Chemical Engineering Science 104 (2013) 975–987

Contents lists available at ScienceDirect

Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces

OSN Designer, a tool for predicting organic solvent nanofiltration

technology performance using Aspen One, MATLAB and CAPE OPEN
Dimitar Peshev 1, Andrew G. Livingston n
Department of Chemical Engineering, Imperial College London, Exhibition Road, London SW7 2AZ, UK

H I G H L I G H T S G R A P H I C A L A B S T R A C T

The Designer enables process simula-

tion for organic solvent nanofiltration.

Necessity for experimental properties
determination in the initial stage of
OSN process development is removed.

OSN is more energy efficient than
evaporation for concentration of rose-
mary extract.

The thermodynamic non-ideality of
the nanofiltered solution can affect the
overall process performance signifi-
cantly.

art ic l e i nf o a b s t r a c t

Article history: The development of OSN Designer is presented: a software tool, which enables commercially available and
Received 31 July 2013 open source suites of chemical process simulation software for organic solvent nanofiltration (OSN) process
Received in revised form modelling and simulation. A modelling framework, comprising state of the art membrane transport models,
27 September 2013
membrane module flow dynamics, and material balance equations over entire units, was embedded in custom
Accepted 17 October 2013
Available online 24 October 2013
unit operations (UOs) for batch concentration (BC), constant volume semi-batch (CVS), and steady-state (SS)
filtration. The numerical solutions of the resulting systems of equations were programmed in MATLAB and
Keywords: implemented in several process simulation environments such as Aspen Plus, HYSYS and COFE. The computer
OSN Designer aided process engineering open (CAPE OPEN) standards for communication between chemical engineering
Aspen One
software components were used to ensure interoperability amongst MATLAB UOs, process simulation
Matlab
environments and different thermophysical properties servers. Three example simulations using OSN Designer
CAPE OPEN
Process design were carried out in the environment of Aspen Plus with dual aims. Firstly, OSN Designer was validated through
Organic solvent nanofiltration comparison to available experimental data. Secondly, the example simulations addressed important topics such
as energy efficiency of OSN compared to evaporation, use of pore flow or solution diffusion membrane
transport models, and effect of thermodynamic non-ideality on the overall process performance.
& 2013 Elsevier Ltd. All rights reserved.

1. Introduction process development usually involves several stages starting from

OSN is an emerging technology, which has already gained
considerable attention for applications in fine organic synthesis,
the pharmaceutical industry, production of natural biologically
active products, catalyst recycle and solvent recovery. Membrane
n
Corresponding author. Tel.: þ 44 20 7594 5582; fax: þ 44 20 7594 5639.
E-mail addresses: d.peshev@uctm.edu (D. Peshev), a.livingston@imperial.ac.uk
(A.G. Livingston).
1
Present address: Department of Chemical Engineering, University of Chemi-
cal Technology and Metallurgy, 8 Kl. Ohridski Blvd., 1756 Sofia, Bulgaria.
feasibility tests at laboratory scale, passing through pilot plant
tests and finishing with a full production scale process. A reliable
modelling framework and simulation tools are required for a
smooth transition between these stages.
As it is illustrated in Fig. 1, three levels can be distinguished
within the general process modelling framework: molecular
membrane transport, flow dynamics and mass transfer in mem-
brane modules, and process performance.
In the past decade many studies have been published dealing
with OSN membrane transport modelling. Even though attention
has been drawn to the importance of the thermodynamic

