7094 Ind. Eng. Chem. Res.

2008, 47, 7094–7103

Mathematical Modeling of the Destabilization of Crude Oil Emulsions Using

Population Balance Equation
Roberto E. P. Cunha,†,§ Montserrat Fortuny,†,‡ Claudio Dariva,†,‡ and Alexandre F. Santos*,†,‡
Programa de Engenharia de Processos, UniVersidade Tiradentes, Aracaju 49032-490 SE, Brazil, Laboratório
de Engenharia de Petróleo, Instituto de Tecnologia e Pesquisa, AV. Murilo Dantas 300, Aracaju 49032-490 SE,
Brazil, and PETROBRAS/ENGP/EIPA, R. Acre 2504, Aracaju 49080-010 SE, Brazil

In this work, the stability of water-in-crude oil emulsions generated in laboratory was investigated using a
phenomenological mathematical model based on the population balance equation, considering different
phenomena such as the binary coalescence of water droplets, the interfacial coalescence with the resolved
water phase, the diffusion of the dispersed phase, and droplet settling. The resulting population balance equation
(PBE) was a nonlinear hyperbolic integro-partial differential equation, which for our particular case required
numerical techniques for resolution. The PBE was converted into a system of partial differential equations
using Kumar’s fixed-pivot technique. The spatial coordinate was discretized using the finite volume method
and a first order upwind scheme, while the discretization of the time coordinate was based on a semi-implicit
approach. On the basis of this algorithm, the mathematical model was solved against experimental results of
water-in-crude oil emulsion destabilization runs, providing suited predictions of droplet size distribution profiles,
and of both emulsified water and free-water volumes.

1. Introduction water droplets is drained, allowing the contact between droplets

In the petroleum industry, the crude oil production is often
associated with high salinity produced water, gas, sediments,
and other contaminants. To achieve refinery specification, all
these oil contaminants must be segregated and removed. The
water hindered in the petroleum can be either under the free or
emulsified form with the oil. The emulsified water can represent
about 60% of the total water and, therefore, should be separated
in the production gravitational separators and coalescing treaters,
to achieve a maximum of 1% water content.1 Depending upon
the crude oil source and on the production scheme, very tight
water-in-crude oil emulsions can be formed during the crude
oil production, making the breakup of these emulsions a difficult
task. The stability of water-in-crude oil emulsions has been
viewed as one of the greatest challenge problems in the primary
separation of crude oils.2 The comprehension of the demulsi-
fication mechanisms involved in these systems is very compli-
cated due to several factors such as the complex composition
of the natural emulsifiers, the not well-known stabilization
mechanisms, and the strong influence of the experimental
conditions (water content, aqueous phase composition, droplet
size distribution, temperature, emulsion age, etc.) on the stability
of the water-in-crude oil emulsions.3
The general demulsification procedure found in the petroleum
production facilities to resolve a water-in-crude oil emulsion
into bulk phases of oil and water can be viewed as a three stage
process involving, (i) destabilization, (ii) coalescence, and (iii)
gravity separation. In short, the destabilization stage involves
the weakening of the stabilizing effect of the natural emulsifiers
that form a film surrounding the dispersed water droplets. This
is usually accomplished by adding heat and/or a properly
selected interfacially active chemical compound to the emulsion.
The coalescence stage occurs when the film surrounding the
* To whom correspondence should be addressed. Tel.: +55 79
32182115. Fax: +55 79 32182190. E-mail: alexfersantos@yahoo.com.br.
†
Universidade Tiradentes.
‡
Instituto de Tecnologia e Pesquisa.
§
PETROBRAS/ENGP/EIPA.
10.1021/ie800391v CCC: $40.75  2008 American Chemical Society
Published on Web 08/06/2008
that coalesce into larger droplets. This is generally accomplished
by imposing a period of moderate agitation or by subjecting
the destabilized emulsion to an alternating electric field, yielding
high rates of contact of dispersed water droplets. The gravity
separation stage requires sufficient residence time and a favor-
able flow pattern in a tank or vessel that will allow the coalesced
droplets of water to separate from the oil.1 As a consequence,
demulsification plants depend on the treating temperature,
chemical usage and physical size of treating equipment. The
design of a system for treating crude oil emulsions has been
traditionally relied on experience and empirical data from other
wells or fields in the area and on laboratory experiments.4,5
Numerous experimental and theoretical studies have also been
presented in the literature in order to improve the understanding
of mechanisms that may influence the crude oil emulsion
destabilization. Traditional laboratory techniques based on
droplet size distribution analyses and the so-called “bottle test”,
which consists of determining by inspection the amount of water
that separates from a quiescent emulsion sample with time, are
among the main experimental techniques used in laboratory-
scale investigations to evaluate the demulsification process. Due
to their simplicity, these laboratory techniques have been also
employed in the petroleum field to tune the operation conditions
of the separation facilities, including the temperature, the amount
of demulsifier, and the residence time. Nevertheless, these
experimental approaches are unable to provide information on
the desestabilization mechanisms. For instance, “bottle tests”
provide a means for estimating ranges of treating temperature
and retention time for design purposes. However, these tests
are static in nature and do not model closely the dynamic effects
of dispersed water droplets and the coalescence that occur during
actual flow of crude oil emulsions.1,4 On the other hand,
theoretical studies based on the joint use of mass balance and
Stokes law have been proposed in the literature to simulate the
destabilization of petroleum emulsions or to design separation
vessels.4,6 In fact, the main phenomenon taken into account in
these models is the droplet settling, which can be easily probed
by experimental measurements of average droplet sizes and

Table 1. Crude Oil Characteristics
property/component units value
density (deg API) 22.8
WC wt % 0.6
salinity mg NaCl/L of crude oil 451.4
TAN mg KOH/g crude oil 0.74
viscosity cP 246.4 at 38 °C
134.0 at 50 °C
92.2 at 60 °C
SARA analysis
saturated wt % 47.4
aromatics wt % 21.7
resins wt % 25.5
asphaltenes wt % 5.4
retention times. Despite interesting pieces of information
concerning the demulsification process can be obtained by either
“bottle tests” or mass balance combined with Stokes law, much
work remains to be done in terms of generating robust
mathematical descriptions of the water-in-crude oil destabiliza-
tion process.
