Fluid Phase Equilibria 194–197 (2002) 783–803

Conceptual design and analysis methodology for crystallization

processes with electrolyte systems
Kiyoteru Takano a,1 , Rafiqul Gani a,∗ , Takeshi Ishikawa b , Petr Kolar c
a
CAPEC, Department of Chemical Engineering, Technical University of Denmark, DK-2800, Lyngby, Denmark
b
Yokohama Research Center, Mitsubishi Chemical Corporation, 1000 Kamoshida-cho, Aoba-ku, Yokohama, 227-8502, Japan
c
Mitsubishi Chemical Corporation, Ushiodori 3-10, Kurashiki, Okayama, Japan
Received 20 May 2001; accepted 6 September 2001

Abstract
A methodology for multi-level conceptual design and analysis of crystallization-based separation processes
involving electrolyte systems is presented. The methodology consists of three main parts: thermodynamic part
(level 1), flowsheet design/analysis part (level 2) and flowsheet validation (simulation) part (level 3). In this paper,
the integration of the thermodynamic and design/analysis parts is presented. This integration allows the analysis of
the thermodynamic behavior of a given system together with design/analysis of flowsheet alternatives in the same
calculation environment. The integrated system is highlighted through four practical examples where features of the
algorithm related to creation of property model package and its usage in flowsheet design/analysis are highlighted.
© 2002 Elsevier Science B.V. All rights reserved.
Keywords: Crystallization; Electrolyte system; Method of calculation; Solid–fluid equilibria; Model; Design

1. Introduction

Solution chemistry and solid–liquid (phase) equilibrium play a very important role in design, synthesis
and analysis of crystallization-based separation processes involving electrolytes. The solid–liquid equilib-
rium (SLE)-phase diagrams can be used to identify the feasible operation paths for a desired product from
a specified feed mixture. They also help to identify the separation boundaries, the condition of operation
(such as temperature of operation), the list of solids that are most likely to precipitate and many more. In-
formation from the generated SLE-phase diagrams, such as phase boundary data and saturation point data,
may be used to solve graphically the mass balance equations related to a crystallization operation. Since the
graphical solution of the mass balance equations is related to the operational paths on the phase diagrams,
∗
Corresponding author.
E-mail addresses: [email protected] (K. Takano), [email protected] (R. Gani).
1
Present address: On leave from Yokohama Research Centre, Mistubishi Chemical Corporation, 1000 Kamoshida-cho,
Aoba-ku, Yokohama, Japan. Tel.: +81-45-963-3269; fax: +81-45-963-3947.

0378-3812/02/$ – see front matter © 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 8 - 3 8 1 2 ( 0 1 ) 0 0 7 0 5 - 1
784 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803

the temperature dependent phase boundaries and the solubility index (indicates which solid is likely to pre-
cipitate first), it is possible to simultaneously (graphically) design, analyze (including evaluation of flow-
sheet economics) and simulate both continuous flowsheets with crystallization operations and sequences
of batch operations. The simultaneous graphical solution not only provides a visualization of the process
operation, but also, provides very good initial estimates for future simulations with a rigorous model.
Reliability of the developed flowsheet design methodology largely depends on the accuracy of SLE-
phase diagrams and reliable computation of SLE-phase diagrams requires the use of an appropriate
set of property models and corresponding model parameters. Electrolyte systems are characterized by
the presence of inorganic and/or organic salts, water, organic chemicals and non-condensable gases.
Because of the complex nature of thermodynamic behavior of electrolyte systems, the application range
of property models are very limited. Furthermore, especially in the early stage of the development of
the pharmaceutical/biochemical processes, only limited number of experimental data is available and
in many cases the usage of the property models outside the range of the available experimental data
may be necessary. This makes the development of a computer aided approach difficult for processes
with electrolyte systems. Therefore, the selection of appropriate property models for the given problem
and system, and, the determination of the correct set of corresponding model parameters from limited
experimental data is an important step.
In earlier works related to the design of crystallization-based separation, Fitch [1] and Dye et al. [2] have
proposed synthesis methods for fractional crystallization processes using various types of solid–liquid
equilibrium based phase diagrams. Fitch [1] and Dye et al. [2] assumed solubility to be a linear function
of temperature. More recently, Samant et al. [3] have shown the use of the component ramping concept
where multicomponent systems are transformed into sub-sets of ternary and quaternary systems for use
in a graphical solution approach.
In the multi-level computer aided system, the thermodynamic part (level 1) “creates” the problem
specific property model package, which is a collection of pure component and mixture property models,
analyzes the solubility of salts and generates phase diagrams. The design and analysis part (level 2)
generates process (flowsheet) alternatives, evaluates/analyzes feasibility of separation and provides a
visual operation path for the desired separation based on thermodynamic insights, which are given by
solid–liquid equilibrium phase diagrams. The simulation part (level 3) consists of a simulation engine
that allows the screening and validation of process alternatives generated in level 2.
This paper mainly deals with the application of the thermodynamic part and the integration of thermo-
dynamic part with the flowsheet design/analysis part. A detailed description of the design/analysis part and
the models for the flowsheet simulation part can be found in Takano [4] and Takano et al. [5]. In the ther-
modynamic part of the integrated methodology, the problem and system specific property model package,
which is the set of the appropriate property models and the corresponding model parameters, is created
in a systematic way for use by the calculation routines as well as for design, analysis and simulation. The
appropriate property models are selected according to a set of selection rules that take into account, appli-
cation range of the model and the “known” limitations of the models. When experimental data is available,
the model parameters are determined through data regression combined with a sensitivity analysis, which
identifies the most sensitive model parameters to the needed property values. The created property model
package is used for solubility analysis, for process simulation and for computation of (SLE)-phase dia-
grams, which is used for flowsheet design/analysis. In the integrated methodology, the thermodynamic
part and flowsheet design/analysis part have been combined such that it is possible to analyze the ther-
modynamic behavior of the given system as well as of the process flowsheet in the same environment.
 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803 785

