Solids Handling

Streamlining
Crystallization
Process Design
Use phase diagrams to develop
reliable crystallization processes
quickly and cost-effectively.

Christianto Wibowo
Lionel OYoung
ClearWaterBay Technology, Inc.
Ka M. Ng
Hong Kong University of
Science and Technology

ECENT ADVANCES IN CHEMISTRY AND

biochemistry have created new challenges for
the chemical processing industries regarding the
production of high-molecular-weight chemicals, which
are normally recovered as solids via crystallization.
Consequently, the focus of R&D must deviate from the
familiar vapor/liquid system into a newer territory
the solid/liquid system. This is in light of the numerous problems often encountered in the development of
crystallization processes.
Just as vapor/liquid equilibrium (VLE) plays a key
role in the design of distillation processes, solid/liquid
equilibrium (SLE) forms the thermodynamic basis for
the design and synthesis of crystallization processes.
In both cases, the use of phase diagrams to visualize
regions in composition space where the system exists
as a single phase or a mixture of multiple phases allows better understanding of the thermodynamic limitations imposed by the phase behavior. However, the
differences between the typical phase behavior of
solid/liquid systems and vapor/liquid systems lead to
different roles of the phase diagrams in the two cases.
In distillation-based separations, the liquid and
vapor phases typically contain all the components that
are present in the feed, although in different compositions. Most of the time, high-purity products can only
be obtained via a multi-stage operation, for which the

30

www.cepmagazine.org

January 2004

CEP

McCabe-Thiele diagram or residue curve map are used,

especially to identify feasible product compositions.
Solid/liquid equilibrium, on the other hand, often allows the recovery of pure solids in a single-stage operation. By combining SLE with other driving forces (e.g.,
VLE) in operations such as cooling, heating, evaporation
and solvent addition, one can devise separation schemes
for obtaining one or more solid products in a pure form.
These operations can be represented as movements on the
phase diagram. Many different schemes can be generated for a given system, depending on which movements
are selected. Drawing on insights from phase behavior, a
combination of feasible movements can be selected to
construct flowsheet alternatives. Once all such combinations are identified, one should come up with a handful of
feasible separation process alternatives that can be further
evaluated. While generating the solid/liquid phase diagram, one should focus on the key characteristics that are
controlling the feasibility of the separation, so that the
number of required experiments can be minimized, leading to a shorter development time.
This article discusses a strategy for rapid development of reliable crystallization processes. The methodology combines the application of solid/liquid phase
equilibrium principles with the execution of a minimum number of experiments in a workflow for process
development. The features of SLE phase diagrams are

reviewed, along with the basic movements used to design

a crystallization-based separation process. A framework
that integrates modeling and experimental efforts to calculate and validate SLE phase diagrams is also presented.

Basic types of solid/liquid phase diagrams

The phase behavior of a system can be represented using
phase diagrams in the temperature-pressure-mole fraction (or
weight fraction) space, which is more conveniently referred
to as composition space. Since pressure does not have a significant effect on solid/liquid phase behavior, it is generally
sufficient to consider isobaric SLE phase diagrams. Figure 1
shows the most common types of binary SLE phase diagrams. Over 50% of binary systems of organic petrochemicals exhibit simple eutectic behavior (Figure 1a), about 25%
exhibit compound formation (Figure 1b) and approximately
7% exhibit peritectic behavior (Figure 1c) (1). Other types of
phase behavior, such as solid solution (i.e., the solid phase is
not a pure component) or partial immiscibility, are much less
common. However, the situation is different for pharmaceuticals, many of which are chiral compounds. In this vein, the
formation of racemic or enantiomeric compounds is the most
common type of binary phase behavior (2).
The simple eutectic phase diagram (Figure 1a) features
two solubility curves, one each for compounds A and B. The
two curves meet at the binary eutectic (point E), where both
A and B are saturated. Cooling a mixture with this composition causes simultaneous crystallization of A and B; and the
solution composition remains constant until all liquid is
gone. In other words, no separation occurs. In this sense, the
eutectic is the SLE analog of an azeotrope in VLE.
The phase diagram of a compound-forming system
(Figure 1b) resembles two simple eutectic phase diagrams
placed side-by-side. In addition to the solubility curves for
A and B, there is a solubility curve for compound AB,
which has a peak corresponding to the congruent melting
point of the compound, Tmc. At this temperature, the compound coexists with a liquid of the same composition.
A compound is also formed in a peritectic system
(Figure 1c), but since it decomposes above its incongruent
melting point, Tmi, it cannot coexist with a liquid of the
same composition. This phase behavior is typically

Nomenclature
a, b
A, B
AD, BD
ABD
D
E, E1, E2
F
H
GE
Gf
Hf
i, j and k
KSP
L
m
M
P
q
R
(R), (S)
(s)
Q
S
T
Tc
Tm
Tmc
Tmi
x
z

