR, promotes the methanation
reaction. The interaction between the methanation reaction
and the WGSR enhances the methanation reaction and also
improves
CH
4
. Internal diffusion also affects the methana-
tion reaction.
To put it briey, the effectiveness factors for parallel
equilibrium-restrained reactions cannot describe the utiliza-
tion degree of catalyst internal-surface accurately. Therefore,
the intraparticle distributions of the temperature, the concen-
trations of species and so on should be taken into account.
Table 5
Two-dimensional parameter distribution within catalyst pellet for No.
3-r
CH
4
10
3
(T = 704.35 K, P = 10.28 atm)
z x
0.0000 0.2390 0.5262 0.7639 0.9275 1.0000
0.0000 0.1672 0.1891 0.2690 0.4486 0.7055 0.8575
0.2390 0.1922 0.2129 0.2889 0.4617 0.7112 0.8575
0.5262 0.2844 0.3015 0.3621 0.5058 0.7256 0.8575
0.7639 0.4739 0.4849 0.5206 0.6098 0.7616 0.8575
0.9275 0.7212 0.7260 0.7378 0.7682 0.8251 0.8575
1.0000 0.8575 0.8575 0.8575 0.8575 0.8575 0.8575
38 G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339
5. Conclusions
Based on the studies on the KD306-type sulfur-resisting
methanation catalyst, the non-isothermal one-dimensional
and two-dimensional reaction-diffusion models for the
key-components have been established, which were solved
using an orthogonal collocation method. The simulation val-
ues of the effectiveness factors for the methanation reaction

CH
4
and the shift reaction
CO
2
are in fair agreement with
the experimental values, which indicates that both mod-
els are able to predict intraparticle transport and reaction
processes within catalyst pellets.
The effectiveness factor
CH
4
of the methanation reac-
tion is large, while the effectiveness factor
CO
2
of the
WGSR is very small and nearly equal to zero. The inter-
action between the methanation reaction and the WGSR
leads to the existence of the direct and reverse WGSR
in catalyst pellet, which is called equilibrium-restrained
WGSR.
The effectiveness factors calculated for the parallel,
equilibrium-restrained reaction systems are not able to
predict the catalyst internal-surface utilization accurately.
Therefore, the intraparticle distributions of the temperature,
the concentrations of species and so on should be taken into
account.
Appendix A. Notations
A, B, AX, transition matrices
BX, BZ
C
i
concentration of component i (mol/m
3
)
C
is
interface concentration of component
i (mol/m
3
)
D
Aj
binary diffusion coefcient of a mixture A
and j (m
2
/s)
D
Ak
Knudsen diffusion coefcient of
component A (m
2
/s)
D
effA
effective diffusivity of component A (m
2
/s)
f(r) aperture distribution function
H
CH
4
heat of methanation reaction (J/mol)
H
CO
2
heat of WGSR (J/mol)
H
p
semi-length of pellet (m)
J
P
, K
P
partial pressure ratio and equilibrium
constant of WGSR, respectively
N
R
, N
A
numbers of radial and longitudinal
collocation point, respectively
R
p
radius of pellet (m)
(r
CH
4
)
w
, methanation reaction rate and WGSR rate,
(r
CO
2
)
w
respectively (mol s
1
kg
1
)
T
s
interface temperature (K)
T temperature (K)
x dimensionless coordinate along radial
distance
y
i
mole-fraction of component i
z dimensionless coordinate along
longitudinal distance
Greek letters
tortuosity

CO
2
,cone
predicted effectiveness factor of WGSR of
one-dimensional model

CO
2
,ctwo
predicted effectiveness factor of WGSR of
two-dimensional model

CO
2
,e
experimental effectiveness factor of WGSR

CH
4
,cone
predicted effectiveness factor of methanation
of one-dimensional model

CH
4
,ctwo
predicted effectiveness factor of methanation
of two-dimensional model

CH
4
,e
experimental effectiveness factor of
methanation
porosity

fm
gas heat conductivity in the pore (W/(mK))

s
heat conductivity of solid (W/(mK))

eff
thermal conductivity (W/(mK))

p
grain density (kg/m
3
)
References
[1] G.W. Roberts, The selectivity of porous catalysts: parallel reactions,
Chem. Eng. Sci. 27 (1972) 13481420.
[2] G.W. Roberts, H.H. Lamb, The effect of reversibility on the selectivity
of parallel reactions in a porous catalyst, Chem. Eng. Sci. 51 (1996)
441448.
[3] G.D. Papakonstantopoulos, G.P. Androutsopoulos, C.J. Philippopou-
los, Reaction selectivity in a porous catalyst pellet: analysis of a
kinetic model of two parallel, rst order, irreversible reactions with
a second order inhibition kinetic term in one of them, Chem. Eng.
Sci. 56 (2001) 54135417.
[4] I. Copelowitz, R. Aris, Communications on the theory of diffusion
and reaction-VI. The effectiveness of spherical catalyst particles in
steep external gradients, Chem. Eng. Sci. 25 (1970) 885.
[5] R.S. Dinit, L.L. Tavlaridis, Integral method of analysis of
FischerTropsch synthesis reactions in a catalyst pellet, Chem. Eng.
Sci. 37 (1982) 539.
[6] M.-S. Xu, Z.-M. Du, X.-F. Gao, False transient method for determin-
ing effectiveness factors of high-temperature shift reaction catalyst
B109, J. Chem. Ind. Eng. (China) 44 (1993) 465.
[7] J.P. Sorensen, W.E. Stewart, Collocation analysis of multicomponent
diffusion and reactions in porous catalysts, Chem. Eng. Sci. 37 (1982)
1103.
[8] A. Madzvamuse, A.J. Wathen, P. Maini, A moving grid nite element
method applied to a model biological pattern generator, J. Comput.
Phys. 190 (2003) 478500.
[9] G.-S. Yu, Study on Town Gas Methanation of KD306-type
Sulfur-tolerant Catalyst [D] (China), East China University of Sci-
ence and Technology, Shanghai, 1998.
[10] G.-S. Yu, J.-G. Yu, Z.-H. Yu, The measurement of effective diffusivity
for sulfur-tolerant methanation catalyst, Chem. Eng. J. 78 (2000)
141146.
[11] G.-S. Yu, J.-G. Yu, Q.-Z. Tan, Z.-H. Yu, Study on thermal conduc-
tivity for sulfur-tolerant methanation catalyst, Chem. Eng. (China)
28 (1) (2000) 2224, 32.
[12] Z.-H. Yu, Hydrocarbon-steam Conversion Engineering, Hydrocarbon
Processing Press, Beijing, 1989 (in Chinese).
G. Yu et al. / Chemical Engineering and Processing 44 (2005) 3339 39
[13] C.R. Wilke, Diffusional properties of multicomponent gases, Chem.
Eng. Progr. 46 (1950) 95.
[14] W. Woodside, J.H. Messmer, Thermal conductivity of porous
media I. Unconsolidated sands, J. Appl. Phys. 32 (1961)
1688.
[15] B.A. Finlayson, Orthogonal collection in chemical reaction engineer-
ing, Cat. Rev.-Sci. Eng. 10 (1974) 69138.
[16] Z.-X. Xu, Proximate Solution of Differential Equation, East China
Institute of Science and Technology Press, Shanghai, 1990 (in Chi-
nese).
[17] R. Aris, On shape factors for irregular particles-I. The steady state
problem. Diffusion and reaction, Chem. Eng. Sci. 6 (1957) 262.
[18] D.-G. Liu, J.-G. Fei, Y.-J. Yu, FORTRAN Arithmetic Compilation,
National Defence Industry Press, Beijing, 1980 (in Chinese).
Chemical Engineering and Processing 44 (2005) 4149
PID controller tuning using mathematical programming
George Syrcos
a
, Ioannis K. Kookos
b,
a
Controls Department, TEI Pireaus, 250 Petrou Ralli & Thivon Ave, 12244 Athens, Greece
b
Department of Chemical Engineering, UMIST, PO Box 88, Manchester M60 1QD, UK
Received 19 December 2003; received in revised form 18 February 2004; accepted 1 April 2004
Available online 18 May 2004
Abstract
Proportional integral derivative (PID) controllers are extensively used in the process industry. As a result a large number of general purpose
tuning methodologies are available. These tuning methodologies can offer initial estimates of the parameters of the PID controllers. However,
the design objectives used for the development of these tuning methods can be quite different from the performance objectives specic to a
process under investigation. As a result, the control engineer often needs customized tuning methods in order to speed up or even eliminate the
retuning procedure, and thus, minimize the time and effort needed to design a satisfactory closed loop system. This paper presents a general
mathematical programming formulation for the development of customized PID controller tunings. A reformulation of the mathematical
formulation is proposed that transforms the initially nonlinear formulation to a linear one that can be solved to global optimality. A number
of case studies are presented to clarify the proposed methodology.
2004 Elsevier B.V. All rights reserved.
Keywords: PID-controller tuning; Mathematical programming
1. Introduction
Proportional integral derivative (PID) controllers are ex-
tensively used in the process industry. It was recently esti-
mated that more than 90% of all control loops involve PID
controllers [1]. The simplicity and transparency of PID con-
trol mechanism, the availability of a large number of highly
efcient, reliable, and cost-effective commercial PID con-
trol modules and their acceptance from the operators are
among the reasons for their success [2]. Tuning of the PID
controllers is achieved in two steps. In the rst step a simpli-
ed model of the process is obtained and initial tunings are
calculated using well-known tuning rules (such as the mod-
ied Ziegler-Nichols [2] or IMC-based [3]). In the second
step, the performance of the loop is assessed and the initial
tunings are modied in order to take into account the oper-
ating specications of the process. The success or failure of
this procedure strongly depends on the judicious choice of
initial tunings. However, all available tuning methods have
been developed to satisfy specic design objectives that, in
most cases, cannot be translated into the design objectives of

Corresponding author. Tel.: +44-1612004346; fax: +44-1612004399.


E-mail address: i.kookos@umist.ac.uk (I.K. Kookos).
a specic application. Thus, control engineers often face the
need to develop customized tuning rules that speed up the
controller tuning procedure by minimizing the time needed
for retuning or even eliminating retuning.
The aim of this paper is to develop mathematical pro-
gramming formulations for the development of tuning rules
for PID controllers with arbitrary design objectives posed in
the time domain. The resulting mathematical formulation is
a nonlinear programming problem (NLP). A reformulation
of the problem is proposed that transforms the formulation
to a mixed-integer linear problem. If the objective function
chosen is linear (convex in general) then the resulting for-
mulation can be solved to global optimality using currently
available software. The main advantage of the proposed
methodology stems from its generality since it can be ap-
plied to obtain PID controllers tunings for any linear system.
2. PID controller tuning
2.1. Tuning of proportional only (P-only) controllers
In this work we consider single-input, single output
(SISO) linear time invariant (LTI) systems described by the
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.001
42 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149
following discrete-time state space model
x
k+1
= Ax
k
+bu
k
+Ev
k
y
k
= Cx
k
+du
k
+Fv
k
(1)
where x is the n vector of state variables, u the input variable,
v the m vector of disturbances, y the output variable, and
k is the sampling time. First, we consider the general class
of proportional-only feedback controllers. In this case, the
control moves are related to the deviation from the set point
through the following:
u
k
= k
C
e
k
, k (2)
e
k
= r
k
y
k
, k (3)
where r is the reference signal and k
C
the controller gain.
The design objective is to select the controller gain so as to
minimize a scalar performance index (which is a function
of the output and control variables) subject to constraints on
the input and output variables, i.e.
min
k
C
=
N

k=1
J[x
k
, u
k
], s.t.
_

_
x
k+1
= Ax
k
+bu
k
+Ev
k
y
k
= Cx
k
+du
k
+Fv
k
e
k
= r
k
y
k
u
k
= u
k1
+u
k
u
k
= k
C
e
k
y
L
k
y
k
y
U
k
u
L
k
u
k
u
U
k
u
L
u
k
u
U
(4)
Problem (4) is a nonconvex, nonlinear programming prob-
lem. The nonconvexities are due to the bilinear form of the
control law. This term can be linearized at the expense of
optimizing over a discrete (rather than continuous) set of
controller gains.
First, we assume that an initial controller gain k
C,0
is
available and calculated using classical control techniques.
Then we consider a discrete set of q tuning factors

that
lie in the interval [
L
,
U
] and dened as follows:

=
L
+
_
1
q 1
_
(
U

L
), = 1, 2, . . . , q (5)
Furthermore, we dene q binary variables

{0, 1} to
denote the selection of tuning factor

= 1) or not (

=
0). Then, the control gain can be expressed as follows:
k
C
=
q

=1

(6)
q

=1

= 1 (7)
where

= k
C,0

. Substituting Eq. (6) into Eq. (2) we


obtain
u
k
=
_
q

=1

_
e
k
, k (8)
Finally, we dene the variable E
,k
to be equal to the product
of the binary variable

and the continuous variable e


k
, i.e.
E
,k
=

e
k
, , k (9)
and Eq. (8) can be written in the following linear form
u
k
=
q

=1

E
,k
, k (10)
Then, the nonconvex Eq. (9) can be replaced by the fol-
lowing set of equivalent linear inequalities [4,5]
e
L

E
,k
e
U

(11a)
e
k
e
U
(1

) E
,k
e
k
e
L
(1

) (11b)
It should be noted that when

= 0 then E
,k
= 0, k,
while when

= 1 E
,k
= e
k
, k. As a result, the equiva-
lence of Eqs. (9), (11a), and (11b) can be established easily.
In summary, the nonconvex formulation given by Eq. (4)
can be replaced by the following linearly constrained for-
mulation
min

=
N

k=0
J[x
k
, u
k
],
s.t.
_

_
x
k+1
= Ax
k
+bu
k
+Ev
k
y
k
= Cx
k
+du
k
+Fv
k
e
k
= r
k
y
k
u
k
= u
k1
+u
k
u
k
=

q
=1

E
,k
e
L

E
,k
e
U

,
e
k
e
U
(1

) E
,k
e
k
e
L
(1

),

q
=1

= 1
y
L
k
y
k
y
U
k
u
L
k
u
k
u
U
k
u
L
u
k
u
U

{0, 1} ,
(12)
Depending on the form of the objective function, formu-
lation (12) corresponds to a mixed integer, nonlinear pro-
gramming (MINLP) problem if the objective is nonlinear,
to a mixed integer quadratic programming (MIQP) problem
if the objective is quadratic or to a mixed integer linear pro-
gramming (MILP) problem if the objective is linear. In this
paper consideration is restricted to the case of convex objec-
tive functions for which the global optimum solutions can
be found using available algorithms [4].
G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149 43
2.2. Tuning of proportional-integral (PID) controllers
The PID control law considered here contains the P, PI,
PD, and PID control laws as special cases. The velocity form
of the discrete approximation of an ideal PID controller is
given by [2]
u
k
=k
C
__
1 +

S

I
+

D

S
_
e
k

_
1 +
2
D

S
_
e
k1
+

D

S
e
k2
_
(13)
where
S
is the sampling time, k
C
the controller gain,
I
the
integral time, and
D
is the derivative time. We dene the
following variables:
P
k
= e
k
e
k1
(14)
I
k
=
S
e
k
(15)
D
k
=
1

S
(e
k
2e
k1
+e
k2
) (16)
and Eq. (13) can be written as
u
k
= K
P
P
k
+K
I
I
k
+K
D
D
k
(17)
K
P
= k
C
, K
I
=
k
C

I
, K
D
= k
C

D
(18)
The complete formulation of the closed loop system op-
timization using a PID controller is as follows
min
K
P
,K
I
,K
D
=
N

k=0
J[x
k
, u
k
],
s.t.
_

_
x
k+1
= Ax
k
+bu
k
+Ev
k
y
k
= Cx
k
+du
k
+Fv
k
e
k
= r
k
y
k
u
k
= u
k1
+u
k
u
k
= K
P
P
k
+K
I
I
k
+K
D
D
k
P
k
= e
k
e
k1
I
k
=
S
e
k

S
D
k
= e
k
2e
k1
+e
k2
y
L
k
y
k
y
U
k
u
L
k
u
k
u
U
k
u
L
u
k
u
U
K
P,L
K
P
K
P,U
K
I,L
K
I
K
I,U
K
D,L
K
D
K
D,U
(19)
This is a nonlinear formulation due to the bilinear terms
involved in Eq. (17) and the (possibly) nonlinear objective.
Following the same procedure as in the previous subsection
we can replace Eq. (17) by a set of linear inequalities.
We rst assume that initial values of the controller param-
eters have been obtained using any classical control tech-
nique. We then introduce tuning factors
P

,
I

and
D

as
well as the binary variables
P

,
I

, and
D

to denote the se-


lection -th term of the proportional, integral and derivative
term of the PID controller, i.e.
K
m
=
q
m

=1

, m (20)
q
m

=1

= 1, m (21)

= k
m
C,0

, , m (22)

,k
=
P

P
k
, , k (23)
O
,k
=
I

I
k
, , k (24)

,k
=
D

D
k
, , k (25)
where m {P, I, D}. Eq. (17) can then be written in the
following linear form:
u
k
=
q
P

=1

,k
+
q
I

=1

O
,k
+
q
D

=1


,k
, k (26)
P
L


,k
P
U

P
k
P
U
(1
P

) E
,k
P
k
P
L
(1
P

)
_
, , k
(27)
I
L

O
,k
I
U

I
k
I
U
(1
I

) O
,k
I
k
I
L
(1
I

)
_
, , k
(28)
D
L


,k
D
U

D
k
D
U
(1
D

)
,k
D
k
D
L
(1
D

)
_
, , k
(29)
Depending on the form of the objective function, the nal
formulation obtained by replacing the nonlinear Eq. (17) by
the set of linear inequalities corresponds to a MINLP (non-
linear objective), to a MIQP (quadratic objective) or to a
MILP (linear objective). For the cases where the objective
function is linear, solution to global optimal solution is guar-
anteed using currently available software. The same holds
true for the more general case where the objective function
is a convex function.
The proposed formulation addresses the problem of PID
controller design solely in the time domain. It is important to
44 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149
emphasize that no approximations have been introduced in
the proposed formulation. This is not the case, for instance,
when the design is based on the frequency domain where
conservative (in most cases) approximations are introduced
in order to translate constraints fromthe time domain into the
frequency domain [6,7]. A further advantage of the proposed
formulation is the fact that the problem can be solved to
global optimality using direct optimization. Classical control
design techniques do not share these desirable characteristics
[6].
3. Case studies
In order to apply the proposed methodology the control
engineer needs to supply the model of the process under
investigation in discrete form (see Eq. (1)), an initial con-
troller tuning (that can be obtained using any classical tech-
nique) and a performance criterion. Then, the optimization
problem can be solved using any commercial optimization
software. In this work GAMS [8] interface to CPLEX and
MINOS solvers was used to solve the proposed formulation.
All results were obtained on an Intel P4 2.66 GHz/512 Mb
computer and the computational time was of the order of
few seconds.
3.1. Control of integrating processes with time delay
(IPTD)
Integrating processes with time delay are among the most
commonly encountered processes. Liquid levels, for in-
stance, exhibit integrating behavior. A typical plant involves
a large number of liquid level loops whose performance
affects strongly the overall performance of the regulatory
control systems. The transfer function of an integrating
process with time delay is the following:
g(s) =
k
P
s
e
ds
(30)
PID controller tunings for this model have been given
by a number of researchers [913]. Chen and Fruehauf
[9] have given an industrial example of the level control
in a distillation column where the open loop dynamics
follows the IPTD model with parameters k
P
= 0.2 and
d = 7.4 min.
The proposed methodology has been applied using the
following objective function:
min
K
P
,K
I
=
S
N

k=0
(|e
k
| + |u
k
|) (31)
where N
S
=
f
= 300 min. A unit step disturbance at t =
0 min was assumed followed be a unit step change in the set
point at t = 150 min. The optimization problem was solved
with = 1/
S
and
S
= 0.2 min giving
K
P
= 0.553, K
I
= 0.0178 (32)
Time (min)
-1
-1.5
0
-2
3
2
1
0
150 100 50 0
-1
y(t)
0
-0.5
u(t)
50 100 150
Time (min)
Fig. 1. Load response of the closed loop IPDT system (proposed: con-
tinuous line, Luyben [10]: dotted line).
Figs. 1 and 2 compare the closed loop behavior obtained
using the tuning proposed in this work and the tunings
proposed by Tyreus and Luyben [10]. Fig. 1 shows that a
signicant improvement in the closed loop load response is
obtained when the proposed methodology is used. The value
of the objective function (Eq. (31)) is = 59.4 for the pro-
posed and = 197.7 for the Luybens tunings. Fig. 2 com-
pares the two methodologies for a unit step set point change.
The responses are comparable and the proposed method
results in = 28.6 while for Luybens tunings = 29.7.
3.2. Systems with inverse response and time delay
Luyben [14] has recently investigated a class of mod-
els that involve both inverse response and time delay. More
specically he has considered the following type of pro-
cesses:
g(s) = k
P
1
z
s
(1 +
P
s)
2
e
ds
(33)
that have a positive zero at s = +1/
z
as well as time a
delay d. Using the objective function given by Eq. (31) with
= 1/
S
and
S
= 0.1 we obtained
K
P
= 0.356, K
I
= 0.154 (34)
for a hypothetical system with
z
= d = 1.6, k
P
= 1 and

P
= 1. A unit step disturbance at t = 0 min was assumed
followed be a unit step change in the set point at t = 50 min.
Fig. 3 compares the closed loop response obtained using the
proposed tunings with the tunings given by Luyben [14].
Luybens tunings correspond to = 19.5 while the ones
proposed in this work to = 16.7. It is interesting to note
that the proposed tunings result in improved load as well as
set point response.
G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149 45
Time (min)
1
0
-0.5
150 100 50 0
2
1.5
1
0.5
0
150 100 50 0
y(t)
0.5
u(t)
Time (min)
Fig. 2. Unit step set point response for IPDT system (proposed: continuous line, Luyben [10]: dotted line).
3.3. Control of integrating processes with time delay and
inverse response (IPTD&IV)
The open loop transfer function of an IPDT&IV process
is the following:
g(s) = k
P
1
z
s
s(1 +
P
s)
e
ds
(35)
and combines the adverse effects associated with having a
right half plane zero at 1/
z
, a pole at zero and a time delay
d. Recently, Luyben [15] proposed a successful and simple
enough tuning techniques for PID controllers and IPDT&IV
systems.
0 10 20 30 40 50 60 70 80 90 100
0
0.5
1
0 10 20 30 40 50 60 70 80 90 100
-1.5
-1
-0.5
0
0.5
y(t)
Time
u(t)
Time
Fig. 3. Closed loop responsecase study 2 (proposed: continuous line, Luyben [14]: dotted line).
In Figs. 4 and 5 the closed loop response obtained using
the tunings proposed by Luyben (PI controller with k
C
=
0.854 and
I
= 23) is compared with the one obtained using
the criterion given by Eq. (31) (k
C
= 1.117 and
I
= 6.32)
with = 1/
S
and
S
= 0.025 for the system:
g(s) = 0.457
1 0.418s
s(1 +1.06s)
e
0.1s
(36)
It is clear that for the set point change the two method-
ologies offer comparable performance ( = 7.4 for the
proposed and = 6.6 for Luybens method). However, as it
becomes clear from inspecting Fig. 4, the proposed method-
ology offers massive improvement for the load response
46 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149
0 10 20 30 40 50 60 70 80 90 100
-0.5
0
0.5
1
1.5
0 10 20 30 40 50 60 70 80 90 100
-2
-1.5
-1
-0.5
0
0.5
Time (min)
Time (min)
y(t)
u(t)
Fig. 4. Load response of the closed loop IPDT&IV system (proposed:
continuous line, Luyben [15]: dotted line).
case. The value of the objective function for the Luybens
method ( = 28.2) is 3.5 times the value of the objective
function obtained using the proposed method ( = 8.1).
3.4. Control of an unstable process
Most chemical processes exhibit stable open-loop behav-
ior. However, there are some important processes, such as
chemical and biological reactors, that might be operated
around an unstable steady state. The tuning of PID controller
for unstable systems has recently attracted attention [1619].
Lee et al. [18] have presented a powerful IMC-based tun-
ing methodology that appears to surpass most of available
tuning techniques for unstable systems with low order dy-
0 2 4 6 8 10 12 14 16 18 20
-0.05
0
0.05
0.1
0.15
0.2
0 2 4 6 8 10 12 14 16 18 20
-2
-1.5
-1
-0.5
0
Time
Time
y(t)
u(t)
Fig. 6. Load response of the open-loop unstable system (proposed: continuous line, Lee et al. [18]: dotted line).
0 5 10 15 20 25 30 35 40 45 50
0
0.5
1
1.5
0 5 10 15 20 25 30 35 40 45 50
-0.5
0
0.5
1
Time (min)
Time (min)
y(t)
u(t)
Fig. 5. Unit step set point response for IPDT&IV system (proposed:
continuous line, Luyben [15]: dotted line).
namics. They have, among other case studies, considered
the following system:
g(s) =
e
0.5s
(5s 1)(2s +1)(0.5s +1)
(37)
and have demonstrated clearly the superiority of their
method over most of the available tuning techniques. The
same process was considered here with the objective func-
tion given by Eq. (31) ( = 0.1 and
S
= 0.01) and the
proposed method gave the following tunings:
K
P
= 6.905, K = 1.707, K
D
= 11.824 (38)
which are very close to the result reported by Lee et al.
[18]. In Fig. 6, the closed loop response obtained using the
G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149 47
0 10 20 30 40 50
0
0.2
0.4
0.6
0.8
1
0 10 20 30 40 50
-1
-0.8
-0.6
-0.4
-0.2
0
Time
Time
y(t)
u(t)
Fig. 7. Load response of the FOPTD/PI system (proposed: continuous line, Tang et al. [20]: dotted line).
proposed methodology is compared to that obtained using
the PID controller parameters reported in [18]. The improve-
ment obtained using the proposed methodology is again
clear.
3.5. Control of rst order processes with large time delay
(FOPTD)
The transfer function of a rst order plus time delay pro-
cess is given by
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
0 5 10 15 20 25 30
-1
-0.8
-0.6
-0.4
-0.2
0
Time
Time
y(t)
u(t)
Fig. 8. Load response of the FOPTD/PID system (proposed: continuous line, Tang et al. [20]: dotted line).
g(s) =
k
P

P
s +1
e
ds
(39)
By dening the dimensionless parameter = d/
P
Eq. (39) can be written in the following form:
g(s) =
k
P
s

+1
e
s

(40)
where s

corresponds to time scaled with respect to


P
.
has been used as a measure of the difculty in controlling
FOPTD systems [2]. Systems with close to zero are easy
48 G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
0 5 10 15 20 25 30
-1
-0.8
-0.6
-0.4
-0.2
0
Time
Time
y(t)
u(t)
=2 =3 =4 =5
=2 =3 =4 =5 q
q q q q
q q q
Fig. 9. Load response of the FOPTD/PID system for different values of .
to control while when approaches or exceeds 1 then the
corresponding system becomes progressively more difcult
to control. A process with = 4 and unit gain is selected
in this case study. The optimization problem (using the ob-
jective function given by Eq. (31)) was solved for a PI con-
trol structure and a unit step load disturbance introduced at
t = 0. The solution obtained is
K
P
= 0.42, K
I
= 0.163 (41)
0 10 20 30 40 50 60
-0.5
0
0.5
1
0 10 20 30 40 50 60
-1.5
-1
-0.5
0
Time
Time
y(t)
u(t)
Fig. 10. Load response of the FOPTD/PID system using the proposed (continuous line), CM (dash dotted line), and the IMC method (dotted line) for
= 10.
Tang et al. [20] have also investigated the same case
study. The closed loop performance obtained using the two
methodologies is compared in Fig. 7. A clear improvement
in the performance is observed from Fig. 7 when the pro-
posed technique is used. In Fig. 8 the same comparison is
shown for the case of PID control.
FOPTD with large time delay systems is an important
class of systems for the process industries. A number of
techniques are indeed available for tuning PID controllers
G. Syrcos, I.K. Kookos / Chemical Engineering and Processing 44 (2005) 4149 49
that feature some form of predictive capability. However, a
limited number of tuning techniques are available for sim-
ple PID controllers. The proposed mathematical formulation
was solved for a number of process models with [2, 5]
for unit load disturbances and the following tuning rules
were derived:
k
C
k
P
= 0.31 +0.6
_
1

_
(42)

P
= 0.777 +0.45 (43)

P
= 0.44 0.56
_
1

_
2.2
(44)
In Fig. 9 a number of responses to a unit step load dis-
turbances are shown. The performance of the closed loop
system is acceptable since the controller eliminates quickly
the deviation with a reasonable amount of oscillation.
Based on the success of the proposed tuning formulas
(Eqs. (42)(44)) within the limits of for which they were
derived an extreme test was used to investigate its applicabil-
ity by selecting a process with = 10. Luyben [21] has also
applied different PID controller forms and tuning techniques
such as the IMC method [7] and a method reported by Mar-
lin [22] (CianconeMarlin or CM method) to this process
and the reader is referred to [21] for an interesting discus-
sion on the subject. The different closed loop responses to a
unit step load disturbance are compared in Fig. 10. For the
CM method the PID2 from [21] was used to derive the sim-
ulation results (k
C
= 0.3,
I
= 5,
D
= 2.7, a = 0.1). For
the IMC method the PID3 from [21] was used to derive the
simulation results (k
C
= 0.48,
I
= 6,
D
= 0.83,
F
= 1).
The proposed tuning formulas offer, to our surprise, an im-
proved closed loop response despite the fact that they were
applied to a case which is well outside the limits for which
they were derived.
4. Conclusions
This paper presents a general mathematical programming
formulation the can be used to obtain customized tuning for
PID controllers. A reformulation of the initial NLP prob-
lem is presented that transforms the nonlinear formulation
to a linear one. In the cases where the objective function is
convex then the resulting formulation can be solved easily
to global optimality. The usefulness of the proposed formu-
lation is demonstrated in ve case studies where some of
the most commonly used models in the process industry are
employed. It was shown that the proposed methodology of-
fers closed loop performance that is comparable to the one
achieved by using the best available tunings for a number
of simplied but yet challenging process models.
References
[1] K.J. strm, T. Hgglund, The future of PID control, Control Eng.
Practice 9 (2001) 11631175.
[2] K.J. strm, T. Hgglund, PID Controllers, second ed., Instrument
Society of America, New York, 1995.
[3] S. Skogestad, Simple analytic rules for model reduction and PID
controller tuning, J. Process Control 13 (2003) 291309.
[4] C.A. Floudas, Nonlinear and Mixed-Integer Optimisation, Oxford
University Press, New York, 1995.
[5] H.P. Williams, Model Building in Mathematical Programming, third
ed., Wiley, Chichester, 1997.
[6] S. Skogestad, I. Postlethwaite, Multivariable Feedback Control, Wi-
ley, Chichester, UK, 1996.
[7] M. Morari, E. Zariou, Robust Process Control, Prentice Hall, New
York, 1989.
[8] A. Brooke, D. Kendrick, A. Meeraus, R. Raman, GAMS, a
Users Guide, GAMS Development Corporation, Washington, DC,
1998.
[9] I. Chen, P.S. Fruehauf, Consider IMC tuning to improve controller
performance, Chem. Eng. Prog. 10 (1990) 3341.
[10] B. Tyreus, W.L. Luyben, Tuning PI controllers for integrator/dead
time processes, Ind. Eng. Chem. Res. 31 (1992) 26252628.
[11] I.K. Kookos, A.I. Lygeros, K.G. Arvanitis, On-line pi controller
tuning for integrator/dead time processes, Eur. J. Control 5 (1999)
1931.
[12] W. Zhang, X. Xu, Y. Sun, Quantitative performance design for
integrating processes with time delay, Automatica 35 (1999) 719
723.
[13] M. Chidambaram, R. Padma Sree, A simple method of tuning PID
controllers for integrator/dead-time processes, Comput. Chem. Eng.
27 (2003) 211215.
[14] W.L. Luyben, Tuning of proportional-integral controllers for pro-
cesses with both inverse response and deadtime, Ind. Eng. Chem.
Res. 39 (2000) 973976.
[15] W.L. Luyben, Identication and tuning of integrating processes with
deadtime and inverse response, Ind. Eng. Chem. Res. 42 (2003)
30303035.
[16] G.E. Rotstein, D.R. Lewin, Simple PI and PID tuning for open-loop
unstable systems, Ind. Eng. Chem. Res. 30 (1991) 18641875.
[17] H.P. Huang, C.C. Chen, Control system synthesis for open-loop
unstable processes with time delay, IEE Control Theory Appl. 144
(1997) 334338.
[18] Y. Lee, J. Lee, S. Park, PID controller tuning for integrating and
unstable processes with time delay, Chem. Eng. Sci. 55 (2000) 3481
3493.
[19] V. Saraf, F. Zhao, B.W. Bequette, Relay autotuning of
cascaded-controlled open-loop unstable reactors, Ind. Eng. Chem.
Res. 42 (2003) 44884494.
[20] W. Tang, S. Shi, M. Wang, Simple autotuning PID control for large
time-delay processes and its application to paper basis weight control,
Ind. Eng. Chem. Res. 41 (2002) 43184327.
[21] W.L. Luyben, Effect of derivative algorithm and tuning selection on
the PID control of dead-time processes, Ind. Eng. Chem. Res. 40
(2001) 36053611.
[22] T.E. Marlin, Process Control, McGraw Hill, New York, 1995.
Chemical Engineering and Processing 44 (2005) 5158
Separation of acetic acidwater mixtures through acrylonitrile
grafted poly(vinyl alcohol) membranes by pervaporation
N. Alghezawi
a
, O. Sanl
a,
, L. Aras
b
, G. Asman
a
a
Gazi Universitesi Fen-Edebiyat Fakltesi Kimya Blm 06500 Teknik Okullar Ankara, Turkey
b
Orta Do gu Teknik Universitesi Fen-Edebiyat Fakltesi Kimya Blm, Ankara, Turkey
Received 19 May 2003; received in revised form 29 March 2004; accepted 30 March 2004
Available online 15 June 2004
Abstract
The pervaporation separation of acetic acidwater mixtures was performed over a range of 1090 wt.%acetic acid in the feed at temperatures
ranging 2550

C using acrylonitrile (AN) grafted poly(vinyl alcohol) (PVA) membranes. The permeation and separation characteristics of
PVA-g-AN membranes were studied as a function of membrane thickness, feed composition, operating temperature and pressure. When the
downstream pressure increased permeation rate increased with decreasing separation factor and at high acetic acid concentrations PVA-g-AN
membranes shows grater tendency for the separation of acetic acidwater mixtures. Depending on the membrane thickness, feed composition
and temperature PVA-g-AN membranes gave separation factors 2.314 and permeation rates 0.181.17 kg/m
2
h. It was also determined that
PVA-g-AN membranes were found to have lower permeation rate and grater separation factors than PVA membranes.
2004 Elsevier B.V. All rights reserved.
Keywords: Membranes; Selectivity; Separations; Transport processes; Pervaporation; Acetic acidwater mixtures; Poly(vinyl alcohol) membranes; Graft
copolymerization
1. Introduction
Pervaporation is a membrane separation process where
the liquid feed mixture is in contact with the membrane in
the upstream under atmospheric pressure and permeate is re-
moved from the downstream as vapor by vacuum or a swept
inert gas. Most of the research efforts of the pervaporation
have concentrated on the separation of alcoholwater sys-
tem [120] but the separation of acetic acidwater mixtures
has received relatively little attention [2134]. Acetic acid is
an important basic chemical in the industry ranking among
the top 20 organic intermediates. Because of the small dif-
ferences in the volatilitys of water and acetic acid in dilute
aqueous solutions, azeotropic distillation is used instead of
normal binary distillation so that the process is an energy
intensive process. From this point of view, the pervapora-
tion separation of acetic acidwater mixtures can be one of
the alternate processes for saving energy.

Corresponding author. Tel.: +90-312-2122900;


fax: +90-312-2122279.
E-mail address: osanli@gazi.edu.tr (O. Sanl).
PVA is a possible candidate for the separation of aque-
ous mixtures [5,7,1214,2125,35] because of its good
chemical stability, lm forming ability and high hy-
drophilicity, except for its poor stability in aqueous solu-
tion. Therefore PVA must be insolubilized by crosslink-
ing, grafting or other modication reactions to create
a stable membrane with good mechanical properties
and selective permeability to water. Huang and Yeom
[23,24] studied the separation of ethanolwater and acetic
acidwater mixtures using amic acid as the crosslink-
ing agent. Yeom and Lee [36] prepared PVA membranes
by crosslinking with gluteraldehyde for the pervapo-
ration of acetic acidwater mixtures. Aminabhavi and
Naik [37] prepared poly(vinyl alcohol)-g-poly(acrylamide)
copolymeric membranes to separate acetic acidwater
mixtures.
Acrylonitrile based membranes were also used in acetic
acidwater separation. Lee and Oh [11] copolymerized
4-vinylpyridine with acrylonitrile in order to prepare a
membrane for the dehydration of wateracetic acid mixture
by pervaporation. Yoshikava et al. [25] reported that mem-
branes prepared from poly (acrylic acid-co-acrylonitrile)
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.007
52 N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158
were effective in the selective separation of water from
acetic acidwater mixtures.
In the present study we have aimed to separate acetic
acidwater mixtures using AN grafted PVA membranes and
studied the permeation characteristics of PVA-g-AN mem-
branes as a function of membrane thickness, temperature,
feed composition and pressure.
2. Experimental
2.1. Materials
PVA (Merck) with molecular weight of 72,000 and degree
of saponication of 98.599.2%, was used as supplied, AN
(Merck) was puried by reduced pressure distillation. Ceric
ammonium nitrate (CAN) (Merck) was used after vacuum
drying. Acetic acid (Merck), dimethylformamide (DMF) and
dimethylsulfoxide (DMSO) (Merck) were used without fur-
ther purication. Other reagents used in the study were also
all Merck products.
2.2. Synthesis of PVA-g-AN
The reaction was carried out in a three-necked ask
equipped with stirrer, condenser and nitrogen gas inlet. Five
grams of PVA was dissolved in 95 mL of distilled water at
80

C. Fifteen grams of AN was added to the solution. Stir-


ring was continued for 1 h at room temperature under N
2
atmosphere. Ten mililiter portion of 1.0 10
2
M HNO
3
containing 10 mL, 1.0 10
2
M of CAN was added to the
reaction vessel and polymerization was carried out for a pe-
riod of 3 h at 30

C. The reaction was quenched with 10 mL


of aqueous ferrous salt solution (FeSO
4
) (5.0 10
2
M)
and pH of graft copolymer latexes were adjusted to 1.5 with
1.0 M HCl solution, then dried in air at room temperature.
The product was extracted with water at 80

C then with
DMF at 55

C for 2 days to remove the unreacted PVA and


polyacrylonitrile (PAN) homopolymer. After the extraction,
graft copolymer was dried in vacuum oven at 60

C [5,38].
The proposed grafting mechanism between PVA and AN is
given in Fig. 1.
2.3. Preparation of membranes
1.2% (w/v) solution of PVA-g-AN in DMSO were
prepared. The solutions of 2030 mL were poured on to
petri-dishes (9.2 cm in diameter) and the solvent was evap-
orated at 80

C to form the membrane (15 m40 m).


2.4. Pervaporation experiment
The apparatus used in this study were illustrated in Fig. 2.
The pervaporation cell was a two compartment cell with a
150 mL upper compartment, 75 mL of lower compartment.
Effective membrane area was 12.5 cm
2
. The pressure at the
Fig. 1. Proposed reaction mechanism between PVA and AN for the
synthesis of PVA-g-AN copolymer.
downstream side was kept at 23 Pa by a vacuum pump
(Vacuu Brand RD 15, GMBH Co., Germany). Pervapora-
tion experiments were conducted at constant temperatures
ranging 2550

C using different feed compositions. Upon


reaching steady state ow conditions, the permeate vapor
was collected in liquid nitrogen traps.
The composition of permeates was deduced refractomet-
rically by digital differential refractometer (Atago DD-5,
Atago Co. Ltd., Japan).
2.5. Representation of the results
The permeation and separation characteristics of
PVA-g-AN membranes toward acetic acidwater mixtures
were expressed as permeation rate (ux) (Q), separation
factor () and pervaporation separation index (PSI).
The permeation rate, Q was determined by using the equa-
tion
Q =
m
At
(1)
where m, A and t represent the weight of the permeate (kg),
Fig. 2. Schematic diagram of the pervaporation apparatus: (1) vacuum
pump; (24, 6) permeation traps; (5) Mc Leod manometer; (7) vent; (8)
permeation cell; (9) constant temperature water bath.
N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158 53
effective membrane area (m
2
) and operating time (h) respec-
tively.
Separation factor was calculated using:
=
Y
H
2
O
/Y
CH
3
COOH
X
H
2
O
/X
CH
3
COOH
(2)
where X
H
2
O
, X
CH
3
COOH
, Y
H
2
O
and Y
CH
3
COOH
denote the
weight fractions of water and acetic acid in the feed and in
the permeate, respectively.
Pervaporation separation index (PSI), which is a mea-
sure of the separation ability of a membrane was dened by
Huang and Yeom [7] and expressed as the product of sepa-
ration factor and permeation rate.
PSI = Q (3)
3. Results and discussion
3.1. PVA-g-AN copolymer
PVA-g-AN copolymer was synthesized by using PVA and
ANin a weight ratio of (PVA/AN) (1/2.5), (1/3.0) and (1/5.0)
in the polymerization solution; compositions of the copoly-
mer were determined by elemental analysis (Table 1). FTIR
spectrum of the copolymer was taken (Fig. 3). The stretch-
ing vibrations of OH and CH appears at 3340 cm
1
and
29472910 cm
1
, respectively, and stretching vibration of
CN appears at 22402225 cm
1
were taken as the evidence
of grafting.
Table 1
Elemental analysis of PVA-g-AN
Membrane PVA/AN (w/w) C (%) H (%) N (%)
PVA-1 1/2.5 59.71 6.56 13.62
PVA-2 1/3.0 60.47 7.13 13.80
PVA-3 1/5.0 60.77 7.34 16.29
Fig. 3. FTIR spectrum of PVA-g-AN.
Table 2
Effect of membrane thickness on permeation rate and separation factor
Thickness (m) Q (kg/m
2
h)
15 Low mechanical strength
25 1.09 2.30
35 0.66 3.00
40 0.11 5.50
20 wt.% CH
3
COOH; P = 23 Pa, T = 30

C.
3.2. Effect of membrane thickness
Membranes with varying thickness (1540 m) were pre-
pared from the copolymer by casting method. The perme-
ation rate and separation factor as a function of membrane
thickness was studied for 20 wt.% acetic acid solutions at
30

C and the results were presented in Table 2. As reected


from the table as the membrane thickness increases perme-
ation rate decreases whereas separation factor increases as
expected from the Ficks rst law [39].
Different results concerning the effect of membrane thick-
ness on pervaporation performance was reported in the lit-
erature [39,40,42].
Koops et al. [39] investigated the pervaporation selectiv-
ity as a function of membrane thickness for the polysulfone,
poly(vinyl chloride) and poly(acrylonitrile) membranes in
the dehydration of acetic acid and reported that selectiv-
ity decreases with decreasing membrane thickness, below a
limiting value of about 15 m.
Aptel et al. [40] observed a reduction of separation fac-
tor as a function of decreasing membrane thickness for
grafted polytetrauoroethylene lms using a water/dioxane
mixture. They have shown in a different study [41] that
transport rate is inversely proportional to thicknes in
the pervaporation separation of water through poly(tetra
uoroethylene)poly(4-vinylpyridine) membranes.
Brun et al. [42] studied the inuence of membrane thick-
ness on the separation factor using nitrile rubber membranes
for 60/40 mixture of butadiene and isobutene. They have
concluded that the separation factor was constant above
100 m membrane thicknesses.
In this present study membranes of 35 m thicknesses
were preferred in the rest of the study due to their acceptable
ux and separation factor.
Table 3
Comparison of PVA and PVA-g-AN membranes
T (

C) Q (kg/m
2
h)
PVA PVA-g-AN PVA PVA-g-AN
25 0.95 0.18 1.76 2.58
45 4.47 1.10 1.23 3.03
50 5.70 1.17 1.05 3.03
20 wt.% CH
3
COOH; P = 23 Pa, t = 1.5 h.
54 N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158
0
0.1
0.2
0.
0.4
0.5
0 10 30 50 70 90 100
CH
3
COOH (wt %)
Q

(
k
g
/
m
2
h
)
0.0
2.0
4.0
6.0
8.0
10. 0
12. 0
14. 0
16. 0

Fig. 4. Effect of PVA:AN ratio on the permeation rate and separation


factor (T = 30

C): () (PVA-2); () Q (PVA-2); () (PVA-1);


() Q (PVA-1).
Table 4
Effect of temperature on permeation rate and separation factor
T (

C) Q (kg/m
2
h)
25 0.18 2.60
30 0.66 3.00
40 1.07 3.00
50 1.17 3.00
20 wt.% CH
3
COOH; P = 23 Pa, t = 1.5 h.
3.3. Effect of AN grafting on the performance of the PVA
membranes
In order to understand the effect of grafting on the perfor-
mance of PVA membranes, PVA-g-AN (PVA-2) membranes
were compared with the heat treated PVA membranes that
were prepared by casting method from 7% (w/v) PVA solu-
tion (Table 3).
As it is reected fromthe table that the presence of ANde-
creases the permeation rate while increasing the separation
factor. Increase in the separation factor is caused from the
0. 0
0. 1
0. 2
0. 3
0. 4
0. 5
0 10 30 50 70 90 100
CH
3
COOH (wt %)
0. 0
2. 0
4. 0
6. 0
8. 0
10 .0
12 .0
14 .0
16 .0
Q

(
k
g
/
m
2
h
)

Fig. 5. Change of total permeation rate and separation factor with feed composition: () Q (30

C); () Q (40

C); () (30

C); () (40

C).
hydrogen bonding between acetic acid and CN groups of
AN; so it becomes difcult for acetic acid molecules to dif-
fuse through the membrane and separation factor toward wa-
ter increases. Decrease in permeation rate can be explained
by the dense structure of the grafted membrane. In Fig. 4
permeation rate of grafted membranes at different grafting
ratio with respect to the feed composition was given. As it
seen from the graph although permeation rate decreases with
the amount of AN separation factors were almost the same.
3.4. Effect of temperature on pervaporation
The effect of temperature on the separation performance
of PVA-g-AN (PVA-2) membranes was studied in the tem-
perature range of 2550

C and the results were given in


Table 4. As it is reected from the table, as the tempera-
ture increases permeation rate increases, however, separa-
tion factor stays constant above 30

C.
Aminabhavi and Naik [37] obtained similar results in the
pervaporation of acetic acidwater mixtures by using PVA
membranes. They reported that for 20 wt.% acetic acid so-
lutions separation factor stayed constant at 35 and 45

C al-
though permeation rate increased with temperature.
The permeation at various operating temperatures is
thought to be governed by three factors [7,23].
1. Change in the free volume of the polymer membrane
swollen by pure components, which affects mainly per-
meation rates.
2. Change in the free volume due to the plasticizing effect.
3. Change in the interaction between permeants in the mem-
brane.
According to the free volume theory as the temperature
increase the frequency and amplitude of chain jumping (i.e.
thermal agitation) increase and the resulting free volumes
become larger. Increase in temperature also decreases the
interaction between acetic acid and water molecules so it
will be easy for both acetic acid and water molecules to
N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158 55
0
0.1
0.2
0.3
0.4
10 30 50 70 90
CH
3
COOH (wt %)
Q

(
H
2
O
)

(
k
g
/
m
2
h
)
0.0
1.0
2.0
3.0
4.0
5.0
Q
(
C
H
3
C
O
O
H
)

(
k
g
/
m
2
h
)
x
1
0
2
Fig. 6. Effect of feed composition on permeation rate of water and acetic acid: () Q (H
2
O) (30

C); () Q (H
2
O) (40

C); () Q (CH
3
COOH) (30

C);
() Q (CH
3
COOH) (40

C).
diffuse through the free volumes resulted by the increase
in temperature. This situation leads to constant separation
factors.
3.5. Effect of feed composition
At 30 and 40

C the effect of feed concentration, in the


concentration range of 1090 wt.% of acetic acid solutions,
on the permeation rate and the separation factor was inves-
tigated and the results were given in Fig. 5. From the gure
it is clear that when acetic acid concentration in the feed so-
lution increases the permeation rate decreases, whereas the
separation factor increases. It can also be seen that the sep-
aration factors obtained at low temperatures are higher than
those obtained at high operating temperatures and they in-
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30 40 50 60 70 80 90
CH
3
COOH (wt%)
P
S
I
Fig. 7. Change of pervaporation separation index with feed composition (T = 30

C).
crease steadily with the increase in acetic acid content of
feed.
Similar results were obtained in the pervaporation separa-
tion of acetic acidwater mixtures using blended polyacrylic
acidnylon 6 membranes [22], poly(4-vinylpyridine-co-
acrylonitrile) membranes [11] and in the permeation and
separation of aqueous alcohol solutions through PVA-AN-
HEMA latex membranes [5].
However, in the studies of Huang and Yeom [23,24] using
chemically crosslinked PVA membranes, maximum perme-
ation rates were reported at low acetic acid concentrations
(10 wt.%).
Looking at the permeation rate of individual components
(Fig. 6), the permeation rate of water component decreases
with the decrease in the water content of the feed solution
56 N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158
Table 5
Comparison of the performance of the membranes based on pva for the pervaporation separation of acetic acidwater mixtures
Polymer Temperature (

C) Mass % of acetic
acid in the feed
Permeation rate
(kg/m
2
h)
Separation factor () Ref. no.
PVA-g-PAAm 35 80 0.056 3.90 [45]
PVA/PAA (Blend membranes) 25 90 0.300 6.60 [46]
PVA/PHC 25 90 0.140 7.90 [46]
PVA/PVP 25 90 0.800 2.40 [46]
PVA 35 80 0.056 6.71 [37]
PVA 45 80 0.124 6.39 [37]
PVA-g-AAm (48%) 35 80 0.086 5.38 [37]
PVA-g-AAm (93%) 35 80 0.094 3.81 [37]
PVA-g-AN (52%) (PVA-2) 30 90 0.090 14.60 This study
PAAm, poly(acryl amide); PAA, poly(acrylic acid); PHC, poly(hydroxycarboxylic acid); PVP, poly(N-vinyl-2-pyrolidone); PVA, poly(vinyl alcohol); Aam,
acryl amide; AN, acrylonitrile.
and the permeation rate of water is higher than that of acetic
acid regardless of feed composition. These phenomena can
be explained in terms of plasticizing effect of water. As the
water concentration in the feed is increased, the amorphous
regions of the membrane becomes more swollen; hence the
exibility of polymer chains increases the energy required
for diffusive transport through the membrane decreases. This
could be the reason why low separation factors were ob-
tained at low acetic acid concentrations.
Additionally in pervaporation separation follows the
solution-diffusion mechanism. Therefore the molecular size
of the permeating molecules becomes very important to
characterize the permeation behavior [43]. It is known that
acetic acid has larger molecular size (0.40 nm) than water
molecules (0.28 nm). As the amount of acetic acid increases
in the feed mixture it becomes difcult for acetic acid
molecules to diffuse through the less swollen membrane, so
separation factor increases at high acid concentrations.
Pervaporation separation index values were calculated us-
ing Eq. (3) and presented in Fig. 7. As it is reected from
0.0
0.2
0.4
0.6
2 4 6 8
P (Pa x 10
-3
)
Q

(
k
g
/
m
2
h
)
0.0
1.0
2.0
3.0
4.0

Fig. 8. Effect of down stream pressure on the total permeation rate and separation factor (20 wt.% CH
3
COOH; T = 30

C): () Q () .
the gure that PSI values of the membranes increases with
the acetic acid concentration which shows that the perfor-
mance of PVA-g-AN (PVA-2) membranes are good at high
acetic acid concentrations
For comparison purposes several results regarding
the separation of acetic acidwater mixtures obtained
by various authors with different type of membranes
listed in Table 5. One can see that PVA-g-AN mem-
branes have high selectivity although they have almost the
same permeation rate depending on the feed composition
and temperature than the membranes studied up in the
table.
3.6. Effect of pressure on pervaporation
Effect of the down stream pressure on the permeation rate
and the separation factor was shown in Fig. 8 for 20 wt.%
acetic acid solution at 30

C. Permeation rate increases with


the pressure which complies well with theoretical prediction.
However separation factor decreases as the down stream
N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158 57
pressure increases. Similar results were also reported in the
literature [44].
4. Conclusions
Acrylonitrile grafted poly(vinyl alcohol) membranes were
found to be capable of permeating water in preference to
acetic acid from aqueous acetic acid mixtures. The perme-
ation rate of water increases whereas permeation rate of
acetic acid decreases as the water content of the feed in-
creases. It was also determined that increase in temperature
increased the permeation rate without affecting the separa-
tion factor much. As the downstream pressure increased per-
meation rate increased whereas separation factor decreased
and from the PSI values of the membranes it could be said
that especially at high acetic acid concentrations membranes
behaved more separable.
Acknowledgements
Three of grateful to Gazi University Research Fund for
the support of this study.
References
[1] M. Yoshikawa, T. Yukoshi, K. Sanui, N. Ogata, Separation of
water and ethanol by pervaporation through poly(acrylic acid-co-
acrylonitrile) membrane, J. Polym. Sci. Polym. Sci. Ed. 22 (1984)
473475.
[2] M. Yoshikawa, T. Yukoshi, K. Sanui, N. Ogata, Selective separation
of water from waterethanol solution through quarternized poly(4-
vinyl pyridine-co-acrylonitrile) membranes by pervaporation tech-
nique, J. Appl. Polym. Sci. 33 (1987) 23692392.
[3] M. Yoshikawa, T. Yukoshi, K. Sanui, N. Ogata, Selective separa-
tion of waterethanol mixture through synthetic polymer membranes
having carboxylic acid as a functional group, J. Polym. Sci. Part A:
Polym. Chem. 24 (1986) 15851597.
[4] C.E. Reineke, J.A. Jagodzinski, K.R. Denslow, Highly water selec-
tive cellulosic polyelectrolyte membranes for the pervaporation of
alcoholwater mixtures, J. Membr. Sci. 32 (1987) 207221.
[5] G.H. Hsiue, Y.S. Yang, J.F. Kuo, Permeation and separation of
aqueous alcohol solutions through grafted poly(vinyl alcohol) latex
membranes, J. Appl. Polym. Sci. 34 (1987) 21872196.
[6] Y.F. Xu, R.Y.M. Huang, Pervaporation separation of ethanolwater
mixtures using ionically crosslinked blended poly(acrylic acid)
(PAA)-nylon 6 membranes, J. Appl. Polym. Sci. 36 (1988) 1121
1128.
[7] R.Y. Huang, C.K. Yeom, Pervaporation separation of aqueous mix-
tures using crosslinked poly(vinyl alcohol) (PVA). Part II. Permeation
of ethanolwater mixtures, J. Membr. Sci. 51 (1990) 273292.
[8] T. Uragami, H. Shinomiya, Concentration of aqueous alcoholic solu-
tions through a modied silicone rubber membrane by pervaporation
and evaporation, Makromol. Chem. 192 (1991) 22932305.
[9] T. Uragami, T. Morikawa, Permeation and separation characteristics
of alcoholwater mixtures through poly(dimethyl siloxane) mem-
brane by pervaporation and evaporation, J. Appl. Polym. Sci. 44
(1992) 20092018.
[10] H. Ohya, K. Matsumoto, Y. Negishi, T. Hino, H.S. Choi, The sepa-
ration of water and ethanol by pervaporation with PVAPAN com-
posite membranes, J. Membr. Sci. 68 (1992) 141148.
[11] Y.M. Lee, B.K. Oh, Pervaporation of wateracetic acid mixture
through poly(4-vinylpyridine-co-acrylonitrile) membrane, J. Membr.
Sci. 85 (1993) 1320.
[12] B. Will, R.N. Lichtenthaler, Comparison of the separation of mix-
tures by vapor permeation and pervaporation using PVA composite
membranes. Part I. Binary alcoholwater systems, J. Membr. Sci. 68
(1992) 119125.
[13] C.K. Yeom, R.Y.M. Huang, Modelling of the pervaporation separa-
tion of ethanolwater mixtures through crosslinked poly(vinyl alco-
hol) membrane, J. Membr. Sci. 67 (1992) 3955.
[14] W.Y. Chiang, C.C. Huang, Separation of liquid mixtures by using
polymer membranes. Part IV. Wateralcohol separation by pervapo-
ration through modied acrylonitrile poly(vinyl alcohol) copolymer,
J. Appl. Polym. Sci. 48 (1993) 199203.
[15] W.H. Jo, H.J. Kim, Y.S. Kang, Separation of waterethanol mix-
ture through poly(acrylonitrile-co-acrylic acid)/poly(ethylene oxide)
membranes by pervaporation, J. Appl. Polym. Sci. 51 (1994) 529
535.
[16] S.Y. Nam, H.J. Chun, Y.M. Lee, Pervaporation separation of
waterisopropanol mixtures using carboxymethylated poly(vinyl al-
cohol) composite membranes, J. Appl. Polym. Sci. 72 (1999) 241
249.
[17] H.C. Park, R.M. Meertens, M.H.V. Mulder, C.A. Smolders, Perva-
poration of alcoholtoluene mixtures through polymer blend mem-
branes of poly(acrylic acid) and poly(vinyl alcohol), J. Membr. Sci.
90 (1994) 265274.
[18] Y.M. Sun, T.L. Huang, Pervaporation of ethanolwater mix-
tures through temperature sensitive poly(vinyl alcohol-g-n-
isopropyacrylamide) membranes, J. Membr. Sci. 110 (1996) 211
218.
[19] E. Ruckenstein, L. Liang, Pervaporation of ethanolwater mixtures
through poly(vinyl alcohol)poly(acrylamide) interpenetrating poly-
mer network membranes unsupported and supported on polyether-
sulfone ultraltration membranes: a comparision, J. Membr. Sci. 110
(1996) 99107.
[20] J.W. Rhim, C.K. Yeom, S.W. Kim, Modication of poly (vinyl
alcohol) membranes using sulfursuccinic acid and its application to
pervaporation separation of water/alcohol mixtures, J. Appl. Polym.
Sci. 68 (1998) 17171723.
[21] T.Q. Nguyen, A. Essamri, R. Clement, J. Nell, Synthesis of mem-
branes for the dehydration of wateracetic acid mixture by perva-
poration. Part 1. Polymer material selection, Makromol. Chem. 188
(1987) 19731984.
[22] R.Y.M. Huang, A. Moreira, R. Notarfonzo, X.F. Xu, Pervaporation
separation of acetic acidwater mixtures using modied membranes.
Part I. Blended poly(acrylic acid) (PAA)nylon 6 membranes, J.
Appl. Polym. Sci. 35 (1988) 11911200.
[23] R.Y.M. Huang, C.K. Yeom, Pervaporation separaion of aqueous mix-
tures using crosslinked poly(vinyl alcohol) membranes. Part III. Per-
meation of acetic acidwater mixtures, J. Membr. Sci. 58 (1991)
3347.
[24] R.Y.M. Huang, C.K. Yeom, Development of crosslinked poly(vinyl
alcohol) (type II) and permeation of acetic acidwater mixtures, J.
Membr. Sci. 62 (1991) 5973.
[25] M. Yoshikawa, T. Shimidzu, Y. Maeda, K. Magara, H. Tsugaya,
Pervaporation of aqueous acetic acid mixture through poly(acrylic
acid-co-acrylonitrile) membranes, J. Membr. Sci. 82 (1993) 157162.
[26] I.J. Ball, S.C. Huang, R.A. Wolf, J.Y. Shimano, R.B. Kaner, Perva-
poration studies with polyaniline membranes and blends, J. Membr.
Sci. 174 (2000) 161176.
[27] U.S. Toti, M.Y. Kariduraganavar, K.S. Soppimath, T.M. Aminabhavi,
Sorption, diffusion and pervaporation separation of water/acetic acid
mixtures through the blend membranes of sodium alginate and guar
gum-g-poly(acrylamide), J. Appl. Polym. Sci. 83 (2002) 259272.
58 N. Alghezawi et al. / Chemical Engineering and Processing 44 (2005) 5158
[28] S. Li, V.A. Tuan, R.D. Noble, J.L. Falconer, A Ge-substituted ZSM-
5 zeolite membrane for the separation of acetic acid from water, Ind.
Eng. Chem. Res. 40 (2001) 61656171.
[29] T. Sano, S. Ejiri, K. Yamada, Y. Kawakami, H. Yanagishita, Sepa-
ration of acetic acidwater by pervaporation through silicate mem-
brane, J. Membr. Sci. 123 (1997) 225233.
[30] S.K. Ray, S.B. Sawant, J.B. Joshi, V.G. Pangarkar, Dehydration of
acetic acid by pervaporation, J. Membr. Sci. 138 (1998) 117.
[31] S.P. Kusumocahyo, M. Sudoh, Dehydration of acetic acid by perva-
poration with charged membranes, J. Membr. Sci. 161 (1999) 77
83.
[32] T.M. Aminabhavi, H.G. Naik, Pervaporation separation of wa-
ter/dimethyl formamide mixtures using poly(vinyl alcohol)-g-
poly(acrylamide) copolymeric membranes, J. Appl. Polym. Sci. 83
(2002) 273282.
[33] S.Y. Lu, C.P. Chiu, H.Y. Huang, Pervaporation of acetic acid/water
mixtures through silicate lled polydimethylsiloxane membranes, J.
Membr. Sci. 176 (2002) 159167.
[34] X.P. Wang, Modied alginate composite membranes for the dehy-
dration of acetic acid, J. Membr. Sci. 170 (2000) 7179.
[35] G. Asman, O. Sanl, Ultraltration of Fe (III) solutions in the presence
of poly(vinyl alcohol) using modied poly(methyl methacrylate-co-
methacrylic acid) membranes, J. Appl. Polym. Sci. 64 (1997) 1115
1121.
[36] C.K. Yeom, K.H. Lee, Pervaporation separation of wateracetic acid
mixtures through poly(vinyl alcohol) membranes crosslinked with
gluteraldehyde, J. Membr. Sci. 109 (1996) 257265.
[37] T.M. Aminabhavi, H.G. Naik, Synthesis of graft copolymeric mem-
branes of poly(vinyl alcohol) and poly(acrylamide) for the pervapo-
ration separation of water/acetic acid mixtures, J. Appl. Polym. Sci.
83 (2002) 244258.
[38] M.D. Kurkuri, T.M. Aminabhavi, Polyacrylonitrile-g-poly(vinyl al-
cohol) membranes for the pervaporation separation of dimethyl for-
mamide and water mixtures, J. Appl. Polym. Sci. 91 (2004) 4091
4097.
[39] G.H. Koops, J.A.M. Nolten, M.H.V. Mulder, G.A. Smolders, Se-
lectivity as a function of membrane thickness: gas separation and
permeation, J. Appl. Polym. Sci. 53 (1994) 16391651.
[40] P. Aptel, J. Cuny, J. Jozefonvicz, G. Morel, J. Neel, Liquid transport
through membranes prepared by grafting of polar monomers onto
PTFE lms. Part II. Some factors determing pervaporation rate and
selectivity, J. Appl. Polym. Sci. 18 (1974) 351364.
[41] P. Aptel, J. Cuny, J. Jozefenvicz, G. Morel, J. Neel, Liquid transport
through membranes prepared by grafting of polar monomers onto
PTFE lms. Part III. Steady state distribution in membrane during
pervaporation, J. Appl. Polym. Sci. 18 (1974) 365378.
[42] J.P. Brun, G. Bulvestre, A. Kergreis, M. Guillou, Hydrocarbon sep-
arations with poly membranes. Part I. Butadiene-isobutene separa-
tion with nitrile rubber membranes, J. Appl. Polym. Sci. 18 (1974)
16631683.
[43] G. Asman, O. Sanl, Characteristics of permeation and separation for
acetic acidwater mixtures through poly(vinyl alcohol) membranes
modied with poly(acrylic acid), Sep. Sci. Technol. 38 (2003) 1963
1980.
[44] I. Blume, J.G. Wijmans, R.W. Baker, The separation of dissolved
organics from water by pervaporation, J. Membr. Sci. 49 (1990)
253286.
[45] G.H. Koops, J.A.M. Nolten, M.H.V. Mulder, G.A. Smolders,
Poly(vinyl chloride) poly(acrylonitrile) composite membranes for the
dehydration of acetic acid, J. Membr. Sci. 81 (1993) 5770.
[46] A. Rudin, The Elements of Polymer Science and Engineering, Aca-
demic Press, New York, 1982, Chapter 3, 102 pp.
Chemical Engineering and Processing 44 (2004) 5969
Countercurrent ow distribution in structured packing
via computed tomography
Shaibal Roy
a
, A. Kemoun
a
, M.H. Al-Dahhan
a,
, M.P. Dudukovic
a
,
Thomas B. Skourlis
b
, Frits M. Dautzenberg
b
a
Chemical Reaction Engineering Laboratory, One Brookings Drive, Campus Box 1198, Washington University, St. Louis, MO 63130-4899, USA
b
ABB Lummus Global, Inc, 1515 Broad Street, Bloomeld, NJ 07003-3096, USA
Received 21 July 2003; received in revised form 9 March 2004; accepted 25 March 2004
Available online 26 June 2004
Abstract
A newly developed -ray computed tomography scanner was implemented to study liquid ow distribution in a 30.48-cm-diameter column
packed with corrugated structure using countercurrent airwater ow. Validation experiments conrm that spatial resolution of 2.5 mm can
be achieved by the new CT unit.
Within the gas (010 cm/s supercial velocity) and liquid (0.62.5 cm/s supercial velocity) ow ranges investigated, a good liquid distri-
bution was observed at all conditions, as manifested by uniformity factor in excess of 70%. The liquid saturation increases with increasing
supercial liquid velocity as well as down the column height. Within the conditions studied, the effect of gas velocity was, in general, found
to be minimal.
2004 Elsevier B.V. All rights reserved.
Keywords: Hydrodynamics; Structured packing; Holdup; Computed tomography
1. Introduction
Fixed-bed catalytic reactors and reactive distillation
columns are widely used in many industrial processes. Re-
cently, structured packing (e.g., monoliths, katapak, mella-
pak etc.) has been suggested for various chemical processes
[14,14].One of the major challenges in the design and op-
eration of reactors with structured packing is the prevention
of liquid ow maldistribution, which could cause portions
of the bed to be incompletely wetted. Such maldistribution,
when it occurs, causes severe under-performance of reac-
tors or catalytic distillation columns. It also can lead to hot
spot formation, reactor runaway in exothermic reactions,
decreased selectivity to desired products, in addition to the
general underutilization of the catalyst bed.
To overcome these problems, ow distribution in reactors
with structured packing of a desired type needs to be quan-
titatively studied and understood over a desired range of op-
erating conditions. The non-invasive computed tomography

Corresponding author. Tel.: +1 314 935 7187; fax: +1 314 935 7211.
E-mail address: muthanna@che.wustl.edu (M.H. Al-Dahhan).
(CT) scanner with -ray source at the Chemical Reaction
Engineering Laboratory (CREL), Washington University [5]
has been used successfully in the past to characterize ow
distribution of liquid and gas over non-structured packing
[6]. Hence, it is the tool of choice for studies on structured
packing.
Many studies on the ow distribution in random packed
beds have been reported in the literature. Mercandelli et al.
[8] published a short review of the ow distribution work in
random packed trickle bed, which includes the list of various
techniques used to determine and quantify the ow distri-
bution. Conventional methods include, for example, collect-
ing liquid at the bottom of the column from different zones
while advanced methods include tomographic techniques.
Mercandelli et al. [8] used several techniques to quantify
liquid distribution in columns of diameters up to 30 cm with
three different distributor designs. They used global pres-
sure drop measurements, global residence time distribution
(RTD) of the liquid, local heat transfer probes, capacitance
tomography and a collector at the bottom of the column.
Boyer and Fanget [7] used computed tomography to
quantify liquid distribution in a 60-cm-diameter trickle bed
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.010
60 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969
reactor (downow of gas and liquid over randomly packed
particles). The apparatus was very similar to that being used
in the Chemical Reaction Engineering Laboratory (CREL)
at Washington University in St Louis, albeit, with a higher
number of detectors (32 of 1.27-cm-diameter BGO detectors
as against 9 of 5-cm-diameter Nal detectors) and increased
radioactive source strength (300 mCi as against 85 mCi).
A comprehensive study was performed to quantify the error
associated with the measurements and the authors were able
to detect solids holdup within 3% accuracy. The authors
also proposed a correlation to predict liquid holdup based
on a force balance on liquid lm and showed that tomog-
raphy can reproduce the liquid holdup as predicted by the
model for liquid ow rates up to 1.7 cm/s (no gas owing).
Yin et al. [15] investigated liquid distribution in a col-
umn of 60 cm in diameter containing pall rings. The authors
found that the distributor type had a signicant effect on the
distribution, which was veried using computational uid
dynamics (CFD).
Examples of the use of tomographic techniques to deter-
mine gas/liquid distribution in structured packing is limited
in the open literature. Mewes et al. [9] studied ow distribu-
tion in monoliths using capacitance tomography. A spatial
Fig. 1. Schematic diagram of the experimental setup: (a) Process diagram (all dimensions in centimeter); (b) packing element; (c) liquid distributor. DP,
differential pressure transmitter; FI, ow indicator; TI, temperature indicator.
resolution of about 510% of the column diameter in the
measurement plane was claimed [912]. Mewes et al. [9]
used water which was introduced through a large number of
injection nozzles into a 120 mm diameter reactor containing
monolith. With increasing liquid velocity, the liquid distri-
bution improved. The authors did not demonstrate the effect
of gas velocity on the liquid distribution.
Toya et al. [13] used X-ray tomography to scan gas/liquid
ow in a corrugated structured packing known as Sulzer
Mellapak 250 Y. The solids prole of the dry structured
packing was reproduced with fair accuracy. The scan was
performed in a 60-cm-diameter column. Very thin liquid lm
owing past the structure at 0.006 m/s was also detected by
the tomography unit.
Iluita and Larachi [4] developed a hydraulic mechanistic
model for prediction of the irrigated pressure drop, liquid
holdup and the effective interfacial area in the preloading
zone of structured packing operated in countercurrent ow
of gas and liquid under partial wetting condition. The model
mimics the two phase ow situation by a network of double
slits consisting of dry and wet slits. The model was success-
ful in predicting the irrigated pressure drop, liquid holdup
and effective interfacial area under various conditions.
S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969 61
This work focuses on investigating the ow distribution in
a structured packed bed using gamma ray computed tomog-
raphy. Gas and liquid ow in a countercurrent mode over
a bed packed with corrugated structured packing is usually
used in reactive and catalytic distillation columns. This in-
vestigation includes determining the liquid holdup and its
distribution in the cross-section of the reactor at a range of
gas and liquid velocities.
2. Experimental work
2.1. Experimental setup and operating conditions
The experimental setup, schematically shown in Fig. 1a,
was developed to measure gasliquid ow distribution in
countercurrent ow structured packed bed. It consists of a
packing section, distributor and liquid and gas delivery sys-
tems. The Plexiglas column is 152.4 cm high and 30.48 cm
in diameter and holds four packing elements each 27.3 cm
in diameter and height. The packing is of corrugated type
made of stainless steel and frequently used in modern re-
active distillation columns, commercially known as Norton
packing (Fig. 1b). The corrugations are at a 45

angle to the
vertical axis. The packing elements are held in position us-
ing a donut ring at the bottom of the Plexiglas column. The
liquid is recirculated and introduced at the top through a dis-
tributor having seven holes of 2.5-cm-diameter; the middle
one is covered with a mesh (Fig. 1c).
Water was used as the liquid phase. The liquid delivery
system consists of a feed tank, pump, and a ow indicator.
Water is recycled, as well as added if necessary, to maintain
a constant suction head at the pump. The in-house air system
was used as the gas delivery system through a rotameter.
Air enters the setup at a point below the packing. Two taps
are mounted at the inlet and outlet of the packed column for
pressure drop measurement.
Flow scans were performed at three axial levels (1.5, 2.5
and 3.5D) from the top packing surface (D being the diam-
eter of the packing), for various ow conditions. Supercial
liquid and gas velocities used, denoted as U
L
and U
g
, re-
spectively, are listed in Table 1.
2.2. The newly developed gamma ray computed
tomography (CT) scanner design
A newly installed computed tomography (CT) scan-
ner was used in this study. This is a modied unit of the
third-generation fan-beam conguration originally devel-
Table 1
Range of operating conditions available for the experimental setup used.
U
G
0 to 10 cm/s
U
L
0.63 to 2.17 cm/s
Scan Levels: 1.5D, 2.5D, 3.5D from the top, D 27.3 cm.
Fig. 2. Schematic diagram of the newly developed CREL computed
tomography.
oped by Kumar et al. [5] at CREL. The key system elements
are illustrated in Fig. 2. The CT scanner consists of an ar-
ray of Nal (TI) detectors with a diameter of 5 cm as well
as an encapsulated 70 mCi Cs
137
source located opposite
to the center of the array of detectors. Nine detectors are
used in the present study to cover the cross-section of the
30.48-cm-diameter column. The detectors and the source
are mounted on a plate, which can be rotated around the
axis of the column by a stepping motor that is controlled
through a microprocessor. Moreover, the whole assembly
can be moved in the axial direction along the column to
perform a scan at different axial levels of the column. The
source collimator provides a collimated fan beam of 40

in the horizontal plane and 5 mm in the vertical plane.