Four case studies are used to highlight the creation of the problem- and system-specific property
model package and the use of the thermodynamic insights of electrolyte systems for design/analysis of
crystallization-based separation processes. The validation of the generated alternatives through the use
of the simulation part is also highlighted for one example.

2. Methodology

As shown in Fig. 1, the methodology for design and analysis of crystallization-based separation
processes consists of three parts (or levels), a thermodynamic calculation part (level 1), a process de-
sign/analysis part (level 2) and a simulation part (level 3). The thermodynamic calculation part mainly
includes definition of the electrolyte system, creation of an appropriate property model package,

Fig. 1. Overall structure of the algorithm to design the fractional crystallization processes.
786 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803

Table 1
Main steps for creation of property model package
Step number Step action

Step T1 Specify the problem (what needs to be determined or calculated?)

Step T2 Define the system (compound expansion in terms of species, ions, and elements present in the system)
Step T3 Select the property model “package” (selection based on problem and electrolyte system from three
alternative model packages)
Step T4 List the physical properties and models (identification of the needed properties and property models for
the selected model package)
Step T5 Data regression and sensitivity analysis (tuning of available model parameters and estimation of
unavailable model parameters)

solubility analysis and phase diagram calculation. The process design/analysis part deals with gener-
ation and analysis of feasible flowsheet alternatives including conditions of operations. The simulation
part provides the calculation engine necessary for analysis, screening and validation of generated/specified
flowsheets. A brief description of the thermodynamic and design/analysis parts is given below together
with aspects of integration of these parts.

2.1. Thermodynamic part

This part of the methodology creates the problem and system specific property model package, performs
calculations related to solubility analysis, SLE-phase diagrams and many more. Since the generated
property model package is used in the steps involving solubility analysis and computation of phase
diagrams, creation of the appropriate property model package is a very important first step.
As listed in Table 1, the “creation” of the problem and system specific property model package consists
of five main steps. The creation of the problem specific property model package is based on the observation
that electrolyte systems have special behavior that cannot always be captured through a general property
model. Therefore, identification of the system specific features and the corresponding property model
that allows these features forms the basis of the creation of the appropriate property model package.
In the first step, the compounds present in the process (inorganic salts, organic salts, water, organic
solvents and gases or non-condensables) are identified together with the list of problems (calculations)
that are to be solved. In the second step, a compound expansion is performed to identify the ionic species
present in the system, the likely phases, the likely solids together with the association/dissociation reac-
tions. Based on the compound expansion information and the problem specification, a property model
package is selected in the third step. The four property model packages considered in this work are
classified in terms of activity coefficient model present in the package. The four activity coefficient
models considered are the Pitzer–Bromley–Debye Huckel model [6], the electrolyte-NRTL model [7],
the extended UNIQUAC model (aqueous) [8] and the extended UNIQUAC model (general). In the de-
veloped property model library, the extended UNIQUAC model (general) [4] has been developed by
modifying the extended UNIQUAC model (aqueous) by adding a “Born” term. For mixed-solvent sys-
tems, either the electrolyte-NRTL model or the extended UNIQUAC model (general) is applicable, as
other two models in the model library are not applicable for such problems (see Table 2). For the cho-
sen property model package and identified type of the problem, the necessary physical properties and
 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803 787

Table 2
Selection rule based on the application range (system type and ionic strength limitation)
Model name Ionic strength limit Aqueous system Non-aqueous system

Aqueous 1 Aqueous 2 Mixed Organic

Pitzer–Bromley–Debye Huckel 6 mol/kg Possible Possible

Extended-UNIQUAC (general) No limitation Possible Possible Possible Possible
Extended-UNIQUAC (aqueous) No limitation Possible
Electrolyte-NRTL No limitation Possible Possible Possible Possible

corresponding property models are selected in the fourth step based on the availability of data and/or
model parameters, and application range of the models. In the fifth step, if measured phase equilibrium
data is available, model parameters are validated or fine-tuned if all parameters are available. Otherwise,
a sensitivity analysis is performed for a range of conditions of interest to the specified problem in or-
der to identify the most sensitive set of parameters that have the highest values of right hand sides of
Eq. (1)
 exp exp 
dF
dFj
d i |(pi − pi )/pi | j
cal
= = , i = 1, . . . , NC and j = 1, . . . , NE (1)
dτ j
dτ j
dτ