= number of cations and anions, respectively,

in an electrolyte mixture, Eqs. 34
= chemical components
= binary eutectic, Fig. 2
= ternary eutectic, Fig. 2
= solvent
= eutectic points, Fig. 1
= molar flowrate of stream, mol/h, Eq. 1
= represents a hydrogen atom and is part of
components name, e.g., (S)AH.(R)B
= excess Gibbs free energy, J/mol
= Gibbs free energy of formation, J/mol
= heat of fusion, J/mol, Eq. 6
= indices, Eqs. 35
= solubility product, dimensionless
= point in binary phase diagram, where A is
saturated in the liquid phase, Fig 1 and Eq. 1
= liquid molality, mol solute/kg solvent
= point in binary phase diagram, representing a
two-phase solution, Fig 1 and Eq. 1
= a weak electrolyte
= number of non-dissociating components
in an electrolyte mixture, Eq. 5
= universal gas constant = 8.314 J/mol-K
= enantiomers, Fig. 6
= solid phase, Fig. 1
= non-dissociating components, Eq. 5
= point in binary phase diagram
representing solid A, Fig 1 and Eq. 1
= temperature, C or K
= crystallizer temperature,C or K
= melting point, C or K
= congruent melting point of a mixture, C or K
= incongruent melting point of a mixture, C or K
= mole fraction of component in mixture, dimensionless
= absolute value of the electric charge on an ion, dimensionless

Greek letters
ik
= number of ion i in a molecule of weak electrolyte k,
dimensionless
ik
= number of ion i in a molecule of complex ion k, dimensionless
i
= activity coefficient of component i, dimensionless
i
= chemical potential of component i at unit molality, dimensionless
= stoichiometric coefficient of ion i in
i
dissociation reaction, dimensionless
Superscripts
*
= asymmetric standard state
s
= solid phase
sat
= saturated component, Eq. 6
The

meanings of subscripts are redundant of symbols in general nomenclature.

observed when the melting points

of the components forming the
compound are very different.
The information represented by
the solubility curve of A in Figure
1a is the same as what is routinely
plotted in a solubility diagram
i.e., the solubility of A in B (in either weight fraction or mole fraction) as a function of tempera-

A(s) +

Figure 1. Three most common types of binary system phase diagrams: (a) simple eutectic,
(b) compound formation, and (c) peritectic.
CEP

January 2004

www.cepmagazine.org

31

Solids Handling

ture, T (in either C or K). The phase diagram is more

complete, since it covers the entire range of temperatures
between the pure component melting point and the eutectic temperature. In addition, it shows the solubility curve
of B in A. The region below the solubility curves represents a two-phase region, where phase split occurs. For example, point M splits to give crystals of A (point S) and a
saturated solution (point L). The relative amount of solid
and liquid can be determined using the lever rule:
FS/FL = ML/SM = (xL xM)/xM
(1)
implying that
FS/FM = (xL xM)/xL

(2)

This expression gives the amount of crystallized solid relative to the amount of the original solution, which is closely
related to the yield or recovery in a crystallization process.

Applications in crystallization
process development
A crystallization-related operation, such as heating,
cooling, stream combination, splitting, solvent addition or
solvent removal, can be represented on top of the phase diagram as a movement in composition space i.e., a transition from the feed composition point(s) to the product
composition point(s). Such a representation allows one to
quickly identify thermodynamic limitations and select a
set of movements to create a feasible process.
Unfortunately, real systems rarely contain just one solute
and one solvent, rendering binary phase diagrams a somewhat inadequate way to represent crystallization systems.
However, they are useful as building blocks to construct
phase diagrams for systems with more than two components. The features of multicomponent phase diagrams are
discussed below, along with examples illustrating their applications in the development of crystallization processes.
Example 1. Maximum recovery of compound A in
cooling crystallization. Consider a natural product extract
containing an active ingredient A, a byproduct B and a
solvent D. Laboratory experiments indicate no compound
formation among the components, so simple eutectic behavior is assumed. The phase diagram of this ternary system takes the shape of a three-dimensional triangular
prism (Figure 2a), the three rectangular faces of which
represent the three binary pairs. In the interior of the
prism lie three solubility surfaces, bounded by eutectic
troughs connecting the binary eutectics to a ternary eutectic, where all three components are saturated. To allow
convenient viewing in two dimensions, the diagram can
be projected parallel to the temperature axis onto the
base. This is called a polythermal projection (Figure 2b).
The surfaces appear as compartments that indicate the
composition range in which a pure component (A) can be
obtained via crystallization.

32

www.cepmagazine.org

January 2004

CEP

Suppose that a pure product is to be obtained via crystallization from a feed at temperature T1 (point 1 in Figure 2a)
containing about 61% A. For economic reasons, the maximum recovery of A is sought. Cooling of the solution from
point 1 is represented by a vertical movement down the temperature axis. When the axis hits the solubility surface of A
at point 1a (represented by the same point in the projection),
the solution becomes saturated in A (i.e., A becomes saturated). Upon further cooling, A supersaturates the solution,
which causes A to crystallize (Figure 2c). The liquid composition slides along the solubility surface of A as a result of
both decreasing temperature and loss of A. In the polythermal projection, this is a movement away from apex A.
If the crystallizer is operated at temperature T2, the
mother liquor composition is given by point 2, which contains about 34% A. From the lever rule (Eq. 1) or via material-balance calculations, it can be found that if 1 kg of
feed (containing 61% A) gives x kg of solid A, then 0.61 =
x + 0.34(1 x) or x = 0.41. In other words, 0.41 kg of
solid is recovered per kg of feed.