Useful mathematical models based on population balance
equations (PBE) have been developed and implemented to
describe the evolution of dispersed systems, taking into account
their dynamics as a result of a set of events such as coalescence,
breakage, nucleation, settling, and growth of particles or flocs,
droplets, bubbles, crystals, etc.7-15 Each particular event should
be described in accordance with the dispersed phase properties,
for instance, the rate of coalescence between particles and the
settling rate of one particle depend on a set of parameters
including the diameter of the involved particle, and the physical
properties of the continuous phase (e.g., viscosity, density,
temperature). As a consequence, PBE models usually consider
two types of coordinate systems: internal coordinate, which
represents quantitative properties of the dispersed phase (e.g.,
volume, mass, density, age, etc.), and the external coordinate
which stands for the particle position in the physical space. The
resolution of the PBE model has allowed for predicting the entire
particle size distribution in a number of systems.15 However,
in this class of models, the number of works published dealing
with crude oil emulsions is rather limited.
In the current work, a general mathematical model based on
population balance equations is proposed to describe water-in-
crude oil destabilization processes. This model includes binary
coalescence of water droplets, interfacial coalescence with the
resolved water phase, the diffusion of the dispersed phase and
droplet settling. The solution techniques for solving the model
are outlined, and simulation results are presented for a typical
case of crude oil emulsion destabilization, as well as compari-
sons with experimental results.
2. Experimental Details
2.1. Crude Oil Characterization. An experimental study
was conducted to evaluate the destabilization performance of
water-in-crude oil emulsions prepared in the laboratory. A
Brazilian crude oil was sampled in the petroleum field and used
in this study. The characteristics of this crude oil are summarized
in Table 1, and the techniques used in its characterization are
described as follows.
The dynamic viscosity of the crude oil was determined using
a rotational rheometer (Physica Rheolab MC1, from Anton Paar)
equipped with a heating jacket. The analysis consisted of
measuring the shear stress as a function of the shear rate keeping
the sample at the desired temperature ((0.3 °C).
The water content (WC) of the crude oil was determined by
the Karl Fischer (KF) reagent method, in accordance with
Ind. Eng. Chem. Res., Vol. 47, No. 18, 2008 7095
ASTM D-1744 procedures.16 The solvent used during the
analysis was a mixture of dry methanol and chloroform (20%,
v/v). For standardization of the KF reagent, distilled water was
solubilized into the solvent cited before. A Metrohm KF titrator
(model 836 Titrando) equipped with a double platinum electrode
was employed during the water content determination tests.
The salinity of the crude oil was determined using the
electrometric method according to ASTM D-3230 procedures.17
The total acid number (TAN) was determined by potentio-
metric titration of the crude oil with alcoholic potassium
hydroxide (KOH) solution, in accordance with the ASTM D-664
test method.18 Prior to each titration, crude oil samples were
dissolved in a 50% v/v toluene/isopropanol solution. The same
automatic titrator used for Karl Fischer analysis (Metrohm 836
Titrando) was employed for the acid number determination,
however, equipped with a combination electrode suitable for
nonaqueous titrations. Such an electrode consisted of a glass
electrode combined with a silver/silver chloride (Ag/AgCl)
reference electrode built in the same electrode body. The acid
number is then described by the mass of KOH used to titrate
sample per weight of crude oil (mgKOH/g oil). SARA analysis
was performed at Petrobras R&D Centre, where a standard
chromatographic procedure has earlier been developed for the
semipreparative separation of crude oils and related materials
into the four SARA fractions.
2.2. Emulsion Preparation and Characterization. The
emulsions were produced according to a standard protocol
considering a pre-emulsification stage and an emulsification
stage. Prior to the pre-emulsification stage, the crude oil was
shaken vigorously to ensure homogeneity before sampling and
heated (<50 °C) to lower the viscosity to a manageable level.
During the pre-emulsification stage, a known amount of water
was added stepwise to the hydrocarbon phase, while the mixture
was hand-shaken to completely incorporate the water, resulting
in a pre-emulsion with a total volume of 300 mL. In the second
stage, this mixture was homogenized using the selected emul-
sifying equipment. The ratio of water and crude oil in the
mixture was varied to make up an emulsion with the desired
values of WC (40-45%) and of average droplet size. To yield
emulsions with distinct droplet size distributions (DSD), two
different types of emulsifying equipment were considered in
this stage: an Ultra-Turrax T-25 homogenizer fitted with a S25-
25G dispersing tool provided by IKA and a variable speed
mechanical agitator from FISATOM 713 D fitted with a three-
bladed airscrew impeller. The mixing conditions, such as the
stirring frequency and the time of homogenization were set
around 18 000 rpm for the Ultra-Turrax, 6000 rpm for the
mechanical agitator, with homogenization times of 5 min for
both devices. Finally, a known amount of a commercial
demulsifier (Dissolvan provided by Clariant) was dispersed into
the emulsion sample by means of the mechanical agitator during
1 min, with concentrations of 0, 12, and 24 ppm.
The water content of the emulsions was determined by the
Karl Fischer (KF) reagent method presented before. The DSD
of emulsion samples was obtained with a Malvern Instrument
Model 2000 Mastersizer particle size analyzer based on the laser
diffraction technique. The detection range of the instrument
varies from 0.020 to 2000 µm. The apparatus is appropriated
to perform measurements of systems with low concentrations
in a transparent environment but also in slightly dark media.
Because the emulsion samples used were dark and concentrated,
a diluent was required. The main purpose of dilution is to avoid
multiple scattering by getting the droplets far away from one
another. Typically, the solutions were prepared by dispersing
7096 Ind. Eng. Chem. Res., Vol. 47, No. 18, 2008
Figure 1. Sampling scheme used to gauge the extent of emulsion
destabilization.
5-10 droplets of emulsion in 200 mL of a white mineral oil
(EMCA 70, 19.3 cP, 25 °C) used as a diluent.
2.3. Destabilization Tests. Emulsion destabilization tests
were conducted using gravity separation with and without the
addition of demulsifier. The tests were performed in batch using
a 500 cm3 glass beaker as the separation vessel (9.0 cm in height
and 8.8 cm in internal diameter). The standard procedure for
demulsification runs involved the addition of 420 cm3 of water-
in-crude oil emulsion with WC ) 40 wt %, resulting in a 6.9
cm liquid column height inside the separation vessel. All
operation occurred at ambient conditions (25 ( 2 °C).