In Eq. (1), F, p and τ represent the objective function value, property value and model parameter, respec-
tively. NC and NE are number of components and number of experimental data, respectively. This is a very
useful option since in most practical problems, availability of measured data is limited and therefore, not
all model parameters can be estimated. Also, for electrolyte systems, the property models need quite a
large number of model parameters. The sensitivity analysis provides initial estimates for data regression
and the analysis results can be used for evaluation of the effect of model uncertainties on the calculated
thermodynamic properties [4].
Once the property model package is created, solubility analysis identifies whether a selected crystal-
lization technique (fractional crystallization technique is considered in Fig. 1) is feasible for the desired
separation by calculating the effect of temperature/pH on the solubility of the products to be separated.
Since a fractional crystallization technique relies on the change of the solubility of salts with the change
of temperature, if solubility is not sensitive to the change of temperature, large changes in temperatures
are needed to move the solubility curve in a certain direction and consequently the technique may not be
feasible. Therefore, solubility analysis helps to screen out potentially uneconomical process flowsheet
alternatives before starting any flowsheet design work.
When solubility is identified to be sensitive to the change of temperature, the next step is computation
of solid–liquid phase diagrams. The algorithm for computation of solid–liquid phase equilibrium dia-
grams is applicable for three different types of systems: type 1, H2 O with two inorganic and/or organic
salts (ternary system); type 2, H2 O with three inorganic and/or organic salts (quaternary system); type
3, two solvents (H2 O and organic solvent) with one salt (ternary system). When the system is defined
and temperature is given, the algorithm calculates the solubility of solids by solving Eqs. (2)–(5) (given
below) simultaneously. Eqs. (2) and (3) represent the phase equilibrium relation and the molecular disso-
ciation/solid formation, while Eq. (4) represents the chemical charge balance and Eq. (5) represents the
788 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803

normalization of compositions
liq
fim = γ (T , P , x)xim Pism = fisoild
m
, im = 1, . . . , NM (2)
NC
 ν
Kj = ai ij , i = 1, . . . , NC and j = 1, . . . , NR (3)
i

NC


xi zi = 0 (4)
i

NC


xi = 1 (5)
i

In Eqs. (2)–(5), NM is number of molecular species, NC is number of chemical species, and NR is number
of chemical reactions (molecular dissociation/solid formation) to be considered. Also, T is the system
temperature and Ps is vapor pressure of pure component, zi the number of chemical charges of component
i, Kj the equilibrium constant for molecular dissociations and solubility product for reaction j. At a given
temperature and pressure, compositions, xi , are unknowns in Eqs. (2)–(5) and the solution of Eqs. (2)–(5)
gives solubility of the salt under consideration. The methodology to generate the correct number of the
equations is discussed in Takano [4]. Once the compositions are determined, pH is calculated as the
logarithm of the activity of hydrogen ion.

2.2. Process design/analysis part

Synthesis and design/analysis based on phase diagrams allow a quick, visual generation of fea-
sible flowsheet alternatives. Also, the solution from the synthesis/design algorithm serves as a very
good initial estimate for the simulation engine. Fractional crystallization is a commonly used sep-
aration technique for downstream processing of reactor effluent in pharmaceutical and biochemical
processes. They involve precipitation of solids from a liquid solution, which may also contain solid
particles. As more than one solid may precipitate out of the liquid solution, it is important to know
which solid should be allowed to precipitate and under what conditions (for example, feed compo-
sition and/or temperature). Solubility diagrams and solubility indices, both of which are based on
solid–liquid equilibrium behavior of the system, provide important thermodynamic insights to these
questions.
Consider the solid–liquid phase diagram for quaternary system consisting of one solvent and three salts
A, B, and C (see Fig. 2). Since the solid–liquid phase equilibrium for quaternary systems is represented by
tetrahedral diagram, the Janecke projection diagram is very often used for easy visualization of operation
paths (see Fig. 2) [9].
In Fig. 2, the three vertices indicate three solids, A, B and C and the three sides indicate the possible solid
complexes between two solids. The inference points also indicate solid complexes. The three saturation
lines, B1, B2 and B3, which indicate the precipitation of a solid from the liquid solution, indicate also the
separation boundaries. For example, with a feed at point F (see Fig. 2), solid C is possibly crystallized by
operating the crystallizer at 373.15 K after adjusting the amount of solvent. To calculate the amount of
solvent to be added/removed, solubility index is used. In Fig. 2, assume that feed point F is located above
 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803 789

Fig. 2. Janecke projection diagram for quaternary systems.

the saturation surface (solubility index of solid C at point F is less than 1), and therefore, not saturated
with respect to solid C. For the recovery of solid C, some solvent needs to be removed. Minimum amount
of solvent to be removed can be determined by identifying the point Fs , which is intersection of the line
connecting pure solvent, the feed point F and the saturation surface. At point Fs , the solubility index of
solid C is equal to one. On the other hand, when too much solvent is removed, the composition of the
product stream after evaporation may go into the region, where two solids precipitate together. Therefore,
the maximum value of the amount of solvent to be removed exists and consequently the range of the
amount of solvent to be removed for the recovery of desired product C is determined. Once the amount
of solvent to be removed (the composition of the product stream after evaporation is denoted by point F∗
in Fig. 2) is chosen within the determined range and then crystallizer is operated, the resulting solid (to
be precipitated) and the liquid composition can be determined graphically by mapping the mass-balance
line (see line connecting solid C and point M in Fig. 2) on the phase diagrams. Note that points F, Fs and
F∗ are represented by the same point in Fig. 2, since the compositions of these three points on solvent-free
basis are the same. The amounts of solid and liquid are calculated by the lever rule and only saturation
point data is needed in this procedure.
Analysis of the generated phase diagrams in terms of sensitivity to changes in temperature/pH also pro-
vides useful design insights. That is, how does the solubility of the specified product(s) change with respect
to temperature and/or pH? Consider the solid–liquid phase equilibrium diagrams for H2 O–alanine–serine–
l-asparagine quaternary system at three different temperatures (see Fig. 3a). Considering point 1 to de-
note the composition of the feed stream, it note that it is located in the region where (dl)alanine can be
crystallized at 343.15 K by adjusting the amount of H2 O, while at 313.15 K it is located in the region
where (l)asparagine can be crystallized by adjusting the amount of H2 O. This is explained by the shift
of phase separation boundary with the change of temperature and the developed (phase-diagram based)
790 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803

