COLUMN: A REVIEW Swati Mohanty Regional Research Laboratory C.S.I.R.), Bhubaneswar-751013, India E-mail: [email protected]ABSTRACT Mathematical models ar e reviewed for different types of commonly used extraction columns viz. pulsed siev e plate column rotating disc contactor uhni column spray column Scheibel extractor packed column Oldshue-Rushton contactor and reciprocating plate column. In addition numerical techniques proc ess simulators and some estimation methods for model parameters have also b een included. Th e review cites over 120 references. Keywords: mathematical modeling extrac tion column population balance diffusion model solvent extraction INTRODUCTION Liquid-liquid extraction is the second most important separation process after distillation in chemical industries. Its early application was in oil-refining but it has now been extended to such varied applications as the petrochemical industry Pharmaceu ticals hydrometallurgy nuclear industry an d environmental protection. Despite years of study a design procedure employing sound basic principles has not been developed and until today the design of extraction column is b ase d on pilot plant data and desig n exp erie nce 1 9 9 Brought to you by | SUNY Buffalo Libraries Authenticated | 128 205 114 91 Download Date | 3/20/13 3:27 PM
50
Embed
MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
Vol. 16 No. 3 2 Modeling of Liquid-Liquid Extraction Column:
A Review
and hence is far from the optimum. Polydispersity of turbulent drop swarm is
the main obstacle in designing liquid-liquid extractors.Mathematical models that adequately define the system lead to better
design of these equipments. However, adequate research has not been carriedout in this area, although attempts have been made to develop models basedon certain simplified assumptions. This is mainly due to the complexity o f thesystem where the dispersed phase constantly loses its identity due to breakage
and coalescence. Insufficient attention has been paid to the experimentaleffort required to verify the utility of the models. Much remains to be donebefore dispersed phase behaviour is adequately modelled and columnperformance can be confidently predicted. Recent books edited by Godfreyand Slater / / and Thorton 2 present detailed analysis of various types of
extraction equipment. Steiner and Hartland /3,4/ have discussed the
advantages and disadvantages o f two early models proposed by Sleicher 15 61
for liquid liquid extraction. One is the backmixing model, which assumes thatthe dispersed phase consists of uniform sized drops without breakage and
coalescence. The other is the forward mixing model, which takes intoconsideration the different drop size, drop breakage and coalescence, varyingvelocities of the drops due to varying size. Some of the stagewise and
differential models and their solution methods have also been reviewed byPratt and Baird Ð I and Steiner and Hartland /8/. Use of stochastic simulationtechniques (e.g., Monte Carlo techniques has been introduced for simulatingliquid liquid extraction, but these have been primarily applied to stirred tankcontactors and hence have no t been included in this review. The purpose o fthis paper is to present a general review of various models available for somemost widely used liquid-liquid extractors.
TYPES O MO ELS
Mathematical models for liquid-liquid extraction are categorised intothree basic types.a) Empirical: These are the simplest of all the models and are obtained by
fitting experimental data with empirical correlations which are functionsof liquid physical properties, column geometry and operating conditionsof the column. The drawback of this type of model is that this cannot be
extrapolated beyond the stipulated range of application. This type of
Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
where the birth, B d), and death, D d) of drops of size d, relates to
breakage and coalescence in the volume element of height, dZ A goodreview of population balance models applied to chemical processes is
presented by Ramakrishna 1 9 1
c) Stagewise: In this type of model, the column is described as a series of
completely mixed stages and the model equations are developed based on
the principle of mass conservation applied to each stage Fig.lc). Theseare similar in form to equations 2), 3) and 4) 161 The stages may be
real, as in agitated stagewise columns, or cascades with a large number of
hypothetical stages, as in differential extractors.
M AT H E M AT I C A L MO ELS
The first models of extraction column that were reported were simple innature. Both the phases were assumed to be in differential contact with eachother, with each phase assumed to be a continuum. The dispersed phase was
treated as pseudo-homogeneous and backflow was accounted for by the axialdispersion coefficient In the subsequent models, the dispersed phase was
represented by monodispersed or polydispersed drops without undergoingbreakage or coalescence. The complex behaviour of the dispersed phase,
consisting of swarms of droplets with a wide size distribution, varying
velocities relative to the continuous phase, and varying mass transfer rates,could be represented by the hydrodynamic and mass transfer behaviour of
some representatively-sized single drop, normally the surface-volume or
Sauter mean diameter drop. In the later models, the coalescence was
accounted for by introducing a coalescence height, which indicates the
average distance the drops actually travel before they coalesce and redisperseto maintain the original drop size distribution. The concentration of the
dispersed phase was assumed to be constant throughout a given cross-section
of the column due to sufficiently intensive coalescence and redispersion.More recently, population balance models have been developed to predictdrop-size distribution and hold-up profiles along the length of the columntaking into account drop breakage and coalescence. In addition, a number of
203rought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
Vol. 16 No. 3 2000 Modeling of Liquid-Liquid Extraction Column:
A Review
empirical correlations for predicting the hold-up in different types of
extractors have also been reported. A unified correlation which can predictthe hold-up in eight different types of extraction columns has been proposed
by Kumar and Hartland /l O/. A summary of d ifferent models for various types
of liquid liquid extractors is presented in Table 1.