To follow the extent of emulsion destabilization, small
aliquots of the emulsion were collected from two levels of the
separation vessel for determining droplet sizes and WC in
accordance with the characterization methods described before.
These aliquots were sampled with the help of two syringes
connected to needles with distinct lengths, enabling thus to take
samples near the top and the bottom of the vessel, as illustrated
in Figure 1. According to this figure, the sampling points were
located at 6.0 and 2.0 cm from the bottom of the separation
vessel. These distances are quite important when we are to
compare the experimental data with the DSD and WC values
calculated by the mathematical model along the height of the
separation vessel.
The volume of free-water, if any, was monitored by visual
inspection of the graduated wall of the vessel along the
destabilization run. Typically, the destabilization runs took
1-12 h.
2.4. Mathematical Model. The water-in-crude oil emulsion
separation was described through a phenomenological math-
ematical model based on the population balance equation (PBE)
which included a general set of physical mechanisms that cover
a wide range of emulsion separation systems. Among the main
phenomena included in the PBE, besides the droplet settling,
are binary coalescence of water droplets (in which droplets
coalesce against large droplets), interfacial coalescence with the
resolved water phase (in which droplets coalesce against a layer
of bulk fluid), and diffusion of the dispersed phase.
The PBE presented on the current work was written consider-
ing time (t) and the axial coordinate (z) as external coordinates
and the droplet volume (V) as an internal coordinate. For a
settling column, the dynamic population balance equation for
the droplet size distribution can be described by eq 1,7,8 where
n is the number distribution function, given as the number
concentration of droplets, with w, Dz, and θ representing
respectively the settling velocity, the vertical diffusion coef-
ficient, and the source term. The last term in eq 1 stands for the
interfacial coalescence phenomenon, which depends on the
interfacial coalescence velocity of a given droplet (Vf) and on
the position of the free-water-emulsion interface (zint), as
described by the Dirac Delta function. The source term is given
by eq 2 and accounts for the formation and death of droplet
classes due to binary coalescence. This term depends on the
collision frequency function (β) that defines the rate at which
droplets of volume V and V′ will coagulate to form a droplet of
volume V + V′.
∂n ∂ (wn) ∂ ∂ (n)
∂t
+
∂z
) D
∂z z ∂z ( )
+ θ - Vfnδ(z - zint) (1)
∫
1 V
θ) β(V - V ′ , V′)n(V - V ′ , z, t)n(V ′ , z, t) dV ′ -n
2 0
∫
∞
(V, z, t)
0
β(V, V′)n(V ′ , z, t) dV′ (2)
The fixed pivot technique proposed by Kumar and Ramkrishna19
was used to discretize the PBE with respect to the internal
coordinate, based on a nonuniform geometric grid given by xk+1
) qxk, where xk represents the droplet volume at the pivot of
the kth interval and q is the adjustable geometric size interval.
The discretized version of the PBE is shown in eq 3, in which
the original integro-partial differential equation is transformed
into a set of Np partial differential equations, where nk represents
the droplet number concentration over the kth interval and Np
stands for the number of discrete intervals of the internal
coordinate. Equation 4 accounts for the binary coalescence
phenomenon, where ηk is the contribution function adopted on
the fixed pivot technique19 to define the distribution among the
volume classes in accordance with eq 5.
Equation 6 describes the settling velocity, which is assumed
to follow Stokes law. In this equation, the term K1 is an
adjustable parameter that stands for the settling efficiency. Note
that according to this equation, the settling velocity possesses
negative values, which matches with the orientation of the
droplet movement downward with respect to the z axial
coordinate, which in turn is oriented in the ascendant direction.
Equation 7 represents the collision frequency between a
droplet from the ith interval with a droplet from the kth interval,
assuming the droplet contact mechanisms of Brownian motion
and differential sedimentation. This latter mechanism is related
to the interaction between droplets of different settling velocities.
On this equation, K2 is an adjustable parameter that stands for
the collision efficiency, βBr(i,k) is the collision frequency due to
Brownian motion, and βDS(i,k) is the collision frequency due to
differential sedimentation. These collision frequencies are
calculated through eqs 8 and 9. Note that eqs 7-9 are still valid
in the event that the interaction occurs between a droplet from
the ith interval and another from the jth interval, requiring only
a change of subscripts from k to j. Similar collision frequency
expressions have been employed in the literature.7,8 On these
equations, the subscripts i, j, and k denote droplet volume
classes; d is the droplet diameter; K is the Boltzman constant;
T is the temperature; µc is the continuous phase viscosity; Fd is
the dispersed phase density; Fc is the continuous phase density;
g is the gravity acceleration; and Vk is the droplet volume of
the kth interval.
The interfacial coalescence velocity of a given droplet (Vf)
was assumed to be proportional to the height of the dense-
packed zone (∆h) and independent of the droplet diameter, in
accordance with eq 10. This equation takes into account the
effect of the weight of the dense-packed layer, which compresses
the droplets located in the interface. Similar reasoning was
employed by Jeelani and Hartland20 on their modeling stud-
ies concerning a batch liquid-liquid separation system. Also
in this equation, the term K3 is an adjustable parameter which
accounts for the frequency of coalescence of a droplet into the
 Ind. Eng. Chem. Res., Vol. 47, No. 18, 2008 7097
free-water layer; the term ∆z stands for the height of one volume Table 2. Experimental Conditions of Separation Tests
element and is employed to normalize the contribution of the initial emulsion properties preparation condition
thickness of the dense-packed zone. emulsifying demulsifier,

( )
run WC, % D(4,3), µm equipment ppm
∂nk ∂ (wknk) ∂ ∂ (nk)
+ ) D + θk - Vfnkδ(z - zint) E1 43.9 29 agitator 0.00
∂t ∂z ∂z z ∂z
E2 42.0 53 homogenizer 11.2
with k ) 1, ..., Np (3) E3 41.8 45 homogenizer 23.8
igj Np E4 40.0 70 agitator 23.8
θk )
1
∑η β i,jninj - nk ∑β i,kni (4)
E5 44.2 134 agitator 11.2

{
k
2 i,j i)1
transforms the population balance equation into a set of linear
xk+1 - V algebraic equations, which can be written in a standard
, if : xk e V e xk+1
xk+1 - xk tridiagonal form, enabling us to use the standard Thomas
ηk ) (5)
V - xk-1 algorithm22 to solve the resulting system. All numerical routines
, if : xk-1 e V e xk were written in FORTRAN 90 (Compaq Visual Fortran 6.6),
xk - xk-1
and calculations were performed by means of an Intel Core 2
(Fc - Fd)dk g
2
Duo, 2.0 GB RAM, 1.73 GHz computer.