Fig. 3. (a) Solid–liquid phase diagram for H2 O–alanine–serine–l-asparagine quaternary system; (b) generated flowsheet for
H2 O–alanine–serine–l-asparagine quaternary system.

flowsheet design methodology exploits the movement of phase separation boundaries with the change of
operation condition such as temperature and/or solution pH.
The flowsheet design algorithm is applicable to the design of process flowsheets with/without recycle
for both ternary and quaternary systems. The algorithm uses both ternary and quaternary SLE-phase
diagrams. The main steps of the algorithm for the design of process flowsheets for ternary systems
are presented in Fig. 4. The algorithm is applicable to quaternary systems can be obtained from the
authors [4]. As shown in Fig. 4, the developed algorithm consists of seven main steps. In the first step,
the order of recovery, which is one of the important design variables in the flowsheet design problem, is
 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803 791

Fig. 4. Algorithm for design of flowsheet using ternary solid–liquid phase diagrams

identified through calculation of solubility index and physical insights. In the steps 5.2–5.4, the crystallizer
operations needed to obtain the desired solid products are determined together with the corresponding
operation temperatures and compositions/amounts of the product streams (solid and liquid stream). In
steps 5.5–5.7, the recycle streams are added, if necessary.

2.3. Integration aspects and simulation part

The proposed methodology gives a common environment for analysis part of thermodynamic behavior
of the system and design/analysis/simulation part of process flowsheet since these parts are developed
in an integrated manner. At the early stage of process design, in many cases, only limited number of
792 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803

experimental data is available for thermodynamic analysis. As the research goes on, the number of
available experimental data increases. Therefore, iterative interaction between thermodynamic analysis
part and flowsheet design/analysis/simulation part is necessary for increasing the reliability of process
design. In the proposed methodology, the information from thermodynamic analysis is easily incorporated
into the results of flowsheet design/analysis/simulation through the flexible data-flow of the methodology.
Consequently the design of process flowsheet is accelerated and performed in a systematic way.
In the developed methodology, the simulation engine (ICAS [10]) is mainly used for screening of
processing alternatives, for verification of design and for producing the operational data needed for
analysis. The simulation engine employs a generalized crystallizer model [5], which considers multiple
phases (vapor, liquid, solid), decanter and employs the created property package.

3. Application examples

Four case studies are presented. All the results were obtained through the computer aided tools imple-
mented in ICAS [10].

3.1. Case study A

In this case study, the combined effect of ethanol and potassium chloride on the solubility of ampicillin in
H2 O is studied [11,12]. In this example, the methodology to create the property model package discussed
in the thermodynamic calculation part (see Table 1) is highlighted. The problem description is given in
Table 3.

3.1.1. Step T1 (see Table 1): specify the problem type

The problem to be solved is, to determine the combined effect of ethanol and potassium chloride (KCl)
on the solubility of ampicillin in H2 O at 298.15 K. This problem can be solved in two steps. As a first step,
generate H2 O–KCl–ampicillin solid–liquid equilibrium diagram. In the next step, investigate the effect
of ethanol on H2 O–ampicillin–KCl phase diagram. Therefore, this problem is defined as solid–liquid
equilibrium (SLE) problem from the rule for the classification of the problem [4].

3.1.2. Step T2 (see Table 1): define the system

Table 4 shows the list of all chemical species considered in this example together with the information of
phase(s), ion type and salt type. The hatched part shows the input component while the other components
are generated through molecular dissociation. Since both H2 O and ethanol are identified as solvents, this

Table 3
Problem description for case studies A–D (case studies A and B are phase diagram computation problems)
Case study System Products Temperature (K)

A H2 O–ethanol–ampicillin–KCl 298.15
B H2 O–Na2 CO3 –NaCl–Na2 SO4 303.15
C H2 O–A–B A, B 393.15
D H2 O–serine–l-asparagine–alanine l-Asparagine, (dl)alanine, (dl)serine 393.15
 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803 793

Table 4
Definition of the system for case studies A–D
Serial no. Compound/ion names Ion type

A B C D A B C D

1 H2 O H2 O H2 O H2 O
2 Ampicillin Na2 CO3 A Asparagine Dipole Dipole Dipole
3 KCl Na2 SO4 B dl–Serine – Dipole Dipole
4 EtOH NaCl H(+) dl–Alanine – Cation Dipole
5 H+ Na(+) OH(−) H+ Cation Cation Anion Cation
6 OH− CO3 (2−) A(−) OH− Anion Anion Anion Anion
7 Ampicillin+ SO4 (2−) A(+) Asparagine(−) Cation Anion Cation Anion
8 Ampicillin− H(+) B(−) Serine(−) Anion Cation Anion Anion
9 K+ OH(−) B(+) Alanine(−) Cation Anion Cation Anion
10 Cl− Cl(−) Asparagine(+) Anion Anion Cation
11 Na2 CO3 –2Na2 SO4 Serine(+) Cation
12 Na2 CO3 –H2 O Alanine(+) Cation
13 Na2 CO3 –7H2 O
14 Na2 CO3 –10H2 O
15 Na2 SO4 –10H2 O