The detector side collimator provides 2 mm width colli-
mated -rays to the detector. The movement of the colli-
mator rack and detector assembly is controlled by another
stepping motor. In each movement, this assembly rotates
62 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969
by 0.2

, thus producing large number of projections per


detector.
The new scanner introduced the following modications
(Fig. 3).
2.2.1. Collimator and detectors rotate simultaneously
In the old CT unit, the detectors are stationary in a partic-
ular view and the collimators move by a desired angle to ob-
tain projections for various chords. The intensity of radiation
as detected by the detectors is not uniform; it is maximum
when the collimator passes the center of the detector and
minimum at the two edges. To overcome this non-uniformity
in the new setup, the collimator position has been xed at
the center of the detector and the whole detectorcollimator
rack is rotated (Fig. 3a).
2.2.2. Eliminating the need to interpolate between detectors
In the old CT unit, the angular distance between two de-
tectors is 0.5

whereas a typical angle between two pro-


jections is approximately 0.2

. Therefore, the source in-


tensity in the space between the detectors has to be in-
terpolated. Hence, two articial projections are incorpo-
rated into the reconstruction process, which may lead to
Fig. 3. (a) New CT: the whole assembly moves; old CT: only the collimator rack moves; (b) new CT: no interpolation needed; old CT: interpolation
needed between detectors; (c) new CT: 2 mm collimator slit thickness; old CT: 5 mm collimator slit thickness.
some errors. In the new setup, the angular distance between
two consecutive detectors is 0.2

, which is the same as


the angle between two consecutive projections. Therefore,
there is no need to interpolate between the detector spaces
(Fig. 3b).
2.2.3. Collimator slit width and radioactive source strength
In order to achieve the best possible resolution, the thick-
ness of the ray being attenuated should have innitely small
width. However, this is practically impossible since in that
case, we need to have an innitely small slit width of the
detector collimator also. If the slit width is too small, the
counts registered by the detectors will be very small, which
in turn will affect the resolution. Generally, the maximum
resolution that can be attained is equivalent to the slit width.
In the new setup, a collimator with a slit width of 2 mm is
used instead of 5 mm, which was the case in old CT unit.
This is expected to result in a higher resolution of the order
of 2 mm (Fig. 3c).
Moreover, the new CT along with all the associated elec-
tronics is mounted on wheels. Being completely made of
aluminum instead of stainless steel, the whole assembly is
lightweight and can be moved around in different locations
S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969 63
Fig. 4. CT validation scan using: (a) jar lled with water; (b) cross-sectional attenuation prole of the liquid in the jar; (c) azimuthally averaged attenuation
in the radial direction (- actual attenuation, average attenuation).
for conducting holdup measurements at various experimen-
tal setups.
The beam attenuation is measured along a number of
beam paths through the column, which originate from dif-
ferent angles. Once a set of attenuation measurements is
completed, the density distribution image is reconstructed
by using the EstimationMaximization algorithm [5]. This
reconstruction algorithm is used, since it has the following
advantages: (1) it accounts for statistical variations associ-
ated with radiation measurements; (2) it readily incorporates
non-uniform beam effects; and (3) it ensures that the nal
reconstruction will contain positive values. To obtain statis-
tically signicant results, and to reduce the effect of position,
the CT scans were obtained by scanning 360

around the
column using a total scanning time of about ve hours and
collecting numerous beam path attenuations (approximately
2500 projections). The nal holdup distribution images are
obtained from the attenuation data using the relationships
and the procedure developed by Chen et al. (2000) [6] for
packed bed.
2.3. CT scan evaluation
Since this was a new CT scanner developed to be used
for the rst time, it was imperative to evaluate the perfor-
mance of the unit in terms of spatial and density resolution.
Two CT scans were performed to evaluate the system. In the
rst case, a 5.08 cm jar was lled with water and scanned
(Fig. 4). In the second case, a hollow ball dipped in water
was scanned and gas holdup was evaluated (Fig. 5). Further,
a third validation, which will be discussed in the following
section of the Results and discussion, was also performed
by determining the solids fraction of the packing using com-
puted tomography, which was compared to the one measured
by water displacement method.
For the rst case of a jar lled with water, the results are
as follows.
The collimator used in the present CT has an aperture of
2 mm and therefore, theoretically, spatial resolution should
be around 2 mm. In this test case, a 70 mm domain con-
taining 5.1 cm outer diameter glass jar lled with water was
reproduced (as shown in Fig. 4b). The diameter of the ob-
ject obtained from CT was 5.3 cm, which indicates that we
have a spatial resolution of 2 mm in this case. The attenua-
tion in the water region should be uniform. In this case, the
scan gives a slightly non-uniform attenuation as shown in
Fig. 4c. The calculated error in the attenuation coefcient
was found to be within 6.5%.
For the second case, a hollow ball was dipped inside a
jar lled with water and scanned. The diameters of the ball
and jar were 6.98 and 18.95 cm, respectively, as shown in
Fig. 5a. A scan area of 27 cm in diameter was reproduced
using the reconstruction algorithm. The dimensions of the
objects as reproduced by the scan were 6.97 and 19.20 cm,
respectively, as shown in Fig. 5b, which give a maximum
spatial error of about 2.5 mm. This is good enough to re-
solve relatively small maldistribution, if it exists, inside the
30.48-cm-diameter column used in this study. The gure
shows that the overall error in the estimated total holdup is
within 12.8%.
It must be noted that the resolution changes with the size
of the object being scanned due to different attenuation of
the gamma rays. However, the obtained resolutions should
be adequate for a column of 30.5-cm-diameter.
64 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969
Fig. 5. CT validation using: (a) hollow ball in water Jar; (b) density distribution of the object; (c) gas holdup prole of the system.
3. Results and discussion
3.1. Solids fraction of the packing
Solids fraction of the packing was rst evaluated using CT
at the same axial levels used for the actual ow scans, i.e.,
at 1.5, 2.5 and 3.5D, from the top packing surface, where
D denote the diameter of the packing (30.48 cm). Fig. 6a
shows the solids distribution at those three axial positions.
The azimuthally averaged solids fraction is shown in Fig. 6b
and is approximately equal to 2.5%. The solid fraction was
also measured experimentally by immersing the packing el-
ement in water-lled bath and measuring the volume of wa-
ter displaced, which was found to be approximately 3%.
This overall average value is close to the values obtained
by CT scanning. Considering the error associated with the
measurement of small amount of the displaced water, this
indicates that CT provides adequate measurements of the
solids distribution in the bed despite its small values.
3.2. Liquid distribution in the bed of corrugated structured
packing
Fig. 7ac shows the time-averaged cross-sectional liquid
saturation distribution in the column at supercial gas ve-
locities of 0, 5 and 10 cm/s, respectively. For each gas veloc-
ity, the liquid saturation distribution is shown at three axial
positions and three liquid velocities. Also provided are the
S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969 65
Fig. 6. Determination of solid distribution at three axial positions: (a) cross-sectional distribution; (b) azimuthally averaged solids holdup and (c) overall
solids holdup. The horizontal line indicates experimentally determined solids fraction.
time and cross-sectionally averaged liquid saturation values,

L
, as well as the uniformity factor expressed as percentage

f
(discussed later). Liquid saturation is the ratio of liquid
volume to the bed void volume. It can be seen that, for the
ow conditions used, as supercial liquid velocity increases,
the cross-sectional averaged liquid saturation increases. For
all the supercial gas and liquid velocities used, the liquid
saturation varies between 6 to 15%.
Fig. 7ac also illustrates the liquid saturation distribution
along the length of the column. In general, it is observed
that as the liquid moves downward, liquid saturation in-
creases. At all conditions studied, the uniformity factor
shows that the liquid distribution improves at the bottom
section of the bed (i.e. 3.5D). This could be due to the
liquid distributor design, which produces seven liquid jets
at the top of the bed. These jets disperse while moving
downward. At high liquid supercial velocity (2.17 cm/s),
the uniformity factors are smaller than those obtained at
lower liquid supercial velocities, since, at this high liquid
velocity, the jets require longer bed length to disperse and
distribute.
Fig. 8 shows the time and azimuthally averaged radial
liquid saturation proles at varying supercial gas and liq-
uid velocities at the middle axial position (2.5D). The gure
shows that liquid saturation is nearly at, which suggests
a fair uniformity of liquid distribution. Moreover, with in-
creasing liquid velocities, liquid saturation increases. Simi-
lar trends were obtained at all scan heights.
66 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969
Fig. 7. Time averaged cross-sectional liquid saturation distribution in structured bed at (a) Ug: 0 cm/s; (b) Ug: 5 cm/s; (c) Ug: 10 cm/s. In the table,
L
shows the liquid saturation and
f
% show uniformity factor percentage.
S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969 67
Fig. 7. (Continued ).
Fig. 8. Effect of supercial gas and liquid velocities on the liquid saturation
radial prole at axial position of 2.5D.
Fig. 9 illustrates the effect of supercial gas and liquid ve-
locities on the cross-sectionally averaged liquid saturation at
the middle of the column (2.5D axial position). It is obvious
that the effect of gas velocity on the liquid saturation is not
signicant within the range of ows studied. This could be
due to the fact that solid and liquid holdups are very small,
leaving enough space for the gas to ow upwards without
signicant interactions with the liquid phase owing down-
ward.
The measured cross-sectionally averaged liquid holdup
(i.e. measured liquid saturation times the bed voidage) are
consistent with previously reported results [4]. For a sim-
ilar packing structure, Illiuta and Larachi [4] found liquid
holdups, as calculated by their mechanistic model and com-
pared with the reported experimental results, slightly lower
than the values obtained in this study. However, there are
some differences between the two studies: this work was
performed in a smaller diameter column (30 cm versus
Fig. 9. Effects of gas and liquid supercial velocities on the cross-
sectionally averaged liquid saturation.
68 S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969
100 cm), and higher supercial gas velocity was used. Il-
liuta and Larachi [4] mentioned that gas ow rate had a
marginal effect on liquid holdup, a fact reected in this work
as well.
Experiments were also performed to compare the holdup
and ow distribution in a bed, randomly packed with 3 mm
spherical alumina particle, under the same ow conditions
as was done for structured packing. However, it was evident
that the successful operating conditions for structured pack-
ing were too severe for random packed bed, due to very high
pressure drop. For very low liquid velocity (1 mm/s) and
no gas ow, when the experiment was possible, the liquid
distribution was poor as indicated by a low uniformity factor
(40%). However, this information is insufcient to com-
pare the distribution characteristics of structured and random
packings.
3.3. Uniformity factor (
f
)
The uniformity of the liquid ow distribution across the
column cross-section was quantied in this work. This is
presented in Fig. 7ac in a tabular form along with time and
cross-sectionally averaged liquid saturation values. Various
methods have been used in the literature to quantify the ow
uniformity across a column cross-section. Mercandelli et al.
[8] used a maldistribution factor, which was a variation of
the standard deviation among nine sectors that the ow do-
main was divided into. The factor varied between 0 and 1.
This method is suitable for ow domains, which are divided
into small number of sectors. For cases where large num-
ber of sectors are used (e.g. holdup obtained by tomography
where large number of pixels are used), the maldistribution
values invariably give small values, thus making it impos-
sible to ascertain the quality of distribution. Jiang proposed
the relative standard deviation to quantify the uniformity of
ow [1]. However, since standard deviation can take values
up to innity, this quantity will also not provide a proper
feel for the uniformity.
In the present work, the degree of uniformity is cal-
culated based on the t-test method. The circular domain
representing the reactor is divided into several sub-domains,
each having a xed number of pixels. In this work, 10 pix-
els were arbitrarily taken. This constitutes several groups
of data points for liquid saturation; t-test was then carried
out between the group representing the overall averaged
cross-sectional distribution of liquid saturation and the av-
erage of one sub-domain group. If the group passes the
t-test, this indicates that the null hypothesis is satised and
the two groups are statistically the same. Then 1 is assigned
to that pair. If the two groups are statistically not simi-
lar, then 0 is assigned. Uniformity factor is then dened
as the percentage number of groups that are statistically
similar (i.e. passes the t-test) to the group representing the
whole cross-sectional distribution of liquid saturation and
its average. In this method, a uniform distribution yields
a uniformity factor of 100%, and a completely maldis-
tributed domain would give a uniformity factor very close
to 0%.
The domain consisted of 60 by 60 pixels (3600) out of
which 2936 pixels represent the actual cross-section of the
column, and these are the pixels which represent the liquid
saturation distribution. Fig. 7ac show the uniformity factor
in percentage (
f
%) for different liquid and gas supercial
velocities for the scans performed in this work. In most of the
cases, the uniformity factor is large (above 70%), suggesting
a fairly uniform distribution of liquid achieved in the studied
setup.
3.4. Pressure drop in corrugated structured packing
The pressure drop across the length of the packing was
extremely small and was beyond the detectable limit of the
available pressure gauges. The full scale of the available
transducer (validyne differential transducer) was 200 units,
which correspond to a pressure drop of 1 cm of water. In
our study, it was observed that the maximum pressure drop
within the gas and liquid ow rates studied was about 1 mm
of water (20 calibration units). A trend was seen wherein
the pressure drop increases with increasing gas and liquid
velocities.
4. Final remarks
It was established that the newly developed CT unit has
a spatial resolution of about 2 2.5 mm. This is quite ac-
ceptable given the size of the column used in this study. The
solid fraction as determined by CT scan was about 2.5%,
which was close to the value of 3% as determined by water
displacement method.
The liquid saturation increases with increasing supercial
liquid velocity. Moreover, the liquid saturation increases as
the liquid phase moves downward. The liquid distribution
was found to be fairly uniform in general as expressed by the
uniformity factor which was relatively large, between 70 to
95%. Liquid distribution was better at the bottom of the bed,
compared to the upper section. The effect of gas velocity,
was in general, found to be very small at the conditions used
in this study.
Acknowledgements
The nancial support provided by ABB Lummus Global,
Inc. is acknowledged which made this work possible.
References
[1] Y. Jiang, Flow distribution and its impact on performance of
packed-bed reactors, Ph.D thesis, Washington University in St Louis,
Missouri, 2000.
S. Roy et al. / Chemical Engineering and Processing 44 (2004) 5969 69
[2] R. Krishna, Reactive Separation: a New Paradigm in an Old Bottle,
Report from Chemical Engineering department, University of Ams-
terdam, 1999.
[3] G.G. Podrebarac, F.T.T. Ng, G.L. Rempel, The production of diace-
ton alcohol with catalytic distillation:part II. A rate-based catalytic
distillation model for the reaction zone, Chem. Eng. Sci. 53 (3)
(1998) 1077.
[4] I. Iliuta, F. Larachi, Mechanistic model for structured-packing con-
taining columns: Irrigated pressure drop, liquid holdup and pack-
ing fractional wetted area, Ind. Eng. Chem. Res. 40 (2000) 5140
5146.
[5] S.B. Kumar, M.P. Dudukovic, Jamal Chaouki, Faical Larachi, Milorad
P. Dudukovic, Computer assisted gamma and X-ray tomography:
applications to multiphase ow systems, in: Non-Invasive Monit.
Multiphase Flows, Elsevier, 1997, pp. 47103.
[6] J. Chen, R. Novica, M.H. Al-Dahhan, M.P. Dudukovic, Study of
particle motion in packed/ebullated beds by computed tomography
(CT) and computer automated radioactive particle tracking (CARPT),
AIChE J. 47 (5) (2000) 9941004.
[7] C. Boyer, B. Fanget, Measurement of liquid ow distribu-
tion in trickle bed reactor of large diameter with a new
gamma-ray tomographic system, Chem. Eng. Sci. 57 (2002) 1079
1089.
[8] C. Marcandelli, A.S. Lamine, J.R. Bernard, G. Wild, Liquid distri-
bution in trickle-bed reactor, Oil Gas Sci. Technol.-Rev IFP 55 (4)
(2000) 407415.
[9] D. Mewes, T. Loser, M. Millies, Modeling of two-phase ow in
packings and monoliths, Chem. Eng. Sci. 54 (21) (1999) 47294747.
[10] N. Reinecke, D. Mewes, Flow regimes of two phase ow in mono-
lith catalyst, in: Proceedings of 5th World Congress on Chemical
Engineering, San Diego, CA, 1996.
[11] N. Reinecke, G. Petritsch, D. Schmitz, D. Mewes, Tomographic
measurement techniquesvisualization of multiphase ows, Chem.
Eng. Technol 21 (1998) 718.
[12] N. Reinecke, D. Mewes, Oscillatory transient two-phase ows in
single channels with reference to monolithic catalyst supports, Int.
J. Multiphase Flow 25 (67) (1999) 13731393.
[13] D. Toye, P. Marchot, M. Crine, M. Pelsser, G. LHomme, Local
measurements of void fraction and liquid holdup in packed columns
using X-ray computed tomography, Chem. Eng. Process. 37 (6)
(1998) 511520.
[14] L. Spiegel, W. Meier, Distillation columns with structured packings
in the next decade, Chem. Eng. Res. Des. 81 (A1) (2003) 3947.
[15] F. Yin, A. Afacan, K. Nandakumar, K.T. Chuang, Liquid holdup
distribution in packed columns: gamma ray tomography and CFD
simulation, Chem. Eng. Process. 41 (5) (2002) 473483.
Chemical Engineering and Processing 44 (2005) 7179
Effect of cycling operations on an adsorbed natural gas storage
O. Pupier, V. Goetz

, R. Fiscal
IMP-CNRS UPR 8521, Institut de Science et Gnie des Matriaux et Procds, Rambla de la Thermodynamique, 66100 Perpignan, France
Received 18 February 2004; received in revised form 7 May 2004; accepted 7 May 2004
Available online 1 July 2004
Abstract
Adsorbed natural gas (ANG) is an interesting opportunity for developing natural gas vehicles technology. In this case, adsorbents such as
activated carbons are used to store natural gas at moderate pressure, 3.5 MPa, compared to the high-pressure (20 MPa) required for current
compressed natural gas technology. Many studies are devoted to the elaboration of suitable adsorbent materials to optimise the methane storage
capacity. Nevertheless, since natural gas (NG) is composed of about 95%of methane mixed with other components, an important deterioration
of the storage performance is observed after successive cycles of an ANG system (lling and delivery). It is the result of adsorption of the
other components present in NG that are mainly higher molecular-weight hydrocarbons, carbon dioxide and nitrogen. In the present study,
the evolution of the storage capacity of a 2 l vessel cycled with NG is experimentally measured. As a function of the cycle number, the gas
composition at the outlet of the ANG system is determined using gas chromatography analysis.
2004 Elsevier B.V. All rights reserved.
Keywords: Activated carbon; Adsorption; Storage; Natural gas; Cycling operation
1. Introduction
Natural gas is a possible alternative as a transportation
fuel. Moreover, natural gas vehicles (NVGs) provide bene-
ts as cleanliness and safety [1,2]. However, the disadvan-
tage of NG is its low volumetric energy density compared to
that of conventional liquid fuels. Current gas vehicles em-
ploy storage vessels at high pressure (20 MPa). This implies
high manufacturing and lling costs. Adsorbed natural gas
(ANG) on a suitable microporous adsorbent offers an inter-
esting opportunity of developing NGVs technology. ANG
technology can provide adequate energy density at a low
pressure, that is, 3.5 MPa and at room temperature. Natural
gas storage capacity of an adsorbent is currently expressed
in terms of the delivered volume of natural gas per unit of
volume containing the adsorbent, measured at standard con-
ditions (P = 0.101 MPa, T = 273 K). To be a commercially
attractive system, the nal objective is to reach a delivered
quantity equal to 150 (v/v) under real operating conditions
[3]. Numerous aspects occur in ANG storage that affect its
performance, and hence, its viability. The rst one is the

Corresponding author. Tel.: +33 4 68 68 22 36;


fax: +33 4 68 68 22 13.
E-mail address: goetz@univ-perp.fr (V. Goetz).
nature of the microporous solid. The microporous structure
of the activated carbon denes the adsorption capacity per
unit of mass of adsorbent. Many theoretical [4,5] and ex-
perimental [6,7] studies and researches have been devoted
to the denition and the practical elaboration of the microp-
orous texture best adapted for methane storage. The second
one is the packing adsorbent scale that denes the storage
capacity per unit of volume. Densication is necessary to
try to reach the requirement of 150 (v/v) delivered by the
system. Numerous methods of densication are provided
in literature [810]. To be really effective, the densication
procedure should lead to a compromise between a high vol-
umetric capacity and high heat and mass transfer properties.
The third aspect corresponds to the scale of the ANG ves-
sel. Good adequacy is necessary between the design of the
ANG vessel, the transfer properties of the packing adsorbent
and the dynamic working conditions of the storage process
[11,12]. Finally, the last aspect concerns the inuence of the
composition of NG during successive cycles of charge and
discharge. This aspect is taken into consideration in the gen-
eral design of a NGVs process through, for example, the
implementation of a guard bed [13,14]. A theoretical study
[15] has shown the determining inuence of the gas compo-
sition. Nevertheless, in the literature, very few experimental
results quantify and analyse the evolution of performance
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.005
72 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179
of a storage vessel during real dynamic cycling operation
with NG [1618]. The purpose of the present work is to in-
vestigate this last point. An automatic cycling experimental
device has been developed. The global storage performance
of the vessel is measured as a function of the cycle number.
For each cycle, the gas composition at the outlet of the ves-
sel during the discharge is determined by gas chromatogra-
phy. The inuence on the performance of the system of the
different gas species that compose NG is described.
2. Experimental
2.1. Experimental device and procedure
Fig. 1 shows the experimental laboratory device. A cylin-
drical stainless steel vessel of about 2 l (diameter: 110.3 mm;
length: 215 mm) equipped with a central gas diffuser was
lled with the adsorbent packing. Four thermocouples were
distributed radially throughout the vessel with a spacing of
11 mm and one on the reactor wall to evaluate the thermal
gradient during charge and discharge phases. Heat exchange
through the vessel wall took place with water at 298 K inlet
temperature owing in a double casing around the reservoir.
Two mass ow controllers (MFC) recorded and controlled
the gas ow at the inlet and the outlet of the vessel. A pres-
sure transducer recorded the pressure in the vessel. At the
outlet of the vessel a micro-gas chromatograph allowed the
analysis of the composition of the delivered gas during dis-
charge. For safety reasons, a safety valve (SV) (calibrated at
4.0 MPa) was on the top of the storage vessel, and the appa-
ratus was placed in a ventilated safety closed room equipped
with gas detectors.
The experimental procedure is described as follows. Be-
fore the start of the rst run, the storage vessel was degassed
during several hours at a temperature higher than 423 K at a
pressure lower than 10 Pa. Before the cycling operation, the
data acquisition system was activated. During each charge,
Micro-Computer
Water
T
T
MFC MFC
P
micro-GC
NG
SV
VESSEL
Fig. 1. Experimental laboratory device for cycling operation.
0 1 2 3
0.00
0.02
0.04
0.06
0.08
0.10
0.12
Q
(
C
H
4
)

a
d
s
o
r
b
e
d

(
g
/
g
)
Pressure (MPa)
4
Fig. 2. Adsorption isotherm of methane in the activated carbon at 298 K.
a constant volumetric ow rate equal to 25 sl min
1
was in-
troduced into the vessel. Charge was considered over when
the pressure in the vessel was equal to 3.5 MPa and when the
difference between the mean temperature of the adsorbent
minus the heat transfer uid temperature was lower than
0.5 K. During each discharge, an almost constant volumet-
ric ow rate of 1 sl min
1
was delivered at the outlet of the
vessel. Every 4 min, the gas composition was measured with
a micro-gas chromatograph. Discharge was considered over
when the pressure in the vessel was lower than 0.15 MPa
and when the difference between the mean temperature of
the adsorbent minus the heat transfer uid temperature was
lower than 0.5 K. As soon as these conditions were reached
a charge procedure happened.
2.2. Preparation and characterisation of the adsorbent
A commercial activated carbon, manufactured by NORIT
was selected for the cycling operation. It is available in a
grain formwith particle sizes in the range of fewmillimetres.
Fig. 2 presents the adsorption isotherm of pure methane at
O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179 73
Table 1
Characteristics of the adsorbent composite block with and k, respectively,
the heat transfer conductivity and the permeability

AC
(kg m
3
)
ENG
(kg m
3
) (Wm
1
K
1
) k 10
12
(m
2
)
720 90 2.4 0.023
298 K measured with a high-pressure volumetric device [19]
under pressure up to 5 MPa.
From this raw material, a consolidated adsorbent com-
posite block was prepared following a procedure previously
detailed [10] in the case of Maxsorb

like activated car-


bon (produced and sold by the Kansai Coke and Chemical
Company Ltd). First, the activated carbon crushed in powder
form with particle sizes around 100 m was coated with a
thermoplastic binder according to Boses patent [20]. Then,
it was mixed with expanded natural graphite (ENG) and
the obtained mixture was compressed directly in the storage
vessel and then heated up to the melting point of the me-
chanical binder. Thereafter, the vessel was air-cooled under
applied stress. The mechanical binder solidication ensures
all mechanical cohesion of the composite. ENG, which is
an excellent thermal binder ensures high values of the heat
transfer conductivity of the composite and of the heat ex-
change coefcient between the composite block and the wall
of the vessel. As discussed in previous papers [10,12], this
procedure of preparation of composite block leads to two
main advantages. The density of the activated carbon in the
composite is increased. The thermal properties are improved
avoiding too much severe thermal limitation during dynamic
charge and discharge.
The two binders, the ENG and the thermoplastic polymer,
are inert concerning the methane adsorption. The methane
quantity delivered depends only on the activated carbon ad-
sorption isotherm and its apparent density in the adsorbent
composite block (Table 1). For pure methane, the quantity
delivered at 298 K between 3.5 and 0.1 MPa is equal to 89
(v/v).
3. Results and discussion
3.1. Cycling operation with pure methane
A preliminary cycling test was performed with pure
methane. It is illustrated in Fig. 3. According to the de-
scribed experimental procedure, in the rst cycle, the vessel
is charged from vacuum to 3.5 MPa at 298 K. During the
rst discharge, when the pressure decreases from 3.5 to
0.1 MPa, around 20% of the total amount of gas adsorbed
Table 2
Natural gas composition (mol%)
CH
4
N
2
CO
2
C
2
H
6
C
3
H
8
i-C
4
H
10
n-C
4
H
10
i-C
5
H
12
n-C
5
H
12
C
6+
92.18 1.82 0.75 4.23 0.78 0.13 0.09 0.012 0.006 0.002
0 10 20 30
0
1
2
3
4
0
15
30
45
60
75
90
Q
d
d

(
V
/
V
)
P
r
e
s
s
u
r
e

(
M
P
a
)
Time (hours)
Cycle 1 Cycle 5 Cycle 10
Fig. 3. Cycling operation with pure methane: pressure proles (continuous
line and left hand scale), adsorbed volume of methane per volume of
composite block (symbol () and right-hand scale).
into the composite at 3.5 MPa is retained. This corresponds
to the difference between the delivered and the stored quan-
tity of methane. As expected, during cycling operation, the
delivered quantity of methane is constant and is not inu-
enced by the number of cycles. This is the result of the
reversible adsorption/desorption phenomena of a pure gas
on an activated carbon. The experimental delivered quantity
(Fig. 3), directly obtained by the integration of the instanta-
neous ow rate of methane at the outlet of the vessel is in
agreement with the calculated one.
3.2. Cycling operation with natural gas
Gaz De France provided natural gas. It is composed of
92.2% of methane. The other components analysed with
the micro-gas chromatograph are: CO
2
, N
2
, C
2
H
6
, C
3
H
8
,
i-C
4
H
10
, n-C
4
H
10
, i-C
5
H
12
, n-C
5
H
12
and C
6+
(Table 2).
3.2.1. Comparison of three typical cycles
Fig. 4 shows experimental proles of the mean tempera-
ture, the pressure and the ow rate during charges of the cy-
cles 5, 100 and 700. Whatever the cycle, during the charge
performed at high ow rate, the exothermic adsorption of
the different components of natural gas entails an increase
of the mean temperature of the vessel (Fig. 4a). The range
in variation of the mean temperatures is the consequence
of the coupling between the power delivered by the gas ad-
sorption and the heat transfer inside the composite block.
The decrease of this range during the charge as a function of
the cycle number is a rst indication of the evolution of the
74 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179
0 2 4 6 8 10 12 14
295
300
305
310
M
e
a
n

t
e
m
p
e
r
a
t
u
r
e

(
K
)
Charge time (min)
0 2 4 6 8 10 12 14
0
1
2
3
4
P
r
e
s
s
u
r
e

(
M
P
a
)
Charge time (min)
0 2 4 6 8 10 12 14
0
5
10
15
20
25
30
F
l
o
w

r
a
t
e

(
s
l
/
m
i
n
)
Charge time (min)
(a)
(b)
(c)
Fig. 4. Experimental proles of the mean temperature (a), the pressure
(b) and the ow rate (c) during the cycle numbers 5 (), 100 () and
700 ().
behaviour of the storage vessel. At the end of the charge, as
soon as the constant ow rate at the inlet of the vessel can-
not be ensured any more, the mean temperature decreases
(Figs. 4a and c). Due to the high values of the heat con-
ductivity and the heat exchange coefcient at the wall, the
duration necessary to return to 298 K is very short and less
than 10 min. During discharge, the thermal properties of the
composite associated with the low regulated ow rate, al-
low a quite isothermal working mode. The three presented
charges were realised for the same regulated ow rates at the
inlet of the vessel (Fig. 4c). The evolution of the duration
necessary for complete charge as well as the pressure pro-
le (Fig. 4b) demonstrate the decrease of the cycled mass
of natural gas at the vessel scale.
Fig. 5 presents in the case of the selected cycles, the
evolution of the molar ratio for three different components.
It is dened as follows,
r
i
=
M
i outlet
M
i inlet
,
where M
i outlet
and M
i inlet
are, respectively, the outlet and
the inlet molar percentage of the component i. While the
inlet molar percentage is constant (Table 2), the outlet mo-
lar percentage of each species was measured during the dis-
charges with the micro-gas chromatograph. Every 4 min, an
analysis was performed. In order to be able to compare these
proles, they are presented as a function of the pressure at
the outlet of the vessel.
Whatever the gas and the cycle are, at the beginning of the
discharge, the ratios of all components are all equal to one.
This is the consequence of the dynamic cycling operation
with a discharge happening immediately at the end of the
charge. The rst volume of gas owing out of the vessel is
not desorbed by the activated carbon, but comes from the
dead gas volume of the vessel. Its composition is equal to
the inlet composition of the gas.
During a discharge, the composition of the gas at the
outlet is a consequence of the multi-component adsorption
equilibria between the previously adsorbed phase and the
gas phase. The ratio proles are necessarily linked to the
dynamic working mode of the vessel (like the ow rates
during charge and discharge) and the heat and mass transfer
properties of the composite adsorbent. Nevertheless, general
tendencies can be observed.
The outlet gas composition is not constant and depends
on the level of pressure. Desorption of the high-molecular
weight hydrocarbons happens mainly for pressure lower than
1 MPa. Between 3.5 MPa and 1 MPa, the equilibrium condi-
tions between the gas and the adsorbed phase entail values
of the different ratios systematically lower than 1. This is
especially pronounced for the i-butane and the i-pentane.
Ratio proles depend highly on the cycle number. They
clearly indicate the evolution of the composition of the ad-
sorbed phase during cycling operation. As an example, at
cycle number 5, whatever the level of the pressure, the
i-pentane (Fig. 5c) is never detected in gas mixture at the
outlet of the vessel. This means that the whole i-pentane in-
troduced in the vessel during the previous charges remains
in the adsorbed phase and is stored inside the activated car-
bon. This is not any longer the case for cycles 100 and 700.
As the cycling operation goes on, the composition of the ad-
O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179 75
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
Pressure during desorption (MPa)
r
C
2
H
6
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0

r
i
-
C
4
H
1
0
Pressure during desorption (MPa)
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0

r
i
-
C
5
H
1
2
Pressure during desorption (MPa) (c)
(b)
(a)
Fig. 5. Evolution of the ratio C
2
H
6
(a), i-C
4
H
10
(b), i-C
5
H
12
(c) at the
cycle numbers 5 (), 100 () and 700 ().
sorbed phase changes. It is progressively saturated by the hy-
drocarbons with molecular weight higher than methane and,
as a consequence, the ratio of these components increase.
During the discharges, the mean values (rm) of the ratios
corresponding to the pressures ranging from 3.5 to 0.1 MPa
were calculated (Table 3). These values indicate if the gas is
Table 3
Evolution of the mean ratio (rm) during all the study
rm
C
2
H
6
rm
i-C
4
H
10
rm
i-C
5
H
12
Cycle 5 0.80 0.35 0.03
Cycle 100 1.04 0.78 0.48
Cycle 700 1.04 0.99 0.77
or is no longer stored at the scale of the vessel during a suc-
cessive phase of charge and discharge. In the case of ethane,
at cycle number ve, the value of 0.80 for rm
C
2
H
6
points
out that around 20% of the ethane introduced in the vessel
remains stored during the discharge. At cycle number 100,
the value close to 1 indicates that a steady state is reached
for this gas. It is conrmed by the same value of rm
C
2
H
6
at
the cycle number 700. Moreover, the ratio prole at cycle
number 700 is strictly identical if compared to that of cycle
100 (Fig. 5a). The i-butane corresponds to the intermediate
case (Fig. 5b, Table 3). This gas continues to be stored in
the adsorbed phase after cycle number 100. After 700 cy-
cles, the steady state is reached. Finally, as demonstrated by
its ratio prole (Fig. 5c), even after 700 cycles, the activated
carbon is not saturated by i-pentane with around 20% of this
gas remained in the adsorbent during the discharge phase
(Table 3).
3.2.2. Efciency of the ANG process
The performance of ANG systems is usually dened as
the volume of methane or natural gas delivered, measured
at standard conditions per adsorbent volume. The impact of
the cycling operation on performances is evaluated by the
efciency (Fig. 6) dened as follows,
=
Q
n
Q
1
,
where Q
1
and Q
n
are the delivered gas volume per adsor-
bent volume at cycle number 1 and n, respectively. Q
1
is
Fig. 6. Experimental prole of the efciency as a function of the cycle
number (H.H. is for heavy hydrocarbons).
76 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179
Cycle number
0 10 20 30 40 50
0.5
0.6
0.7
0.8
0.9
1.0
0 10 20 30 40 50
0.0
0.2
0.4
0.6
0.8
1.0
1.2
r
m
C
O
2
Cycle number
Carbon dioxide ( CO
2
)
(a) (b)
Fig. 7. Relation between the mean ratio of CO
2
in the gas at the outlet of the vessel (a) and the rate of variation of the efciency (b).
equal to 83 (v/v), close to the delivered quantity of pure
methane.
First it is interesting to note the continuous decrease of
the efciency. After 700 cycles, the steady state conditions
are not yet completely reached and the efciency goes down
very slowly. At the end of the cycling operation, less than a
half of the initial delivered quantity of gas is still cycled by
the storage vessel. This is the result of the adsorption of the
gases other than methane present in the mixture at low per-
centage. Secondly, the rate of decrease is not constant and
depends on the number of cycles already performed. The
straight lines in Fig. 6 show that several rates of variation
can be drawn. These regimes are linked to the storage in the
vessel of the different species that compose natural gas. It is
conrmed by the evolution of the mean ratios of the gas as a
function of the number of cycles. The rst rate of variation
corresponds to the progressive saturation of the adsorbent by
the carbon dioxide. During this step CO
2
, like all the gases
with a molecular weight higher than methane is stored in the
vessel. As demonstrated by the Fig. 7a, at the scale of the
vessel a steady state is reached for CO
2
at a cycle number
around 10. As soon as the mean ratio of CO
2
is equal to one,
the rate of variation of the efciency changes (Fig. 7b). The
second step corresponds to the number of cycle necessary
to reach the steady state for ethane (Fig. 8a), which is after
methane, the lightest hydrocarbon in the gas mixture. Be-
0 10 20 30 40
0.5
0.6
0.7
0.8
0.9
1.0
Cycle number
CO
2
50
0 0 10 20 30 40 5
0.0
0.2
0.4
0.6
0.8
1.0
1.2
r
m
C
2
H
6
Cycle number
Ethane ( C
2
H
6
)
(a) (b)
Fig. 8. Relation between the mean ratio of C
2
H
6
in the gas at the outlet of the vessel (a) and the rate of variation of the efciency (b).
tween the cycles 10 and 30, a second rate of variation char-
acterised by a second straight line can be drawn (Fig. 8b). As
for CO
2
, as soon as the quantity of C
2
H
6
stored in the vessel
at the end of the discharge phase does not change any more,
another regime begins. Saturation of the vessel with propane
that ends at cycle number 150 denes the third rate of vari-
ation (Fig. 9). Then, the evolution of the global efciency is
closely linked to the successive saturations of the activated
carbon with the different gases. At the end of the cycling op-
eration, the slow decrease of the efciency is the result of the
slow accumulation of n-butane and gases with higher molec-
ular weight (cf. Appendix A). If the cycling operation had
been continued, the last part of the efciency prole (Fig. 6)
should probably be decomposed with the different gases still
stored.
The global efciency of the ANG process is inuenced
greatly by the natural gas composition. Nevertheless, the
molar percentage of methane at the outlet of the vessel dur-
ing discharge changes little during the cycling operation
(Fig. 10). It is a little bit higher than 0.92, the gas supply
composition, during the rst 50 cycles. Since nitrogen is not
adsorbed at the working levels of pressure and temperature
(cf. Appendix A), this is the consequence of the storage of
CO
2
and C
2
H
6
, the gas having the highest molar percent-
ages in the gas feed. When this process is nished, because
of the very low percentage of the others components of
O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179 77
0 50 100 150 200
0.5
0.6
0.7
0.9
0.8
1.0
Cycle number
CO
2
C
2
H
6
0 50 100 150 200
0.0
0.2
0.4
0.6
0.8
1.0
1.2
r
m
C
3
H
8
Cycle number
Propane (C
3
H
8
)
(a) (b)
Fig. 9. Relation between the mean ratio of C
3
H
8
in the gas at the outlet of the vessel (a) and the rate of variation of the efciency (b).
0 200 400 600 800
91
92
93
94
95
Cycle number
CH
4
concentration (molar %)
at the outlet of the vessel
CH
4
concentration (molar %)
at the inlet of the vessel
CH
4
concentration
(molar %)
Fig. 10. Mean molar percentage of methane during discharge as a function
of the cycles number.
natural gas, the molar percentage of methane at the outlet is
almost constant and equal to the inlet one.
4. Conclusion
The obtained experimental results display the great im-
pact of the gas composition on the performance of an ANG
storage system. After 700 cycles, corresponding approxi-
mately to 250,000 km for a vehicle with a fuel tank allowing
400 km of autonomy, the efciency of the storage system
falls down by 50%. The number of cycles necessary to reach
a steady state working mode, as well as the accurate value of
the nal performance, probably depends on the composition
of natural gas and the dynamic working mode selected for
the cycling operation. Moreover, the microporous structure
of the activated carbon should inuence the result. Never-
theless, basically, the nal performance is the result of the
multi-component adsorption equilibriums at a temperature
of 298 K for the two characteristic working pressures, 3.5
and 0.1 MPa. Gas chromatography analysis of the stream
composition at the outlet of the vessel shows that, as ex-
pected, the different components contained in the natural
gas, slowly obstruct the adsorption sites for methane. With
the system of analysis retained, it was not possible to detect
the admixed sulphur-containing odorants which are present
in natural gas at the level of some parts per million. Like the
alkanes with molecular weight higher than that of methane,
they can be suspected to participate in the slow decrease of
the cycled mass of gas. Adsorption of these components by
the activated carbon is not irreversible. Modications in the
working conditions make it possible to restore the initial per-
formance. As an example, at the end of the discharge when
the pressure is equal to 0.1 MPa, heating the vessel to a tem-
perature equal to 473 K entails a partial desorption of the
different species. Following this treatment, a cycle leads to
an experimental value of the efciency equal to 0.65. Nev-
ertheless, a complete regeneration of the activated carbon
necessitates heating the vessel at 473 K under vacuum. Stor-
age capacity is currently the main limitation to the practical
development of ANG technology. It is therefore crucial to
design an efcient process, adapted to the practical use of
this storage system. At the present time, a guard bed at the
inlet of the storage vessel is the proposed solution.
Acknowledgements
This work has been done with the nancial support of
Gaz De France and ADEME.
Appendix A. Experimental proles of the mean ratios
for nitrogen, i-butane, n-butane, i-pentane, n-pentane,
and the hydrocarbons with molecular weight higher
than n-pentane (C
6+
)
For safety reasons, the cycling operation is interrupted
during the weekend. After an interruption, always happening
at the end of a discharge phase, several cycles are necessary
to join the continuous curve representative of the evolution of
the behaviour of the tank. This phenomenon tends to demon-
strate the importance of the kinetics of the multi-component
adsorption equilibrium in the case of a complex mixture
of gas.
78 O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179
O. Pupier et al. / Chemical Engineering and Processing 44 (2005) 7179 79
References
[1] J. Wegrzyn, M. Gurevich, Adsorbent storage of natural gas, Appl.
Energy 55 (1996) 7183.
[2] O. Talu, An overview of adsorptive storage of natural gas, in: M.
Suzuki (Ed.), Proceedings of the fourth International Conference on
Fundamentals of Adsorption, Kyoto, 1992, pp. 655622.
[3] N.D. Parkins, D.F. Quinn, Natural gas adsorbed on carbon, in: J.W.
Patrick (Ed.), Porosity in Carbons, Great Britain, Edward Arnold,
1995, pp. 291325.
[4] K.R. Matranga, A.L. Myers, E.D. Glandt, Storage of natural gas by
adsorption on activated carbon, Chem. Eng. Sci. 47 (1992) 1569
1579.
[5] S. Bilo, V. Goetz, A. Guillot, Optimal design of an activated carbon
for an adsorbed natural gas storage system, Carbon 40 (2002) 1295
1308.
[6] D. Lozano-Castello, D. Cazorla-Amoros, A. Linares-Solano, D.F.
Quinn, Inuence of pore size distribution on methane storage at
relatively low pressure: preparation of activated carbon with optimum
pore size, Carbon 40 (2002) 9891002.
[7] D. Lozano-Castello, D. Cazorla-Amoros, A. Linares-Solano, Can
highly activated carbons be prepared with homogeneous micropore
size distribution, Fuel Process. Technol. 7778 (2002) 325330.
[8] J.A.F. Mac Donald, D.F. Quinn, Carbon adsorbent for natural gas
storage, Fuels 77 (1998) 6164.
[9] T.D. Burchell, M.R. Rogers, Carbon bers based natural gas stor-
age monoliths, Extended Abstract, Gas storage workshop, Kingston,
2001.
[10] S. Bilo, V. Goetz, S. Mauran, Characterisation of adsorbent com-
posite blocks for methane storage, Carbon 39 (2001) 16531662.
[11] K.J. Chang, O. Talu, Behaviour and performances of adsorptive
natural gas storage cylinders during discharge, Appl. Therm. Eng.
16 (1996) 359374.
[12] S. Bilo S, V. Goetz, S. Mauran, Dynamic discharge and performance
of a new adsorbant for natural gas storage, AIChE J. 47 (2001)
28192830.
[13] K. Dvorak, R.C. Hodrien, Development of adsorbed natural gas
technology for large scale diurnal storage application, in: Proceedings
of IGRC, Amsterdam, 2001.
[14] K. Seki, Y. Sumie, Development of adsorptive natural gas storage
systemApplication to gas holder and natural gas vehicle, in: Pro-
ceedings of IGRC, Amsterdam, 2001.
[15] J.P.B. Mota, Impact of gas composition on natural gas storage by
adsorption, AIChE J. 45 (1999) 986996.
[16] A. Golovoy, E.J. Blais, Natural gas storage on activated carbon, in:
Proceedings of SAE Conference, Warrendale, 1983, pp. 4753.
[17] A.S. Pedersen, B. Larsen, Adsorption of methane and natural gas on
six carbons, Rap. Riso-M-2781, Riso National Laboratory, Denmark,
1989.
[18] L. Czepirski, Some aspects of a sorbent-containing storage system
for natural gas, Indian J. Tech. 29 (1991) 266268.
[19] A. Guillot, S. Follin, L. Poujardieu, Adsorption studies of carbon
dioxide on KF1500, X2MH6/8 and PX21 by means of a high-pressure
volumetric device, in: B. Mc Enaney, T.J. Mays, J. Rouquerol, F.
Rodriguez-Reinoso, K.S.W. Sing, K.K. Unger (Eds.), Proceedings
on Characterisation of Porous Solid IV, London, 1997, pp. 573
580.
[20] T.K. Bose, R. Chahine, J.M. St-Arnaud, High density Adsorbent and
Method Producing the Same, US Patent 4,999,330, 1991.
Chemical Engineering and Processing 44 (2005) 8187
Effect of internal on the hydrodynamics in external-loop airlift reactors
Tongwang Zhang, Jinfu Wang

, Tiefeng Wang, Jing Lin, Yong Jin


Department of Chemical Engineering, Tsinghua University, Beijing 100084, PR China
Received 12 March 2004; received in revised form 4 May 2004; accepted 5 May 2004
Available online 19 June 2004
Abstract
Small bubbles and ow uniformity are important for gasliquid and gasliquidsolid multiphase reactors. A reactor internal was designed
and installed in an external-loop airlift reactor (EL-ALR) to enhance bubble breakup and owredistribution and improve reactor performance.
Hydrodynamic parameters, including local gas holdup, bubble rise velocity, bubble Sauter diameter and liquid velocity were measured. A
radial maldistribution index was introduced to describe radial non-uniformity in the hydrodynamic parameters. The inuence of the internal
on this index was studied. Experimental results show that The effect of the internal is to make the radial proles of the gas holdup, bubble
rise velocity and liquid velocity radially uniform. The bubble Sauter diameter decreases and the bubble size distribution is narrower. With
increasing distance away from the internal, the radial proles change back to be similar to those before contact with it. The internal improves
the ow behavior up to a distance of 1.4 m.
2004 Elsevier B.V. All rights reserved.
Keywords: Airlift reactor; Internal; Radial prole; Gas holdup; Bubble size; Bubble breakup
1. Introduction
External-loop airlift reactors (EL-ALRs) have drawn
much attention due to their simple construction, good heat
and mass transfer, and good mixing characteristics as the
gas phase in the reactor serves the dual functions of aer-
ation and agitation [1]. EL-ALRs have been used in a
variety of industrial applications where intimate contact
of gasliquidsolid phases is necessary in the chemical,
petrochemical, minerals processing and biochemical pro-
cesses [2]. The special feature distinguishing EL-ALRs
from bubble columns is the recirculation of the liquid
through a downcomer that connects the gasliquid separa-
tor and the bottom of the riser. The density difference due
to the gas passing through the riser induces circulation of
the liquid and the solid in the EL-ALR. The circulation
creates good mixing in all phases and provides good mass
transfer.
Mass transfer is an important issue in the reaction pro-
cesses in EL-ALRs, especially in the heterogeneous ow
regime. The volumetric mass transfer rate is determined

Corresponding author. Tel.: +86 10 62785464;


fax: +86 10 62772051.
E-mail address: wangjf@otu.org (J. Wang).
by the mass transfer coefcient and the interfacial area.
Since smaller bubbles and larger gas holdups have higher
specic interfacial area, a decrease in bubble size or an
increase in gas holdup can improve mass transfer and in-
crease liquid production at the same supercial gas velocity
[3]. A uniform radial prole of the gas holdup decreases
bubble-bubble collisions, which in turn decreases bubble
coalescence. Therefore, it is valuable to provide an approach
to reduce bubble size and make the gas holdup radial prole
more uniform.
Previous workers have studied the inuence of the ratio
of the cross-section area of the downcomer to the riser [4,5],
the reactor height [6,7], the gasliquid separator congura-
tion [8], and the distributor type and location [9]. All these
affect the ow characteristics and mass transfer. Most pre-
vious works focus on global parameters, such as the liquid
circulation velocity [1013] and the average gas holdup in
the riser [1416]. Although much work has been carried out
on EL-ALRs, the proper design and scale-up of an EL-ALR
is still difcult because any variation in the physical proper-
ties of the gas or the liquid and the reactor structural feath-
ers can have a considerable effect on the hydrodynamics
[7].
Smaller bubbles and higher gasliquid interface re-
newal frequency are benecial to improving mass transfer.
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.003
82 T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187
Intensied turbulence can destroy the stability of large
bubbles, which leads to smaller bubbles, and increase the
frequency of bubble coalescence and breakup, which im-
proves the surface renewal frequency of bubbles [17,18].
Intensied turbulence can be achieved by changing the gas
sparger to decrease the initial bubble size and improve its
radial distribution, but the effective region of the gas sparger
is only limited to a certain height above the distributor [19].
Liquid mixing time decreases sharply for an initial in-
crease in the gas sparging rate and approach an asymptotic
value that is determined by the height and diameter of the
downcomer and the liquid properties [5]. A higher liquid
velocity shortens the gas residence time and results in a de-
crease of gas holdup and interfacial area. The radial pro-
le of the liquid is parabolic. These are disadvantageous for
mass transfer. The mounting of internals in a xed bed is
often used to improve the radial prole of the liquid veloc-
ity. This motivates us to mount internals in an EL-ALRs to
improve the radial prole of the gas holdup and the liquid
velocity and to intensify turbulence.
As noted above, small bubbles, a uniform gas holdup ra-
dial distribution and an appropriate liquid circulating veloc-
ity can intensify mass transfer between the gas phase and
the continuous phase and improve the production efciency
in EL-ALRs. In order to reduce the bubble size and obtain a
more uniform radial distribution of the local gas holdup and
the liquid and bubble rise velocities, and regulate the liq-
uid circulating velocity appropriately as well, in this work, a
novel internal is used and mounted in the riser column to im-
prove hydrodynamics and mass transfer. The hydrodynamic
behavior and mass transfer characteristics of an EL-ALR
with the new designed internal are investigated.
2. Experimental setup
An EL-ALR was set up for the experimental study and
fabricated using plexiglas. The schematic diagram of the
experimental setup is shown in Fig. 1. The riser and the
downcomer are 0.230 and 0.190 m in diameter, respectively,
and 4.8 m in height. The height and the diameter of the
airliquid separator are 0.960 and 0.480 m, respectively.
The downcomer is connected to the riser 0.2 m above the
distributor.
Air was used as the gas phase and was introduced into the
system through a distributor with holes of diameter 1 mm
and a perforation of 0.25%. The supercial gas velocity,
based on the riser cross-section area, varied from 0.0067 and
0.0535 m/s.
The internal has been described in patent CN 1,403,190A
[20], and is shown in Fig. 2. It is 230 mm in diameter and
100 mm in height. The angle between the bafe of the in-
ternal and the vertical axis is 45

. Each bafe is 30 mm in
width and 1 mm in thickness. There are some semicircular
holes on the plate and each hole has a tongue-like plate fac-
ing the upstream to break bubbles. Because of the returning
Fig. 1. Schematic of the experimental equipment.
ow of the downcomer, the ow near the sparger is not axial
symmetry. But the unsymmetry will disappear about 130 cm
above the sparger. So the internal is installed 165 cm above
the sparger.
Fig. 2. Photo of the internal used in this work.
T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187 83
The local gas holdup and bubble behavior were measured
by a reective optic ber probe developed by Wang and
co-workers [21,22]. It can be known whether the probe is im-
merging in the gas. The rate of the time that probe immerg-
ing in the gas and the total sample time is gas holdup. Gas
velocity can be got by the time difference that one bub-
ble touch two probes and the distance between two probes.
Chord length can be obtained from one bubble velocity and
the time that the probe stays in the bubble. Bubble size dis-
tribution is got from the probability density of the chord
length based on some numerical method. The local liquid
velocity in the riser was measured by a backward scattering
LDA system (system 9100-8, model TSI). Details have been
given by Lin et al. [23].
3. Results and discussion
3.1. Gas holdup
The upper surface of the internal is dened as the zero
of the axial position. Axial positions are positive above this
and negative below this. The radial proles of the gas holdup
at ve axial positions were measured, as shown in Fig. 3.
The radial prole of the gas holdup becomes much atter
after owing through the internal, with an increase in the
gas holdup near the wall and a decrease in the center region
as compared with the gas holdup below the internal. As
the distance above and away from the internal increases,
the prole becomes more and more similar to that before
contact with the internal. The local radial proles of the gas
holdup at axial positions 144 and 209 cm show almost no
difference. This shows that the inuence of the internal on
the radial prole of the gas holdup becomes weaker and
weaker with increasing distance away from the internal and
is no longer felt beyond a certain distance. The analysis of
the local gas holdup proles shows that the inuence of
the internal persists to a distance of about 1.4 m. In order
to show the effect of the internal on the radial prole of
Fig. 3. Radial prole of the gas holdup at different axial positions.
Fig. 4. Radial maldistribution index of the local gas holdup at different
axial positions.
the gas holdup clearly, a radial maldistribution index, , is
introduced, dened as:
=

1
n 1
n

i=1
(y
i
y)
2
(1)
where y can be gas holdup, bubble velocity, bubble diameter
or liquid velocity. Fig. 4 shows the change in maldistribu-
tion index

with axial position at different supercial gas


velocities. It can be seen that the radial maldistribution has
a minimum a short distance after passing the internal and
then it increases gradually. The

difference between 144


and 209 cm is negligible. Therefore, it can be concluded that
the region of the inuence of the internal on the gas holdup
is within 140 cm.
Fig. 5 shows axial changes in the cross-sectional aver-
aged gas holdup at different supercial gas velocities with
an increase in the distance above the internal. The average
gas holdup rst increases to a maximum at about 40 mm
above the internal and then decreases gradually. Note that
at the supercial gas velocity of 0.0067 m/s, the maximum
gas holdup appears somewhat earlier. At a low supercial
Fig. 5. Cross-sectional gas holdup at different axial positions.
84 T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187
velocity, the inuence of the internal on liquid turbulence
and its effect on breaking bubbles are weak, which results
in the maximum gas holdup appearing somewhat earlier. As
expected, the average gas holdup is almost unchanged when
the axial height exceeds 140 cm. An enhancement in turbu-
lence intensity was observed after liquid owed through the
internal. Enhanced turbulence intensity effectively forces the
bubble breakup and results in an increase in the gas holdup
[24]. In the population balance model [17], bubble size is
determined by an equilibrium between bubble breakup and
coalescence. Due to enhanced turbulence by the internal,
bubble breakup will be dominant in the downstream region
after the internal, which leads to a decrease in the bubble
size and an increase in the local gas holdup. With increasing
axial height, the turbulent intensity decreases and the local
gas holdup decreases. When the axial height is more than
144 cm above the internal, the inuence of the internal has
vanished and the local gas holdup is the same as in the state
below the internal.
Fig. 6 presents the experimental results on the gas holdup
proles with and without the internal at the axial position
of 74 cm for different supercial gas velocities. The inter-
nal causes extra ow resistance, which in turn decreases the
liquid circulation velocity and increases gas holdup [5]. In
addition, the experimental results show that the radial pro-
les of the local gas holdup with an internal are atter than
those without the internal. Therefore, a properly designed
internal can have dual function of increasing the gas holdup
and improving its radial prole.
3.2. Bubble rise velocity
It is important to study the bubble rise velocity and its
radial prole in a gasliquid system as these are closely re-
lated to the hydrodynamics, and mass and heat transfer [25].
Bubble rise velocity and its radial prole have also signif-
icant inuences on gas and liquid residence time distribu-
tions. A suitable bubble rise velocity and radial prole can
improve production efciency. Bubble rise velocities in a
Fig. 6. Inuence of the internal on the local gas holdup for different
supercial gas velocity.
Fig. 7. Bubble rise velocity radial prole with axial positions.
gasliquid system are affected by the physical properties of
each phase, bubble size, local gas holdup and liquid veloc-
ity. Larger bubbles have higher rise velocities and the bubble
swarm velocity is higher than that of a single isolated bubble
[26]. Bubble breakup is caused by collisions between bub-
bles and turbulent eddies [24]. Thus, the turbulence inten-
sity has an important effect on the bubble rise velocity and
methods that intensify the turbulence intensity can improve
the bubble rise velocity and its radial prole.
Fig. 7 shows the radial proles of the bubble rise velocity
at different axial positions. The radial proles of bubble rise
velocity become much more uniform after the gasliquid
mixture owed through the internal, with an increase near
the wall and a decrease in the central region compared with
that below the internal. As the distance above the internal
increases, the radial prole of the bubble rise velocity be-
comes more and more similar to that below the internal. At
the axial position of 144 cm, the radial prole of the bubble
rise velocity is about the same as that below the internal.
To show the effect of the internal on the average bubble
rise velocity more clearly, the cross-sectional average bub-
ble rise velocity, u
b
, and the radial maldistribution index of
bubble rise velocity,
u
b
are introduced. u
b
is dened as:
u
b
=
2

R
0

g
(r)u
b
(r)rdr
A
g
(2)
where A is the cross-sectional area.
Fig. 8 shows the average bubble velocity as a function
of the axial distance at different supercial gas velocities.
The average bubble velocity decreases after contact with the
internal to a minimum, then it increases gradually to the
original velocity above the position of 144 cm. Fig. 9 shows

u
b
as a function of the axial distance at different supercial
gas velocities. It decreases notably after contact with the
internal to a minimum at 39 cm above the internal, then
increases gradually. According to the analysis of Figs. 79,
the radial prole of the bubble rise velocity above the internal
are much atter than that below, and the average bubble rise
velocity increases after the bubble had passed through the
internal. The change caused by the internal gets smaller and
T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187 85
Fig. 8. Average bubble rise velocity as a function of the axial position.
then vanishes gradually with increasing distance from the
internal.
3.3. Bubble Sauter diameter
Precise knowledge of the bubble size is important for a
better understanding of the hydrodynamics, mass transfer,
and reactor design. Zun [27] concluded that small bubbles
tend to have uniform radial proles while large bubbles tend
to rise in the central region. Smaller bubbles and a uniform
radial prole lead to a larger interfacial area and a lower rise
velocity, which increases mass transfer.
The Sauter bubble diameter, d
32
, is commonly used to
show the statistic size distribution of the bubbles [28].
d
32
=

n
i
d
3
i

n
i
d
2
i
(3)
Fig. 10 shows the radial prole of the bubble Sauter di-
ameter at ve axial positions. The bubble size is very much
decreased after owing through the internal and then in-
creases with an increase of the distance from the internal.
The difference between the values at the positions of 144
and 209 cm above the internal is negligible and their radial
Fig. 9. Axial change in the radial maldistribution index of the bubble rise
velocities.
Fig. 10. Bubble Sauter diameter radial prole at different axial positions.
prole is similar to that below the internal. According to
the analysis of bubble diameter, it can be concluded that the
effective distance of the internal is about 140 cm.
While owing through the internal, bubbles rise along the
undersurface of the bafes and collide with the tongue-like
bars, and are broken up into smaller bubbles as shown in
Fig. 10. The difference in the direction of adjacent bafes
increases the liquid turbulent intensity, which is benecial
to the breakup of large bubbles. With increasing distance
from the internal, the turbulent intensity decreases and coa-
lescence becomes dominant until a new equilibrium between
the breakup and coalescence of bubbles is reached.
3.4. Bubble size distribution
The bubble size detected by the probe is subject to a
probability distribution [28,29] and only the mean statistical
value of the bubbles is given by d
32
. The analysis of the
bubble size distribution helps us understand the uniformity
of the bubble sizes and its axial evolution. Fig. 11 shows
the axial change of the bubble size distribution at r/R = 0
and r/R = 0.97, respectively. Below the internal, the peak
Fig. 11. Bubble size distribution at different axial positions.
86 T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187
of the distribution at r/R = 0.97 is to the left of that at r/R
= 0. Due of bubble breakup by the internal, the bubble size
distributions at the center and near the wall are narrower than
that obtained below the internal and the peaks are shifted to
the left after passing through the internal. This means that the
bubble size decreases and the bubble size distribution is more
uniform above as compared with that below. With increasing
distance from the internal, the effect of the internal vanishes
gradually and the bubble size distributions become wider and
the peaks shift to the right. There is no difference between
the size distributions obtained at the position of 144 and
209 cm above the internal, which means that the region of
the inuence of the internal on bubble size distribution is
about 1.4 m.
3.5. Liquid velocity
Fig. 12 shows the radial prole of the liquid velocity at
different axial positions. Due of the bafes, the liquid is re-
distributed in the radial direction and the turbulent intensity
is increased. The radial prole of the liquid velocity is al-
most uniform after passing the internal. Liquid velocity is
lower at the center and higher near the wall as compared
with that below the internal. With increasing distance from
the internal, the turbulence intensity diminishes and the wall
effect becomes more apparent, that is, the liquid velocity in-
creases at the center and decreases near the wall. The radial
prole obtained at the position of 114 cm from the internal
is similar to that obtained below the internal and is the same
as that at the position of 144 cm.
The radial maldistribution index of the liquid velocity,

u
l
, is dened according to Eq. (1). Fig. 13 shows the ax-
ial changes in
u
l
at different supercial gas velocities.
u
l
is much decreased after passing through the internal and
reaches a minimum at 5 cm above the internal, then it in-
creases gradually with increasing axial height. According to
the analysis of the radial prole of the liquid velocities, it
can be concluded that the region in which the internal inu-
ences the liquid velocity is only within 110 cm.
Fig. 12. Radial prole of the liquid velocity.
Fig. 13. Axial change of the radial maldistribution index of liquid veloc-
ities.
4. Conclusions
Mass transfer is essential in EL-ALRs. Smaller bubbles
and a uniform gas holdup radial distribution increase the in-
terfacial area and improve mass transfer. Intensied turbu-
lence increases the surface renewal frequency and decreases
bubble size. A novel internal to improve mass transfer and
the hydrodynamic behavior in a gasliquid system is re-
ported. Experiments were carried out to study the effect of
the internal on the bubble behavior and liquid velocity in an
EL-ALR.
The radial proles of the gas holdup and bubble rise veloc-
ity become more uniform after passing through the inter-
nal. With increasing distance from the internal, the radial
proles of the gas holdup and bubble rise velocity change
back to be similar to that below the internal. At 144 cm
above the internal, the radial proles of the gas holdup
and the bubble rise velocity are the same as those below,
while

and
u
b
reach their minimum 39 cm above the
internal.
The bubble Sauter diameter very much decreases after
passing the internal, then increases with increasing dis-
tance from it. The bubble size distribution becomes nar-
rower after the internal and its peak is shifted to the left.
The bubble size at 144 cm above the internal has increased
back to be the same as that below.
Turbulence is intensied after passing through the inter-
nal. The radial prole of the liquid velocity becomes at-
ter.
u
l
reaches a minimum 5 cm above the internal. The
liquid velocity 114 cm above the internal is the same as
that below, that is, the internal affects the liquid velocity
up to about 110 cm above the internal.
The internal used in this work has good performance in
intensication of gasliquid mass transfer and in improv-
ing the ow distribution, and has a good perspective in
application in airlift reactors. Meanwhile, further study on
the internal should be carried out.
T. Zhang et al. / Chemical Engineering and Processing 44 (2005) 8187 87
Acknowledgements
The authors gratefully acknowledge nancial support by
the National Natural Science Foundation of China (No.
20276035) and by the Ministry of Education Ph.D. Fellow-
ship Program (No. 20020003060).
Appendix A. Notations
A cross-section of the reactor (m
2
)
d
32
Sauter bubble diameter (mm)
H axial distance from the upper surface of
the internal (m)
r radial position in the reactor (m)
R radius of the riser (m)
u local velocity (m/s)
U supercial velocity (m/s)
local gas holdup
radial maldistribution index
Subscriptions
b bubble
g gas phase
l liquid phase
u velocity
gas holdup
References
[1] J.C. Merchuk, S. Yehuda, Local holdup and liquid velocity in air-lift
reactors, AIChE J. 27 (1981) 377388.
[2] A.G. Livingston, S.F. Zhang, Hydrodynamic behavior of three-phase
(gasliquidsolid) airlift reactor, Chem. Eng. Sci. 48 (1993) 1641
1654.
[3] S. Guet, G. Ooms, R.V.A. Oliemans, R.F. Mudde, Bubble injector
effect on the gas-lift efciency, AIChE J. 49 (2003) 22422252.
[4] K.H. Choi, W.K. Lee, Circulation liquid velocity, gas hold-up, vol-
umetric oxygen transfer coefcient in external-loop airlift reactors,
J. Chem. Technol. Biotechnol. 56 (1993) 5158.
[5] M.A. Young, R.G. Carbonell, D.F. Ollis, Airlift bioreactors: analysis
of local two-phase hydrodynamics, AIChE J. 37 (1991) 403428.
[6] A.B. Russell, C.R. Thomas, M.D. Lilly, The inuence of vessel
height and top-section size on the hydrodynamic characteristics of
airlift fermentors, Biotechnol. Bioeng. 43 (1994) 6976.
[7] C. Bentifraouine, C. Xuereb, J.P. Riba, An experimental study of
the hydrodynamic characteristics of external-loop airlift contactors,
J. Chem. Technol. Biotechnol. 69 (1997) 345349.
[8] M.H. Siegel, J.C. Merchuk, Hydrodynamics in rectangular airlift
reactors: scale-up and the inuence of gasliquid separator design,
Can. J. Chem. Eng. 69 (1991) 465473.
[9] S. Becker, A. Sokolichin, G. Eigenberger, Gasliquid ow in bubble
column and loop reactors. Part II. Comparison of detailed experiments
and ow simulations, Chem. Eng. Sci. 49 (1994) 57475762.
[10] M.Y. Chisti, B. Hallard, M. Moo-Young, Liquid circulation in airlift
reactors, Chem. Eng. Sci. 43 (1988) 451457.
[11] A.G. Jones, Liquid circulation in a draft tube bubble column, Chem.
Eng. Sci. 40 (1985) 449462.
[12] Y.C. Hsu, M.P. Dudukovic, Gas holdup and liquid recirculation in
gas-lift reactors, Chem. Eng. Sci. 35 (1980) 134141.
[13] M. Gavrilescu, R.Z. Tudose, Study of the liquid velocity in
external-loop airlift reactor, Bioprocess Eng. 14 (1995) 183193.
[14] J.H. Hills, The operation of a bubble column at high throughputs. I.
Gas hold-up measurement, Chem. Eng. J. 12 (1976) 8999.
[15] M.H. Siegel, J.C. Merchuk, K. Schugerl, Airlift reactor analysis
interrelationships between riser, downcomer and gasliquid behavior,
including gas recirculation effects, AIChE J. 32 (1986) 15851595.
[16] J.B. Joshi, V.V. Ranade, S.D. Gharat, S.S. Lele, Sparged loop reactors,
Can. J. Chem. Eng. 68 (1990) 705741.
[17] M.J. Prince, H.W. Blanch, Bubble coalescence and break-up in
air-sparged bubble columns, AIChE J. 36 (1990) 14851499.
[18] C. Maretto, R. Krishna, Modeling of a bubble column slurry reactor
for FischerTropsch synthesis, Catal. Today 52 (1999) 279289.
[19] S.M. Shavaraju, T.W.F. Russel, H.W. Blanch, The design of gas
sparged devices for viscous liquid systems, AIChE J. 24 (1978)
454466.
[20] Y. Jin, T. F. Wang, J. F. Wang, T. W. Zhang, CN Patent 1,403,190A
(2003).
[21] T.F. Wang, J.F. Wang, W.G. Yang, Y. Jin, Bubble behavior in
gasliquidsolid three-phase circulating uidized beds, Chem. Eng.
J. 84 (2001a) 397404.
[22] T.F. Wang, J.F. Wang, W.G. Yang, Y. Jin, Experimental study on
gas-holdup and gasliquid interfacial area in TPCFBs, Chem. Eng.
Commun. 187 (2001b) 251263.
[23] J. Lin, M.H. Han, T.F. Wang, T.W. Zhang, J.F. Wang, Y. Jin, Exper-
imental study on the local hydrodynamic behavior of a three-phase
external-loop airlift reactor, Ind. Eng. Chem. Res., in press.
[24] T.F. Wang, J.F. Wang, Y. Jin, A novel theoretical breakup kernel
function for bubbles/droplets in a turbulent ow, Chem. Eng. Sci.
58 (2003) 46294637.
[25] K. Tsuchiya, A. Furumoto, L.S. Fan, J. Zhang, Suspension viscosity
and bubble rise velocity in liquidsolid uidized beds, Chem. Eng.
Sci. 52 (1997) 30533066.
[26] R. Krishna, M.I. Urseanu, J.M. van Baten, J. Ellenberger, Rise
velocity of a swarm of large gas bubble in liquids, Chem. Eng. Sci.
54 (1999) 171183.
[27] I. Zun, The transverse migration of bubble inuenced by walls in
vertical bubbly ow, Int. J. Multiphase Flow 6 (1980) 583588.
[28] S.A. Patel, J.G. Daly, D.B. Bukur, Holdup and interfacial area mea-
surements using dynamic gas disengagement, AIChE J. 35 (1989)
931942.
[29] D. Sabtana, A. Macias-Machin, Local bubble-size distribution in
uidized beds, AIChE J. 46 (2000) 13401347.
Chemical Engineering and Processing 44 (2005) 89100
Multicriteria synthesis of exible heat-exchanger networks
with uncertain source-stream temperatures
Cheng-Liang Chen