wk ) K1 (6) The mathematical model was finally validated for batch
18µc
separation processes conducted under ambient and constant
βi,k ) K2(βBr(i,k) + βDS(i,k)) (7) temperatures. However, as actual emulsion separation processes

( )
2KT 1 1 commonly involve heating, the mathematical model was
βBr(i,k) ) + (d + dk)
µc di dk i
π the application of the model to the simulation of separation
βDS(i,k) ) (di + dk)2|wi - wk|
4 processes involving thermal treating, providing appropriate
∆h
Vf ) K3
∆z
For model resolution, an initial condition (t ) 0) and two
boundary conditions (the lower boundary at z ) 0 and the upper
boundary at z ) H, where H is the emulsion height in the
separation vessel) were considered. For the experimental system
under study, the initial condition describes the values of nk for
k ) 1,..., Np at t ) 0. Such values are provided by means of
DSD measurements performed in the laboratory during the
emulsion sample preparation; the lower boundary condition
relates to the zone of interfacial coalescence, while the upper
boundary condition relates to the zone of air-liquid interface.
Equation 11 was applied as the boundary condition at z ) 0
and H. Solving the mathematical model at z ) 0 with the use
of eq 11 results in a zero net change of the distribution, this
being modified at z ) 0 solely by binary and interfacial
coalescence phenomena. The same approach done at z ) H
results in zero axial flux in the top of the separation vessel,
with the distribution being modified only by binary coalescence.
These boundary conditions were found to yield the best
description of the experimental data, allowing us to solve the
mathematical model without any problem of discontinuity or
instability and related issues.
∂nk
wknk ) Dz
∂z
The finite volume method was chosen for discretization of the
model with respect to the external axial coordinate. For the
diffusive term, the first spatial derivative of each face was
approximated by a first order central differencing scheme and
for the advective term an upwind scheme was applied, since
the central differencing schemes adopted on this term are known
to cause excessive numerical oscillations.7 For further details
on finite volume method and related approximations, the reader
is referred to the work of Leveque.21 Besides, the fixed pivot
technique was employed to discretize the internal coordinate
(droplet volume) and a semi-implicit scheme based on the finite
differencing approach was used for integration of the model
with time. The application of these numerical techniques
(8) formulated considering the temperature as an input variable, as
explicitly shown in eq 8 (Brownian motion), making possible
(9)
parameters K1, K2, K3, and physical properties (µc, Fc, and Fd)
(10) at a given temperature are available.
3. Results and Discussion
3.1. Laboratory-Scale Separation Tests. Five destabiliza-
tion runs were carried out using the separation vessel described
before. Experimental conditions are reported in Table 2, which
includes WC measurements, average droplet sizes by
volumesD(4,3), demulsifier concentration, and type of emulsify-
ing equipment. All runs were performed with demulsifier
addition, except in the first test. This run E1 was used as a
control test allowing us to evaluate the destabilization process
without chemical treating and to observe how stable the
emulsion under study was.
As expected, the DSD evolution of the emulsion at the bottom
and at the top of the separation vessel during run E1 revealed
that changes of DSD are negligible after 4 h of settling. For the
sake of brevity, results for run E1 are not shown here, but they
indicate that very low rates of binary coalescence occurred
between droplets. Moreover, no free water was observed even
after 7 h of settling, which means that very low rates of
interfacial coalescence occurred during this run. Analogously,
the WC evolution of the emulsion during E1 was essentially
kept constant at both the bottom and the top of the separation
(11) vessel. Although slight differences of WC between top and
bottom do exist, one may conclude that the droplet settling
during E1 was negligible.
According to these preliminary results, it may be said that
the natural surfactants (eg, asphaltenes, resins, etc.) originally
present in the crude oil sufficed to yield stable emulsions. In
particular, the emulsion preparation procedure adopted for E1
provided a dispersed phase composed of droplets with very low
rates of binary and interfacial coalescence. As a consequence,
the addition of demulsifier into the emulsion is of practical
importance, because this would allow us to study the destabi-
lization dynamics, providing information about coalescence
phenomena.
Following the test with no demulsifier addition, two desta-
bilization runs (E2 and E3) were conducted using emulsion
7098 Ind. Eng. Chem. Res., Vol. 47, No. 18, 2008
Figure 2. Time-dependent droplet size distributions of the water-in-crude
oil emulsion sample during run E2: top and bottom of the separation vessel.
Figure 3. Time-dependent droplet size distributions of the water-in-crude
oil emulsion sample during run E3: top and bottom of the separation vessel.
samples prepared by means of the homogenizer as the emulsify-
ing equipment, but containing distinct amounts of demulsifier.
Even though the emulsification experience indicates that the
homogenizer device tends to generate very fine droplets during
the homogenization, the presence of demulsifier contributes to
modify the rheological properties of the interfacial film sur-
rounding the droplets promoting the drainage of this film and
the formation of coarser droplets.23 Note from Table 2 that the
average droplet sizes by volume of emulsions prepared for runs
E2 and E3 are significantly higher than in the case of run E1,
indicating apparently a demulsifier effect on the droplet
coalescence. Additionally, comparing the average droplet sizes
of emulsions prepared using the mechanical agitator (runs E1,
E4, and E5), one may conclude that a coalescence effect does
occur during the demulsifier addition and dispersion, even when
a very low concentration of demulsifier is used.