Pulsed Sieve Plate xtraction olumn PSE)
This type of liquid liquid extractor finds application mainly in nuclear
processing industries. Most of the models reported for this type of extractorare based on population balance of drops. Models by Haverland et al l\ I/,
Dimitrova AI Khani et al I\2I and Zimmermann et al /13/ are based on thebasic population balance model given by equation 4). In the steady state
hydrodynamic model, Haverland et al /I I/ have neglected drop-drop
coalescence and have proposed a normalised Beta-distribution for the drop
size distribution which was found to fit best with the experimental data. A
correlation for break-up probability for drops of different sizes has been
proposed based on experimental data. Although the model predictions forhold up and drop size distribution agree quite well for the system
toluene-water and under operating conditions where coalescence is negligible,
it would not be applicable for systems and operating conditions where
coalescence is significant. Dimitrova A l Khani et al. I12I have made certain
assumptions for drop-drop interaction, such as that a mother drop breaks into
three equisized drops and the breakage rate is proportional to the eighth
power of the drop diameter. For coalescence, experimental values for the two
parameters, one for collision efficiency and another for coalescenceprobability, have been used but no detailed method of determination of these
two parameters has been reported. For the mass transfer model, a linear
equilibrium relationship and the mass transfer coefficient independent of the
drop diameter have been assumed. The model contains two fitting parameters,
one for the drop transport and the other for the drop breakage which limits the
applicability of the model. Zimmermann et al /13/ have applied the
population balance model to multi-component extraction. The model involves
too many parameters to be known. The authors have validated the model
against pilot plant data, but the values used for various model parameters
have not been given. Methods of determination of these parameters would
have been useful for applying the model to other sieve plate extractors.
204 Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
Mohanty and Vogelpohl /14/ have simplified the differential population
balance model of equation 4) into a stagewise model which considerably
reduces the computational time. The space between two consecutive
sieve-plates is taken to be a single stage. Both breakage and coalescence have
been considered. A beta distribution for the daughter drop size distribution
has been used and the experimental breakage probability has been used.
However, the coalescence parameters have been taken as adjustable. Qian and
Wang 5 have developed a stagewise model where it is assumed that thedrop size distribution does not change throughout the column. At the end of
each stage, the drops coalesce to equalise the concentration and then
redisperse to maintain the original drop size distribution. For the mass
transfer coefficient of the dispersed phase, the Handlos and Baron /16/
correlation is used, while the Higbie penetration model is used for the
continuous phase. Although the model is quite simple, the predicted
concentration profile agrees well with the experimental data. The stagewise
population balance model by Garg and Pratt /17/ accounts for drop breakage
and coalescence and uses experimentally determined coalescence and
breakage parameters, which is unlike other models where these are used as
adjustable parameters. The breakage and coalescence rate constants have
been expressed as functions of hold-up and drop size. However, the
assumption of binary breakage of drops and coalescence occurring only
between adjacent and same drop sizes does not adequately represent the
actual system. Moreover, the colorimetric method used for determination of
coalescence rates, where drops of two different colours yellow and green)
coalesce to give rise to a third colour red), makes it difficult to distinguish
the colours when multiple coalescence takes place. Although the prediction of
drop size distribution at the outlet compared well with experimental data,
comparison of predicted number of coalesced red drops with experimental
data showed that the agreement was good only for the first few plates.
Blass and Zimmermann /18/ have presented a simple stagewise, backflowmass transfer model for a pulsed sieve-plate extraction column assuming a
constant holdup. A recirculation regime model developed by Prabhakar et al 9 compares well with experimental data obtained for mixer-settler andemulsion regime. Experimental determination of hold-up shows that in themixer settler region, the hold-up is affected by the hole diameter whereas inthe emulsion region it is affected mainly by the free area. Also, the drop
formation takes place in the mixer-settler region due to dispersed phase flow
211Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
Vol. 16. No . 3, 2000 Modeling of Liquid-Liquid Extraction Column :A Review
through the perfora tions and drop breakage takes place in the em ulsion region
due to pulsing action within the plate stack. Hold-up was found to be highwhen the solute transfer w as from the continuo us to the dispersed phase.
The empirical correlation fo r predicting th e hold-up in a PSE proposed b yKumar and H artland /20/, w hich was compared w ith other available empiricalcorrelations, is
Af - Af) m
(5)
where
Af)m =
O
(6)
The constants k b k2 an d k3 that appear above are given in Table 2 for thecase of no solute transfer, solute transfer from continuous to dispersed phase,and solute transfer from the dispersed to continuous phase. The correlation isbased on the available data for column diameters of 25-213 mm and isapplicable fo r prediction of hold-up in the mixer settler, transition andemulsion regions of operation within an average error of 17.8 . Theinfluence of hole diameter and column diameter w as insignificant.
ble
Parameter values for no solute transfer, solute transfer from continuous todispersed phase and solute transfer from dispersed to continuous phase.