, Ping-Sung Hung
Department of Chemical Engineering, National Taiwan University, Taipei 10617, Taiwan, ROC
Received 23 May 2003; received in revised form 28 July 2003; accepted 31 March 2004
Available online 2 July 2004
Abstract
A multi-criteria synthesis strategy for heat-exchanger networks (HENs) simultaneously considering minimum utility consumption, max-
imum source-stream temperature exibility, and even minimum number of matches is proposed. The exible HEN synthesis problem is
formulated as a multi-objective mixed-integer linear programming (MO-MILP). For handling the multiple conict design targets, a two-phase
fuzzy multi-criteria decision-making method is presented to attain a best compromised solution. Two numerical examples with exibility pref-
erences in source-stream temperatures are supplied, demonstrating that the proposed strategy can provide denite and feasible compensatory
solutions for multi-criteria HEN synthesis problems.
2004 Elsevier B.V. All rights reserved.
Keywords: Heat-exchanger network; Synthesis; Flexibility; Superstructure; MILP; Multi-criteria decision-making; Fuzzy optimization
1. Introduction
A heat exchanger network (HEN) synthesis problem can
be described as the one that synthesize a HEN congura-
tion to reach some assigned targets such as minimum util-
ity consumption, minimum total number of heat exchangers,
etc., with given heating/cooling utilities and hot/cold process
streams be cooled/heated from nominal inlet temperatures
to specied target temperatures [1].
Most of the existing HEN synthesis methods rely on ei-
ther heuristic rules (for example, pinch analysis method [2])
or mathematical programming (for example, simultaneous
optimization approach [36]). And further, to some typical
objectives considered in the HEN synthesis such as utility
consumption, total number of matches, and total exchanger
area, the exibility of the HENs for feasible operation un-
der possible variation of source-stream temperatures and/or
heat-capacity ow rates has been emphasized in some re-
cent articles [610]. For HEN synthesis, the analysis of this
exibility, dened as the size of the region of feasible oper-
ation in the space of desired or undesired deviations of pa-

Corresponding author. Tel.: +886 2 23636194;


fax: +886 2 23623040.
E-mail address: ccl@ntu.edu.tw (C.-L. Chen).
rameters from their nominal values [10], however, attracts
attention usually indirectly as test examples basing on math-
ematical programming synthesis. Therein, [10] explored the
HEN synthesis problem with simultaneous exibility target-
ing and minimum-utility objective based on an MILP for-
mulation. The optimal solution is examined on the basis of
the vertices of the polyhedral uncertainty region in the space
of source-stream temperatures. It is found, however, that the
resulting HEN structures with increasing exibility require-
ments are prone to variation, and the true maximum exi-
bilities of resulting HEN structures are usually greater than
the assigned targets. Thus, for a given exibility target, it is
very often to obtain a more conservative HEN design by the
method of simultaneous exibility targeting and synthesis
of minimum-utility HENs proposed in [10].
In this paper, we extend the work of [10] by simultane-
ously considering minimization of the total utility consump-
tion, maximization of operational exibility to source-stream
temperatures, and even minimum number of matches as
multiple design objectives. The exible HEN synthesis
problem is thus formulated as the one of multi-objective
mixed-integer linear programming (MO-MILP). This for-
mulation also assumes that the feasible region in the space
of uncertain input parameters is convex, so that the optimal
solution can thus be explored on the basis of the vertices
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.017
90 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100
of the polyhedral region of uncertainty [10]. Under the as-
sumption of convexity, only the source-stream temperatures
of the HENs are considered to be the uncertain input pa-
rameters. With this formulation, the standard denition of
the HEN synthesis problem with minimal total utility con-
sumption and even minimum number of units is extended
to include a exibility specication for the potential HEN
structure and can be stated as: given hot/cold streams to be
cooled/heated from nominal supply temperatures to speci-
ed target temperatures and hot/cold utility specications,
synthesize a HEN such that it has minimal utility consump-
tion (considering nominal case or average of all vertical
operating points), minimal number of matches if desired,
and maximal exibility for feasible operation.
For handling the multiple and conict design objectives,
a fuzzy decision-making method is adopted to attain the
compromised solution among all conict objectives [11].
Therein each design objective is treated as a fuzzy goal,
and a specic membership function is used to characterize
the transition from the numerical objective value to the de-
gree of satisfaction for the fuzzy objective. The nal deci-
sion, therefore, is interpreted as a fuzzy aggregation of these
multiple objectives and measured by the overall degree of
satisfaction. And the best compromised solution is nally
reached by maximizing the overall degree of satisfaction for
the decision. In the course of nding the solution, two pop-
ular operators, the minimum and the average, are applied as
the fuzzy intersection operators, the effects of which are ex-
amined as well. We also proposed an interactive two-phase
fuzzy decision-making method by combining these two op-
erators to take advantages of the both [11,12]. The minimum
operator is used in phase I to maximize the degree of satis-
faction for the worst objective, and the average operator is
Fig. 1. The two-stage superstructure.
then applied in phase II to simultaneously promote satisfac-
tion levels of all objectives with guaranteed least satisfaction
value.
Two numerical examples with exibility preference in
source-stream temperatures is presented here to demonstrate
that the proposed interactive two-phase fuzzy optimization
method can provide a feasible and better compensatory so-
lution for multi-objective HEN synthesis.
2. Model formulation
Consider the standard HEN synthesis problem with N
H
hot and N
C
cold process streams along with hot and cold
utilities. Since it is suitable for formulating the simultaneous
solution which involving the consideration of both utility
consumption and operational exibility, the HENsuperstruc-
ture proposed by [3,4] is applied for modeling the structure.
Therein, the isothermal mixing assumption in the simplied
superstructure eliminates the need for nonlinear/nonconvex
energy balance. The minimum number of superstructure
stages, N
T
, corresponds to max{N
H
, N
C
}, as suggested by
[3]. Fig. 1 illustrates a 2-hot/2-cold/2-stage superstructure.
The mathematical programming formulation for minimizing
utility consumption with specied uncertain source-stream
temperature ranges can be summarized as follows: [3,4,10]:
min
x
J
(0)
=

iHP
qcu
(0)
i
+

jCP
qhu
(0)
j
(1)
x
_
_
_
z
ijk
, zcu
i
, zhu
j
; t
ik
, t
jk
; dt
ijk
, dtcu
i
, dthu
j
;
q
ijk
, qcu
i
, qhu
j
;
i HP, j CP, k ST
_
_
_
(2)
C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100 91
=
_

_
x

(T
in
i
T
out
i
)FCp
i
=

kST

jCP
q
ijk
+qcu
i
(T
out
j
T
in
j
)FCp
j
=

kST

iHP
q
ijk
+qhu
j
_

_
overall heat balances
(t
ik
t
i,k+1
)FCp
i
=

jCP
q
ijk
(t
jk
t
j,k+1
)FCp
j
=

iHP
q
ijk
_

_
stage heat balances
t
i,1
= T
in
i
{T
in(0)
i
, T
in(0)
i
+}
t
j,N
T
+1
= T
in
j
{T
in(0)
j
, T
in(0)
j
+}
_
_
_
uncertain inlet temperatures
t
ik
t
i,k+1
t
jk
t
j,k+1
T
out
i
t
i,N
T
+1
T
out
j
t
j,1
_

_
feasibility of temperatures
(t
i,N
T
+1
T
out
i
)FCp
i
= qcu
i
(T
out
j
t
j,1
)FCp
j
= qhu
j
_
_
_
utility loads
q
ijk
z
ijk
0
qcu
i
zcu
i
0
qhu
j
zhu
j
0
_

_
logical constraints
dt
ijk
t
ik
t
jk
+(1 z
ijk
)
dt
ij,k+1
t
i,k+1
t
j,k+1
+(1 z
ijk
)
dtcu
i
t
i,N
T
+1
T
out
CU
+(1 zcu
i
)
dthu
j
T
out
HU
t
j,1
+(1 zhu
j
)
_

_
approach temperatures
dt
(0)
ijk
T
min
dtcu
(0)
i
T
min
dthu
(0)
j
T
min
_

_
nominal approach temp. bounds

iHP,jCP,kST
z
ijk
+

iHP
zcu
i
+

jCP
zhu
j
MEU
max
_
_
_
maximumexchanger units
z
ijk
, zcu
i
, zhu
j
{0, 1}
t
ik
, t
jk
, dt
ijk
, dtcu
i
, dthu
j
, q
ijk
, qcu
i
, qhu
j
; 0
i HP, j CP, k ST
_

_
(3)
where x and denote variables for design and the feasible
searching space, respectively; is the targeted exibility for
source-stream temperatures [10]; the superscript
(0)
denotes
the nominal condition; and the upper bound for unit num-
bers is MEU
max
. The searching space is comprised of all
heat balances constraints and relevant logical constraints. As
pointed out in [610], this problem is difcult to solve di-
rectly since it involves a max-min-max constraint that leads
to a non-differentiable global optimization problem. For the
HEN synthesis problem with uncertain source-stream tem-
peratures, the feasible region dened by the reduced inequal-
ity constraints is convex [6], so the critical point that limits
the solution lies at a vertex of the polyhedral region of un-
certainty. For the problem of Eq. (7) with N ( N
H
+N
C
)
uncertain source-stream temperatures, the vertex-based for-
mulation is given as follows [10]:
min
x
V

V
J
(0)
=

iHP
qcu
(0)
i
+

jCP
qhu
(0)
j
(4)
92 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100
or
min
x
V

V
J
(ave)
=
1
N
V
+1

n{0}VT
_
_

iHP
qcu
(n)
i
+

jCP
qhu
(n)
j
_
_
(5)
x
V
=
_

_
z
ijk
, zcu
i
, zhu
j
; dt
(n)
ijk
, dtcu
(n)
i
, dthu
(n)
j
;
t
(n)
ik
, t
(n)
jk
; q
(n)
ijk
, qcu
(n)
i
, qhu
(n)
j
;
i HP, j CP, k ST, n {0} VT
_

_
(6)

V
=
_

_
x
V

(T
in(n)
i
T
out
i
)FCp
i
=

kST

jCP
q
(n)
ijk
+qcu
(n)
i
(T
out
j
T
in(n)
j
)FCp
j
=

kST

iHP
q
(n)
ijk
+qhu
(n)
j
(t
(n)
ik
t
(n)
i,k+1
)FCp
i
=

jCP
q
(n)
ijk
(t
(n)
jk
t
(n)
j,k+1
)FCp
j
=

iHP
q
(n)
ijk
T
in(n)
i
= (T
in(0)
i
+r
(n)
i
) = t
(n)
i,1
T
in(n)
j
= (T
in(0)
j
+r
(n)
j
) = t
(n)
j,N
T
+1
t
(n)
ik
t
(n)
i,k+1
t
(n)
jk
t
(n)
j,k+1
T
out
i
t
(n)
i,N
T
+1
T
out
j
t
(n)
j,1
(t
(n)
i,N
T
+1
T
out
i
)FCp
i
= qcu
(n)
i
(T
out(n)
j
t
(n)
j,1
)FCp
j
= qhu
(n)
j
q
(n)
ijk
z
ijk
0
qcu
(n)
i
zcu
i
0
qhu
(n)
j
zhu
j
0
dt
(n)
ijk
t
(n)
ik
t
(n)
jk
+(1 z
ijk
)
dt
(n)
ij,k+1
t
(n)
i,k+1
t
(n)
j,k+1
+(1 z
ijk
)
dtcu
(n)
i
t
(n)
i,N
T
+1
T
out
CU
+(1 zcu
i
)
dthu
(n)
j
T
out
HU
t
(n)
j,1
+(1 zhu
j
)
dt
(0)
ijk
T
min
dtcu
(0)
i
T
min
dthu
(0)
j
T
min

iHP,jCP,kST
z
ijk
+

iHP
zcu
i
+

jCP
zhu
j
MEU
max
z
ijk
, zcu
i
, zhu
j
{0, 1}
t
(n)
ik
, t
(n)
jk
, dt
(n)
ijk
, dtcu
(n)
i
, dthu
(n)
j
, q
(n)
ijk
, qcu
(n)
i
, qhu
(n)
j
0
i HP, j CP, k ST, n {0} VT
_

_
(7)
Here, r
(n)
i
and r
(n)
j
are the vertex identiers which take val-
ues of N
V
= 2
(N
H
+N
C
)
combinations of +1 and 1, see
Table 1 in [10]. Therein, Eq. (4) considers nominal utilities
and Eq. (5) takes into account the average of all vertical op-
erating points.
For the benchmark example mentioned in [10], a
2-hot/2-cold streams problem along with heating steam and
cooling water, the maximal allowable variation of various
possible minimum-utility HEN structures is examined by
C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100 93
Table 1
Problem data of example 1
Process streams and utilities Heat-capacity ow rate FCp (kW/K) Input temperature T
in
(K) Output temperature T
out
(K)
Hot stream 1 (H1) 10 650 370
Hot stream 2 (H2) 20 590 370
Cold stream 1 (C1) 15 410 650
Cold stream 2 (C) 13 350 500
Hot utility (HU) 680 680
Cold utility (CU) 300 320
T
min
= 10 K.
Table 2
The resulting HEN structure of example 1 with increasing exibility target
( 0, see Table 2 of [10] for labeling HEN structures)
Specied value Resulting HEN structure Maximum exibility

0 36.95 A 36.95
36.95 + B1
a
70.63
40 B7
a
90.00
50 B6 90.00
60 B1
a
70.63
70 B7
a
90.00
80 B7
a
90.00
90 B7
a
90.00
90 + C2 120.0
120150 D 150.0
increasing the value over [0, 150] and solving Eq. (4)
or Eq. (5). There are 18 sets of structure-determining bi-
nary variables and 11 of them possess unique structures
(see Table 2 of [10]). However, we found that the resulting
HEN structures with increasing exibility targets are prone
to variation, and the true maximal exibilities of these HEN
structures are usually greater than the required targets, as
illustrated in Table 2. Similar results can also be found
should restricted heat-load constraints on vertices, as shown
in Eqs. (8) and (9), be taken into consideration in the HEN
synthesis.
min
x
V

load
J
(0)
=

iHP
qcu
(0)
i
+

jCP
qhu
(0)
j
(8)
or
min
x
V

load
J
(ave)
=
1
N
V
+1

n{0}VT

_
_

iHP
qcu
(n)
i
+

jCP
qhu
(n)
j
_
_
(9)

load
=
_

_
q
(n)
ijk
, qcu
(n)
i
, qhu
(n)
j

q
(0)
ijk
(1 ) q
(n)
ijk
q
(0)
ijk
(1 +)
qcu
(0)
i
(1 ) qcu
(n)
i
qcu
(0)
i
(1 +)
qhu
(0)
j
(1 ) qhu
(n)
j
qhu
(0)
j
(1 +)
i HP, j CP, k ST, n VT
_

_
(10)
In this paper, the targeted source-stream temperatures
are directly treated as individual design objective, and the
multi-criteria optimization approach is adopted for HEN
synthesis. The minimizing utility and the maximizing oper-
ational exibility can be simultaneously considered as two
conict objectives for synthesis of the network structure.
Furthermore, other targets such as minimizing number of
matches can also be considered, such as,
min
x
V

V
J
(0)
1
=

iHP
qcu
(0)
i
+

jCP
qhu
(0)
j
or
min
x
V

V
J
(ave)
1
=
1
N
V
+1

n{0}VT

_
_

iHP
qcu
(n)
i
+

jCP
qhu
(n)
j
_
_
and
max
x
V

V
J
2
= and
min
x
V

V
J
3
=

iHP

jCP

kST
z
ijk
+

iHP
zcu
i
+

jCP
zhu
j
(11)
In such a case, a unique HEN structure with satisfactory
levels in nominal or average utility consumption and opera-
tional exibility as well as unit numbers will be obtained. A
two-phase fuzzy optimization method is proposed to nd a
best compromised solution for the multi-criteria HEN syn-
thesis problem, as discussed in the next section. The basic
number of constraints and variables for the multi-objective
MILP formulation are summarized in the following.
1. The number of constraints:
(a) for linear equality constraints: (N
V
+ 1)[N
T
(N
H
+
N
C
) +4(N
H
+N
C
)];
94 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100
(b) for linear inequality constraints: (N
V
+ 1)[(N
H
+
N
C
)(N
T
+3) +3N
T
N
H
N
C
] +N
T
N
H
N
C
+N
H
+N
C
.
2. The number of variables:
(a) binary variables: N
H
N
C
N
T
+N
H
+N
C
;
(b) positive continuous variables: (N
V
+1)[N
T
(2N
H
N
C
+
N
H
+N
C
) +2(N
H
+N
C
)] +1.
3. Fuzzy multi-criteria optimization
Consider the multi-criteria optimization problem de-
ned in Eq. (11). Because of the fact that these objective
functions usually conict with each other in practice, the
optimization of one objective implies the sacrice of other
targets; it is thus impossible to attain their own optima, J
s
,
s S = [1, . . . , S], simultaneously. Therefore, the decision
maker (DM) must make some compromise among these
goals. In contrast to the optimality used in single objec-
tive optimization problems, Pareto optimality characterizes
the solutions in a multi-objective optimization problem
[13].
The weighting-sum method, among methods found in lit-
eratures for solving multi-objective optimization problems,
is the one that is used most often. Basing on the subjective
comprehension for each objective, the DM of this method
can weigh and sum up these objectives into a scalar form,
and then nd the solution by any existing single-objective
optimization method. However, as the situation of combin-
ing weighting factors becomes more complex, this method
becomes more tedious and the solution could be still in-
valid. Moreover, it is difcult for the DM to attribute a set
of incompatible objectives, such as utility consumption,
operational exibility, or number of matches in a heat ex-
changer network, without knowledge of the possible level
of attainment for these objectives. The physical mean-
ing of the nal scalar objective function is thus usually
vague.
In this work, we adopt the fuzzy set theory [14] to deal
with the multi-objective optimization problem. By consid-
ering the uncertain property of human thinking, it is quite
natural to assume that the DM has a fuzzy goal, J
s
, to
describe the objective J
s
with an interval [J
1
s
, J
0
s
]. For the
sth objective to be minimized, it is quite satised as the
objective value J
s
J
1
s
, and is unacceptable as J
s
J
0
s
.
The degree of satisfaction decreases as the objective value
increases from J
1
s
to J
0
s
. A strictly monotonic decreas-
ing membership function,
J
s
(J
s
(x
V
)) [0, 1], can be
used to characterize such a transition from the objective
value to the degree of satisfaction, therein value of 1 de-
notes absolutely satisfactory, and 0 means unacceptable.
Notably, an interval of [J
0
s
, J
1
s
] and a monotonic increas-
ing membership function should be used for dening a
fuzzy objective to be maximized. Without loss of gener-
ality, we will adopt linear membership functions in the
following.

J
s
(J
s
(x
V
))
=
_

_
1; for J
1
s
J
s
J
0
s
J
s
J
0
s
J
1
s
for J
1
s
J
s
J
0
s
0; for J
s
J
0
s
for J
s
to be minimized
(12)

J
s
(J
s
(x
V
))
=
_

_
1; for J
s
J
1
s
J
1
s
J
s
J
1
s
J
0
s
for J
0
s
J
s
J
1
s
0; for J
0
s
J
s
for J
s
to be maximized
(13)
The original multi-criteria optimization problem is now con-
verted to the one that looks for a suitable decision variable
vector that can provide the maximal degree-of-satisfaction
for the multiple fuzzy objectives.
max
x
V

V
(
J
1
(x
V
), . . . ,
J
S
(x
V
)) (14)
Under incompatible objective circumstances, a DM must
make a compromise decision that provides a maximal
degree-of-satisfaction for all these conict objectives. The
new optimization problem, Eq. (14), can be interpreted as
the synthetic notation of a conjunction statement (maximize
jointly all objectives). The result of this aggregation can be
viewed as a fuzzy intersection of all fuzzy goals, J
s
, s S,
and is still a fuzzy set, D.
D = J
1
. . . J
S
(15)
The nal degree-of-satisfaction resulting from certain vari-
able set,
D
(x
V
) can be determined by aggregating the
degree-of-satisfaction for all objectives,
J
s
(x
V
), s S, via
specic fuzzy intersection operator, T.

D
(x
V
) = T
_

J
1
(x
V
), . . . ,
J
S
(x
V
)
_
(16)
The fundamental properties for a fuzzy set and the related
operators can be found in [15]. As the ring level for each
policy is determined by the above procedure, the best so-
lution, x

V
, with the maximal ring level,
D
(x

V
), can be
selected.
max
x
V

D
(x
V
) = max
x
V

V
T(
J
1
(x
V
), . . . ,
J
S
(x
V
)) (17)
Using the fuzzy intersection operator, the original multi-
objective optimization problem, Eq. (11), is converted into a
single objective problem, Eq. (17). Several operators for im-
plementing fuzzy intersection can be selected for T, therein
two most popular ones are shown below.
T(
J
1
, . . . ,
J
S
)
=
_
min(
J
1
, . . . ,
J
S
) T = minimum
(
J
1
+. . . +
J
S
)
S
T = average
(18)
The average operator simultaneously takes all membership
values into account. But, no guarantee can be made for any
C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100 95
single objective. The minimum operator concentrates on im-
proving the worst scenario. However, the minimum operator
may result in multiple solutions since contribution of those
objectives with membership values greater than the mini-
mum one would not be cared. We thus combine these two
fuzzy intersection operators to take advantages of both and
propose a two-phase optimization procedure [11,12]. The
minimum operator is rst used in phase I to nd the least
degree of satisfaction for the worst objective, then the aver-
age operator is applied in phase II to promote satisfaction
levels of all objectives with guaranteed least membership
value. The merit of this method is that we can not only ob-
tain the unique optimal solution by using average operator
but also guarantee each objective to go after their own max-
imum on the basis of taking the least degree of satisfaction
as the lower-bound constraint. So, now we can summarize
the procedure of the two-phase fuzzy satisfying approach
for the multi-criteria optimization problem.
Step 1. For a function to be minimized, determine its ideal
solution and anti-ideal solutions by directly mini-
mizing and maximizing the objective function.
min
x
V

V
J
s
= J

1
s
(ideal solution ofJ
s
, totally acceptable
value) (19)
max
x
V

V
J
s
=

J
0
s
(anti ideal solution ofJ
s
,
unacceptable value) (20)
Notably, direct maximization and minimization
should be taken for a maximizing objective to
obtain the ideal and anti-ideal solutions.
Step 2. Based on the importance of different objective func-
tions and the acceptable ranges for objective val-
ues, subjectively select suitable lower/upper bounds,
J

1
s
J
1
s
J
0
s


J
0
s
for minimizing objective and
J

0
s
J
0
s
J
1
s


J
1
s
for maximizing objective. De-
ne membership functions for multiple fuzzy ob-
jectives as given in Eqs. (12) and (13).
Step 3. (Phase I). Use the minimum operator to nd the
maximal degree of satisfaction for the worst objec-
tive,
min
.
max
x
V

D
= max
x
V

V
min(
J
1
, . . . ,
J
S
) =
min
(21)
Step 4. (Phase II). Use the average operator to simultane-
ously promote satisfaction levels of all objectives
with new constraints of guaranteed minimal degree
of satisfaction.
max
x
V

+
V

D
= max
x
V

+
V

J
1
+ +
J
S
S
(22)
where

+
V
=
V
{
J
s

min
, s S} (23)
The new feasible space guarantees the least degree
of satisfaction for each fuzzy objective.
4. Numerical example
Two numerical examples adapted from [4,10] are supplied
to demonstrate the efcacy of proposed HEN synthesis strat-
egy. To solve this MO-MILP for the HEN model, GAMS
[16] and CPLEX are used as the modeling environment and
the MIP solver, respectively. The computing machine is a
personal computer with an Intel Pentium IV 2.26 GHz CPU.
Example 1. The 2-hot/2-cold streams example studied by
[10], with problem data presented in Table 1, is illustrated.
With these parameters, the multi-objective MILP formula-
tion has 408 linear equality constraints, 760 linear inequal-
ity constraints, 12 binary variables, and 545 positive con-
tinuous variables. Notably, the restriction of MEU
max
= 6
in Eq. (7) will be removed should the minimum number of
matches be simultaneously taken into account as one of the
design objectives.
According to the model formulation and the problem data,
we solve the multi-criteria MILP synthesis problem by us-
ing the fuzzy optimization procedure discussed in Section
3. The ideal and anti-ideal solutions, obtained by sequen-
tially minimizing and maximizing each objective function,
are shown in Table 3. Within these lower/upper limiting val-
ues, several preference intervals are subjectively selected to
establish linear membership functions for the fuzzy objec-
tives. In implementing the two-phase fuzzy optimization, we
rstly use the minimum operator to maximize the degree of
satisfaction for the worst target,
min
. The average operator
is then applied to optimize the aggregated objectives with
guaranteed level of satisfaction.
At rst, only two conict objectives are considered: the
minimal utility consumption and the maximal exibility to
all source-stream temperatures. And then the third objective,
the minimal number of matches, would be appended. Results
of two-phase fuzzy optimization with preference intervals of
[2550, 12750] or [2550, 8850] for utility, [0, 150], [40, 90]
or [40, 70] for exibility, and [4, 7] for unit numbers, along
with either or not considering restrictions on heat loads at
extreme operating points, are listed in Table 4. The resulting
HEN structures are also depicted in Fig. 2. Notably, the re-
duced range of exibility, [40, 90], implies that the required
minimum tolerance for temperature deviation is at least 40 K
and a tolerance of maximum temperature deviation for 90 K
Table 3
The ideal/anti-ideal solutions of various objectives of example 1
Objective function Ideal solution J

1
i
Anti-ideal solution

J
0
i
J
(0)
1
utility (kW) 2550 12750
J
2
exibility (K) 150 0
J
3
units 4 12
96 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100
Table 4
Results of HEN synthesis for example 1 using two-phase optimization when simultaneously considering minimal utility and maximal exibility (cases
IIV), and additional objective of minimal units (cases V and VI) with different preference intervals, and with or without considering restriction on heat
loads at vertices ( = = 0.6)
Case Preference intervals Heat load Phase Minimum utility Maximum exibility Minimum units
Hot Cold Total Satisfaction Flexibility Satisfaction Units Satisfaction
I [2550, 12750] No I 1300 2950 4250 0.833 120.0 0.800 6
(C2) [0, 150], [, 6] II 1300 2950 4250 0.833 120.0 0.800 6
II [2550, 8850] No I 1050 2700 3750 0.810 80.5 0.810 6
(B7) [40, 90], [, 6] II 1050 2700 3750 0.810 90.0 1.00 6
III [2550, 8850] Yes I 2190 3840 6030 0.448 62.4 0.448 6
[40, 90], [, 6] II 2190 3840 6030 0.448 62.4 0.448 6
IV [2550, 8850] Yes I 1893 3543 5436 0.542 56.3 0.542 6
[40, 70], [, 6] II 1893 3543 5436 0.542 56.3 0.542 6
V [2550, 8850] No I 1300 2950 4250 0.730 73.3 0.667 5 0.667
[40, 90], [4, 7] II 1300 2950 4250 0.730 74.2 0.685 5 0.667
VI [2550, 8850] Yes I 2550 4200 6750 0.333 56.7 0.333 6 0.333
[40, 90], [4, 7] II 2550 4200 6750 0.333 69.8 0.597 6 0.333
Fig. 2. The HEN structures for cases IVI of example 1.
C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100 97
Table 5
Problem data of example 2
Process streams
and utilities
Heat-capacity
ow rate FCp
(kW/K)
Input
temperature
T
in
(K)
Output
temperature
T
out
(K)
Hot stream 1 (H1) 6 500 320
Hot stream 2 (H2) 4 480 380
Hot stream 3 (H3) 6 460 360
Hot stream 4 (H4) 20 380 360
Hot stream 5 (H5) 12 380 320
Cold stream 1 (C1) 18 290 660
Hot utility (HU) 700 700
Cold utility (CU) 300 320
T
min
= 10 K.
Table 6
The ideal/anti-ideal solutions of various objectives of example 2
Objective function Ideal solution J

1
i
Anti-ideal solution

J
0
i
J
(0)
1
utility (kW) 3780 9860
J
2
exibility (K) 20 0
J
3
units 6 31
is absolutely satised. We choose 90 as the maximum de-
viation for most cases since it is the maximal tolerance for
the HEN with heat load restrictions on vertices.
As shown in Table 4 and Fig. 2, the HEN structure in
case I is the same as C2 of [10], therein the maximum ex-
ibility to temperatures is explicitly given as 120. In case II,
the B7 structure of [10] can be obtained when the tolera-
ble upper bound for utility consumption is decreased from
12750 to 8850, and the preference interval for exibility is
reduced from [0,150] to [40,90]. The resulting utility con-
sumption will be reduced from 4250 (Case I) to 3750 due
Table 7
Results of HEN synthesis for example 2 using two-phase optimization when simultaneously considering minimal utility, maximal exibility and minimal
units with different preference intervals, and with or without considering restriction on heat loads ( = = 0.6)
Case Preference intervals Heat load Phase Minimum utility Maximum exibility Minimum units
Hot Cold Total Satisfaction Flexibility Satisfaction Units Satisfaction
I [3780, 9860] No I 3863 403 4266 0.920 18.4 0.920 8 0.920
[0, 20], [6, 31] II 3793 333 4126 0.943 20.0 1.00 8 0.920
II [3780, 4930] No I 3712 252 3964 0.840 16.8 0.840 10 0.840
[0, 20], [6, 31] II 3660 200 3860 0.930 20.0 1.00 10 0.840
III [3780, 4930] No I 3811 351 4162 0.667 13.3 0.667 8 0.667
[0, 20], [6, 12] II 3793 333 4126 0.699 20.0 1.00 8 0.667
IV [3780, 4930] No I 3811 351 4162 0.667 16.7 0.667 8 0.667
[10, 20], [6, 12] II 3793 333 4126 0.699 20.0 1.00 8 0.667
V [3780, 9860] Yes I 3980 520 4500 0.882 10.6 0.882 8 0.920
[0, 12], [6, 31] II 3980 520 4500 0.882 12.0 1.00 8 0.920
VI [3780, 4930] Yes I 3742 282 4024 0.787 9.4 0.787 11 0.800
[0, 12], [6, 31] II 3742 282 4024 0.787 9.4 0.787 11 0.800
VII [3780, 4930] Yes I 3860 400 4260 0.583 6.99 0.583 8 0.667
[0, 12], [6, 12] II 3860 400 4260 0.583 10.0 0.833 8 0.667
VIII [3780, 4930] Yes I 3860 400 4260 0.583 9.5 0.583 8 0.667
[6, 12], [6, 12] II 3860 400 4260 0.583 10.0 0.667 8 0.667
to the restriction of preference intervals, and the exibility,
90, is equivalent to the maximum of the B7 structure.
All other conditions are equal to those in case II, case
III includes heat-load restrictions ( = = 0.6) on vari-
ous vertical operating points as additional constraints. The
prices of such additional restrictions are increased utility
consumption from 3750 to 6030 and decreased exibility to
temperature deviation from 90 to 62.4, which is still signi-
cantly greater than the minimum targeted value, 40. In case
IV, the preference interval for temperature deviation is fur-
ther reduced to [40, 70]. It is found that the resulting HEN
has smaller utility consumption, 5438, with the expense of
further reduction on exibility, 56.3, since our desideratum
for exibility has been made lower.
Cases V and VI take into account the unit number as
the third design objective, where heat-load restrictions on
vertices are either included or not. In case V, it is found
that the unit number is only ve, exibility to temperature
deviation is 74.2, and the utility consumption is 4250, a little
more than B4 and B5 of [10]. In case VI where heat-load
restrictions on vertices are further included, the total unit
number becomes six, the total utility is increased to 6750
with a reduced maximum temperature exibility of 69.8.
From these results, it is found that the proposed
multi-criteria synthesis strategy can attain a denite and
compromised solution for a problem with assorted conict
objectives. The preference intervals of various objectives
have signicant effects on nal HEN structures. Such ac-
ceptable and/or preference intervals can also reect the
importance of different objective functions. Should one
specic objective is emphasized, a tighter restriction or
shrinking span should be placed on its acceptable ranges.
98 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100
Fig. 3. The HEN structures for cases IVIII of example 2.
Observing the results shown in Table 4, we also discover
that the integrated two-phase method can bring the merits of
the minimum and average operators together. The minimum
operator maximize the degree of satisfaction for the worst
objective, and can result in a harmonious solution with
satisfaction levels for objective functions that are equal or
close to each other. The average operator, on the other hand,
can sometimes promote satisfaction levels for other objec-
tives with guaranteed minimum value, such as Cases II, V,
and VI. Owing to these advantages, the two-phase method
can thus provide the best compromised HEN conguration.
Example 2. This problem consists of ve hot streams and
one cold stream, (N
H
= 5, N
C
= 1), along with steam and
cooling water as utilities [3]. The problem data are listed
in Table 5. The number of superstructure stages is set as
N
T
= 5. With these parameters, the MO-MILP formulation
has 3510 linear equality constraints, 8026 linear inequality
constraints, 31 binary variables, and 5981 positive continu-
ous variables. The ideal and anti-ideal solutions are shown
in Table 6. Various preference intervals are sequentially se-
lected for dening the membership functions, as shown in
Table 7.
We directly use the ideal/anti-ideal solutions as the pref-
erence intervals in Case I. The acceptable utility range is
reduced from [3780,9860] to [3780,4930] in Case II. With
such a smaller allowance interval, the resulting utility con-
sumption will be reduced slightly from 3793 to 3660 at the
expense of unit numbers increased from 8 to 10. In Case III,
C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100 99
the maximum allowable unit number is changed from 31 to
12. The resulting HEN structure is slightly different to that
of Case I, but the required unit numbers, utilities, and exi-
bility are the same as Case I. In Case IV, when the minimum
exibility requirement is upgraded from 0 to 10, the same
HEN structure of Case I results with the same levels of util-
ity, exibility and unit numbers. Cases VVIII give similar
results with additional heat-load restrictions ( = = 0.6)
on various vertical operating points. With such additional
constraints, the HEN structures use similar levels of utilities
and unit numbers, but the exibility levels are dramatically
reduced. The resulting HEN structures are depicted in Fig. 3.
5. Conclusion
In this paper, we use the fuzzy multi-criteria optimization
approach to synthesize the heat-exchanger network where
some conict design objectives such as the total utility
consumption, the exibilities to source-stream temperature
variations, and even the total number of heat exchange units
can be considered simultaneously. Such a exible HEN
synthesis problem can be formulated as a multi-objective
mixed-integer linear programming (MO-MILP). For han-
dling the multiple conict design objectives, a two-phase
fuzzy optimization method is proposed to attain the best
compromised solution. The attractive features of the pro-
posed MO-MILP model are that it not only considers the
trade-off among the utility consumption, the source-stream
temperature exibility, and even the number of matches,
but also avoids the determination of structural boundaries,
as discussed in [10]. Two numerical examples with various
cases are studied, demonstrating that the proposed strat-
egy can provide a feasible compensatory solution for the
multi-criteria HEN synthesis problem.
Acknowledgements
This work is supported by the National Science Council
(ROC) under Contract NSC91-ET-7-002-004-ET. Partial -
nancial support of Ministry of Economic Affairs under grant
92-EC-17-A-09-S1-019 is also acknowledged.
Appendix A. Nomenclature
CP index set of cold process stream
dt
ijk
temperature approach for match i and j in stage k
dtcu
i
temperature approach for match i and cold utility
dthu
j
temperature approach for match j and hot utility
FCp heat capacity owrate
HP index set of hot process stream
J objective function
MEU
max
maximum number of heat-exchange units
N number of uncertain parameters
N
C
number of cold streams
N
H
number of hot streams
N
T
number of superstructure stages
N
V
number of vertices, = 2
N
q
ijk
heat exchanged between stream i and j in stage k
qcu
i
heat exchanged between stream i and cold utility
qhu
j
heat exchanged between stream i and hot utility
r directional identier for vertices
ST index set of superstructure stages
S number of objectives
T
min
minimum approach temperature
t
ik
temperature of stream i at hot end of stage k
t
jk
temperature of stream j at hot end of stage k
T temperature
VT index set of vertices
x, x
V
vector of variables
z
ijk
binary variable for existence of unit for match i
and j in stage k
zcu
i
binary variable for existence of unit for match i
and cold utility in stage k
zhu
i
binary variable for existence of unit for match j
and hot utility in stage k
Greek letters
, parameters used for restriction of heat-load
deviations
exibility index

exibility index (scalar)

J
s
membership function for J
s
J
s
a fuzzy goal
S index set of multiple objectives
D a fuzzy set

D
degree of satisfaction
upper bound for temperature difference
upper bound for heat exchange
the feasible searching region
Superscripts
in
inlet
(n)
identier for vertices
out
outlet
(0)
identier for nominal operating condition
Subscripts
CU
cold utility
HU
hot utility
i
index for hot process streams
j
index for cold process streams
k
index for superstructure stages
s
index for objectives
References
[1] K.C. Furman, N.V. Sahinidis, A critical review and annotated bib-
liography for heat exchanger network synthesis in the 20th century,
Ind. Eng. Chem. Res. 41 (2002) 23352370.
100 C.-L. Chen, P.-S. Hung / Chemical Engineering and Processing 44 (2005) 89100
[2] B. Linnhoff, E. Hindmarsh, The pinch design method for heat ex-
changer networks, Chem. Eng. Sci. 38 (5) (1983) 745763.
[3] T.F. Yee, I.E. Grossmann, Z. Kravanja, Simultaneous optimization
models for heat integration-I. Area and energy targeting and modeling
of multi-stream exchangers, Comp. Chem. Eng. 14 (10) (1990) 1151
1164.
[4] T.F. Yee, I.E. Grossmann, Simultaneous optimization models for heat
integration-II. Heat exchanger network synthesis, Comp. Chem. Eng.
14 (10) (1990) 11651184.
[5] C.A. Floudas, Nonlinear and Mixed-Integer Optimization: Funda-
mentals and Applications, Oxford University Press, New York, 1995.
[6] L.T. Biegler, I.E. Grossmann, A.W. Westerberg, Systematic Methods
of Chemical Process Design, Prentice Hall, Englewood Cliffs, NJ,
1997.
[7] K.P. Halemane, I.E. Grossmann, Optimal process design under un-
certainty, AIChE J. 29 (3) (1983) 425433.
[8] R.E. Swaney, I.E. Grossmann, An index for operational exibility
in chemical process design, AIChE J. 31 (4) (1985) 621630.
[9] I.E. Grossmann, C.A. Floudas, Active constraint strategy for exi-
bility analysis in chemical processes, Computers Chem. Eng. 11 (6)
(1987) 675693.
[10] A.E.S. Konukman, M.C. Camurdan, U. Akman, Simultaneous ex-
ibility targeting and synthesis of minimum-utility heat exchanger
networks with superstructure-based MILP formulation, Chem. Eng.
Processing 41 (2002) 501518.
[11] C.L. Chen, B.W. Wang, W.C. Lee, Multi-objective Optimization for
Multi-enterprise Supply Chain Networks, Ind. Eng. Chem. Res. 42
(2003) 18791889.
[12] R.J. Li, E.S. Lee, Fuzzy multiple objective programming and com-
promise programming with Pareto optimum, Fuzzy Sets and Systems
53 (1993) 275288.
[13] Sakawa, M., Fuzzy Sets and Interactive Multi-Objective Optimiza-
tion, Plenum Press, New York, 1993.
[14] L.A. Zadeh, Fuzzy sets, Information and Control 8 (3) (1965) 338
353.
[15] G.L. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logics-Theory and Ap-
plication, Prentice Hall, New York, 1995.
[16] A. Brooke, D. Kendrick, A. Meeraus, R. Raman, R.E. Rosen-
thal, GMAS: A Users Guide, GAMS Development Corporation,
1988.
Chemical Engineering and Processing 44 (2005) 101114
On-line dynamic optimization and control strategy for
improving the performance of batch reactors
A. Arpornwichanop
a
, P. Kittisupakorn
a
, I.M. Mujtaba
b,
a
Department of Chemical Engineering, Chulalongkorn University, Bangkok 10330, Thailand
b
School of Engineering, Design & Technology (EDT 3), University of Bradford, Bradford, West Yorkshire BD7 1DP, UK
Received 23 December 2002; received in revised form 25 August 2003; accepted 28 April 2004
Available online 2 July 2004
Abstract
Since batch reactors are generally applied to produce a wide variety of specialty products, there is a great deal of interest to enhance batch
operation to achieve high quality and purity product while minimizing the conversion of undesired by-product. The use of process optimization
in the control of batch reactors presents a useful tool for operating batch reactors efciently and optimally. In this work, we develop an approach,
based on an on-line dynamic optimization strategy, to modify optimal temperature set point prole for batch reactors. Two different optimization
problems concerning batch operation: maximization of product concentration and minimization of batch time, are formulated and solved using
a sequential optimization approach. An Extended Kalman Filter (EKF) is incorporated into the proposed approach in order to update current
states from their delayed measurement and to estimate unmeasurable state variables. A nonlinear model-based controller: generic model
control algorithm (GMC) is applied to drive the temperature of the batch reactor to follow the desired prole. A batch reactor with complex
exothermic reaction scheme is used to demonstrate the effectiveness of the proposed approach. The simulation results indicate that with the
proposed strategy, large improvement in batch reactor performance, in term of the amount of a desired product and batch operation time, can
be achieved compared to the method where the optimal temperature set point is pre-determined.
This paper gathers and integrates efcient well-known methodologies such as EKF, GMC and on-line optimization together, resulting in
an applicable, reliable and robust control technique for batch reactors.
2004 Elsevier B.V. All rights reserved.
1. Introduction
In many chemical industries, there is an increasing trend to
place a consideration on the production of high value prod-
ucts (e.g. polymers, pharmaceuticals, and specialty chemi-
cals) in batch processes. As an important main unit in such
processes, a batch reactor is generally involved in manufac-
turing of these products. The use of batch reactors offers
many advantages. Firstly, a batch reactor is quite exible, it
can adapt to small volume production of various products,
which are greatly submitted to the rapid changes in market
conditions and the advent of new technology. Secondly, a
batch reactor provides the natural way to scale-up processes
from laboratory experiment where complex chemical syn-
thesis is studied, to industrial manufacturing. Finally, it is

Corresponding author. Tel.: +44 1274 233645;


fax: +44 1274 235700.
E-mail address: i.m.mujtaba@bradford.ac.uk (I.M. Mujtaba).
also especially suitable to carry out reactions where materi-
als involved are dangerous and difcult to handle [1,2].
As batch reactors are used to produce a wide variety of ex-
pensive products, it is known that this process involves sev-
eral competing reactions which may cause undesired prod-
uct or waste. As a result, there is a great deal of interest
to enhance batch operation to achieve high quality and pu-
rity products while minimizing the conversion of undesired
by-products. Recently, the use of process optimization in the
control of batch reactors has received much attention in the
literature. This provides a useful tool for operating batch re-
actors efciently and optimally. For this purpose, it is desir-
able to optimize the process conditions during its operation
in order to meet the desired product and safety specica-
tions. A control system is an essential part to ensure that the
desired operating conditions can be maintained as close as
possible during the course of a batch run.
However, achieving such a proposed method for an op-
timal operation of batch reactors is quite difcult, and still
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.010
102 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
provides challenging and interesting problems. This is
mainly due to the inherent complexity of the batch reac-
tors which can be characterized by (i) highly nonlinear
behavior resulting from the dependence of reaction rates on
concentrations and temperature, (ii) time-varying system;
the process variables (e.g. concentration, temperature) and
parameters change with time, (iii) no steady state operating
condition, (iv) imperfect model; complex kinetic reactions
occurring within batch reactors are rarely well understood
that leads to an inaccuracy in developing the system model,
and (v) lack of measurement information; the product
qualities or key properties to be controlled (e.g. molecular
weight) cannot be measured until the end of batch run or
even if they can be measured sometime (e.g. concentration),
there is a signicant time delay. Only a few of physical
quantities such as temperature and pressure are available
for direct on-line measurement. That makes direct con-
trol of product properties difcult [2]. Although, in recent
years there have been signicant advances in developing
new sensors for measuring these product properties, they
have rarely been used in industrial processes due to high
operating cost and expensive investment on the measure-
ment devices. Thus, the usual practice is to control other
variables that can be measured rapidly in order to obtain
desired product properties instead.
In general, optimal batch reactor operation can be car-
ried out by two-step approach; rstly, determining an opti-
mal set point prole of key operating process variables such
as temperature [3]and secondly, tracking the desired prole
by a control system [4]. The optimal prole can usually be
determined off-line by solving an optimal control problem.
However, as mentioned above, because of the complexity
of chemical reaction schemes, modeling errors are always
present and in addition, process disturbances can occur dur-
ing the process operation. Due to the existence of this error
and disturbance, the nal product may signicantly differ
from the desired value, even though the pre-specied opti-
mal prole is tracked perfectly [5]. To realize this fact, it
is necessary to recalculate the optimal prole as an on-line
optimization strategy whenever new feedback information
is available. This strategy could compensate the modeling
error leading to process operation improvement.
In order to perform the on-line optimization strategy, the
knowledge of current state variables and/or parameters in
the process models is required. Due to the fact that some of
these variables cannot be known exactly or sometime can
be measured with time delay, it is essential to include an
on-line estimator to estimate these process variables using
available process measurements as well. The sequence of
an estimation and optimization procedure is known as an
estimation-optimization task [6]. As in several estimation
techniques, an Extended Kalman Filter (EKF) has become
increasingly popular because it is relatively easy to imple-
ment. It has been found that the EKF can be applied to a
number of chemical process applications with great success.
Once the estimate of unknown process variables is deter-
mined and then the model is updated, the optimization is
performed on-line to generate a new optimal input prole.
With the modied optimal prole, a designed controller is
used to control the system to follow this prole until the
new one is available.
Apart from specifying the optimal set point prole, a con-
trol system used to implement such a prole is another im-
portant issue to be considered. This is because the deviation
fromthe desired prole may cause an off-spec product. How-
ever, since it is well known that the control of batch reactors
is difcult due to the inherently nonlinear behavior, the use
of a linear control technique may give a poor performance.
For this reason, many advanced control techniques have been
developed and applied to the control of batch reactor. These
include, for example, nonlinear feedforward-feedback con-
trol [7], generic model control [8], adaptive control [1], glob-
ally linearizing control [9], dynamic matrix control [10],
linear model predictive control [11], or inverse model con-
trol [12]. Reviews on the progress in control techniques that
have been applied to the control of batch reactors as well as
their importance and performance are given by [13]. Among
these advanced control methods, a generic model control
(GMC) technique is one of the most studied control algo-
rithm. This is because nonlinear process models can be in-
terpreted straightaway in the GMC control algorithm so that
they do not need to be linearized. Furthermore, its implemen-
tation is relatively easy when compared to other model-based
control methods; consequently, the application of this con-
trol technique appears in many chemical processes.
In this work, an approach based on an on-line dynamic op-
timization strategy to modify optimal temperature set point
prole is developed for improving batch operation perfor-
mance. To demonstrate the effectiveness of the developed
approach, the batch reactor studied by [8], where two paral-
lel exothermic reactions occur, is chosen here as a case study.
For solving the on-line optimization problem, it needs the
knowledge of the current states of the system. Although most
physical quantities (e.g. temperature, owrate) can be mea-
sured frequently and are available as on line measurement,
some other properties (e.g. concentration) are measured in-
frequently with time delay. To overcome this difculty, the
Extended Kalman Filter (EKF) is incorporated into the pro-
posed strategy in order to estimate the concentrations from
their delayed measurements. The resulting optimal control
problem is solved by the sequential solution and optimiza-
tion method which is often referred to as the control vector
parameterization (CPV) method. Once the optimal tempera-
ture prole is modied, a controller based on generic model
control algorithm (GMC) is applied to control the batch re-
actor temperature following the desired prole. In the GMC
formulation, the EKF is also used to estimate the heat re-
leased from the chemical reactions. It should be noted that
the proposed strategy would be one of several strategy stud-
ied to promote the applicability of on-line dynamic opti-
mization with set point tracking for improving a batch pro-
cess; the aim of this work is to integrate efcient methods
A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 103
including the EKF and GMC and on-line optimization lead-
ing to an applicable, reliable and successful strategy for an
actual implementation.
2. Batch reactor
A reactor system considered by [8] which consists of a
batch reactor and jacket cooling system is chosen here as
a case study. The typical diagram of this system is shown
in Fig. 1. It is assumed that two parallel highly exothermic
reactions are carried out in the reactor
A + B
k
1
C
A + C
k
2
D
where A and B are raw materials, C and D are desirable
product and undesirable by-product, respectively. The rate
constant k
1
and k
2
are temperature dependence according to
the Arrhenius relation.
2.1. Reactor model
The batch reactor is modeled by the following equations:
Material balances in the reactor
dM
A
dt
= k
1
M
A
M
B
k
2
M
A
M
C
(1)
dM
B
dt
= k
1
M
A
M
B
(2)
dM
C
dt
= +k
1
M
A
M
B
k
2
M
A
M
C
(3)
dM
D
dt
= +k
2
M
A
M
C
(4)
Energy balances around the reactor
dT
r
dt
=
Q
r
+ Q
j
M
r
Cp
r
(5)
dT
j
dt
=
F
j

j
Cp
j
(T
jsp
T
j
) Q
j
V
j

j
Cp
j
(6)
Tr
Tj
TRC
Heat exchanger Batch reactor
Heating fluid
Cooling fluid
TJC
Fig. 1. Batch reactor system.
with
k
1
= exp

k
1
1

k
2
1
T
r
+ 273.15

k
2
= exp

k
1
2

k
2
2
T
r
+ 273.15

W = MW
A
M
A
+ MW
B
M
B
+ MW
C
M
C
+ MW
D
M
D
M
r
= M
A
+ M
B
+ M
C
+ M
D
Cp
r
=
(Cp
A
M
A
+ Cp
B
M
B
+ Cp
C
M
C
+ Cp
D
M
D
)
M
r
Q
r
= H
1
(k
1
M
A
M
B
) H
2
(k
2
M
A
M
C
)
Q
j
= UA(T
j
T
r
)
A =
2W
r
where M
i
is the amount of mole of component i, T
r
is the
reactor temperature, T
j
is the jacket temperature, and T
jsp
is
a set point value of the jacket temperature control system.
The meaning of other variables and parameters are explained
in the nomenclature.
The dynamic behavior of the reactor can be simulated
by solving Eqs. (1)(6). The differential-algebraic solver
DASSL [14] is used to give the solution of these equations.
The initial conditions for M
A
, M
B
, M
C
, M
D
used in all
simulation studies are 12, 12, 0, and 0 kmol, respectively.
The initial values of both reactor and jacket temperature are
set to 20

C. Other process parameter values used in the


reactor models are listed in Table 1.
In this work, it was assumed that all temperature can be
measured frequently without delay. The sampling time of
Table 1
Process parameter values
MW
A
= 30 kg/kmol
MW
B
= 100 kg/kmol
MW
C
= 130 kg/kmol
MW
D
= 160 kg/kmol
k
1
1
= 20.9057
k
2
1
= 10000
k
1
2
= 38.9057
k
2
2
= 17000
r = 0.5 m
F
j
= 0.348 m
3
/min
U = 40.842 kJ/(min m
2
C)
Cp
A
= 75.31 kJ/(kmol

C)
Cp
B
= 167.36 kJ/(kmol

C)
Cp
C
= 217.57 kJ/(kmol

C)
Cp
D
= 334.73 kJ/(kmol

C)
H
1
= 41840 kJ/kmol
H
2
= 25105 kJ/kmol
= 1000 kg/m
3

j
= 1000 kg/m
3
Cp
j
= 1.8828 kJ/(kg

C)
V
j
= 0.6912 m
3
104 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
0.2 min is used for temperature measurements. It is noted
that measurement noise is also included in the temperature
sensors having Gaussian noise with zero mean and 1

Cstan-
dard deviation. In addition, it was assumed that the concen-
tration (amount of mol) of reactants in the reactor is mea-
sured infrequently and has a sampling time and measure-
ment delay of 10 min.
3. On-line dynamic optimization strategy
The aim of a dynamic optimization problem is to deter-
mine a control prole minimizing (or maximizing) a given
objective function subject to process constraints. With the
optimal control policy, the controlled system is driven from
the initial state to a nal desired state in an optimal way.
However, in the presence of modeling error, the pre-specied
control prole may lose its optimal character [2]. For this
reason, an on-line optimization strategy through a receding
horizon scheme is employed in this work to compensate
such an error. The basic concept is to compute the optimal
control prole based on current feedback information. How-
ever, only the initial value of the optimal trajectory is sent
to the system as a set point for the controller. After new
information of states is available from either measurement
or estimation, the optimization is repeated again to generate
updated optimal set point prole at the next time interval.
The method proposed for improving the batch operation
can be divided into two phases: on-line modication of the
reactor temperature trajectory and on-line tracking of the de-
sired temperature trajectory. The rst phase involves deter-
mining an optimal temperature set point prole by solving
the on-line dynamic optimization problem and will be de-
scribed in this section. The other phase involves designing a
nonlinear model-based controller to track the obtained tem-
perature set point and will be presented in the next section.
Since both the on-line dynamic optimization and the
model-based control strategy rely on process models, the
knowledge of current states and/or model parameters is
required. However, in most industrial processes, state vari-
ables are not all measurable and some parameters are not
known exactly. As a consequence, there is a need for es-
timating these states and parameters. In this work, two
Extended Kalman Filters (EKF) are implemented. The rst
one is applied to predict the reactant concentration, which
will be used for on-line dynamic optimization, from its
delayed measurement. The other one is applied to esti-
mate the unknown heat of reaction, which will be used
for model-based controller, from the frequently available
measurements of temperature.
3.1. Problem statements
As a batch reactor is utilized for the production of a wide
variety of high value products, an optimization of batch op-
erating conditions, e.g. temperature, operating time, etc. is
the more efcient approach to obtain a maximum yield in a
minimum time or minimum cost and also to reach the spe-
cic nal conditions of the products in terms of quality and
quantity.
In general, an objective function in the optimization prob-
lem can be chosen, depending on the nature of the problem.
Here, two practical optimization problems related to batch
operation: maximization of product concentration in a xed
batch time and minimization of batch operation time given
amount of desired product, are considered to determine an
optimal reactor temperature prole. The rst problem for-
mulation is applied to a situation where we need to increase
the amount of desired product while batch operation time is
xed. This is due to the limitation of complete production
line in a sequential processing. However, in some circum-
stances, we need to reduce the duration of batch run to allow
the operation of more runs per day. This requirement leads
to the minimum time optimization problem. These problems
can be described in details as follows.
3.1.1. Maximum product concentration problem (P1)
In this type of problem, the objective is to compute the
optimal temperature policy maximizing the amount of a de-
sired product concentration for a given xed batch time sub-
ject to bounds on the reactor temperature. The problem can
be written mathematically as
max
T(t)
J = X(t
f
)
subject to
x = f(x(t), T, p, t) process model
x(t
0
) = x(0) initial conditions
T
L
T T
U
t
f
= t

f
where X is the amount of the desired product at a given nal
batch time, x is the vector of state variables, x is the derivative
of x with respect to time (t), T is the reactor temperature, p
are process parameters, t

f
is the xed batch time, and T
L
and
T
U
are lower and upper bounds of the reactor temperature.
3.1.2. Minimum batch time problem (P2)
The purpose of this optimization problem is to determine
the optimal temperature proles to achieve the desired -
nal product concentration in minimum batch time, thus the
performance index is the nal time whereas the desired pro-
duction concentration is dened as a terminal constraints.
The formulation of the minimum batch time problem can be
shown as
min
T(t)
J = t
f
subject to
x = f(x(t), T, p, t) process model
A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 105
x(t
0
) = x(0) initial conditions
T
L
T T
U
X(t
f
) = X

where X

is the desired product concentration at the end of


batch run and t
f
is nal batch time.
3.2. Solution of dynamic optimization problems
Computational techniques for the solution of a dynamic
optimization problem as formulated above have been an ac-
tive area of research. There are a number of different tech-
niques that have been proposed in the literature to solve this
class of problems. In general, they are mainly classied into
three classes.
The rst one is based on a classical variation method.
This approach is also known as an indirect method as it
focuses on obtaining the solution of the necessary conditions
rather than solving the optimization directly. Solution of
these conditions often results in a two-point boundary value
problem (TPBVP), which is accepted that it is difcult to
solve [15]. Although several numerical techniques have been
developed to address the solution of TPBVP, e.g. control
vector iteration (CVI) and single/multiple shooting method,
these methods are generally based on an iterative integration
of the state and adjoint equations and are usually inefcient
[16]. Another difculty relies on the fact that it requires an
analytical differentiation to derive the necessary conditions.
The second class of solutions is based on dynamic pro-
gramming. Unlike the variation method, this approach
applies the principle of optimality to formulate an op-
timization problem, leading to the development of the
HamiltonJacobiBellman partial equations that determine
the solution of the optimal control problem. However,
this approach is quite limited to a simple control prob-
lem [17] because of a difculty in obtaining the solution
of the optimality equations. [18] extended the idea of the
optimality principle to develop an alternative technique,
named as iterative dynamic programming (IDP). Although
the implementation of the IDP for solving many optimal
control problems can be found in the literature, it is known
that the IDP algorithm would be slower than most other
gradient-based algorithms [19].
The last one is based on discretization techniques, re-
ceived major attention and considered as an efcient solu-
tion method. The concept of this approach is to transform the
original optimal control problem into a nite dimensional
optimization problem, typically a nonlinear programming
problem (NLP). Then, the optimal control solution is given
by applying a standard NLP solver to directly solve the op-
timization problem. For this reason, the method is known as
a direct method. The transformation of the problem can be
made by using discretization technique on either only con-
trol variables (partial discretization) or both state and con-
trol variables (complete discretization). Based on this con-
sideration, this approach can be divided into two categories:
sequential and simultaneous strategy.
In the sequential strategy, a control (manipulated) variable
prole is discretized over a time interval. The discretized
control prole can be represented as a piecewise constant,
a piecewise linear, or a piecewise polynomial function. The
parameters in such functions and the length of time subin-
terval become decision variables in optimization problem.
This strategy is also referred to a control vector parameter-
ization (CVP).
One advantage in the sequential approach is that only the
parameters that are used to discretize the control variable
prole are considered as the decision variables. The opti-
mization formulated by this approach is a small scale NLP
that makes it attractive to apply for solving the optimal con-
trol with large dimensional systems that are modeled by a
large number of differential equations. In addition, this ap-
proach can take the advantage of available IVP solvers. How-
ever, the limitation of the sequential method is a difculty
to handle a constraint on state variables (path constraint).
This is because the state variables are not directly included
in NLP.
In contrast to the sequential solution method, the simul-
taneous strategy solves the dynamic process model and the
optimization problem at one step. This avoids solving the
model equations at each iteration in the optimization algo-
rithm as in the sequential approach. In this approach, the dy-
namic process model constraints in the optimal control prob-
lem are transformed to a set of algebraic equations which
is treated as equality constraints in NLP problem [20]. To
apply the simultaneous strategy, both state and control vari-
able proles are discretized by approximating functions and
treated as the decision variables in optimization algorithms.
The main advantage of the simultaneous approach is a ca-
pability in handing constraints on the state variables. This
is because these constraints can be dealt with by including
them directly in the NLP as additional constraints. However,
due to the discretization on both state and control variables,
this leads the simultaneous approach to a large scale opti-
mization problem consisting of a large set of algebraic con-
straints and decision variables and needs a special solution
strategy.
As the ability of the sequential approach to handle large
systems without the need to solve excessively large opti-
mization problem, this approach is utilized in this study to
solve the optimal control problem. The formulation of the
optimal control problem as a nonlinear programming is de-
scribed below.
3.2.1. Sequential approach
Consider a dynamic process model as in the form of an
implicit function
f(t, x(t), x(t), u(t), p) = 0 [t
0
, t
F
] (7)
where x(t) is the set of all state variables, x denotes the
derivative of x(t) with respect to time, u(t) is a vector of con-
106 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
trol variables, and p is a vector of time invariant parameters.
The time interval of interest is [t
0
, t
F
] and the function f is
assumed to be continuously differentiable with respect to all
its arguments. For given initial condition x(t
0
), the optimal
control of a dynamic process represented by Eq. (7) can be
determined by computing u(t) maximizing (or minimizing)
an objective function J of the form
J = F( x(t
F
); x(t
F
); u(t
F
); t
F
) (8)
subject to bounds on u(t) and terminal constraints. To trans-
form such a dynamic optimization problem into a nonlin-
ear programming problem via the sequential approach, the
control u(t) is approximated by a nite dimensional repre-
sentation. The time interval [t
0
, t
F
] is divided into a nite
number of subintervals (P). In each subinterval, the control
u(t) is represented by a set of basis functions involving a
nite number of parameters
u(t) =
j
(t, z
j
), t [t
j1
, t
j
] with j = 1, 2, . . . , J
(9)
where t
J
= t
F
. The control prole is dened by the param-
eters z
j
and switching times t
j
. In this study, the piecewise
constant control is assumed and used because the form of
the solution is ideally suited for implementation on a digital
computer. Thus the set of decision variables for the nonlin-
ear program can be written as
y = {z
1
, z
2
, . . . , z
J
, t
1
, t
2
, . . . , t
J
} (10)
The resulting nonlinear programming problemof the form
min(or max
y
) J(y) (11)
subject to equality constraints (process model), inequality
constraints, bounds on control, etc. is solved using an effec-
tive decomposition algorithm named the successive reduced
quadratic programming developed by [21]. Computational
procedure of the sequential approach for the optimization
problem (Eq. (11)) is illustrated in Fig. 2. With the initial
guess of the decision variables (y), an integrator based on
Model Solver
Evaluate
Objective function
Constraints
Gradient
NLP
Check
tolerance
Set initial conditions
Guess initial control parameters
Optimal control
parameters
no
yes
Fig. 2. Optimal control solution via the sequential approach.
Gears type method is used for solving the process model
which provides the value of the objective function and con-
straints. Gradient information of the objective function and
constraints with respect to the decision variables is eval-
uated in an efcient way using adjoint variable approach.
Then, a NLP solver determines a new set of control param-
eters and sends it back to the model solver. This procedure
is repeated until the optimal value is found satisfying a
specied accuracy.
3.3. State and parameter estimation
The implementation of the on-line optimization strategy
requires the knowledge of current states and/or parameters in
nonlinear process models in order to modify a new optimal
prole dened as a set point for a controller. It is known
that some measurements i.e. concentration are available at
low sampling rate with signicant time delay. To overcome
this difculty, state and parameter estimation is incorporated
into the proposed on-line optimization algorithm.
In this work, an Extended Kalman Filter (EKF), an ex-
tension of the Kalman Filter, is designed to reconstruct the
current state variables from the delayed state measurement.
The advantage of the EKF is that it requires information only
from the previous sampling time and allows prior knowledge
of a system via process models to be used for the estimation.
The algorithm of the EKF can be seen in Appendix A.
3.3.1. Application to the batch reactor
Since the concentrations of reactants (M
A
, M
B
, M
C
and
M
D
) in batch reactor are assumed to be measurable with a
delay of one sampling time; that is, at time k of the reactor,
only information at time k 1 is available. Thus, the EKF
is applied to estimate the value of reactant concentration at
current time k from their delayed measurements at sampling
time k 1. However, since it is expected to exhibit uncer-
tainty in reaction rate constant (i.e. k
1
2
and k
2
2
) in real plant,
the EKF is also used to estimate these uncertain parame-
ters. The following equations, therefore, are appended for
parameter estimation
dk
1
2
dt
= 0 (12)
dk
2
2
dt
= 0 (13)
Eqs. (1)(4), (12) and (13) correspond to Eq. (A.1) in
the EKF algorithm. Based on the estimate of the current
information, the dynamic optimization problem is resolved
to generate a new optimal temperature trajectory.
4. Generic model control (GMC)
This section presents the design of a controller to control
the reactor temperature following a desired temperature tra-
A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 107
jectory. It is accepted that the use of linear control techniques
in highly nonlinear and time variant chemical processes (e.g.
batch reactors) is quite limited to their performances and
may give a poor control response. Therefore, in this work,
a nonlinear control technique based on generic model con-
trol (GMC) algorithm is utilized. This model-based control
methodology has been received much interest during the last
decade and a number of applications of GMC to the con-
trol of batch processes have been reported in the literature
(e.g. [4,8,22,23], etc.). However, most of these works focus
on using the GMC to track the pre-determined optimal pro-
le (off-line calculation) of a reactor temperature. No effort
has been made to apply the GMC to implement an on-line
optimal set point prole. Therefore, in this work, the perfor-
mance of the GMC controller with optimal temperature set
points determined by on-line optimization strategy is also
evaluated and compared to that of the GMC controller with
pre-determined set point.
4.1. Control algorithm
Let us consider a process based on the following model
equations:
dx
dt
= f(x, p, t) + g(x, t)u (14)
y = h(x) (15)
where x is a vector of state variables, y is a vector of output
variables, u is a vector of input variables, p is vector of
process parameters, and f, g, and h are generally nonlinear
function vectors. The general form of GMC algorithm can
be written as
dy
dt
= K
1
(y
sp
y) + K
2

t
f
0
(y
sp
y) dt (16)
The GMC control response can be designed via the tuning
parameters K
1
and K
2
based on the tuning curve given by
[24]. It should be noted that the GMC approach is a special
case of the global input output linearizing control technique
in which a transformed control action is chosen properly
with the external PI controller. The use of Eq. (16) forces y
toward its set point, y
sp
, with zero offset. If Eq. (15) is dif-
ferentiated, and the Eq. (16) is substituted into the resulting
equation, the GMC control law is
u =
[K
1
(y
sp
y) + K
2