Figures 2 and 3 report the DSD measurements taken from
the top and the bottom of the separation vessel during demul-
sification runs E2 and E3, respectively. As observed, the DSD
curves of these runs are similar and evolve in both tests toward
higher droplet sizes probably due to the binary coalescence
phenomenon. Also, note that differences between DSD curves
at the top and the bottom are negligible along the first 60 min,
Figure 4. Free-water resolution with time during runs E2-E5 (the curves
are linear fittings).
Figure 5. Time-dependent WC values of emulsion samples during runs E2
and E3: top and bottom of the separation vessel (the lines are visual aids).
in both runs. This fact is indicative of little settling of droplets
during this interval and may be confirmed later when regarding
the WC profiles of these runs. It should be pointed out that
droplet sizes with peaks beyond 1000 µm are rather inaccurate
and should be seen with caution, because these sizes are located
around the upper detection limit of the particle size analyzer.
The free-water resolution profiles during runs E2-E5 are
shown in Figure 4. Comparing runs E2 and E3, when higher
demulsifier concentration is employed, the free-water formation
is faster (run E3 is faster than run E2, as expected). Different
from run E1 in which no chemical was fed into the emulsion,
both runs E2 and E3 involved interfacial coalescence phenom-
ena, leading to the formation of decanted water (we will return
to Figure 4 later when discussing about runs E4 and E5).
The settling of water droplets during demulsification runs E2
and E3 can be inferred from measurements of WC at the top
and bottom of the separation vessel, as illustrated in Figure 5.
It becomes clear from the data reported in this figure that the
WC values decay with time in the top region as a result of the
droplet settling, reaching a steady value around 15% (wt) for
both runs. On the other hand, the WC values increase with time
at the bottom region, slowing down and reaching the 100% mark
asymptotically, which means the formation of a clear layer of
free-water at the bottom. Note that the WC profiles are very
similar in both runs, principally if we are to consider the top
region. As a final remark concerning these runs, one may verify
that the settling effects were still more pronounced after 60 min,
probably leading to the accumulation of water at the bottom.

Figure 6. Time-dependent droplet size distributions of the water-in-crude
oil emulsion sample during run E4: top and bottom of the separation vessel.
Figure 7. Time-dependent droplet size distributions of the water-in-crude
oil emulsion sample during run E5: top and bottom of the separation vessel.
These results indicate that both binary coalescence and settling
phenomena do evolve in a similar fashion for runs E2 and E3.
However, distinct water separation rates were assessed on these
tests, which can be seen as an indication of an effect of the
demulsifier concentration over the interfacial coalescence rate.
Following the tests E2 and E3 discussed above, two additional
demulsification runs (E4 and E5) were performed considering
initial emulsion charges containing bigger droplets, as reported
in Table 2. These additional runs involved initial DSD profiles
very distinct from runs E2 and E3. The DSD curves obtained
for the initial emulsion charges of each demulsification run
summarized in Table 2 and involving demulsifier addition can
be retrieved from Figures 2, 3, 6, and 7. A simple visual analysis
shows that the DSDs of the water-in-crude oil emulsions are
very sensitive to the change of the emulsifying equipment,
indicating also a marked widening of the distribution when the
homogenization has been performed with the mechanical
agitator. These distributions are of special interest to evaluate
the droplet size effect on the settling, binary and interfacial
coalescence phenomena.
DSD measurements taken from the top and the bottom of
the separation vessel during demulsification runs E4 and E5
Ind. Eng. Chem. Res., Vol. 47, No. 18, 2008 7099
Figure 8. Time-dependent WC values of emulsion samples during runs E4
and E5: top and bottom of the separation vessel (the lines are visual aids).
are shown in Figures 6 and 7, respectively. The DSD curves
are shifted toward higher sizes in both runs, indicating that the
binary coalescence phenomenon plays a role in the separation
process. It is interesting to observe that the coalescence events
during these destabilization runs lead to the formation of droplet
distributions with similar shapes in both tests, but at distinct
rates, as one may infer from time labels informed in the plots.
These rate differences can be linked to the differences on the
initial droplet sizes between runs, which in turn influence the
water resolution, as discussed on the following paragraphs.
The free-water resolution profiles during runs E4 and E5 were
shown in Figure 4. A simple visual inspection shows that the
volume of resolved water grows linearly with time in both cases,
with run E5 keeping a higher demulsification rate, despite of
the lower demulsifier concentration employed on this run. This
apparent contrast can be explained by the higher droplet sizes
involved in the initial emulsion of run E5, as reported in Table
2. On the other hand, as one can deduce from Figure 4, note
that the free-water formation rates of runs E4 and E5 are higher
than those obtained during runs E2 and E3, with runs E4 and
E5 involving higher droplet sizes in their initial emulsions (as
shown in Table 2). Recall that the settling rate is proportional
to the square of the droplet diameter (Stokes law). As a
consequence, large droplets travel faster throughout the medium
and collide with a greater number of droplets, thus increasing
the number of successful collisions leading to coalescence.
Figure 8 presents measurements of WC at the top and bottom
of the separation vessel during demulsification runs E4 and E5.
Note that the settling of water droplet is intensified after 30
min and is relatively fast in both runs, with run E5 involving a
slightly higher settling rate. Even though measurements for run
E4 are more infrequent, which could lead to some uncertainties
on the WC profile, trends are clearly seen. Moreover, these WC
results reinforce the discussion presented before (concerning
comparisons with runs E2 and E3), indicating the sharp influence
of the initial DSD on the settling rate of the water droplets with
consequences on the demulsification dynamics. Finally, these
experimental results were partially employed to validate the
mathematical model developed on the current work.