0009-2509/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ces.2013.10.033
976 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987
Fig. 1. Levels of modelling used in OSN and numbers of relevant publications in the period 2003–2013.
non-ideality of the solutions on the separation performance at
membrane transport level (Peeva et al., 2004; Stamatialis et al.,
2006), its impact at process level has not been explored. There are
limited studies on the hydrodynamic and mass transfer character-
istics in OSN spiral-wound modules (Santos et al., 2007; Silva et al.,
2010). Finally, there are few attempts to implement these models
to describe process performance, and the reports that do exist
exclusively use simple, non-predictive membrane transport mod-
els (Peshev et al., 2011; Sereewatthanawut et al., 2010; Vanneste
et al., 2013).
The objective of the present work was to streamline OSN process
design by making OSN unit operations (UOs) available in process
modelling environments (PMEs), such as Aspen Plus, HYSYS, ProSim
Plus. This required development of custom OSN UOs for common
membrane separation processes such as batch concentration (BC),
constant volume semi-batch (CVS) and steady state (SS) filtration,
based on state of the art predictive membrane transport models.
Moreover, the aim was to develop a general methodology, which can
provide a flexible tool for process simulation in terms of membrane
transport or process model selection, choice of thermodynamic and
physical property databases and engines, as well as the possibility
to use any molecular structures in the simulations. The latter is
important in the field of OSN.
2. From quantum chemical level to process design –
integration of modelling tools and process simulation
environments
Nowadays, the conceptual design of a new industrial process
usually starts with process simulation using state of the art PMEs. The
latter provide a user interface, where all unit operations comprising
the process flow diagram are connected by material, energy and
information streams. They are mainly designed to simulate steady
state processes by ensuring that the mass and energy balances
between the interacting unit operations are closed at the end of the
simulation. There are two major ways in which the PMEs can be used
for facilitating the prototyping of new OSN processes or optimization
of existing ones. Firstly, they contain libraries with built-in models
commonly used in industrial unit operations. With minor effort, they
can be efficiently combined in order to simulate entire plants. This
removes the necessity for specific model development, which might
otherwise require a high level of proficiency in mathematical
modelling and computer programming from the process developers.
Secondly, they have underlying thermodynamic and physical proper-
ties prediction capability, which removes the need for costly and time
consuming experimental property determination in the initial stages
of process development.
There are numerous commercial and freeware packages for
process simulation, which have proved efficient for design of well-
established processes. However, their applicability to emerging
or non-conventional technologies is relatively unexplored. One
reason is the fact that despite the wide selection of built-in UOs
available in many of the PMEs, they cannot provide any model,
which may be required by the end user. To resolve this issue, most
of the process simulators have embedded tools for development of
custom UOs, which usually employ basic programming languages.
The UOs created are applicable only in the simulation environment
they were created for. All these substantially limit their applica-
tion. A second limitation is imposed by the limited number of
databank components included in the simulation applications.
Even though some commercial data bases such as Aspen Proper-
tiess (AspenTech, 2013) provide access to properties for more than
20 000 pure components, they are still insufficient in some areas
such as fine organic synthesis and the pharmaceutical industry,
where new and very often not-previously-existing compounds
may need to be involved in the simulations.
With the advancement of computational technologies in the
past two decades resulting in a diversity of process simulation,
modelling and thermophysical properties prediction packages,
process engineering software developers have realized the need
for standard interoperability interfaces. This was the incentive for
the EU-funded CAPE-OPEN (CO) project of January 1997, and the
following Global CAPE-OPEN of July 1999, which have developed
common interface specifications in order to ensure interoperabil-
ity for Computer Aided Process Engineering (CAPE) software tools,
which are currently maintained and promoted by the non-for-
profit organization, (CO LaN, 2013). Both the previously mentioned
hurdles to the efficient use of state of the art simulation tools for
design of innovative technologies can be addressed by means of
the CO interoperability standards.
A comprehensive summary of the ways CO can be applied in
engineering practice to plug in or export 3rd party unit operation
models or thermophysical properties can be found elsewhere (Pons,
2010; van Baten and Szczepanski, 2011). When using CO as an
architectural basis for developing a custom unit operation, the UO
 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987
will subsequently be capable for evaluation in any CO compliant
simulator. Two alternatives for development were distinguished. The
first one supposes use of programming languages, such as Microsoft
Component Object Model (COM), commonly used by software
developers to create re-usable software components and use them
to build applications. There are also wizards, which facilitate the use
of other common programming languages such as FORTRAN, Cþ þ,
Visual Basic, etc. However, it is more likely that process engineers are
not familiar with any of the above, but may wish to rely on the so
called “Modellers”, i.e. software tools, which have their own pro-
gramming facilities. More importantly they contain advanced ready
to use toolboxes or functions which can significantly streamline
model development and increase its robustness. Some of the
commercially available Modellers such as gPROMS (from Process
Systems Enterprise Ltd, UK) are readily CO compatible while others
such as Matlab, SciLab and Excel can be CO enabled through the
respective complementary software applications (AmsterCHEM,
2013).
Once integrated in the simulation environment, the unit opera-