Figure 2. (a) Ternary phase diagram for example 1. (b) Polythermal projection of crystallizer. (c) Process synthesis for recovery of pure A in a
cooling crystallizer.

Considering that 1 kg of feed contains 0.61 kg of A, the

recovery of A at this point is about 67%. Upon further
cooling, the composition moves further away from A. In
other words, operation at a lower temperature would increase the recovery of A. Operation at T3 gives a composition at the eutectic trough (point 3), which corresponds to
about 78% recovery of A. This is the maximum recovery
of A, because cooling beyond point 3 causes B to crystallize with A. For example, cooling to temperature T4 causes
the composition to follow the eutectic trough to point 4.
Example 1 illustrates that the maximum possible recovery of the desired product via cooling crystallization
can be found by locating the compartment boundaries. It
is therefore desirable to look at the polythermal projection, which clearly displays the boundaries. From a design perspective, there are two important points concerning the phase diagram. If crystallization of pure A is desired, then (1) the feed point has to lie in compartment A,
and (2) the operating temperature must be above that at
which the compartment boundary and a straight line connecting A and the feed point intersect.
Example 2. Maximum recovery of compound A in
evaporative crystallization. An alternative method of generating supersaturation in a crystallization process is to
evaporate solvent from the solution. Consider another
feed containing the same three components (A, B and D)
at a different composition, as indicated by point 1 in Figure 3a. An evaporative crystallizer operating at Tc is used
to separate pure A. Removal of solvent D in the crystallizer is represented in the phase diagram as a movement
away from D. After reaching the solubility surface of A at
point 2, A becomes supersaturated and the composition
delves into the two-phase region. Simultaneous removal
of D and crystallization of A results in a movement along
the saturation curve as indicated. The endpoint is determined by how much D is removed from the solution. The
maximum recovery of pure A using evaporative crystallization at Tc is achieved when point 3 is reached, because, at that point, B also becomes saturated. Note that at
any operating temperature, the amount of evaporated solvent can always be adjusted such that the maximum recovery of A is achieved. The phase diagram can be used
to identify the location of the double saturation point at
that temperature (i.e., when A and B are both saturated).
Since the entire process occurs at Tc, an isothermal cut at
this temperature (Figure 3b) is sufficient for the representation. In the cut, solubility surfaces appear as single saturation curves. The curves meet at the double saturation point,
which is actually a point on the eutectic trough. The yellow
area signifies the two-phase region, where a mixture would
phase-split to give solid A (Figure 3c) and a saturated solution along the single saturation curve A-3. Triangle A-B-3 is
a three-phase region; a mixture with a composition lying
within this area would phase-split to form solid A, solid B
and a double-saturated solution represented by point 3.

Example 3. Development of a process for complete

separation of two solids. The usefulness of phase diagrams is more evident in the synthesis of crystallizationbased separation processes. For example, a possible strategy for maximizing the recovery of A from the previously
discussed natural product extract is to also crystallize B
out and recycle the mother liquor. Figure 4 illustrates the
process on the polythermal projection of the phase diagram. Starting with a feed at point 1, a recycle stream is
added to give point 2. Note that such stream combination
results in a point on the tie-line connecting the two feed
points. The location of point 2 is determined by the relative amount of the two feed streams. Solvent is then
added to obtain point 3, which still lies in compartment A.
The solution is then cooled to crystallize pure A, until the
composition reaches point 4 lying close to the A/B bound-

Figure 3. (a) Ternary phase diagram for example 2. (b) Isothermal cut of
phase diagram. (c) Process synthesis for recovery of pure A in an
evaporative crystallizer.

CEP

January 2004

www.cepmagazine.org

33

Solids Handling

Figure 4. (a) Ternary phase diagram for example 3. (b) Synthesis of a

process for separating components A and B using a polythermal
phase diagram.

rated are visited. It can be seen in the polythermal projection

that solvent removal causes point 4 in compartment A to
move across the compartment boundary to give point 5 in
compartment B. Stream combination is used to move point 6
back into compartment A. By knowing the phase diagram,
particularly the location of compartment boundaries that need
to be crossed, feasible separation process alternatives can be
generated. It is also apparent that the curvature of the AB
compartment boundary determines the feasibility of using
solvent removal for boundary crossing. In other words, it is
important to accurately determine this boundary to make sure
that the process is feasible.
The selection of the most suitable crystallization solvent is
an important design issue, because it affects both separation
feasibility and product quality, such as impurity inclusion and
particle-size distribution. From a thermodynamics point of
view, a good solvent must provide a favorable phase behavior
for separation. Since the location and curvature of compartment boundaries determine the feasibility of separation, phase
diagrams can be used as a rough criterion for solvent screening. As an illustration, assume that there are three solvents D1,
D2 and D3, that can be used to extract the natural product discussed earlier. Since the extract always contains both A and
B, it is desirable to choose the best solvent for complete separation of these two components.
Figure 5 shows the polythermal phase diagram of the
ternary systems formed with D1, D2 and D3. Although the
location of the binary eutectic AB is unaffected by the solvent, the orientation of the AB eutectic trough is different
for each solvent. As mentioned previously, the curved
shape of this trough allows the crossing from compartment
A to compartment B using solvent removal (Figure 5a).
When D2 is used instead of D1, the trough becomes more
curved (Figure 5b) and it is possible to obtain point 5 well
inside compartment B. Using the lever rule to compare
these two cases, it is obvious that the maximum yield of B
is higher for the system with D2, because points 5 and 6
are farther apart. The phase diagram shown in Figure 5c
suggests that D3 is not a suitable solvent, because solvent
removal or addition cannot cause compartment crossing.
Example 4. Chiral separation using crystallization. Up