no solutec-»d
d-»c
k,2.10xl0 6
2.14x10l . l O x l O6
k2
44.53
44.53
50.56
k 3
9.69 ÷ 10°
9.69 xlO' 3
9.69 xlO' 3
212 Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
This type of extractor is commonly used for deasphalting of petroleum,for desulfurisation o f gasoline and for recovery of phenol from wastewater. Anew Open Turbine Rotating Disk Contactor OTRDC) has been developed by
Zhu et al I \l that has higher efficiency than the RD C and is also suitable fo rliquid-liquid-solid as well as liquid-solid systems. Modeling studies on the
RD C include pseudo-homogenous and population balance models as well as
those that take into account the interfacial reaction. The complexity of modelsaccounting for interphase chemical reaction increases since the observed rate
is a function of both chemical kinetics and diffusion.The differential pseudo-homogenous dispersion model by Chartres and
Korchinsky 22 includes a mass balance equation fo r each drop size fraction
for the dispersed phase and a mass balance equation for the continuous phase.Drop coalescence has been accounted for by introducing a coalescence heightto equalise the solute concentration and to then redisperse maintaining the
original drop size distribution. The drop size distribution has been estimatedusing Mugele-Evans functions /98/. The effect of drop size distribution was
found to decrease with a decrease in the coalescence height. The study alsoshows that the axial dispersion in the dispersed phase does not influence the
mass transfer significantly and can be neglected. However no verification of
the model equation has been made with experimental data.Korchinsky and Cruz Pinto /23/ have improved this model by introducing
a settling zone above the agitated section of the column, and by eliminatingthe assumption o f constant continuous phase concentration fo r estimating themass transfer coefficient by replacing it with rigid drop and turbulent
circulating drop model. The predicted number of transfer units was, on theaverage, ca. 10 higher than the experimental values. In order to improve the
agreement the mass transfer coefficient and the axial dispersion coefficient of
the continuous phase were adjusted. Thus four parameters, two for the
agitated zone and two for the settling zone, can be adjusted. Cruz-Pinto and
Korchinsky Ã Ë É have also solved the model equation of a RDC byincorporating one of the two diffusion models, i.e., either the Newman rigiddrop model 25 or the Handlos-Baron /16/ turbulent circulating drop model,
but taking into account the variation in the solute concentration of thecontinuous phase. This has been done by solving simultaneously the
hydrodynamic and mass transfer equation fo r swarm of liquid drops. Th e
213Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
Vol. 16 No. 3 2 Modeling of Liquid-Liquid Extraction Column:
A Review
model takes into consideration the axial mixing and the time dependent mass
transfer coefficient. The mass balance equation for the continuous phase issolved simultaneously by the drop diffusion model using a finite differencemethod and the results compared with that obtained by exact solution. The
predicted results have been compared with experimental data. The operating
condition of the RDC was so chosen that drop coalescence and drop breakage
were negligible. The discrepancy has been attributed to an inaccurate estimate
of the extent of mixing in the continuous phase and by accounting for therange of drop mass transfer that depends on the size.
A population balance model developed by Casamatta and Vogelpohl 7267
for a RDC accounts for drop coalescence and drop breakage with an
assumption that the drops break-up into three equal size daughter drops. The
factors accounting for the coalescence and breakage have been taken as
adjustable parameters.
The population balance model by Ghalehchian and Slater 7277 considers
drop breakage but no coalescence. The characteristic velocity predicted from
correlation from literature has been modified to minimise the discrepancy
between the predicted and experimental hold-up. Also the parameter, m, used
to account for the drop swarm, to predict the drop velocity has been taken as
the adjustable parameter. The beta function has been assumed for daughter
drop size distribution. The error in predicting hold-up is between ± 20 and
Sauter mean diameter between ± 18 .
A steady-state differential pseudo-homogenous dispersion model
involving extraction with interphase chemical reaction is given by Sarkar et al/28,29/. The model treats the disperse phase as pseudo-homogeneous. The
surface area for mass transfer is based on the average drop size measurement.
Expressions for extraction rates for very slow reactions, slow reactions, andfast reactions based on the film penetration, and Dankwerts models have been
given. The predicted extent of reaction has been compared with experimental
data for RDC with a maximum deviation being 20 . However, the model has
certain limitations, one being that it assumes the resistance to mass transfer
lies only in the aqueous phase and experimental values of mass transfer
coefficient without reaction under actual conditions of turbulence arerequired. Also the effect of droplet interaction has not been taken into
consideration.
The stagewise model by Azimzadeh-Khatayloo 7307 which is animprovement over Misek's model /31/ includes the possibility of entrainment
of smaller drops by the continuous phase. A mean diameter based on d 43 =
Znjdj4
/Znjdj3
has been used instead of the Sauter mean diameter for betterestimation of the characteristic velocity. The model incorporates into it the
influence of drop size distribution on mass transfer.