3.2. Simulations of Laboratory-Scale Destabilization of
Water-in-Crude Oil Emulsions. The experimental data ob-
tained during run E4 were chosen for model validation purposes
because the DSD obtained for the initial emulsion charge
prepared for run E4 was very similar to those encountered for
samples collected from the inlet stream of a production
gravitational separator in the petroleum field.24 As reported by
7100 Ind. Eng. Chem. Res., Vol. 47, No. 18, 2008
Table 3. Parameters Used in the Numerical Method and for Model
Simulation No. 1
parameter value
number of finite volumes of spatial Nz ) 400
coordinate
number of discrete intervals of the Np ) 67
internal coordinate
adjustable geometric size interval q ) 1.5
height of one volume element of the ∆z ) 0.017 cm
spatial coordinate
time interval ∆t ) 2s
settling efficiency K1 ) 8.1 × 10-2
binary coalescence efficiency K2 ) 2.5 × 10-2
interfacial coalescence efficiency K3 ) 1.66 × 10-5
Cunha,24 the same crude oil employed in the current work may
generate stable emulsions in the field because of the presence
of a choke valve in the production line, leading to the formation
of droplets with average sizessD(4,3)saround 60-75 µm, being
very similar to that obtained for the emulsion prepared in run
E4sD(4,3) ) 70 µm (from Table 2). On the basis of the
separation results obtained on this run, the values of the
parameters K1, K2, and K3 (related respectively with Stokes law,
binary coalescence, and interfacial coalescence) used in the
simulations were chosen to give the best fit to the experimental
results of run E4. Such parameters are unknown and represent
to a certain extent a measurement of the efficiency of each
phenomenon, taking into account the viscous and concentration
effects which are likely to occur in concentrated water-in-crude
oil emulsion systems.
The initial distribution fed into the mathematical model was
the experimental DSD curve obtained for the initial emulsion
charge prepared for run E4. A first simulation test was performed
considering this initial distribution and the set of numerical
parameters reported in Table 3, which includes the parameters
related to the numerical discretization of time, spatial and
internal coordinates: Nz ) 400, Np ) 67, q ) 1.5, ∆z ) 0.017,
and ∆t ) 2 (with ∆z given by the height of the liquid column
inside the separation vessel divided by Nz). The application of
these parameters into the simulation of a demulsification run
of 4 h and 20 min yielded a computational time of 48 h and 28
min.
Table 3 also summarizes the estimates obtained for the
adjustable parameters of the mathematical model. The parameter
fitting was done mostly based on the results of a sensitivity
analysis. Pure statistical parameter fitting was not performed
due to a number of reasons. First, the CPU time for the
calculation of WC and DSD profiles along a typical demulsi-
fication run was too high, making impossible the application
of optimization techniques based on successive evaluations of
the model. Second, the lack of experimental data makes difficult
the application of regression techniques. Third, some of these
parameters are completely unknown, making difficult the
definition of initial guesses and intervals for accurate parameter
estimation.
Figure 9 shows the experimental and predicted behavior of
the emulsion DSD at the top and the bottom of the separation
vessel at distinct instants. The correspondent D(4,3) of each DSD
is reported in Table 4. According to this figure, the distributions
predicted by the mathematical model evolve from the initial
curve (experimental input) toward higher droplet sizes keeping
its monomodal shape. At 30 min, both experimental and model
predicted curves were widened and no differences were observed
between top and bottom samples. Moreover, the correspondent
D(4,3) value predicted by the mathematical model at 30 min is
in very good agreement with the experimental value, as reported
in Table 4. These results are indicative of the proper description
Figure 9. Time-dependent droplet size distributions of the water-in-crude
oil emulsion during run E4: experimental and model predictions at the top
and bottom of the separation vessel.
Table 4. Average Droplet Sizes by Volume along Run E4:
Experimental vs Model Predictions
D(4,3) (µm)
time (min) level experimental simulation
0 70 70
30 top 94 92
30 bottom 98 92
120 top 675 388
120 bottom 760 936
of the binary coalescence phenomenon, which is supposed to
be the main effect that may govern the destabilization at this
period of time. Concerning predictions at 120 min, note that
the DSD curves keep widening, with bimodal distribution curves
being measured by the particle size analyzer. However, these
bimodal curves should be viewed with care. Recall that as
discussed before in section 3.1, peaks beyond 1000 µm are rather
inaccurate because these sizes are located around the upper
detection limit of the particle size analyzer. A further proof of
the reliability of the mathematical model is given by the proper
predictions of D(4,3) at 120 min, which indicate coarser droplets
at the bottom and smaller droplets at the top of the separation
vessel, this being clearly influenced by the sedimentation
phenomenon. It is interesting to observe that the experimental
D(4,3) values also follow the abovementioned behavior, despite
the uncertainties inherent to these coarse emulsion samples in
particular.
Figure 10 shows experimental data measured during run E4
and simulation results for free-water volume-versus-time curves.
Model predictions are in very good agreement with the
experimental values of free-water volume. The linear behavior
of the free-water formation during run E4 is well explained by
the mathematical model, which indicates the accuracy of the
parameters adopted here, at least with respect to the water
separation. Both experimental and simulated data predict the
formation of decanted water since the very beginning of the
demulsification run. It should be mentioned that proper predic-
tion of the water separation rate depends on the correct
description of the interfacial coalescence phenomenon in the
system, among others. Experimental evidence of water-in-crude
oil (W/O) emulsions shows that the water droplets sediment
downward when water is denser than the oil to form a dense-
packed layer at the bottom of the dispersion. Since the thickness

Figure 10. Free-water formation: model predictions vs experimental (run
E4).
Figure 11. Time-dependent WC values at distinct levels during the
destabilization process: model predictions vs. experimental (run E4).
∆h of the dense-packed layer increases with time during settling
so does the interfacial coalescence rate. This point is ac-
complished in the model by defining the interfacial coalescence
velocity as a linear function of ∆h, as shown in eq 10. A value
of the minimum water content in the dense-packed layer (WCdpl)
in which the droplets coalesce to form the free-water zone can
be assumed. Similar reasoning has been proposed in the
literature to describe the coalescence phenomenon during
separation processes of W/O dispersions.20 Thus, a minimum
water content in the dense-packed layer WCdpl ) 40% was used
as a model input for all simulations performed in this work.
This minimum value coincides with the WC of the emulsion
initially charged into the separation vessel. Of course, this
assumption should be revised if we are to treat with different
emulsion and/or vessel systems.
Concerning actual petroleum treating processes typically
found in the petroleum production plants, the free-water
dynamics represents one of the most important pieces of
information for gravitational separator design and residence time
specification. Consequently, model results presented before are
Ind. Eng. Chem. Res., Vol. 47, No. 18, 2008 7101
encouraging and indicate that our modeling approach can
constitute a useful tool for designing oil emulsion separation
systems.