Serial no. Phases Salts

1 Vapor/solvent Vapor/solvent Vapor/solvent Vapor/solvent

2 Solid Liquid/solid Liquid/solid Liquid/solid Organic Inorganic Organic Organic
3 Solid Solid Liquid/solid Liquid/solid Inorganic Inorganic Organic Organic
4 Solid Vapor/solvent Liquid Liquid/solid Inorganic Organic
5 Liquid Liquid Liquid Liquid
6 Liquid Liquid Liquid Liquid
7 Liquid Liquid Liquid Liquid
8 Liquid Liquid Liquid Liquid
9 Liquid Liquid Liquid Liquid
10 Liquid Liquid Liquid
11 Solid Liquid Inorganic
12 Solid Liquid Inorganic
13 Solid Inorganic
14 Solid Inorganic
15 Solid Inorganic

system is classified as “mixed-solvent system”. The hatched part denotes the input information and all
the other information are generated automatically from the input information.

3.1.3. Step T3 (see Table 1): select the property model for activity coefficients
From step T2, the given system is defined as “mixed solvent system”, which is shown in hatched part.
Based on the selection rules given in Table 2, any of four activity coefficient models is applicable to the
given problem (SLE problem), while only the extended-UNIQUAC(general) or the electrolyte-NRTL
model can be used for the mixed solvent system. In this example, the electrolyte-NRTL model is
selected.
794 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803

Table 5
List of problem specific physical properties for case study A
Properties name VLSE SLE LLE VLE

Pure properties
Density ∗
Vapor pressure ∗ ∗
Enthalpy of formation ∗
Heat capacity ∗
Gibbs energy of formation
Mixture properties
Fugacity coefficient ∗ ∗
Activity coefficient ∗ ∗ ∗ ∗
Dissociation constant ∗ ∗ ∗ ∗
Solubility ∗ ∗
Henry’s constant

3.1.4. Step T4 (see Table 1): list the physical properties and select the models
In this step, the list of necessary physical properties is created and the corresponding property models are
selected from the list of the models. The creation of the list is made in three steps. First, problem specific
properties are added to the list. Since the type of problem is “SLE”, the shaded column in Table 5 lists the
necessary properties for SLE-phase diagram calculations—activity coefficient, solubility and dissociation
constants. In the next step, since the electrolyte-NRTL model has been selected as the activity coefficient
model, the model specific properties that are needed are listed in Table 6 (shaded column)—Born radius,
density, and dielectric constant. Since this system does not include non-condensable gases, to which
Henry’s law may be applicable, and all of equilibrium constants are known (which means that they do not
have to be calculated from thermodynamic properties), no additional properties are needed. The complete
list of necessary physical properties for the created property model package is listed in Table 7 together
with the selected property models. Once the list of properties and property model parameters has been
made, analyses of the pure component properties and of mixture properties are made by comparing the

Table 6
List of model specific physical properties for case study A
Properties name Pitzer NRTL UNIQ(aq) UNIQ

Pure properties
Born radium ∗ ∗
Van der Waals volume ∗ ∗
Van der Waals surface area ∗ ∗
Density ∗ ∗ ∗ ∗
Dielectric constant ∗ ∗ ∗
Mixture properties
Density ∗ ∗
 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803 795

Table 7
List of necessary physical properties and corresponding models for case studies A and D
Case study A Case study D

Property name Property model Property name Property model

Pure properties
Born radius Data Density Racket
Density Racket Vapor pressure Antoine
Dielectric constant Correlation Enthalpy Data
Heat capacity Correlation
Gibbs energy Data
Born radius Data
Dielectric constant Correlation
Mixture properties
Activity coefficient Electrolyte-NRTL Activity coefficient Electrolyte-NRTL
Dissociation constant Data Dissociation constant Data
Solubility Solubility product Solubility Solubility product

calculated values with the experimental data, if available. Since limited experimental data is available,
data regression is recommended.

3.1.5. Step T5 (see Table 1): sata regression with sensitivity analysis
In step T4, the analysis of the selected model showed that the model did not describe the entire range
of experimental data and consequently, data regression was recommended. That is, regression of activity
coefficient data was necessary for the H2 O–KCl binary system. Sensitivity analysis was made based on
the calculation results of Eq. (1). After the sensitivity analysis, the interaction parameter between H2 O
and a pair of K+ and Cl− were found to be the most sensitive. Therefore, the interaction parameter
between H2 O and a pair of K+ and Cl− were adjusted to match the experimental data, as shown in
Fig. 5.

3.1.6. Final result

Through these five steps, the property model package was completed for the given problem and system.
Once the property model package is created, the given problem is solved. Fig. 6 shows the solid–liquid
phase diagrams at 298.15 K for H2 O–KCl–ampicillin ternary system under two different ethanol condi-
tions: one is without ethanol and the other is with 10 wt.% ethanol. From the calculation results, it can be
noted that the solubility of ampicillin in H2 O is decreased by adding ethanol, while, adding KCl increases
the solubility of ampicillin in H2 O.