(y
sp
y) dt (dh/dx)f(x, d, t)]
(dh/dx)g(x, t)
(17)
4.2. Application of GMC controller to the batch reactor
To implement the GMC, an energy balance around the
reactor is required; it gives the relation between the reactor
temperature (controlled variable) and the jacket temperature
(manipulated variable). Based on the assumption that the
amount of the heat accumulated in the walls of the reactor
is negligible compared to the heat transferred in the reactor,
the energy balance equation becomes
dT
r
dt
=
Q
r
+ U
r
A
r
(T
j
T
r
)
W
r
Cp
r
(18)
where U
r
is the heat transfer coefcient, A
r
is the heat trans-
fer area, W
r
is the mass of the reactor contents, Cp
r
is the
mass heat capacity of the reactor content, and Q
r
is the heat
released by the reactions.
Rearranging the Eq. (18) as in the form of GMC algo-
rithm, the following functions, f, g, and h can be dened
f(x, p, t) =
Q
r
U
r
A
r
T
r
W
r
Cp
r
(19)
g(x, t) =
U
r
A
r
W
r
Cp
r
(20)
h(x) = T
r
(21)
Replacing these equations in Eq. (17), we have
T
j
=T
r
+
W
r
Cp
r
U
r
A
r

K
1
(T
rsp
T
r
)
+K
2

t
0
(T
rsp
T
r
) dt

Q
r
U
r
A
r
(22)
In order to make the GMC control law available for an
on-line control implementation, the integral term in Eq. (22)
is approximated by numerical integration. This leads to the
discrete-time form of the GMC algorithm as given in the
following equation.
T
j
(k) =T
r
(k) +
W
r
Cp
r
U
r
A
r

K
1
(T
rsp
T
r
(k))
+K
2
k

0
(T
rsp
T
r
(k)) t

Q
r
(k)
U
r
A
r
(23)
where t is the sampling time chosen to be equal to the
frequency of the temperature measurement. It is sufciently
small to approximate the continuous-time form without re-
quiring a large number of sampled values.
However, Eq. (23) gives the actual jacket temperature
(T
j
(k)) required at the next sampling time to control the re-
actor temperature (T
r
(k)) at the desired trajectory (T
rsp
). As
in usual practice, the reactor temperature control is cascaded
with the jacket temperature control (heating and cooling sys-
tem); the output of reactor temperature controller (master
loop) is the set point of the jacket temperature controller
(slave loop), as demonstrated in Fig. 3. In addition, as the
model of the heat exchanger system for heating and cooling
is not included in Eq. (18), if T
j
(k) is applied directly as the
set point for jacket temperature control system without con-
sidering on its dynamic, the resulting control response would
be sluggish. To accommodate such an effect, it is reason-
able to assume that the dynamic of the jacket control system
108 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
REACTOR
Dynamic
compensation
GMC
EKF
with simplified
reactor model
Tr(k) Tj(k)
Mi (K-1)
EKF
On-line dynamic
optimization problem
Updated Mi (K)
Initial
setpoint
New reactor
temperature setpoint
Trsp (K)
Tj Tjsp
Qr(k)
Fig. 3. The proposed strategy for on-line update and control of reactor temperature prole.
can be approximated by a rst order model with time con-
stant (
j
) [25]. Consequently, the T
jsp
(k) can be computed
by
T
jsp
(k) = T
j
(k 1) +
j

T
j
(k) T
j
(k 1)
t

(24)
With this jacket temperature set point, a jacket temper-
ature controller (setting as a PI controller) through a heat
exchanger system opens or closes control valve reecting to
the owrate of heating and cooling uid. However, in real-
ity the ability of the heat exchanger in adjusting the jacket
temperature is always limited, thus, in this work the jacket
temperature is bounded between 0

C and 120

C. The tun-
ing parameters of GMC are given in Table 2.
As shown in Eq. (23), the heat released (Q
r
(k)), which
cannot be measured, is needed in the GMC algorithm. Here,
the EKF algorithm, as used in on-line optimization strategy,
coupled with the simplied reactor model, given by [26],
is also applied to estimate the heat released (Q
r
(k)). The
reason of using the simplied model, not the exact model
of the plant, is because if the exact model were used, too
many uncertain/unknown parameters as well as too many
unmeasurable states would be involved. That may lead to
poor performance of the EKF. Hence, the simplied model
with less uncertain/unknown parameters and unmeasurable
Table 2
Parameters in GMC algorithm
W
r
= 1560 kg
Cp
r
= 1.8828 kJ/(kg

C)
U
r
= 40.842 kJ/(min m
2
C)
A
r
= 6.24 m
2

j
= 2 min
K
1
= 2.4
K
2
= 10
4
states as shown below is used instead.
dT
r
dt
=
Q
r
+ U
r
A
r
(T
j
T
r
)
W
r
Cp
r
(25)
dT
j
dt
=
F
j

j
Cp
j
(T
jsp
T
j
) U
r
A
r
(T
j
T
r
)
V
j

j
Cp
j
(26)
dN
dt
= bNT
r
(27)
dQ
r
dt
= N
dT
r
dt
+ T
r
dN
dt
(28)
db
dt
= 0 (29)
where N = bM
r
(H), b is a pseudo reaction rate constant,
M
r
is the total reactant concentration, and His heat of reac-
tion. It should be noted that the variable N representing two
unknown parameters (M
r
and H) can be estimated instead
of these parameters so that the number of state equations for
estimation decrease and an error of estimation correspond-
ing to the uncertainty of each parameters can be reduced.
Eqs. (25)(29) correspond to Eq. (A.1) in the EKF algo-
rithm. Once the reactor and jacket temperature measurement
are available, the EKF with simplied model estimates the
heat release of reactions (Q
r
(k)). Table 3 summarizes the
initial conditions and tuning parameters of the EKF used in
this simulation work.
5. Simulation results
5.1. Maximum conversion problem (P1)
All simulation results given here are based on the opti-
mization problem P1. The objective in the problem formu-
lation is to nd the optimal reactor temperature prole, such
A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 109
Table 3
Parameters and initial conditions in EKF
Estimate M
i
, k
1
2
and k
2
2
M
A
(0) = 12 kmol P = diag[100 10 100 100 100 2000]
M
B
(0) = 12 kmol Q = diag[100 1 100 1 500 5000]
M
C
(0) = 0 kmol R = diag[10 10 10 10]
M
D
(0) = 0 kmol
k
1
2
(0) = 38.9057
k
2
2
(0) = 17000
Estimate Q
r
T
r
(0) = 20

C P = diag[1 1 100 20 10]


T
j
(0) = 20

C Q = diag[10 10 2500 100 100]


N(0) = 1.8462 R = diag[10 10]
Q
r
(0) = 0
b(0) = 1.8386 10
6
that the amount of mole of product C is maximized in a xed
batch time with respect to a constraint on the temperature
due to the reason of safe operation. In this case study, the
specied nal batch time (t
f
) of 200 min is used and the reac-
tor temperature is bounded according to 20 T (

C) 120.
5.1.1. Temperature set point prole determined off-line
with perfect tracking
First, simulation studies have investigated the case where
the optimal temperature prole is determined by off-line
computation and perfect tracking of such a prole is as-
sumed. This results in the maximum amount of product C
(maximum conversion) that can be achieved at the end of
batch run and is served as a reference to be compared with
results obtained from the proposed strategy. The optimal
control problems have been solved using time interval with
equal length varied from one to 40 intervals to discretize the
prole. The switching time is xed and the length of each
interval is specied by dividing the batch operation time (t
f
)
by a number of time intervals (P). Thus, the problem is to
seek an optimal temperature value (decision variables) in
each subinterval.
Simulation results with different time interval (P) are re-
ported in Table 4. Optimal control policy in reactor tem-
perature for each case is shown in Fig. 4. As shown in
Table 4, when one time interval (P = 1) is used, the amount
of product C obtained at the nal time (t
f
= 200 min) is
7.0171 kmol and the optimal temperature (isothermal opera-
tion) set point is 88.01

C whereas using P = 20, the amount


of product C achieved is 7.0379 kmol. It was found from
Table 4
Summary of the results: off-line optimization and perfect tracking
Time
interval
Product
C (kmol)
By-product
D (kmol)
CPU
time

(s)
1 7.0171 1.3464 0.1591
5 7.0281 1.3605 1.2600
10 7.0339 1.3594 1.9778
20 7.0379 1.3585 2.8066
40 7.0402 1.3579 5.0519

Run on Pentium III/850 computer.


0 20 40 60 80 100 120 140 160 180 200
86
88
90
92
94
96
98
Time (min)
1
5
4
3
2
T
e
m
p
e
r
a
t
u
r
e

(
C
)
Fig. 4. Optimal temperature prole: 1 interval (1), 5 intervals (2), 10
intervals (3), 20 intervals (4), 40 intervals (5)
these results that the amount of the desired product C in-
creases as a number of interval increases. This is due to that
as the number of intervals enlarges (more degrees of free-
dom in the optimization), the approximated optimal prole
with piecewise constant policy is closer to the actual optimal
prole.
5.1.2. Temperature set point prole determined on-line
with GMC controller
Next, the proposed strategy using an on-line dynamic op-
timization to update optimal temperature set point prole is
implemented. Rather than assuming the reactor temperature
trajectory can be tracked perfectly as in previous studies, the
GMC controller is applied here to drive the system to follow
the desired trajectory. Considering the time elapsed in the
determination of the optimal control problem, with P = 20,
the temperature set point prole is updated every 10 min.
This means that, to apply this strategy on-line, the computa-
tional time for the updated temperature set point prole must
be less than 10 min. Here, the computational time based on
Pentium III/850 mHz is approximately 3 s. As a result, this
strategy is applicable for on-line implementation. Regard-
ing to the GMC control performance, it was found that the
GMC controller can drive the system from current set point
value to a new one and maintain it at this set point. There-
fore, the GMC controller can be used for tracking the prole
obtained from the proposed strategy.
Simulation studies carried out to compare the amount
of desired product C obtained from on-line dynamic opti-
mization strategy with that from off-line strategy, are cases
where the perfect model (all parameters correctly specied)
is used (nominal case), and where plant/model mismatch
is introduced by changing parameters in actual plant i.e.
pre-exponential rate constant (k
0
) decreased by 50% and ac-
tivation energy (Ea) increased by 20% from their nominal
values, as shown in Table 5.
110 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
Table 5
Comparisons of the results obtained from off-line and on-line optimization strategy with GMC controller (problem P1)
Case studies Product C (kmol)
Off-line On-line
1. Nominal case
All parameters specied correctly 6.9478 6.9584
2. Plants/model mismatch case
50% k
0
of reaction 2 in plant model (k
1
2
= 38.2125) 7.6751 7.8851
+20% Ea of reaction 2 in plant model (k
2
2
= 20400) 8.5825 10.2131
50% k
0
and +20% Ea of reaction 2 in plant model 8.5827 10.2137
Note: k
0
= exp(k
1
2
) and Ea = (k
2
2
)(R) where k
0
is pre-exponential rate constant, Ea = activation energy, and R is ideal gas constant.
In the nominal case, the product obtained from the off-line
strategy (C = 6.9478) is close to that obtained by the on-line
strategy (C = 6.9584). Fig. 5(a) and (b) shows the response
of GMC controller to track the reactor temperature trajectory
that is pre-specied by off-line calculation, and the compar-
Fig. 5. (a) Control response (off-line, nominal case): T
rsp
(dash), T
r
(solid), T
j
(dot). (b) Heat released (off-line, nominal case): actual (solid),
estimated (dot).
ison of the actual and estimated heat released by reactions,
respectively. It can be seen that the EKF provides an excel-
lent estimation of the heat released. With this heat released,
the GMC controller gives very good temperature control.
Similarly, in the case that optimal temperature is modied
via the on-line optimization strategy based on the current in-
formation of M
i
, k
1
2
and k
2
2
obtained from delayed measure-
ment of M
i
, the GMC controller is able to successfully track
the reactor temperature (Fig. 6(b)). The performance of the
EKF to predict the amount of M
A
, M
B
, M
C
, and M
D
at cur-
rent time from their measurement with time delay of 10 min
is illustrated in Fig. 6(a). Also, the EKF gives good estima-
tions of k
1
2
and k
2
2
as shown in Fig. 6(c). It is interesting to
note that since the initial reactor temperature starts at 20

C
which is below the optimal temperature set point, the GMC
controller attempts to drive the reactor temperature from this
condition to the specied set point. However, due to the dy-
namics of the reactor temperature, the reactor temperature
differs from the optimal set point during the rst 20 min,
causing that the amount of the product C obtained from the
on-line strategy with the GMC controller (C = 6.9584) is
less than that obtained from the off-line strategy with perfect
tracking (see Table 4 for P = 20, C = 7.0379)
For the mismatch in k
1
2
, the value of the desired prod-
uct C = 7.8851 can be achieved at the end of batch for the
on-line optimization strategy which is higher than that ob-
tained from off-line strategy where the mismatch is not no-
ticed (C = 7.6751). Similar results can be observed under
the case of plant model mismatch in k
2
2
as shown in Table 5.
These results indicate clearly that the performance of batch
reactor operation is improved via the proposed strategy. Due
to similarity in their control responses, only the result for
change in k
1
2
is shown in Fig. 7.
Finally, with a change in both k
1
2
(50% k
0
) and k
2
2
(+20% Ea) in plant model, the results using the on-line opti-
mization strategy show that the GMC controller is able to ac-
commodate this change very well as can be seen in Fig. 8(a).
Fig. 8(b) presents the performance of the EKF for estima-
tion of k
1
2
and k
2
2
. Since the EKF estimates these parameters
close to the true values, the mismatch is eliminated. That
leads to high product C obtained at the nal batch time (C
= 10.2137) compared to the value of C = 8.5827 obtained
from the off-line optimization strategy.
A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 111
Fig. 6. (a) Product prole (on-line, nominal case): actual (solid), estimated (). (b) Control response (on-line, nominal case): T
rsp
(dash), T
r
(solid), T
j
(dot). (c) Estimate of k
1
2
and k
2
2
(on-line, nominal case): actual (solid), estimated ().
Fig. 7. Control response (on-line, mismatch in k
1
2
): T
rsp
(dash), T
r
(solid),
T
j
(dot).
5.2. Minimum time problem (P2)
The results presented here correspond to the case where
the objective is to minimize the batch time of operation
subject to the terminal constraints on the desired amount
of mole of product C (M
C
(t
f
) = 6.00 kmol). The reactor
temperature constraints are same as in problem P1.
Several simulations have been carried out under process
parameter uncertainties e.g. in pre-exponential rate constant
(k
0
) and activation energy (Ea). In all case studies we consid-
ered 10 time intervals when reactor temperature and switch-
ing time are optimized while minimizing the nal batch op-
eration time. Results, reported in the value of minimumbatch
time to obtain the desired product C and the amount of the
desired product C at the end of batch operation, from on-line
dynamic optimization strategy are also compared with those
from the off-line strategy.
With an 20% increase of parameter k
1
2
in plant model, it
can be seen from Table 6 that the nal batch time needed
to achieve the desired product C from the proposed on-line
112 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
Fig. 8. (a) Control response (on-line, mismatch in k
1
2
and k
2
2
): T
rsp
(dash),
T
r
(solid), T
j
(dot). (b) Estimate of k
1
2
and k
2
2
(on-line, mismatch in k
1
2
and k
2
2
): actual (solid), estimated ().
modication of temperature set point prole (t
f
= 49.4 min)
is shorter compared to the result with the off-line strategy (t
f
= 64.8 min). This is because the EKF can acknowledge this
parameter uncertainty, so that the temperature set point pro-
le is updated corresponding to a modied parameter value
close to the actual value. Fig. 9 shows, for this mismatch,
the control response of GMC controller to deliver the reac-
tor temperature from initial condition to the temperature set
point determined off-line. For the on-line temperature set
Table 6
Comparisons of the results obtained from off-line and on-line optimization strategy with GMC controller (problem P2)
Case studies Final time (min)/product C (kmol)
Off-line On-line
(1) 50% k
0
of reaction 2 in plant model (k
1
2
= 38.2125) 64.8/6.0049 49.4/6.0089
(1) +20% Ea of reaction 2 in plant model (k
2
2
= 20400) 56.4/6.0147 44.6/6.0104
(1) 50% k
0
and + 20% Ea of reaction 2 in plant model 56.7/6.0144 45.2/6.0121
Fig. 9. Control response (off-line, mismatch in k
1
2
, problem P2): T
rsp
(dash), T
r
(solid), T
j
(dot).
Fig. 10. Control response (on-line, mismatch in k
1
2
, problem P2): T
rsp
(dash), T
r
(solid), T
j
(dot).
point modication, as expected, GMC controller can also
control the reactor corresponding to the changes in temper-
ature set point as can be seen in Fig. 10
The results for the remaining case studies are summa-
rized in Table 6. An important aspect obtained from these
results is that in all cases, the minimum batch time to obtain
the desired product concentration in the on-line set point
A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114 113
modication strategy decreases. This points out the effec-
tiveness of the proposed method to improve the operation
of batch reactor.
6. Conclusions
In this work, the method using an on-line dynamic opti-
mization and control strategy to enhance batch reactor oper-
ation has been proposed. The on-line dynamic optimization
by the idea of receding horizon scheme is performed. The ap-
proach is based on the updated current information of states
of the system which are estimated from the delayed mea-
surement of the amount of reactants in the reactor using an
Extended Kalman Filter (EKF) technique, at specied time
interval to provide a new updated optimal reactor temper-
ature set point prole. Two types of optimization formula-
tion related to batch operation (maximum concentration and
minimum time problem) were considered in the proposed
on-line set point modication strategy. The obtained reac-
tor temperature set point was implemented using a generic
model control (GMC). The EKF was also incorporated into
the GMC algorithm in order to estimate the heat released by
reactions using the direct measurement of reactor and jacket
temperature. A batch reactor with highly exothermic reac-
tions was used as a simulation case study to demonstrate the
effectiveness of the proposed approach. Simulation studies
have been carried out in both nominal case and plant/model
mismatch case and the results show that the performance of
the batch reactor in terms of the amount of a desired prod-
uct and batch operation time can be improved signicantly
by the proposed strategy. In addition, they also clearly in-
dicate the capability of the GMC controller to control the
reactor temperature along the specied trajectory and that
of the EKF to estimate the states and parameters of the
system.
Acknowledgements
The nancial support to A. Arpornwichanop through local
graduate scholarship from the national science and technol-
ogy development agency (NSTDA) is gratefully acknowl-
edged. A. Arpornwichanop also acknowledges the support
of the Department of Chemical Engineering, University of
Bradford, UK during his stay as a visiting researcher at the
University in 2000.
Appendix A. The EKF algorithm
Since in this work the process models used are in time
continuous form and measurements are in discrete form, the
EKF in discrete/continuous formulation [27] is used. The
basic algorithm of the EKF can be summarized as follows:
For nonlinear systems, the process model can be described
by differential equations
x = F(x(t), u(t), t) + (t) (A.1)
y = G(x(t)) + (t) (A.2)
where F is a vector of system function, G is a vector of
measurement function, is a zero mean Gaussian process
noise with covariance Q, and is a zero mean Gaussian
measurement noise with covariance R.
The equations for the EKF are given by a set of correction
and prediction equations as shown in the following:
Correction phase: Correct the prior estimates of state at
k 1 and update the weighting matrix
K
k1
= P
k1/k2
C
k1
T
[C
k1
P
k1/k2
C
k1
T
+ R]
1
(A.3)
x
k1/k1
= x
k1/k2
+ K
k1
[y
k1
C
k1
x
k1/k2
]
(A.4)

P
k1/k1
=[I K
k1
C
k1
]

P
k1/k2
[I K
k1
C
k1
]
T
+K
k1
RK
T
k1
(A.5)
Prediction phase: Integrate the nonlinear state and co-
variance equations from time k 1 to k in order to acquire
estimate x
k/k1
and

P
k/k1

x = F(x, u) (A.6)

P = A
k1

P +

PA
T
k1
+ Q (A.7)
where x
k/k1
denotes the estimate of state x at t = k from
information at t = k 1, K is Kalman gain matrix, P is
covariance matrix, and matrices
A
k1

=
F
x

x
k1|k1

and C

=
G
x

x
k1|k2

are the Jacobians of function F and G with respect to the


state vector, respectively.
Appendix B. Nomenclature
k
i
rate constant for reaction i (kmol
1
/s)
k
1
i
rate constant 1 for reaction i
k
2
i
rate constant 2 for reaction i
t time (min)
t sampling time (min)
u input variables
x state variables
y output variables
A heat transfer area (m
2
)
Cp mass heat capacity (kJ/(kg

C))
Cp
i
molar heat capacity of component i (kJ/(kmol

C))
114 A. Arpornwichanop et al. / Chemical Engineering and Processing 44 (2005) 101114
F owrate (m
3
/min)
H
i
heat of reaction of reaction i (kJ/kmol)
K
1
, K
2
GMC controller parameters
M
i
number of moles of component i (kmol)
MW
i
molecular weight of component i (kg/kmol)
P covariance matrix
Q covariance matrix of process noise
Q
r
heat released from reactions (kJ/min)
R covariance matrix of measurement noise
R
i
rate of reaction i (kmol/min)
T reactor temperature (

C)
U heat transfer coefcient (kJ/(min m
2
C))
V reactor volume (m
3
)
W Reactor content (kg)
density (kg/m
3
)
Subscripts
f lter
j jacket
r reactor
sp set point
References
[1] G.E. Rotstein, D.R. Lewin, Control of an unstable batch chemical
reactor, Comp. Chem. Eng. 16 (1) (1992) 2749.
[2] D. Bonvin, Optimal operation of batch reactors a personal view,
J. Proc. Cont. 8 (5/6) (1998) 355368.
[3] N. Aziz, I.M. Mujtaba, Optimal operation policies in batch reactors,
Chem. Eng. J. 85 (2002) 313325.
[4] N. Aziz, M.A. Hussain, I.M. Mujtaba, Performance of different
types of controllers in tracking optimal temperature proles in batch
reactors, Comp. Chem. Eng. 24 (2000) 10691075.
[5] C. Loeblein, J.D. Perkins, B. Srinivasan, D. Bonvin, Performance
analysis of on-line batch optimization systems, Comp. Chem. Eng.
21 (Suppl) (1997) s867s872.
[6] D. Ruppen, D. Bonvin, D.W.T. Rippin, Implementation of adap-
tive optimal control operation for a semi-batch reaction system,
Comp. Chem. Eng. 22 (1998) 185189.
[7] C. Kravaris, R.A. Wright, J.F. Carrier, Nonlinear controller for tra-
jectory tracking in batch processes, Comp. Chem. Eng. 13 (1/2)
(1989) 7382.
[8] B.J. Cott, S. Macchietto, Temperature control of exothermic batch
reactors using generic model control, Ind. Eng. Chem. Res. 28 (1989)
11771184.
[9] Z.H. Liu, S. Macchietto, Model based control of a multipurpose
batch reactor an experimental study, Comp. Chem. Eng. 19(Suppl)
(1995) s477s482.
[10] S. Yuce, A. Hasaltun, S. Erdogan, M. Alpbaz, Temperature control
of a batch polymerization reactor, Trans. IChemE 77 (Part A) (1999)
413420.
[11] P. Kittisupakorn, M.A. Hussain, A. Arpornwichanop, Temperature
control of an exothermic batch reactor by model predictive exother-
mic batch reactor by model predictive control, in: Proceedings of the
Third Asian Control Conference, Shanghai, China, 47 July 2000,
pp. 25062511.
[12] N. Aziz, M.A. Hussain, I.M. Mujtaba, Optimal control of batch
reactor: comparison of neural network based GMC and inverse model
control approach, in: Proceedings of the Sixth World Congress of
Chemical Engineering, Melbourne, Australia, 2327 September 2001.
[13] R. Berber, Control of batch reactors: a review, Trans. IChemE 74 (Part
A) (1996) 320.
[14] L.R. Petzold, A description of DASSL: a differential/algebraic system
solver, SAND82-8637, Sandia National Laboratories, 1982.
[15] W.H. Ray, Advanced Process Control, McGraw Hill, New York,
1981.
[16] J.G. Renfro, A.M. Morshedi, O.A. Asbjornsen, Simultaneous opti-
mization and solution of systems described by differential/algebraic
equations, Comp. Chem. Eng. 11 (5) (1987) 503517.
[17] V. Nevistic, Ph.D. thesis, Constrained control of nonlinear systems,
Swiss Federal Institute of Technology, 1997.
[18] R. Luus, Optimal control by dynamic programming using systematic
reduction in grid size, Int. J. Control 51 (1990) 9951013.
[19] S.A. Dadebo, K.B. Mcauley, Dynamic optimization of con-
strained chemical engineering problems using dynamic programming,
Comp. Chem. Eng. 19 (5) (1995) 513525.
[20] J.E. Cuthrell, L.T. Biegler, On the optimization of differential-
algebraic process systems, AIChE J. 33 (8) (1987) 1257
1270.
[21] C.L. Chen, Ph.D. thesis, A class of successive quadratic programming
methods for owsheet optimization, University of London, 1988.
[22] L.S. Kershenbaum, P. Kittisupakorn, The use of a partially simulated
exothermic (PARSEX) reactor for experimental testing of control
algorithms, Trans. IChemE. 72 (Part A) (1994) 5563.
[23] J.X. Shen, M.S. Chiu, Q.G. Wang, A comparative study of model-
based control techniques for batch crystallization process, J. Chem.
Eng. Jpn. 32 (4) (1999) 456464.
[24] P.L. Lee, G.R. Sullivan, Generic model control (GMC), Comp. Chem.
Eng. 12 (6) (1988) 573580.
[25] B.G. Liptak, Controlling and optimization chemical reactors, Chem.
Eng. Magazine (May 1986) 6981.
[26] P. Kittisupakorn, Ph.D. thesis, The use of nonlinear model predic-
tive control techniques for the control of a reactor with exothermic
reactions, University of London, 1995.
[27] A. Gelb, Applied Optimal Estimation, The MIT Press, Cambridge,
MA, 1974.
Chemical Engineering and Processing 44 (2005) 115119
Synthesis of nano-sized particles from metal carbonates
by the method of reversed mycelles
Christo Karagiozov, Dana Momchilova

Departement of Chemical Engineering, Faculty of Technical Sciences, University Prof. Dr. Assen Zlatarov Bourgas,
Bul: Prof. Iakimov 1, 8010 Bourgas, Bulgaria
Received 2 December 2003; received in revised form 30 April 2004; accepted 5 May 2004
Available online 25 August 2004
Abstract
The nano-sized particles of calcium carbonate and barium carbonate have specic characteristics. They are important materials for the
industry. The main object of this investigation is to obtain nanoparticles of calcium carbonate and barium carbonate by chemical reaction
carried out in microemulsion of water in oil. The nanoparticles obtained are spherical. Their sizes vary from 20 to 30 nm. The shape and size
of particles are determinated by electron microscopy.
2004 Elsevier B.V. All rights reserved.
Keywords: Calcium carbonate; Barium carbonate; Nanoparticles; Microemulsion; Surface tension
1. Introduction
The concepts of the nano-world suggested recently re-
vealed new elds of scientic research. The number of the
publications, patents, projects, implementations and compa-
nies involved with nano-technology has been exponentially
increasing during the last 10 years. Nano-technology is a
transition eld between the concepts of the classic quantum
mechanics and physical chemistry of solid state [1].
The investigations of materials obtained from nano-sized
particles showed that they have unknown properties or
enhanced characteristics compared to common materials.
Many alkaline carbonates have take in especially application
in practice. Calcium carbonate is widely used for production
of toothpaste, binding agents, etc. [12,13]. Barium carbon-
ate is used mainly in glass production industries, as well as
water softening agent, etc. Both substances are crystalline,
with rhombic lattice and practically insoluble in water [12].
Many methods have been used for synthesis of nano-sized
particles: vapour transport [2], precipitation of homogeneous
solution [3], synthesis by sol-gel process [4], freeze-drying

Corresponding author. Tel.: +359 56 858351/424;


fax: +359 56 880249.
E-mail addresses: chcar@abv.bgs, da mom@yahoo.com
(D. Momchilova).
[5,6], electrolysis [10], synthesis in microemulsion or the so
called method of reversed micelles [710].
Microemulsion is used as a special microreactor to limit
the nano-sized particles growth. The shape of the microre-
actor depends on reaction conditions [9]. This method in-
creases the homogeneity of the chemical composition at
nano-level and facilitates the preparation of nano-particles
with comparatively equal sizes [11]. The specic properties
of the nano-particles make them suitable for microelectron-
ics, ceramics, catalysis, medicine, cosmetics, as piezoelec-
tric materials, conductors, etc.
The aim of the present investigation is to synthesize
nano-sized particles from CaCO
3
and BaCO
3
by the mi-
croemulsion method.
2. Experimental
The synthesis of nano-sized particles from CaCO
3
and
BaCO
3
was carried out in a recurrent reactor equipped with
stirrer and was close to reactor with ideal mixing. Chemical
interaction between CO
2
and the corresponding alkali sus-
pension (dispersed in organic phase) takes place in the reac-
tor. The process is endothermic so the device was equipped
with cooling jacket. A diagram of the device is shown in
Fig. 1.
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.004
116 C. Karagiozov, D. Momchilova / Chemical Engineering and Processing 44 (2005) 115119
Fig. 1. Diagram of the facility for preparation of nano-sized particles. 1:
Bottle with carbon dioxide, 2: rotameter, 3: recurrent reactior with jacket,
4: bubbler, 5: magnetic stirrers, 6: drecssel bottle.
The experiments started with preparation of solutions of
calcium (barium) hydroxide in water and storing them for 1
day. Their exact concentration was determined by titration
with hydrochloric acid. The microemulsion used was aque-
ous alkali solution/oil (w/o) where the water drops contained
solved calcium or barium hydroxide and organic phase
n-hexane.
The total volume of the microemulsion was 50 ml for all
the experiments. The volumes of the two phases were varied
from 10 to 40 ml, respectively. The volumes of both organic
and inorganic phases of the system for each experiment are
shown in Table 1.
The liquid mixture was stirred with the magnetic stirrer
for 60 min at constant speed. The emulsion formed con-
tained only one of the reagents. The other was fed into the
reactor as gas (CO
2
) at constant ow rate for 60 min. The
second reagent and the cooling water were counterowing
(Fig. 1). The experiments were carried out at temperature of
2023

C.
3. Results and discussion
Solid nano-sized particles of CaCO
3
and BaCO
3
were ob-
tained from the experiments carried out using chemical re-
Table 1
Ratio of volumes of organic and inorganic phases
Experiment
no.
Inorganic
phase
Organic
phase
Volume of inorganic
phase V
1
(ml)
Volume of organic
phase V
2
(ml)
V
1
/V
2
1 2 3 4 5 6
1 Ca(OH)
2
n-C
6
H
14
5 45 0.111
2 Ca(OH)
2
n-C
6
H
14
10 40 0.250
3 Ca(OH)
2
n-C
6
H
14
15 35 0.429
4 Ca(OH)
2
n-C
6
H
14
20 30 0.667
1 Ba(OH)
2
n-C
6
H
14
30 20 1.5
2 Ba(OH)
2
n-C
6
H
14
25 25 1
3 Ba(OH)
2
n-C
6
H
14
20 30 0.667
4 Ba(OH)
2
n-C
6
H
14
15 35 0.429
5 Ba(OH)
2
n-C
6
H
14
35 15 2.333
7 Ba(OH)
2
n-C
6
H
14
40 10 4
Note: 1. All the experiments were carried out under stirring for 60 min at 400 rpm, 2. CO
2
bubbling time was 60 min.
action in a microemulsion of type water in oil. The qual-
itative analyses of the organic and inorganic phases for all
the experiments showed the presence of calcium or barium
carbonate. The shape and size of the particles were deter-
mined by electron microscopy. Photographs of nano-sized
particles of BaCO
3
are presented in Fig. 2.
The nano-particles synthesized were with spherical shape
and almost equal sizes varying from 20 to 30 nm.
The electron-microscopy analysis (Fig. 2) revealed that
processes of particles agglomeration took place. Since the
opalescence observed with the organic phase from the ex-
periments was not very good, the agglomeration was sup-
posed to occur during the electron microscopy analysis it-
self. It may result from the irradiation of the samples by the
high-energy electron beam used for the analysis. Photograph
A was taken from samples obtained from the experiment
carried out at a ratio of inorganic phase Ba(OH)
2
to organic
phase n-C
6
H
14
15/35 ml. Photograph B was taken at phase
ratio Ba(OH)
2
/n-C
6
H
14
= 40/10 ml.
The results sustained the opinion that the synthesis of
nano-particles in microemulsion (w/o) is to be preferred.
In common emulsions, both phases spontaneously separate
from each other while microemulsions are thermodynami-
cally stable, do not segregate and appear to be transparent.
This can be explained with the size of the water droplets
[14,11]. Microemulsions are also characterized by the so
called dynamic exchange process. The emulsion droplets
in such emulsions constantly integrate and disintegrate, thus
exchanging substance between each other.
The substance dissolved in the emulsion droplets was
quite small amount, i.e. the reagent concentration in the
droplet was low so it could be expected to obtain small par-
ticles with almost equal sizes.
In the case studied, the following physical model was sug-
gested. The microemulsion water solution/organic phase was
formed preliminarily. It contained only one of the reagents
(in this case solution of Ca(OH)
2
or Ba(OH)
2
). The sec-
ond reagent was added in the form of a gas (CO
2
). The
gas passed through the organic liquid, diffused through the
C. Karagiozov, D. Momchilova / Chemical Engineering and Processing 44 (2005) 115119 117
Fig. 2. Electron microscopy photographs of nano-sized particles of BaCO
3
.
interphase surface oil/water solution in the drops and then
interacts with the inorganic phase present in them. Thus,
a chemical reaction was initiated which gave slightly solu-
ble products (CaCO
3
and BaCO
3
). This hypothesis can be
schematically represented as follows (Fig. 3).
The summarized equation of the chemical reaction pro-
ceeding between the corresponding hydroxide and CO
2
is
Me(OH)
2
+CO
2
MeCO
3
+H
2
O (1)
The chemistry of the separate stages of the process within a
droplet can be described by the following reactions:
Me(OH)
2
Me
2+
+2OH (2)
CO
2
+H
2
O H
2
CO
3
(3)
H
2
CO
3
HCO

3
+H
+
(4)
CO
2
+OH

HCO

3
(5)
HCO

3
H
+
+CO
2
3
(6)
Me
2+
+CO
2
3
MeCO
3
(7)
where Me = Ca, Ba.
The basic parameter characterizing nano-particles is their
specic area, which can be determined by the formula [15]
A = 4r
2
N(m
2
/kg) (8)
where A: specic area (m
2
/kg), r: particles radius (m), N:
number of particles per unit of weight.
The number of particles can be calculated from
N =
1
(4/3)r
3

(9)
Substituting Eq. (11) into Eq. (10) gives the following ex-
pression for calculation of the specic area:
A =
3
r
(m
2
/kg) (10)
The size of the nanoparticles of BaCO
3
obtained was
about 2030 nm. Their specic area was calculated to be
from 6.82 10
7
to 4.55 10
7
m
2
/kg. The particles with
20 nm sizes have specic area of 6.82 10
7
m
2
/kg while
particles with 30 nm size 4.55 10
7
m
2
/kg. Therefore, the
specic area decreased with the increase of particle radius
(Fig. 3).
For the microemulsion system inorganic solution/
n-C
6
H
14
, it is important to prove that the inorganic so-
lution was dispersed in the organic phase in the form of
microdrops. To determine the distribution of the phases in
the microemulsion, the forces of surface tension for each
microdrop should be estimated, as well as the forces acting
at the interphase surface.
Forces of surface tension act on the surface of the parti-
cles in the liquid phase. It can be expressed as the work for
the formation of a unit of interphase surface under constant
thermodynamic parameters of the state (temperature, pres-
sure, chemical potentials of the components). This process
is reversible and isothermal. The surface tension forces can
be regarded also as free energy per unit area, i.e. specic
free energy (G
S
). Then, the free energy per unit weight of
particles would be
G
S
= G
S
A (11)
Fig. 3. Probable mechanism of contact between the reagents.
118 C. Karagiozov, D. Momchilova / Chemical Engineering and Processing 44 (2005) 115119
Assuming that G
S
is constant value, then free energy
G
S
should increase with the decrease of particle radius A
[16,17].
The surface tension of 0.02 N solution of Ba(OH)
2
was
determined by the formula [18]

Ba(OH)
2
=
0
+0.049(1 a
W
) (12)
where : surface tension of Ba(OH)
2
(dyne/cm),
0
: sur-
face tension of water at 25

C. It is considered to be
71.97 dyne/cm [18], a
w
: activity of water in the solution.
The water activity in the solution was determined by the
well-known equation [18]
ln a
W
=
tm
Ba(OH)
2
M
Ba(OH)
2
55510RT

B
1
+at +bt
2
+

B
2
1000 +M
Ba(OH)
2
m
Ba(OH)
2

m
Ba(OH)
2
M
Ba(OH)
2
+
1
E

(13)
where R: universal gas constant = 8.314441 J/mol, T: tem-
perature (K), m
i
: solution molality (g/l), M
i
: mole weight
(g/mol), B
1
, a, b, B
2
, E: coefcients calculated by mathe-
matical treatment of the experimental data for heat capacity.
Value of the surface tension of Ba(OH)
2
was calculated to
be 71.88 dyne/cm. Reference data showed that
n-C
6
H
14
=
32.21 [18]. The value of Ba(OH)
2
surface tension was close
to that of water (
0
) since the concentration of Ba(OH)
2
was
low and the solution was very diluted.
Forces of interphase tension
T
1
/T
2
act at the contact sur-
face between the two liquid phases of Ba(OH)
2
(or Ca(OH)
2
)
and n-C
6
H
14
. The thermodynamic condition for a liquid to
spread over the surface of another liquid is [19]

T
2
/
>
T
1
/
+
T
1
/T
2
(14)
where
T
1
/
,
T
2
/
: surface tensions of the liquids related
to their environment (dyne/cm),
T
1
/T
2
: surface tension at
the interphase boundary between the liquids (dyne/cm).
The interphase tension at the phase boundary between an
inorganic solution and oil was assumed to be 50.5 dyne/cm,
according to literary data [19]. The two phases are quite
different by polarity. The basic process taking place at the
contact surface between them is the change of the total con-
tact surface area due to the decrease of the total energy of
the system, i.e. the conditions became suitable for process
of wetting to take place. The coefcient of spreading S can
be calculated from the values of the surface and interphase
tensions according to [20]
S =
T
2
/

T
1
/

T
1
/T
2
, S = 10.853 (15)
At S < 0, the wetting angle can be determined by the equa-
tion [20]
cos
0
=

2
T
2
/

2
T
2
/

2
T
1
/T
2
2
T
1
/

T
2
/
(16)
i.e. cos
0
= 0.3717
Its value was calculated to be
0
= 68.18

. At
0
< 90

(as is in this case), the process called limited wetting or


wetting with impregnation. Therefore, the change of the free
energy per unit area can be found by the expression
G =
T
B
/T

T/
(17)
The interphase tension
T
1
/T
2
corresponds to the so-called
boundary coefcient of spreading S
k
, after equilibrium was
established at the interphase surface between the two liq-
uids. The higher the coefcient of spreading is, the faster is
the spreading of the wetting liquid. In the system studied,

T
1
/T
2
= S
k
= 50.5 > 0. It means that the surface tension of
n-C
6
H
14
remained lower than that of the inorganic solution,
i.e. n-C
6
H
14
wetted the polar phase of aqueous solution of
Me(OH)
2
.
4. Conclusions
The following conclusions can be formulated on the basis
of the results obtained
1. A technique and process for preparation of nano-sized
particles from alkali carbonates was developed on the
basis of the method of reversed micelles and a probable
mechanism was suggested.
2. Nano-sized particles of barium and calcium carbonate
were obtained by a chemical reaction in a microemulsion.
The particles were studied by electron microscopy and
were found to possess spherical shape and diameters from
20 to 30 nm.
3. The nano-sized particles were obtained at various ratios
of the phases forming the microemulsion. They were
about the same size, which conrmed the advantage of
the method selected.
4. The particles synthesized were observed to agglomer-
ate, probably due to effects induced by the electron mi-
croscopy analysis.
5. The values of the surface tensions of the individual phases
were calculated, as well as the value of the interphase
tension. The results conrmed that the inorganic phase
was dispersed in the organic phase.
Appendix A. Nomenclature
a
w
activity of water in the solution
m
i
solution molality (g/l)
r particles radius (m)
o oil
w water
A specic area (m
2
/kg)
B
1
, a, b, B
2
, E coefcients calculated by mathematical
treatment of the experimental data for
heat capacity
G change of the free energy per unit area
C. Karagiozov, D. Momchilova / Chemical Engineering and Processing 44 (2005) 115119 119
G
S
specic free energy
M
i
mole weight (g/mol)
N number of particles per unit of
weight
R Universal gas constant (J/mol)
T temperature (K)
S coefcient of spreading
surface tension of Ba(OH)
2
(dyne/cm)

0
surface tension of water at 25

C (dyne/cm)

T
1
/T
2
surface tension at the interphase boundary
between the liquids (dyne/cm)

T
1/
,
T
2
/
surface tensions of the liquids related
to their environment (dyne/cm)
References
[1] Xp. Kapaoob, . Momoba, Hahoact ha akah
kapohat chtepah ocpectbom mkpoemycoheh meto,
Scientic Conference with International Participation Stara Zagora,
Technical Sciences, Sea Sciences and Ecology, vol. II, 2003,
pp. 179184.
[2] X. Dong, C. Choi, B. Kim, Chemical synthesis of Co nanoparticles
by chemical vapor condensation, Powder Technol. 113 (2002) 75
79.
[3] Y. Du, J. Fang, M. Zhang, J. Hong, Z. Yin, Q. Zhang, Phase
character and structural anomaly of Bi
4
Ti
3
O
12
nanoparticles prepared
by chemical coprecipitation, Mater. Lett. 57 (2002) 802806.
[4] A. Hartridge, A. Bhattacharya, Preparation and analysis of zirconia
doped ceria nanocrystal dispersions, J. Nanoparticle Res. 1 (2001)
7580.
[5] E. Bermejo, T. Becue, C. Lacour, M. Quarton, Synthesis of
nanoscaled iron particles from freeze-dried precursors, Powder Tech-
nol. 94 (Issues 1) (1997) 2934.
[6] R. Valencuela, G. Bueno, A. Solbes, Nanostructured ceria-based
catalysts for oxydehydrogenation of ethane with CO
2
, Topics Catal.
15 (2-4) (2001) 181187.
[7] Y. Mori, Y. Okastu, Y. Tsujimoto, Titanium dioxide nanoparticles
produced in water-in-oil emulsion, J. Nanoparticle Res. 3 (23)
(2001) 219225.
[8] B. Palla, D. Shah, P. Garcia-Casillas, J. Matutes-Aquino, Preparation
of nanoparticles of barium ferrite from precipitation in microemul-
sions, J. Nanoparticle Res. 1 (2) (1999) 215221.
[9] K. Song, J. Kim, Synthesis of high surface area tin (Sn) oxide
powders via water-in-oil microemulsions, Powder Technol. 103 (3)
(2000) 268272.
[10] W. Wang, X. Fu, J. Tang, L. Jiang, Preparation of submicron spherical
particles of silica by the water-in-oil microemulsion method, Colloids
Surf. A: Physicochem. Eng. Aspects 81 (1993) 177180.
[11] J. Shmidt, Reaktionstechnishe Untersuchungen zur Herstellung von
Nanopartikeln in Mikroemulsionen, Dissertation, TU Berlin, 2000.
[12] Xp. Koapob, Heopahha xm, Texhka, C., 1970.
[13] M. Dagaonkar, Effect of a Microphase on GasLiquid Mass Transfer,
Proefschrift (Rijksuniversiteit Groningen) Printed by: Print Partners
IPSKamp, Enschede, 3 December 2001.
[14] M. Roth, Dissertation, Synthhesen zur Herstellungneuartiger-
nanoskaliger Materialien und Dispersionen mittels kontrollierter Fal-
lung, 1998.
[15] K. Tamapy, Kapha xm, Mockba Mp, 1983.
[16] A. Akoh, Xmecka tepmohamka, Mockba: Bica
koa, 1963.
[17] A. pock, 1. Cboctba ha matepta. Xmecka
tepmohamka, 2. Patbop, Eektpoxm, Xmha
khetka, 1951
[18] . a eb, .. Aceeb ko xmecke cboctba haphix
mhookomohehthix pactbopob heopaheckx beectb,
Mockba "Xm", 1988.
[19] H. Backes, J. Ma, E. Bender, G. Maurer, Interfacial tensions in binary
and ternary liquidliquid systems, Chem. Eng. Sci. 45 (1) (1990)
275286.
[20] . Cymm, .B. ophob, ko xmecke ochobi
cmabah pactekah, Mockba Xm, 1976.
Chemical Engineering and Processing 44 (2005) 121130
Mechanism of mass transfer from bubbles in dispersions
Part II: Mass transfer coefcients in stirred gasliquid reactor
and bubble column
V. Linek

, M. Korda c, T. Moucha
Department of Chemical Engineering, Prague Institute of Chemical Technology, Technick 5, 166 28 Prague 6, Czech Republic
Received 15 December 2003; accepted 14 May 2004
Available online 2 July 2004
Abstract
Experimental data on the average mass transfer liquid lm coefcient (k
L
) in an aerated tank stirred by Rushton turbine and in bubble
column are presented. Liquid media were used as 0.8 M sodium sulphite solution, pure or with the addition of Sokrat 44 (copolymer of
acrylonitrile and acrylic acid) or short-ber carboxymethylcellulose (CMC) for the Newtonian and long-ber CMC for the non-Newtonian
viscosity enhancement and ocenol (cis-9-octadecen-1-ol) or polyethylenglycol (PEG) 1000 for surface tension change. Volumetric mass
transfer coefcient (k
L
a) and specic interfacial area (a) were measured by the Danckwerts plot method. Coefcients k
L
measured by pure
oxygen absorption in pure sulphite solution and Newtonian viscous liquids are well tted by the eddy model in the form of k
L
= 0.448
(ev/)
0.25
(D/v)
0.5
with a mean deviation of 20%. Surface-active agents (ocenol and PEG) and non-Newtonian additive (long-ber CMC)
reduced k
L
value signicantly but their effect was not described satisfactorily neither by surface tension nor by surface pressure. It is shown
that the decisive quantity to correlate k
L
in the stirred tank and bubble column is power dissipated in the liquid phase rather than the bubble
diameter and the slip velocity. Absorption of air did not yield correct k
L
data, which did not depend on or slightly decreased with increasing
power. This is due to the application of an improper gas phase mixing model for absorption data evaluation.
2004 Elsevier B.V. All rights reserved.
Keywords: Gasliquid contactor; Bubble column; Agitated vessel; Mass transfer coefcient; Viscosity; Surfactants
1. Introduction
Mass transfer from swarm of bubbles into turbulent liq-
uid controls the rate of many chemical and biochemical pro-
cesses. It is assumed that the mechanism of mass transport
in liquid phase is due to a renewal of the liquid at the bub-
ble surface. Models of the process differ in the scale of ow,
which is responsible for the renewal.
The rst group of models (eddy models) assumes that
the liquid renewal is due to small-scale eddies of the turbu-
lent eld. These models are based on idealized eddy struc-
tures of turbulence in the bubble vicinity. Lamont and Scott
[1] have assumed that the small scales of turbulent motion,
which extend from the smallest viscous motions to the iner-

Corresponding author. Tel. +420 2 2435 3298;


fax: +420 2 3333 7335.
E-mail address: linekv@vscht.cz (V. Linek).
tial ones, affect the mass transfer. In any case, these motions
are much smaller in scale than the gas bubbles. As a result,
the size of the gas bubble is not a very critical parameter for
the estimation of k
L
. They deduced [1] the following rela-
tion for k
L
:
k
L
= c
1

ev

0.25

D
v

0.5
(1)
Different authors predict different values of the constant c
1
:
0.301 [2], 0.4 [1], 0.531 [3], 0.592 [4] and 1.13 [5].
The second group of models (slip velocity models) as-
sumes a gross mean ow of uid relative to the bubble (slip
velocity) and a bubble surface mobility control of this re-
newal rather than the small-scale eddies of the turbulent
eld. The model proposed by Calderbank and Moo-Young
[6] belongs to this group. They [6] have divided bubbles by
size into two categories: the small bubbles (d < 1 mm),
which behave always as a rigid sphere, and the large
ones (d > 2.5 mm), which always have a completely mobile
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.009
122 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130
surface. An expression for the calculation of k
L
for the
large bubbles has been deduced fromdimensionless analy-
sis of mass transfer from rising particle in gravitational eld.
k
L
= 0.42

1/3

D
v

1/2
(2)
For the small bubbles, an expression for the calculation of
k
L
has been obtained by the equation proposed by Frssling
[7] for dissolution of a rigid sphere.
k
L
= 0.31

1/3

2/3
(3)
The transition between these two bubble categories was
unclear and strongly dependent on the presence of surfac-
tants. An expression for calculation of the mass transfer
coefcients in this transition region was not given.
Recently, Vasconcelos et al. [8] and Alves et al. [9] pro-
posed another variant of the slip velocity model. Their
model is based on experimentally observed phenomena of
an abrupt change of the dissolution rate of free-oating bub-
bles held stationary in a downward water ow. Starting from
these experiments, they have interpreted the mass transport
from bubbles in terms of bubble contamination kinetics, us-
ing a stagnant cap model, according to which bubbles sud-
denly change from a mobile interface to a rigid condition
when surface tension gradients, caused by surfactant accu-
mulation, balance out shear stress. The fresh bubble entering
the dispersion has clean, completely mobile interface. For
the bubbles with completely mobile surface, k
L
is given by
the equation, which follows from the well-known penetra-
tion model with exposition time equal to d/v
sl
:
k
mobil
L
= 1.13

v
sl
d
D
1/2
(4)
After some time (
mobile
), enough contaminant for transition
to rigidity is accumulated on the surface. For the rigid bub-
bles, an expression for the calculation of k
L
follows from
Frssling [7] equation.
k
rigid
L
= 0.6

v
sl
d
D
2/3
v
1/6
(5)
Gasliquid slip velocities v
sl
are assumed to be close to
single-bubble terminal velocities in still water, v
t
, on which a
correction for turbulence is introduced, v
sl
= 0.65v
t
[9]. As-
suming monosized bubble dispersion and considering only
two possible values of k
L
(k
mobil
L
and k
rigid
L
, depending on
surface mobility), they deduced the following relation for
the average mass transfer coefcient:
k
L
=
k
mobil
L

mobile
+k
rigid
L
(
R

mobile
)

R
(6)
where
R
is the average bubble residence time.

R
=
V
L

q(1 )
(7)
The time
mobile
depends on bubble diameter and surfactant
concentration.

mobile
= k
d
1/2
ln (d/h
trans
)
c
surf
(8)
where d is an average bubble diameter, k is an empirical con-
stant related with surfactant properties, c
surf
is the concen-
tration of surfactant and h
trans
is the bubble clean segment
height at the transition point from mobile to rigid.
The main difference between the eddy and the slip
velocity models is in the inuence of turbulence inten-
sity on mass transfer coefcient: the eddy models predict
an increase while the slip velocity models a decrease of
k
L
with increasing turbulence intensity, i.e. with increasing
power dissipated in the liquid. This is due to the fact that
k
L
from rigid bubbles is cca four times lower than from the
ones with a mobile surface. The small bubbles, which ei-
ther have lower k
L
(according to Calderbanks conception)
or become rigid more quickly (according to Alves concep-
tion,
mobile
diminishes with diminishing bubble diameter
d, see Eq. (8)), are generated in large amount just at higher
turbulence intensities. As a result, this leads to the decrease
of average mass transfer coefcient with increasing power
dissipated in the liquid. Literature data on k
L
in stirred tanks
supporting both these models can be found.
In dispersions with no mechanical agitation (bubble
columns, air-lifts), the only source of energy is the expansion
of the entering gas which gives the relation e = gv
s
. Reith
[10], Dillon and Harris [11], Kawase et al. [5] and Vzquez
et al. [12] revealed strong dependence of liquid-side mass
transfer coefcient on gas ow rate in bubble columns:
k
L
v
s
0.25/0.5
. Slight dependence or independence of k
L
on v
s
was found by Schumpe and Deckwer [13], Bouai
et al. [14], Wongsuchoto et al. [15] and Vasconcelos et
al. [8]. The values of k
L
corresponded to values predicted
by Calderbanks model for large bubbles. Values which
correspond also to small bubbles were found by Vzquez
et al. [12] and Vasconcelos et al. [8] only, who worked with
solutions of surfactants. Vasconcelos et al. [8] interpreted
their results in terms of bubble contamination kinetics.
In dispersions with mechanical agitation, following au-
thors found an increase of k
L
with increasing energy dissi-
pation e: Prasher and Wills [4] (k
L
e
0.25
); Figueiredo and
Calderbank [16] (k
L
e
0.33
); Bouai and Roustan [17] (k
L
e
0.22
); Panja and Rao [18] (k
L
e
0.15
); and Linek et al.
[19] (k
L
e
0.14
). Power dissipated in the liquid by an agi-
tator is proportional to the third power of impeller rate (e
f
3
). Thus, we can add the results of authors that presented k
L
as a function of impeller rate f: Koetsier and Thoenes [20]
(k
L
f
0.9
e
0.3
) and Yoshida and Miura [21] (k
L
f
0.6
e
0.2
). Robinson and Wilke [22] measured k
L
that did not
depend on power dissipated and Hassan and Robinson [23]
obtained coefcients that decreased with increasing power
dissipated. It was shown [32] that the reported independence
on or the decrease of k
L
with increasing power dissipated
was a result of misinterpretation of k
L
a data.
V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130 123
Figueiredo and Calderbank [16] have analysed the suit-
ability of Calderbanks model [6] for small and large
bubbles to the interpretation of mass transfer coefcients in
stirred vessels. They analysed an effect of the gas phase mix-
ing model used for k
L
a evaluation by the dynamic method
based on the inlet gas exchange N
2
/air. In their original
work [6], the k
L
a evaluation assumed that the interchange
of the gases in gas hold-up is immediate (steady-state model
of the gas phase). Corresponding k
L
values were indepen-
dent of power dissipated in the liquid by the agitator and
matched the constant value of 4.03.10
4
ms
1
as predicted
by Calderbanks model [6] for large bubbles (Eq. (2)).
When the gas phase dynamics was taken into account (the
ushing out of an old gas by a new one), the correspond-
ing values of k
L
were higher and well tted by the eddy
model (Eq. (1)) with c
1
= 0.5, see Fig. 1. (It was shown
by Linek et al. [24] that faulty sampling of liquid used by
Figueiredo and Calderbank [16] leads to considerably over-
estimated k
L
a values for k
L
a > 0.1 s
1
. Therefore, k
L
values
corresponding to this range are not included in the gure.)
They concluded [16] (in difference with the original work
[6]) that k
L
is dependent on the turbulent intensity expressed
as a function of the power dissipated per unit volume and
that the coefcient is considerably higher than would be
found for bubbles in free rise under gravitational force as
given by the model of small and large bubbles.
Vasconcelos variant of the slip velocity model based
on bubble contamination kinetics, Eqs. (4)(8), was used
by Alves et al. [9] to interpret k
L
data in a double Rush-
ton turbine-stirred tank. The application of Vasconcelos
model to the interpretation of unreliable mass transfer co-
Fig. 1. Effect of gas phase mixing model used for k
L
a evaluation; k
L
values calculated from k
L
a and a presented by Figueiredo and Calderbank [16].
efcients in the stirred vessel has been analysed by Linek
et al. [3]. Alves et al. [9] measured the overall volumetric
mass transfer coefcients k
L
a by the peroxide decomposi-
tion steady-state technique with air as a gas phase, show-
ing that k
L
decreased slightly with power dissipated. Linek
et al. [25] showed, however, that the steady-state methods
with air absorption/desorption give substantially lower k
L
a
values at high mixing intensities as compared to pure oxy-
gen absorption. If the coefcients k
L
a were measured by the
non-problematic pressure dynamic method [26] with pure
oxygen absorption in the same apparatus as in Alves et al.
[9], the relevant k
L
data were well interpreted in terms of the
eddy model (Eq. (1)) with c
1
= 0.531 [3]. The corrected
data have shown that the decisive quantity to correlate k
L
in
stirred tanks is power dissipated in the liquid phase rather
than the bubble diameter and their rigidity and that k
L
in-
creases with increasing turbulence intensity.
It seems from the literature survey that in mechanically
agitated dispersions, the intensity of energy dissipation plays
a dominant role in the value of the mass transfer coefcient
and that the successful applications of the slip velocity
models in stirred tanks are probably only a result of errors
in evaluation of k
L
a. On the contrary, in dispersions with no
mechanical agitation, both conceptions (eddy and slip
velocity) are equally successful. This may indicate that the
mechanism of mass transfer without mechanical agitation
differs from that in the presence of agitation, as was sug-
gested by Bouai et al. [14]. The same indicate the results by
Lamont and Scott [1]. They compared rates of mass trans-
fer from turbulent liquid into bubbles owing in vertical
and horizontal tube where the bubbles can and cannot rise
124 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130
freely through the turbulent liquid, respectively. If the bub-
bles could rise freely through the liquid, the mass transfer
coefcient did not depend on intensity of turbulence at low
Reynolds number. With increasing Re number, the role of
the intensity increased and at high Re number, the dependen-
cies for both arrangements (with and without the free rise of
bubbles) coincided. It means that both mechanisms operate
simultaneously but they apply at different extent, depending
on intensity of turbulence.
Surface-active substances reduce the mass transfer coef-
cient markedly in gasliquid dispersions. Although a con-
siderable effort was made to identify the mechanism of mass
transfer in the presence of surfactants, no consistent conclu-
sion had been drawn. Two possibilities were offered [27]:
surfactant molecules affect mass transfer via hydrodynam-
ics effects or via the formation of an interfacial barrier layer.
The barrier effect is due to a monolayer of the surfactants
formed at the interface, which can offer a resistance to be
crossed by the gas molecule. The resistance exerted by the
monolayer is a complex phenomenon, possibly depending on
the effects of absorbing gassurfactantwater interactions,
which are related to the molecular structure of the surfac-
tant: the polarity of the hydrophilic group, the molecular
weight of the hydrophilic group, the hydrophilic chain length
and the orientation of the molecules at interface. Hydrody-
namics effects are induced by the difference in composition
and physical properties of the liquid at the interface and in
the bulk of liquid. Surfactants (but also electrolytes [28,29])
tend to create different spatially oriented structures in close
proximity of the interface and in bulk liquid and thus, their
presence causes an energetic barrier for the surface renewal
due to additional requirements of reordering the liquid en-
tering or leaving the interface. For example, if the liquid
turbulent eddies penetrate to the interface, they would bring
about an increase in the free energy per unit area of inter-
face due to the replacement of the interfacial liquid layer
by inner liquid of higher surface tension. This dampens the
rejuvenation of the interfacial liquid by fresh bulk liquid,
renders mass transfer more difcult and lowers the value
of mass transfer coefcient. Llorens et al. [27] decided the
hydrodynamic effect to be the principal. They offered the
surface pressure (i.e. the difference in surface tension of
the solution without and with
s
, the surfactant) to rep-
resent a measure of the additional work needed to replace
a surface element: (k
L
)
s
/k
L
= 1/(1.25 + 0.073 ) for 0.4 <
< 16 mNm
1
and (k
L
)
s
/k
L
= 0.41 for > 16 mNm
1
.
The form of the correlation did not depend on the type of
surfactant. Zieminsky and Whittemore [28] offered the ionic
strength and Machon and Linek [29] the electrical surface
potential to correlate the reduction of k
L
in electrolyte solu-
tions as measures of the additional work needed to replace a
surface element in these solutions. Vzquez et al. [12] used
a simple correlation between the mass transfer coefcient
and surface tension of solution under the form k
L

1.35
.
Vasconcelos et al. [8] interpreted mass transfer coefcients
in air-lift and bubble columns with solution of surfactants by
the model of gradual surface contamination (Eqs. (4)(8))
using three empirical parameters: h
trans
, k/c
surf
and v
t
/v
sl
.
The aim of this paper is to make measurements with liq-
uids of various physical properties in order to dene the
effect of the liquid properties and operating conditions on
the parameter k
L
and the limits of validity of the literature
models for the interpretation of mass transfer coefcients in
bubble dispersions. The method, which is used for the mea-
surements, was veried in Part I to minimize misinterpreta-
tions.
2. Experimental
The experiments were performed in a at-bottomed
cylindrical vessel with internal diameter 0.19 m, equipped
with standard Rushton turbine of 0.075 m in diameter lo-
cated 0.08 m above the bottom. Four bafes (1/10 of the
vessel diameter width) were symmetrically mounted. A
membrane-covered polarographic oxygen probe was sit-
uated in the bottom to measure the equilibrium partial
pressure of oxygen in the bulk of liquid. Gas (pure oxygen
or air) was taken from a pressure ask and introduced 2 cm
under the impeller through a tube of i.d. 4 mm. The bubble
column was realized by keeping zero rotational frequency of
the agitator. The experiments were performed with the same
solutions as those used in the kinetics measurement: 0.8 M
sodium sulphite solution, pure or with the addition of vis-
cosity or surface tension-changing additives (see Table 1).
Physical properties of the solutions are given in Part I of
this work. The vessel was lled with liquid to the height of
226 mm, which amounts to 5.9 l. Supercial velocities of
aeration gas were 1.8, 3.6 or 5.4 mms
1
, which correspond
to volumetric ow rates 5.1 10
5
, 10.2 10
5
and 15.3
10
5
m
3
s
1
, respectively. The impeller rotational speeds
were 0, 5.5, 10, 14.33 and 18.83 l/s. Temperature was kept
at 30 1

C by a cooling coil.
The specic power dissipated by agitators in the liquid,
e
agit
, was measured by a strain gauge mounted on the im-
peller shaft. The total specic power e dissipated in the liq-
uid by the agitator and the rising bubbles was calculated as
e = e
agit
+ gv
s
. The volumetric mass transfer coefcient
and interfacial area were measured by the Danckwerts plot
method described in detail in Part I.
3. Results
3.1. Volumetric mass transfer coefcient
The overall volumetric mass transfer coefcients mea-
sured by pure oxygen absorption were expressed as a func-
tion of the total specic power dissipated in the liquid as
follows:
k
L
a =
1
e

2
(9)
V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130 125
Table 1
Solutions used in this work and pertinent symbols and lines used in gures and parameters , and in correlations (9), (10) and (11) obtained by
regression of the experimental data
Solution Symbol Lines Parameter i = 1 i = 2 Mean
deviation (%)
Mean
deviation
a
0.8 kmol m
3
Na
2
SO
3

i
3.35 10
5
1.18 34 21

i
0.523 0.863 36 36

i
1.06 10
4
0.238 19 20
0.8 kmol m
3
Na
2
SO
3
+ Sokrat 44 (3 vol.%)
i
3.26 10
5
1.16 44 43

i
0.507 0.913 40 43

i
6.70 10
5
0.243 11 11
0.8 kmol m
3
Na
2
SO
3
+ CMC TS.5 (0.2 wt.%)
i
3.65 10
5
1.08 30 29

i
1.21 0.632 33 49

i
1.04 10
4
0.276 25 26
0.8 kmol m
3
Na
2
SO
3
+ CMC TS.20 (0.6 wt.%)
i
2.21 10
4
0.771 14 14

i
1.80 0.689 13 14

i
1.15 10
4
0.090 5 5
0.8 kmol m
3
Na
2
SO
3
+ ocenol (3 ppm by volume) +
i
1.27 10
4
0.732 27 14

i
2.35 0.492 12 9

i
7.97 10
5
0.188 17 13
0.8 kmol m
3
Na
2
SO
3
+ PEG 1000 (100 ppm by mass)
i
2.65 10
5
1.15 1

i
1.14 0.903 2

i
0.237 0.246 1
a
Mean deviation when the gas ow rate v
s
is taken into account as an additional parameter (e.g., k
L
a =
1
e

2
v

3
s
).