ary. Further cooling or removal of solvent would cause A

and B to precipitate together. However, heating of this
mixture, followed by evaporation of solvent D, gives
point 5 on the solubility surface of B. A second crystallizer can then be used to crystallize out pure B until point 6
is reached. This point is then recycled to the feed and the
cycle repeats itself. With such a recycle scheme, a theoretical recovery of 100% can be achieved for both A and
B. Of course, in practice the recovery is lower due to the
presence of a purge stream for discharging any impurity
present in the feed.
Example 3 illustrates
the construction of a
separation process by
combining basic movements: cooling, heating,
solvent addition and solvent removal. The
movements are carefully
picked to allow crossings across compartment boundaries, such
that all compartments of Figure 5. Comparison of phase diagrams for example 3 is a quick way of screening for a suitable solvent for separating
the products to be sepa- A and B.
34

CEP

January 2004

www.cepmagazine.org

to this point, the discussion has focused on molecular systems with no reaction involved. A different set of coordinates
is needed for electrolyte systems, because electrolytes partially or completely dissociate in water to give ions, which
can rearrange themselves to form another electrolyte. For example, consider a racemic mixture of a chiral acid compound: (R)AH and (S)AH. A racemic mixture of an amine,
(R)B and (S)B, is used as a resolving agent to form a mixture
of four diastereomeric salts: (R)AH.(R)B, (R)AH.(S)B,
(S)AH.(R)B, and (S)AH.(S)B. Since all of them dissociate to
give ions, specifying the concentrations of molecular species
would not provide an accurate representation of the solutions composition. For instance, an equimolar mixture of
(R)AH.(R)B and (S)AH.(S)B in solution is indistinguishable
from an equimolar mixture of (R)AH.(S)B and (S)AH.(R)B.
For an electrolyte mixture containing a simple cations, b
simple anions and q non-dissociating components (and
maybe some complex ions and weak electrolytes), it is convenient to use (a 1) cationic coordinates, (b 1) anionic
coordinates and q inert coordinates as follows:

z +
Mi M i

z
z M Mi

=
z
zM M j
j

Mi

Mj

] +
+

Mi , k

[ Pk ] + k M , k [Ck ]

+ M , k [ Pk ] + M , k [Ck ]
k
k

(3)

where i = 1,2,, a 1.

z
Ni Ni

z N Ni
+ N , k [ Pk ] + N , k [Ck ]
k
k

=
z

+ N , k [ Pk ] + N , k [Ck ]
zN N j
j
k
k

Ni

Nj

(4)

where i = 1,2,, b 1.

R(Qi ) =

zM
j

z +
M j Nj

[Qi ]

+ M , k [ Pk ] + M , k [Ck ]
k
k

(5)

where i = 1,2,, q.
In these equations, M+ and N denote a cation and an anion,
respectively; P is a weak electrolyte; C is a complex ion; Q is
an inert solvent, such as water; z is the absolute value of an
ionic charge (e.g., z = 1 for H+ and z = 2 for SO42); ik is the
number of ion i in a molecule of weak electrolyte k. Similarly,
ik is the number of ion i in a molecule of complex ion k. The
concentrations are expressed in molality (mole of solute/kg
solvent). This choice is consistent with the set of coordinates
that has been conventionally used for strong electrolyte sys-