A new approach by Zhang et al 7327 uses stochastic process theory to
account for forward mixing, backmixing an d mass transfer behaviour of thedrops in the model. The drops move in the column in a stochastic manner an dthe displacement of the drops is described by the Fokker-Planck equation 7337
where q is the transition probability density, t and ô are time (1>ô), and both w
an d Æ are positions of the drop at time ô and t respectively. Also, U is theaverage velocity of the drop and Ed is the axial dispersion coefficient of thedispersed phase. The solution to this equation gives the age distribution of
drops at different heights. From the age distribution of different drops, the
RTD of drops at any height can be determined. The axial dispersion
coefficient is a function of the RDC rotor diameter, the height of a
compartment, the drop axial velocity and the free cross-sectional area. The
concentration of the dispersed phase is then calculated by solving the mass
transfer equation for a single drop an d then integrating over al l drops at anyheight to get the overall concentration of the dispersed phase. Both
coalescence and redispersion have also been neglected. The drop size
distribution was measured experimentally an d found to fit well with
Mugele-Evans upper limit log-normal distribution function. Three types of
model, namely, those of Newman 7257 Kronig an d Brink /34/, an d Handlos
and Baron /16/ have been used for estimating the mass transfer coefficient of
the dispersed phase, depending on the drop Reynold s number. The mass
transfer coefficient for the continuous phase is based on the Calderbank and
Moo Young /35/ correlation. For the axial dispersion coefficient, the
correlation suggested by Zhang et al 7367 has been used. The mass transfer
resistance in the continuous phase has been assumed to be negligible. A
comparison between the number of transfer units when the solute transfer is
from the continuous phase to the dispersed phase has been found to agree
well with experimental data, but the deviation was appreciable when the
transfer was in the opposite direction, i.e., from the dispersed to the
215rought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
continuous phase. This may be due to coalescence as coalescence is high
when the solute transfer is from the dispersed phase to the continuous phase.Zhu et al. I2\l have studied the hydrodynamics axial mixing and mass
transfer in OTRDC using the combined forward mixing and backflow model
fo r axial mixing.
In addition to the above discussed models there are some empirical
correlations that predict the hold-up in an extraction column. Sarkar et al /37/
have correlated the experimentally determined hold-up as a function of rotor
or impeller speed column height and dispersed phase velocity using the
toluene-water system. The correlation cannot be taken as a general one sincethe constants in the correlation may not be same for a RDC with a different
configuration and when a different system is used. However Kumar and
Hartland 738 39/ have suggested correlations for predicting the mean drop
size and hold-up in a RDC as a function of column geometry operating
conditions and physical properties.
ii ni Column
This type of column has been used for separation of aromatic and aliphatic
hydrocarbons. Modeling studies reported for this type of column are
relatively few in number an d include both pseudo-homogenous an dpopulation balance models. A dynamic pseudo-homogenous dispersion model
has been applied for simulating a Kiihni column by Hufnagl et al /40/ and
Hufnagl and Blass /41/ assuming constant flow rate of both the phases and
constant levels in the column. The hold-up and the Sauter mean diameter have
also been assumed to be constant along the length of the column. At thecolumn inlet and exit the model includes a mixer and a dead-time term to
describe the separation chamber. Correlations for estimating the Sauter mean
diameter relative phase velocity axial dispersion coefficient and mass
transfer coefficient have been taken from literature and in some cases thecorrelation constants have been modified to obtain a best fit with
experimental data. Correlations for the distribution coefficient and interfacial
tension have been suggested that are dependent on solute concentration. This
model is a component part of the DIVA simulator /42 43/. Simulated
responses fo r disturbances in feed concentration and flow rate have been
compared with experimental data. In the case of a disturbance in the feedconcentration the simulated curve at feed inlet attained steady state faster
216 Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
than the experimental curve. This may be due to the mixing in the feed line
that has not been accounted for by the model. When an increase in flow ratesof both the phases occurs, there is a sudden overshoot in the concentration
due to the level controller action to maintain constant interface level. The
deviation of the simulated values from the experimental value are due to the
assumption that the flow rate of both the phases are constant, whereas in
actual practice, due to the counteraction to maintain the interface level, there
is significant deviation in hydrodynamics. The model has not taken into
consideration the level controller.
In addition to the pseudo-homogenous dispersion models, several modelsbased on drop population balance have been reported. The differential
population balance model given in equation 4) was used to perform a steady
state simulation of the Kühni column by Haunold et al /44/. Coalescence has
been neglected whereas breakage has been described by a breakage
probability function function of impeller speed and Weber number) and a
daughter drop size distribution function. All the parameters have been
determined experimentally. However, comparison between the predicted and
experimental hold-up shows that the agreement is satisfactory only at lowflow rates. Steiner /45/ has presented a population balance model based on
geometrical droplet size classification for estimating the hold-up in Kühn i
column. The advantages and disadvantages of population balance models
have also been discussed.
The transient stagewise population balance model by Gerstlauer et al /46/
neglects coalescence and assumes a binary breakage of drops. The model
considers each drop in the dispersed phase to be characterised by the mass of
the solvent and the mass fraction of the solute. The breakage rate is predictedby the model proposed by Tsouris and Tavlarides /47/ and the daughter drop
distribution is predicted using the method of Coulaloglou and Tavlarides /48/.
Although the model considers polydispersity of the drops, all drops are
assumed to have the same velocity, which is not true in reality. The mass
transfer coefficient has been calculated using the Handlos and Baron /16/ and
Heeitjes et al /49/ models. The Gerstlauer et al /46/ model has been used to
simulate the startup behaviour of a Kühni column for drop number
distribution, dispersed phase hold-up, and solute mass fraction. No
comparison has been made with experimental data in order to check the
validity of the model. In their later work /50,51/, the same model has been
simplified by assuming that the solute concentration is same in all the drops.
217rought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
Vol. 16 No. 3 2 Modeling of Liquid-Liquid Extraction Column:
A Review
The backflow of the continuous phase has been calculated by the wake
volume of the continuous phase carried away by the dispersed phase. Thesimulated start-up and step change behaviour of the column agreed well with
measured solute concentration profile. This model has been incorporated into
the DIVA simulator.