1, 2
values together with mean deviations of the data are
given for all batches in Table 1. Taking into account gas
ow rate v
s
as additional independent parameter has not
improved delity of the correlation as follows from a com-
parison of the mean deviations of the relation (9) with and
without taking into account v
s
(see Table 1). Fig. 2 clearly
shows that k
L
a data for pure sulphite solution agree well
with a recommended correlation for non-coalescent batches
Fig. 2. k
L
a/(v
s
)
0.4
measured by pure oxygen absorption as a function of total power dissipated in the liquid, e. Symbols and lines (see Table 1).
of electrolyte solutions [24]. Marked reductions of k
L
a
are observed with ocenol (up to 10 times) and long-ber
carboxymethylcellulose (CMC) TS.20 (up to four times).
The ocenol has a strong antifoaming effect that enhances
the coalescence rate of bubbles in dispersion. The resulting
coefcients are even lower than those predicted by a corre-
lation suggested for coalescent batches like pure water [24].
The reduction of k
L
a due to CMC TS.20 cannot be ascribed
126 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130
Fig. 3. Interfacial area a measured by pure oxygen absorption as a function of total power dissipated in the liquid, e. Symbols and lines (see Table 1).
unequivocally to higher viscosity of the batch as the addi-
tion of Sokrat 44 (which increased the viscosity to the same
extent as CMC) made no reduction in k
L
a.
3.2. Specic interfacial area
The specic interfacial areas measured by pure oxygen
absorption were expressed as a function of the total specic
power dissipated in the liquid as follows:
a =
1
e

2
(10)

1, 2
values together with mean deviations of the data are
given for all batches in Table 1. Taking into account gas
ow rate v
s
as an additional independent parameter has not
improved delity of the correlation as follows from a com-
parison of the mean deviations of the relation (10) with and
without taking into account v
s
(see Table 1). This agrees
with the nding of Westersterp et al. [30], who stated that
the interfacial area is independent of gas ow rate at im-
peller speeds higher than a critical one. The critical im-
peller speed for Rushton turbine and sulphite solution equals
9.4 s
1
[29]. All but one of the impeller speeds used in this
work were higher than this critical value. The results given
in Fig. 3 show that ocenol and short-ber CMC TS.5 pro-
duce a marked decrease of the interfacial area (up to ve and
three times, respectively), and that a marked increase is pro-
duced by polyethylenglycol (PEG) (up to three times) and
a smaller one by the addition of Sokrat (up to two times).
3.3. Mass transfer coefcient
Mass transfer coefcients were calculated by the ratio
k
L
a/a. The coefcients evaluated from the data measured by
pure oxygen absorption were expressed as a function of the
total specic power dissipated in the liquid as follows:
k
L
=
1
e

2
(11)

1, 2
values together with mean deviations of the data are
given for all batches in Table 1. The results show clearly that
k
L
increases with increasing power dssipated in the liq-
uid with exponent
2
equal approximately to 0.25 in agree-
ment with the eddy model (1) (see Fig. 4). Exceptions
are the solutions with the addition of long-ber CMC TS.20
and ocenol. In these solutions, the coefcients increase with
power but the exponent
2
is lower than 0.25 (
2
= 0.09
and 0.188, respectively). The correlation (11) also ts well
the data measured in the bubble column, which are repre-
sented by the points for e < 100 Wm
3
in Fig. 4. k
L
data
for all Newtonian batches without surface-active agents (i.e.
pure sulphite solution and with the addition of Sokrat and
short-ber CMC) are well tted by the equation
k
L
= 0.915

ev

0.243

D
v

0.623 (12)
with a mean deviation of 19%. If the theoretical value 0.25 of
the power exponent and 0.5 of Schmidt number exponent
are used, the equation changes to
k
L
= 0.448

ev

0.25

D
v

0.5
(13)
with a mean deviation of 20%. Since the deviation is prac-
tically the same as the deviations of correlation (11), cor-
relations (12) and (13) describe the effects of viscosity and
diffusivity satisfactorily. Experimental and calculated values
from Eq. (13) of k
L
are compared in Fig. 5.
V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130 127
Fig. 4. Liquid lm mass transfer coefcient measured by pure oxygen absorption vs. total power dissipated in the liquid, e. Symbols and lines (see Table 1).
Literature data for Newtonian batches (water [3,4,16,17]
and for sodium sulphite solutions [3,19]) together with
our Newtonian data were also evaluated according to
relation
k
L
= c
1

ev

c
2

D
v

c
3
(14)
yielding the regression constants c
1
, c
2
and c
3
given in
Table 2. Correlation (14) with the theoretical exponents c
2
= 0.25 and c
3
= 0.5 ts our and literature data on Newto-
nian batches with a mean deviation of 20%. Experimental
and calculated values of k
L
from Eq. (14) are compared in
Fig. 6.
Fig. 5. Comparison of experimental and calculated values of mass transfer
coefcient k
L
from Eq. (13). Symbols (see Table 1).
Table 2
Parameters c
1
, c
2
and c
3
in correlation (14) obtained by regression of
literature and our data for Newtonian batches
k
L
= c
1
(ev/)
c
2
(D/v)
c
3
c
1
c
2
c
3
Mean deviation (%)
0.363 0.213 0.542 19
0.265 0.210 0.500 19
0.463 0.250 0.500 22
3.4. Effect of surfactants
Additions of substances for surface tension depression
(ocenol, PEG, CMC TS.20, see Table 1 of Part I) induced
a signicant reduction of the mass transfer coefcient (by
Fig. 6. Comparison of our and literature experimental data for Newtonian
batches with those calculated from Eq. (14) with c
1
= 0.463, c
2
= 0.25
and c
3
= 0.5. Hollow symbols (see Table 1) [31].
128 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130
Fig. 7. Relative decrease of mass transfer coefcient in the presence of
surfactant (in comparison to its value in pure sulphite solution) as a
function of surface pressure . Comparison with literature data measured
in bubble columns [12,8] and in wetted-wall column [27].
40 up to 80%). The results are given in Fig. 7 together with
literature data measured in bubble columns [12,8] and in
wetted-wall column [27]. In the gure, the relative decrease
of k
L
(in comparison to its value in pure sulphite solution,
(k
L
)
s
/k
L
) is plotted as a function of surface pressure . It is
apparent that neither the surface pressure nor surface tension
alone (cf. the values of given in Table 1 of Part I) is
Fig. 8. Comparison of k
L
a, a and k
L
data measured by pure oxygen and air absorption in pure sulphite solution and with addition of ocenol. Air
absorption, lled symbols; oxygen absorption, hollow symbols; see Table 1.
the proper parameter to correlate the effect of surfactants
on k
L
.
3.5. Effect of diluted gas (air) absorption
Experiments with air were performed in pure sulphite so-
lution (non-coalescent batch) and with the addition of ocenol
(coalescent). The results are plotted in Fig. 8 together with
those measured by pure oxygen absorption. While the mass
transfer coefcients measured with oxygen increase with in-
creasing dissipated power, the k
L
values determined with air
either do not depend (in pure sulphite solution) or slightly
decrease with increasing power (in sulphite with ocenol).
The coefcients roughly coincide at the lowest stirring
intensity (330 rpm). With increasing stirring intensity, the
difference between them increases up to 70% at the highest
stirring intensity (1130 rpm). The differences can be caused
by application of an improper gas phase mixing model of the
dispersion, leading to an incorrect value of the driving force
for absorption, which proliferates to k
L
values. Application
of pure oxygen absorption avoids the problem as the oxygen
concentration and, therefore, the absorption efciency are
the same for all bubbles. Thus, k
L
data used in this work are
measured by pure oxygen. The absorption of air has only
been used to demonstrate the possible reason of discrepancy
in the literature data. Fig. 9 shows that k
L
data measured
with air can be interpreted in terms of Calderbanks model:
data measured in pure sulphite solution are independent
of bubble diameter and correspond to Eq. (2) for large
V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130 129
Fig. 9. Comparison of mass transfer coefcients measured by air absorption in pure sulphite solution and in the solution with addition of ocenol with
Calderbanks model.
bubbles. (The average bubble diameter was calculated from
gas hold-up and interfacial area as d = 6/[(1)a].) The
data measured in solution with ocenol fall between the
values for large and small bubbles. This is usually
interpreted by partial contamination of bubble surface by
surfactant, which shifts the transition between large and
small bubbles to higher bubble diameters. The decrease
of k
L
values with increasing stirring intensity (see Fig. 8 for
ocenol) can be explained by an increasing amount of small
bubbles. Such interpretation is frequently used in literature
for the data measured by absorption of diluted gas.
4. Conclusions
Mass transfer coefcients measured by absorption of pure
oxygen increased in all batches with increasing power dis-
sipated in the liquid raised approximately to 0.25. An ex-
ception was the addition of long-ber CMC for which the
exponent is only 0.09. k
L
data for all Newtonian batches
without surface-active agents (i.e. pure sulphite solution and
with addition of Sokrat and short-ber CMC) are well t-
ted by the eddy model of Lamont and Scott [4], Eq. (13).
Substances that depress the surface tension of the absorp-
tion solution (ocenol, PEG, CMC TS.20) induced a signif-
icant reduction of the mass transfer coefcient (by 40 up
to 80%). Absorption of air (diluted gas) did not yield cor-
rect k
L
data, not depending on or slightly decreasing with
increasing power. This is due to the application of an im-
proper model of gas ow pattern in the dispersion (i.e. per-
fect mixing of gas phase), which leads to an incorrect value
of the driving force for absorption that proliferates to k
L
values. Since this data are well interpreted by Calderbanks
model of small and large bubbles, this model seems to
be the result of interpretation of physically inconsistent k
L
data.
Acknowledgements
This work was supported by the Grant agency of Czech
Republic through the project no. 104/98/1126 and by Czech
Ministry of Education through the scientic project no.
223400007.
Appendix A. Nomenclature
a specic interfacial area based on the liquid
volume, m
1
c
1, 2, 3
constants in correlation (14)
d bubble diameter, m
D gas diffusivity in the liquid, m
2
s
1
e total specic power dissipated in the liquid
volume, Wm
3
e
agit
specic power dissipated by agitators in the
liquid volume, Wm
3
f impeller rate, s
1
g gravitational constant, ms
2
k
L
liquid-side mass transfer coefcient measured by
oxygen absorption, ms
1
(k
L
)
s
liquid-side mass transfer coefcient in the
presence of surfactant, ms
1
(k
L
)
air
liquid-side mass transfer coefcient measured by
air absorption, ms
1
k
L
a volumetric mass transfer coefcient referred to
the liquid volume, s
1
k empirical constant of the model of bubble
contamination kinetics, mol m
7/2
s
130 V. Linek et al. / Chemical Engineering and Processing 44 (2005) 121130
q volumetric gas ow rate, m
3
s
1
v
s
supercial velocity, ms
1
v
t
single bubble terminal velocity in still water, ms
1
v
sl
slip velocity, ms
1
Greek symbols
, , constants in correlations (9), (10) and (11)
gas hold-up
kinematics viscosity of the liquid phase, m
2
s
1
=(
s
), surface pressure, Nm
1
density of the liquid phase, kg m
3
=(
liquid

gas
), kg m
3

s
, surface tension of solution with or without
surfactant, Nm
1

mobile
time for transition of bubble surface to rigidity, s

R
average bubble residence time, s
References
[1] J.C. Lamont, D.S. Scott, An eddy cell model of mass transfer into
surface of a turbulent liquid, AIChE J. 16 (1970) 513519.
[2] Y. Kawase, B. Halard, M. Moo-Young, Liquid-phase mass trans-
fer coefcients in bioreactors, Biotech. Bioeng. 39 (1992) 1133
1140.
[3] V. Linek, M. Fujasov, M. Korda, T. Moucha, Gasliquid mass
transfer coefcient in stirred tank interpreted through models of
idealized eddy structure of turbulence in the bubble vicinity, Chem.
Eng. Proc. (in press).
[4] B.D. Prasher, G.B. Wills, Mass transfer in an agitated vessel, I&EC
Process Des. Develop. 12 (1973) 351354.
[5] Y. Kawase, B. Halard, M. Moo-Young, Theoretical prediction of
volumetric mass transfer coefcients in bubble columns for Newto-
nian and non-newtonian uids, Chem. Eng. Sci. 42 (1987) 1609
1617.
[6] P.H. Calderbank, M.B. Moo-Young, The continuous phase heat and
mass transfer properties of dispersions, Chem. Eng. Sci. 16 (1961)
3961.
[7] N. Frssling, ber die Verdnstung fallenden Tropfen, Gerlands
Beitage Geophys. 52 (1938) 170216.
[8] J.M.T. Vasconcelos, J.M.I. Rodrigues, S.C.P. Orvalho, S.S. Alves,
R.L. Mendes, A. Reis, Effect of contaminants on mass transfer
coefcients in bubble column and airlift contactors, Chem. Eng. Sci.
58 (2003) 14311440.
[9] S.S. Alves, C.I. Maia, J.M.T. Vasconcelos, Gasliquid mass transfer
coefcient in stirred tank interpreted through bubble contamination
kinetics, Chem. Eng. Proc. 43 (7) (2004) 823830.
[10] T. Reith, Physical aspects of bubble dispersions in liquids, Ph.D.
Thesis, Delft Techn. Univ., 1968.
[11] G.B. Dillon, I.J. Harris, The determination of mass transfer coef-
cients and interfacial areas in gasliquid contacting systems, Can. J.
Chem. Eng. 44 (1966) 307312.
[12] G. Vzquez, M.A. Cancela, C. Riverol, E. Alvarez, J.M. Navaza,
Application of the Danckwerts method in a bubble column. Effects
of surfactants on mass transfer coefcient and interfacial area, Chem.
Eng. J. 78 (2000) 1319.
[13] A. Schumpe, W.D. Deckwer, Gas holdups, specic interfacial areas,
and mass transfer coefcients of aerated carboxymethyl cellulose
solutions in a bubble column, I&EC Process Des. Develop. 21 (1982)
706711.
[14] M. Bouai, G. Hebrard, D. Bostoul, M. Roustan, A comparative
study of gas hold-up, bubble size, interfacial area and mass transfer
coefcients in gasliquid reactors and bubble columns, Chem. Eng.
Proc. 40 (2001) 97111.
[15] P. Wongsuchoto, T. Chrinpanitkul, P. Pavasant, Bubble size distribu-
tion and gas/liquid mass transfer in airlift contactors, Chem. Eng. J.
92 (2003) 8190.
[16] M.M.L. Figueiredo, P.H. Calderbank, The scale-up of aerated mixing
vessels for specied oxygen dissolution rates, Chem. Eng. Sci. 34
(1979) 13331338.
[17] M. Bouai, M. Roustan, Bubble size and mass transfer coefcients
in dual-impeller agitated reactors, Can. J. Chem. Eng. 76 (1998)
390397.
[18] N.C. Panja, D.P. Rao, Measurement of gasliquid parameters in a
mechanically agitated contactor, Chem. Eng. J. 52 (1993) 121129.
[19] V. Linek, J. Mayrhoferov, J. Monerov, The inuence of diffusivity
on liquid phase mass transfer in solutions of electrolytes, Chem.
Eng. Sci. 25 (1970) 10331045.
[20] W.T. Koetsier, D. Thoenes, Mass transfer in a closed stirred gas/liquid
contactor part 2: the liquid phase mass transfer coefcient k
L
, Chem.
Eng. J. 5 (1973) 7175.
[21] F. Yoshida, Y. Miura, Gas absorption in agitated gasliquid contac-
tors, I&EC Process Des. Develop. 2 (1963) 263268.
[22] C.W. Robinson, C.R. Wilke, Simultaneous measurement of interfacial
area and mass transfer coefcients for a well-mixed gas dispersion
in aqueous electrolyte solutions, AIChE J. 20 (1974) 285294.
[23] I.T.M. Hassan, C.W. Robinson, Mass transfer coefcients in mechan-
ically agitated gasaqueous electrolyte dispersions, Can. J. Chem.
Eng. 58 (1980) 198205.
[24] V. Linek, V. Vacek, P. Bene, A critical review and experimental
verication of the correct use of the dynamic method for the de-
termination of oxygen transfer in aerated agitated vessels to water,
electrolyte solutions and viscous liquids, Chem. Eng. J. 34 (1987)
1134.
[25] V. Linek, P. Bene, J. Sinkule, Critical assessment of the steady-
state Na
2
SO
3
feeding method for k
L
a measurement in fermentors,
Biotechnol. Bioeng. 35 (1990) 766770.
[26] V. Linek, P. Bene, J. Sinkule, T. Moucha, Non-ideal pressure step
method for k
L
a measurement, Chem. Eng. Sci. 48 (1993) 15931599.
[27] J. Llorens, C. Mans, J. Costa, Discrimination of the effect of sur-
factants in gas absorption, Chem. Eng. Sci. 43 (1988) 443450.
[28] S.A. Zieminski, R.C. Whittemore, Behavior of gas bubbles in aqueous
electrolyte solutions, Chem. Eng. Sci. 26 (1971) 509520.
[29] V. Macho n, V. Linek, Effect of salts on the rate of mass transfer
across a plane interface between a gas and mechanically agitated
aqueous solutions of inorganic electrolytes, Chem. Eng. J. 8 (1974)
5361.
[30] K.R. Westerterp, L.L. Van Dierendonck, J.A. de Kraa, Interfacial
areas in agitated gasliquid contactors, Chem. Eng. Sci. 18 (1963)
157176.
[31] A.A. Kulkarni, J.B. Joshi, Simultaneous measurement of ow pattern
and mass transfer coefcient in bubble columns, Chem. Eng. Sci.
59 (2004) 271281.
[32] V. Linek, P. Bene, V. Vacek, F. Hovorka, Analysis of differences
in k
L
a values determined by steady-state and dynamic methods in
stirred tanks, Chem. Eng. J. 25 (1982) 7788.
Chemical Engineering and Processing 44 (2005) 131137
Separation of n-hexaneethyl acetate mixtures by azeotropic batch
distillation with heterogeneous entrainers
I. Rodriguez-Donis
a
, J. Acosta-Esquijarosa
a
, V. Gerbaud
b,
,
E. Pardillo-Fondevila
a
, X. Joulia
b
a
Centro de Quimica Farmaceutica, Ave. 200 y 21 Atabey Apdo. 16042, Playa, C. Habana, Cuba
b
Laboratoire de Gnie Chimique UMR CNRS 5503, ENSIACETUPSCNRS, BP1301, 5 rue Paulin Talabot, 31106 Toulouse Cedex 1, France
Received 26 February 2004; received in revised form 4 May 2004; accepted 14 May 2004
Available online 2 July 2004
Abstract
In this article, a systematic study of the separation of the n-hexaneethyl acetate mixture with an entrainer by heterogeneous azeotropic
batch distillation is performed. Based upon the thermodynamic behaviour of the ternary mixtures, potential entrainers partially miscible with
one or two original azeotropic components are chosen. In all cases, the entrainer adds a heterogeneous binary or ternary azeotrope that is the
lowest boiling point in the ternary diagram. Therefore, it leaves the column by the overhead streamwhich is subcooled to get two liquid phases
in the decanter. The phase with the highest amount of the original component is removed as distillate product whereas the entrainer-rich phase
is continuously reuxed to the column. Considering methanol, acetonitrile, water and nitromethane as heterogeneous entrainers, screening was
performed based on the composition of the unstable heteroazeotropic mixture, the ratio of both liquid phases in the condensed top vapour and the
purity of the distillate product determined by the liquidliquid envelope at the decanter temperature. The process feasibility analysis is validated
by using rigorous simulation with the batch process simulator ProSimBatch. Simulation results are then corroborated in a bench experimental
column for the selected entrainer, showing several advantages of heterogeneous batch distillation (HBD) compared to homogeneous systems.
2004 Elsevier B.V. All rights reserved.
Keywords: Batch distillation; Heterogeneous entrainer; Azeotropic distillation
1. Introduction
Batch processes are again becoming important because
of recent expansion of the pharmaceutical and speciality
chemical industries in many developing countries. Besides,
recycling of liquid waste streams is becoming a key issue in
all processing plants to meet stricter environmental regula-
tions. Batch distillation is a widely used separation method
in batch processes because of its inherent operational ex-
ibility due to some extent to its time dependent operation.
The separation of azeotropic mixtures or close boiling com-
ponents is a challenging task in many chemical processes
as it is impossible using a single conventional column or
as a pressure swing distillation is an uneconomical process.
Alternative techniques based on the addition of an auxiliary

Corresponding author. Tel.: +33 5 62 88 58 26;


fax: +33 5 62 88 58 29.
E-mail address: Vincent.Gerbaud@ensiacet.fr (V. Gerbaud).
substance in the original mixture have been developed and
are usually known as azeotropic and extractive distillation.
Process synthesis and design of these non-conventional
distillation processes proceed in two steps. The rst
stepprocess synthesisis the selection of one or more
candidate entrainers along with the computation of ther-
modynamic properties like residue curve maps that help
assess many column features such as the adequate column
conguration and the corresponding product cuts sequence.
The second stepprocess designinvolves the search for
optimal values of batch distillation parameters such as the
entrainer amount, reux ratio, boiler duty and number of
stages. The complexity of the second step depends on the
solutions obtained at the previous level, because efciency
in azeotropic and extractive distillation is largely deter-
mined by the mixture thermodynamic properties that are
closely linked to the nature of the entrainer. Hence, we
have established a complete set of rules for the selection of
feasible entrainers for the separation of non ideal mixtures
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.006
132 I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137
by homogeneous azeotropic batch distillation [1]. In con-
trast to batch homogeneous distillation processes, studies
on heterogeneous batch distillation (HBD) were scarce
[24] until a new method for the synthesis of heterogeneous
distillation in a batch rectier was dened and a complete
set of rules for the selection of heterogeneous entrainers for
the separation of non-ideal mixtures was established [5,6].
Simulation and experimental results have shown key dif-
ferences between homogeneous and heterogeneous batch
distillation [6]. Indeed, some advantages of heterogeneous
batch processes compared to homogeneous systems are: (i)
more suitable candidate entrainers and hence, more design
alternatives for the separation of non-ideal mixtures, (ii)
simplied distillation sequences thanks to the liquidliquid
phase split occurring in some parts of the column and in the
decanter, (iii) the addition of a smaller amount of entrainer to
the original mixture, (iv) the use of more exible reux poli-
cies through any combination of the entrainer-rich phase and
the distillate-rich phase, (v) the still path can cross unstable
separatrices allowing the separation of components located
in different basic distillation regions. The overall cost re-
duction generated by these advantages makes heterogeneous
batch distillation an attractive option for the separation of
non-ideal mixture generated by specialty industries.
The aim of this paper is to study the performance of het-
erogeneous batch distillation for the separation of the mix-
ture n-hexaneethyl acetate. This azeotropic mixture is often
used in pharmaceutical industries for separating bioactive
substances by liquid chromatographic processes. The sepa-
ration of this mixture by batch distillation was rst studied
by Rodriguez-Donis et al. [7], by using acetone as a homo-
geneous entrainer. As the separation by using a single batch
distillation process with acetone was not possible, a hybrid
process was proposed involving a homogeneous azeotropic
distillation column with acetone and a liquidliquid extrac-
tion column with water. To circumvent this complex pro-
cess, we propose in this paper a simpler process by using
heterogeneous azeotropic batch distillation. Based on en-
trainer selection rules established by Rodriguez-Donis et al.
[5], which consider the thermodynamic behaviour of mul-
ticomponent mixtures, an initial list of potential entrainers
partially miscible with one or both original azeotropic com-
Table 1
Potential heterogeneous entrainers for the separation of n-hexaneethyl acetate mixture with thermodynamics data from Gmehling et al. [8]
Entrainer bp Binary azeotrope with
n-hexane (68.7

C)
Binary azeotrope with
ethyl acetate (77.1

C)
Ternary azeotrope
(

C) x
molar
a
T (

C) x
molar
a
T (

C) x
molar
a
T (

C)
Methanol 64.7 0.5100
b
50.6 0.708 62.05
Acetonitrile 81.5 0.5755
b
54.4 0.6730 76.1
Water 100.0 0.7860
b
61.8 0.7000
b
70.5 0.5764
b
60.7
0.2558
Nitromethane 101.2 0.7271
b
62.0 Zeotropic
a
The mole fraction indicated is that of the lightest component in the mixture.
b
Heteroazeotrope.
ponents are selected. From the analysis of residue curve map
features, in particular the decanter phases composition and
the distillation region separatrices curvature, the sequence of
products to be drawn from the rectier column can be pre-
dicted. The number of distillation tasks in the whole process
is set by the products targeted purity.
2. Selection of a suitable entrainer for the separation
of n-hexaneethyl acetate mixtures by heterogeneous
azeotropic batch distillation
Liquidvapour equilibrium data for the binary mixture
n-hexaneethyl acetate have been scarcely reported in the
scientic literature [8]. Those authors reported a mini-
mum boiling temperature azeotrope with an average mole
fraction of n-hexane equal to 0.6565 and a temperature
of 65.15

C at 101.3 kPa. A more detailed experimental


study, at atmospheric pressure, of liquidvapour equilib-
rium was performed by Acosta et al. [9]. Their estimation
of the azeotropic mixture corresponds to a temperature of
64.85

C with a molar fraction of n-hexane of 0.657 with


an average experimental error for the temperature and com-
position measurements of 0.02

C and 0.003 mole fraction,


respectively.
The candidate homogeneous and heterogeneous en-
trainers are investigated applying the rules reported by
Rodriguez-Donis et al. [1,5]. The existence of binary and
ternary azeotropes is sought by using literature data [8]
and by simulating liquidvapour and liquidliquidvapour
equilibrium by using the BibPhyAddIn Macro in an Excel
spreadsheet [10]. Conrming the need for a complex hybrid
process in case a homogeneous entrainer like acetone is
used [7], no homogeneous entrainer matching the reported
rules has been found whereas some suitable heteroge-
neous entrainers have been identied; the composition and
temperature of their azeotropic mixtures with the original
components are displayed in Table 1 [8].
As seen in Table 1, the boiling temperature of the pos-
sible entrainers is either lower or higher than that of the
original components. All candidates form a binary hetero-
geneous azeotrope with n-hexane. Methanol and acetonitrile
I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137 133
Table 2
Binary interaction parameters for NRTL and UNIQUAC models
Binary parameters A
ij
A
ji
a
ij
Binary coefcients for NRTL model (cal/mol)
n-Hexaneethyl acetate 720.731 33.844 0.1521
n-Hexaneacetonitrile 924.616 1212.64 0.2000
Ethyl acetateacetonitrile 580.783 253.247 0.1997
Binary coefcients for UNIQUAC model (cal/mol)
n-Hexaneethyl acetate 371.892 133.73
n-Hexanemethanol 1451.225 7.831
Ethyl acetatemethanol 770.688 147.777
n-Hexanewater 2292.190 324.669
Ethyl acetatewater 768.058 214.271
n-Hexanenitromethane 1080.270 7.335
Ethyl acetatenitromethane 181.416 42.973
add a homogeneous minimum boiling azeotrope with ethyl
acetate. Nitromethane forms a zeotropic mixture with ethyl
acetate, whereas water is the only component that forms an
additional heterogeneous azeotrope with ethyl acetate and
also adds a heterogeneous ternary azeotrope to the result-
ing mixture. Experimental liquidvapour equilibrium data
are available for all binary entrainer- ethyl acetate mixtures
[11]. However, for the binary system entrainer-n-hexane that
exhibits a signicant miscibility gap, the only experimental
thermodynamic information is given in Table 1, except for
methanol for which liquidvapour equilibrium data were re-
ported by Gmehling et al. [11]. All previous experimental
data are used to regress binary interaction parameters for the
UNIQUAC and NRTL models with ChemCAD IV minimiz-
ing the mean square deviation between the experimental and
calculated mole fraction in the liquid-phase (Table 2). Then,
ternary liquidliquidvapour equilibrium is determined by
using these binary coefcients with BibPhyAddIn [10]. As
a result of the use of these binary interaction parameters
regressed on binary systems to predict the behavior of a
ternary system, we have detected the existence of the ex-
perimental heterogeneous ternary azeotrope in the mixture
n-hexaneethyl acetatewater, which has the lower boiling
point in the system (unstable node).
For the synthesis of heterogeneous batch distillation
the liquidliquid envelope at the decanter temperature is
considered in addition to the residue curve map. There-
fore, the binary interaction parameters used in predicting
liquidliquid equilibrium are estimated from binary hetero-
geneous azeotrope or liquidliquid equilibrium data [8,10].
Table 3 shows the calculated purity of original compo-
nents in each phase split at 25

C for all heterogeneous


azeotropes reported in Table 1. The thermodynamic mod-
els and binary coefcients used in the calculation of the
liquidliquidvapour equilibrium, liquidliquid equilibrium
at 25

C and the separatrices are reported in Table 2.


Fromthe thermodynamic information given in Tables 13,
the residue curve maps are drawn in Fig. 1 for each entrainer
and the batch distillation task sequence required to perform
the separation of the original components is deduced using
published rules [5]. The residue curve maps of the ternary
Table 3
Purity of n-hexane-rich phase (x
I
) and entrainer-rich phase (x
II
) at 25

C
Entrainer Binary azeotrope with
n-hexane
Ternary azeotrope
x
I
nhexane
x
II
nhexane
x
I
x
II

Methanol 0.9152 0.1198 0.5346


Acetonitrile 0.9176 0.0605 0.3991
Water 0.7028 0.2481 0.1265
0.1754 0.0324
Nitromethane 0.9356 0.0893 0.2464 -
mixture n-hexaneethyl acetate-entrainer displayed in Fig. 1
indicate the temperature and the topological stability of all
singular points. As the heteroazeotropes are unstable nodes,
they are recovered in the decanter where the phase rich in
one of the original components L
W
is removed as distillate,
whereas the other original component is retained into the
still at the process end. In some cases, the composition of the
phase rich in original component dened by the liquidliquid
equilibrium at the decanter temperature may not meet purity
requirements; then, further distillation tasks are needed.
One of the key thermodynamic parameter in heteroge-
neous batch distillation is the decanter split ratio . It is de-
ned by the position of the heteroazeotrope composition on
its liquidliquid tie line at the decanter temperature or alter-
natively by the mole ratio of the entrainer-rich phase L
R
to
the overall liquid phase L
o
into the decanter as follows:
L
R
= L
o
(1)
Then, the molar quantity of entrainer-weak/original
component-rich phase L
W
is given by equation (2):
L
W
= (1 )L
o
(2)
As stated by Rodriguez-Donis et al. [6], the reux policy
to be used is strongly inuenced by the split ratio at the
decanter. If the liquid reux needed at the top of the column
is lower than L
R
, then the distillation can be performed by
using only the reux of entrainer-rich phase. Otherwise, the
separation of original components requires the reux of a
combination of both decanted phases.
Nitromethane shows the simplest residue curve map with
one unstable curved separatrix dividing the triangle in two
basic distillation regions. Methanol and acetonitrile give rise
two binary azeotropic mixtures and three distillation regions
that are bounded by two unstable curved separatrices. Water
shows the most complicated residue curve maps, due to the
presence of a ternary azeotrope and a miscibility gap with
both the n-hexane and the ethyl acetate component. In all
four cases, the heteroazeotrope (binary or ternary) has the
lowest boiling temperature of the system. As it can be seen in
Table 3, all entrainers except water provide the n-hexane-rich
phase L
W
as distillate product with a purity better than 0.91.
Water is not a desirable entrainer because of the existence
of ternary azeotrope whose n-hexane-rich phase has a water
purity much lower (0.70). Considering in Table 3 the split
134 I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137
Fig. 1. Residue curve maps of ternary systems n-hexaneethyl acetate-heterogeneous entrainer.
ratio at the decanter (inside the range 0.30.6) and the
purity of both liquid phases at 25

C, methanol and acetoni-


trile are better entrainers than nitromethane. Methanol is a
light entrainer that would result in a ternary system with a
low boiling temperature that in turn will likely increase the
cost of the condenser cooling system and the make-up of
fresh entrainer to be added in the next batch. Besides, the
mole ratio of the product-to-be n-hexane in the methanol
rich phase (x
II
nhexane
in Table 3) is higher than the one cal-
culated for the acetonitrile. Therefore, acetonitrile seems the
best entrainer for the separation of n-hexaneethyl acetate
by heterogeneous azeotropic batch distillation.
3. Simulation of the separation of the mixture
n-hexaneethyl acetate by heteroazeotropic batch
distillation using acetonitrile
We study the separation of n-hexaneethyl acetate mix-
ture by using acetonitrile as a heavy heterogeneous entrainer.
The simulation of the process is performed with the batch
process simulator ProSimBatch [10]. It enables to evaluate
operational parameters like the entrainer amount that are
not provided by the feasibility and synthesis analysis The
column model consists of usual plate by plate Material bal-
ance, Equilibrium, Summation of fractions and Heat balance
(MESH) equations which are solved for the whole column,
decanter included and taking into account the liquidliquid
phase split. Numerical treatment of the Differential Alge-
braic Equation (DAE) system and discrete events handling is
performed with DISCo, a numerical package for hybrid sys-
tems with a DAE solver based on Gears method. The col-
umn technical features and operating conditions are shown in
Table 4. A sequence of two operational batch steps, namely
Table 4
Operating conditions of heteroazeotropic batch distillation with acetonitrile
Parameters Value
Initial charge (mol) 12.2
Initial composition (molar) 0.705/0.173/0.122
Number of theoretical trays 50
Operating pressure (atm) 1
Holdup on the trays (mol) 0.0072
Tray efciency 1
Vapour ow rate (mol/h) from 9.7 to 7.5
Extra entrainer-rich phase (mol/h) 1.2
Decanter holdup (mol) 3.4
Decanter temperature (

C) 25
Distillation time (h) 4.8
0.3991
1.01
0.7
I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137 135
Fig. 2. Simulation and experimental results of the separation of n-hexane
ethyl acetate by using batch distillation with acetonitrile as a heavy
heterogeneous entrainer.
an innite reux step and a distillate removal step described
in detail below are simulated.
The amount of heterogeneous entrainer to be added to
the original binary mixture is established from material bal-
ances considering that the entrainer composition in the still
should be less than 0.01 at the end of innite reux batch
step 1. Simulation shows that this composition is reached
after two hours, which is the end of batch step 1. It should
be noted that the initial ternary composition F is placed in
distillation region I that is not limited by the vertex of ethyl
acetate (see Fig. 2). Therefore, in order to reach the vertex
of ethyl acetate located in region II at the end of the distil-
lation process, the still path must cross the unstable separa-
trix connecting the heteroazeotrope with the homoazeotrope
n-hexaneethyl acetate. As established by Rodriguez-Donis
et al. [6], such a crossing is possible with different reux
policies, either the reux of entrainer-rich phase or a com-
Table 5
Simulation and experimental results (x
1
: n-hexane; x
2
: ethyl acetate; x
3
: acetonitrile)
Simulation results Experimental results
Quantity (mol) Molar fraction Quantity (mol) Molar fraction
Initial mixture 12.2 x
1
0.705 12.2 x
1
0.705
x
2
0.173 x
2
0.173
x
3
0.122 x
3
0.122
n-Hexane-rich phase (distillate + decanter) 8.6 x
1
0.951 8.4 x
1
0.968
x
2
0.002 x
2
0.001
x
3
0.047 x
3
0.031
Acetonitrile-rich phase (decanter) 1.1 x
1
0.107 0.99 x
1
0.082
x
2
0.157 x
2
0.190
x
3
0.736 x
3
0.728
Final still content (ethyl acetate) 1.7 x
1
0.001 1.6 x
1
0.001
x
2
0.992 x
2
0.991
x
3
0.007 x
3
0.008
Ethyl acetate recovery yield 81.04% 75.1%
n-Hexane recovery yield 95.1% 94.5%
Acetonitrile recovery yield 54.5% 48.4%
bination of both decanted liquid phases. In this example, the
distillate removal step (batch step 2) is performed reuxing
a combination of entrainer-rich L
R
and entrainer-weak L
W
phases coming from the decanter, the reux policy 2.3 of
Rodriguez-Donis et al. [6]. As in Rodriguez-Donis et al. [6]
the acetonitrile-rich phase L
R

and n-hexane-rich phase L


W

reux ow rates are a fraction and of the amount


of each phase (L
R
and L
W
) contained in L
o
, the condensed
vapour ow rate entering the decanter. Table 4 displays the
and values that enable to withdraw the binary heteroge-
neous azeotrope during the whole distillation process. The
process requires that the reux of the acetonitrile-rich phase
is larger ( >1) than the acetonitrile quantity in the vapour
ow rate leaving the column top L
o
, while the amount of
n-hexane to be recycled at the top of the column is deter-
mined by the calculated value. > 1 means that drying
of the acetonitrile-rich phase in the decanter during opera-
tion can occur and care should be taken to avoid it in the
denition of an optimal reux policy.
Fig. 2 shows the simulation results, the still path and the
column liquid prole, at the end of batch step 1 (innite
reux) and at the beginning and end of batch step 2 (dis-
tillate removal). Table 5 displays signicant process values.
As can be seen in Fig. 2 and Table 5, only 12.2% of acetoni-
trile is added to the initial azeotropic binary mixture and the
resulting ternary mixture lies in the distillation region I. it
has been observed [6] that heterogeneous batch distillation
requires less entrainer than homogeneous batch distillation.
The liquid prole in the column at the end of the innite
reux operation links the still composition F
R
to the bi-
nary heteroazeotrope n-hexaneacetonitrile in the rst top
tray of the column through the vapour line. Then, applying
the reux policy detailed in Table 4, distillate removal starts.
Five minutes after the beginning of the distillation step,
the column liquid prole changes suddenly and crosses the
136 I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137
unstable separatrix due to the recycle of both decanted liquid
phases. The reason is that the liquid phase leaving the rst
tray of the column top has a composition (x
T
) inbetween the
heteroazeotrope and the acetonitrile-rich phase composition
and is located in a distillation region (III) different from re-
gion (I) which contains the still composition at innite reux
F
R
. Consequently, the simulated still path crosses the un-
stable separatrix separating distillation regions I and II (full
black line in Fig. 2). At the end of batch step 2, the still
path reaches the ethyl acetate vertex when all n-hexane is
drawn as distillate product. The operating parameters and
are set to obtain the binary heteroazeotrope at the column
top during the whole step 2. Therefore, the split ratio is
kept constant. The n-hexane-rich phase is drawn as distillate
product whereas the acetonitrile-rich phase can be recov-
ered at the decanter at the end of the step 2. Table 5 reports
the simulated purity and recovery yield of all components.
Simulation computation time is typically less than 4 min on
a Pentium III-based computer.
4. Experimental verication of the separation of the
mixture n-hexaneethyl acetate by heteroazeotropic
batch distillation in a bench scale rectication column
The technical features of the column and the operating
conditions are similar for the experiments and the simula-
tions. The glass SCHOTT distillation column has a total
height of 1.7 m and an internal diameter of 0.026 m. The to-
tal height is packed with stainless steel wire mesh rings of
3 mm 3 mm. The total packed height is considered to be
equivalent to 45 theoretical plates and the total liquid hold
goes up of 52.5 ml. The condenser is subcooled at 25

C and
total liquid reux is provided at the top of the column by a
solenoid valve. The open/close time of the solenoid valve is
dened considering that 70% of the condensed vapour is re-
turned to the column. The 30% remaining condensed vapour
is collected into the decanter (70 ml) where the light phase
is drawn as distillate (n-hexane-rich phase) whereas a por-
tion of the heavy entrainer-rich phase is recycled to the top
of the column with a peristaltic pump.
Experiments taking into account these operating condi-
tions (Table 4) are carried out in the bench distillation col-
umn and the results are presented in Fig. 2 and Table 5.
Analysis of the still composition is performed every 0.5 h.
The analysis of purity of the light and heavy phases in the
decanter and of the distillate product is made at the end of
the process (4.8 h). The acetonitrile, n-hexane and ethyl ac-
etate compositions are determined using a FISONS HRGC
chromatograph with a FID detector and with a column HP
INNOWax (30 m 0.53 mm, 0.25 m lm). The injector
and detector temperature are held at 180

C while the col-


umn temperature is set equal to 65

C. Prior calibration is
done using internal standards; the samples were prepared by
weighing on a SARTORIUS BP 211D balance with a pre-
cision of 10 mg. The standard accuracy and reproducibility
in the composition for all components are found to be 0.1
and 0.3%, respectively.
As shown in Fig. 2, the still path obtained experimentally
(circles) is in excellent agreement with the still trajectory
calculated by simulation. The selected reux policy permits
the still path to cross the separatrix into another distillation
region than the initial feed region. Hence, the still path is
able to reach the ethyl acetate vertex and this component re-
mains pure into the still at the end of the distillate removal
step. Such a behavior is not possible with a homogeneous
entrainer that gives rise to a similar residue curve map be-
cause the distillation process is restricted to the distillation
region where the initial composition of the mixture is lo-
cated. In this case, ethyl acetate could not be obtained as an
isolated product.
Table 5 displays the quantity and composition of the main
streams of the process. n-Hexane is drawn of as distillate
product with a molar purity of 96.8%. Another binary batch
distillation step is required to obtain n-hexane with a high-
est purity. Ethyl acetate with 99.6% of purity is recovered in
the still at the end of the process, while the acetonitrilerich
phase obtained from decanter can be reused in the next batch
distillation process. The make-up of fresh entrainer for the
next cycle of n-hexaneethyl acetate separation is approxi-
mately 52%.
5. Conclusions
Acetonitrile was chosen as a suitable heterogeneous
entrainer for the separation of the azeotropic mixture
n-hexaneethyl acetate by heteroazeotropic batch distil-
lation from the comparison of the residue curve maps of
several candidate entrainers. We have conrmed the feasi-
bility of the heteroazeotropic batch distillation via simula-
tion and by experimentation in a bench batch distillation
column. Good agreement was obtained between simulation
and experimental results. The novel process with a single
heterogeneous distillation column shows a great potential
over a reported hybrid process with a homogeneous distilla-
tion column and a liquidliquid extraction column. Indeed,
the consumption of entrainer is low and the separation of
all three components can be performed in a single batch
distillation step using the dened reux policy. The ethyl
acetate purity is high enough to enable its recycling in a
chromatography process whereas the n-hexane-rich phase
must be puried in a simple non azeotropic batch distillation
process.
References
[1] I. Rodriguez-Donis, V. Gerbaud, X. Joulia, Entrainer selection rules
for the separation of azeotropic and close boiling temperature mix-
tures by homogeneous batch distillation, Ind. Eng. Chem. Res 40
(2001) 27292741.
I. Rodriguez-Donis et al. / Chemical Engineering and Processing 44 (2005) 131137 137
[2] R. Dssel, J. Stichlmair, Separation of azeotropic mixtures by batch
distillation using an entrainer, Comp. Chem. Eng. 19 (1995) s113
s118.
[3] J. Khler, H. Haverkamp, N. Schadler, Zur diskontinuierlichen Rekti-
kation azeotroper Gemische mit Hilfsstoffeinsatz, Chem. Ing. Tech.
67 (1995) 967971.
[4] S. Skouras, S. Skogestad, Separation of ternary heteroazeotropic
mixtures in a closed multivessel batch distillation-decanter hybrid,
Chem. Eng. Proc. 43 (2004) 291304.
[5] I. Rodriguez-Donis, V. Gerbaud, X. Joulia, Heterogeneous entrainer
selection rules for the separation of azeotropic and close boiling
temperature mixtures by heterogeneous batch distillation, Ind. Eng.
Chem. Res. 40 (2001) 49354950.
[6] I. Rodrguez-Donis, V. Gerbaud, X. Joulia, Feasibility of heteroge-
neous batch distillation, AIChE J. 48 (2002) 11681178.
[7] I. Rodriguez-Donis, U. Juregui-Haza, E. Pardillo-Fondevila, Sepa-
ration of n-hexane-ethyl acetate mixtures by azeotropic distillation,
Latin Am. Appl. Res. 29 (1999) 119127.
[8] J. Gmehling, J. Menke, J. Krafczyk, K. Fischer, Azeotropic Data,
Verlag Chemie, Weinheim, 1994.
[9] J. Acosta, A. Arce, J. Martnez-Ageitos, E. Rodil, A. Soto, Vapour
liquid equilibrium of the ternary system ethyl acetate + hexane
+ acetona at 101.32 kPa, J. Chem. Eng. Data 47 (2002) 849
854.
[10] S.A. ProSim, 2000, http://www.prosim.net, Reference Manual.
[11] J. Gmehling, U. Onken, VapourLiquid Equilibrium Data Collection,
DECHEMA Chemistry Data Series, vol. 1, 12 parts, DECHEMA,
Frankfurt am Main, 1982.
| |
| |
CHEMICAL ENGINEERING AND PROCESSING
Editors:
Prof. Dr.-lng. A. Go rak
Universitat Dortmund
Lehrstuhl fur Thermische Verfahrenstechnik
Fachbereich Chemietechnik
Emil-Figge-Strasse 70
D-44221 Dortmund
Phone: 49 721 608 2391
Fax: 49 721 608 3490
e-mail: gorak.CEP@ct.uni-dortmund.de
Prof. Dr.-Ing. A. Wild
Laboratoire des Science du Genie Chimique
Groupe ENSIC
1 rue Grandville
BP 451
F-54001 Nancy
France
Phone: 49 231 755 2323
Fax: 49 231 755 3035
e-mail: Gabriel.Wild@ensic.inpl-nancy.fr
Founding Editor:
Prof. Em. Dr.-lng. Dr. h.c. mult. E.-U. Schlunder, Universitat Karlsruhe, Germany
Honorary Editor:
Dr. lng. V. Gnielinski, Universitat Karlsruhe, Germany
Editorial Board:
P. Aptel, Universite Paul Sabatier, France
P.I. Barton, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA
P. Bekiaroglou, University of Thessaloniki, Thessaloniki, Greece
P. Burghardt, Instytut Inzynierii Chemicznei l, Gliwice, Poland
K.J. Carpenter, Syngenta Global Specialist Technology, Hudderseld, UK
E. Drioli, Universita Della Calabria, Italy
J. Gyenis, Hungarian Academy of Sciences, Veszprem, Hungary
K. Jelemensky, Slovak Technical University, Bratislava, Slovak Republic
J.T.F. Keurentjes, Eindhoven University of Technology, The Netherlands
B. Kraushaar-Czarnetzki, University of Karlsruhe, Karlsruhe, Germany
F. Larachi, Universite Laval, Canada
R. Marr, Technische Universitat Graz, Graz, Austria
D. Mewes, University of Hanover, Hanover, Germany
H. Mu ller-Steinhagen, German Aerospace Centre, Stuttgart, Germany
K.D.P. Nigam, Indian Institute of Technology Delhi, New Delhi, India
A. Olcay, Ankara University, Ankara, Turkey
L. Puigjaner, Universitat Politencnica de Catulunya, Barcelona, Spain
H. Schoenmakers, BASF AG, ZAT-L540, Ludwigshafen, Germany
S. Skogestad, Norwegian University of Science and Technology, Trondheim, Norway
A. Stankiewicz, DSM Research, The Netherlands
K. Sundmacher, Universitaet Magdeburg, Germany
J. Werther, Technische Universitat of Hamburg-Harburg, Hamburg, Germany
K.R. Westerterp, Twente University of Technology, Enschede, The Netherlands
Aims and Scope
Chemical Engineering and Processing is a journal intended for practising engineers in industry and academia. The aim of the published articles is to demonstrate
how novel discoveries, developments and theories may be used for improved analysis and design of chemical engineering equipment and processes. The Journal
offers space for articles on any branch of chemical engineering and is particularly concerned with mechanical, thermal, chemical and bioengineering processes and
also includes analyses relating to environmental chemical engineering and sustainable industrial development. Advanced knowledge on engineering fundamentals
and processes is presented in such a form that it can be readily implemented into practical application. From time to time, review papers will be published for
individual elds of activity to illustrate and summarise scientic and/or technological progress.
*1 The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper).
Printed in The Netherlands
Publication information: Chemical Engineering and Processing (ISSN 0255-
2701). For 2005, volume 44 is scheduled for publication. Subscription prices are
available upon request from the Publisher or from the Regional Sales Ofce
nearest you or from this journals website (http://www.elsevier.com/ locate/cep).
Further information is available on this journal and other Elsevier products
through Elseviers website: (http://www.elsevier.com). Subscriptions are
accepted on a prepaid basis only and are entered on a calendar year basis.
Issues are sent by standard mail (surface within Europe, air delivery outside
Europe). Priority rates are available upon request. Claims for missing issues
should be made within six months of the date of dispatch.
Author enquiries
Submissions
For information regarding the submission of manuscripts to Chemical
Engineering and Processing, please refer to the Instructions to Authors in the
back of this issue.
These instructions are also available on the World Wide Web: http://
www.elsevier.com/locate/cep
For detailed instructions on the preparation of electronic artwork, consult the
Elseviers author gateway at http://authors.elsevier.com
Other enquiries
Visit Elseviers Author Gateway (http://authors.elsevier.com) for the facility
to track accepted articles and set up e-mail alerts to inform you of when an
articles status has changed. The Author Gateway also provides detailed artwork
guidelines, copyright information, frequently asked questions and more.
Contact details for questions arising after acceptance of an article, especially
those relating to proofs, are provided after registration of an article for publication.
USA mailing notice: Chemical Engineering and Processing (ISSN 0225- 2701)
is published monthly by Elsevier B.V. (P.O. Box 211, 1000 AE Amsterdam, The
Netherlands). Annual subscription price in the USA US$1,965 (valid in North,
Central and South America), including air speed delivery. Periodical postage paid
at Jamaica, NY 11431.
USA POSTMASTER Send address changes to Chemical Engineering
and Processing, Publications Expediting Inc., 200 Meacham Ave, Elmont, NY
11003.
AIRFREIGHT AND MAILING in the USA by Publications Expediting Inc., 200
Meacham Avenue, Elmont 11003, NY 11003.
Orders, claims, and journal enquiries: please contact the Customer Service
Department at the Regional Sales Ofce nearest you:
Orlando: Elsevier, Customer Service Department, 6277 Sea Harbor Drive,
Orlando, FL 32887-4800, USA; phone: (+1) (877) 8397126 [toll free number
for US customers], or (+1) (407) 3454020 [customers outside US]; fax: (+1) (407)
3631354; e-mail: usjcs@elsevier.com
Amsterdam: Elsevier, Customer Service Department, PO Box 211, 1000 AE
Amsterdam, The Netherlands; phone: (+31) (20) 4853757; fax: (+31) (20)
4853432; e-mail: nlinfo-f@elsevier.com
Tokyo: Elsevier, Customer Service Department, 4F Higashi-Azabu, 1-Chome
Bldg, 1-9-15 Higashi-Azabu, Minato-ku, Tokyo 106-0044, Japan; phone: (+81) (3)
5561 5037; fax: (+81) (3) 5561 5047; e-mail: jp.info@elsevier.com
Singapore: Elsevier, Customer Service Department, 3 Killiney Road, #08-01
Winsland House I, Singapore 239519; phone: (+65) 63490222; fax: (+65)
67331510; e-mail: asiainfo@elsevier.com
Advertising Information: Advertising orders and enquiries can be sent to:
Elsevier, 84 Theobalds Rd, London, WC1X 8RR, tel: +44 (0) 20 7611 4117;
fax: +44 (0) 20 7611 4463; e-mail: k.barton@elsevier.com
Reprints
Reprint orders can be sent to currentreprints@elsevier.com, tel: +44 (0) 20 7611
4494; fax: +44 (0) 20 7611 4463
Chemical Engineering and Processing 44 (2005) 139140
Editorial
Special issue on conveying and handling
of particulate solids
Dear reader,
Already almost 1 year has own away since the Fourth
International Conference on Pneumatic Conveying and Han-
dling of Particulate Solids took place in Budapest, Hungary,
2730 May 2003. This was the very rst event that took
place in Europe followed by previous meetings held in Israel
(Herzilya 1995, Jerusalem 1997, Dead Sea 2000). In
addition to the habitually pleasant and friendly atmosphere
of these conferences, this latter was even greater from pro-
fessional point of view. During this 4 full days meeting,
more than 200 presentations were delivered by excellent sci-
entists. Among them, 20 plenary or keynote lectures, 120
oral presentations in four parallel streams of sessions, as
well as 60 posters gave detailed information on the research
and industrial practice carried out and applied recently in the
eld. In addition, the marvelous conference center of Ho-
tel Inter-Continental Budapest and its surroundings with the
historical district of the Hungarian capital, one of the most
beautiful part of the worlds heritage, offered more than suit-
able milieu for the meeting and satellite workshops.
The objective of the event was to bring together re-
searchers, practicing engineers, industrialist and students
from all over the world. The rich scientic and technical
program have attracted great attention: more than 220 par-
ticipants joined to the conference from 28 countries, most
of them from Germany, Israel, United Kingdom, Poland,
USA, Hungary, Australia, Japan, Norway and France.
Full text papers have been published in two thick vol-
umes of conference proceedings, having altogether more
than thousand pages. However, to ensure broader publicity,
the International Scientic Council (ISC) and the Interna-
tional Organizing Committee (IOC) have decided to select
the most suitable papers for publication in several interna-
tionally respected periodicals, too.
Chemical Engineering & Processing was among the jour-
nals best suited to the topics of the majority of papers sub-
mitted to the conference. Selection of papers was the task
of ISC and invited guest editors on basis of quality and con-
tent. All papers submitted for publication in this special is-
sue were peer-reviewed by two independent referees. After
reviewing and nal revisions, 27 papers were accepted for
publication. These papers give representative sample on the
conference topics, especially from respect to the scope of
this journal.
The rst, memorial paper written by Prof. David Mills
gives honour to the oeuvre of Prof. Predrag Marjanovic, one
of the most distinguished personalities in particle technol-
ogy, late member of ISC, who left us tragically not long ago.
The next three papers by M.J. San Jos et al., I. Descamps
et al., as well as J. Li et al. deal with gassolids systems and
disperse particle ow. Gas ow pattern in spouted bed, mod-
eling of particle take-off in turbulent ow, as well as solids
deposition during pneumatic conveying were discussed, re-
spectively. During pneumatic conveying various transforma-
tions of particles can also take place, as is shown in two
next papers: L. Frye and W. Peukert describe specic attri-
tion mechanism in dilute phase process, while I. Skuratovsky
et al. show the results of a numerical simulation study on
drying process during pneumatic conveying.
Particle production is an important process of particle
technology. The paper of J. Tth et al. discusses the results
of investigation on salting-out precipitation of glycine and
sodium chloride. In another paper, production of micronized
cocoa butter particles by supercritical process was reported
by J.-J. Letourneau et al.
Relatively big number of contributions has been selected
for this special issue dealing with processing particulate
materials. J. Fitzpatrick and L. Ahrn gave comprehensive
summary on food powder handling and processing. Microen-
capsulation of heat sensitive ne particles was reported by
H. Krber and U. Teipel in a novel uidized-bed process
using supercritical carbon dioxide as solvent for coating ma-
terial. Plasma spheroidization of ceramic particles was de-
scribed by Z. Karoly and J. Szpvlgyi, while thermal plasma
processing was applied by I. Mohai and J. Szpvlgyi for
treatment of particulate metallurgical wastes. M.J.R. Aguado
et al. studied conical spouted bed reactor for pyrolysis of
plastic wastes using sand particles as uidizing packing to
avoid agglomeration.
Energy demand of granulation process was studied by T.
Gluba, described in the next paper. Elimination of particle
aggregation during nanomilling was realized and described
by W. Peukert et al. Four more papers are dealing with size
reduction processes: milling, grinding, attrition and particle
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.001
140 Editorial / Chemical Engineering and Processing 44 (2005) 139140
degradation (R.K. Sahoo and D. Roach, A. Heim et al., V.
Mizonov et al., and P.B. Kis et al.). Separation and classi-
cation are inevitable processes in particle technology. M.
Shapiro and V. Galperin gave a review on various techniques
of separation of particles. The paper of W. McBride and S.
Keys have applied vertical Venturi separator to improve re-
cycling of automotive tyres.
A worldwide revival of research activities nowadays on
mixing, blending, homogenization processes of particulate
solids is evidenced by three papers. A new pulsated bottom
mixer and its investigation were described by M. Nemenyi
and A.J. Kovacs. Process dynamics and homogenization ef-
ciency of mammoth silos were studied by D.L. Schott et
al. Assessment of homogeneity in powder mixtures by elec-
trical capacitance measurement was proposed N. Ehrhardt
et al., as a good tool to examine segregation during particle
ow or the efciency of a mixer.
As a special topic, solids transport in a pilot-scale rotary
kiln was studied and outlined in a paper of N. Descoins et al.
Finally, two papers give examples on environmental aspects
of this eld. A nice overview on dust control and explosions
in solids handling plants are given P. Wypych et al. with
description of some own experimental results. Application
of uidized bed process for photo-catalytic degradation of
harmful gas components are described by T.H. Lim and S.D.
Kim.
We hope that this special volume offers a good cross-
section on the professional work of the conference for the
readers, giving useful information on the results achieved
recently in the eld, and will also stimulate the participation
at the next event to be held in 2006 in Rome.
April 2004.
Janos Gyenis
Research Institute of Chemical and Process Engineering
University of Veszprem, P.O. Box 158
Veszprem 8201, Hungary
Corresponding author. Tel.: +36 88 425 206
fax: +36 88 424 424
E-mail address: gyenis@mukki.richem.hu
(J. Gyenis)
Avi Levy
Ben-Gurion University of the Negev
Beer-Sheva, Israel
Available online 19 June 2004
Chemical Engineering and Processing 44 (2005) 141151
A review of the research work of Professor Predrag Marjanovi c
David Mills

Pneumatic Conveying Consultant, Old Wives Lees, Canterbury CT4 8BQ, UK


Received 6 August 2003; received in revised form 18 December 2003
Available online 20 June 2004
Abstract
Predrag Marjanovi c was born on 1 May, 1951 in Belgrade, Yugoslavia, where he lived until November 1992. He moved with his wife
Mara and daughter Nevena to Glasgow, Scotland, where he continued to reside throughout his time in the UK. I was supervisor of his Ph.D.
programme and colleague in Glasgow for about 5 years. Predrags involvement in research covered a diverse range of topics, including
hydraulic transport, uid mechanics, bulk material characterisation and hopper design. This review will mostly feature the work undertaken
for his Ph.D. and that associated with pneumatic conveying. A denitive list of Predrags references, however, is appended. Predrag died
suddenly on 14 September, 2001.
2004 Elsevier B.V. All rights reserved.
Keywords: Fluid mechanics; Hopper design; Vertical pipeline; Pneumatic conveying; Inclined pipeline; Pipeline bends; Rubber hose; Dense phase
1. Belgrade
Predrag obtained a 1st class Honours degree in Mechan-
ical Engineering in 1975 from the University of Belgrade.
He then joined the Fluid Mechanics Department of the Uni-
versity of Belgrade as a teaching assistant, where he was
employed for 18 years, rising to the position of Associate
Professor. During this time he obtained an M.Sc. with dis-
tinction in Process Engineering. His M.Sc. was on the Ex-
perimental Determination of GasSolids Friction Factors. In
these early years, he also published a number of papers on
hydraulic conveying with Professor M. Sasic.
I met Predrag in Belgrade in October 1979, being in-
troduced by Dr. Z. Bukerov from the University of Novi
Sad. Predrag expressed the desire to undertake a Ph.D. pro-
gramme on pneumatic conveying and particularly wanted
to investigate conveying vertically up in dense phase. There
was the opportunity of undertaking such a programme of
research at Thames Polytechnic (now the University of
Greenwich) in London, where I was a Senior Lecturer in
the Department of Mechanical Engineering, and Professor
Stan Mason was Head of Department. Predrag obtained
sabbatical leave from the University of Belgrade and came
to London early in 1982.

Tel.: +44 1227 732493; fax: +44 1227 732504.


E-mail address: drdavid.mills@btinternet.com (D. Mills).
2. Ph.D. programnme
A high-pressure (7 bar gauge) top discharge blow tank
having a 1 m
3
capacity was made available and a pipeline
was built on the outside wall of the building, which allowed
a 17 m vertical rise. With a need to re-circulate materials for
test purposes it was necessary to run the pipeline both up
and down the wall to complete the circuit. It was decided,
therefore, to investigate pneumatic conveying performance
in both the vertically up and vertically down sections of the
pipeline as it involved little additional effort in building and
instrumenting the test facility for the added data. There was
also a lot of interest at the time in the pneumatic conveying
of various materials down vertical mine shafts. A sketch of
the test pipeline facility is provided in Fig. 1.
Two pipelines, in fact, were built, one of 2 in. nominal
(53 mm actual) bore and another of 3 in. nominal (81 mm ac-
tual) bore. This was to allow the inuence of pipeline bore to
be additionally investigated. The two pipelines ran alongside
one another and so the pipeline routing and geometry was
almost identical for the two pipelines. The pipelines were
built on the rear wall of the building and this had a re escape
access with a large platform. As a consequence, because of
the convenience, it was decided to install sight glasses in the
two pipelines in both the vertically up and vertically down
sections. I had full video facilities available from the UK
Department of Trade and Industry for the research work.
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.05.002
142 D. Mills / Chemical Engineering and Processing 44 (2005) 141151
Fig. 1. Test facility and pipeline employed.
In order to derive data for the individual sections of
pipeline in isolation from the total pipeline, all four sections
of vertical pipeline were provided with pressure tappings
along their entire length. In the vertically down sections of
pipeline, there were seven sets of pressure tappings and in
the vertically up sections, there were eight sets. A ring of
four pressure tappings was provided at every location and
these were inter-connected. Every pressure tapping was t-
ted with a lter pad and provided with a high-pressure air
purging facility, which was routinely operated after each
and every test run. Bearing in mind that this was 1982 and
that electronic pressure transducers were both in their in-
fancy and very expensive, mercury manometers were used
for all vertical pipeline pressure measurements.
Two typical sets of pressure measurement data for the
vertically down and vertically up sections of pipeline are
presented in Fig. 2. This shows the location of the pressure
tappings and their proximity to the various bends in the
pipeline. The data relate to the pneumatic conveying of a ne
grade of pulverised fuel ash. Five different bulk particulate
materials were investigated in the research programme, the
other four being barytes, bentonite, cement and uorspar.
All ve materials were capable of being conveyed in dense
phase and hence at low velocity.
Typical conveying data obtained for the total pipeline sys-
tems are presented in Figs. 36.
Several compressors, each capable of delivering 200 ft
3
/
min (0.095 m
3
/s) of free air at a pressure of 100 psig (7 bar
gauge) were available, together with a desiccant-type air
drier. For the 53 mm bore pipeline, one compressor was
used and for the 81 mm bore pipeline two were used.
Convergentdivergent choke ow nozzles were used to mea-
sure and control the ow rate of air used for conveying, as
well as to control the discharge rate of material fromthe blow
tank. The receiving hopper was mounted on load cells and
these were used to measure the conveyed material ow rate.
Fig. 2. Pressure gradient results for a ne grade of pulverised fuel ash.
The discharge limit of the blow tank used was about
26 tonne/h with the 53 mm bore pipeline and approximately
50 tonne/h with the 81 mm bore pipeline. Within this capa-
bility of the conveying facility, however, tests were carried
out with conveying line pressure drop values of well over
two bar and the materials were all capable of being con-
veyed at solids loading ratios of well over one hundred. A
minimum of fty individual tests were undertaken with ev-
ery material/pipeline bore combination, in order to draw the
various families of curves required.
There was no lower limit on pressure drop, material ow
rate or solids loading ratio that the test facility could operate,
Fig. 3. Conveying characteristics for bentonite in 53 mm bore pipeline.
D. Mills / Chemical Engineering and Processing 44 (2005) 141151 143
Fig. 4. Conveying characteristics for uorspar in 53 mm bore pipeline.
but Predrag chose to investigate dense phase conveying for
his Ph.D. study and he just did not have the time available
to extend the work down into the dilute phase range of con-
veying. As a consequence very few tests were undertaken
with a conveying line pressure drop below 08 bar.
3. Pressure gradient data
Data was recorded from the manometer banks from each
test and this was analysed in terms of a pressure gradient
in mbar/m. The resulting data was also plotted in the form
of conveying characteristics, and representative sets of data
obtained are presented in Figs. 710. In Figs. 7 and 8, data
obtained for vertically upward ow are shown, with cement
in the 53 mmbore pipeline in Fig. 7 and barytes in the 81 mm
Fig. 5. Conveying characteristics for cement in 81 mm bore pipeline.
Fig. 6. Conveying characteristics for pulverised fuel ash in 81 mm bore
pipeline.
bore pipeline in Fig. 8. As will be seen, almost the entire
dense phase conveying capability of the materials has been
covered in the programme of work.
Conveying data presented in this form clearly show
the capability of pneumatic conveying systems and the
inter-relating effects of pressure, material concentration
and pipeline bore, as well as air ow rate when designing
a system to convey a material at a given ow rate, over a
specied distance. Since there is generally a limit on air
supply pressure, a compromise has to be made between
solids loading ratio and pipeline bore.
Figs. 9 and 10 showdata obtained for vertically downward
ow in the 53 mm bore pipeline, with cement in Fig. 9 and
barytes in Fig. 10. From these two sets of data, it will be
Fig. 7. Pressure gradient for cement in 53 mm bore vertically up pipeline.
144 D. Mills / Chemical Engineering and Processing 44 (2005) 141151
Fig. 8. Pressure gradient for barytes in 81 mm bore vertically up pipeline.
seen that when the materials are conveyed at a solids loading
ratio of about 35, there is no pressure drop associated with
the conveying of the materials. Results with the other three
materials also showed the same pattern. At higher solids
loading ratios, the pressure gradient is negative and hence
there is a pressure recovery in downward ow. At lower
values of solids loading ratio, however, the pressure gradient
is positive, and hence by inference over the entire dilute
phase range of conveying, there will be a pressure drop.
4. Subsequent collaboration
Following Predrags return to the University of Belgrade,
and the award of his Ph.D. in 1984, Predrag maintained
Fig. 9. Pressure gradient for barytes in 53 mm bore vertically down
pipeline.
Fig. 10. Pressure gradient for pulverised fuel ash in 81 mm bore vertically
down pipeline.
his collaboration with myself and published numerous pa-
pers on the many aspects of his Ph.D. research programme,
including dense phase conveying, pipeline bore, material
inuences and blow tank performance. The collaboration
was continued with Predrag being involved in my contin-
uing research work, and this included papers on pipeline
orientation, pipeline material and the inuence of bends in
pneumatic conveying pipelines. The nature of the work and
the range of topics will be seen from the list of date ordered
references and publications at the end of this paper.
5. Scaling parameters
With my work on horizontal conveying, and Predrags
work on vertical conveying, using many of the same materi-
als, it was a natural development of our joint work to com-
pare data and derive scaling parameters for vertical pipelines
in terms of horizontal pipelines. In Fig. 11, data are pre-
sented on pressure gradients for the conveying of barytes in
a horizontal pipeline of 53 mm bore. Predrag had obtained
identical data for conveying vertically up and so to provide
a comparison of the two sets of results a rectangular grid
was placed on both sets of curves and the value of the ra-
tio of the two pressure gradients was noted. Fig. 12 shows
the value of the ratios of the vertical line pressure gradient
divided by the horizontal line pressure gradient, determined
for every grid point.
It will be seen that the relationship obtained covers a very
wide range of conveying conditions, although it is unfortu-
nate, in this case, that the data could not be extended to cover
dilute phase conveying also. This exercise was also carried
out with a number of other materials and very similar results
were obtained. In all cases, the ratio between vertical and
horizontal pressure gradients was in a very narrow band, at
D. Mills / Chemical Engineering and Processing 44 (2005) 141151 145
Fig. 11. Pressure gradient for barytes in 53 mm bore horizontal pipeline.
Fig. 12. Ratio of vertical to horizontal pressure gradient for barytes in
53 mm bore pipeline.
a mean value of about 2, over the entire range of conveying
conditions considered.
6. Pipeline bends
In my work for the Department of Trade and Industry in
the UK, I had generated a considerable amount of data with
a number of different materials conveyed through pipelines
of approximately 100 m in length but having differing num-
ber of bends. Predrag worked with me to analyse the data
and isolate the inuence of the straight pipeline so that the
effects of the bends could be determined. Recognising that
engineers design pneumatic conveying systems in different
ways, the data were analysed in order to provide values both
in terms of an equivalent length and in terms of an actual
pressure drop value.
Fig. 13. Equivalent length of bends.
Data obtained with cement and analysed in terms of an
equivalent length of straight horizontal pipeline are pre-
sented in Fig. 13. This is for 90