The simulation results obtained for WC at four distinct levels
of the separation vessel during the demulsification run are shown
in Figure 11. Experimental values of WC at the top and bottom
of the separation vessel are also plotted for comparison. Given
the relatively small number of experimental measurements
concerning run E4, experimental data coming from run E5 are
also overlaid in this figure. As seen before in Figure 8,
experimental WC profiles of runs E4 and E5 are comparable,
principally in the top of the separation vessel. This is obviously
a rough approximation, but is adequate for the analysis of model
results. Moreover, note that none of the levels in which model
outputs are calculated necessarily match with those two levels
in which emulsion samples were taken and characterized
experimentally. Due to experimental limitations, small random
variations of the sampling system position are likely to occur,
making impossible a perfect correspondence between the heights
established for model predictions and those used for emulsion
sampling and characterization. Also, as emulsion samples are
taken from the separation vessel, the total volume of emulsion
in the vessel diminishes, which may eventually lead to gross
measurement errors.
As verified in Figure 11, model predictions accurately
describe the dynamic behavior observed during the demulsifi-
cation runs conducted in the laboratory, in which the WC values
decay with time at the top region and increase at the bottom of
the separation vessel as a result of the droplet settling. Besides,
improved agreement with experimental WC values was observed
for model predictions around the top of the vessel, probably
because this region is less subjected to experimental errors.
Nevertheless, the mathematical model is able to capture the
asymptotical profile of the WC data at the distinct levels.
It is important to observe that with minimal data fitting, the
model gives a quantitative explanation for many of the demul-
sification properties, including free-water volumes and D(4,3),
WC, and DSD profiles. Improved results would be obtained if
pure statistical parameter fitting had been done. However, this
would require a reliable experimental data set and improved
measurement devices.
To reduce the computational effort involved in the numerical
resolution of the mathematical model, three further simulations
(nos. 2-4) were performed to evaluate how the destabilization
results and CPU time would change upon slight modifications
of the numerical discretization parameters, Nz, q, ∆z, and ∆t.
Table 5 summarizes the results obtained through these simula-
tions and also includes the results related to the first simulation
already discussed, for comparison. It is seen from the results
that changes of these parameters can produce very distinct
computational times. Despite these computational time differ-
ences, note that the free-water volume estimates are in very good
agreement from simulation to simulation. It was also observed
that both WC and DSD profiles were essentially similar between
simulations (for the sake of brevity, these profiles are not
included here). As a consequence, the optimization of these
numerical parameters should be established to minimize the

Table 5. Numerical Parameters and Results Obtained during Free-Water Formation Simulations
simulation Nz ∆z q ∆t free-water volume (mL) after 140 min free-water volume (mL) after 260 min CPU time
1 400 0.017 1.5 2s 21.8 mL 39.8 mL 48 h 28 min
2 400 0.017 1.7 2s 22.2 mL 39.5 mL 22 h 00 min
3 400 0.017 1.7 1s 22.2 mL 39.5 mL 41 h 29 min
4 50 0.138 1.5 2s 21.8 mL 39.5 mL 06 h 13 min
7102 Ind. Eng. Chem. Res., Vol. 47, No. 18, 2008
computational effort, principally if we are to apply this
mathematical model into a process simulator framework.
4. Conclusions
Although there is a very large amount of literature dealing
with different aspects of water-in-crude oil emulsion destabiliza-
tion, there are very few papers that have attempted to present
comprehensive models useful for design and operation of
emulsion separation processes. In this paper, a mathematical
model that includes in detail the different physical phenomena
occurring in the emulsion separation vessel was presented. The
approach was to derive a phenomenological mathematical model
based on the population balance equation (PBE), considering
different phenomena such as the binary coalescence of water
droplets, the interfacial coalescence with the resolved water
phase, the diffusion of the dispersed phase, and droplet settling.
The resulting PBE was converted into a system of partial
differential equations using Kumar’s fixed-pivot technique. The
spatial coordinate was discretized using the finite volume method
and a first order upwind scheme, while the discretization of the
time coordinate was based on a semi-implicit approach. On the
basis of this algorithm, the mathematical model was solved
against experimental results of water-in-crude oil emulsion
destabilization runs, with minimal data fitting.
Simulation results demonstrated that the model was able to
accurately describe the dynamics of the coalescence and settling
phenomena involved in the gravitational separation of crude oil
emulsions. Indeed, the model allowed us to perform proper
predictions of DSD and water content profiles and of free-water
volumes in good agreement with experimental data, by using
the initial DSD curve measured experimentally as input. Also,
it was verified that the computational time can be deeply
reduced, without affecting model accuracy, if the numerical
parameters are optimized, which makes the model an interesting
tool for practical applications. In this sense, the mathematical
model presented here may be useful to size new separators and
to analyze existing equipment for changing conditions, providing
insight into both the free-water production and the quality of
the crude oil at different heights of the vessel.
Further studies will be conducted to optimize the parameters
of both the numerical method and the model and to validate
the mathematical model in a broader range of operation
conditions.
Acknowledgment
The authors thank Petrobras (Petróleo Brasileiro S.A., Brazil)
for supporting the present work. A.F.S. also thanks FAPITEC/
SE and CNPq (Brazilian funding agencies).
Notation
d ) droplet diameter, cm
dk ) droplet diameter of droplets of the kth interval, cm
Dz ) vertical diffusion coefficient, cm2 s-1
g ) gravity acceleration, cm s-2
H ) emulsion height in the separation vessel, cm
K ) Boltzman constant
K1 ) parameter in the settling velocity
K2 ) parameter in the coalescence rate function
K3 ) parameter in the interfacial coalescence velocity, s-1
n ) number distribution function, cm-3
nk ) droplet number concentration over the kth interval, cm-3
Np ) number of discrete intervals of the internal coordinate
Nz ) number of finite volumes of spatial coordinate
q ) adjustable geometric size interval
T ) temperature, K
t ) time, s
V, V′ ) droplet volume, cm3
Vf ) interfacial coalescence velocity, s-1
Vk ) droplet volume of the kth interval, cm3
w ) settling velocity, cm s-1
wk ) settling velocity of droplets of the kth interval, cm s-1
x ) droplet volume at the pivot, cm3
xk ) droplet volume at the pivot of the kth interval, cm3
z ) axial coordinate, cm
zint ) free-water and nonresolved emulsion interface, cm
Greek Letters
β ) collision frequency function, cm3 s-1
βBr(i,k) ) collision frequency due to Brownian motion between
droplets from the ith and kth intervals, cm3 s-1
βDS(i,k) ) collision frequency due to differential sedimentation
between droplets from the ith and kth intervals, cm3 s-1
βi,k ) collision frequency between droplets from the ith and kth
intervals, cm3 s-1
δ ) Dirac Delta function
∆h ) height of the dense packed zone, cm
∆z ) height of one volume element of the spatial coordinate, cm
ηk ) contribution function at the kth interval
θ ) source term, cm-3 s-1
θk ) source term of the kth interval, cm-3 s-1
µc ) continuous viscosity, g cm-1 s-1
Fc ) continuous phase density, g cm-3
Fd ) dispersed phase density, g cm-3
Literature Cited
(1) Smith, H. V.; Arnold, K. E. Crude oil emulsions. In Petroleum
Engineering Handbook, 3rd ed; Bradley, H. B., Ed.; SPE: Richardson, TX,
1992; pp 19.1-19.34.