tems. Note that because of the electroneutrality constraint, the

denominators in Eqs. 3 and 4 are the same. Both can be used
interchangeably as the denominator in Eq.5.
The system of diastereomeric salts discussed previously
(example 4) contains two simple cations ((R)BH+and
(S)BH+), two simple anions ((R)A and (S)A) and an inert
solvent. Therefore, the isobaric phase diagram involves
cationic, anionic and inert coordinates, plus a temperature
coordinate. Because it is impossible to view a four-dimensional diagram on a page, projections and cuts must be used.
An isothermal cut of this phase diagram is shown in Figure
6a. There are four solubility surfaces, corresponding to the
four salts. Similar to a two-dimensional isothermal cut, one
can identify two-, three- and four-phase regions where one
or more salts coexist with a saturated liquid. To maintain
clarity, these regions are not shown in Figure 6. For easier
visualization, it is convenient to take a Jnecke projection
from the water apex, as shown in the Figure 6. Points on this
projection can be obtained by renormalizing the composition
to a solvent-free basis. Note that all points in the Jnecke
projection of an isothermal plot have the same temperature
they represent saturated solutions with different solvent
concentrations.
Phase diagrams for electrolyte systems can also be used to
generate separation process alternatives. Figure 6b shows the
Jnecke projection of two isothermal phase diagrams of the
diastereomeric salt system, taken at T1 and T2. With equimolar feed, a mixture of diastereomeric salts indicated by point 1
is formed. This feed is mixed with a recycle stream to give
point 2, which lies on the saturation region of (S)AH.(R)B at
T1. After crystallizing this salt in an evaporative crystallizer,
water is added in a dissolver to give point 3, and the mixture
is cooled down to T2. At this temperature, the double saturation curves have shifted in such a way that point 4 is now on
the saturation region of (R)AH.(S)B at T2. Crystallization of
this salt gives point 5, which is recycled to allow the cycle to
repeat itself. The two salts can then be sent to a recovery system, such as by adding a strong acid, to decompose them and
obtain the desired products. This example illustrates the use
of solvent addition to cross a compartment boundary. Clearly,
the saturation regions of the salts (defined by the double saturation curves) need to be known in order to design a resolution process based on such a movement. For this process, the
shift of the (S)AH.(R)B/(R)AH.(S)B double-saturation curve is
the key to the process feasibility. The location of the other
double-saturation curves does not affect the process. This example also demonstrates that one can synthesize a process
with the aid of isothermal cuts at different temperatures, instead of using a polythermal projection.
In summary, it has been shown that the use of phase diagrams to represent the phase behavior of various systems,
ranging from molecular to electrolyte systems, allows the
identification of regions or ranges of compositions where a
pure solid or a mixture of solids can be obtained using crystallization. Such identification provides an ability to determine

www.cepmagazine.org

January 2004

CEP

35

Solids Handling

SLE data for selected binary pairs, mostly petrochemicals, are

available in the DECHEMA handbook (3); there are also
some binary- and ternary-phase diagrams reported in journal
articles. But, in general, the user would have to generate the
phase diagram. During the conceptual design stage, in the interest of saving development time and effort, it is preferable to
first estimate the phase diagram using thermodynamic information, such as melting point (Tm), heat of fusion (Hf,) or
Gibbs free energy of formation (Gf). Laboratory experiments
can then be performed to validate selected portions of the
phase diagram.
A system reaches SLE when the chemical potentials for all
components in the liquid and solid phases are equal. The
chemical potential can then be related to the solute concentration as well as to thermodynamic entities, such as Tm and Hf.
Starting from this equality, relevant equations for predicting
SLE can be derived.
For molecular systems, it is convenient to relate chemical
potential to mole fraction and apply a symmetric standard
state convention for the activity coefficient in the liquid phase,
that is, = 1 for x = 1 (pure component). The resulting expression is the solubility equation, which gives the mole fraction
of component i in a solution saturated with that component:
xisat =

H f , i 1
is xis
1

exp

i ( x, T )
R Tm, i T

(6)

where is activity coefficient, x is mole fraction, Hf is heat of

fusion, T is temperature, and Tm is melting point. The superscript s indicates solid phase. R is the universal gas constant
(8.314 J/molK), If the solids are completely immiscible,
Figure 6. (a) Shown is an isothermal cut of the phase diagram involving four which is true for most systems, then ixis= 1. The values of acdiastereomeric salts in aqueous solution. (b) The Jnecke projection is shown tivity coefficients are obtained from excess Gibbs free energy
on the base of the pyramid. (c) Synthesis of a process for separating a chiral
GE, which can be calculated using various thermodynamic
electrolyte mixture using isothermal cuts at two different temperatures.
models (4).
separation limits and synthesize flowsheet alternatives for
A polythermal projection can be obtained by calculating
crystallization-based separations. Depending on the applicaeutectic points and troughs. A eutectic can be found by applytion, it may be preferable to view a polythermal or isothermal
ing Eq. 6 to all components in the mixture, because of all of
phase diagram.
them are saturated at this point. For example, a binary eutectic
However, most real systems involve more components than
consisting of A and B can be found by solving:
the systems discussed above, resulting in multi-dimensional
phase diagrams. The best way to handle these complex repre H f , A 1
1
1
sentations is to use multiple projections and cuts. Depending
xA + xB =
exp
+

A
R Tm, A T
on the application, a particular type of projection or cut may
be preferable. For example, cuts taken by fixing the solvent
H f , B 1
1
1
composition are useful in the design of a crystallization pro = 1
exp
(7)

cess using a mixed solvent because it allows the solvent mixB

R Tm, B T
ture to be represented as a single component, thus reducing
the dimensionality of the phase diagram.
Along the eutectic trough, all components but one component are saturated. Thus, the trough can be calculated by varyCalculation of SLE phase diagrams
ing the mole fraction of the unsaturated component and applyOne should have the SLE phase diagram for a system in
ing Eq. 6 for the saturated components.
hand before trying to develop a crystallization process. UnforCalculation of SLE phase diagrams for electrolytes differs
tunately, the suitable solid/liquid phase diagram for the particfrom that for non-electrolytes for two reasons. First, the
ular application is seldom available in the literature. Some
molecules dissociate in solution to form ions, the effect of