Spray Column
This is the simplest of all extractors and is used for washing treating and
neutralisation. The disadvantage of this type of column is its poorperformance due to intensive continuous phase backmixing. Not many models
have been reported for this type of column.
A hydrodynamic model for a spray column operating in the dense packing
region of drops has been developed by Noworyta and Kroti 1521 taking into
account the drop coalescence. The model assumes uniform sized drops across
any cross-section and constant flow rates of both the phases. The analytical
solution of the differential equation has been presented for predicting the
hold-up along the length of the column which is found to agree well withsystems with low coalescence.
Seibert and Fair 1531 developed mechanistic hydrodynamic and mass
transfer models based on fundamental principles fo r spray columns. The keyassumptions are that the drops are spherical and drop size can be represented
by the Sauter mean diameter. Models for determining the drop diameter drop
velocity drop-drop interaction hold-up and flooding velocity along with a
model for estimation of the mass transfer coefficient from fundamental
principles with limited number of empirical correlations have been presented.Comparison between experimental and predicted hold-ups shows that the
agreement is good at low hold-up but deviates significantly near flooding
regions. The predicted and experimental overall mass transfer coefficients
based on the continuous phase agreed quite well.
Schermuly and Blass /54/ have developed a model for a 3-component
mass transfer in a spray column. The system chosen is
glycerol-acetone-water. The model takes into account the variation of
concentration and mass flow rate along the column and liquid phase
backmixing The mass transfer coefficients are based on both the theoretical
and experimental results available in literature on mass transfer in saturated
and unsaturated phases. A correction for the eddy diffusivity parameter
218 Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
calculated using the Handlos Baron formula and the backmixing parameter
has been determined by adjusting these parameters using a best fit criterionbetween the calculated and experimental data. Since adjustable parameters
have been used, the model cannot be confidently used for predicting the
column performance adequately.
Kumar and Hartland 7557 have proposed a correlation for predicting the
hold-up for a spray column in the loose and dense packed region of operation
for with and without mass transfer. The correlation is a function of column
geometry, physical properties and known operating parameters.
Scheibe Extractor
Literature on modeling of Scheibel extractor is relatively sparse and
includes works by Alatiqi et al 756,577 and Pang and Johnson 7587. Alatiqi et
al 7567 have developed a stagewise, backflow model for mass transfer in a
Scheibel extractor. Each stage is assumed to consist of a mixing zone and a
settling zone. In the mixing zone, mass transfer takes place while coalescence
takes place in the settling zone. A correlation for hold-up which is a functionof agitation speed and phase ratio has been suggested for predicting the
hold-up at different heights of the column. The distribution coefficient is also
expressed as a function of solute concentration. Physical properties, such as
viscosity and surface tension, have been correlated as functions of solute
concentration. For estimating the continuous phase and drop-side mass
transfer coefficients, the correlations by Gamer and Tayeban 7597 and Rose
and Kintner 7607 have been used, respectively. The three parameters that are
adjusted to minimise the discrepancy between the simulated and experimentalresults are the two backmixing parameters one for the continuous and the
other for the dispersed phase) and the overall mass transfer coefficient. It was
found that the two backmixing coefficients were close to zero and hence were
neglected and the simulation was done with one adjustable parameter. This
shows that extrapolating a single droplet mass transfer correlation to droplet
swarms does not accurately predict the mass transfer coefficient. Steady-state
analysis studies were carried out and they found that the agitation speed and
the phase ratio had considerable influence on the control structure.
In their further studies, Alatiqi et al 7577 have studied two types of
dynamic stage-wise models, one equilibrium and the other non-equilibriumwith backflow, assuming constant mass transfer coefficient flow rates and
219rought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
Vol. 16 No. 3. 2000 Mo deling o f Liquid-Liquid Extraction Co lumn:A Review
hold-up. Response to step disturbances in flow rates, solute concentration,
agitation speed and backmixing coefficients were studied. However, noverification has been m ade w ith experim ental data.
Transient response studies have been reported in simulation of
liquid-liquid extraction column. Pang and Johnson /58/ have developed a
stagewise dynam ic m odel for a m odified Scheibel ex tractor and have studiedthe frequency response to a pulse input at the feed concentration in the rangezero to 1.2 radian/minute. The model equations have been transformed to the
frequency domain to obtain the frequency response instead of the time
domain transient response for any disturbance in the feed concentration. Thestages were assumed to be non-identical i.e., the overall mass transfercoefficient was assumed to vary from stage to stage) and undernon-equilibrium conditions. The mass transfer coefficient, the time delay andthe number of mixing stages at both ends of the column, as well as thefraction of the feed and the solvent entrained at two ends of the column, weretaken as fitting parameters and were obtained by obtaining a best fit withexperim ental and predicted freq uency response am plitude ratio and phase
shift). When the predicted frequency response was transformed to the timedomain and compared with experimental transient response, the agreementwas good.