bends having a bend diam-


eter D, to pipe bore d, ratio of 24:1 in 53 mm bore pipeline
in horizontal plane. Almost identical results were obtained
when a similar analysis was carried out for the conveying of
barytes. A simple correlation in terms of the conveying line
inlet air velocity was not expected, but it was not possible
to determine any effect of the position of the bends in the
pipeline. Data obtained with barytes and analysed in terms
of a pressure drop across a bend are presented in Fig. 14.
Once again, almost identical results were obtained for the
conveying of cement. These data are for the same bends re-
ported in Fig. 13.
Fig. 14. Pressure loss for bends.
146 D. Mills / Chemical Engineering and Processing 44 (2005) 141151
7. Pipeline material
Another area of interest at this time was the transport of
cement and drilling mud powders, with a particular require-
ment for the transferring of these materials onto off-shore
drilling platforms. This involves the use of exible hose for
a signicant proportion of the pneumatic conveying pipeline
for the transport of these materials. There was, therefore, a
need to know how pressure drop through rubber hose com-
pared with the pressure drop through steel pipeline. Oil well
cement and barytes were the materials used for this partic-
ular investigation. 40 m of rubber hose of 53 mm bore was
available and so a 40 m long pipeline was built of 53 mm
bore steel pipeline and a full set of conveying characteris-
tics were obtained for each material. The rubber hose was
then strapped to the steel pipeline to create a pipeline hav-
ing exactly the same routing and number of bends and bend
geometries. Full sets of conveying characteristics were then
obtained for the two materials conveyed through this rubber
hose pipeline.
The conveying characteristics for the oil well cement con-
veyed through the steel pipeline are presented in Fig. 15.
Once again as wide a range of conveying conditions were
investigated as possible. Similar sets of conveying character-
istics were obtained for the barytes in the steel pipeline and
for both materials in the rubber hose pipeline. The rubber
hose was rated to a 10 bar capability.
A rectangular grid was constructed on each set of convey-
ing characteristics, as in the above analysis, and the ratio of
pressure drops was evaluated at corresponding grid points
of material and air ow rates. In Fig. 16, a comparison of
the pressure drop required to convey the cement through
the rubber hose, compared with that to convey the cement
through the steel pipeline, is presented. It will be seen that
Fig. 15. Conveying characteristics for cement in 40 m long steel pipeline.
Fig. 16. Comparison of pressure drop data for steel and rubber hose lines
for cement.
the percentage differences between the two pipeline mate-
rials follow a regular pattern. With increase in air ow rate
there is a signicant increase in the pressure drop required
for the rubber hose line. The lines drawn through the data
also approximate to lines of mean value of conveying air
velocity through the pipeline.
It is believed that this can be related to the differences in
coefcient of restitution between the conveyed particles and
the pipeline walls. On impact with the rubber, the particles
will be decelerated, since the rubber will absorb much of
the energy of impact. As a consequence, the particles will
have to be re-accelerated back to their terminal velocity. The
coefcient of restitution of the particles against the steel
pipeline wall will be very much lower. This effect is clearly
magnied by increase in velocity and explains why there is
little difference between the two pipeline materials in low
velocity dense phase conveying, but differ by 50% in high
velocity dilute phase conveying. The results obtained with
the barite were very similar.
8. Glasgow
In 1988, I and a number of colleagues moved to Glasgow
College (now Glasgow Caledonian University) to join Pro-
fessor Stan Mason who had moved there the previous year
and was later appointed Principal and Vice Chancellor of
the University. A major teaching aim was to provide a post-
graduate Masters Degree course in Bulk Solids Handling
in Glasgow. To help in the preparation and development of
this new and innovative venture, and early teaching on the
course, Pregrad obtained sabbatical leave for a year from the
University of Belgrade in 1990/91.
D. Mills / Chemical Engineering and Processing 44 (2005) 141151 147
Predrag joined Glasgow Caledonian University in 1992
as Senior Research Fellow and was appointed Professor in
1999. In Glasgow, Predrags research work expanded con-
siderably, with the supervision of Ph.D. programmes, con-
tract research and in the undertaking of consultancy work.
In 1998, for instance, his CV lists 13 publications. His
work in this period included major programmes on inclined
pipelines, bend losses, very high-pressure conveying sys-
tems, hopper design, on-site measurements, and material
degradation and segregation. I have concentrated on the rst
three of these topics as they represent the largest proportion
of Predrags published output.
9. Inclined pipelines
The performance of inclined pipelines has long been a
topic of mystery and contention and so Predrag set up an
experimental facility in the laboratory at Glasgow Caledo-
nian University in order to study the problem. A 100 m long
pipeline of 81 mm bore was built, having a section in the
central area which was 8 m long and that could conveniently
be adjusted to provide inclinations ranging from 20

to
+90

. Material was fed by means of a rotary valve, and a


Roots-type blower capable of delivering up to 033 m
3
/s of
free air was used, giving supercial air velocities from 10
to 20 m/s. A series of seven papers over a period from 1994
to 1997 were written specically on this subject, with var-
ious distinguished co-authors, as will be seen from the list
of publications.
A major feature of the work was the analytical modelling
of the inclined pipeline situation using computational uid
dynamics, and comparing the numerical predictions with the
experimental results. Test results obtained with 3 mm poly-
mer pellets are presented in Figs. 17 and 18. In Fig. 17, the
results are presented in terms of the difference between the
Fig. 17. Variation of pressure gradient difference with angle of inclination.
Fig. 18. Variation of normalised pressure gradient with angle of inclination.
total pressure gradient for an inclined pipe and that for a
horizontal pipe, under the same ow conditions, as a func-
tion of the angle of inclination. Data are presented for solids
loading ratios of 5, 10 and 20. It should be noted that the
pressure gradient for horizontal ow is also dependent upon
the solids loading ratio and, in addition, increases continu-
ously as the solids loading ratio increases.
In Fig. 18, the results are presented in terms of a nor-
malised pressure gradient, which is the ratio of the pressure
gradient for a particular angle of inclination divided by that
for the horizontal. Once again, solids loading ratios of 5, 10
and 20 were considered.
10. Pipeline bends
Pipeline bends are another area of particular interest
to pneumatic conveying engineers on which Predrag has
worked. A series of six papers over a period from 1995
to 2000 were written specically on this subject, working
mostly with the same group of co-authors as above. For
the majority of the work, very little new experimental work
was undertaken, for once again the main thrust of the work
was on modelling, and for this purpose two sets of existing
data were used.
One of the sets of data used was that for cement, obtained
by the author in a 50 m long pipeline of 53 mm bore that
contained 99

bends having a D/d ratio of 24:1. The other


set of data was for polyethylene pellets conveyed through a
50 m long pipeline of 80 mm bore that contained six ninety
degree bends having a D/d ratio of 85:1 and obtained later
from the laboratory in Glasgow. The modelling employed a
solids friction coefcient and the authors attempted to model
the experimental data.
The model was capable of separating the constituent ele-
ments of the total pipeline pressure drop and so the authors
148 D. Mills / Chemical Engineering and Processing 44 (2005) 141151
Fig. 19. Cement in 53 mm bore pipeline with nine bends.
Fig. 20. Polyethylene in 80 mm bore pipeline with six bends.
were able to demonstrate the proportion of the total pres-
sure drop that could be attributed to the bends. The results
for the cement are presented in Fig. 19 and those for the
polyethylene pellets in Fig. 20. Both materials were capable
of being conveyed in dilute as well as dense phase, although
at low velocity, the cement was conveyed in a sliding bed
mode of ow and the pellets in plug ow. Inuence of bends
on pressure drop required to convey materials in 50 m long
pipelines is shown in Figs. 19 and 20.
11. High-pressure conveying
Very high-pressure conveying is a subject of particular in-
terest of mine, and this was another topic on which Predrag
worked. A series of ve papers over a period from 1995 to
1997 were written specically on this subject. The test fa-
cility used was designed and built in the laboratory at Glas-
gow Caledonian University under my supervision, with test
work carried out in which material was conveyed from the
Fig. 21. High-pressure conveying facility and pipeline.
high-pressure blow tank and into a reception vessel main-
tained at pressures of up to 20 bar gauge. The blow tank
feeding the conveying system was rated at 25 bar and the
system was tested to 32 bar. The lter unit was located in-
side the pressurised reception vessel.
In the work undertaken by Predrag, crushed coal was
conveyed, and so nitrogen was used as the conveying gas.
Modelling of system performance was a major part of the
programme once again. The conveying pipeline employed
was 78 m long, of 49 mm bore and included eight 90

bends.
A sketch of the test facility and pipeline is given in Fig. 21.
A number of sets of test results obtained for different back
pressures and line pressure drops are presented in Fig. 22.
Minimumconveying air velocity was also determined dur-
ing the test programme. It was found that this velocity is in-
uenced by system pressure, with its value decreasing from
10.5 m/s at atmospheric pressure to 6 m/s at a pressure of
20 bar gauge. This effect was expected due to a signicant
increase of gas density at higher pressures.
Fig. 22. Conveying characteristics for coal at different back pressures.
D. Mills / Chemical Engineering and Processing 44 (2005) 141151 149
12. What could have come
We are naturally left wondering what might have been
achieved with so many more years that should have been his,
to spend on the loves of his life: his family and his research
work, and not forgetting the golf that he discovered in his
new home in Scotland. Thank you for being such a valued
colleague and friend. You will be long remembered by us
all.
Further reading
References and publications
[1] P. Marjanovi c, M. Sasic, Hydraulic transport of ash in thermo-
power stations, Proceedings of the 4th Yugoslav Thermal Symposium,
Belgrade, Yugoslavia, 1975 (in Serbian).
[2] M. Sasic, P. Marjanovi c, On the methods for calculation of hydraulic
transport and their reliability in practice. Part 1, Proceedings of the
5th International Conference on Hydraulic Transport of Solids in
Pipes, vol. 1, Hanover, Germany, 1978, pp. A5:61A5:76.
[3] M. Sasic, P. Marjanovi c, On the methods for calculation of hy-
draulic transport and their reliability in practice. Part 2, 5th Interna-
tional Conference on Hydraulic Transport of Solids in Pipes, vol. 2,
Hanover, Germany, 1978, pp. X2X3.
[4] P. Marjanovi c, Experimental determination of gassolids friction fac-
tor, M.Phil. Thesis, University of Belgrade, Yugoslavia, 1978 (in
Serbian).
[5] M. Sasic, P. Marjanovi c, Hydromechanics of gassolid systems, Jnl
Tehnika-Masinstvo, no. 6, Belgrade, Yugoslavia, 1979, pp. 19 (in
Serbian).
[6] P. Marjanovi c, J. Jovanovic, R. Askovic, Sur lecoulement dun
liquide conducteur dans la region dentree un tube circulaire en
presence dun champ magnetique. Publ. of Mathematical Institute
SANU, new series, vol. 3, no. 11, Belgrade, Yugoslavia, pp. 7580.
[7] M. Sasic, P. Marjanovi c. One more general approach for investigation
of hydro-mechanics of two-component systems, J. Powder Bulk
Solids Tech. 5 (1) (1981) 1319.
[8] M. Sasic, P. Marjanovi c, Non-isothermal compressible ow in pipes,
ZAMM 62 (1982) 226228.
[9] P. Marjanovi c, A comparative study of performance characteristics for
horizontal and vertical pneumatic conveying in pipelines, Proceedings
of the Pneumatech 1. PAC Conference Stratford-upon-Avon, UK,
1982.
[10] D. Mills, J.S. Mason, P. Marjanovi c, Pneumatic conveying vertically
down. Proceedings of the 8th Powder and Bulk Solids Conference,
Atlanta, USA, May 1983, pp. 546557.
[11] D. Mills, J.S. Mason, P. Marjanovi c, An analysis of the dense phase
pneumatic conveying of cement in vertical pipelines. Proceedings of
the 9th Powder and Bulk Solids Conference, Chicago, USA, May
1984, pp. 125147.
[12] D. Mills, J.S. Mason, P. Marjanovi c, The inuence of pipeline
bore on dense phase pneumatic conveying in vertical pipelines.
Proceedings of the 16th Yugoslav Congress on Theoretical and
Applied Mechanics, Becici, Yugoslavia, May/June 1984, pp. 261
269.
[13] P. Marjanovi c, An investigation of the behaviour of gassolids mix-
ture ow properties for vertical conveying in pipelines, Ph.D. Thesis,
CNAA, Thames Polytechnic, London, UK, 1984.
[14] D. Mills, J.S. Mason, P. Marjanovi c, The inuence of product type
on dense phase pneumatic conveying in vertical pipelines, Proceed-
ings of the Pneumatech 2, Powder Advisory Centre Conference,
Canterbury, UK, September 1984, pp. 193210.
[15] D. Mills, J.S. Mason, P. Marjanovi c, A comparison of pressure drops
in horizontal and vertical dense phase pneumatic conveying, Proceed-
ings of the 3rd International Conference on Pneumatic Conveying,
Pecs, Hungary, March 1985.
[16] D. Mills, P. Marjanovi c, J.S. Mason, An analysis of line pressure
gradient for dense phase vertical gassolids ow, Proceedings of the
GAMM Congress, Dubrovnik, Yugoslavia, April 1985.
[17] P. Marjanovi c, D. Mills, J.S. Mason, An investigation of high con-
centration vertical pneumatic conveying, J. Procesna Tehnika, vol. 1,
no. 2, Belgrade, Yugoslavia, December 1985, pp. 3949 (in Serbian).
[18] P. Marjanovi c, On the choking velocity for vertical pneumatic con-
veying, Proceedings of the 17th Yugoslav Congress of Theoretical
and Applied Mechanics, Zadar, Yugoslavia, 1986 (in Serbian).
[19] P. Marjanovi c, The design of vertical pneumatic conveying using
high-pressure blow tank system, Proceedings of the International
Colloquium on Modern Trends in the Development of Pneumatic
Conveying, Sarajevo, Yugoslavia, June 1986, pp. 4564.
[20] P. Marjanovi c, D. Mills, J.S. Mason, A method of calculating
ow parameters for pneumatic conveying in pipelines, Proceedings
of the Pneumatech 3, PAC Conference, Jersey, UK, March 1987,
pp. 143167.
[21] P. Marjanovi c, I. Vuskovic, L. Bodiroga, Hydraulic Transport of
granular activated carbon in water renery plant, Symposium on
Water Supply, Belgrade, Yugoslavia, 1987 (in Serbian).
[22] P. Marjanovi c. Bends in gassolids mixture ow in pipes a view to
the prediction of pressure loss, GAMM Congress, Vienna, Austria,
1988.
[23] P. Marjanovi c, Analysis of the change of ow parameters in dis-
continuous pneumatic conveying systems, Proceedings of the 18th
Yugoslav Congress of Theoretical and Applied Mechanics, Vrnjacka
Banja, Yugoslavia, 1988, pp. 8992 (in Serbian).
[24] P. Marjanovi c, Determination of performance characteristics of dilute
phase pneumatic conveying system, Proceedings of the 3rd Sympo-
sium of Process Industry Applications, Belgrade, Yugoslavia, 1988,
pp. 215223 (in Serbian).
[25] P. Marjanovi c, A comparison between pneumatic and hydraulic con-
veying, Proceedings of the 19th Yugoslav Congress of Theoretical
and Applied Mechanics, Ohrid, Yugoslavia, 1990, pp. 6166 (in Ser-
bian).
[26] P. Marjanovi c, D. Mills, J.S. Mason, The inuence of bends on the
performance of a pneumatic conveying system, Proceedings of the
15th Powder and Bulk Solids Conference, Chicago, USA, June 1990,
pp. 391399.
[27] P. Marjanovi c, D. Mills, J.S. Mason, The inuence of pipeline ma-
terial on the performance of pneumatic conveying systems, Proceed-
ings of the Pneumatech 4, Glasgow, UK, June 1990, pp. 453464.
[28] N. Barbalic, P. Marjanovi c, E. Dzaferovic, Z. Mujcinovic, A. Cam-
para, Measurement of P.F. ash pneumatic conveying system parame-
ters in thermo-power station Ugljevik-1, J. Procesna Tehnika, vol.
7, no. 1, Belgrade, Yugoslavia, 1991, pp. 2326 (in Serbian).
[29] P. Marjanovi c, M. Stanojevic, B. Todorovic, A. Vlajcic, The inu-
ence of variable operating conditions on the design and exploitation
of air-slide systems in thermo-power stations, Proceedings of the
RelPowFlo 2, EFChE Publication Series no. 96, Oslo, Norway, 1993,
pp. 659671.
[30] P. Marjanovi c, Modelling the transient behaviour of blow tank pneu-
matic conveying system. Powder handling and processing, Trans.
Tech. Pub. 5 (3) (1993) 219226.
[31] P. Marjanovi c, Pneumatic conveying in an inclined pipeline: inappro-
priate, unfortunate or wrong concept? Proceedings of the 19th Powder
& Bulk Solids Conference, Chicago, USA, May 1994, pp. 235251.
[32] P. Marjanovi c, D.J. Mason, Gassolid ows in an inclined pipeline,
Proceedings of the 1st International Particle Technology Forum, vol.
3, Denver, USA, August 1994, pp. 466471.
[33] P. Marjanovi c, Assessing the ow properties of powdered materials,
IMechE Seminar on Hoppers and Silos, London, UK, 1994.
150 D. Mills / Chemical Engineering and Processing 44 (2005) 141151
[34] P. Marjanovi c, V. Djordjevic, On the compressible ow losses through
abrupt enlargements and contractions, J. Fluids Eng. Trans ASME
116 (1994) 756762.
[35] P. Marjanovi c, D. Geldart, J.L.R. Orband, T. Mooney, A comparative
analysis of two hopper design methods, Proceedings of the Interna-
tional Congress for Particle Technology PARTEC 95, Nrnberg,
Germany, March 1995, pp. 6978.
[36] B. Armstrong, M.G. Jones, P. Marjanovi c, G. Welford, P.J. Blenkin,
R.G. Holder, An evaluation of the effects of high system pressure
on the performance of pneumatic coal conveyors, Proceedings of the
20th International Conference on Coal Utilization & Fuel Systems,
Clearwater, USA, 1995, pp. 393403.
[37] T. Mooney, P. Marjanovi c, Bend pressure loss in a pneumatic con-
veying system, Proceedings of the 20th Powder & Bulk Solids Con-
ference, Chicago, USA, May 1995, pp. 6173.
[38] P. Marjanovi c, G. Welford, M.G. Jones, Pneumatic conveying of coal
against high back pressures, Proceedings of the 12th International
Pittsburgh Coal Conference, Pittsburgh, USA, September 1995.
[39] M.G. Jones, P. Marjanovi c, G. Welford, The prediction of pneumatic
conveying performance when conveying into back pressures up to
20 bar, Proceedings of the 1st International Conference BULK ASIA,
Singapore, 1995, pp. 107115.
[40] R.J. Hitt, P. Marjanovi c, A computer technique to develop a model
of vertical dense phase gassolids ow from experimental data.
Proceedings of the 1st International Symposium on Two-Phase Flow
Modelling and Experimentation, vol. 1, Rome Italy, October 1995,
pp. 6774.
[41] P. Marjanovi c, M.G. Jones, Assessment of a new technique for
measuring ow properties of powdered materials stored in hoppers,
IMechE J. Process. Mech. Eng. 210 (1996) 18.
[42] P. Marjanovi c, Theoretical and practical silo design, Seminar on
Storage and Transportation of Bulk Solids, Moreton-in-Marsh, UK,
March 1996.
[43] M.G. Jones, P. Marjanovi c, Maintaining product quality in pneumatic
conveying, Proceedings of the Conveyorex 96 Seminar, Harrogate,
UK, April 1996.
[44] P. Marjanovi c, G. Welford, M.G. Jones, Pneumatic conveying of
solids at high system pressures, Proceedings of the 21st Powder &
Bulk Solids Conference, Chicago, USA, May 1996, pp. 9198.
[45] M.G. Jones, P. Marjanovi c, The modication of material properties
for improved ow characteristics, Proceedings of the Seminar on
Solving Problems in Hopper and Silo Systems, IMechE, London,
UK, June 1996.
[46] P. Marjanovi c, D.J. Mason, T. Mooney, The Performance of a pneu-
matic conveying system which incorporates an inclined pipeline sec-
tion, Proceedings of the 1st International Conference on Pneumatic
and Hydraulic Conveying Systems, Florida, USA, 1996.
[47] D.J. Mason, P. Marjanovi c, A. Levy, A simulation system for pneu-
matic conveying systems, Proceedings of the 1st International Con-
ference on Pneumatic and Hydraulic Conveying Systems, UEF Con-
ference, Florida, USA, 1996.
[48] P. Marjanovi c, D.J. Mason, The transient ow conditions during
feeding pneumatic conveying system using high-pressure blow tank,
Proceedings of the 12th International Congress of Chemical and
Process Engineering. Prague, Czech Republic, August 1996.
[49] D.J. Mason, A. Levy, P. Marjanovi c, Modelling the inuence of bends
on the ow of gas-solids mixture through pipelines, Proceedings of
the 12th International Congress of Chemical and Process Engineering,
Prague, Czech Republic, August 1996.
[50] A. Levy, P. Marjanovi c, D.J. Mason, A Comparison of analytical
and numerical models for gas-solid ow through straight pipe of
different inclinations with experimental data, Proceedings of the 12th
International Congress of Chemical and Process Engineering, Prague,
Czech Republic, August 1996.
[51] P. Marjanovi c, Determination of bulk solids indices using the Johan-
son indicizer system, Proceedings of the Bulk 96 Design Seminar,
Manchester, UK, December 1996, pp. 233238.
[52] T. Mooney, A. Levy, P. Marjanovi c, D.J. Mason, An investigation
of gassolids ow through inclined pipes, Proceedings of the 1997
Jubilee Research Event, vol. 1, Nottingham, UK, 1997, pp. 425428.
[53] P. Marjanovi c, M.G. Jones, G. Welford, Pneumatic conveying of
solids at high system pressures, J. Powder/Bulk Solids Technol. 1 (1)
(1997) 37.
[54] P. Marjanovi c, M. McGarvey, R.B. McKay, Development of lab-
oratory methodology to determine ow properties of organic pig-
ment powders, Proceedings of the 2nd Israel Conference for Con-
veying and Handling of Particulate Solids, Jerusalem, Israel, 1997,
pp. 2.282.33.
[55] A. Levy, T. Mooney, P. Marjanovi c, D.J. Mason, Analytical, Nu-
merical and experimental investigations for gassolid ow through
straight pipe of different inclinations, Proceedings of the 2nd Is-
rael Conference for Conveying and Handling of Particulate Solids,
Jerusalem, Israel, 1997, pp. 4.304.35.
[56] D.J. Mason, A. Levy, P. Marjanovi c, The inuence of bends on the
ow of gas-solids mixtures through pipelines, Proceedings of the
2nd Israel Conference for Conveying and Handling of Particulate
Solids, Jerusalem, Israel, 1997, pp. 4.364.41.
[57] P. Marjanovi c, D.J. Mason, Gas solids ow in an inclined pipeline.
Powder handling and processing, Trans. Tech. Pub. 9 (3) (1997)
217220.
[58] A. Levy, T. Mooney, P. Marjanovi c, D.J. Mason, A comparison of
analytical and numerical models with experimental data for gassolid
ow through a straight pipe at different inclinations, Powder Technol.
93 (1997) 253260.
[59] D.J. Mason, P. Marjanovi c, A. Levy, A simulation system for pneu-
matic conveying systems, Powder Technol. 95 (1998) 714.
[60] D.J. Mason, A. Levy, P. Marjanovi c, The inuence of bends on the
performance of pneumatic conveying systems, Adv. Powder Technol.
9 (3) (1998) 197206.
[61] D.J. Mason, P. Marjanovi c, Re-Visit of the fundamental denitions
of uid-solids ow properties in freight pipelines, Proceedings of
the 9th International Symposium on Freight Pipelines, Monterrey,
Mexico, April 1998.
[62] P. Marjanovi c, A. Levy, D.J. Mason, An investigation of the ow
structure through abrupt enlargement of circular pipe, Proceedings
of the 9th International Symposium on Freight Pipelines, Monterrey,
Mexico, April 1998.
[63] P. Marjanovi c, M. McGarvey, R.B. McKay, Development of labora-
tory methodology to determine ow properties of organic pigment
powders. Powder handling and processing, Trans. Tech. Pub. 10 (2)
(1998) 151154.
[64] M.G. Jones, P. Marjanovi c, The inuence of rotary valve air leakage
on pneumatic conveying system performance, Proceedings of the
23rd Powder & Bulk Solids Conference, Chicago, USA, May 1998,
pp. 231239.
[65] M.G. Jones, P. Marjanovi c, The inuence of rotary valve air leakage
on pneumatic conveying system performance, J. Powder/Bulk Solids
Tech. 2 (2) (1998) 310.
[66] M.G. Jones, P. Marjanovi c, The optimisation of an existing pneumatic
conveying system, J. Powder/Bulk Solids Technol. 2 (2) (1998) 12
16.
[67] P. Marjanovi c, D. Geldart, J.L.R. Orband, Techniques for assess-
ing powder owability a comparison, Proceedings of the World
Congress on Particle Technology 3, Brighton, UK, July 1998.
[68] D.J. Mason, A. Levy, P. Marjanovi c, The inuence of bends on
the ow of gas-solids mixture trough pipelines, Proceedings of the
World Congress on Particle Techn 3, Brighton, UK, July 1998.
[69] M.G. Jones, P. Marjanovi c, D. McGlinchey, R. McLaren, Segregation
in handling processes of blended industrial coal, Proceedings of the
6th International Conference on Bulk Material Storage, Handling
and Transportation, Wollongong, Australia, September 1998.
[70] P. Marjanovi c, M. McGarvey, R.B. McKay, Development of labora-
tory methodology to determine ow properties of organic pigment
D. Mills / Chemical Engineering and Processing 44 (2005) 141151 151
powders, Hopper & Silo Discharge: Successful Solutions. IMechE,
London, UK, November 1998.
[71] E.A. Knight, M.G. Jones, P. Marjanovi c, On-site measurement for
troubleshooting pneumatic conveying systems, Proceedings of the
Bulk Design Seminar, Sutton Coldeld, UK, December 1998.
[72] D.J. Mason, P. Marjanovi c, Re-visit to the fundamental denitions
of uid-solids ow properties in conveying pipelines, Proceedings
of the 2nd International Conference on Pneumatic and Hydraulic
Conveying Systems. UEF Conf. Davos, Switzerland, June 1999.
[73] P. Marjanovi c, E. McGee, Determination of powder ow properties
using different shear cells, Proceedings of the RelPowFlo 3, Pors-
grunn, Norway, August 1999, pp. 151158.
[74] M.G. Jones, P. Marjanovi c, D. McGlinchey, D. Morrison, Investiga-
tion of the discharge pattern of industrial coals from wedge shaped
hoppers, Proceedings of the RelPowFlo 3, Porsgrunn, Norway, Au-
gust 1999, pp. 543550.
[75] P. Marjanovi c, A. Levy, D.J. Mason, An investigation of the ow
structure through abrupt enlargement of circular pipe, Powder Tech-
nol. 104 (1999) 296303.
[76] P. Marjanovi c, M. McGarvey, R.B. McKay, Determination of the
inuence of surface coating and particle size on ow properties of
organic pigment powders, Proceedings of the 3rd Israel Conference
for Conveying and Handling of Particulate Solids, Dead Sea, Israel,
vol. 1, May 2000, pp. 3.393.44.
[77] M.G. Jones, P. Marjanovi c, D. McGlinchey, An investigation of
degradation and segregation in typical coal handling processes, Pro-
ceedings of the 3rd Israel Conference for Conveying and Handling
of Particulate Solids, vol. 1, The Dead Sea, Israel, May 2000, pp
8.558.60.
[78] D.J. Mason, J. Li, P. Marjanovi c, Numerical simulation of solids
feeding in a gas-solids pneumatic transport system, Proceedings of
the 3rd Israel Conference for Conveying and Handling of Particulate
Solids, vol. 2, The Dead Sea, Israel, May 2000, pp. 10.9310.101.
[79] D. McGlinchey, P. Marjanovi c, M. G. Jones, S. Cook, Particle seg-
regation in pneumatic conveying lines, IMechE Conference Trans-
actions, June 2000, pp. 331340.
[80] M.G. Jones, A.G. Mason, P. Marjanovi c, E. A. Knight, Bend Ef-
fects in Pneumatic conveying, IMechE Conference Transactions, June
2000, pp. 351361.
[81] R.O. Ansell, P. Marjanovi c, Hazardous substances COSHH revis-
ited, Powder Rep. 3 (4) (2000) 1314.
[82] P. Marjanovi c, E.A. Knight, J.R. Pugh, On-site measurement of
pneumatic conveying system performance: why, when, what and how,
Proceedings of the 16th International Conference on Material Flow,
Machines and Devices in Industry, Belgrade, Yugoslavia, December
2000, pp. 2025.
[83] J.S. Xiang, P. Marjanovi c, Modelling and simulation for pyroow
type circulating uidised bed boiler, Proceedings of the Interna-
tional Congress for Particle Tech PARTEC 2001, Nrnberg, Ger-
many, March 2001.
[84] P. Marjanovi c, Technology transfer between academia and industry
two-way street leading to mutual benet and success, MHEA
Bulk 2001 Technical Awareness Seminar, Manchester, UK, April
2001.
[85] P. Marjanovi c, E.A. Knight, J.R. Pugh, Benchmarking, optimising
and uprating of discontinuous pneumatic conveying system perfor-
mance through on-site measurement, Proceedings of the 6th World
Congress of Chemical Engineering, Melbourne, Australia, September
2001.
[86] J. Li, P. Marjanovi c, J.S. Xiang, E.A. Knight, An experimental tech-
nique for the analysis of plugs travelling through pneumatic pipelines
using pressure measurements, Proceeding of the 7th International
Conference on Bulk Materials Storage, Handling and Transportation,
Newcastle, Australia, October 2001, pp. 5367.
[87] E. McGee, P. Marjanovi c, L. Bates, Modifying ow behaviour in
hoppers using inserts and novel wall proles, Proceedings of the 7th
International Conference on Bulk Materials Storage, Handling and
Transportation, Newcastle, Australia, October 2001.
[88] J.S. Xiang, P. Marjanovi c, Hydrodynamic model of gas-solid ow
in circulating uidised bed, Proceedings of the 7th International
Conference on Bulk Materials Storage, Handling and Transportation,
Newcastle, Australia, October 2001, pp. 825832.
Chemical Engineering and Processing 44 (2005) 153157
Inuence of the particle diameter and density in the
gas velocity in jet spouted beds
Mara J. San Jos

, Sonia Alvarez, Alvaro Ortiz de Salazar, Martn Olazar, Javier Bilbao


Departamento de Ingeniera Qumica, Universidad del Pas Vasco, Facultad de Ciencias, Apartado 644, 48080 Bilbao, Spain
Received 30 September 2003; received in revised form 25 February 2004; accepted 19 March 2004
Available online 20 June 2004
Abstract
A gas ow pattern in the regime of jet spouted bed (or dilute spouted bed) has been analyzed in conical spouted beds using materials of
different density and particle diameter. The experimental results have been tted to an empirical equation for calculation of the interstitial
velocity at the contactor axis in jet spouted beds. The general equation proposed relates this property to the geometric factors of the contactor
(angle, ; and inlet diameter, D
o
) and to the operating variables (particle size, d
p
; particle density,
s
; particle shape, ; stagnant bed height,
H
o
; and air relative velocity referred to that minimum spouting, u
o
/u
oms
).
2004 Elsevier B.V. All rights reserved.
Keywords: Spouted beds; Jet spouted beds; Gas ow
1. Introduction
Markowski and Kaminski [1] identied the hydrodynamic
characteristics of the regime of jet spouting obtained by
expansion of the spouted bed in conical contactors: high
velocity; high average bed voidage, over 0.75 depending on
the operating conditions; systematic cyclic movement of the
particles; a hydrodynamic behaviour different to that of the
conventional spouted bed.
The jet spouted bed, Fig. 1, has a vigorous movement for
the solid and a uniform gas-solid contact, which is necessary
in operations with sticky solids and in highly exothermal
reactions. In the same way, it allows for working in a close to
uniform way with a wide particle size distribution. In short,
the hydrodynamic differences between the jet spouted bed
and the conventional conical contactors are:
Non-existence of hysteresis in the pressure drop versus
velocity curves.
Deviation of the experimental results of minimumvelocity
for which the correlations deduced for the spouted bed
are not applicable.
Lesser instability problems, despite the fact the hydrody-
namic regime is greatly affected by the contactor dimen-

Corresponding author. Tel.: +34 94 6015362; fax: +34 94 6013500.


E-mail address: iqpsaalm@lg.ehu.es (M.J. San Jos e).
sions, by the characteristics of the inlet, by the particle
diameter and by other operation and design parameters.
The design of the jet spouted bed requires the rigorous
denition of the gas ow pattern in order for the residence
time distribution to be considered. In previous papers, the
regime of jet spouted bed and its hydrodynamics correlations
have been dened [28]. The minimum jet spouting velocity
is calculated by the following correlation [7].
(Re
o
)
mj
= 6.891 Ar
0.35
(D
b
/D
o
)
1.46
[tan(/2)]
0.53
(1)
The geometric factors and the design requirements for
stability have been delimited and the operational conditions
of the jet spouted bed have been compared with those of the
other gas-solid contact techniques [2,3].
The good behaviour of this regime has been proven in a
satisfactory way in the obtaining of thermally stable poly-
benzyls by polymerization of gaseous benzyl alcohol on
solid acid catalyst [9,10]. The jet spouted bed has also
been successfully used in coal gasication [1113]. The jet
spouted bed allows for working with short gas residence
times, by using a simpler and more economical device than
that of other innovative design reactors, such as the cyclonic
reactors [1416], spout-uid bed reactors with draft tube
[17], spouted bed reactors with draft tube [18], or the im-
pinging stream reactors [19,20]. Consequently, they have a
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.011
154 M.J. San Jos e et al. / Chemical Engineering and Processing 44 (2005) 153157
Fig. 1. Solid circulation in jet spouting regime.
potential interest for fast reactions in which selectivity is the
factor for optimization.
A model for gas ow using glass beads has been pro-
posed in previous papers [21]. In order to describe the gas
ow in jet spouted bed regime in conical contactors. In this
paper, the vertical component of the gas interstitial velocity
has been calculated in jet spouted beds and the experimental
results have been tted to an empirical equation for calcu-
lation of the interstitial velocity at the contactor axis in jet
spouted beds.
2. Experimental
The pilot plant scale experimental unit, Fig. 2, described
in previous papers [2,3] is provided with a blower that sup-
plies a maximum air owrate of 300 Nm
3
/h at a pressure of
1500 mm of water column. The ow measurement is carried
out with two rotameters, used in the ranges from 2.5 to 30
and from 30 to 250 Nm
3
/h. In the range from 0 to 4 Nm
3
/h a
gas volume meter provided with paddles is used. A system
of valves allows for choosing the suitable rotameter for the
desired owrate. This owrate is xed by closing or opening
a buttery valve.
Five contactors of poly(methyl methacrylate) have been
used, whose geometric characteristics are denided in the
Fig. 2. Experimental equipment.
Fig. 3. They have made to the following dimensions: column
diameter D
c
, 0.36 m; contactor base diameter D
i
, 0.62 m;
cone angles, , 33, 36, and 45

; height of the conical section


H
c
, 0.50, 0.45, and 0.36 m; gas inlet diameter D
o
, 0.03, 0.04,
0.05 and 0.06 m; stagnant bed height H
o
, between 0.05 and
0.30 m.
The solids used correspond to group D and B of Geldart
classication [22,23] and they are set out in Table 1.
The velocity readings are carried out by four probes whose
radial and longitudinal position can be established at will
inside the contactor by means of a displacement device
controlled by computer, which established the coordinates
Fig. 3. Geometric factors of the contactors.
M.J. San Jos e et al. / Chemical Engineering and Processing 44 (2005) 153157 155
Table 1
Properties of the solids used
Material
s
(kg/m
3
)
d
p
(mm)

o
Geldart
classication
Glass beads 2420 1.0 1 0.322 B
2420 2.0 1 0.328 D
2420 3.0 1 0.345 D
2420 3.5 1 0.351 D
2420 4.0 1 0.355 D
2420 5.0 1 0.358 D
Polypropylene 890 3.5 0.90 0.390 D
Extruded polystyrene 1030 1.0 0.80 0.395 B
1030 2.0 0.80 0.385 D
1030 3.5 0.80 0.380 D
Expanded polystyrene 65 3.5 0.95 0.507 D
Sawdust 242 1.0 0.90 0.324 B
242 1.5 0.90 0.328 D
242 2.0 0.90 0.330 D
of the point to be measured by each probe following a given
sequence.
3. Results
A wide experimental study of the vertical component of
the gas interstitial velocity in the contactor has been carried
out with the aim of obtaining a correlation that relates this
property to the geometric factors of the contactor (angle, ,
and inlet diameter, D
o
) and to the operating variables (parti-
cle size, d
p
; particle density,
s
; particle shape, ; stagnant
bed height, H
o
; and air relative velocity referred to that min-
imum spouting, u
o
/u
oms
).
The general equation proposed to calculate the vertical
component of the interstitial velocity at any position of the
contactor is a Gaussian curve, in which the velocity in each
position is a function of radius, r; of the maximum velocity
at the axis of the contactor, u
j
(0) and of the velocity along
the ow at the wall, u
W
:
u
j
= (u
j
(0) u
W
) exp[0.5(r/)
2
] +u
W
(2)
The evolution of the maximum velocity, u
j
(0), with the
dimensionless longitudinal position at the axis,

, has a
similar tracing for all the systems, the velocity decreases as
the longitudinal position at the axis increases, Fig. 4. These
velocity proles are more pronounced than those of conical
spouted beds [24].
The maximum velocity at the axis of the contactor, u
j
(0),
is a function of the geometric factors and operating condi-
tions. The experimental results have been tted to an em-
pirical equation for calculation of the interstitial velocity
at the contactor axis in jet spouted beds. The tting has a
Fig. 4. Evolution with the dimensionless position in the bed, *, of the
vertical component of the gas interstitial velocity at the axis, u
j
(0), in jet
spouting regime. = 36

, D
o
= 0.04 m, H
o
= 0.23 m for the different
materials of particle diameter d
p
= 3.5 mm.
regression coefcient of r
2
= 0.97 and the standard devia-
tion of 3%.
u
j
(0) =2.14 10
5

2.78 10
2
D
2
o
+H
8
o

d
p

0.17

K
1 +K

(3)
K=H
0.81
o

0.1

(
0.85
0.22D
0.5
o
+0.75d
0.5
p
+0.21
0.5
0.248 1.17 10
6

s
)

(4)
The expression proposed for the vertical component of
velocity near the wall, u
W
, which has been chosen taking
into account simplicity and because it gives a satisfactory
tting for all the experimental data is only function of the
dimensionless spherical radius,

:
u
W
= 20
2
exp(1.4
2
) (5)
From the vertical components of points velocity in the
contactor and by using the continuity equation, the radial
component, u
r
, and consequently, the modulus and direction
of velocity vector at each point in the contactor are calcu-
lated.
The continuity equation applied to differential volume el-
ement in cylindrical coordinates is:
1
r

r
(ru
r
) +
u
z
= 0 (6)
In Fig. 5, the map of velocities and of uid direction is
shown as an example, which corresponds to the system =
156 M.J. San Jos e et al. / Chemical Engineering and Processing 44 (2005) 153157
Fig. 5. Map of gas interstitial velocity vector in jet spouting regime.
= 36

, D
o
= 0.04 m, H
o
= 0.23 m and extruded polystyrene of particle
diameter d
p
= 3.5 mm.
36

, D
o
= 0.04 m, H
o
= 0.23 m and extruded polystyrene
d
p
= 3.5 mm. The length of the arrows is proportional to
the value of the corresponding velocity vector modulus.
From the results it has been proven for all the experimen-
tal systems studied that the ow rate is constant along the
streamtubes dened in the annular zone, such as has been
proposed in the model.
In order to solve the model conservation equations, only
the knowledge of the average interstitial velocity in the
spout zone at each level, u
s
, is required. This is calculated
as:
u
s
=
1
A
s

u dA
s
=
1
A
s

D
o
/2
0
2u
r
dr (7)
The values of u
s
calculated using Eq. (7) are very similar
(differ in less than 3%) to the values of velocity at the axis
u
j
(0) for beds of sawdust and expanded polystyrene. Conse-
quently, the velocity prole in the spout zone is almost at
in beds of these solids. Nevertheless, for beds of glass beads
and extruded polystyrene the velocity proles in the spout
zone are more pronounced.
4. Conclusions
In jet spouted bed regime in conical spouted bed, the gas
interstitial velocity at the axis, u
j
(0), decreases as the dimen-
sionless longitudinal position at the axis,

, increases; be-
ing this decreasing more pronounced as the solid density in-
creases. Besides these velocity proles are more pronounced
than those of conical spouted bed regime.
Empirical correlations proposed in this paper have been
proven to be valid for calculating the interstitial velocity
at any position of the contactor. These equations takes into
account, as well as the geometric factors of the contactor
the operating variables such as particle size, particle density
and particle shape.
The calculated average velocity in the spout at each level
are very similar to the velocity at the axis for beds of low
density solids. Nevertheless, for beds of solids of higher
density, the velocity proles in the spout zone are more
pronounced.
Acknowledgements
This work was carried out with the nancial sup-
port of the University of the Basque Country (Project
9/UPV00069.310-13607/2001).
Appendix A. Nomenclature
A
r
Archimedes number gd
3
p

g
(
s

g
)/
2
A
s
cross-sectional area of the spout (m
2
)
d
p
particle diameter (m)
D
b
, D
c
, D
i
, D
o
diameter of the upper level of the
stagnant bed, of the column, of the bed
base and of the inlet, respectively (m)
g gravity constant (ms
2
)
H
c
, H
o
height of the the conical section of the
contactor, and of the stagnant bed,
respectively (m)
r, z radial and longitudinal coordinates (m)
R contactor radius at z height (m)
(Re
o
)
mj
Reynolds modulus of minimum jet
spouting referred to D
o
u, u
j
(0), u
j
, u
w
vertical component of interstitial velocity,
maximum velocity at the axis, at any
position and at the contactor wall (ms
1
)
u
r
, u
s
radial component of interstitial velocity
and vertical component of the average
interstitial velocity in the spout at z level,
respectively (ms
1
)
u

ratio between velocity and minimum


spouting velocity, dimensionless
Greek letters

o
bed voidage of the static bed,
dimensionless
particle shape factor
contactor angle (deg)
viscosity (kg m
1
s
1
)
M.J. San Jos e et al. / Chemical Engineering and Processing 44 (2005) 153157 157
,
N
,
o
longitudinal spherical coordinates,
maximum and minimum longitudinal
spherical coordinates (m)

g
,
s
,
v
density of the gas, of the solid and of
the glass beads, respectively (kg m
3
)

dimensionless longitudinal spherical


coordinate, (
o
)/(
N

o
)
parameter of Eq. (2) (m).
References
[1] A. Markowski, W. Kaminski, Hydrodynamic characteristics of jet
spouted beds, Can. J. Chem. Eng. 61 (1983) 377381.
[2] M. Olazar, M.J. San Jos, A.T. Aguayo, J.M. Arandes, J. Bilbao,
Stable operation conditions for gassolid contact regimes in conical
spouted beds, Ind. Eng. Chem. Res. 31 (1992) 17841792.
[3] M. Olazar, M.J. San Jos, A.T. Aguayo, J.M. Arandes, J. Bilbao,
Design factors of conical spouted beds and jet spouted beds, Ind.
Eng. Chem. Res. 32 (1993) 12451250.
[4] M.J. San Jos, Operation regimes in conical spouted beds. Stability
conditions and hydrodynamics, Ph.D. Dissertation, University of the
Basque Country, Bilbao, Spain, 1991.
[5] M.J. San Jos, M. Olazar, A.T. Aguayo, J.M. Arandes, J. Bilbao,
Design and hydrodynamics of conical jet spouted beds, Rcents
Progrs en Gnie des Procds, La Fluidisation 5 (1991) 146153.
[6] M.J. San Jos, M. Olazar, A.T. Aguayo, J.M. Arandes, J. Bilbao, in:
D. Behrens (Ed.), Gassolid contact regimes in conical spouted beds,
Proceedings of the Fourth World Congress of Chemical Engineering,
Fluidization Session, Chemische Technik und Biotechnologie e.V.,
Frankfurt am Main, Germany, 1991, pp. 9.513.
[7] M.J. San Jos, M. Olazar, A.T. Aguayo, J.M. Arandes, J. Bilbao,
Hydrodynamics correlations of conical jet spouted beds, in: O.E.
Potter, D.J. Nicklin (Eds.), Fluidization VII, Engineering Foundation,
New York, 1992, pp. 831838.
[8] M.J. San Jos, M. Olazar, A.T. Aguayo, J.M. Arandes, J. Bilbao,
Expansion of spouted beds in conical contactors, Chem. Eng. J. 51
(1993) 4552.
[9] J. Bilbao, M. Olazar, A. Romero, J.M. Arandes, Design and operation
of a jet spouted bed reactor with continuous catalyst feed in the benzil
alcohol polymerization, Ind. Eng. Chem. Res. 26 (1987) 12971304.
[10] J. Bilbao, M. Olazar, A. Romero, J.M. Arandes, Optimization of
the operation in a reactor with continuous catalyst circulation in the
gaseous benzyl alcohol polymerization, Chem. Eng. Commun. 75
(1989) 121134.
[11] O. Uemaki, T. Tsuji, in: K. Ostergaard, A. Sorensen (Eds.), Flu-
idization V, Engineering Foundation, New York, 1986, p. 497.
[12] O. Uemaki, T. Tsuji, in: Proceedings of the 41st Canadian Chem-
ical Engineering Conference, Vancouver, BC, CSChE Publications
Department, Ottawa, Canada, 1991, pp. 117.
[13] T. Tsuji, T. Shibata, K. Yamaguchi, O. Uemaki, in: Proceedings of
International Conference on Coal Science, vol. I, 2327 October
1989, Tokyo, p. 457.
[14] J. Lede, H.Z. Li, J. Villermaux, Le Ciclon reactor. Partie I: measure
directe de la distribution des temps de sjour de la phase Gaseuze-
Lois dExtrapolation, Chem. Eng. J. 42 (1989a) 3755.
[15] J. Lede, K.B. Mathur, J. Villermaux, Le Ciclon reactor. Partie II:
measure directe de la distribution des temps de sjour de la phase
Solide-Lois dExtrapolation, Chem. Eng. J. 42 (1989b) 103117.
[16] T.A. Gauthier, C.L. Briens, M.A. Bergougnou, P.A. Galtier, Distri-
bution des temps de sejour du gaz dans un cyclone a co-courant,
Rcents Progrs en Gnie des Procds La Fluidisation 5 (1991)
185192.
[17] R.K. Stoecker, A. Rastogi, L.A. Behie, W.Y. Svreck, M.A. Bergoug-
nou, A computer simulation of propane cracking in a spout-uid bed
reactor with draft tube, in: K. Ostergaard, A. Sorensen (Eds.), Flu-
idization V, Engineering Foundation, New York, 1986, pp. 465472.
[18] R.K. Stoecker, J.H. Eng, W.Y. Svreck, L.A. Behie, Ultrapyrolysis
of propane in a spouted-bed reactor with a draft tube, AIChe J. 35
(1989) 16171624.
[19] D. Sonnet, S. Afara, C.L. Briens, J.F. Large, M.A. Bergougnou, 1988.
Micromixing of particles in an ultra-rapid uidized (URF) reactor,
in: P. Basu, J.F. Large (Eds.), Circulating Fluidized Bed Technology
II, Pergamon Press, Oxford.
[20] A. Tamir, Process and phenomena in impinging-stream reactors,
Chem. Eng. Prog. 86 (9) (1989) 5361.
[21] M.J. San Jos, M. Olazar, F.J. Peas, A.T. Aguayo, J.M. Arandes,
J. Bilbao, A model for gas ow in jet spouted beds, Can. J. Chem.
Eng. 71 (1993) 189194.
[22] D. Geldart, Types of uidization, Powder Technol. 7 (1973) 285292.
[23] D. Geldart, Gas Fluidization Technology, John Wiley, New York,
1986.
[24] M.J. San Jos, M. Olazar, F.J. Peas, A.T. Aguayo, J.M. Arandes, J.
Bilbao, Correlation for calculation of the gas dispersion coefcient
in conical spouted beds, Chem. Eng. Sci. 50 (1995) 21612172.
Chemical Engineering and Processing 44 (2005) 159166
Taking-off model of particles with a wide
size distribution
Isabelle Descamps, Jean-Luc Harion

, Bernard Baudoin
Dpartement Energtique Industrielle, Ecole des Mines de Douai, 941 rue Bourseul, B.P. 838, 59508 Douai, France
Received 23 June 2003; received in revised form 4 April 2004; accepted 15 April 2004
Available online 2 July 2004
Abstract
The aim of this work is to estimate more accurately fugitive dust emissions due to wind erosion of exposed aggregate storage piles. The
model constructed allows to quantify the temporal evolution in the mass ux when a bed of materials is exposed to a turbulent ow. It is based
on the interaction between particles take-off and wall turbulence. The model allows as well to take into account materials having a wide size
distribution, which is typical of materials such as coal or ore. Some experimental studies have shown a temporal decrease in the mass ux as
the bed contains a wide range of particle sizes. This decrease, due to large particles, is predicted by the model. The rate depends on the ow
velocity and the characteristics of the particles.
2004 Elsevier B.V. All rights reserved.
Keywords: Fugitive dust emissions; Wind erosion; Particle take-off; Mass ux
1. Introduction
Atmospheric diffuse dust emissions constitute a pollution
source, which is difcult to be quantied. In fact, a diffuse
emission is dened as a polluting atmospheric ux which
is spread out over a large area or which is not concentrated
[1]. In the case of steelwork sites, diffuse dust emissions can
represent many as 20% of airborne particle pollution, they
result from conveyances and handling of materials and from
wind erosion of exposed aggregate storage piles. This latter
category of emission source is widely predominant.
The aims of the investigations presented in this paper are
to estimate accurately fugitive dust emissions resulting from
wind erosion. Thus, a model which allows to quantify emis-
sions resulting from an exposed particle bed to a turbulent
ow has been developed. Its originality is to take into ac-
count the wide particle size distribution of materials used,
as example, at steelwork sites. Typically, the ner particles
can have a size of about 10 m and the larger running to
centimetres. This characteristic is very important. In fact,
Meunier [2] observed for various experimental tests carried

Corresponding author. Tel.: +33 3 2771 2379; fax: +33 3 2771 2915.
E-mail address: harion@ensm-douai.fr (J.-L. Harion).
out in a wind tunnel with coal and ore, that the emitted mass
ux of particles decreases with time. He remarked the very
fast take-off of ne particles staying at the bed surface, then
the increasingly difcult take-off of the particles being be-
low. This can be explained by the presence of coarse par-
ticles close to the surface, which dont take-off because of
their inertia. These large particles constitute a cover and this
phenomenon is called the pavement. The pavement of the
bed, which is inherent to the wide size distribution of parti-
cles (especially to the presence of non-erodible particles at
the bed surface), generates a temporal decrease in emitted
mass ux.
The temporal decrease in emitted mass ux in the case of
particles with a wide size distribution has been shown by a
few experimental studies [35]. These authors assume that
the temporal decrease in emitted ux is due to modica-
tion of soil surface during wind erosion. These experimen-
tal studies underline the importance to consider the tempo-
ral evolution of the emitted mass ux in order to accurately
estimate particles emissions with a wide size distribution.
The rst part of the present work describes the wind ero-
sion model of a horizontal particle bed. A simple test of the
model is presented in the second part, which reports the in-
uence of a wide size distribution on the emitted mass ux
of particles.
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.04.007
160 I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166
2. Wind erosion model of a particle bed
2.1. Aerodynamic entrainment modelling
The modelling of aerodynamic entrainment is based on
the close link between particles take-off and turbulent co-
herent structures above the surface. In fact, some authors
[6,7] have experimentally observed that a particle take-off
can be associated to the ejection of uid from the wall re-
gion due to the presence of streamwise counter rotating vor-
tices. If it is assumed that the presence of two streamwise
counter rotating vortices produces only one ejection, each
pair of streamwise vortices is considered as a possibility that
a particle takes-off. Thus, for each of these possibilities, a
take-off criterion is tested.
The vortex structures appear with statistical spatial peri-
odicities, whose values are dependent on the ow properties
and velocity, and on the presence or not of particles. Jimenez
and Moin [8] have obtained, by numerical simulations, the
values of average spatial periodicities of appearing in the
spanwise and streamwise directions of the ow which are
respectively,
z
+
100 and
x
+
250300.
The periodicities are adimensionalized with the two scale
parameters of the inner layer of the turbulent boundary, the
kinematic viscosity and the friction velocity u

, by the
equation:

+
=
u

(1)
Although, these values are established in the case of a
smooth wall, they will be considered in the rst approxima-
tion like the spanwise and streamwise spatial periodicities
of the coherent structures over the particle bed. The parti-
cle bed is thus subdivided into boxes having spanwise and
streamwise sizes given by average spatial periodicities of
the turbulent coherent structures.
Fig. 1 shows the number of taking-off possibilities over
the bed surface if it is assumed that each box, which contains
an ejection, is associated with a possibility of taking-off. The
numbers of possibilities of taking-off along the spanwise
and streamwise directions of the bed are equal respectively
to l/
z
and L/
x
. Consequently, they depend on the ow
velocity.
The frequency of the take-off criterion, which is tested for
each box, is supposed to be equal to the burst frequency in
Fig. 1. Number of taking-off possibilities at the bed surface.
the wall region of the turbulent boundary layer. Blackwelder
and Haritonidis [9] have shown that the average periodicity
of occurence is given by T
B
+
250 whose expression is:
T
+
B
=
T
B
u
2

(2)
So the time step of the model is equal to T
B
.
2.2. The take-off criterion
Before the take-off criterion, a size range of representa-
tive diameter D
p
is selected randomly from the total number
of size ranges of the particle size distribution. The take-off
criterion, which is established from the balance of forces
exerted on a particle lying at a at surface, requires the
lift force from the ow to exceed the forces holding the
particle down, namely the adhesive forces and the particle
weight. The expression of the lift force used in the model
is given by the experimental expression of Mollinger and
Nieuwstadt [10]. They have developed a new technique for
measuring the lift force and its uctuations. Consequently,
the instantaneous lift force value is equal to the sum of the
average lift force F
aero
and the uctuating lift force F

aero
,
which is selected randomly fromits probability density func-
tion which has been measured by Mollinger and Nieuwstadt
[10]. In the same way, the adhesive force is selected ran-
domly from its probability density function given by Zimon
[11].
The aerodynamic entrainment model is based on the com-
parison between the distributions of forces holding the par-
ticles on the bed and of the lift force exerted on the particles
by the ow. The expression of the take-off criterion is given
for each size range by:
F
aero
(D
p
) +F

aero
(D
p
) P(D
p
) +F
ad
(D
p
) +F

ad
(D)
Particle take-off
The take-off test of the model developed in the present
study allows to simulate the stochastic nature of the phe-
nomenon. The establishment of the take-off criterion is de-
tailed more precisely in the thesis of Descamps [12].
One of the characteristic parameters of particle take-off
is the threshold ow velocity, at which particles start to take
off. The threshold ow velocity can be dened from the
take-off criterion.
I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166 161
Fig. 2. Take-off curve [13].
Fig. 2 shows the take-off curve obtained by Foucaut [13].
It traces out the adimensionalized threshold friction velocity
u

according to the adimensionalized particle diameter



D
p
.
These two parameters are given by the following equations:

D
p
=
D
p
D
p
ref
u

=
u

ref
(3)
Fig. 3. Integration of the effect of the pavement by diminishing by graduated steps the lift force exerted on a particle.
Fig. 4. Principle of the model of simulation of the emitted mass ux.
The values of D
p
ref
and U

ref
are, respectively, (
2
/
p
g)
1/3
and (
p
g)
1/3
, where
p
= (
p
)/ [13]. The take-off
curve allows to estimate by knowing the adimensional-
ized friction ow velocity the granulometric slice liable to
take-off. In fact, the diameter noted

D
pcrit1
is the diameter
from which the particles take-off and the diameter

D
pcrit2
162 I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166
is the diameter from which the particles dont take-off be-
cause of their inertia. Thus, if the value of

D
pcrit2
is known,
the proportion of non-erodible particles can be determinate.
2.3. Modelling of the pavement
The temporal decrease in emitted mass ux can be ex-
plained by the phenomenon of pavement, which is linked to
the strong inertia particles, which cannot take-off and which
form a protection, called cover. It is supposed that these
non-erodible particles remain at the bed surface, and conse-
quently, overlay it completely from a certain depth, which
is called erosion depth, E
erosion
. From this depth, the vortex
structures bump only particles which they are not able to
entrain. No more take-off is possible.
In order to determinate the erosion depth, we assume that
the particles have a spatial uniform distribution along the
height Hof the bed. For each size range, the number of layers
N
c
(D
p
) including into the height of the bed is calculated by
dividing the height of the bed by the particle diameter. The
number of particles N
part
(D
p
) of each size range, given by
the size distribution of beds particles, is distributed equally
on each layer of the bed:
N
p
(D
p
) =
N
part
(D
p
)
N
c
(D
p
)
(4)
Now, the depth from which all the non-erodible particles
cover the bed surface has to be determined. Kurose and
Komori [14] dene the rate of cover like the ratio between
the surface equal to the projected area of particles on the wall
and the total area of the considered surface. Consequently,
the rate of cover is determined for each size range by:
T
C
(D
p
) =
N
p
(D
p
) D
2
p
/4
L l
(5)
Thus, the bed surface covered by the non-erodible parti-
cles is calculated by successive summation on all the size
ranges of the non-erodible particles in the increasing order
of diameters. For a y depth of the bed given, the surface
covered by the non-erodible particles can be determinate
since the number of layers, included by y, is known for
each size range. The evolution of the surface, covered by the
non-erodible particles, according to the depth can be estab-
lished. The depth from which the covered surface is equal
to the bed surface, namely the value of the erosion depth
E
erosion
, is deduced from this evolution. We suppose that, if
there are not non-erodible particles in the bed, the erosion
depth is equal to innity. The erosion depth is then divided
by the diameter D
p
of each size range. The number of layers
likely to be affected by erosion is given by:
N
erosion
(D
p
) =
E
erosion
D
p
(6)
For each size range, the value of the lift force is then de-
creased as soon as all the particles of a layer have taken-off.
The increment is given by:
F
aero
(D
p
) =
Max pdf(F
aero
initial
) Min pdf(P +F
ad
)
N
erosion
(D
p
)
(7)
Fig. 3 shows schematically the modelling of the bed pave-
ment.
This formulation allows to include in the model the lift
force inuence decrease for particles which are in a deeper
position in the bed. When the erosion depth is attained, the
take-off criterion can no more be respected. Indeed, the val-
ues of the probability density functions of the instantaneous
Fig. 5. Algorithm of the model.
I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166 163
lift force are always smaller than the sum of the adhesion
and weight forces. In fact, the maximum of the probability
density function of the lift force is lower than the minimum
of the sum of the probability density function of the adhe-
sion force and the weight force.
2.4. Synthesis and algorithm of the model
This part describes concisely the model presented in this
paper. The modelling is constituted of three parts (Fig. 4):
the aerodynamic entrainment of particles, which is based
on the interaction between the coherent structures appear-
ing at the bed surface and the particles take-off;
the denition of the take-off criterion which depends on
the particles diameters and density for a given ow veloc-
ity;
the pavement modelling as a result of the presence of
coarser particles at the bed surface.
The principle of the model is to scan the bed surface,
which is subdivided into boxes whose the width and the
length are equal respectively to the spanwise and streamwise
statistical periodicities of appearing of the coherent struc-
tures. In fact, some authors have shown that the phenomenon
of ejection in a turbulent boundary layer could be connected
with the particles take-off.
For each box, namely for each ejection, the criterion for
taking-off is tested. Previously, it is necessary to determine
which particle, that is to say, which size range will be tested.
The size range of representative diameter D
p
of the particle
likely to be carried away by the ow is selected randomly
fromthe size distribution. If all the particles of this size range
have taken-off, then an other size range is randomly selected.
For each take-off, a particle is substracted from its size
range then it is counted in the emitted mass ux. Propor-
tionately to these takes-off, the ow has more and more dif-
culty to carry away particles, this difculty is taken into
account by the modelling of the bed pavement.
The model algorithm is showed on Fig. 5.
Fig. 6. Size distributions tested by the model.
The input variables are:
the exposition time to the turbulent ow, d
expo
;
the ow velocity, U

;
the particles density, the representative diameters of the
size distribution and their mass percentages, the number
of size ranges and the value of their depth;
the adimensionnalized values of the streamwise, spanwise
and temporal appearing periodicities of coherent struc-
tures;
the sizes of the bed (height, width and length);
the kinematic viscosity and the density of the ow.
Afterwards, the modelling is divided into various parts:
the calculation of N
part
(D
p
), N
c
(D
p
) and N
p
(D
p
);
the establishment of E
erosion
and N
erosion
(D
p
),
the calculation of F
aero
(D
p
) and of the average lift force,
the standard deviation of the probability density function
of the lift force, the average adhesion force, the standard
deviation of the probability density function of the adhe-
sion force, the weight force,
the bed is scanned along width and length. The iterations
number is equal to the number of coherent structures ap-
pearing along the width multiplied by the number of co-
herent structures appearing along the length multiplied by
the ratio between the exposition period and the occurence
period of the vortex structures. The take-off criterion is
tested for each iteration and the modelling of the pave-
ment is taken into account.
The results obtained by the model are the temporal varia-
tion of the emitted mass and of the mass ux as well as the
temporal evolution of the surface size distribution and the
size distribution of eroded particles.
3. Numerical results
This part aims at showing that the model allows to
estimate the temporal decrease of the emitted mass ux
164 I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166
observed in presence of non-erodible (relatively coarse)
particles.
A test has been investigated with a simple size distri-
bution constituted by a blend of erodible (60160 m)
and non-erodible sand particles (540640 m). The rate of
non-erodible particles, which is the cover rate noted tdc,
varies from 0 to 80%. Fig. 6 shows the size distributions
which are tested by the model.
The input variables are:
the exposition time to the turbulent ow, d
expo
= 60 s,
the ow velocity, U

= 10 m/s,
the particles density,
p
= 2500 kg/m
3
, the representative
diameters of each size range, D
p
, and the mass percentages
corresponding, P
mass
, which are given by Fig. 6,
the statistical spatial and temporal periodicities of appear-
ing of ejections
x
+
=300,
z
+
=100 and T
B
+
= 250,
the sizes of the bed H = 2.5 cm, L = 48.5 cm and l = 8.7
cm.
Fig. 7 shows the results obtained by the model. The gure
traces out the evolutions of the mass ux in kg/m
2
s according
to the time in s for a velocity of the ow equal to 10 m/s
and for various cover rates, tdc = 0, 20, 40, 60 and 80%.
Several remarks can be made from the numerical results.
If the particles bed does not contain any non-erodible parti-
cles, then the emitted mass ux is constant with time. When
the particles bed contains non-erodible particles, the emitted
mass ux decreases with time.
Fig. 7 shows that the rate of the temporal decrease is
higher when the initial percentage of large particles in-
creases, which corresponds to experimental observations.
Moreover, the initial ux of particles decreases with the
presence of non-erodible particles.
Other tests, which are not represented in this paper, have
shown that, when the velocity is higher, the rate of the tem-
Fig. 7. First results obtained by the model in the case of a wide size distribution: U

=10 m/s;
p
=2500 kg/m
3
; tdc = 0, 20, 40, 60 and 80%.
poral decrease and the initial percentage of the emitted mass
ux increase.
All these results can be deduced from the modelling of
the wind erosion of the bed. The decrease rate goes higher,
when the cover rate, noted tdc, increases. This can be ex-
plained by the modelling of the pavement, described in part
2.3. The reduction increment of the lift force is inversely
proportional to the erosion depth (Eq. (6)). According to the
denition of the erosion depth, the higher is the proportion
of non-erodible particles, the smallest is the erosion depth.
Consequently, when the proportion of non-erodible particles
is increased, the reduction increment becomes greater and
the temporal decrease is faster.
When the mass percentage of non-erodible particles
is greater, the emitted mass ux is lower. In fact, be-
fore the take-off criterion, a size range is randomly se-
lected from the size distribution. So, the proportion of
non-erodible particles is taken into account. When cover
rate is increased the chances that the particle is erodible are
weaker.
As ow velocity is increased, the emitted mass ux in-
creases. In fact, the number of taking-off possibilities equals
the number of boxes at the bed surface times the number
of time step of the model. Moreover, the number of boxes
at the bed surface multiplied by the number of time step is
proportional to U
4

.
When the takes-off are more numerous, the layers of par-
ticles are blown off faster. Consequently, the decrease of
emitted mass ux is greater.
Fig. 7 also shows a temporal decrease in mass ux in
graduated steps. The bound from one step to another is due
to the passage of one layer to another in the numerical model
and consequently to the decrement of the probability density
function of the lift force. This is not directly representative of
the reality. However, a general trend of the temporal decrease
I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166 165
Fig. 8. Interpolation of the temporal decrease in emitted mass ux obtained by the model: U

=10 m/s;
p
=2500 kg/m
3
; tdc= 20, 40, 60 and 80%.
in the mass ux can be obtained by interpolating the data
by a negative exponential function.
In Fig. 8 it is observed that the temporal decrease of emit-
ted mass ux can be interpolated by an exponential negative
function:
ux(t) = ux(t
initial
)exp(k
d
t) (8)
k
d
is the decrease rate and ux(t
initial
) is the emitted mass
ux at t
initial
0 s.
4. Conclusions
The pavement modelling allows to introduce into the
model the temporal evolution of the size distribution of
materials at the bed surface. By a progressive decrease of
the probability density function of the lift force, this model
successfully predicts the temporal decrease in mass ux
that occurs with the presence of coarse particles at the sur-
face. The rate of this decrease depends on the ow velocity
and the characteristics of the particles. In order to improve
the accuracy of the estimation of fugitive particle emissions
with a wide size distribution, it is necessary to take into
account this temporal decrease.
An experimental study, whose aim is to measure the tem-
poral decrease in mass ux, is on running and will allow to
compare the results obtained by the model with the measure-
ments. Thus, it will allow to adjust the numerical results.
Various improvements could be brought to the model.
Another study, whose objective is to measure the statistical
periodicities of the coherent structures on a rough wall, will
allow to adjust the results obtained by the model with the
experimental results.
This study constitutes a rst step on studying the mod-
elling of wind erosion of multiple size grains bed.
Acknowledgements
This work was carried out with the nancial support
of ADEME (the French Agency for Environment and
Energy Management), ARCELOR Dunkerque and Fos-
sur-Mer (steelworks in France) and European Community
(FEDER).
Appendix A. Nomenclature
D
p
particle diameter (m)
E
erosion
erosion depth (m)
F
ad
adhesion force (N)
F
aeroinitial
initial instantaneous lift force (N)
F
aero
mean lift force (N)
F

aero
uctuating lift force (N)
H bed height (m)
l bed width (m)
L bed length (m)
P particle weight (N)
P
mass
mass percentage (%)
t time (s)
T
B
average periodicity of burst
occurence (s)
tdc cover rate: mass percentage of non-erodible
particles (%)
T
C
size range cover rate (%)
u

friction velocity (m s
1
)
U

ow velocity (m s
1
)
166 I. Descamps et al. / Chemical Engineering and Processing 44 (2005) 159166
Greek letters

x
+
adimensionalized average streamwise
periodicity of appearing of vortex structure

z
+
adimensionalized average spanwise
periodicity of appearing of vortex structure
kinematic viscosity (m
2
s
1
)
References
[1] R. Bouscaren, Emissions diffuses and fugitives in Journe dtudes
du CITEPA, 2002.
[2] M.A. Meunier, Envol et transport de particules en couche limite
turbulente instationnaire, Ph.D. Thesis, University of Valenciennes,
France, 1999.
[3] L. Li, L.W. Martz, Aerodynamic dislodgement of multiple size grains
over time, Sedimentology 42 (1995) 683694.
[4] M.V. Lpez, Wind erosion in agricultural soils: an example of limited
supply of particles available for erosion, Catena 33 (1998) 1728.
[5] L.Y. Liu, P.J. Shi, X.Y. Zou, S.Y. Gao, H. Erdon, P. Yan, X.Y. Li,
Z.B. Dong, J.H. Wang, Short terms dynamics of wind erosion of
three newly cultivated grasslands soils in Northern China, Geoderma
115 (2003) 5564.
[6] Schet, P., Contribution ltude des structures cohrentes en tur-
bulence de paroi, De leur inuence sur le transport des sdiments
dans le cas du charriage, Ph.D. Thesis, INPT, France,. 2000.
[7] Y. Nio, M.H. Garcia, Experiments on particle-turbulence interactions
in the near-wall region of an open channel ow: Implications for
sediment transport, J. Fluid Mech. 326 (1996) 285319.
[8] J. Jimenez, P. Moin, The minimal ow unit in near wall turbulence,
J. Fluid. Mech. 225 (1991) 213240.
[9] R.F. Blackwelder, J.H. Haritonidis, Scaling of the bursting frequency
in turbulent boundary layers, J. Fluid. Mech. 316 (1983) 285
306.
[10] A.M. Mollinger, F.T.M. Nieuwstadt, Measurement of the lift force
on a particle xed to the wall in the sublayer of a fully devel-
oped turbulent boundary layer, J. Fluid. Mech. 316 (1996) 285
306.
[11] A. Zimon, Adhesion of dust and powder, Plenum Publishing Corpo-
ration, 1982.
[12] I. Descamps, Erosion olienne dun lit de particules large spectre
granulomtrique, Ph.D. Thesis, University of Valenciennes France,
2004.
[13] J.M. Foucaut, M. Stanislas, Take-off threshold friction velocity of
solid particles lying under a turbulent boundary layer, Exp. in Fluids.
20 (1996) 377382.
[14] R. Kurose, S. Komori, Turbulence structure over a roughness, Int. J.
Mult. Pha. Flow. 27 (2001) 673683.
Chemical Engineering and Processing 44 (2005) 167173
Solids deposition in low-velocity slug ow pneumatic conveying
J. Li
a,
, C. Webb
a
, S.S. Pandiella
a
, G.M. Campbell
a
,
T. Dyakowski
a
, A. Cowell
b
, D. McGlinchey
b
a
Department of Chemical Engineering, Satake Centre for Grain Process Engineering, UMIST, P.O. Box 88, Manchester M60 1QD, UK
b
Centre for Industrial Bulk Solids Handling, School of Engineering, Science and Design, Glasgow Caledonian University, Glasgow G4 0BA, UK
Received 26 August 2003; received in revised form 4 December 2003; accepted 11 February 2004
Available online 25 September 2004
Abstract
Solids deposition in the horizontal pipeline of a pneumatic conveying system was studied both mathematically and experimentally. Math-
ematically modelled results using the coupled discrete element method (DEM) and computational uid dynamics (CFD) approach have
demonstrated an intensive exchange of particles between the stationary layer (deposited particles) and the moving slug and a variation of
solids concentration and pressure and velocity distributions across the slug. Slug ows were also visualised experimentally through a glass
section and analysed by a high-speed video camera. The amount of particle deposition in the pipeline after a conveying was calculated
by controlling the solids feeding rate using a rotary valve and by monitoring the solids ow out of the system using dynamic load cells.
Experimentally generated data have quantitatively shown a tendency of more solids deposition with lower gas mass ow rate in slug ows
except that, below a certain amount of solids mass ow rate, the deposition becomes independent of gas ow rate.
2004 Published by Elsevier B.V.
Keywords: Pneumatic conveying; Slug ow; Solids deposition; CFD model; DEM simulation
1. Introduction
Low-velocity slug ow pneumatic conveying has become
more and more common in many industrial sectors ranging
across chemical, pharmaceutical and food industries, due to
its advantages over conventional high-velocity suspension
ow in preventing product degradation and plant wear and
in delivering high throughput and efcient power utilisation
[1]. One of the distinguishing characteristics in such an op-
eration is the gravitational deposition of solid particles in
pipelines, which is due to the low gas velocity used, usu-
ally below the solids saltation velocity. A layer of particles
at the bottom of the horizontal conveying line is commonly
seen in these systems and this solids deposition signicantly
affects the transition and ow of slugs along the pipeline.
This paper presents mathematical and experimental stud-
ies of solids deposition characteristics in the horizontal
pipeline of a dense phase slug ow pneumatic conveying