36

www.cepmagazine.org

January 2004

CEP

which has to be taken into account. Ionic interactions are the

dominant source of non-ideality in electrolyte solutions. Second, infinite dilution is normally taken as the standard state for
electrolyte activity coefficients, that is, 1 as x 0 (asymmetric standard state). For an electrolyte P that dissociates to
form ions according to the reaction
P = MMz

+NNz

the chemical potential in the solid phase should be equal to

the sum of the chemical potentials of the ions in the liquid
phase, leading to:

) (

1 N
*
M mM

1 M
*
N mN

= K SP (T )

(8)

where m is molality in mol solute/kg solvent and the asterisk

indicates asymmetric standard-state activity coefficients. The
use of this expression is convenient for electrolyte systems,
because it eliminates commonly encountered problems such
as unavailability of melting point data because the component
dissociates before melting. The solubility product, KSP, is related to the chemical potential at unit molality, *, of the ions
and the solid electrolyte,
1 *
P* ( s )
*
M
K SP (T ) = exp
+ N

RT M N M N

(9)

Various models are available for predicting the activity coefficients for electrolyte systems. Just as for molecular systems,
the activity coefficients can also be predicted from excess
Gibbs free energy models (46).

Experimental validation of phase diagrams

The two most common experimental methods for determining SLE are the synthetic or stationary method and differential thermal analysis (DTA) or differential scanning
calorimetry (DSC). The conventional synthetic method which
relies on visual observation of the melting process, has several
variants. One of them is step-warming, during which a solid
sample is slowly heated and the temperature at which the solid
phase disappears is recorded (7, 8). Step-warming can be used
to generate a polythermal phase diagram by repeating the experiment using different compositions of the solid sample. In
the solvent-addition method, temperature is kept constant
while a solvent is added to a solid mixture until the solid disappears (9). In a similar method, an excess amount of solid
mixture is partially dissolved in a solvent at a constant temperature, and the liquid phase concentration is measured (10).
The last two techniques are suitable to determine an isothermal phase diagram.
The DSC method has been widely used in the experimental studies on metal, polymer and organic compounds. It has
the advantage over the synthetic method because it is capable

of quickly determining simultaneously the liquidus and

solidus points of a mixture. Consequently, the DSC results can
be used to determine whether a binary pair forms a eutectic
system or solid solution (11). In a differential scanning
calorimeter, a small amount of solid sample and a reference
material are simultaneously heated or cooled at a fixed rate.
The temperature difference between the sample and the reference, and the amount of heat added to or removed from the
sample are recorded in the DSC curve. Analysis of the location and area of the peaks in theSC curve allows the determination of the Tm, eutectic point and Hf. Only polythermal
phase diagrams (as opposed to isothermal phase diagrams)can
be obtained from a DSC analysis. The preparation of samples
in small amounts can be a challenge in using DSC.

Crystallization process development

Properly executed, the combination of modeling and experimental effort can lead to the development of promising process alternatives in less time and at lower cost. The workflow
for crystallization process development consists of an iterative
procedure with five major steps: (1) estimation of phase behavior; (2) process synthesis; (3) identification of the need for
validation; (4) collection of experimental data; and (5) assessment of estimation accuracy. The level of complexity in each
iteration is matched with the availability of information. A
simple model should be used to model the phase behavior in
the beginning of the development process when little information is available on the process, while a rigorous one is reserved for a later stage when more experimental data have
been generated.
To estimate the phase behavior, one should first collect the
necessary thermodynamic information. For molecular systems, these include melting point and heat of fusion, which
are available in many handbooks and databases. For electrolyte systems, the key piece of information is the solubility
product (KSP), whose values at different temperatures can be
obtained from various sources (1214). Alternatively, KSP can
be calculated from the standard-state Gibbs free energy of formation, the values of which can be found in various thermodynamic tables (15). Information on thermodynamic properties is usually available for common solvents and salts, but can
be hard or even impossible to find for large molecules such as
drugs and natural products. In such cases, efforts should be
made to obtain a reasonable estimate. Methods for estimating
these properties based on group contributions have been developed (16, 17).
The thermodynamic information, along with an appropriate
activity coefficient model, is used to generate the phase diagram. A limited number of experimental data, such as the solubility of pure components at one or several temperatures,
may be available or easy to obtain. These data may be used to
back out the binary interaction parameters in the activity coefficient model, thus enhancing the prediction. If more data are
available, a model with more parameters can be used. As an
example, consider the phase diagram for synthesizing the sepCEP