Packed Column
Packed columns are preferred to spray columns since they increase the
drop coalescence and breakage and decrease axial mixing, which results in
improved performance. Inspite of their popularity, not many modelingstudies have been reported in the literature. The mechanistic hydrodynamicand mass transfer model fo r spray columns developed by Seibert and Fair 1531
has also been extended to packed columns and has been applied to varioustypes of packings. However, the axial mixing of the continuous and dispersed
phases has been neglected and a tortuosity factor has been introduced in the
drop velocity equation. The discrepancy between the predicted hold-up and
overall mass transfer coefficient was higher than that for spray columns.
Steiner et áú . /61/ have presented simplified differential as well as stagewisemodels for predicting the hydrodynamic and mass transfer performance of an
extraction column filled with regular packings. For the differential model, thecoalescence and breakage time, the drop velocity, and the mass transfer
220 Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
Vol. 16 No. 3 2 Modeling o f Liquid-Liquid Extraction Column:A Review
áú. 73Ë For estimating the hold-up, the following model is proposed:
1-Ö) 2 9)
where L = -=-;
where Kk =30 for rigid drop and 15 for circulating drops. The modelpredicted the hold-up with reasonable accuracy for agitation rates greater than3 cm/s. The model does not take into account transfer of solute from the
dispersed to the continuous phase which enhances the coalescence rate and
hence the model is not suitable fo r mass transfer from dispersed to continuous
phase. For the mass transfer, a stagewise model has been proposed with axialdispersion in both the phases. A linear equilibrium relationship has beenassumed and the aqueous phase mass transfer resistance has been neglected.The axial dispersion coefficient for both the phases were estimated using thecorrelation given by Hafez /74/. The overall mass transfer coefficient was
taken as the adjustable parameter to obtain a best fit between the experimentaland predicted aqueous phase outlet solute concentration. The system chosenwas water-acetic acid-kerosene. This shows that the correlation used for
estimating the mass transfer coefficient is not suitable for the system chosen.Sovov et al. Ð5 / have proposed a model for prediction of mean drop size
in any stage for a vibrating plate extractor VPE) or Prochazka RFC. In thismodel, an exponential rate of change of mean drop size has been assumed
12)
where the constant, Kp depends on the plate type and dm is the limiting dropsize. The total hold-up in any stage is the sum of the hold-up in the denselypacked layer zone and the hold-up in the loosely packed zone. A semi-
224 Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
redispersion are accounted for by assuming that all the drops coalesce and
redisperse thereby maintaining the original drop size distribution. The numberof such mixing stages introduced into the column is left as a free parameter
and is obtained from the measured concentration profiles. A logarithmic-
normal distribution of drop size is assumed. In addition, it is assumed that
coalescence takes place between two equal sized drops and a drop breaks into
two equal sized drops. The rate constants for the breakage and coalescence
have been taken as the adjustable parameters and were fitted to get good
agreement with experimental data obtained in a pilot-plant agitated column.
Caminos et al. ÐºÉ have developed a steady-state stagewise model forextraction columns assuming each stage to be in equilibrium using phase
equilibria obtained by the UN1FAC group contribution method /78/. A total
and component balance has been made for each stage. The model can be
applied to both stage extractors as well as sieve plates and has been applied to
industrial extractors. A solution method that requires less computation time
has been suggested. Sereno et al. /79f have assumed a constant hold-up along
the length of the column and have obtained a dynamic mass balance equation
fo r describing t he concentration in both phases o f the system. Instead o f using
a rigorous thermodynam ic model for describing the distribution ratios of the
ternary system, local non-linear models have been used that are valid over
moderate range of composition, which saves considerable computer time. The
distribution ratio, which is first evaluated by using a rigorous method such as
UNIFAC is then fitted by polynomial approximation in the follow ing form
In K-t = LljJc^f L2y?ef +L3, i = 1 2 3
where LI , L 2 and L 3 are adjustable empirical local model parameters, and xrefand yref are mole fractions of the reference component in the raffmate andextract phases, respectively. The dominant component is chosen as the
reference component. Introducing time derivative of the distribution ratios inthe general mass balance equation leads to a differential mass balance of
component i in the equilibrium unit. The differential equation is solved byGEMS General Equation Modelling System) /80/. The equation for a single
equilibrium unit has also been modified to account for a multistage model.Each stage is treated as an equilibrium stage and no backmixing between the
stages has been assumed. The model-predicted concentration profiles of the
226 Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
raffmate and extract phases in multistage extraction columns ha s been
compared with those predicted by other workers for extraction of cyclohexane
from ç-heptane using furfural as the solvent and extraction of acetone from
water using 1 1 2-trichloroethane as the solvent. The model predictions were
in exact agreement with that of other workers.
PROCESS SIMUL TORS
Simulators with built-in models applicable to various process industriesare also available. Black /81/ gives a brief description of a process simulator
/82/ and its application to various petroleum and petrochemical industries for
Vol. 16 No. 3 2 Modeling of L iquid-Liquid Extraction Column:
A Review
this method, the convergence of the solution of the model equations is very
fast without making use of any special convergence-promotion techniques.Chatzi and Lee 7867 have suggested a numerical solution for thehomogeneous model developed for a batch agitated liquid liquid extractionprocess taking into account the drop breakage and coalescence. An algorithmfor solving stiff differential equations using a semi-implicit Runge Kutta
method and its application to liquid-liquid extraction has been proposed by
Michelsen/87, 887.Kronberger et al 7897 have suggested a new solution method for solving
the system of differential population balance equations. The system of eachIntegra Differential Equation IDE) mass balance and volume balanceequations) is discretised with respect to the drop diameter by applying the
Galerkins projection technique to give n Partial Differential Equations(PDE s) where n is the number of drop size intervals considered. The coupledPDE s are then space-time discretised by using finite-volume technique. The
spatial derivatives in the dispersive flux are approximated by centraldifferences and integration in time is performed by the Euler method. This
method ha s been implemented for simulation of an extraction column that iscalculated using the flux-extrapolation method with a time increment of 0.1 s.