tions and the embedded mathematical models usually obtain all
the required thermodynamic and physical properties for the mate-
rial phases involved in the simulation from the so called “property
package”. The latter is specified by the user and represents a
consistent set of compounds, methods, model parameters and pure
component library data sufficient to calculate all the relevant
material properties. As mentioned above, in many cases databanks
of pure components are not readily available in the simulation
environment, or a particular compound of interest is missing. For
this reason most of the process simulation packages also provide
interfaces for user defined components, where a new compound
can be defined by its molecular structure. All of them however rely
on group contribution methods (GCMs) to estimate at least some
of the pure component properties and binary parameters. While
the GCMs are presently the most reliable and therefore the most
widely used methods for properties prediction, they still have
limited applicability. They can be applied only to systems, for
which structure can be described by functional groups with known
interaction parameters. This automatically excludes many aromatic
heterocyclic compounds, transition states, large bio-structures,
isomers, organic salts, etc.
On the other hand, some advanced computational chemistry
methods such as COSMO RS have been proved equally adequate,
and do not have theoretical restrictions on their applicability to
any structures, which can be calculated by quantum chemistry
(COSMOlogic, 2013; Eckert and Klamt, 2002). Another COSMO
based method for prediction of thermodynamic and physical
properties in liquid phase is the COSMO-SAC method (Wang and
Sandler, 2007), which has also been available in Aspen Plus since its
version 12.0 (Ferro et al., 2012). Their quantum chemical basis is
the so called Conductor-like Screening Model (COSMO), available
in many of the commercial molecular simulation software
packages. In addition to the gas phase simulations, some of them
(e.g. Turbomole, Gaussian, DMOL3, GAMESS-US, PQS, Molpro,
Columbus, ORCA, Q-Chem (according to COSMOthermX manual,
version C30_1301)) can also calculate a discrete surface around a
molecule, surrounded by a virtual conductor (dielectric constant
infinity), each segment of which is characterized by its area and
screening charge density (SCD). The liquid is then considered as an
ensemble of closely packed ideally screened molecules, and the
macroscopic thermodynamic properties are predicted based on
statistical averaging on the possible pairwise electrostatic interac-
tions of segments with different SCDs (Eckert and Klamt, 2002;
Wang and Sandler, 2007). This makes it more computationally
efficient and therefore more attractive for process engineering
purposes compared to force field based molecular simulation
methods such as molecular dynamics and Monte Carlo modelling,
977
in which the solute and solvents are treated explicitly. A major
drawback of the COSMO based methods is the user expertise and
complementary specialized software required for the parameteriza-
tion of any non-databank component, as explained in Section 4.3.
Since the present work uses Aspen Plus V7.3 as the simulation
environment, use of COSMO-based methods is further explored.
In order to evaluate the COSMO SAC method on a user defined
component in Aspen Plus, the pure component records have to be
upgraded with its so called s-profile, i.e. a two-dimensional histo-
gram projection of the three-dimensional SCD distribution, as well
as with data for its COSMO molecular volume. Up to version V7.3,
however there is no implemented user interface and models in
Aspen's Property Estimate System (PES) capable of calculating the
s-profile of a specified molecular structure. Therefore, the user must
rely on supplementary software for this purpose. COSMOlogic GmbH
& Co. KG commercialized the COSMOthermX software, which is
based on the COSMO RS method and can calculate s-profiles and
s-potential for particular compounds as well as thermodynamic
and physical properties of pure compounds and their mixtures
(COSMOlogic, 2013). COSMOthermX uses as input the so called
“.cosmo” files exported from the above mentioned molecular
simulation programs and containing information for the SCD dis-
tribution of the structure of interest as well as the total energy of the
ideally screened molecule obtained as a result of quantum mechan-
ical COSMO calculations. A direct way to use COSMO RS predictions,
and particularly activity coefficients, into CO compliant PMEs
(including Aspen Plus V7.3) is through the COSMOthermCO (the
CAPE-OPEN interface of COSMOthermX) program (COSMOlogic,
2013; COSMOthermCO-Manual-C30-1301). However, this implies
that another PME basic property method shall be used in the
simulation and therefore, the COSMO RS method cannot be used
alone. Another hybrid approach was successfully applied in Ferro
et al. (2012) and has been used in the present study to specify the
COSMO SAC method on TOABr (an organic salt). The COSMO RS
s-profiles and molecular volumes for nondatabank components
were used in order to fulfill the minimum input required for the
COSMO SAC property method in Aspen Plus.
Obviously, the CO standards for interfacing process modelling
software components can ensure unconstrained interoperability
amongst CO enabled PMEs, Modellers and thermodynamic and
physical properties servers. Some examples for CO compliant
PMEs are Aspen Plus and Aspen HYSYS (Aspen Tech), gPROMS
(PSE), ProSimPlus (Simulis), PRO/II (SimSci/Invensys), CHEMCAD
(Chemstations), COFE (AmsterCHEM), UnSim Design (Honeywell),
DWSIM (Open Source), etc. However, not all of them implement
CO interface specifications for both UOs and thermodynamic
and physical properties, and comply with all available versions
of the CO standard (versions 0.93, 1.0, 1.1). Amongst the CO
enabled Modellers are Excel (Microsoft), Matlab (MatWorks),
gPROMS (PSE), SciLab (Scilab Enterprises) and ANSYS (ANSYS
Inc.). Compliance of the PME, the property package and the
Modeller (i.e. the UO) with one and the same version of the CO
standard is required for an efficient interoperation. Alternatively, a
component such as COFE (AmsterCHEM), which supports both
CAPE-OPEN version 1.0 and CAPE-OPEN version 1.1 standards, can
be plugged in so as to automatically apply conversions between
the two standards.
3. OSN Designer
3.1. Modeling framework
As shown above, the modelling at process level presumes a
choice of membrane transport model, membrane module flow
978 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987