January 2004

www.cepmagazine.org

37

Solids Handling

a few data points and trying to find the

location where a change of slope or discontinuity occurs. This method is generally effective for simple-eutectic or compound-forming systems, but may be difficult to implement for a peritectic system,
+
+
in which the change of slope may not be
obvious.
The second method is to determine the solid identity for a few data
points around which the double saturation point is suspected to lie. Since all solutions with compositions to the right of
+
the double saturation point must be saturated with A while those to the left must
be saturated with B, the point at which
the solid identity changes is the double
saturation point. Once the double saturation point has been located, it can be
drawn on the polythermal projection. By
repeating this procedure for several values of xD, the compartment boundary can
be located. Note that after obtaining one
or two points, one should have a good
idea of the approximate location of the
Figure 7. Incorporating experimental data in the experimental validation of phase diagrams.
double saturation point at the next value
aration process in example 3. Experimental data on the soluof xD, so that less data would be required to pinpoint its exact
bility of B in D (B/D) at temperature T1 is indicated in Figure
location.
7. By adjusting model parameters to fit this data point, a better
For high-melting compounds with
estimate of the B/D binary phase diagram can be obtained.
low solubility, a more practical approach is to measure the
Similarly, three data points on solubility of A in D (A/D; red
double saturation point directly by preparing a mixture of A
squares) lead to a better fit of the A/D binary parameters.
and B, then adding a small amount of solvent to dissolve some
These parameters can then be used in the calculation of the
solids. As long as both A and B exist in the solid phase, the
ternary phase diagram.
liquid composition always corresponds to the double saturaUsing the estimated phase diagram, one can proceed with
tion composition. Since the main objective in the first iteration
process synthesis, as discussed in example 3. As pointed out
is mainly to verify the process feasibility, a few data points
previously, the most important feature of the phase diagram is
taken over a wide range of xD may be sufficient. After all, the
the location and curvature of the boundary between compartresult may, contrary to the original estimate, indicate that the
ments A and B. The other two eutectic troughs are not imporprocess is not feasible, in which case the use of other solvents
tant because they do not have any effect on the process. Theremust be considered.
fore, it is very important to accurately determine the boundary,
Upon positive identification that the phase diagram supespecially in the range of mole fractions of D, xD, between 0
ports a feasible process, one can proceed to determine the
and 0.5. Once the boundary is verified, it is also necessary to
exact location of the boundary, as well as the crystallization
know the temperatures at points 4 and 6 (Figure 4), since they
operating temperatures, since they would affect the material
determine crystallization operating temperatures.
and energy balances. Additional data points can be taken to
With this in mind, a set of experiments can be designed.
increase the accuracy of the estimate. However, there is a
These experiments involve a series of mixtures for which: xD
limit beyond which additional effort has a rapidly diminish= x1; the relative amounts of A and B are different; and the dising return. There is always a degree of uncertainty until the
solution temperatures are determined using the step-warming
commercial process is up and running for a period of time.
method. The obtained data points (shown as open circles) lie
The expected financial return from the increased accuracy
on a vertical cut of the three-dimensional phase diagram,
must be compared against the expenses for performing addishown on the lower right corner of Figure 7. A double saturational experiments.
tion point (indicated by a diamond in the figure), can be locatPhase diagrams play a central role in the development of
ed using two methods, the first of which is by observing the
crystallization processes, because they offer useful insights
trend of increasing or decreasing dissolution temperature over
on the thermodynamic limitations for separations and the

38

www.cepmagazine.org

January 2004

CEP

Literature Cited
1. Matsuoka, M., Developments in Melt Crystallization, in Advances in Industrial Crystallization, Butterworth-Heinemann,
Oxford, U.K. (1991).
2. Jacques, J., et al., Enantiomers, Racemates, and Resolutions,
John Wiley and Sons, New York (1981).
3. Knapp, H., et al., Solid-liquid Equilibrium Data Collection,
Chemistry Data Series Vol. 8, DECHEMA, Frankfurt, Germany
(1987).
4. Prausnitz, J. M., et al., Molecular Thermodynamics of FluidLiquid Equilibria, 3rd ed., Prentice-Hall, Upper Saddle River, NJ
(1999).
5. Tester, J. W., and M. Modell, Thermodynamics and Its Applications, 3rd ed., Prentice-Hall, Upper Saddle River, NJ (1996).
6. Nicolaisen, H., et al., Correlation and Prediction of Mineral
Solubilities in the Reciprocal Salt System (Na+, K+)(Cl, SO42)H2O at 0100C, Chem. Eng. Sci., 48, pp. 3,1493,158 (1993).
7. Jakob, A., et al., Solid/liquid Equilibria in Binary Mixtures
of Organic Compounds, Fluid Phase Equil., 113, pp.
117126 (1995).
8. Wittig, R., et al., Binary Solid/liquid Equilibria of Organic Systems Containing e-Caprolactone, J. Chem. Eng. Data, 46, pp.
1,4901,493 (2001).
9. Zhang, L., et al., Measurement of Solid/liquid Equilibria
by a Flow-Cloud-Point Method, J. Chem. Eng. Data, 43, pp.
3237 (1998).
10. Fuyuhiro, A., et al., Solubility Isotherms of Reciprocal SaltPairs of Optically Active Cobalt (III) Complexes, Bull. Chem.
Soc. Japan, 52(1), pp. 9093 (1979).
11. Matsuoka, M., and R. Ozawa, Determination of S-L Phase
Equilibria of Binary Organic Systems by DSC, J. Cryst. Growth,
96, pp. 596604 (1989).
12. Linke, W. F., and A. Seidell, Solubilities of Inorganic and
Metal Organic Compounds, American Chemical Society, Washington, DC (1965).
13. Silcock, H. L., Solubilities of Inorganic and Organic Compounds, Pergamon, Oxford (1979).
14. Zemaitis, J. F., et al., Handbook of Aqueous Electrolyte Thermodynamics, AIChE, New York (1986).
15. Karapetyants, M. K., and M. L. Karapetyants, Thermodynamic Constants of Inorganic and Organic Compounds,
Humphrey Science Publishers, Ann Arbor, MI (1970).
16. Constantinou, L., and R. Gani, New Group Contribution
Method for Estimating Properties of Pure Compounds, AIChE J.,
40, pp. 1,6971,710 (1994).
17. Joback, K. G., and R. C. Reid, Estimation of Pure-Component Properties from Group Contributions, Chem. Eng. Commun., 57, pp. 233243 (1983).
18. Price, C. J., Take Some Solid Steps to Improve Crystallization, Chem. Eng. Prog., 93 (9), pp. 3443 (1997).