Stable discretisation of the convective term has been incorporated. Ribeiro et
al . 7907 suggest an algorithm for solving the population balance equation pbe) that adequately predicts the drop distribution function for a perfectlymixed continuous or batch vessel with spatially homogeneous, stirred
turbulent dispersion. A first-order, explicit finite-difference method has beenused to solve the pbe. For the chosen systems, toluene-water, and methyl-isobutylketone-water, a logarithmic grid was used for discretization of the
pbe s. This algorithm has also been extended to simulate a trivariate (drop
volume, age and solute concentration distribution) with unsteady statebehaviour of the dispersed phase by Ribeiro et al 7917 and was found to be
fast enough to be applied to control of extraction columns. Regueiras et al
1921 have proposed a new simple algorithm for drop population balance thatfurther reduces the computation time and is therefore suitable for process
control. Instead of calculating the solute concentration distribution for each
drop class at every time interval, it calculates the mean solute concentrationand standard deviation for the solute concentration. However lack of
experimental verification limits confidence in this algorithm.
228 Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
Vol. 16, No. 3, 2000 Modeling of L iquid-Liquid Extraction Column:A Review
transfer studies, where the mass transfer rate depends on the surface area ofthe drop and the am ount of solute transferred into the drop volum ed
32 seems
to be most suitable and has been used by most of the authors. Howeverd43 defined as d43 = ZnidjV £n dj3 has been used for estimation of slip velocity.Misek /31/, discusses the error involved in characterising the dispersion bySauter m ean drop diameter and has justified the use of d43.
Purely empirical correlations based on column geometry and operatingconditions have been developed for predicting th e mean drop sizes in varioustype of extraction colum ns. Ku mar and H artland /38/ have critically reviewedthe available correlations and have proposed a correlation fo r predicting thedrop size in a RDC. Prediction of the drop size in a spray column from amulti-nozzle distribution has been suggested by Dalingaros et al. 1951. Forpulsed perforated plate columns the correlation proposed by Kumar an dHartland /95/ and Sovov Ð5É are more reliable. Kumar and Hartland /97/
have presented a unified correlation for predicting the drop size in eightdifferent types of extraction column.
rop Size istribution
Several au thors have tried to fit drop size distribution with different typesof statistical distributions of which the most commonly used ones are givenbelow.
The arithm etic normal distribution function has been used by Qian /15/ todescribe the drop size distribution for a pulsed plate extraction column. Thefunction is given as
8 d) = \ e Xp\- d-d 43 ) 2 /2a 2 ] 17)
\° ·4 3
< / 4 3 0.0123 Ì fA/;~°· 70 18)\Pc)
19)
where ó is the standard deviationThe basic equation of Mugele and Evans /98/ to describe the drop
230 Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
back to the column. The axial mixing of the dispersed phase is due to the
difference in drop sizes that results in difference in drop velocities. Severalcorrelations have been reported in literature for predicting the axial mixing
coefficients of the continuous and dispersed phases for different types of
extraction columns, but the percentage error in some cases is very high, lying
between 25-70%/ /.
. For continuous phase:
Kumar and Hartland /l 067 have presented a correlation for prediction of
the axial dispersion coefficients for RDC by making use of steady state tracerinjection and dynamic tracer injection technique. For the Kühni column,
Breysse et al /107/ have proposed a correlation for both single and two phase
flow. The correlation by Kumar and Hartland /108/ and Prvcic et al /109/ is
applicable to pulsed sieve-plate columns. A physical model has been
proposed by Stevens and Baird /1107 for single-phase axial mixing in a
single-stage Karr column. Two regimes, i.e., poorly mixed and well-mixed
regimes, have been considered to predict the overall axial dispersion
coefficient. The hydrodynamic model predicts the distance for which the
mixing is poor and the distance for which the mixing is good. Two
parameters must be estimated. The model is proposed based on flow rate,
frequency, amplitude, plate spacing, hole size, and fractional free area. These
authors give a good summary of work carried out by other workers on the
estimation of the axial dispersion in PSE and RFC. A correlation has been
proposed by Karr et al 71117 for axial mixing in a Karr column.
Steiner et al 71127 have suggested a method for experimental
determination of the axial dispersion co efficient for the continuous phase and
have fitted the data with an empirical correlation. For the dispersed phase, the
data seemed to be scattered with the correlation proposed.
Godfrey et al 71137 have reviewed the work carried out by various
authors on single-phase axial mixing in pulsed sieve plate extraction columns
and have proposed a correlation for stationary phase axial mixing with the
objective of getting better co-efficient required in calculating the mass
transfer performance. The phenomenon of axial mixing is generally described
by the usual effective diffusional model. They have used the idea and
technique developed by May I I147 Misek 71157 and Miyauchi and Oya 71167
for determining the stationary phase axial mixing coefficient.