3.2. Case study B

The phase diagram for solid–liquid equilibrium for the quaternary system involving three inorganic
salts has been computed to highlight the phase diagram generation aspects of the developed algorithm.
The problem description is given in Table 3.
796 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803

Fig. 5. Activity coefficient of KCl in H2 O at 298.15 K (case study A).

3.2.1. Steps T1 and T2 (see Table 1): specify the system and problem
In steps T1 and T2, the system and problem are specified. As shown in Table 3, the system involves
H2 O and three inorganic salts, which are Na2 CO3 , NaCl and Na2 SO4 . The problem is to generate the
solid–liquid phase equilibrium diagram for the given system.

3.2.2. Step T3 (see Table 1): creation of property model package

Table 4 shows the definition of the given system. This system involves 15 chemical species, which
are H2 O, six ionic species and eight inorganic salts. For all of these salts, complete dissociation can be

Fig. 6. Combined effect of ethanol and KCl on the solubility of ampicilin in H2 O at 298.15 K (case study A).
 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803 797

Fig. 7. Solid–liquid phase diagram for H2 O–Na2 CO3 –NaCl–Na2 SO4 quaternary system (case study B).

assumed. The hatched part in Table 4 shows the input components. Since this system is classified as
“aqueous” system and all of the salts are inorganic, the extended-UNIQUAC (aqueous) is selected for
computation of the phase diagram.

3.2.3. Final result

Finally, the solid–liquid phase diagram computed with the extended-UNIQUAC model (aqueous) is
shown in Fig. 7.

3.3. Case study C

This example highlights the solid–liquid phase diagram based flowsheet design algorithm (see Fig. 4)
in the methodology. In this example, two organic salts, which are salts A and B (the identities of these
salts cannot be given due to proprietary information), need to be recovered as products. The problem
description is shown in Table 3. The methodology steps shown in Fig. 1 are employed for this case
study.

3.3.1. Steps 1 and 2 (see Fig. 1): specify the system and problem
In steps 1 and 2, the system and problem are specified. As shown in Table 3, the system involves H2 O
and two organic salts, which are salts A and B. The problem is to design the thermodynamically feasible
flowsheet for the recovery of two organic salts.

3.3.2. Step 3 (see Fig. 1): creation of property model package

According to the methodology in the thermodynamic part (see Table 1), property model package needed
to simulate this process was created. This system includes nine chemical species involving six ionic
798 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803

species. The definition of the system is given in Table 3. Hatched part in Table 4 is input components.
Since this system is classified as an “aqueous 1” system, any model except the extended-UNIQUAC
(aqueous) model can be used for this problem. In this example, the electrolyte-NRTL is selected for the
calculation.

3.3.3. Steps 4 and 5 (see Fig. 1): generate the phase diagram and operation path
In this example, two flowsheets, one without recycle and one with recycle, are generated according to
the steps in Fig. 4. In this example, salt A is recovered first and a flowsheet without recycle is generated
as the first alternative.

3.3.4. Step 5.2 (see Fig. 4): specification of operation temperatures to recover salt A
The temperature of the fresh feed stream is 393.15 K. Therefore, operation temperatures of the crys-
tallizers must be below 393.15 K. In this example, 283.15 K is selected to recover salt A.

3.3.5. Step 5.3a ans b (see Fig. 4): computation of solid–liquid phase diagram at Ti and identification
of the feed location to recover salt A
In Fig. 8, the solid–liquid phase diagram is generated at the condition, where temperature is 283.15 K.
First, the composition of the feed stream to the crystallizer 1 to recover salt A, which is (0.4, 0.4, and
0.2), is located in the two phase region, where one liquid phase and salt A coexist, at 283.15 K. According
to rule 1 in step 5.3b [4], neither dilution nor evaporation is required. However, in order to obtain a higher
recovery rate, some of solvent was evaporated (see step 5.3b). The algorithm calculated the maximum
evaporation ratio in order to keep the shifted feed point in the same region as original feed point, which is
0.241. Note that the minimum evaporation ratio is zero. In this case, 0.241 was selected as the maximum

Fig. 8. Operation path for case study C.

 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803 799

evaporation ratio for recovery of salt A. After evaporation of solvent, composition of the shifted feed
stream was found to be (0.208, 0.528, and 0.264).

3.3.6. Step 5.4 (see Fig. 4): operation of crystallizer at Ti to recover salt A
First, the liquid stream, whose composition is (0.208, 0.528, and 0.264), is sent to the crystallizer 1 in
order to recover salt A. From mass balance, the composition of mother liquor found to be (0.124, 0.386,
and 0.490). This is now fed to crystallizer 2 for recovery of salt B and the same procedure (steps 5.2–5.4)
is repeated, yielding a continuous flowsheet for the separation process.

3.3.7. Step 5.5 (see Fig. 4): implementation of recycle stream

The above procedure generated a flowsheet without recycle for recovery of salts A and B. In this step,
addition of recycle streams is considered by mixing mother liquor from crystallizer 2 with the fresh feed
stream.

3.3.8. Step 5.6 (see Fig. 4): identification of maximum mixing ratio
In this step, the maximum value of mixing ratio [4] is calculated and was found to be 0.6786 and the
minimum value was found to be zero (that is, no recycle). In this example, 0.2 was selected as the mixing
ratio.