(2) Sjoblöm, J.; Aske, N.; Auflem, I. H.; Brandal, Ø.; Havre, T. E.;
Sæther, Ø.; Westvik, A.; Johnsen, E. E.; Kallevik, H. Our Current
Understanding of Water-in-Crude Oil Emulsions Recent Characterization
Techniques and High Pressure Performance. AdV. Colloid Interface Sci.
2003, 100-102, 399.
(3) Mclean, J. D.; Kilpatrick, P. K. Effects of Asphaltene Solvency on
Stability of Water-in-Crude-Oil Emulsions. J. Colloid Interface Sci. 1997,
189, 242.
(4) Arnold, K. E.; Koszela, P. J. Droplet-Settling vs. Retention-Time
Theories for Sizing Oil/Water Separator. SPE Prod. Eng. 1990, 2, 59.
(5) Kokal, S. Crude Oil Emulsion: A State-Of-Art Review. SPE Prod.
Facilities 2005, 2, 5.
(6) Grødal, E. O.; Realff, M. J. Optimal Design of Two- and Three-
Phase Separators: A Mathematical Programming Formulation. Proceedings
of the SPE Annual Technical Conference and Exhibition, Houston, TX,
October 3-6, 1999.
(7) Attarakih, M. M.; Bart, H. J.; Faqir, N. M. Numerical Solution of
the Spatially Distributed Population Balance Equation Describing the
Hydrodynamics of Interacting Liquid-Liquid Dispersion. Chem. Eng. Sci.
2004, 59, 2567.
(8) Sterling, M. C.; Bonner, J. S.; Page, C. A.; Fuller, C. B.; Ernest,
A. N. S.; Autenrieth, R. L. Modeling Crude Oil Droplet - Sediment
Aggregation in Nearshore Waters. EnViron. Sci. Technol. 2004, 38, 4627.
(9) Tang, X.; Luo, G.; Li, H.; Wang, J. A Dynamic Interaction Mass
Transfer Model for Simulating the Mass Transfer Process in Extraction
Columns. Comput. Chem. Eng. 2006, 30, 978.
(10) Alopaeus, V.; Koskinen, J.; Keskinen, K. I. Simulation of the
Population Balances for Liquid-Liquid Systems in a Nonideal Stirred Tank.
Part 1 Description and Qualitative Validation of the Model. Chem. Eng.
Sci. 1999, 54, 5887.
(11) Kronberger, T.; Ortner, A.; Zulehner, W.; Bart, H. J. Numerical
Simulation of Extraction Columns Using a Drop Population Models.
Comput. Chem. Eng. 1995, 19, S639.
(12) Scott, W. T. Analytical Studies of Cloud Droplet Coalescence I. J.
Atmos. Sci. 1968, 25, 54.

(13) Patil, D. P.; Andrews, J. R. G. Analytical Solution to Continuous
Population Balance Model Describing Floc Coalescence and Breakage - A
Special Case. Chem. Eng. Sci. 1998, 53, 599.
(14) Campos, F. B.; Lage, P. L. C. A Numerical Method for Solving
the Transient Multidimensional Population Balance Equation Using an
Euler-Lagrange Formulation. Chem. Eng. Sci. 2003, 58, 2725.
(15) Ramkrishna, D. Population Balances -Theory and applications to
Particulate systems in Engineering, 1st ed.; Academic Press: New York,
2000.
(16) ASTM D-1744. Standard Test Method for Determination of Water
in Liquid Petroleum Products by Karl Fischer Reagent; ASTM: Consho-
hocken, PA, 1992.
(17) ASTM D-3230. Standard Test Method for Salts in Crude Oil
(Electrometric Method); ASTM: Conshohocken, PA, 2000.
(18) ASTM D-664. Standard Test Method for Acid Number of Petroleum
Products By Potentiometric Titration; ASTM: Conshohocken, PA, 2004.
(19) Kumar, S.; Ramkrishna, D. On the Solution of Population Balance
Equations by Discretization - I. A Fixed Pivot Technique. Chem. Eng. Sci.
1996, 51, 1311.
Ind. Eng. Chem. Res., Vol. 47, No. 18, 2008 7103
(20) Jeelani, S. A. K.; Hartland, S. Effect of Dispersion Properties on
the Separation of Batch Liquid-Liquid Dispersions. Ind. Eng. Chem. Res.
1998, 37, 547.
(21) Leveque, R. J. Finite Volume Methods for Hyperbolic Problems,
1st ed.; Cambridge University Press: Cambridge, 2002.
(22) Hoffman, J. D. Numerical Methods for Engineers and Scientists,
2nd ed.; Marcel Dekker: New York, 2001.
(23) Peña, A. A.; Hirasaki, G. J.; Miller, C. A. Chemically Induced
Destabilization of Water-in-Crude Oil Emulsions. Ind. Eng. Chem. Res.
2005, 44, 1139.
(24) Cunha, R. E. P. Modelagem Matemática da Separação Gravitacional
de Emulsões de Petróleo. M.Sc. Dissertation, Universidade Tiradentes,
Brazil, 2007.
ReceiVed for reView March 10, 2008
ReVised manuscript receiVed June 17, 2008
Accepted June 18, 2008
IE800391V