233rought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
Correlations available for predicting the dispersed phase axial mixingcoefficient are fewer in number and are unreliable as the error involved is
quite significant. Most of these correlations apply to specific situations.Kumar an d Hartland /106/ have proposed a correlation fo r axial mixingcoefficients in RDC and have also reviewed other available correlations. For
the pulsed sieve-plate extraction column, the correlation by Miyauchi and
Oya /Ð5/ is more commonly used.
ss Transfer oefficient
Mass transfer coefficient measurements are based on two theories:
a) Whitman s two-film theory and b) Higbie s penetration theory.For the continuous phase, the correlation proposed by Steiner /l 17/ is
recommended. The equation given by Lochiel and Calderbank IIW derivedfrom the diffusion continuity equation is recommended fo r intermediate and
high Reynolds numbers (Re = p d V / ì ).
For single drop mass transfer coefficient of the dispersed phase, theNewman s model 1251 for rigid drop and Kronig and Brink s model /34/ fo rcirculating drops have been used. For mass transfer in drop swarms, the
model by Yaron and Gal-Or /119/ is recommended. Korchinsky and
Cruz-Pinto /120/ have suggested a method fo r applying th e rigid drop an dHandlos-Baron drop models fo r evaluating the mass transfer co-efficientwhen th e continuous phase concentration is not constant and the drop sizes
are not uniform. Bahmanyar et al. /121/ have examined th e rate of masstransfer from a single drop in RDC s, pulsed sieve-plate columns, and packedcolumns and have compared these with those estimated by Handlos and
Baron model to incorporate necessary corrections.
ON LUSIONS
From the various types of models reviewed, it can be concluded that these
fall into tw o groups: 1) the diffusion model with turbulent back diffusion ofsolute superimposed on plug flow of both the phases, and (2) the backflowmodel with well-mixed non-ideal stages between which backflow occurs.The major difficulty lies in proper representation of the dispersed phase. The
234 Brought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
pseudo-homogenous model assumes the dispersed phase to be pseudo-
homogenous whereas the population balance model treats it as a discreteelement. The drop population balance model takes into consideration the
previous history of each drop as it moves along the column, i.e., its breakage
and coalescence, to give birth to new drops of different sizes. The population
balance models have been found to represent the real system more closely
than the pseudo-homogenous model. As far as the population balance model
is concerned, the major difficulties lie in obtaining the break-up an dcoalescence parameters. Several researchers have assumed that the drops
break into 2 or 3 equisized daughter drops. In recent years, however, studieshave been carried out to determine the drop size distribution for various drop
sizes and fitting them to appropriate statistical functions. The constant
parameters depend on the system and operating conditions chosen. These
parameters for some systems are available in literature, but a lo t of
experimental data remain to be generated for determining the parameters.
Except for a few, the coalescence parameters in most models have been
estimated by obtaining best fi t with experimental drop size distribution data.
Even those who have used experimental values have made assumptions that
are not realistic, such as coalescence only between adjacent and same drop
sizes. However, none of the methods available to date is fully satisfactory and
more research in the area of d etermination of coalescence parameters for drop
swarms would be helpful in obtaining a better model. Applicat ion of
advanced modelling techniques, such as, Computational Fluid Dynamics
CFD), may lead to a better model, but may be extremely complex since the
dispersed phase continuously changes in size and composition.
K N O W L E D G E M E N T
The author gratefully acknowledges the f inancial assistance of the
Alexander von Humboldt Foundation, Bonn; Prof. Alfons Vogelpohl, Institutfü r Thermische Verfahrenstechnik, TU Clausthal, Germany, for his help an dcooperation; and the Director, Regional Research Laboratory C.S.I.R.),
Bhubaneswar, for permission to publish this paper.
35rought to you by | SUNY Buffalo LibrariesAuthenticated | 128 205 114 91
Download Date | 3/20/13 3:27 PM
8/12/2019 MohantyS-2000_Modeling of Liquid-Liquid Extraction Column_mohanqwety (1)
105. Hussain, A.A., T. B. Liang and M.J. Slater, Characteristic Velocity
of Drops in a Liquid-Liquid Extraction Pulsed Sieve Plate ColumnChem. Eng. Res. Des. 66,541-554 (1988).
106. Kumar, A. and S. Hartland, Prediction of Axial M ixing Coefficients
in Rotating Disc Extraction Columns , Can. J. of Chem. Eng. 70,
77-87(1992).
107. Breysse, J., U. Bühlmann and J.C. Godfrey, Axial MixingCharacteristics of industrial and Pilot Scale Kiihni Column , AIChESymp. Ser. 80(238), 94-101 (1984).
108. Kumar, A. and S. Hartland, Prediction of Continuous Phase AxialMixing Coefficients in Pulsed Perforated-Plate Extraction Column ,
Ind. Eng. Chem. Res. 28,1507-1513 (1989).
109. Prvcic, L.M., H.R.C. Pratt and G.W. Stevens, Axial Dispersion in
Pulsed-, Perforated-Plate Extraction Column , AIChE J. 35,
1845-1855 (1989).110. Stevens, G.W. and M.H.I. Baird, A Model for Axial Mixing in