3.3.9. Step 5.7 (see Fig. 4): final design specification

When the mixing ratio is set to 0.2, the composition of the feed stream to crystallizer 1 is (0.3544,
0.3802, and 0.2653). In steps 5.3a and b, maximum evaporation ratio is 0.17 (it was 0.241 before recycle
was added). In this case, 0.17 was used. Therefore, the feed composition, which is (0.3544, 0.3802,
and 0.2653), is shifted to (0.427, 0.253, and 0.320). Finally, step 5.4 is repeated for both salt products
until the final design specifications are matched. The composition of mother liquor from both crystallizer
were the same as that before recycle was considered. This is because invariant points at the two selected
temperatures of operation were used as the basis for the design and since the temperatures for the flowsheet
alternatives were the same, the mother liquor compositions from the crystallizers were also the same.
In Figs. 8 and 9a, the operation paths and corresponding continuous process flowsheets for the recovery
of salts A and B are shown. In Fig. 9a, two process flowsheets: one without recycle and the other with
recycle are shown. In the flowsheet without recycle, fresh feed is sent to crystallizer C1 for recovery of salt
A after evaporation of H2 O in evaporator E1 . The mother liquor is next sent to crystallizer C2 for recovery
of salt B after evaporation of H2 O in evaporator E2 . In the flowsheet with recycle, feed stream is mixed
with recycle stream before evaporation of H2 O in evaporator E1. Salt A and then salt B are recovered in
crystallizers C1 and C2 , respectively. A part of the mother liquor from crystallizer C2 is recycled to mixer
M1.
The operation paths in Fig. 8 can also be used to generate the sequence of (two) batch operations needed
to recover the two products. In the first operation, salt A is recovered and in the second operation, salt B
is recovered. Adjustment of amount of solvent is needed to perform the batch operations. Note that the
flowsheet alternative and operational sequence shown in Fig. 8 is just one feasible alternative. In principle,
other alternatives can also be generated by, for example, choosing a different temperature specification or
product order. The final selection of the flowsheet will depend on the economic and operational evaluations
(not considered in this work). The steady-state simulation results obtained through the simulation engine
is given in Fig. 9b for the flowsheet with recycle.
800 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803

Fig. 9. (a) Generated flowsheets with and without recycle for case study C; (b) stream summary of the simulation results for the
flowsheet with recycle (case study C).

3.4. Case study D

This example involves the selective precipitation of three different products from an aqueous solution.
The problem description is given in Table 3. The results are given in terms of the main steps of the design
and analysis methodology (see Fig. 1). The results of step 1 (specify the system) and step 2 (specify the
identities of products) are included in Table 3.

3.4.1. Step 3 (see Fig. 1): creation of property model package

In step 3, a property model package has been created for the given problem and system. The property
model package is created according to the steps explained in the thermodynamic part. From the rule for the
classification of the problem [4], flowsheet design/simulation problem is classified into vapor–liquid–solid
equilibrium (VLSE) problem (step T1 in Table 1). In step T2 (see Table 1), the given system is defined
through information on the presence of solvents, salts and ionic species. This system includes 12 chemical
species including four cations, four anions and three dipole ions (see Table 4). Since only H2 O is the
solvent, this system is classified as “aqueous system”. The hatched part denotes the input information
and all the other information are generated automatically from the input information. The next step is the
selection of a property model package. In this example, the electrolyte-NRTL model was selected (step
3 in Table 1). Once the system and problem are defined and an appropriate activity coefficient model is
 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803 801

Fig. 10. Activity coefficient of serine and alanine in H2 O at 298.15 K (case study D).

selected, the next step is creation of the list of necessary physical properties and selection of the models
for each of the necessary properties (step T4 in Table 1, see Table 7).
After the models for necessary properties are selected, the next step is data regression, if necessary.
Ddata regression for the selected activity coefficient model (the electrolyte-NRTL model) is considered.
In the total 12 by 12 parameter matrix, none of the parameters were found in the open literature. However,
activity coefficient data for H2 O–alanine and H2 O–serine binary systems are available [15,16]. From the
sensitivity analysis, the interaction parameters between H2 O–alanine and H2 O–serine were identified as
sensitive parameters. Therefore, these interaction parameters were adjusted to reproduce the whole range
of experimental data. The final results of data regression are shown in Fig. 10. To calculate the solubility
of three products, the solubility product approach was taken for all of them. The model parameters for the
solubility calculation were adjusted using the solubility data. The solubilities of all these salts in H2 O are
available [13] and by adjusting the model parameters, the selected solubility model was able to describe
the whole range of experimental data [14].

3.4.2. Step 4 (see Fig. 1): solubility analysis

The solubility of l-asparagine, alanine and serine in H2 O does not change over a wide range of pH
[14]. To change the solubility of these products, pH of solution has to be set at below three or above
nine. This behavior is due to two molecular dissociations: one of which is dominant in the strong acid
side while the other is dominant in the strong base side. On the other hand, the solubility is sensitive
to the temperature change. Therefore, it is preferable to separate the three products using temperature
dependency of solubility of salts.