Corresponding author. Tel.: +44 1236 878448;


fax: +44 1236 872837.
E-mail address: jintang.li@devro-casings.com (J. Li).
system. To obtain an insight into the slug ow and solids de-
position characteristics, a coupled discrete element method
(DEM) and computational uid dynamics (CFD) approach
was used to simulate the solids and gas ows. Solids de-
position in the horizontal section of the conveying system
was visualised through a glass section and recorded by a
high-speed video camera. The amount of particle deposition
was calculated by closely monitoring the solids ow into
and out of the system. Mathematically and experimentally
generated data were analysed in order to develop a useful
tool for the future prediction of solids deposition and slug
ow behaviours in low-velocity pneumatic conveying.
2. Characteristics of slug ow - DEM modelling
Although low-velocity slug owpneumatic conveying has
been studied intensively in the past two decades [18], due
to the practical restrictions of applying modern measuring
tools to gas and solids multiphase ows there is still a lack
of understanding of how gases and particles interact inside a
moving slug and how these interactions affect the formation
0255-2701/$ see front matter 2004 Published by Elsevier B.V.
doi:10.1016/j.cep.2004.02.011
168 J. Li et al. / Chemical Engineering and Processing 44 (2005) 167173
Fig. 1. Initial conditions used in the simulation of slug ow in pipes.
and ow of slugs. A recent study by Klinzing [8] demon-
strated that the slugging behaviours of solids in a pipeline
are complex, involving intensive interchanges of particles
between the stationary layer and the moving slug, i.e. pick-
ing up particles at the front of a slug and dropping off at
the tail. He also stated that a thin layer of particles in the
pipeline was essential for the development of stable slugs. To
obtain this realistic experimental condition several pre-runs
were conducted in his experiments to let particles lay down
in the pipeline. Obviously, a quantitative description of this
particle layer, which is equivalent to the amount of particle
deposition after a conveying, would bridge the understand-
ing of solids slugging behaviour and the performance of the
transport system.
To understand the mechanisms of solids slug ows, a
two-dimensional coupled DEM/CFD numerical model was
built to simulate the motion of a pre-formed slug (ca. 0.3 m
long) in a 1 m long horizontal 50 mm bore pipe as shown in
Fig. 1. The pipe was initially lled with a layer of particles,
approximately 15 mm thick at the bottom. (The thickness
of this stationary layer was determined based on experience
from previous experiments and computer test runs).
In the model, the solids phase was modelled as discrete
particles using DEM and the gas phase was modelled as a
continuum by CFD. The inuence of solid particles on the
gas ow was considered as a source of mass, momentum
and energy, via the change to its volume fraction, the aero-
dynamic forces exerting on the gas and the heat transfer with
the gas. The effect of the gas ow on particle motion was
calculated via the uid drag forces. (Detailed description of
the model is found in a previous publication of the authors,
Li and Mason [9]). This method allows individual particle
trajectories and interstitial gas ows to be traced, and lo-
cal gas-particle, particle-particle and particle-wall interac-
tions to be computed with the evolution of the slug ows.
The basic advantage of this coupled approach over conven-
tional continuum techniques is that DEM simulates effects
at the particle level. There is less need for global assump-
tions and the assembly response is a direct output from the
simulation [10]. DEM simulations have been used to model
different granular systems by a number of researchers and
have reported positive results: e.g. Tsuji et al. [11] on a u-
idised bed, Langston et al. [12] on hopper ow, Frank et al.
[13] on channel ow, Cleary [14] on ball mills and Stewart
et al. [15] and Kuo et al. [16] on solids mixing. The main
disadvantage of DEM modelling is that it requires a large
amount of resource to compute increased number of parti-
cles, which makes it difcult to model ne particles or in
three-dimensional geometries. Therefore, discrepancies are
expected when comparing the current two-dimensional sim-
ulations with experiments.
Particles used in this simulation were 5 mm spherical
polyethylene pellets with a particle density of 880 kg/m
3
.
The gas was introduced into the pipe inlet with a constant
supercial gas velocity of 2 m/s and all particles in the solids
slug and the stationary layer were assigned a low initial ve-
locity at 0.01 m/s to represent a dynamic starting condition.
The simulation started with an ambient pressure in the whole
pipe and this ambient pressure was kept constant at the pipe
outlet while the pressure rose in the pipeline when gas was
introduced from the inlet.
Fig. 2 shows six snapshots of slug ows along the pipe
from the DEM/CFD modelling at a time interval of 0.2 s for
a total time length of 1.0 s. This simulation demonstrates a
similar picture as described by Klinzing [8], in terms of par-
ticle motion in the slug. That is, particles in the stationary
layer (dark coloured particles at the bottom of the pipeline)
ahead of the slug wave are drawn into the slug (light coloured
particles) and form part of it when being moved along the
pipe by the pressurised gases. The slug wave actually com-
presses the particle layer just ahead of it, pushing some of
the particles up from the layer into the wave. It is also inter-
esting to see that at the tail of the slug particles fall into the
bottom part of the pipeline and form the stationary layer. The
exchange of particles between the stationary layer and the
slug has shown a unique slugging mechanism in slug ows,
which explains why particles move slower than slugs [8].
A statistical analysis shows that, after the slug ow, the
number of particles deposited in the pipe is about 90% of the
number of particles originally placed in the stationary layer.
Taking into account the entrance effect (much less particle
deposition in the front section from 0 to 0.3 m), this shows a
reasonable comparison between the assumed layer thickness
(15 mm) and the modelled particle deposition, though an
uneven particle layer has been produced after the slug ow.
This also suggests that the uneven particle layer from the
current model should be adopted as initial input for further
modelling.
It can be clearly seen fromFig. 2 that there is a variation of
solids concentration across the slug with a looser front and
a denser back. Some particles are even pushed ahead of the
slug wave at the front and this group of particles behave in a
very dynamic and suspended fashion. This phenomenon has
been conrmed by analysing video footages from experi-
ments in slug ows using similar particulate materials (3 mm
spherical polymer pellets), as shown in Fig. 3. Although the
general slugging behaviour is determined by the aeration,
permeability and de-aeration characteristics of the particu-
late material, the local variation of solids concentration in
the slug is very much dependent upon the actual interstitial
gas ow through the slug and the pressure distribution along
the pipe. Fig. 4 shows a plot of pressure drops along the
pipeline at both top and bottom of the pipe cross-section at
J
.
L
i
e
t
a
l
.
/
C
h
e
m
i
c
a
l
E
n
g
i
n
e
e
r
i
n
g
a
n
d
P
r
o
c
e
s
s
i
n
g
4
4
(
2
0
0
5
)
1
6
7

1
7
3
1
6
9
Fig. 2. Solids slug ow and deposition DEM/CFD simulation (time interval between each picture 0.2 s and total time 1.0 s).
170 J. Li et al. / Chemical Engineering and Processing 44 (2005) 167173
Fig. 3. Snapshots of slug ow from video footages recorded in experi-
ments.
an instantaneous time of 0.4 s. A clear response of pressure
build-up at the back of and across the slug is shown and
a pressure difference between top and bottom of the pipe
cross-section at the back of the slug is observed, both of
which have been conrmed by experiments using accurate
pressure measurement [17]. Obviously the pressure build-up
Fig. 4. Pressure drop along the pipeline DEM/CFD simulation (at 0.4 s).
Fig. 5. Gas velocity prole along the pipe DEM/CFD simulation (at 0.4 s).
is due to the blockage of gas passage by particles in the pipe
cross-section and the gas retention characteristics in front
of the slug plays an important role in determining solids
slugging behaviours. The presence of particles has severely
altered the gas ow eld in the vicinity of the slug. Figs. 5
and 6 show the simulated gas velocity proles horizontally
along the pipeline and vertically in the pipe cross-section
at different locations, respectively. In general the axial gas
velocity is much higher at the top of the pipe cross section
than at the bottom and the highest occurs at the back of the
slug (around 0.4 m), which follows the prole of the solids
assembly at the back of the slug. However, gas velocity pro-
les become much atter in the middle of the slug as shown
in Fig. 5 at 0.5 m. It is also interesting to see the change
of gas ow direction in the pipe cross-section as shown in
Fig. 6 (positive value indicating an up-ow and negative a
down-ow). An up-ow of gas is observed at the front of
the slug (0.6 and 0.7 m) and the far back of the slug (0.3 m)
and a down-ow at the back of the slug (0.4 m) but a com-
plex picture in the middle of the slug (at 0.5 m). In general,
these cross section velocity proles comply with the axial
velocity proles in terms of mass balance in the gas phase.
J. Li et al. / Chemical Engineering and Processing 44 (2005) 167173 171
Fig. 6. Gas velocity prole in pipe cross section DEM/CFD simulation (at 0.4 s).
The gas velocity proles may explain why a loose slug
front is generated. That is particles in the front of the slug
wave are pushed by the slug body and dragged ahead of
the slug by the high local gas velocity at the top pipe cross
section while being agitated through severe interactions in-
curred by the rising particles from the stationary layer be-
neath the slug front and oating over them. The agitated
particle layer underneath the slug front is actually acting as
multiple uidised rolling wheels for the slug, which greatly
reduces the frictional forces comparing with a direct inter-
action with the pipe bottom wall. In some cases, even an
air pocket or a particle swirling ow is created between the
suspended loose slug front and the bottom particle layer, of
which the resulting mechanisms and its consequence on slug
ow need further studies.
3. Experimental analysis
Experiments were conducted to determine quantitatively
the amount of particle deposition in the pipeline of a spe-
cially designed pneumatic conveying system [17,18]. A
schematic layout of the rig is shown in Fig. 7. Briey
Table 1
Solids deposition in a pneumatic conveying system
U
g
(ca.)
(m/s)
M
g
(kg/s) M
s1
M
s2
M
s3
M
s4
Fixed
(kg/s)
Deposition
(kg)
Fixed
(kg/s)
Deposition
(kg)
Fixed SLR5
(kg/s)
Deposition
(kg)
Fixed SLR10
(kg/s)
Deposition
(kg)
25 0.1534 1.2270 0.00 0.6135 0.00 0.7669 0.00 1.5337 0.00
8 0.0491 1.2270 3.94 0.6135 9.35
a
0.2454 11.35
a
0.4908 9.51
6 0.0368 1.2270 6.28 0.6135 7.29 0.1840 9.18
a
0.3681 5.43
4 0.0245 1.2270 6.05 0.6135 8.03 0.1227 8.33
a
0.2454 8.05
2.6 0.0160 1.2270 9.79 0.6135 8.92 0.0798 11.70 0.1595 12.88
a
a
Large difference ( > 2 kg) of load cell readings occurred between repeated runs, which may be due to the unstable ow of slugs in these ow
conditions.
it comprises of a feed vessel, a receiving vessel, a 13 m
long conveying pipeline with interchangeable 5.1, 7.6 and
10.2 cm (i.e. 2, 3 and 4 in.) nominal bore pipes, a bank of air
supply nozzles, a variable speed rotary valve with dropout
box, an air lter unit, and connecting pipes and control
valves. In the system, both gas and solids mass feeding
rates are closely controlled through the choked nozzle bank
and the variable speed rotary valve. A set of three load cells
is mounted underneath the receiving vessel to monitor the
weight of solids conveyed. This particular design allows the
desired solids mass ow rate to be selected by adjusting
the rotary valve speed while the actual conveying rate is
calculated simultaneously from the load cell readings. Fig.
8 shows a typical diagram of a non-suspension wave-like
ow. Each step in the mass collected history is caused by
the arrival of a slug of solids. It is noticed that there is a
difference between the mass of solids owing into and out
of the system and the difference in the nal masses is the
deposition of solids in the system, which usually forms a
particle layer in the pipeline at the end of conveying.
Different combinations of gas and solids ow rates were
tested in a 7.6 cm bore pipe with a common industrial gran-
ular material - polyethylene pellets. The pellets were spher-
172 J. Li et al. / Chemical Engineering and Processing 44 (2005) 167173
Fig. 7. Schematic layout of the pneumatic conveying rig.
ical in shape with an average diameter of 3 mm, a particle
density of 880 kg/m
3
and an approximate bulk density of
550 kg/m
3
. The experiments were started initially with high
supercial gas velocity at 25 m/s, in which very little particle
deposition occurred (This was conrmed with visualisation
via a 3 m glass section in the conveying line). When using
Fig. 8. Mass trace and solids deposition in slug ows.
low gas velocity (< 10 m/s), a particle layer was built at the
bottom of the pipeline after a slug passing and at the end of
each test case. This solids deposition was assumed equiva-
lent to the difference of the nal load cell readings between
the tested cases and that obtained using a high gas velocity
(25 m/s).
J. Li et al. / Chemical Engineering and Processing 44 (2005) 167173 173
Table 1 summarises solids depositions in the four groups
of tests conducted: M
s1
- xed mass ow rate of solids
with a higher rate at 1.2270 kg/s, M
s2
- xed mass ow
rate with a lower rate at 0.6135 kg/s, M
s3
- xed solids
loading ratio (ratio between solids and gas ow rates) SLR
= 5, and M
s4
- xed SLR = 10, totalling 20 experimen-
tal cases. From M
s1
, a strong tendency for more solids de-
position with a lower gas mass ow rate is shown. How-
ever, this effect does not appear to be true for either the
lower solids mass ow rate M
s2
, or the two groups with
xed SLR, M
s3
and M
s4
(also with lower mass ow rate
< 0.5 kg/s). It shows that below a certain amount of solids
mass ow rate the deposition of solids becomes independent
of gas ow rate, which can be explained by the uidised
motion of particles before a full slug moves and the grad-
ual build-up of long slugs in the down stream pipeline as
discussed previously in Li et al. [18]. Further experiments
with a wider range of ow conditions using advanced mea-
suring tools such as tomographic image analysis [19] are
required to validate the model and to build up a quantita-
tive description of solids deposition for different particulate
materials.
4. Conclusions
Gravitational solids deposition in a horizontal pipeline
of a pneumatic conveying system was studied both math-
ematically and experimentally. Mathematically modelled
results using the DEM/CFD simulation have demonstrated
an intensive exchange of particles between the slug body
and the stationary layer as slugs move along the pipeline,
and a large variation of solids concentration and pres-
sure and velocity distributions across the slug. A slug
wave actually compresses the particle layer just ahead of
it, pushing some of the particles up from the layer into
the wave. On the other hand, particles at the tail of the
slug fall into the bottom part of the pipeline and form a
stationary particle layer behind the slug. These phenom-
ena have been conrmed by analysing recorded video
footages.
Experimentally generated data have quantitatively shown
a tendency of increased solids deposition at lower gas mass
ow rate in slug ows except that, below a certain amount of
solids mass ow rate, the deposition becomes independent
of the gas ow rate, which is due to particle uidised motion
before a slug ow commences. Further work will focus on
experiments with a wider range of ow conditions using
sophisticated measuring tools to validate the model and to
build up a quantitative protocol for the description of solids
deposition in conveying pipes.
Acknowledgements
Funding for this project was provided in part by the Satake
Corporation of Japan.
References
[1] K. Konrad, Dense-phase pneumatic conveying: a review, Powder
Technol. 49 (1986) 135.
[2] T. Ramakrishnan, K. Ramakoteswara Rao, M.A. Parameswaran, S.
Sivakumar, Experimental investigation of a dense-phase pneumatic
transport system, Chem. Eng. Process. 32 (1993) 141147.
[3] A.J. Jaworski, T. Dyakowski, Investigations of ow instabilities
within the dense pneumatic conveying system, Powder Technol. 125
(2002) 279291.
[4] R. Pan, P.W. Wypych, Pressure drop and slug velocity in low-velocity
pneumatic conveying of bulk solids, Powder Technol. 94 (1997)
123132.
[5] Y. Tomita, K. Tateishi, Pneumatic slug conveying in a horizontal
pipeline, Powder Technol. 94 (1997) 229233.
[6] O. Molerus, A. Burschka, Pneumatic transport of coarse-grained
materials, Chem. Eng. Process. 34 (1995) 173184.
[7] A. Levy, Two-uid approach for plug ow simulations in horizontal
pneumatic conveying, Powder Technol. 112 (2000) 263272.
[8] G.E. Klinzing, Dense phase (plug) conveying observations and pro-
jections, in: A. Levy, H. Kalman (Eds.), Handbook of Conveying and
Handling of Particulate Solids, vol. 10, Elsevier, 2002, pp. 291301.
[9] J. Li, D.J. Mason, A computational investigation of transient heat
transfer in pneumatic transport of granular particles, Powder Technol.
112 (2000) 273282.
[10] J.P.K. Seville, U. Tuzun, R. Clift, Processing of Particulate Solids,
Blackie Academic & Professional, London, 1997, p. 372.
[11] Y. Tsuji, T. Kawaguchi, T. Tanaka, Discrete particle simulation of
two-dimensional uidized bed, Powder Technol. 77 (1993) 7987.
[12] P.A. Langston, U. Tuzun, D.M. Heyes, Distinct element simulation
of granular ow in 2D and 3D hoppers: dependence of discharge rate
and wall stress on particle interactions, Chem. Eng. Sci. 50 (1995)
967987.
[13] T.H. Frank, K.-P. Schade, D. Patrak, Numerical simulation and ex-
perimental investigation of a gas-solid two-phase ow in a horizontal
channel, Int. J. Multiphase Flow 19 (1993) 187198.
[14] P.W. Cleary, Predicting charge motion, power draw, segregation and
wear in ball mills using discrete element methods, Miner. Eng. 11
(1998) 10611080.
[15] R.L. Stewart, J. Bridgwater, Y.C. Zhou, A.B. Yu, Simulated and
measured ow of granules in a bladed mixer a detailed comparison,
Chem. Eng. Sci. 56 (2001) 54575471.
[16] H.P. Kuo, P.C. Knight, D.J. Parker, Y. Tsuji, M.J. Adams, J.P.K.
Seville, The inuence of DEM simulation parameters on the particle
behaviour in a V-mixer, Chem. Eng. Sci. 57 (2002) 36213638.
[17] D.J. Mason, J. Li, A novel experimental technique for the inves-
tigation of gas-solids ow in pipes, Powder Technol. 112 (2000)
203212.
[18] J. Li, S.S. Pandiella, C. Webb, T. Dyakowski, M.G. Jones, Analysis
of gas-solids feeding and slug formation in low-velocity pneumatic
conveying, Part. Sci. Technol. 21 (2003) 5773.
[19] T. Dyakowski, J.F.C. Jeanmeure, A. Jaworski, Applications of elec-
trical tomography for gas-solids and liquid-solids ows a review,
Powder Technol. 112 (2000) 174192.
Chemical Engineering and Processing 44 (2005) 175185
Identication of material specic attrition mechanisms
for polymers in dilute phase pneumatic conveying
Lars Frye
a,
, Wolfgang Peukert
b
a
Technische Universitt Muenchen, Lehrstuhl fr Feststoff und Grenzchenverfahrenstechnik (LFG),
Boltzmannstr. 15, 85748 Garching, Germany
b
Institute of Particle Technology, Friedrich-Alexander-Universitt Erlangen-Nrnberg, Cauerstr. 4, 91058 Erlangen, Germany
Received 22 July 2003; received in revised form 25 February 2004; accepted 19 March 2004
Available online 27 August 2004
Abstract
In the approach presented, attrition of bulk solids in pneumatic conveying is regarded as a result of a process function (stress conditions)
and a material function (inuence of material properties). For dilute phase conveying the process function was determined by employing
computational uid dynamics (CFD). Opposed to what was expected initially, the numerical simulations indicate that sliding friction apparently
is of importance in dilute phase conveying. This conclusion can be drawn from the low calculated impact angles in a pipe bend (r
B
/D
n
= 5;
D
n
= 80 mm) whichlie between5and35

. Consequently, the tangential impact velocitycomponents rangingfrom33to44 m/s are considerably


higher than the normal ones (525 m/s). These results were at rst endorsed by observations made in experiments to determine the material
function. The relative attrition behavior of four different polypropylenes (PP) under pure sliding friction conditions closely resembles that
observed in attrition experiments carried out in a pipe bend of the above geometry, while differences are observed for normal impact conditions.
But further experiments with polymethylmethacrylates (PMMA) and polystyrenes (PS) showed that here the attrition behavior encountered
under normal impact conditions is similar to that observed in the pipe bend. With dynamic mechanical analysis (DMA) it was possible to
show that PP on one hand and PMMA and PS on the other are attrited by different mechanisms and that the glass transition temperature of
the polymers is a key factor in the determination of the prevailing attrition mechanism.
2004 Elsevier B.V. All rights reserved.
Keywords: Attrition; Pneumatic conveying; Polymers; Process function; Material function; Dynamic mechanical analysis (DMA)
1. Introduction and approach
Pneumatic conveying belongs to the well established
processes for transporting a vast variety of bulk solids.
Especially in the handling of granular polymers it is used
extensively. Nevertheless, the problem of product degra-
dation due to attrition is unresolved to date. This is partly
due to the fact that research was focused on different topics
like bend erosion or pipe wear [1,2]. But with the increas-
Abbreviations: DMA, dynamic mechanical analysis; DSC, differential
scanning calorimetry; He, helium; PE, polyethylene; PMMA, polymethyl-
methacrylate; PP, polypropylene; PS, polystyrene; St-bend, steel disc with
topography similar to that of pipe bend; St-sb, steel disc with sandblasted
surface

Corresponding author. Tel.: +49 89 289 15656;


fax: +49 89 289 15674.
E-mail address: l.frye@lfg.mw.tum.de (L. Frye).
ing importance of engineered products possessing carefully
designed material properties, even smallest amounts of at-
trition are intolerable. Thus, focus has to be and is laid on
the investigation of bulk solids attrition [38].
In the present paper, the approach chosen at our institute
to study the phenomenon of attrition is briey presented.
In the results section, the link between bulk solids mate-
rial properties and attrition results is discussed. The ndings
document the suitability of our approach for identifying ma-
terial specic attrition mechanisms.
The approach chosen goes back to the work of Rumpf [9],
Krekel and Polke [10] and Peukert and Vogel [11]. Rumpf
was the rst who analyzed comminution processes by dis-
tinguishing between machine and material parameters. He
stated that the main parameters affecting the comminution
result are stress mode, stress intensity and number of stress
events as well as particle size and shape and material com-
position. This approach in which a mill is not regarded as a
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.012
176 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185
Fig. 1. Interpretation of attrition formation as a result of a process function
and a material function.
black box, but where the stress conditions of different mills
are used as characteristic features to distinguish between
them was later also applied by Krekel and Polke [10]. They
introduced the term process function which describes the
combined effect of process or machine parameters and ma-
terial properties on the chemical and physical product prop-
erties.
Peukert and Vogel [11] divided the process function into
a machine and a material function. The machine function
describes the connection between design and machine spe-
cic parameters and the driving potential for the engineering
operation, i.e. the stress conditions in a mill. The material
function summarizes the reaction of the materials used on
this driving potential. This concept was transferred to the
problem of attrition in pneumatic conveying. Since no sin-
gle part of machinery can be directly associated with this
process, the machine function was renamed as process or
conveying function in order to obtain an unambiguous termi-
nology. The process function thus differs from what Krekel
and Polke [10] have dened. The term material function is
used as introduced by Peukert and Vogel [11].
Consequently, attrition formation in pneumatic conveying
is interpreted as the result of a process function describing
the stress conditions the particles are subjected to in the con-
Table 1
Overview of different material properties
Name Shape x
p
(mm)
p
(kg/m
3
) H
V
(MPa) E (MPa)
y
(MPa) J
Qd
(kJ/m
2
)
PP 1040 N Elliptical 4.04 869.9 95.5 2,000 40 2.04
PP 1100 RC Elliptical 4.00 869.7 85.0 1,500 34 2.09
PP 1148 RC Elliptical 4.11 867.8 90.0 1,650 35 1.92
PP 2500 H Elliptical 4.06 896.3 68.0 1,100 23 n.a.
PE 2420 H Elliptical 3.28 919.3 61.5 260 11 n.a.
PE 5031 L Elliptical 3.46 946.3 84.9 1,000 26 4.00
PMMA G7 Cylindrical 3.20 1,191.8 208.9 3,200 n.a. 0.77
PMMA G55 Cylindrical 3.03 1,203.9 178.0 3,100 n.a. 0.65
PS 144 C Cylindrical 3.46 1,053.1 177.7 3,300 n.a. 1.54
PS 158 K Cylindrical 3.46 1,059.8 178.8 3,300 n.a. 1.99
veying pipelines and a material function summarizing the
material specic response to the process function in terms of
intrinsic material properties. This approach is schematically
shown in Fig. 1. The process function of course depends on
the mode of conveying. For dilute phase conveying attrition
will be mainly caused by particle wall and inter-particle
impacts, whereas in plug ow conveying friction is the
dominating stress mode. Stress intensities as well as the
number of stress events will vary greatly likewise. For the
material function, it is unknown which properties can be
used to at least qualitatively predict the material reaction
to the stress conditions. Both issues are discussed in detail
posterior to the description of the materials used for the
model experiments. Special attention is given to the deter-
mination of the material function while the process function
is only discussed in short. Both functions are of course
not fully independent, but can in our opinion be separated
satisfyingly for the investigation of attrition processes.
2. Material
Since pneumatic conveying is largely applied to trans-
port granular polymers and on the other hand even smallest
amounts of attrition of these solids cannot be tolerated the
results presented are focused on these materials. Polymers
of four chemically different polymer classes were examined.
Polypropylene (PP) and polyethylene (PE) belong to the
semicrystalline polymers, which possess both, an amorphous
phase and a crystalline phase. The polymethylmethacry-
lates (PMMA) and polystyrenes (PS) are fully amorphous.
Some material properties of the polymers are summarized
in Table 1.
The particle shape was determined by visual inspec-
tion of the particles. To determine the particle diameter
x
p
, the volume of individual particles was measured in a
He-pycnometer and the diameter of the sphere possessing
the same volume was calculated. The solid density
p
was
measured in the He-pycnometer as well. The Vickers hard-
ness H
V
was determined in a microhardness tester while
Youngs modulus E and yield stress
y
were given by
the manufacturer. Finally, the J-integral value J
Qd
which
L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185 177
corresponds to the critical energy release rate G
c
for materi-
als with a nonlinear deformation curve was measured at the
Institute of Polymeric Materials, Martin-Luther University
Halle-Wittenberg by employing a modied Charpy notched
impact test.
3. Comments on the process function
Since a detailed description of the determination of the
process function for dilute phase conveying has been given
elsewhere [12] only the main ndings are presented here.
The stress conditions were calculated by employing compu-
tational uid dynamics (CFD). The uid phase was modeled
by using the Eulerian approach, i.e. the Reynolds averaged
conservation equations for mass and momentum in conjunc-
tion with the k- turbulence model while for the particulate
phase the Lagrangian particle tracking as expressed in the
BassetBoussinesqOseen equation was applied. The parti-
cles were modeled as being spherical, particle wall impacts
as fully elastic and particle lift forces were neglected.
The computational domain consisted of a straight pipe
with a diameter D
n
of 80 mm and a length l
e
of 2400 mm.
Proceeding the pipe is a bend. For the simulations different
r
B
/D
n
ratios of 2, 5 and 10 corresponding to short, standard
and long radius bends were used. Details on the geometry
can also be found in the experimental section, since an in-
stallation possessing the above geometry of the straight pipe
in conjunction with a pipe bend of r
B
/D
n
= 5 was con-
structed. Due to the fact that the simulations were intended
to be as closely related to the experimental conditions as
possible, the gas and particle velocities measured in this in-
stallations were used as boundary conditions for the simula-
tions. Thus the average gas entrance velocity normal to the
inlet face was set to v
f,i
= 40.77 m/s. The particle entrance
velocities v
p,i
were intended to be equal to the gas velocity,
however due to experimental limitations it was not possible
to exactly match them with the gas velocity. Consequently,
Fig. 2. Cumulative number distributions for wall impact angles and normal and tangential impact velocity components for one polymer of each class.
depending on the material v
p,i
ranged from 40.22 to 45 m/s.
The material properties required for the simulations were
taken from Table 1.
In Fig. 2 the cumulative number distributions Q
0
of im-
pact angles and impact velocity components are shown for
selected polymers. It can be observed that with 535

the
calculated angles are comparatively small. This results in
low normal (525 m/s) and high tangential (3344 m/s) im-
pact velocity components which leads to the conclusion that
not only impact stresses contribute to attrition in dilute phase
conveying but that sliding friction stresses might play an im-
portant role as well. The differences in the distributions of
the tangential velocity component can be mainly attributed
to the already mentioned slight variations in the initial par-
ticle velocities.
Since the results were unexpected, the particle wall im-
pact locations were plotted from the data generated in the
simulations for a total of 2000 individual particles. They are
shown in Fig. 3 from two different angles. Due to the large
particle diameters resulting in high inertia forces the parti-
cles do not follow the gas stream in the pipe bend, but pro-
ceed directly to the outer bend wall. Only a slight sinking
caused by the gravity force can be observed. Since this be-
havior can be attributed to the high initial particle velocities,
simulations varying these have been carried out. They lead
to changes in the respective number distributions of impact
angle and velocity components, but did not have any effect
on the maximum angle. It can thus be concluded that not
only the bend geometry is of inuence as simulations with
varying r
B
/D
n
ratio have shown but that the curvature of
the pipe plays an important role for the impact conditions
as well. Although Fig. 3 clearly shows that the results are
reasonable an experimental validation of the simulations is
nevertheless still necessary.
For reasons of completeness, results for the determina-
tion of the process function for plug ow conveying shall
be stated briey as well. Since in this case a two phase ow
with high particle concentrations is regarded the options to
178 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185
Fig. 3. Visualization of particle wall impact locations for 2000 particles of PP 1040 N with a diameter x
p
= 4 mm and initial gas and particle velocities
v
x,i
41 m/s.
simulate this ow are very limited. Therefore, experimental
results obtained by Mi and Wypych [13,14] were used to
obtain at least a rough estimate of the stress conditions the
particles are subjected to. If it is assumed that the particles
in a plug are at least for short times xed relative to each
other, the highest relative velocities will occur between the
particles forming the outer layer of the plug and the pipe
wall. The stress mode in this case is sliding friction. From
measurements of the pressure exerted by a plug on the pipe
wall and by applying the Hertzian equations for elastic con-
tact, a contact pressure between a single particle and the
pipe wall of roughly 4 MPa is obtained.
4. The material function
4.1. Concept for determination
It is generally known that different bulk solids might ex-
hibit widely varying attrition rates even if conveyed in the
Fig. 4. General concept and experimental devices to determine the material function for attrition processes.
same mode, q.v. [3]. Since under these circumstances the
stress conditions are almost identical the signicance of the
material function becomes obvious. Unfortunately it is still
unknown which material properties determine the attrition
behavior. In the case of polymers the identication is dif-
cult due to their viscoelastic nature. Thus the concept for
the determination of the material function shown in Fig. 4
was developed.
The crucial point is to simulate the previously identied
basic stress modes impact and friction under well dened
stress conditions. This way, material properties can be re-
lated to attrition caused by these stress modes and the re-
spective attrition mechanisms in effect. For this purpose it
was chosen to perform single particle experiments in simple
experimental setups to realize the dened stress conditions.
Details on these setups are given in the next section.
Ideally, the stress conditions obtained from the process
function are used as input parameters for the single particle
experiments. But due to practical reasons this is not always
possible, since a compromise between stress conditions
L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185 179
closely related to the conveying process and conditions
where the attrition rates can be measured reliably has to
be found. The knowledge gathered from the single particle
experiments will be broadened by and compared to results
of the dened stressing of a particle ensemble. For this
purpose, experiments in a pipe bend are carried out to in-
vestigate the polymers under conditions closely related to
those in the conveying pipelines.
The results of the attrition experiments are compared to
various material properties in order to identify those rele-
vant for attrition. The long term goal is to develop a phys-
ical model of the attrition process which makes it possible
to predict the amount of attrition that will be encountered
in a pneumatic conveying installation by knowledge of the
process and material functions.
4.2. Experimental setup and parameters
The experimental setups of two out of the three installa-
tions used to date are shown in Fig. 4. The installation on the
left-hand side is used to stress single particles under normal
impact loads. It was originally developed by Schnert and is
described in Marktscheffel and Schnert [15]. The particles
are fed one by one into the center of a rotor (2) by means
of vibration (1). In the rotor, the particles are accelerated in
radial channels and nally hit the impact ring (3) under an
impact angle of 90

. The impact velocity is determined by


the number of revolutions of the rotor. Particles and attri-
tion debris are collected in the impact chamber (4) and can
be discharged through a tube at the bottom of this chamber.
The experiments were carried out under vacuum conditions
to eliminate any effects due to viscous drag.
For the attrition experiments approximately 25 g of poly-
mer particles were used. After three, six and nine consec-
utive impacts with a velocity of 40 m/s, the attrition rate A
was determined as the relative loss of mass (see Eq. (1)).
This procedure was necessary since the attrition rates for
some polymer classes were not measurable with satisfying
accuracy for lower impact numbers. Since the attrition pro-
cess is highly statistical, each experiment was repeated three
times. This also holds for the other experimental setups. In
the following diagrams, the median values of these three re-
peated experiments are plotted with the standard deviations
as error bars. For all experiments, the attrition rate A was
calculated according to Eq. (1), where M
i
denotes the initial
particle mass. M
a
is the mass of particles at the initial size
after the attrition experiment.
A =
M
i
M
a
M
i
(1)
The installation to simulate sliding friction shown on the
right-hand side of Fig. 4 was developed at our institute. In
design it resembles a classic pin-on-disc tribometer. Here,
a rectangular polymer particle cut from tensile bars with a
length of 10 mm, a width of 4 mm and a thickness of 4 mm
is pressed onto a rotating disc with a dened normal force
F. Due to its shape a constant contact area of 16 mm
2
and
thus a dened contact pressure p
c
is maintained throughout
the experiment. Due to sample geometry, the velocity over
the contact face varied by a maximum of 2.5% from the
median value. Through the number of revolutions n of the
disc and the distance r between the particle and the disc axis,
the sliding velocity v
s
is adjusted. The sliding distance s
results from the sliding velocity and the contact time t
c
. The
disc simulating the wall material can be changed to account
for different materials and surface topographies conveying
pipelines might exhibit.
The experimental parameters for the sliding friction ex-
periments, were chosen to simulate dense phase conveying
conditions. Basically, two different sets of experimental pa-
rameters were used. The polypropylenes were tested against
different wall materials. Here, steel possessing different sur-
face roughness values was used. Since a sandblasted surface
lead to high attrition rates, a low contact pressure of 61 kPa
as well as a long sliding distance of 1500 m were used. The
sliding velocities ranged from 4 to 8 m/s. Based on the nd-
ings from the process function for dense phase conveying,
a higher contact pressure of 0.61 MPa in conjunction with
a lower sliding distance of 200 m was used for further ex-
periments. To accomplish similar conditions to the experi-
ments in the pipe bend installation, here a steel disc with
approximately the same roughness value as the pipe bend
was installed. The parameters, especially the comparatively
high sliding distances had to be chosen in such a way, to
make sure that reproducibly measurable attrition rates were
obtained. As already mentioned, here the attrition rate was
determined according to Eq. (1) by weighing the single par-
ticles on a high precision scale.
The third installation was constructed to conduct single
particle attrition experiments under well dened stress con-
ditions which are closely related to conditions in industrial
conveying installations. For this purpose a pipe bend was
chosen. The exact setup is given in Fig. 5. As it has already
been discussed, the geometry of this setup was used for the
simulations to determine the process function in dilute phase
conveying.
The installation was constructed as a negative pressure
system with the blower being located downstream of the l-
ter. This way a high exibility in feeding the particles was
Fig. 5. Design and dimensions of the pipe bend installation.
180 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185
realized. For the experiments presented, the particles were
fed into the pipe along its center axis by a specially designed
feeder in which they are accelerated with pressurized air to
approximately 41 m/s prior to entering the pipe. The veloc-
ity of the particles was measured by photo-electric guards.
The pipe has a diameter D
n
of 80 mm and possesses the di-
mensions already presented in the process function section.
For separating the particles from the gas stream, a lter bag
was used. Particles and attrition debris are discharged at the
bottom of the lter housing. With the blower, gas velocities
in the pipe of up to 41 m/s are reached. They are measured
and monitored with a Prandtls tube.
Similar to the normal impact experiments, the attrition
rates were determined after three, six and nine impacts. Here
it is assumed that only the rst particle wall impact in the
pipe bend contributes signicantly to attrition formation and
that further impacts are of minor importance. Again 25 g
of the granular polymers were used. In order to maintain
conditions close to those in conveying processes the particle
velocity was set to be approximately equal to the gas velocity
of 41 m/s.
4.3. Results of the single particle experiments
The rst results of the single particle experiments carried
out for different polypropylenes have shown that the relative
attrition behavior changes between the normal impact and
the sliding friction tests [12]. This was taken as a chance
to identify which attrition mechanism dominates in the pipe
bend installation and thus presumably in dilute phase con-
veying. In Fig. 6, a comparison of the results obtained for
the normal impact experiments, the sliding friction experi-
ments and the pipe bend installation is given. The normal
impact velocity was 40 m/s. For sliding friction a vanadium
steel disc possessing a sandblasted surface was used. The
contact pressure was 61 kPa, the sliding distance 1500 m and
the sliding velocities ranged from 4 to 8 m/s. The particle
Fig. 6. Comparison of the attrition rates for different polypropylenes as obtained from experiments in the normal impact, sliding friction and pipe bend
experiments.
and gas entrance velocities in the pipe bend installation were
approximately 41 m/s.
In the middle diagram, the error bars were omitted for
reasons of clarity. Although the attrition rates cannot be
compared directly due to the different stress conditions, it
can be observed that the relative attrition behavior of the
polypropylenes in the sliding friction experiments is iden-
tical to that observed in the pipe bend experiments, i.e. in
both cases PP 2500 H is the least attrited material, while
PP 1148 RC is the most attrited one. For normal impacts,
considerable deviations from the results in the pipe bend are
found. This leads to the conclusion that attrition formation
can be mainly attributed to sliding friction. This is in good
agreement with the ndings from the process function for
dilute phase conveying discussed earlier. In case of PMMA
and PS a different behavior is observed (see Fig. 7). Here,
the relative attrition behavior found for the normal impact
experiments corresponds to that in the pipe bend.
These ndings clearly show the signicance of the ma-
terial function, since the material reaction to similar stress
conditions in the pipe bend can vary greatly. In contrast to
PP, PMMA and PS apparently are attrited by the same mech-
anisms in effect under normal impact stress conditions. This
also claries that it is only possible to deduce the stress mode
from the process function but that without knowledge of the
material function no conclusions concerning the governing
attrition mechanisms can be drawn.
4.4. Correlation between single particle experiments and
material properties
One of the most important tasks in the determination of the
material function is to identify those material properties rele-
vant to attrition formation. Acomparison of the above results
to classic mechanical material properties, e.g. Youngs mod-
ulus E or Vickers hardness H
V
as given in Table 1 suggests
that these are not suitable to describe attrition formation.
L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185 181
Fig. 7. Comparison of the attrition rates for different PMMA and PS granules as obtained from normal impact, sliding friction and pipe bend experiments.
This is due to the fact that the respective measurement pro-
cedures for their determination do not take into account the
dynamic nature of the attrition process. Therefore, in the
present study dynamic mechanical analysis (DMA) was em-
ployed.
For the measurements, the single cantilever exure mode
was chosen. For this, rectangular polymer samples of
32 mm 6 mm 2 mm had to be produced. This was done
by injection molding of the original polymers. Fig. 8 shows
the setup of a DMA-experiment in the single cantilever
conguration.
Prior to the experiments, the polymer sample is xed be-
tween two clamps and the sample dimensions are measured
carefully. One of the two clamps is xed, while the other
can be moved up and down. To this clamp, a sinusoidal de-
formation with an amplitude
A
and a frequency is ap-
plied and the corresponding force is measured, which can be
transferred into the stress
A
with the sample dimensions.
From these two quantities the complex modulus of elastic-
ity E

is measured in dependence of sample temperature T


and oscillation frequency .
|E

(, T)| =

A

A
(2)
The viscoelastic behavior of the polymers is reected in
the phase lag between the deformation signal and the cor-
Fig. 8. Setup of a DMA-experiment in the single cantilever conguration.
responding stress signal. For a fully elastic material, would
be zero whereas it would be 90

for a fully plastic material.


With this phase lag, which is measured as well, the com-
plex modulus of elasticity can be divided into the storage
modulus E

and the loss modulus E

.
E

(, T) = |E

| cos (3)
E

(, T) = |E

| sin (4)
The storage modulus is proportional to the amount of
energy which is stored in the material elastically, whereas
the loss modulus corresponds to the energy that is dissipated
during one load cycle. Both quantities are combined in the
damping factor tan which is dened as
tan =
E

(5)
After checking the inuence of various experimental pa-
rameters as well as the reproducibility of the measurements,
a standard procedure was developed. The polymer samples
are cooled quickly with liquid nitrogen to a temperature of
140

C. After an isothermal period of 1 min, a frequency


sweep is carried out for 1, 3, 5, 10, 30, 40, 50 and 60 Hz with
a deformation amplitude of 20 m. Afterwards the temper-
ature is increased by 5

C and after another isothermal pe-


riod of 1 min, the next frequency sweep is carried out. This
182 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185
Fig. 9. Storage (E

) and loss (E

) modulus functions for different PP, PMMA and PS granules.


procedure is repeated until the value of the storage modu-
lus drops down to 10 MPa which was set as the criterion
to end the measurement in time before melting of the sam-
ple starts. The obtained storage and loss modulus functions
for the polymers are depicted in Figs. 9 and 10. The values
given were obtained for a frequency of 10 Hz.
From the measurements the typical viscoelastic behavior
of the polymers becomes obvious. The polymer chains are
xed relative to each other at low temperatures. Thus the
amount of energy that is stored elastically is high which is
reected in a high value of the storage modulus. The corre-
sponding values of the loss modulus are low. The mobility
of the polymer chains increases with increasing tempera-
ture and consequently the storage modulus decreases while
the loss modulus increases. In the glass transition regime
the maximum of energy dissipation is reached (maximum
of loss modulus) and the storage modulus falls rapidly.
The structural differences between the semicrystalline
polypropylenes/polyethylenes and the amorphous poly-
Fig. 10. Storage (E

) and loss (E

) modulus functions for different PP and PE granules.


methylmethacrylates/polystyrenes are reected as well.
After the glass transition, the storage modulus reaches a
plateau for PP and PE due to the crystalline structures which
do not change in stiffness signicantly until melting. The
amorphous polymers on the other hand fully soften during
glass transition leading to very low storage modulus values.
In Table 2 the glass transition temperatures T
g
as obtained
from the storage and loss modulus values are given. For the
storage modulus they were determined by using the inec-
tion point method whereas for the loss modulus the peak
maximum was used [16].
It can be observed that the agreement of the correspond-
ing temperatures is quite well. As Ehrenstein et al. [16] point
out, this deviation is normal and usually T
g
(E

) T
g
(E

)
T
g
(DSC) holds. The materials where a second temperature
value is given in italics do not only possess a glass transi-
tion but also show a second order transition. For PP 2500 H,
PMMA G7 and G55 these second order transitions which
are most likely caused by increased mobility of side-chains
L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185 183
Fig. 11. Comparison of the attrition rates obtained from sliding friction experiments for 6 m/s, 0.61 MPa and 200 m against St-bend with storage (E

) and
loss modulus (E

) values measured at 10 Hz and 25

C.
or chain segments lie below their glass transition tempera-
tures. PE 2420 H and PE 5031 L on the other hand show
their respective second order transitions above their glass
transition temperatures (see Fig. 10). Here it is not quite
clear what causes them.
If the ambient temperature at which the single particle ex-
periments were carried out is regarded as the macroscopic
stress temperature, it can be observed with respect to the
experimentally obtained attrition rates, that the polypropy-
lenes which are attrited by sliding friction are stressed in
their glass transition regimes whereas the glass transition
temperature of PMMA and PS lies well above the stress
temperature. Obviously, the relation between the values of
the macroscopic stress temperature and the glass transi-
tion temperature can be used to evaluate which material
specic attrition mechanism impact or sliding friction
is dominating in attrition formation. This conclusion is
also valid for PE 5031 L, while PE 2420 H which is a
very soft material (see Youngs modulus in Table 1 or stor-
age modulus in Fig. 10) does not t. In this case, addi-
tional effects like adhesion are likely to play an important
role.
For PP it was furthermore found that the attrition rates
observed in the sliding friction experiments decrease with
decreasing values of both storage and loss modulus. A
Table 2
Glass transition temperatures as obtained from DMA measurements ( =
10 Hz)
Name T
g
(E

) (

C) T
g
(E

) (

C)
PP 1040 N 17.54 15.08
PP 1100 RC 15.04 15.10
PP 1148 RC 15.16 15.08
PP 2500 H 37.33/12.43 34.88/9.99
PE 2420 H 122.32/12.02 114.82/9.62
PE 5031 L 107.31/45.03 109.81/55.12
PMMA G7 45.07/107.69 29.90/110.20
PMMA G55 90.12 25.05/95.12
PS 144 C 92.13 95.17
PS 158 K 107.84 110.20
possible explanation for this behavior can be found in
the microscopic attrition mechanisms known from tribol-
ogy and was briey discussed in [17]. As Fig. 11 shows,
PE 5031 L ts well into this tendency while PE 2420 H
again does not. From the materials which are attrited by
an impact mechanism, PMMA G7 and G55 also follow
the above trend while the two polystyrenes do not. It is
currently under investigation how this observation can be
explained.
If the above comparison is made for the attrition results
from the pipe bend installation, again a decrease in attrition
rate for the polypropylenes corresponds to decreasing E

and
E

values. Of course, the scatter gets larger since in this


installation although sliding friction is the most prominent
attrition mechanism, impact also marginally contributes to
overall attrition. Furthermore, PE 5031 L, PMMA G7 and
PMMA G55 do not t quite as well as for the sliding friction
experiments while PE 2420 H, PS 144 C and PS 158 K
shows the same discrepancies as before.
The results presented show that three levels have to be dis-
tinguished when investigating attrition processes. The rst
one is the stress mode as derived from the process function
which is essential to know if the attrition process is to be
simulated successfully in a simple experimental setup. The
second point is the material reaction to this stress mode,
i.e. the material function which varies depending on mate-
rial properties like storage and loss modulus as measured
by DMA. Finally, the microscopic attrition mechanisms (see
[18] for impact and [19,20] for sliding friction) describing
the formation of attrition on a microscopic scale constitute
the bottom level.
5. Conclusions
In the present research project the problem of attrition
in pneumatic conveying was addressed by regarding it as
the result of a process and a material function. The process
function summarizes the inuence of all process parame-
ters and as a result provides the stress conditions in terms
184 L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185
of stress mode, intensity and number of stress events. The
material function on the other hand contains material prop-
erties which govern the attrition behavior of individual bulk
solids and as a result supplies an attrition function. With
knowledge of these two functions it should be possible to
predict the amount of attrition that is encountered in pneu-
matic conveying.
For dilute phase conveying numerical simulations with a
commercial computational uid dynamics code were carried
out. The analysis of particle wall impact conditions in a pipe
bend showed that they take place under low wall impact an-
gles of 535

which results in low normal (525 m/s) and


high tangential (3344 m/s) impact velocity components.
These ndings lead to the conclusion that not only normal
stresses caused by the impacts are important in dilute phase
conveying but that sliding friction stresses play an important
role as well.
This was corroborated by experimental results obtained in
single particle experiments carried out for polypropylenes. It
was possible to showthat the relative attrition behavior under
pure sliding friction stresses was identical to that observed
in a pipe bend whereas deviations to that under pure normal
impact conditions were found. For polymethylmethacrylate
and polystyrene on the other hand, the relative attrition be-
havior in the pipe bend corresponds to that of the normal
impact experiments.
An explanation for this was deduced from material prop-
erties obtained by dynamic mechanical analysis. It was
found that the glass transition temperature or glass tran-
sition regime is a key factor in the determination of the
mechanism the polymer particles are attrited by. If the stress
temperature lies below the glass transition temperature, an
impact-like mechanism is observed (PMMA, PS) while in
the other case friction is the dominating mechanism (PP,
PE). It is thus possible to identify the dominating attrition
mechanism by knowledge of the glass transition and the
macroscopic stress temperature. Furthermore the results in-
dicate that both a low storage modulus as well as a low loss
modulus are feasible for low attrition rates if sliding friction
is the dominating attrition mechanism. This correlation was
also observed for the PMMA granules but not for PS. An
explanation for this has still to be found.
The investigated polypropylenes are difcult to catego-
rize. While PE 5031 L ts in the above ndings and appar-
ently is mainly attrited by sliding friction as expected from
analysis of the DMA measurements, PE 2420 H does not t
into this scheme. Here different mechanisms like adhesion
might play an important role, which can be neglected for
the other polymers.
Nevertheless, the presented results indicate that if it is
carefully distinguished between the process function (stress
mode), the material function (material specic attrition
mechanisms) and microscopic attrition mechanisms (attri-
tion formation on the microscale) in the analysis of attrition
processes, signicant progress in the understanding of the
complex phenomenon of attrition can be made.
Acknowledgements
This project is sponsored by the German Federal
Ministry of Economics and Technology, grant AiF-No.
13252 N/1. The authors appreciate the nancial sup-
port.
Appendix A. Nomenclature
A attrition rate (%)
D
n
pipe diameter (m)
E Youngs modulus (Pa)
E

complex modulus (Pa)


E

storage modulus (Pa)


E

loss modulus (Pa)


F force (N)
G energy release rate (kJ/m
2
)
H microhardness (Pa)
J
Qd
J-integral (kJ/m
2
)
L, l length (m)
M mass (kg)
n number of revolutions (s
1
)
p pressure (Pa)

Q heat ux (W/s)
r radius (m)
s sliding distance (m)
t time (s)
T temperature (

C)
v velocity (m/s)
x
p
particle diameter (m)
x, y, z cartesian coordinates (m)
Greek symbols
angle (

)
phase lag ()
deformation (m)
frequency (Hz)
density (kg/m
3
)
stress (Pa)
Subscripts
a after attrition
A amplitude
B bend
c critical, contact
e entrance
f lter, uid
g glass transition
i initial
p particle
s sliding
V Vickers
x, i initial component in x-direction
y yield
L. Frye, W. Peukert / Chemical Engineering and Processing 44 (2005) 175185 185
References
[1] D. Mills, J.S. Mason, Particle size effects in bend erosion, Wear 44
(1977) 311328.
[2] P.W. Wypych, P.C. Arnold, Minimizing wear and particle damage
in pneumatic conveying, Powder Handl. Process. 5 (1993) 129
134.
[3] H. Kalman, Attrition of powders at various bends during pneumatic
conveying, Powder Technol. 112 (2000) 244250.
[4] H. Kalman, M. Hubert, E. Grant, Fatigue behavior of impact com-
minution and attrition units, in: Proceedings of the 10th Euro-
pean Symposium on Comminution (ESC), Heidelberg, Germany,
2002.
[5] D. Papadopoulos, M. Ghadiri, Impact breakage of polymethyl-
methacrylate (PMMA) extrudates. I. Chipping mechanism, Adv. Pow-
der Technol. 7 (1996) 183197.
[6] A.D. Salman, M.J. Hounslow, A. Verba, Particle fragmentation in
dilute phase pneumatic conveying, Powder Technol. 126 (2002) 109
115.
[7] A.D. Salman, C.A. Biggs, J. Fu, I. Angyal, M. Szab, M.J. Houn-
slow, An experimental investigation of particle fragmentation us-
ing single particle impact studies, Powder Technol. 128 (2002) 36
46.
[8] A.D. Salman, J. Fu, D.A. Gorham, M.J. Hounslow, Impact breakage
of fertiliser granules, Powder Technol. 130 (2003) 359366.
[9] H. Rumpf, Die Einzelkornzerkleinerung als Grundlage einer tech-
nischen Zerkleinerungswissenschaft, Chemie-Ingenieur-Technik 37
(1965) 187202.
[10] J. Krekel, R. Polke, Qualittssicherung bei der Verfahrensentwick-
lung, Chemie-Ingenieur-Technik 64 (1992) 528535.
[11] W. Peukert, L. Vogel, Product Engineering am Beispiel des Zerklein-
erns von Kunststoffen, Chemie-Ingenieur-Technik 73 (2001) 327
331.
[12] L. Frye, W. Peukert, Attrition of bulk solids in pneumatic conveying:
mechanisms and material properties, Part. Sci. Technol. 20 (2002)
267282.
[13] B. Mi, P.W. Wypych, Pressure drop prediction in low-velocity pneu-
matic conveying, Powder Technol. 81 (1994) 125137.
[14] B. Mi, P.W. Wypych, Investigations into wall pressure during slug
ow conveying, Powder Technol. 84 (1995) 9198.
[15] M. Marktscheffel, K. Schnert, Liberation of composite particles by
single particle compression, shear and impact loading, in: Preprints
of 6th European Symposium on Comminution, Nuremberg, Germany,
1986, pp. 2945.
[16] G.W. Ehrenstein, G. Riedel, P. Trawiel, Praxis der Thermischen
Analyse von Kunststoffen, Carl Hanser, Mnchen, second ed., 2003,
p. 269.
[17] L. Frye, W. Peukert, Neue Anstze zum Studium des Frdergutabriebs
bei pneumatischer Frderung, Schttgut 7 (2001) 435442.
[18] J.A.S. Cleaver, M. Ghadiri, Impact attrition of sodium carbonate
monohydrate crystals, Powder Technol. 76 (1993) 1522.
[19] R. Kaneko, T. Miyamoto, E. Hamada, Microwear, in: B. Bhushan
(Ed.), Handbook of Micro/Nano Tribology, CRC Press, Boca Raton,
1995, pp. 183221.
[20] H. Czichos, K.-H. Habig, Tribologie Handbuch: Reibung und
Verschlei, Vieweg, Braunschweig, 1992, pp. 100144.
Chemical Engineering and Processing 44 (2005) 187192
Two-dimensional numerical simulations of the
pneumatic drying in vertical pipes
I. Skuratovsky, A. Levy

, I. Borde
Department of Mechanical Engineering, Ben-Gurion University of the Negev, Beer-Sheva 84105, Israel
Received 5 August 2003; received in revised form 16 December 2003; accepted 11 February 2004
Available online 25 June 2004
Abstract
The Two-Fluid model has been used for modeling the ow of particulate materials through pneumatic dryer. The model was solved for
a two-dimensional steady-state condition and considering axial and radial proles for the ow variables. A two-stage drying process was
implemented. In the rst drying stage, heat transfer controls evaporation from the saturated outer surface of the particle to the surrounding
gas. At the second stage, the particles were assumed to have a wet core and a dry outer crust; the evaporation process of the liquid from a
particle assumed to be governed by diffusion through the particle crust and by convection into the gas medium. As evaporation proceeds, the
wet core shrinks while the particle dries.
Our developed model was applied to simulate drying process of wet sand in a pneumatic dryer. The numerical solutions were compared
with the results of independent numerical and experimental investigations. These comparisons are quite encouraging.
2004 Published by Elsevier B.V.
Keywords: Pneumatic drying; Two-uid model
1. Introduction
Pneumatic drying is a widely used process in chemical
industries and includes simultaneous conveying, and heat
and mass transfer between the particles and the hot gas.
The large surface area for heat and mass transfer and the
high convective heat and mass transfer coefcients, which
appear in these units, are leading to high drying rates and
as a consequence to high drying capacity. The increase in
the use of this unit operation requires reliable mathematical
models to predict processes in the industrial facilities.
Three types of theoretical approaches can be used for
modeling the gas-particles ows in the pneumatic dry-
ers, namely: Two-Fluid Theory [1], Eulerian-Granular [2]
and the Discrete Element Method [3]. Traditionally the
Two-Fluid Theory was used to model dilute phase ow.
In this theory, the solid phase is being considering as a
pseudo-uid. It is assumed that both phases are occupy-
ing every point of the computational domain with its own
volume fraction. Thus, macroscopic balance equations of
mass, momentum and energy for both the gas and the solid

Corresponding author.
E-mail address: avi@bgumail.bgu.ac.il (A. Levy).
phases are formulated. Often the ow in a pneumatic dryer
can be classied as dilute.
One-dimensional mathematical models are widely applied
to calculate drying processes, which can predict average
values of different properties of the phases in cross-sections
of the dryer [49].
Rocha and Paixo [10] presented a pseudo two-dimensional
mathematical model for a vertical pneumatic dryer that
was based on the two-uid approach. Axial and radial
proles for gas and solid velocity, water content, porosity,
temperatures and pressure were considered. Their predic-
tions were not validated with experimental results. Another
two-dimensional numerical investigation was conducted by
Silva and Nebra [11] for drying of particulate material in
cyclone. They used the two-uid approach and presented
steady-state drying model that considered incompressible,
axi-symmetric, turbulent gassolids ow, slip condition
of particles on the wall, particleswall heat transfer and
particles shrinkage. Gravity force effect on the spherical
particle was neglected. The predictions of their numerical
simulations were compared with experimental results and
it was claimed that the parameters most inuencing the
predictions are the particles slip conditions on the walls and
material shrinkage during the drying process.
0255-2701/$ see front matter 2004 Published by Elsevier B.V.
doi:10.1016/j.cep.2004.02.012
188 I. Skuratovsky et al. / Chemical Engineering and Processing 44 (2005) 187192
Blasco et al. [12] proposed two-dimensional mathemati-
cal model for the drying process of dense phase pneumatic
conveying. However, heat and mass transfer were not consid-
ered and therefore their model may be used for dense phase
pneumatic transport only. In their paper, both experimental
and numerical predictions for axial and radial proles for
gas and solid velocity, axial proles for solid concentration
and pressure drop were presented.
Contrary to traditional one-dimensional models, two-
dimensional models are required for taking into consid-
eration the effects of the radial distribution of the most
inuential thermophysical properties. As can be seen from
the above literature survey, not many studies have adopted
the two-dimensional approach for simulation the drying
process in a vertical tube. In addition, their predictions were
not validated with experimental data and in some cases only
the momentum transfer was taken into account.
In the present study, two-dimensional Two-Fluid Eulerian
model was used to describe the steady state, dilute phase ow
of a wet dispersed phase (wet solid particles) in a continuous
gas phase through a pneumatic dryer. The predictions of
the numerical solutions were compared successfully with
the results of other one-dimensional numerical solutions and
experimental data of Baeyens et al. [5] and Rocha [13]. Axial
and the radial distributions of the characteristic properties
were examined.
2. Theoretical model
A two-dimensional, Two-Fluid model has been developed
for modeling the steady-state ow of particulate materi-
als through vertical pneumatic dryer. Two-dimensional ow
eld interpretations provide information about properties of
the continuous and dispersed phases at every point of the
ow eld. Axial and radial proles were obtained for the
ow variables.
The gas phase was treated as a mixture of gas and liquid
vapor. The wet particle is made of porous solid structure
that contains saturated liquid. The drying model considered
a two-stage drying process. In the rst drying stage, heat
transfer controls evaporation fromthe saturated outer surface
of the particle to the surrounding gas. In the second stage,
the particles are assumed to have a wet core and a dry outer
crust. The evaporation of the liquid from a particle assumed
to be governed by diffusion through the particle crust and by
convection into the gas medium. As evaporation proceeds,
the wet core shrinks as the particle dries.
The developed model was solved numerically and applied
to simulate drying of wet sand in a laboratory-scale pneu-
matic dryer.
The drying model is based on the following assumptions:
Steady-state ow.
The ow in the dryer is vertical, two-dimensional,
non-rotational and axi-symmetrical.
Non-slip and non-penetrating wall conditions.
Dispersed phase behavior like continuous phase.
The particles are made from a substance containing solid
porous matrix and liquid or gas with liquid vapor.
The specic densities of the liquid and the solid, which
compose the particle, are constants.
Mass, momentum and heat transfer occur only between
the two phases and not between the particles themselves.
The gas phase is a mixture of water vapor with air.
The gas behaves as ideal gas.
Inuence of gravity on gas phase is neglected.
Heat transfer can occur between the pipe wall and the
continuous gas phase.
The acting forces per unit volume between the phases are
due to drag and virtual-mass.
At the rst drying stage, the particle is isothermal and at
the second drying stage the particle wet core temperature
is different from that of dry crust.
The inuence of the pressure gradient on the inertia of the
solid particles is negligible in comparison to that of the
drag force.
Electrical and surface tension forces are neglected.
Based on the above-mentioned assumptions, the mass,
momentum and energy balance equations for the gas and the
dispersed phases in two-dimensional, two-phase ow were
developed [14]. In order to solve the mass, momentum and
energy balance equations, several complimentary equations,
denitions and empirical correlations were required. These
were presented by [14]. In order to obtain the water vapor
distribution the gas phase the water vapor diffusion equa-
tion was added. During the second drying period, the model
assumed that the particle consists of a dry crust surround-
ing a wet core. Hence, the particle is characterized by two
temperatures i.e., the particles crust and core temperatures.
Furthermore, it was assumed that the heat transfer from the
particles crust to the gas phase is equal to that transferred
from the wet core to the gas phase i.e., heat and mass cannot
be accumulated in the particle crust, since all the heat and
the mass is transferred by diffusion through the crust from
the wet core to the surrounding gas. Based on this assump-
tion, additional heat balance equation was written.
The mass transfer source term was obtained by multiply-
ing the evaporation rate from a single particle by the total
number of particles per unit volume. Two-stage drying pro-
cess was applied for calculating the mass transfer from a wet
particle to the gas phase [6]. In the rst drying period, the
gas phase resistance controls the evaporation rate. When the
liquid evaporated from the particle surface and lls all the
voids inside the porous particle, the second drying period
starts. During this drying period, the dry crust causes addi-
tional resistance to heat and mass transfer. This resistance,
which occurs between the outside diameter of the particle
and the diameter of the wet core is governed by a diffu-
sion process. The evaporation rate from a particle with a dry
crust was expressed by Stephan-type diffusion rule [15]. The
I. Skuratovsky et al. / Chemical Engineering and Processing 44 (2005) 187192 189
Fig. 1. Comparison between the predictions of the present pneumatic drying model, DryPak model, Rocha in [20] and Levy and Borde [8] models and
the experimental data for (a) gas temperature, (b) solid temperature, (c) gas humidity and (d) particles moisture content along the dryer length.
mass transfer coefcient was calculated in analogy to the
heat transfer coefcient from the Sherwood number, which
is equivalent to the Nusselt number. It should also be noted
that both the gas and the dispersed phases are mixtures, and
hence all their thermodynamic properties were calculated
using the mixture theory.
3. Numerical simulations
The pneumatic drying model was solved numerically for
the drying processes of sand particles. The numerical proce-
dure includes discretization of the calculation domain into
torus-shaped nal volumes, and solving the model equa-
tions by implementation of the semi-implicit method for
pressure-linked equations (SIMPLE) algorithm[16]. The nu-
merical procedure also implemented the Interphase Slip Al-
gorithm (IPSA) of [17] in order to account the various cou-
pling between the phases. The simulation stopped when the
moisture content of a particle falls to a predened value or
when the ow reaches the exit of the pneumatic dryer.
In order to validate the theoretical model, the predic-
tions of the numerical simulations were compared with
experimental data. Comparison between the results of the
present two-dimensional model and the results of experi-
mental data and one-dimensional models requires a repre-
sentation model for presenting the two-dimensional data in
one-dimensional terms. Thus, an averaging technique was
used to obtain the average values of the various solution
properties at the dryer cross-section. The average value of
a property, , at the dryer cross-sectional area was obtained
by mass weighted method and an average temperature of
a phase was evaluated by the mean bulk temperature across
the tube.
=
2