dynamics and basic unit operation, so as to define the material rate as the permeate stream. The material balance equations over
balance equations over the entire unit. the entire unit, both total and with respect to the individual
species, were used in the model and are shown in Table 1.

3.1.1. Unit operations

For the efficient use of PMEs in the simulation of OSN processes,
availability of all typical unit operations is required. Alongside
steady-state (SS) operation, which is feasible in large scale applica-
tions, very often in smaller scale applications, common for the
pharmaceutical, cosmetics and food industries, batch or semi-batch
nanofiltration processes are favored, due to the reduced capital
costs and easier control. These non-steady state membrane filtra-
tion processes can be classified as constant volume or variable
volume, depending on the ratio between the feed and permeate
volumetric flow rate ratio (Kovács et al., 2009). The variable and
constant volume operations in practice are most often carried out in
batch concentration (BC) and constant volume semi-batch (CVS)
modes respectively, Fig. 2, which are further considered in the
present study.
The BC is a variable volume operation in which the entire
retentate volume is pressurized and decreases over the course of
the process due to the removal of permeate. In the case of CVS
operation there is a feed stream with the same volumetric flow
3.1.2. Membrane transport models
In the past decade researchers tried to accommodate well-
established reverse osmosis models to describe the transport in
OSN membranes, due to the similar fundamentals of these two
processes. All the approaches for mathematical description of
permeation in the membranes can be considered in the light of
the thermodynamics as driving forces of different nature – pressure,
concentration, electrostatic forces, etc., which all contribute to
the chemical potential gradient. The two membrane transport
models which have found broad application to date are the solution
diffusion and pore-flow models (Wijmans and Baker, 1995). The
solution diffusion model presumes that the pressure in the mem-
brane is uniform and equal to that in the feed side and therefore the
chemical potential gradient is expressed only as a concentration
gradient. The pore flow approach discards the concentration gra-
dient and accounts only for the pressure term in the chemical
potential gradient of the solvent (Wijmans and Baker, 1995). Hybrid
approaches, which represent the solution diffusion model with an

Fig. 2. Process diagrams of typical OSN unit operations which have been integrated in the OSN Designer.

Table 1
OSN Designer modeling framework (in all equations: i ¼ 1; 2; j ¼ 1; 2; i a j).

BC CVS SS

^J ^J
 ^J  J^  J^ ¼ 0;
Mass balance dxR;i dxR;i vP
¼ P ½xR;i  xP;i ; ¼ P ðx  xR;i Þ þ ðxR;i  xP;i Þ ; F R P
dt NM R dt NM R vF F;i ^J xF;i  ^J xR;i  ^J xP;i ¼ 0: (5)
  F R P
dNM R dNM R vP
¼  ^J P ; (3) ¼  1 ^J P ;
dt dt vF (4)
t ¼ 0- NM R ¼ V 0;R =v0;R ; xR;i ¼ x0;R;i : t ¼ 0- NM R ¼ V 0;R =v0;R ; xR;i ¼ x0;R;i :
Parameters: V 0;R ; x0;R;i ; t; p Parameters: V 0;R ; x0;R;i ; xF;i ; t; p Parameters: xF;i