3.4.3. Steps 5 and 6 (see Fig. 1): generate the phase diagram and operation path and validate the
generated flowsheet
In Fig. 3a, the Janecke projection diagram for H2 O–alanine–serine–l-asparagine quaternary system
is shown for three different conditions: temperature = 313.15, 343.15 and 373.15 K. Original feed
802 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803

composition, whose composition is represented by F, is found to be located in the region where the liquid
phase and solid alanine coexist at 373.15 K. alanine is crystallized at the condition where temperature
is 373.15 K. The mother liquor from the first crystallizer is sent to the second crystallizer to recover
l-asparagine at 313.15 K. The first and third crystallizers are operated at low pressure for evaporation of
H2 O. Finally, serine is recovered at 343.15 K and the mother liquor is recycled to mix with fresh feed,
F. In Fig. 3b, the corresponding flowsheet is shown. With the initial estimates of the stream variables
obtained from Fig. 3a, the rigorous simulation converged to the solution [14] very rapidly. Note that
the generated flowsheet may not be optimal but it is thermodynamically feasible. Also, it is based on
consistent and correct thermodynamics and implicitly takes into account the ease of separations through
solubility analysis and, as a result, energy costs.

4. Conclusion

The main features of the developed methodology are, use of correct/consistent property models, gen-
eration of thermodynamically feasible process/operation alternatives and their simultaneous validation
through rigorous simulation models. These features combined together, provides a unique visualization
that provides simultaneously, the problem solution as well as the operational paths for the process flow-
sheet or the batch operation sequence. This visualization helps to understand the feasibility of a desired
operation, to identify the operational limits and to generate the necessary knowledge (insights) for the
synthesis/design algorithms. The main features of the developed methodology also needed the develop-
ment of a number of additional methods and tools. For example, sensitivity analysis based property model
parameter estimation so that the necessary property model can be created when only limited experimental
data is available. Since visualization is based on solubility diagrams, computation algorithms for various
types of phase diagrams have been developed taking into consideration, their use in the design/analysis
part. Also, availability of a rigorous multi-phase separation module, developed earlier [5] to model and
simulate various types of separation operations (crystallization) involving vapor, liquid, solid and gases
helps to verify the generated flowsheet alternatives in an efficient and consistent manner. Finally, the
methods and tools have been integrated into a computer aided system so that the transfer of knowledge
(data) between any two methods/tools is smooth, noise free and efficient, thereby, providing an efficient,
visual and flexible solution of these types of problems.

List of symbols
a activity
f fugacity
F objective function value
K equilibrium constant
NC number of components
NE number of experimental data
NM number of molecules
NR number of reactions
p physical property value
P pressure
T temperature
 K. Takano et al. / Fluid Phase Equilibria 194–197 (2002) 783–803 803

x composition
z number of chemical charges

Greek letters
γ activity coefficient
ν stoichiometric coefficient
τ model parameter

Subscripts
cal calculated value
exp experimental value
liq liquid phase
s saturation
solid solid phase

References

[1] B. Fitch, How to design fractional crystallization processes? Ind. Eng. Chem. 62 (12) (1970) 6.
[2] S.R. Dye, M. Ng Ka, Fractional crystallization: design alternatives and tradeoffs, AIChE J. 41 (11) (1995) 2427–2438.
[3] K.D. Samant, M. Ng Ka, Representation of high-dimensional solid–liquid phase diagrams of ionic systems, AIChE J. 47 (4)
(2001) 861–879.
[4] K. Takano, Computer-aided design and analysis of separation processes with electrolyte systems, Ph.D. Thesis, Technical
University of Denmark, Lyngby, Denmark, 2000.
[5] K. Takano, R. Gani, T. Ishikawa, P. Kolar, Trans. IchemE 78 (Part A) (2000) 763.
[6] K.S. Pitzer, Thermodynamics of aqueous electrolytes at various temperatures, pressures and compositions. Thermodynamics
of aqueous systems with industrial applications. In: B.S.S. Newton (Ed.), American Chemical Society Symposium Series,
Washington, USA, 133 (1980) 451–456.
[7] C.C. Chen, L.B. Evans, A local composition model for excess gibbs energy of aqueous electrolyte systems, AIChE J. 32
(1986) 1655.
[8] H. Nicolaisen, P. Rasmussen, J.M. Sorensen, Correlation and prediction of mineral solubilities in the reciprocal salt system
(Na+ , K+ )(Cl− , SO4 2− )—H2 O at 0−100 ◦ C, Chem. Eng. Sci. 48 (18) (1993) 3149–3158.
[9] W.C. Blasdale, Equilibrium in Saturated Salt Solutions, American Chemical Society Monograph Series, Washington, USA,
31 (1927) 79.
[10] R. Gani, G. Hytoft, C. Jaksland, A.K. Jensen, Comput. Chem. Eng. 21 (1997) 1135.
[11] Z. Yizu, B.L. Evans, Representation of phase equilibrium behavior of antibiotics, Biotechnol. Prog. 6 (1990) 266–272.
[12] J.P. Hou, J.W. Poole, J. Pharm. Sci. 58 (1969) 1510–1515.
[13] J.B. Dalton, C.L. Schmidt, The solubilities of certain amino acids in water. The densities of their solutions at twenty five
degrees, and the calculated heats of solution and partial volumes, J. Biol. Chem. 103 (1933) 549–578.
[14] K. Takano, R. Gani, T. Ishikawa, P. Kolar, Comput. Chem. Eng. 24 (2000) 645–651.
[15] Hutchens et al., J. Biol. Chem. 238 (1963) 1419.
[16] Smith et al., J. Biol. Chem. 121 (1937) 607.