Further Reading
Samant, K. D., et al., Representation of High-Dimensional, Molecular Solid-Liquid Phase Diagrams, AIChE J., 46, pp. 2,4352,455
(2000).
Schroer, J. W., et al., Synthesis of Chiral Crystallization Processes, AIChE J., 47, pp. 369388 (2001).
Wibowo, C., and K. M. Ng, Unified Approach for Synthesizing
Crystallization-based Separation Processes, AIChE J., 46, New
York, pp. 1,4001,421 (2000).
Wibowo, C., and K. M. Ng, Visualization of High Dimensional
Systems via Geometric Modeling with Homogeneous Coordinates, Ind. Eng. Chem. Res., 41, pp. 2,2132,225 (2002).
Wibowo, C., and K. M. Ng, Workflow for Process Synthesis and
Development: Crystallization and Solids Processing, Ind. Eng.
Chem. Res., 41, pp. 3,8393,848 (2002).

possible ways to overcome them. Combined with focused

experimental effort, modeling of phase behavior and visualization using phase diagrams can significantly cut the required development time and effort. Starting with limited information, one can proceed with phase diagram estimation
and conceptual design of a separation process. An iterative
procedure with increasing levels of accuracy is followed,
coupled with systematic generation and incorporation of
new information. In this way, the development efforts are
channeled toward the regions of parameter and composition
space that have the highest impact on the overall process.
Thermodynamic considerations are insufficient for the
development of a complete crystallization process, since
they are not responsible for many issues that can occur during crystallization. For example, impurities can be incorporated into the final product, due to inclusion during crystal
growth or imperfect removal of mother liquor from the
crystals during filtration and washing. Unstable polymorphs
are often found in the crystallization product because they
crystallize faster than the stable ones. Formation of crystal
deposits on the vessel, or incrustation, occurs as a result of
excessive local supersaturation. Therefore, dynamic issues
such as nucleation, growth kinetics and mixing effects in
the crystallizer must be taken into account (18). However,
before adding complexities to the process, it is critical to
ensure that the process is thermodynamically feasible. For
this reason, thermodynamics is the primary concern in the
conceptual design stage, while the dynamic issues are dealt
CEP
with at a later stage of process development.
CHRISTIANTO WIBOWO is a senior engineer at ClearWaterBay Technology,
Inc. (CWB Technology, Inc., 20311 Valley Blvd., Suite C, Walnut, CA
91789, Phone: (909) 595-8928; Fax: (909) 595-6899; E-mail:
[email protected]). He is currently involved in various consulting
projects on the development of crystallization and solid/liquid
separation processes. Prior to joining CWB Tech, Wibowo conducted
research in solids processing and crystallization process development,
and has published many papers on these and other topics. Wibowo
holds a BS from Institute of Technology Bandung, Indonesia and a PhD
from the Univ. of Massachusetts, Amherst, both in chemical
engineering. He is a member of AIChE.
LIONEL OYOUNG is president and co-founder of CWB Tech (E-mail:
[email protected]). He has over 15 years of experience in process
synthesis and development in petrochemical, fine chemicals and
pharmaceuticals and holds several patents for various petrochemical
processes. OYoung earned both BS and PhD degrees in chemical
engineering from the Univ. of Manchester Institute of Science and
Technology (UMIST), Manchester, U.K. He is a member of AIChE.
KA MING NG is professor and head of chemical engineering, as well as
director of the consortium of chemical products and processes, at the
Hong Kong Univ. of Science and Technology (Clear Water Bay, Kowloon,
Hong Kong; E-mail: [email protected]). He has consulted for various
companies and is currently the corporate science and technology
advisor of Mitsubishi Chemical Corp. Ng holds a BS degree from Univ. of
Minnesota and a PhD from Univ. of Houston, both in chemical
engineering, and is a member of AIChE.

Acknowledgment
The authors wish to acknowledge their colleagues Vaibhav V. Kelkar, Ketan
D. Samant and Joseph W. Schroer for their comments and contributions
to the writing of this article.

CEP

January 2004

www.cepmagazine.org

39