Membrane PF SD SIMPLE
transport
λ ¼ d2 =dPore ; Φ ¼ ð1  λÞ ;2 ^J ¼ Area P i ^J ¼ LP Area½ΔP  Δπ ;
i;P
"  # P
vP (8)
ptK d ¼ 1:0  2:30λ þ 1:154λ2 þ 0:224λ3 ; γ ðx ; T; p  ΔPÞ ^J i;P v ΔP
 xi;M  i i;P exp  i ; xP;2 vM ¼ xM;2 vP ð1 RÞ:
2
ptK c ¼ ð2  ΦÞð1:0þ 0:054λ  0:988λ þ 0:441λ Þ; 3
γ i ðxi;M ; T; pÞ ^J i;P þ ^J j;P ^
RT
2
32μK d D12 v2 ðK C þ YÞdPore J^i;P (7)
Y¼ ; Pe ¼ ΔP; xi;P ¼ :
^ 2
RTd 32μK d D12 ^J þ ^J
P i;P j;P
ΦðK C þ YÞ
R ¼ 1 ;
1  ½1  ΦðK C þ Y Þexpð PeÞ
2
J V AreaΔP ef f εd 1 AreaΔP ef f Str Area½ΔP  Δπ
JP ¼ ¼ Pore ¼ ;
vP 32l μ vP μ vP (6)
xP;2 vM ¼ xM;2 vP ð1  RÞ:
Parameters: d2 ; dPore ; ΔP; p; str; φ; Area; T Parameters: P i ; ΔP; p; Area; T Parameters:
" # LP ; φ; R; ΔP; p; Area; T
 
Mass transfer xM;i x x ð^J =AreaÞvP x
¼ R;i  P;i exp P þ P;i :
vM vR vP k vP (9)
 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987
additional viscous term to account for pore transport through the so
called imperfections were also used (Fierro et al., 2012), however at
the expense of an increased number of experimentally determined
model parameters. Additionally, the solution diffusion model with
Maxwell–Stefan multicomponent diffusion theory was used to
predict the performance of polymeric OSN membranes with organic
solvents and their mixtures (Hesse et al., 2013). The latter may be
particularly useful in cases of highly swollen membranes, where the
solvent concentration in the membrane is comparable to the
concentration of the membrane material and no independent
diffusion of the solvent and solute molecules in the membrane
matrix can be assumed. The challenges to its practical application
are related to the estimation of the model parameters (Paul, 2004).
Despite the number of membrane transport models, proposed
in the literature, debate continues on what the best representation
of OSN membrane transport mechanism is, and which models
are more accurate. Most of the studies in the field report a particular
model formulation falling into one of the above mentioned model-
ling approaches, and then test these by comparison to limited
experimental data. Nowadays, there is no membrane transport
model wizard in OSN for use in the process design practice. Its
development requires systematic comparison amongst different
modelling approaches and extensive experimental data. To ensure
meaningful comparison, a uniform level of accuracy is required in
the numerical solutions, model parameters and thermo-physical
properties estimations. Three particular model formulations were
selected in the present work as a reasonable balance between model
complexity (number of parameters) and practical applicability –
steric pore-flow (PF), solution-diffusion (SD) and simple (SIMPLE)
models. Their advantages and limitations are discussed further.
Bowen and Welfoot (2002) suggested a steric pore flow model,
PF, (Table 1) that accounts for diffusive and convective terms in the
solute flux, while the solvent flow was supposed to be due to
pressure drop only. Apart from the purely steric interactions, this
model is not able to account for specific solute–solvent–membrane
molecular interactions, which however were claimed to have a
significant impact on the membrane performance in some cases
(Darvishmanesh et al., 2010). To the knowledge of the authors, the
Bowen and Welfoot model is the only pore-flow model applied for
prediction of both solvent and solute permeation in OSN (Silva and
Livingston, 2006). Input parameters are the membrane structural
parameter, and the effective diameters of the membrane pores and
the solutes. The first two are membrane specific parameters and can
be regressed if limited experimental nanofiltration data are available,
as part of the membrane characterization tests. The solute molecular
diameter can be predicted as a result of molecular simulations or
estimated as a hydraulic diameter based on experimental data for its
diffusivity and the Stokes–Einstein equation. Therefore, it is fully
predictive, but at the expense of its limited applicability due to the
lack of accounting for any solute–solvent–membrane molecular
interactions apart from steric ones. Moreover, as discussed in more
detail in Section 4.2., it is applicable to dilute solutions only.
Alternatively, the solution diffusion model, SD, (Table 1)
accounts for the permeant–membrane interactions through the
experimentally determined permeability coefficients, and can be
derived in a way which accounts for non-ideal behavior of the
filtered solution through the activity coefficients (Peeva et al.,
2004). The permeability coefficients are material characteristics
that can be easily determined during lab-scale feasibility tests and
the activity coefficients can be calculated based on well estab-
lished semi-empirical group contribution methods (UNIFAC,
UNIQUAC, etc.) or predicted based on quantum mechanics and
statistical thermodynamics methods, such as COSMO-RS and
COSMO-SAC. Therefore, for OSN membranes with small enough
free-volume elements, comparable to the molecular size of the
permeants, the solution diffusion model provides a very
979
convenient base for process modelling. A source of limitation for
the applicability of the SD model may be that it does not account
for interaction between the diffusing components (e.g. solute and
solvent) in the membrane matrix.
Due to its simplicity, the approach for constant and known
rejection and solvent permeability (SIMPLE model, Table 1) is
often employed in process modelling based on available experi-
mental data for a particular solute–solvent–membrane system. Its
parameters, solute rejection and solvent permeability, are
assumed constant, which is true only within a narrow range of
the operating parameters around the values for which they were
measured.
The impact of any thermodynamic non-ideality of the solutions
undergoing OSN on the permeate flux can also be accounted for in
the PF and SIMPLE membrane transport models (Eqs. (6) and (8))
through the osmotic pressure difference across the membrane.
Two alternative expressions for the latter (Eqs. (1) and (2)) were
embedded in the OSN Designer.
 