R
0
urdr
2

R
0
urdr
; T
ave
=
2

R
0
uc
p
Trdr
2

R
0
uc
p
rdr
(1)
4. Numerical results and discussion
The predictions of the numerical simulations with
Baeyens et al.s [5] correlation for heat transfer were also
compared with the experimental results of Rocha [13] (pre-
sented by Silva and Correa [18]) that were obtained in a
4-m high pneumatic dryer with diameter of 5.25 cm. In this
study, 380-m sand particles having density of 2622 kg/m
3
and solid mass ow rate of 4.74 10
3
kg/s were dried
190 I. Skuratovsky et al. / Chemical Engineering and Processing 44 (2005) 187192
Fig. 2. The predictions of the pneumatic drying model for (a) gas phase velocity, (b) dispersed phase velocity, (c) gas phase temperature, (d) dispersed
phase temperature, (e) gas humidity, (f) particles moisture content and (g) particles diameter along stream lines and the dryer length.
I. Skuratovsky et al. / Chemical Engineering and Processing 44 (2005) 187192 191
with 3.947 10
2
kg/s air mass ow rate. Comparison
between the predictions of the numerical simulations and
the experimental data for changes of gas temperature, solid
temperature, gas humidity and particles moisture content
with length under known wall temperature operating condi-
tions is presented in Fig. 1ad, respectively. When known
wall temperature operating conditions were simulated, it
was assumed that in average the pipe wall temperature is
just about the outlet air temperature and it is falling linearly
from 360 K at the inlet to 354 K at the outlet. In this way,
heat losses from the dryer walls are considered indirectly.
In these gures, the star symbols represent the experimen-
tal data of Rocha [13] that were published by Silva and
Correa [18] and the other curves and symbols represent the
predictions of various models and numerical simulations.
Form these gures, it can be seen that the predictions of the
two-dimensional numerical simulations t fairly well with
the experimental data and both one-dimensional DryPak
[19] and Levy and Borde [8] models. Detailed comparison
between the various one-dimensional models can be found
at Levy and Borde [20]. Since the theoretical pneumatic
drying model and the numerical simulations done in the
course of this study predicted successfully the most impor-
tant ow parameters (gas temperature, solid temperature,
gas humidity and particles moisture content) of the drying
process, it was concluded that the presented model could
be used for further investigations. The distributions of the
various ow properties along the dryer length for various
stream lines (i.e., xed r/R) are presented in Fig. 2.
Form these gures, it can be seen that a fully developed
ow was obtained after about 3 m from the dryer inlet. In
addition, various thicknesses of boundary layers were ob-
tained for the various phases properties. This may inuence
the nal product properties.
5. Conclusions
Two-Fluid, two-dimensional model was developed for
modeling the dilute ow of particulate materials through
pneumatic dryer. The model was solved for a steady-state
condition. Axial and radial proles for the ow variables
were obtained. The developed model was applied to sim-
ulate the drying process of wet sand in a laboratory-scale
pneumatic dryer. The numerical solutions were compared
with the results of other numerical and experimental in-
vestigations. The successful comparisons validate the theo-
retical and the numerical models. As demonstrated by the
two-dimensional numerical simulations, the radial distribu-
tion of the ow characteristics causes an uneven product
quality. Thus in order to improve the drying process, a re-
liable two-dimensional model is required. Further investi-
gations are needed in order to validate the two-dimensional
distribution of the ow characteristic and to improve the
model to account for particle circulation in the investigated
domain. The present investigation signicantly extends the
ability to simulate and to predict drying processes in pneu-
matic dryer.
Appendix A. Nomenclature
c
p
(J/kg K) specic heat of the gas
R (m) radius of the pipe
T
ave
(K) average temperature
T (K) temperature
u (m/s) velocity
Greek symbols
density
dummy property
References
[1] R.M. Bowen, in: A.C. Eringen (Ed.), Theory of Mixtures in Con-
tinuum Physics, Academic Press, NY, USA, 1976.
[2] D. Gidaspow, Multiphase Flow and Fluidization, Academic Press,
Boston, USA, 1994.
[3] P.A. Cundall, O.D. Strack, A discrete numerical model for granular
assemblies, Geotechnique 29 (1979) 4765.
[4] J. Andrieu, R. Bressat, Experimental, experimental & theoretical
study of a pneumatic dryer, Proc. 3rd Int. Drying Symp. 2 (1982)
1019.
[5] J. Baeyens, D. van Gauwbergen, I. Vinckier, Pneumatic drying: the
use of large-scale experimental data in a design procedure, Powder
Technol. 83 (1995) 139148.
[6] D. Levi-Hevroni, A. Levy, I. Borde, Mathematical modelling of
drying of liquid/solid slurries in steady sate one-dimensional ow,
Drying Technol. 13 (5-7) (1995) 11871201.
[7] A. Levy, D.J. Mason, I. Borde, D. Levi-Hevroni, Drying of wet solids
particles in a steady-state one-dimensional ow, Powder Technol. 85
(1988) 1523.
[8] A. Levy, I. Borde, Steady-state one-dimensional ow for a pneumatic
dryer, Chem. Eng. Proc. 38 (1999) 121130.
[9] Z. Mindziul, A. Kmiec, Modeling gas-solid ow in a pneumatic-ash
dryer, Drying96 Proc. 10th Int. Drying Symp. A (1996) 275
282.
[10] S.C.S. Rocha, A.E.A. Paixo, Pseudo two-dimensional model for a
pneumatic dryer, Drying96 Proc. 10th Int. Drying Symp. A (1996)
340348.
[11] M.A. Silva, S.A. Nebra, Numerical simulation of drying in a cyclone,
Drying Technol. 15 (6-8) (1997) 17311741.
[12] R. Blasco, R. Vega, P.I. Alvarez, Pneumatic drying with superheated
steam: bi-dimensional model for high solid concentration, Drying
Technol. 19 (8) (2001) 20472061.
[13] S.C.S. Rocha, Contribution to the study of pneumatic drying: simula-
tion and inuence of the gas-particle heat transfer coefcient, Ph.D.
Thesis, So Paulo University, So Paulo, Brazil, (in Portuguese),
1988.
[14] I. Skuratovsky, A. Levy, I. Borde, Two-uid, two-dimensional model
for pneumatic drying, Drying Technol. 21 (9) (2003) 16491672.
[15] N. Abuaf, F.W. Staub, Drying of liquidsolid slurry droplets, Drying
86 Proc. 5th Int. Drying Symp. 1 (1987) 227248.
[16] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere
Publishing Corporation, New York, USA, 1980.
[17] D.B. Spalding, Developments in the IPSA procedure for numeri-
cal computation of multiphase-ow phenomena with interphase slip,
192 I. Skuratovsky et al. / Chemical Engineering and Processing 44 (2005) 187192
unequal temperatures, etc. in: Numerical Properties and Method-
ologies in Heat Transfer, Hemisphere, Washington, DC, 1983,
pp. 421476.
[18] M.A. Silva, J.L.G. Correa, Using DryPak to simulate drying process,
in: Proceedings of 11th International Drying Symposium Drying98,
vol. A, 1998, pp. 303310.
[19] Z. Pakowski, DryPak v.3, Program for psychometric and drying
computation, 1996.
[20] A. Levy, I. Borde, Two-uid model for pneumatic drying of
particulate materials, Drying Technol. 19 (8) (2001) 1773
1788.
Chemical Engineering and Processing 44 (2005) 193200
The formation of ne particles by salting-out precipitation
Judit Tth

, Andrea Kardos-Fodor, Susan Halsz-Pter


Research Institute of Chemical and Process Engineering, University of Veszprm, P.O. Box 125, 8201 Veszprm, Hungary
Received 25 August 2003; received in revised form 27 January 2004; accepted 11 February 2004
Available online 17 June 2004
Abstract
A systematic study of salting-out precipitation is carried out to obtain the operational limits within which this precipitation method can
be applied for the production of nes (mean particle size <10 m) with acceptable quality and productivity. The model substances: glycine
and sodium chloride are salted-out from their aqueous solutions by using ethanol as antisolvent. The main operational parameter is the initial
supersaturation of the solutions. It is shown that the smallest particles can be produced at the limits of the metastability domain determined by
three optional process parameters: the initial solution concentration, the equilibrium solubility and the operational time. The product quality
(crystallinity, polymorphic states, aggregation) and productivity considerably change with the operational conditions.
2004 Elsevier B.V. All rights reserved.
Keywords: Salting-out precipitation; Weimarn laws; Metastability; Initial supersaturation; Particle size
1. Introduction
Formation of micron and submicron size particles has
become particularly important in the material-research and
processing. The small size, high surface and surface activity
of these crystals and nanocrystals determine their various
applications as catalysts, ceramics, electronics, pharmaceu-
ticals, cosmetics etc. and this may explain the world-wide
increasing research activities to nd the suitable crystalliza-
tion modes and techniques for processing them [1,2].
Chemical precipitation is a well known method to produce
nes being a rapid solidication process of sparingly solu-
ble substances. Comparatively less known about the other
type of precipitation called salting-out or antisolvent pre-
cipitation although it is also widely used, especially in the
pharmaceutical industry owing to the available high yield,
purity and the fast, simple operation. In salting-out precip-
itation the necessary supersaturation is achieved by the ad-
dition of an antisolvent or electrolyte to the solution to be
crystallized. The produced crystals are small but often ag-
glomerated thus questioning the applicability of the pro-
cess for producing nes. Recently the salting-out precipi-
tation has got new perspectives. According to new experi-

Corresponding author. Tel.: +36 88 405 265; fax: +36 88 424 424.
E-mail address: toth@mukki.richem.hu (J. T oth).
mental ndings [3], it is possible to reduce or even elim-
inate the agglomeration in the presence of protective col-
loids or surfactants and/or by controlling the precipitation
kinetics.
In this work, batch salting-out experiments were per-
formed with the objective to know more about the limits and
possibilities on the formation of ne particles. As model ma-
terial the zwitterionic glycine was investigated within a wide
range of initial supersaturations. Considering the polymor-
phic nature of glycine, the non-polymorphic, ionic sodium
chloride was also investigated as reference system.
2. The precipitation theory
Not long ago, the main crystallization trend was to pro-
duce well-developed, bigger crystals in proper size and habit.
Accordingly, the research activities focused on the growth
process and the role of habit modifying additives. Nowa-
days, when extra ne particles are claimed by several in-
dustrial areas, increased attention has been paid to the early
stages of nucleation by applying even extreme conditions
for the particle formation. This area is the typical eld of
the colloidal state which is not yet enough utilized for the
industrial practice but it is getting more and more impor-
tant due to the increasing demands on the production of
nes.
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.02.013
194 J. T oth et al. / Chemical Engineering and Processing 44 (2005) 193200
In the colloid chemistry practice the antisolvent precip-
itation is called Weimarns method. Weimarn published
his famous work The precipitation laws early in the past
century [4] but his statements seem to be still valid and
useful for reconsidering. His main conception was that the
mean size of precipitates may change with the supersatura-
tion through a maximum. The maximum shifts up and left
with increasing time until the system reaches its stable state
where the precipitation process is completed. The principle
of the Weimarn laws on precipitation is schematically shown
in Fig. 1. According to the Weimarn laws, ne particles can
be produced both at low and high supersaturation depending
on the time of crystallization and the solubility of the pre-
cipitate. Weimarn based his statements on his chemical pre-
cipitation experiences. What is the situation for salting-out
precipitation? It is not necessarily the same because there
is a fundamental difference between the chemical- and salt-
ing out precipitation. In the rst case the supersaturation
changes with the concentration of the reaction partners while
the equilibrium concentration (solubility) remains constant,
but at salting-out processes both the supersaturation and the
equilibrium solubility change with the solution/antisolvent
ratios, therefore, both can inuence the particle size.
The relationship between particle size, solubility
and supersaturation is expressed by the well-known
GibbsThomson equation [5a]:
ln

c(r)
c

= ln S =
2V
kTr
, (1)
where c(r) is the solubility of the particles of size (radius)
r, c* is the normal equilibrium solubility of the substance,
S is the supersaturation ratio, is the surface energy, V is
the molecular volume, k is the Boltzmann constant, T is the
absolute temperature. Eq. (1) does not consider the growth
process which follows the nucleation step. At precipitation,
in most cases, the growth process is rather an aggregation
step. When aiming ne crystals, the growth subprocess has to
be retarded, consequently, the precipitation has to be stopped
Supersaturation, S
M
e
a
n

s
i
z
e

o
f

c
r
y
s
t
a
l
l
i
n
e

i
n
d
i
v
i
d
u
a
l
s
stable state
time
u
n
s
t
a
b
l
e

s
t
a
t
e
s
Fig. 1. The Weimarn laws on precipitation (via [4]).
(quenching) before the particles could grow up their nal,
stable size.
3. Experimental
In the experiments analytical grade chemicals: glycine
(purity 98.5%) or sodium chloride (99.5%), ethanol (99.8%)
and deionized water were used. The precipitation mode
was one-shot: at high supersaturations various amount (V
0
= 0.00412 cm
3
) of saturated or undersaturated aqueous so-
lutions of glycine (c
0
= 20.000.50 g/100 g solution) were
added in one portion to 12 cm
3
ethanol at ambient temper-
ature. At low supersaturations 12 cm
3
saturated solutions of
glycine were added to various amount (V
e
= 0.112 cm
3
) of
ethanol. Depending on the solution/antisolvent ratios, series
of initial supersaturations were created. In each case con-
centrated ethanol, saturated at 298 K (0.0037 g glycine/100 g
solvent) was used. In the most cases the crystals appeared
at once after the shots but 10 min stirring was applied ad-
ditionally. No longer stirring time was allowed not to leave
time for the polymorphic change. Some precipitation runs,
together with the in-line particle size measurements, were
carried out directly in the 15 cm
3
volume MALVERN cu-
vette. When the suspension densities were higher than al-
lowed in direct CSD measurements by MALVERN 2600
the samples were precipitated in an Erlenmeyer ask under
stirring. In these cases the CSD measurements were possi-
ble after the appropriate dilution of the slurry. The diluting
agent was the actual mother liquid or concentrated and satu-
rated ethanol. For characterizing the particle size the volume
mean diameter was used:

d =

nd
4

nd
3
, (2)
where n is the number of particles of equivalent diameter d.
For characterizing the particle size distribution the so called
span was used:
span =
D
90
D
10
D
50
, (3)
where D
10
, D
50
, and D
90
is the diameter below which 10,
50 and 90% of the volume of the particles lies.
The precipitates were separated by glass bre lter,
washed by ethanol and air-dried. In some cases X-ray
diffraction (by Philips 3710) for polymorphism and SEM
analysis (by Philips XL 30 ESEM) for habit determination
were made.
The runs with sodium chloride were carried out by
the same techniques as detailed above but the operational
concentrations were different: c
0
was changed between
26.55 and 7.40 g/100 g solution. A characteristic difference
between the two systems was their solubility in ethanol
(0.0037 g glycine/100 g solvent and 0.0649 NaCl g/100 g
solvent) at 298 K. Because of the slower precipitation abil-
J. T oth et al. / Chemical Engineering and Processing 44 (2005) 193200 195
ity of sodium chloride, longer (60 min) additional stirring
was allowed.
4. Experimental results and discussion
The main driving force of a salting-out precipitation pro-
cess is the supersaturation generated by the different solubil-
ities of the solute in the solvent and antisolvent. The one-shot
mode generally means that the necessary amount of the an-
tisolvent is added in one portion to the initial solution (either
saturated or undersaturated) as it is schematically shown in
Fig. 2. In our cases the aqueous solution was added to the
antisolvent. As a result of the antisolvent addition the orig-
inal solubility of the substance will change until the new
equilibrium solubility, c
eq
has been reached. To characterize
the supersaturation driving force in a precipitation system
the so called initial supersaturation, S
i
is used:
S
i
=
c
0
c
eq
, (4)
i.e. it is a ratio between the initial concentration of the
solution to be crystallized (in this case the aqueous solu-
tion) and the equilibrium concentration (solubility) in the
solute-solvent-antisolvent system. In this way different val-
ues of S
i
can be chosen as experimental parameter by chang-
ing the concentration of the aqueous solution or the ratio of
the aqueous solution and the antisolvent. Although S
i
means
a certain simplication but it is still acceptable, because the
actual operational concentrations are very uncertain accord-
ing to the micromixing insufciencies. The highest applied
initial supersaturations were S
i
5000 for glycine and S
i

400 for NaCl owing to their different solubilities in ethanol.


The solubility data of the model systems were taken from
the literature [6,7]. For the calculation of the missing data
correlation methods were elaborated. The equations which
0 20 40 60 80 100
Mass percent of the antisolvent
S
o
l
u
b
i
l
i
t
y
C
0
-
supersaturation driving
forces in saturated and
undersaturated solutions
'one-shot' addition
antisolvent
C
eq
-
C
0
-
Fig. 2. The scheme of the one shot salting-out precipitation from the
saturated and undersaturated solutions.
best tted the whole range of solubilities have been found
as follows:
for the glycinewaterethanol system:
y =2.618 +2.264 10
2
x
+23.17 exp(4.060 10
2
x), (5)
for the NaClwaterethanol system:
y =24.91 4.348 10
1
x
+18.66 exp[2.433 10
2
(100 x)], (6)
where x is the mass percent of the ethanol, y is the equilib-
rium concentration. In the range of 90100% ethanol con-
tent another type of equation was construed for the sake of
higher accuracy:
for the glycinewaterethanol system:
lg y =2.656 9.476 10
2
x
+1.363 10
3
x
2
9.243 10
6
x
3
, (7)
for the NaClwaterethanol system:
lg y = 6.764 10
1
+7.363 10
2
x 7.861 10
4
x
2
(8)
0
5
10
15
20
25
0 20 40 60 80 100
Mass percent (g ethanol/100 g solvent), (%)

c
0

(
g

g
l
y
c
i
n
e
/
1
0
0

g

s
o
l
u
t
i
o
n
)
0
5
10
15
20
25
30
0 20 40 60 80 100
Mass percent (g ethanol/100 g solvent), (%)
c
0

(
g

N
a
C
l
/
1
0
0

g

s
o
l
u
t
i
o
n
)
(a)
(b)
Fig. 3. Solubilities and applied supersaturation programmes for the
glycinewaterethanol (a) and NaClwaterethanol (b) model systems.
196 J. T oth et al. / Chemical Engineering and Processing 44 (2005) 193200
The solubility curves (at 298 K) as well as the ways of cre-
ating the different supersaturations are shown in Fig. 3, for
the glycinewaterethanol (a) and NaClwaterethanol (b)
model systems, respectively.
The experiments with saturated aqueous solutions show
remarkable differences concerning the two model systems
as it can be seen in Fig. 4, where the corresponding ini-
tial supersaturation, S
i
and average particle size,

d data of
glycine (a) and NaCl (b) are plotted, separately. When only
a very small amount of glycine solution (0.004 cm
3
) was
added to 12 cm
3
of antisolvent the alcohol mass percent
was very high: 99.96% but surprisingly the particle size was
not as small as expected, it was about 50 m. At somewhat
lower S
i
and alcohol ratio a minimum occurred (point A in
Fig. 4a). After the minimum the particle size approached
to a plateau (point B in Fig. 4a) then a peak of 100 m,
and nally decreased down to 20 m. These just mentioned
anomalies in the glycine particle size and S
i
relation can
be caused by several reasons (the change of the nucleation
mechanism or polymorphic states, etc.). The polymorphic
change of glycine itself is already known from the litera-
0
20
40
60
80
100
120
1 10 100 1000 10000
Initial supersaturation, S
i
()
A
v
e
r
a
g
e

p
a
r
t
i
c
l
e

s
i
z
e
,

d

(
m
m
)
A
B
C
0
20
40
60
80
100
120
1 10 100 1000
Initial supersaturation, S
i
()
A
v
e
r
a
g
e

p
a
r
t
i
c
l
e

s
i
z
e
,

d

(
m
m
)
(a)
(b)
Fig. 4. Relations between the initial supersaturation, S
i
and the av-
erage particle size,

d (under the applied operational time) in the
glycinewaterethanol (a) and the reference NaClwaterethanol (b) sys-
tems.
ture [8] where the unstable form is usually reported for
waterethanol systems.
SEM micrographs were made on the representing samples
belonging to the points A, B, C in Fig. 4a, at the same time
XRD investigations were also carried out from these sam-
ples. At point A the polymorphic form is mainly (97%),
the particles seem rather dendritic (Fig. 5). At point B the
particles are pure with the signs of aggregation growth
(Fig. 6), while at point C the pure form was produced
(Fig. 7). As a remarkable new experience: it was possible to
produce the pure form by precipitation.
Comparing the precipitation data of glycine and the refer-
ence non-polymorphic sodiumchloride, the latter shows def-
inite maximum in the S
i


d relation. It has to be mentioned
that ne particles (

d < 10 m) were produced even from sat-


urated NaCl solutions (Fig. 4b). It is interesting since NaCl is
better soluble in ethanol and longer operational time (60 min)
was allowed for its crystallization. But comparing the growth
abilities of the two substances the linear growth rates are dif-
ferent at least in one magnitude: 10
8
ms
1
for NaCl [5b]
and for glycine 10
7
ms
1
was found [9]. This case well
demonstrates the importance of the material properties.
The experiences with saturated solutions did not result in
acceptable small particle size for glycine and the particles
were more or less agglomerated. Not much improvement
was achieved when some agglomeration retarding additives
were used, i.e. gelatine, polyvinylpyrrolidone, Tween 80 etc.
By our experiences these additives seem to be ineffective in
cases when very rapidly high supersaturation generates in the
solution. The same experiences were reported by Garti and
Zour [10]. In the region of the lower supersaturations certain
size-reduction effect was observed (6030 m) but not to the
expected extent. The most signicant effect was achieved
by the application of foreign ions, electrolytes. In glycine
precipitation NaCl, CH
3
COOK and CaCl
2
were tested one
by one. The best result was found when 100 ml ethanol
containing 5 g CaCl
2
was used as antisolvent resulting in
gel-like precipitates. Unfortunately, these type additives can
hardly be applicable in the food- or pharmaceutical industry.
The real turning-point in our precipitation strategy ap-
peared when less and less saturated aqueous solutions were
used. Even keeping the ethanol content at 99.6% (see point
F in Fig. 8a) but starting the salting-out precipitation with
more and more undersaturated aqueous solutions (i.e. de-
creasing the values of c
0
), the initial supersaturation, S
i
will
be also decreasing according to its denition, S
i
= c
0
/c
eq
(in this case c
eq
= constant). With the decreasing initial
solution concentrations the average particle sizes also de-
creased (see points f
1
f
5
in Fig. 8a) down to 6 m and
the particles showed less and less agglomeration. Starting
from other equilibrium points it means another alcohol
contents the tendencies were the same but the average
particle size was over 10 m. In this way, it was possible
to determine experimentally the whole operational areas
both for glycine (Fig. 8a) and sodium chloride (Fig. 8b). In
Fig. 8a and b the upper boundaries of the operational areas
J. T oth et al. / Chemical Engineering and Processing 44 (2005) 193200 197
Fig. 5. SEM micrograph on the glycine sample (mainly form) precipitated at high alcohol content and initial supersaturation (see point A in Fig. 4a).
represent the saturated initial solutions, while the lowest
ones belong to the less saturated solutions. At the same time
the lowest boundaries show the limits under which there
was no crystallization within the applied operational time
(10 min for glycine and 60 min for NaCl). It means that the
metastable limit must be not at a certain point but along a
rather broad operational area.
By determining the boundaries of the whole operational
eld the precipitation area of the expected particle size can
be chosen within these limits. Fig. 8a and b show the pos-
Fig. 6. SEM micrograph on the glycine sample ( form) precipitated at medium alcohol content and initial supersaturation (see point B in Fig. 4a).
sibilities for the investigated model systems. Regarding the
formation of ne particles (d< 10 m) the following ten-
dencies have been found: in the case of glycine, under the
before mentioned experimental conditions, the desired small
size was available only when undersaturated aqueous solu-
tions (c
0
= 3.442.56 g/100 g solution) but high initial su-
persaturations (S
i
800600) were applied. Because of the
low solubility at high ethanol content (at 99.6% ethanol c
eq
= 0.0041 g glycine/100 g solution) it was possible to cre-
ate high supersaturations even in the originally undersat-
198 J. T oth et al. / Chemical Engineering and Processing 44 (2005) 193200
Fig. 7. SEM micrograph on the glycine sample ( form) precipitated at low alcohol content and initial supersaturation (see point C in Fig. 4a).
urated solutions. In the case of NaCl the expected small
size was already obtained from saturated aqueous solution
(c
0
= 26.55 g NaCl/100 g solution) also at high supersatu-
ration (S
i
= 394). Decreasing the initial solution concentra-
tion down to 7.4 g NaCl/100 g solution (S
i
= 107, at 99.6%
ethanol content) the average particle size also decreased
down to 1 m.
When the goal is the production of ne particles it is
important to save the primary size of the crystals as they
appear rst in the solution, i.e. the nucleation has to be pro-
moted over the growth and aggregation steps. In the case
of undersaturated, weak initial solutions the precipitation
takes place near the metastable region where the kinetic
processes are rather slow. For example, the induction time
which was necessary for crystallization in the weakest NaCl
solutions approaches to 60 min. Repeating the precipita-
tion (where the ethanol content was the same 99.6%) with
saturated aqueous solution there was no measurable induc-
tion time and the particle size changed considerably within
the applied 60 min operational time: the

d was 4.41 m
after 10 min, 8.86 m at 20 min and nally16.27 m at
60 min. It is obvious that in the latter case not the small-
est available size was measured after 60 min, but for the
sake of comparison the same operational time had to be
applied.
Beside the average particle size, the size distribution is
also an important parameter of the product. Comparing the
obtained cumulative undersize distribution of the precipi-
tates it has been found that the ratio of the smallest parti-
cles (below 10 m) in the distribution was more than 50%
when

d 10 m. In the other cases (

d > 10 m) the ratio


was less than 50% and showed decreasing tendency with
increasing size. The Malvern 2600 size-analyzer calculates
the so called span see Eq. (3) as the measure of the
width of the particle size distribution. Having checked all
the particle size analysis data the characteristic value of span
changed between 0.9 and 1.6 for glycine and between 0.9
Fig. 8. The operational areas of the glycine (a) and NaCl (b) precipitation.
J. T oth et al. / Chemical Engineering and Processing 44 (2005) 193200 199
0
20
40
60
80
100
120
0.00 0.02 0.04 0.06 0.08 0.10 0.12
Q (gcm
-3
)
A
v
e
r
a
g
e

p
a
r
t
i
c
l
e

s
i
z
e
,

d

(
m
m
)
0
20
40
60
80
100
120
0 0.02 0.04 0.06 0.08
Q (gcm
-3
)
A
v
e
r
a
g
e

p
a
r
t
i
c
l
e

s
i
z
e
,

d

(
m
m
)
(a)
(b)
Fig. 9. The relation between the average particle size,

d and the maximum
available amount of crystals, Q produced in unit volume of solution in
the glycinewaterethanol (a) and the NaClwaterethanol (b) systems
( refers to saturated initial solutions, refer to undersaturated
initial solutions).
and 1.8 for NaCl samples. These values are typical for the
nest precipitates, too.
Summarizing the experiences on the formation of nes
by salting-out precipitation: the supersaturation has the
0
20
40
60
80
100
120
1 10 100 1000 10000
Initial supersaturation, S
i
()
A
v
e
r
a
g
e

p
a
r
t
i
c
l
e

s
i
z
e

d
,

(
m
m
)
glycine
NaCl
metastability
domain
Fig. 10. The operational areas of the two systems studied.
strongest effect on the product particle size as it was ex-
pected. Contrary to the Weimarns precipitation model, ne
particles (

d < 10 m) were produced only at high supersat-


urations while at the low values (at low ethanol content),
where the solubility of the substances were rather high, it
was not possible to get the desired size. The idea to work
with weak, undersaturated initial solutions, but then creat-
ing high supersaturation proved to be useful and has led to
the determination of the lower metastable boundary of the
operational eld. Consequently, the nest and agglomera-
tion free particles can be gained near the metastable region.
In the other part of the operational eld it is possible only if
the particles are quickly removed from the solution before
they could grow up to their nal, stable size and when the
substance has only moderate growing ability like NaCl. Of
course, it means consequences on the available productivity.
The productivity of a crystallization process is usually
expressed as the amount of the produced crystals in unit
volume of solution and unit time. The maximum available
amounts of the crystals produced from unit volume of solu-
tions, Q (g cm
3
) were calculated on the bases of the solu-
bility differences and shown as function of the average size
in Fig. 9a for glycine and Fig. 9b for NaCl. Although it is an
approximation especially for the weakest slurries, where the
measuring error would be also high, but in other cases it is
acceptable because the measured particle sizes were close to
their nal, stable size. Taking the produced amounts of both
model materials as well as the operational time (10 min for
glycine and 60 min for NaCl) into account the available pro-
ductivities are between 1.66 10
6
1.82 10
4
g cm
3
s
1
,
or expressed in the usual industrial unit 6654 kg m
3
h
1
.
The relatively low productivity is quite common in colloid
systems or in nanoparticle formation. It is always a question
of consideration to decide about what productivity is still
acceptable and on what expense, although there is a strong
rule in the crystallization practice: the particle size and pro-
ductivity cant be changed independently of each other.
In Fig. 10 the operational elds of the two model systems
are plotted together to make a comparison possible and assist
in the explanation of the experimental ndings.
5. Conclusion
A systematic study of the salting-out precipitation pro-
cess was made to obtain characteristic particle size and pro-
ductivity data within wide range of the initial supersatu-
rations. The appropriate supersaturations were created by
changing two independent parameters: the initial solution
concentration and the solubility. The operational time was
xed at a given system but in limited cases the effect of
the operational time on the particle size was also investi-
gated.
It has been concluded that the salting-out precipitation is
a exible and acceptable process for producing nes but the
material properties (solubility, nucleation and growth ability)
200 J. T oth et al. / Chemical Engineering and Processing 44 (2005) 193200
may considerably limit the possibilities as it is well seen in
Fig. 10.
The supersaturation particle size relations show denite
maximums just following the Weimarns statement in this
respect. The metastable limit below which there is no crys-
tallization within the applied batch time exists along the
lower boundary of the operational eld. The nest and ag-
glomeration free particles can be produced at the extremes,
i.e. at the edges of the metastability domain determined by
changing three optional parameters: the initial solution con-
centration, the solubility and operational time. Since the
particle size depends on the relative rate of the nucleation
and aggregation growth kinetic processes the application of
growth retarding techniques (quenching) can be useful. The
polymorphic state may change at the extremes: for example
pure -glycine form was precipitated at low alcohol content
although the form is more typical at salting-out precipita-
tion.
Relatively lowproductivity can be expected when nes are
produced by salting-out process although it also depends on
the substance properties. The post-crystallization processing
(separation, concentration etc.) of very weak slurries may
go according to the routines of colloid chemistry practice.
Acknowledgements
The authors are very grateful to the University of
Veszprm, Department of Silicate Chemistry and Material
Sciences, Hungary for the XRD and SEM facilities.
Appendix A. Nomenclature
c(r) solubility of the particles of size (radius) r
(kmol m
3
)
c* normal equilibrium solubility of the substance
(kmol m
3
)
c
0
initial concentration of aqueous solutions
(g solute/100 g solution)
c
eq
solubility of the substance (g solute/100 g solution)
CSD crystal size distribution
d equivalent diameter of the particles (m)

d average particle size (m)


D
x
diameter below which x% of the volume of the
particles lies (m)
k Boltzmann constant (J K
1
)
n the number of particles of equivalent diameter d ()
Q maximum available amount of crystals produced
from unit volume of solution (g cm
3
)
r particle radius (m)
S supersaturation ratio ()
SEM scanning electron microscopy
S
i
initial supersaturation ()
T absolute temperature (K)
V molecular volume (m
3
)
V
0
volume of aqueous solutions (cm
3
)
V
e
volume of ethanol antisolvent (cm
3
)
XRD X-ray diffraction
surface energy (J m
2
)
References
[1] The 15th Symp., Ind. Cryst., Sorrento, Italy, 2002.
[2] The 8th Conf., Colloid Chem., Keszthely, Hungary, 2002.
[3] P.H. Karpinsky, Maneuvering through the Complexities of Batch
Precipitation, Proc. Int. Symp. Ind. Cryst., Tokyo, Japan, 1998,
pp. 1322.
[4] P.P. von Weimarn, The precipitation laws, Chem. Rev. 2 (1926) 217
242.
[5] (a) J.W. Mullin, Crystallization, third ed., Butterworh-Heinemann,
London, 1993. p. 102.;
(b) J.W. Mullin, Crystallization, third ed., Butterworh-Heinemann,
London, 1993, p. 222.
[6] A. Seidell, Solubilities of Organic Compounds, D. van Norstrand
Company Inc., New York, 1941, p. 124.
[7] H. Stephen, Solubilities of Inorganic and Organic Compounds, Perg-
amon Press, 1964, pp. 102103.
[8] W. Hubig, ber die drei Formen des Glycins, Z. Naturforsch. 13b
(1958) 633638.
[9] S. Halsz-Pter, Controlled Precipitation, CSc Thesis, Hung. Ac.
Sci., Budapest, 1994.
[10] N. Garti, H. Zour, The effect of surfactants on the crystallization
and polymorphic transformation of glutamic acid, J. Cryst. Growth
172 (1997) 486498.
Chemical Engineering and Processing 44 (2005) 201207
Micronized cocoa butter particles produced by a supercritical process
J.-J. Letourneau
a,
, S. Vigneau
a
, P. Gonus
b
, J. Fages
a
a
Chemical Engineering Laboratory for Particulate Solids, UMR CNRS 2392, cole des Mines dAlbi, Campus Jarlard, 81013 Albi, France
b
Nestl Product Technology Centre York, UK
Received 25 July 2003; received in revised form 15 February 2004; accepted 19 March 2004
Available online 24 June 2004
Abstract
A process using supercritical CO
2
has been developed to produce form V crystals of cocoa butter. In this process, solid cocoa butter is put
in an autoclave in which supercritical carbon dioxide is then introduced. At high pressure, two liquid phases, one CO
2
-rich and one cocoa
butter-rich are formed. Both of them can be sent to a depressurisation nozzle leading to particle formation in an expansion vessel. Moreover,
particles can also be formed in the rst autoclave undergoing slower depressurisation. Both particle populations (upstream and downstream
the nozzle) have been characterised using calorimetric method (differential scanning calorimetry, DSC) and environmental scanning electron
microscopy (ESEM). The polymorphism of the particles produced as well as their size and morphology show that they are potentially usable
in the tempering step of the manufacturing process of chocolate.
2004 Elsevier B.V. All rights reserved.
Keywords: Cocoa butter; Supercritical; CO
2
1. Introduction
1.1. Polymorphism of cocoa butter and chocolate tempering
Cocoa butter, the fat of chocolate, is responsible for the
pleasant melt in the mouth sensation given by chocolate.
It is a complex mixture of triacylglycerol (TAGs), which
melting points are in the good range (3135

C) to give
this feeling [1].
The most abundant are monounsaturated TAGs, the un-
saturated fatty acid (very often oleic acid) being in the
2-position. 1-Palmitoyl-2-oleyl-3-stearoylglycerol (P-O-S),
P-O-P and S-O-S account for 3638, 1416 and 2328%,
respectively of the total TAGs. It is these TAGs which are
mainly responsible for the behaviour of chocolate when un-
dergoing crystallisation [2].
There are six different ways for these TAGs to arrange
themselves when they crystallise. Six different polymorphs
noted Ithe less stable with the lowest melting point 290 K
(17

C)to VIthe most stable with the highest melting


point 308 K (35

C)can be obtained. The relative amount


of these six polymorphs will condition several features of the

Corresponding author. Tel.: +33 563 493088; fax: +33 563 493025.
E-mail address: letourne@enstimac.fr (J.-J. Letourneau).
URL: http://wwww.enstimac.fr.
end product. If the wrong type of fat is present the chocolate
may exhibit a white powdery surface called fat bloom which
is to be avoided because of its unpleasant appearance. The
most desirable form is known as form V (or
2
) having a
melting temperature of 305307 K (3234

C). The transfer


to the even more stable form VI is not desirable since it
might lead to fat bloom as do the less stable forms.
The processing conditions determine which poly-
morph is formed. To obtain predominantly form V a
pre-crystallisation step also called tempering has been intro-
duced in the chocolate manufacturing process. It is aimed
at the formation of form V seed crystals which in turn
will force the bulk of the remaining fat to crystallise in
this stable form. Two main ways are used: (i) starting with
hot liquid chocolate 318323 K (4550

C) a succession
of cooling to 300 K (27

C) in order to form both stable


and unstable crystals followed by re-heating to 303305 K
(3032

C) to melt only less stable crystals will lead to form


V crystals which, in turn, will seed the mass of chocolate
and (ii) seeding the mass of untempered chocolate by nely
divided form V crystals of cocoa butter.
1.2. Crystallisation with supercritical processes
Supercritical uid (SCF) technology presents a new and
interesting route for particle formation, which avoids most of
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.013
202 J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207
the drawbacks of the traditional methods of milling/crushing
or liquid crystallisation. Supercritical processes are known
to give micro- or even nano-particles with narrow size distri-
bution, and may allow the control of crystal polymorphism.
In addition, several other advantages inherent to these pro-
cesses must be emphasised: they are single-step processes
with simplied downstream processing and they are envi-
ronmentally desirable since they avoid the use of organic
solvents. Moreover, they allow the processing of thermola-
bile molecules and give high purity products [3].
Although most of the applications of supercritical particle
formation are known in the pharmaceutical industry, some
examples can be found with foodstuff. Cocoa butter a rela-
tively high value product has been successfully micronised
using the PGSS (particles from gas-saturated solutions) pro-
cess [4]. In this paper we will investigate the production of
ne cocoa butter crystals in order to use them for the seed-
ing of chocolate.
2. Material and methods
2.1. Experimental apparatus
The experiments were carried out in a versatile pilot plant
(Separex, France). A schematic diagram of the apparatus is
shown in Fig. 1. Carbon dioxide is cooled and stored in a liq-
uid CO
2
storage tank (2). It can be circulated at a maximum
ow of 25 kg h
1
by a membrane pump (Lewa, Germany)
(3). Compressed CO
2
passes then through a heat exchanger
(4) and becomes supercritical. Some of the devices (1, 8, 9,
13) shown in Fig. 1 are used for other experiments such as
SAS or RESS with co-solvent.
The vessel (5) (PARR Instrument, USA) is equipped
with three sapphire windows (16), a magnetic stirrer (15)
Fig. 1. Experimental set-up.
with a maximum torque of 1.8 Nm and a exible mantle
heater.
The ow coming out this vessel is sprayed into the ex-
pansion vessel (7) by means of a nozzle (6) (Spraydry
SK-SIY80-SKY16, orice insert diameter: 340 m, Spray-
ing System). The powder formed is collected in a porous bag
made of PTFE bres. Then, CO
2
is depressurised (10) and
can either be directed to the vent or puried and recycled.
2.2. Cocoa butter particle generation
Pieces of solid cocoa butter are loaded in the vessel (5)
equipped with a stirrer and this vessel is lled with CO
2
.
Then, the operating temperature and the stirrer speed are
xed by the controllers. Further addition of CO
2
allows
reaching the operating pressure. The mixture thus obtained is
stirred during a long enough period of time (at least 5 mn at
constant pressure and temperature) in order to reach the ther-
modynamic equilibrium. In particular if two liquid phases
are present, each of them is assumed to be saturated.
The second vessel (7) is then lled with CO
2
until reach-
ing the second operating pressure. This is the expansion
pressure at which the mixture is placed downstream of the
nozzle. This pressure is xed during all the experiment with
a back-pressure regulator (10).
The temperature of the CO
2
upstream from the stirred
vessel is controlled by the heating device (4) and xed to
be the same as the temperature of the mixture in the stirred
vessel.
During the experiments the valve between the two ves-
sels is open. The saturated mixture sprayed into the expan-
sion vessel via the nozzle, is subjected to a sudden drop
in pressure, resulting in supersaturated conditions and the
crystallisation of the cocoa butter. This powder is collected
in the bag inside the second vessel. This process is known
J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207 203
Fig. 2. Phase diagram of pure CO
2
(Peng-Robinsons model [6]).
as RESS process (rapid expansion of supercritical solutions)
[3].
Two major types of experiments have been achieved:
In the rst one, the rst vessel is let to slowly depres-
surise during its emptying through the nozzle. Therefore,
particles of cocoa butter can also be formed in this vessel.
In the second, the membrane pump (3) feeds the vessel
with CO
2
in order to compensate the emptying via the
nozzle and to maintain a constant pressure.
Another renement in the classication of types of ex-
periments depends on the initial load of cocoa butter. For
small amounts, the mixture appears in a homogeneous single
phase. On the other hand, when the initial amount increases,
two phases are present.
If the temperature is too low, the solid cocoa butter formed
this second phase, but with higher temperatures two uid
phases are in equilibrium: one is rich in CO
2
and the other
is rich in cocoa butter. Kokot et al. [5] showed that at 313 K
and 20 MPa, the mass fractions of CO
2
for these two phases
are 0.44 and 0.98. When working with two uid phases
and without stirring, the results of the experiments depend
on the position of the outlet pipe of the rst vessel: top or
bottom. If the uid is extracted from the top of the vessel,
the cocoa butter powder is produced from the richest phase
in CO
2
. Conversely, it is the cocoa butter rich phase which is
extracted from the bottom. If the mixture is homogenised by
agitation these two phases are mixed and the resulting uid
appears (as seen through the windows) to be homogeneous
just like an emulsion.
The drop in pressure undergone by the mixture in the noz-
zle, leads also the temperature to fall and crystals of CO
2
may appear (CO
2
triple point: T = 216.6 K, P = 0.52 MPa).
To prevent this solid CO
2
formation, the pressure P
7
in the
expansion vessel must be kept beyond the triple point pres-
sure. But this does not always prevent the CO
2
liquefaction
if the temperature upstream the nozzle is not high enough.
As an example shown in Fig. 2 (path A), the nal tempera-
ture of pure CO
2
at T = 333 K submit to an isenthalpic de-
pressurization from P = 10 to 1 MPa is T = 233 K, which
is a point within the boundary of the liquidvapour region.
The presence of two phasesliquid and gasaround the
particles in formation could change the growth of the crys-
tals and therefore their nal morphology. The heating device
(17) is used to prevent the temperature T
7
in the expansion
vessel to be lower than the dew temperature of pure CO
2
at
P
7
. The theoretical path for the experiment #3 of Table 1 is
shown in Fig. 2 (path B). For this experiment boiling water
has been used in the heating device and the nal tempera-
ture of the mixture at T = 373 K submit to a depressuriza-
tion from P = 20 to 1 MPa is T = 288 K. As it can be seen
this depressurization is almost isenthalpic.
The owrate of CO
2
is adapted to t with the operating
conditions. It mainly depends on the characteristics of the
204 J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207
Table 1
Experimental operating conditions
Experiment number Series Diam. (mm) M
0
(g) V
5
(cm
3
) P
5
(MPa) T
5
(K) P
7
(MPa) T
7
(K) Duration (min)
1 E-I 0.34 20.8 2000 20 313 0.6 288 23
2 E-I 0.34 30.6 2000 25 373 0.3 288 28
3 E-I 0.41 18.6 2000 20 318 1 288 22
4 E-I 0.41 36 2000 20 323 0.1 288 21
5 E-I 0.41 19.9 2000 30 313 0.9 288 21
6 E-II 0.41 20.6 2000 30 303 0.1 291 10
7 E-II 1.50 20.2 2000 25 303 0.1 293 6
8 E-III 0.41 19.9 2000 30 313 0.1 289 4
9 E-III 1.50 102.3 2000 31 298 0.1 292 3
10 E-IV 0.34 99.7 2000 20 293 0.7 291 7
11 E-IV 0.34 106.4 2000 27 305 0.4 288 24
12 E-IV 0.41 100.6 2000 20 313 0.2 288 14
13 E-IV 0.41 94.4 2000 32 313 0.3 288 15
14 E-V 0.34 128 1200 20 303 4 283 5
15 E-V 0.34 135 1200 25 308 2 283 5
Diam.: diameter of the nozzle orice; M
0
: initial mass of cocoa butter; V
5
, P
5
, T
5
: volume, pressure and temperature of the tank (5), respectively; P
7
,
T
7
: pressure and temperature of the tank (7), respectively.
nozzle and on the pressure difference P
5
P
7
between the
two vessels. A typical owrate value for the experiences
listed in Table 1 is about 20 kg h
1
.
2.3. Powder characterisation
Differential scanning calorimetry (DSC) has been used to
characterise the samples in a DSC-7 Perkin-Elmer. The DSC
patterns of the samples (25 mg) were obtained between 283
and 323 K at a heating rate of 5 Kmin
1
in a N
2
gas stream.
The picture presented in Figs. 4 and 5 have been made
with an Environmental Scanning Electron Microscope
(XL30 ESEM FEG, FEI Philips, The Netherlands).
In order to use these powders for chocolate seeding, a
characterisation of their mechanical and rheological proper-
ties is needed and this will be done in future work.
2.4. Tempering experiments
Chocolate tempering experiments were carried out by
seeding with various type of cocoa butter powder. The de-
gree of pre-crystallisation (temper) was measured using a
computerised tempermeter (Exotherm 7400 from Systech
Analytics). This measurement is basically a cooling curve
where the change in temperature is measured against the
time. The chocolate sample is taken at processing tempera-
ture in a special container and placed in the instrument cell
at 281 K (8

C). In a typical tempering curve, the exothermic


process of crystallisation generates an inexion point. The
slope at this point is used to determine whether the chocolate
is under-, well- or over-tempered. Usually a well-tempered
curve shows a horizontal plateau. More details on tempering
curve measurements can be found in Bolliger et al. [7].
The process of cocoa butter seeding is usually carried out
at 303 K(30

C) where a mass fraction of 0.52%of ordinary


cocoa butter seeds is mixed with the melted chocolate.
3. Results and discussion
3.1. Particle formation experiments
The rst series of experiments (E-I) was done with small
amounts of cocoa butter (single-phase system) and with con-
tinuous feeding in CO
2
. The second series (E-II) was done
without feeding and top extraction: particles were collected
in both vessels. The third series (E-III) was done without
feeding and bottom extraction: particles were collected in
both vessels. The fourth series (E-IV) was done with larger
amounts of cocoa butter, continuous feeding and bottom ex-
traction. The fth and last series of experiments (E-V) was
done with larger amounts of cocoa butter and homogeneous
mixing.
Two typical experiments have been selected for dis-
cussion: the experiment #5 which belongs to the rst
series (rapid expansion of a single-phase mixture with
continuous feeding in CO
2
) and the experiment #15 of
the fth series (rapid expansion of a two-phase stirred
mixture).
In Fig. 3, the DSC patterns for the original cocoa butter
and for experiments 5 and 15 are shown. Original cocoa
butter shows a peak at T = 307.1 K, an end-set point at
T = 309.6 K and a melting enthalpy of h
fus
= 131 J g
1
,
which are characteristics values for form VI crystals.
The powder obtained in experiment 15 was collected in
the bag of the expansion vessel. Its characteristics are: T
peak
= 304 K, T
end-set
= 306.3 K and h
fus
= 120.3 J g
1
. This
form V powder was white, light, with regular individual
particles and was free-owing. Nearly all experiments were
achieved with the following conditions: nozzle orice di-
ameter 340 m, 298 T
5
308 K, P
5
20 MPa and
0.5 P
7
4 MPa led to the same kind of form V powder
collected in the expansion vessel (7). A typical example of
this kind of particle is shown on Fig. 4. It is made of an
agglomeration of smaller particles.
J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207 205
Fig. 3. Typical DSC curves.
The powder obtained in experiment 5 was collected in
the vessel (5) and has the following characteristics: T
peak
= 300.5 K, T
end-set
= 302.5 K for the rst peak and: T
peak
= 303.7 K, T
end-set
= 305.0 K for the second peak. All the
Fig. 4. ESEM picture of a cocoa butter particle obtained with experiment #15.
particles obtained in this vessel are made of a mixture of both
forms IV and V. The faster the drop in pressure in this vessel,
the larger the quantity of form V powder obtained. A typical
example of this kind of particle is shown in Fig. 5. This
206 J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207
Fig. 5. ESEM picture of a cocoa butter particle obtained with experiment #5.
powder is generally white, extremely light and sometimes
pieces of original yellow cocoa butter can be found among
the particles.
3.2. Seeding of chocolate
The cocoa butter powder obtained by supercritical process
was able to seed using the following conditions.
At 306 K (33

C) and above, the seeding produced an


under-tempered chocolate; at 304 K (31

C) and below, the


chocolate was quickly over-tempered. However, at 305 K
(32

C), good tempering conditions were obtained with less


than 0.3% of cocoa butter seeds. This is at least 2

C higher
than common seeding and several degrees higher than most
industrial tempering step. Therefore, the chocolate is very
much thinner.
The powder samples described in series E-I and E-IV
(obtained with the smallest nozzle) gave the best results. The
seeding was carried out at 305 K(32

C) with a mass fraction


of 0.10.2% and produced a well-tempered chocolate. At
this high temperature, the stirrer velocity is a critical point
as an excess of mechanical energy could melt the seeds,
but this has to be tested using an industrial scale tempering
machine (batch or continuous).
4. Perspectives and conclusion
The best cocoa powders in terms of appearance, DSC,
reproducibility and shelf-stability are those of the E-Vseries.
They have not been tested yet to seed chocolate and this will
be part of a future work.
High shear stresses are known to favour the most stable
forms of the cocoa butter. The same phenomenon is probably
also appearing in the supercritical generation of cocoa butter
particles: when the particles are generated in the expansion
vessel with a huge drop in pressure through a small nozzle,
they are exclusively made of form Vwhen generated in the
rst vessel with a slow drop in pressure form IV is present
and even predominant. This is a track for further studies.
A major interest in using this new process [8] is the value
of the temperature at which seeding can be done with the
cocoa butter particles supercritically generated. At 305 K
the chocolate is less viscous, this is advantageous in term
of owability and energy consumption. An additional ad-
vantage lies in the fact that the time needed to obtain a
well-tempered chocolate is shorter (about 5 min). Moreover,
the overall quantity of seeds to temper the chocolate is very
small. Therefore, a scaling-up of this process to an industrial
scale appears feasible.
J.-J. Letourneau et al. / Chemical Engineering and Processing 44 (2005) 201207 207
References
[1] S.T. Beckett, The Science of Chocolate, RSC Paperbacks, 2000.
[2] M. Lipp, E. Anklam, Review of cocoa butter and alternative fats for
use in chocolate. Part A. Compositional data, Food chem. 62 (1998)
7397.
[3] J. Fages, H. Lochard, E. Rodier, J.-J. Letourneau, M. Sauceau, La
Gnration de Solides Diviss par Fluides Supercritiques, Can. J.
Chem. Eng. 81 (2003) 161175.
[4] Z. Knez, E. Weidner, Precipitation of solid with dense gases, in:
A. Bertucco, G. Vetter (Eds.), High Pressure Process Technology:
Fundamentals and Applications, Elsevier, 2001, pp. 587611.
[5] K. Kokot, Z. Knez, D. Bauman, SLG (solidliquidgas) phase
transition of cocoa butter in supercritical CO
2
, Acta Alimentaria 28
(1999) 197208.
[6] D.-Y. Peng, D. Robinson, A new two-constant equation of state, Ind.
Eng. Chem. Fundam. 15 (1976) 5964.
[7] S. Bolliger, B. Breitschuh, M. Stranzinger, T. Wagner, E.J. Windhab,
Comparison of precrystallization of chocolate, J. Food Eng. 35 (1998)
281297.
[8] J. Fages, J.-J. Letourneau, Procd de production de particules, French
Patent FR 2847486 (2002).
Chemical Engineering and Processing 44 (2005) 209214
Food powder handling and processing: Industry problems, knowledge
barriers and research opportunities
John J. Fitzpatrick
a,
, Lilia Ahrn
b
a
Department of Process Engineering, University College Cork, Cork, Ireland
b
SIK, The Swedish Institute for Food and Biotechnology, Gothenburg, Sweden
Received 26 August 2003; received in revised form 1 March 2004; accepted 19 March 2004
Available online 17 July 2004
Abstract
This paper highlights some of the outcomes from a EU Accompanying Measure focussing on food powders. The main goals of the work
were to outline the major industrial problems, knowledge barriers, and research challenges and opportunities in relation to food powders, and
to promote the creation of a sustainable network of excellence in the area of food powders. Food powders are powders rst and foremost,
thus many of the research challenges and opportunities are similar to handling and processing of other powders. As a result, much can be
learned from other industries that deal with powders. What makes food powders different from many other powders is their composition,
which is mainly of biological origin, and that they are eventually consumed by people and animals. Consequently, stability of food ingredient
functionality and prevention of contamination are major issues right through frompowder production to nal application of the powder, which
is usually in the form of a wet formulation.
2004 Elsevier B.V. All rights reserved.
Keywords: Food powder research; Powder handling; Powder processing
1. Introduction
1.1. Background and objectives
This paper originates from a 1-year project funded by the
EU Commission through an Accompanying Measure, which
was completed in April 2003. The focus of the project was
to highlight industry problems, knowledge barriers and re-
search and development opportunities in relation to food
powders. It was also envisaged that this would act as a nec-
essary step in fostering the potential for future research ideas
and collaborative research.
The specic objectives of the project were
to outline research and development needs and opportu-
nities in terms of
industrial problems with handling, processing and pro-
duction of safe, high quality food powders that require

Corresponding author. Tel.: +353 21 4903089;


fax: +353 21 4270249.
E-mail address: j.tzpatrick@ucc.ie (J.J. Fitzpatrick).
further targeted research initiatives;
areas requiring further science and engineering knowl-
edge so that this can be applied in the develop-
ment of new and more efcient processes to produce
added-value food powders that are safe and of high
quality;
new and emerging food powder technologies requiring
research and development;
to promote the creation of a sustainable network of excel-
lence in the area of food powders.
The main steps in implementing the objectives of the
project consisted of
holding of a workshop on food powders;
creation of a strategic document for research on food pow-
ders;
dissemination of the output from the project.
This paper summarises some of the key ndings from the
strategic document for research in food powders [1].
0255-2701/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2004.03.014
210 J.J. Fitzpatrick, L. Ahrn e / Chemical Engineering and Processing 44 (2005) 209214
1.2. Food powders what is special about them
The development of formulation engineering concepts in
food manufacturing and the demand for diversity in food
products has driven a substantial market increase for food
ingredients. Most ingredients are supplied in powdered form
and therefore powder technology is an increasingly impor-
tant issue both to food ingredient manufacturers and food
producers. The two major aspects about food powders that
make them special are biological origin and the fact that
they are ultimately consumed by humans and animals.
The major reason for production in powder form is sim-
ply to prolong the shelf-life of the ingredient by reducing
water content, otherwise the ingredient will be degraded and
broken down in its natural biological environment. So, the
major function of the powder form is to maintain the sta-
bility of the ingredient functionality until it is required for
utilisation, which is usually in some sort of wet formulation.
The major functionalities of food ingredients can be broadly
classied as
physical/chemical: for example, gelation, emulsication,
foaming, pH control;
nutritional: for example, vitamins, nutraceuticals;
organoleptic: for example, colour, taste, smell, texture.
There are a multitude of food ingredients available with a
whole variety of different functions, and there is huge scope
for R&D into investigating new ingredients, and new and
improved functionalities. As the role of the powder form is
mainly to preserve this functionality over time and to deliver
it when required, the focus of this work deals with the pow-
der issues and not the different ingredient functionalities.
The major powder issue with regard to food ingredient pow-
ders is maintaining the stability of ingredient functionality
from production right through to nal powder application.
Most food powders are not directly consumed by humans
and animals, they are usually mixed with water and other
liquids to produce wet formulations, which are further pro-
cessed to produce products that are directly consumed. As a
result of being ultimately consumed by humans and animals,
prevention of contamination of the powders with microor-
ganisms, viruses and undesirable chemical components is a
huge issue for food powder manufacturers and the food in-
dustry in general. One major case of food poisoning linked
to a contaminated powder can be very detrimental to a pro-
ducer.
Other powder issues are also important in delivering the
food powder, and these include the ability to handle and
transport ingredient powders, dust problems, dust re and
explosion hazards, allergy problems, creation of desirable
powder particle properties, and the ability to dissolve these
powders when required. Many of these issues are in com-
mon with other powders and much can be learnt from other
industries dealing with powders.
1.3. Identication of problems, knowledge barriers and
research challenges and opportunities
When looking at the processes involved in going from
powder production to its nal application, they can be
broadly divided into the following two categories in Table 1.
Category A processes have the dominant inuence in cre-
ating the powder properties that inuence the powders
nal application. Category B processes may inuence the
powder properties by degrading them or by requiring cer-
tain properties to allow effective handling and transport. To
identify problems, knowledge barriers and research chal-
lenges and opportunities, it may be useful to look at these
categories separately and to generically investigate how
research possibilities may come about as a result.
For example, for category A processes, it may be useful
to rstly look at the applications of the powder (how it is
nally applied) and work backwards to the production of
the powder. There are not many food powders that are di-
rectly consumed by the nal consumer in powder form only.
Most are incorporated as ingredients into some sort of wet
formulation somewhere along the chain before being nally
ingested by the consumer. As a result, an important area is
research into the technologies and processes involved in ad-
dition and mixing of powders to produce these wet formu-
lations. Allied to this, is the creation of powder properties
that enable these processes to function and that enhance pro-
cess performance. From this, a major area of research is the
investigation of how powder properties inuence their roles
in forming wet formulations. Some of these properties may
be difcult to dene and measure, thus denition and mea-
surement may become research areas that must be tackled
before effective investigation of how the properties affect
application.
A further step back from application to powder produc-
tion is the processes that create the desired powder proper-
ties. Thus, another major area of research involves the study
of these processes, and investigating how raw material vari-
ables and process input variables affect the engineering of
the desired powder properties. Finally, as mentioned above,
a very important aspect of research for food powders is
component stability, all the way from powder production,
Table 1
Classication of powder processes
A. Processes that give
powder its properties
B. Handling and
transport processes
Powder production processes Storage (e.g. silos, IBCs, bags, cans)
Drying Transport (e.g. feeders and conveyers)
Comminution Packaging equipment
Crystallisation/precipitation
+ drying
Mixing
Separation
Agglomeration/granulation
Coating/encapsulation
J.J. Fitzpatrick, L. Ahrn e / Chemical Engineering and Processing 44 (2005) 209214 211
through storage, handling and processing right through to -
nal application of the powder. There are also holistic issues
of importance for food powders, including health and food
safety, re and explosion hazards, energy efciency and cost
of manufacture especially for basic food powders.
As a result of the above considerations, the following six
sections were created to gain a more focused insight into
typical industrial issues and problems, knowledge barriers
and research opportunities in areas that have importance to
food powders in particular, and which also are relevant to
powders in general
Quality and safety
Powder production processes
Mixing and agglomeration
Other added-value technologies
Powder storage and transport
Characterisation of powders
These areas are explored in detail in the strategic docu-
ment for research in food powders. Some of the major is-
sues that surface from these sections are presented in the
following two sections.
2. Powder issues of practical importance to food
powders
2.1. Maintaining or stabilising the functionality of
ingredients
This is a critical theme for food powder ingredients as
many of them will be added to form wet formulations and
they will be expected to perform. For example, will the egg
white have the same functional performance when dried,
handled and rehydrated as it did before it was dried? How
can polyunsaturated fats be dried, stored and applied with-
out being readily oxidised and producing noxious odours?
What lactic acid culture activity will remain after it has been
dried, stored and rehydrated after 1 year? Thus, research into
maintaining the stability of ingredient functionality during
formulation prior to drying, drying, storage, handling, pro-
cessing and rehydration is key to food powder ingredients.
Some examples of general research topics include
Research into formulation technology prior to drying and
the role of excipients in maintaining and enhancing com-
ponent stability.
Research into the mechanisms of small molecule stabili-
sation of biopolymers during the removal of water during
drying.
Research into improved and alternative drying technolo-
gies that maintain the stability of components (e.g. low
temperature drying).
Research into factors that affect the destabilisation of the
functionality of ingredients in powder form during stor-
age, handling and rehydration.
Research into agglomeration, coating, encapsulation, and
barrier technologies, to maintain component stability and
to provide smart composite powders for triggered re-
lease and targeted delivery.
2.2. Prevention of contamination
Contamination of a food powder with undesirable bio-life
forms and chemical components is a major concern due
to direct impact on human and animal health. As a conse-
quence, one high prole case can cause major damage to
an individual company and a whole industrial sector. Be-
cause of its importance, a lot of work has already been done
to prevent contamination. There are plenty of methods and
procedures available to eliminate most contamination risks,
however the key to success requires constant monitoring and
strict implementation of current good manufacturing prac-
tices (cGMP). The EU hygienic engineering design group
(EHEDG) has produced documentation on the hygienic de-
sign of food equipment, including those dealing with dry
particulate solids [2]. Problems still exist with dust forma-
tion leading to powder settling on equipment, which may
provide a growth environment for microbes when contact
with moisture exists.
2.3. Dust prevention and control
Dust generation can lead to a number of problems for food
powders. These can be summarised as follows: (i) health
problems, in particular allergy problems; (ii) contamination
and plant hygiene issues due to dust settling and sticking
onto equipment; and (iii) re/explosion hazards. As a result,
improvement in dust measurement, prevention and control
is a major issue when dealing with food powders.
Due to their biological origin, there is always the pos-
sibility that a food powder may contain biological active
components, such as enzymes, that may be harmful and
may produce allergy problems. Consequently, this may
eventually lead to the imposition of stricter dust exposure
limits. Self-ignition of food powders that stick onto the
drying chamber is still a big problem. The powder may
be experiencing temperatures below its minimum igni-
tion temperature, however the temperature and moisture
content may be high enough to favour other exothermal
reactions, that may cause the product to heat up over time
until it reaches a temperature that can sustain combustion
reactions. At this stage, it will start to smoulder and then
it can start a re or a dust explosion. This self-ignition
scenario can also occur in other equipment, such as lter
bag houses. There is a need for a greater understand-
ing of these self-ignition mechanisms and how to prevent
their occurrence. There is also a need for in-line sensors
that can detect volatiles coming from these reactions and
give early warning of self-ignition. Sensors for measuring
carbon-monoxide have been developed and are already in-
dustrially implemented in the dairy industry but sensors for
212 J.J. Fitzpatrick, L. Ahrn e / Chemical Engineering and Processing 44 (2005) 209214
measuring other volatiles where CO is not given off are
required.
2.4. Stickiness
Occasionally during spray drying of many different prod-
uct formulations, there may be one formulation whereby the
powder starts sticking onto the wall of the drying chamber
and coating it. In powder mixing where a small amount of
liquid ingredient is sprayed onto a powder, the powder may
stick to the mixing chamber surface and form a crust over
time. Both these examples may appear mundane, however
they cause major problems when they occur, as the chamber
now needs to be cleaned. There is a need for testing that can
predict the likelihood of a powder sticking onto a surface,
and how this is affected by the composition of the pow-
der and processing conditions. Furthermore, there is a need
for determining approaches to overcoming these stickiness
problems.
2.5. Segregation of food ingredient mixes
Many food ingredients are supplied as food ingredient
mixes, which, in some cases, can contain many ingredients.
If the ingredients have different particle properties, in partic-
ular, particle size, this may lead to segregation or demixing
after the mixing process. This problem can seriously limit
the potential for creating ingredient mixes. Industry people
would like to have tools for measuring segregation tendency
and for predicting the maximum difference in particle prop-
erties tolerable before segregation becomes a problem. Fur-
thermore, it is desirable to have a systematic approach to
assist in the development of non-segregating mixes.
2.6. The value of food powders
Food powders are usually perceived to have a low quality
and thus low value. As a result, low cost production drives
powder production and thus there is little investment avail-
able to stimulate innovation and change, so the technology
tends to stagnate and change little over time. From this, there
is a need for investment in powder science and technology in
the food industry to help solve recurring problems, improve
process efciency and engineer added-value properties into
food powders. Solving problems and improving efciency
will reduce cost and added-value properties will increase the
value of food powders. Some of this cost saving and addi-
tional revenue can then be ploughed back in to continue the
cycle of product and process innovation. Simultaneously,
there is a challenge for scientists and engineers to continue
to demonstrate that powder education and R&D in the food
industry is worth the investment.
Powder storage and transport operations are perceived not
to add value to the powder, thus there is little investment
into dealing with and solving storage and transport prob-
lems. Processors appear to simply live with the problems
and do not appear to realise that solving these problems can
save a lot of headache and more importantly, bottom-line
cost. A major underlying problem is education insofar as
many working in the food industry do not have an adequate
knowledge of powder science and technology to be able to
apply it effectively in the food industry.
2.7. Particle engineering
This involves the engineering of food powder ingredient
particles that can maintain ingredient stability and function-
ality, that do not present dust formation problems, that do
not present difcult handling and transport problems, and
can be readily rehydrated to deliver their required functional
performance in a wet formulation. The major limitation to
this type of approach is the low value of food powders and
the low cost driven nature of food powder product that stunts
this type of innovation. On the other hand, particle engineer-
ing is the key to the higher added value of food powders,
e.g. by producing products with excellent instant properties.
3. Issues of importance to powders in general,
including food powders
Food powders are powders rst and foremost, thus many
issues that are relevant to powders in general, are also very
relevant to food powders. These are summarised as follows.
3.1. Education
Most of the industry people dealing with food powders
come from a food science and technology background or an
engineering background. Powder science and technology is
seldom taught in engineering courses or food science and
technology courses. The subject is too important in the prac-
tical life of an engineer or scientist for this to be omitted
from curricula. Thus, there is a need for incorporation of
powders into existing undergraduate and postgraduate edu-
cation. The world of particle and powder technology should
be as important as uids.
The educational scenario is more complex for practic-
ing engineers and scientists. For those wishing to obtain a
greater knowledge of powders, there are a number of short
courses available conducted mainly by industry consultants
and some academic institutions. The question is: Is this suf-
cient or is there a need for more educational opportunities?
There may be potential for a European or even world-wide
distance learning programme dealing with aspects of pow-
der science and technology. Allied to this is the problem that
many practising engineers and scientist do not realise their
deciency in powders or are simply not aware of the edu-
cational opportunities or are not given the time and support
to pursue them. This can only be overcome by promoting a
greater awareness of the importance of particle and powder
J.J. Fitzpatrick, L. Ahrn e / Chemical Engineering and Processing 44 (2005) 209214 213
science and technology, which is denitely making progress
with time.
3.2. Transfer of existing knowledge into application in
industry
Much more attention should be paid to transfer of exist-
ing knowledge from universities, research centres and con-
sultancies into industry. In most cases, knowledge transfer,
mainly in the form of innovative equipment and system de-
sign, is introduced to industry via powder equipment sup-
pliers. Consequently, there is potential for greater synergy
between these suppliers and centres of research.
Non-multinational companies/SMEs, in particular, can
use much more knowledge to escape from the trial and error
approach. There is an enormous amount of wasted time and
effort because the same mistakes and problems appear time
and time again throughout the process industries. There is
a lot of wisdom scattered about that can be applied to solve
many of these problems and prevent many of these mistakes.
However, the limiting steps are education and knowledge
transfer, although the situation is gradually improving with
time and with more and more individuals promoting the
importance of particle/powder science and technology.
3.3. Standardisation of powder property denition and
measurement
In the world of particles and powders, there are many
properties and there is a multitude of property denitions
and measuring techniques, which may even vary from one
industry to another. There is a need to standardise the de-
nition and measurement of these properties. This will help
streamline denitions and measurement techniques to those
that are most applicable. It will help in obtaining greater
reproducibility and also enable greater comparison of work
undertaken by different groups.
3.4. Fundamental science, modelling and design
Description and modelling of the fundamental mecha-
nisms that can be applied in the design of most powder pro-
cesses is still very much at an early stage. It is comparable
to the pre McCabe-Thiele era in distillation technology. This
is particularly true for many common operations like dry
powder mixing, mixing of powders and liquids, granulation
and coating. Many of them have similar mechanisms taking
place, such as particle motion and particle/liquid contact,
thus there is potential for cross-fertilisation. There is much
scope for progress in this area. Modelling techniques, such
as DEM and CFD, are trying to model particle/particle and
particle/uid motions and interactions, however it may take
many years before these techniques are useful for design in-
volving real powders, and some are dubious whether they
will ever be useful for real cohesive powders. The rst step
in the modelling of spray-drying has already been taken.
In addition, there is potential for looking at alternative
mechanisms that may result in more effective ways of doing
things that could lead to the development of new equipment
based on these new concepts. For example, the concepts
behind most present-day dry mixers have been around for
decades. Maybe, there are alternative approaches, such as
the application of chaos in mixing that could lead to more
effective mixing with less energy requirement. Similar com-
ments could be made for other processes, such as granula-
tion, whereby a greater understanding of the fundamental
science may lead to alternative approaches to granulation
process design.
3.5. On-line and in-line monitoring and control
This is very much in its infancy as applied to powder pro-
cesses, and as a result, there is a lot of scope for work in
this area. The advantages of improved monitoring and con-
trol are early warning of problems, reduced costs and more
consistent product quality. Presently, there are a number of
sensors that have potential for use on- or in-line, however
there is scope for the development of on-/in-line versions
of existing sensors and totally new sensors. The major chal-
lenges for the application of existing sensors are either one
or a combination of the following: (i) reducing sensor cost;
(ii) improvement of sensor robustness for real-time opera-
tion in a powder environment, and (iii) development of data
analysis technology that can lter the sensor output and con-
vert it into useful information for control.
3.6. Nanoparticle technology
Nanotechnology is gaining more and more popularity
throughout industry. As a result, nanoparticle technology
represents a new promising eld of application to food pow-
ders, although current research and development is very
much in its infancy.
3.7. Energy efciency
Most powder processes, and in particular drying and com-
minution, are not very energy efcient. An improvement in
the fundamental science, modelling and application to de-
sign coupled with improved on- and in-line monitoring (e.g.
model predictive control) will give opportunities to improve
energy efciency while ensuring product quality.
Acknowledgements
We would like to acknowledge the ve group leaders
in the project for their valued contributions: Prof. Koen
Dewettinck, Ghent University, Belgium; Richard Farnish,
Wolfson Centre for Bulk Solids Handling Technology, UK;
Prof. Peter Lillford, University of York, UK; Dr. Sivert Ose,
POSTEC, Norway; and Prof. Karl Sommer, TU-Munich,
214 J.J. Fitzpatrick, L. Ahrn e / Chemical Engineering and Processing 44 (2005) 209214
Germany. We also acknowledge Prof. Fernanda Oliveira,
UCC Ireland and Prof. Thomas Ohlsson, SIK Sweden for
their assistance and support during the creation and im-
plementation of the project. Finally, we acknowledge all
those who participated in the workshop and the many oth-
ers who made contributions to the creation of the strategic
document.
References
[1] Strategic document for research in food powders. http://www.
foodpowders.net, March 2003.