xM;2 xP;2 ^
Δπ ¼ φ  RT ð1Þ
vM vP
^  
RT γ xP;1
Δπ ¼ ln P;1 ð2Þ
v1 γ M;1 xM;1
Eq. (1) is a Van't Hoff like equation, which relies on the
empirical osmotic coefficient, φ. Eq. (2) allows for purely theore-
tical prediction of the osmotic pressure difference in an incom-
pressible fluid as a function of the solvent concentration and
activity coefficients on respective sides of the membrane.
3.1.3. Fluid dynamics and mass transfer in the membrane modules
The flow dynamics in nanofiltration modules and consequently
their mass transfer characteristics are specific for each module
type. For example, in the case of spiral-wound-modules (SWMs),
the flow in the spacer filled channels can be modeled as plug-flow
(Silva et al., 2010), while many of the lab scale flat-sheet mem-
brane dead-end or cross-flow nanofiltration cells, such as those
used in Peshev et al. (2011), Silva and Livingston (2006), Peeva
et al. (2004), Luthra et al. (2002), Tylkowski et al. (2010), Peev et al.
(2011), Sereewatthanawut et al. (2011) are best described as
perfectly mixed vessels. In the present study, the mass balance
equations were derived based on the assumption of ideally mixed
liquids in the modules, which ensures their general applicability.
Moreover, Silva et al. (2010) have shown that for a typical OSN
application, any axial pressure and concentration gradients in a
single spiral-wound module are negligible and do not affect its
performance. The same speculations are also relevant to tubular
and hollow fiber modules providing sufficiently high retentate
flow rates are applied in these cases in order to minimize the
effects of concentration polarization. Therefore, OSN Designer is
also applicable to tubular and hollow fiber modules if the mass
transfer coefficient is known. Accordingly, the effects of pressure,
velocity and concentration gradients in the main flow direction on
the membrane module performances were ignored. The retentate
concentration throughout the module was assumed to be equal to
the outlet value.
The effect of the formation of a concentration boundary layer
(so called concentration polarization) was accounted for by Eq. (9)
(Table 1), derived based on the two-film theory for the mass
transfer (Peeva et al., 2004), when the mass transfer coefficient
is known.
The governing system of equations for a defined UO comprises
a membrane transport model, a mass transfer equation and a mass
balance equation over the respective unit. They form a non-linear
system of differential algebraic equations (DAEs) for the non-steady
state operations (BC and CVS), or algebraic equations for the SS
980 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987
operation. To ensure their reliable solution, advanced numerical
methods are required. In the present work MATLAB has been
selected as a “Modeller”. It has been recognized as one of the most
common and powerful state of the art software tools for technical
computing (Assi, 2011). On the other hand, parameters such as
activity coefficients, molar volumes, binary coefficients of molecular
diffusivity and dynamic viscosity, are functions of the calculated
concentrations. Therefore access of the numerical method to ther-
mophysical properties data bases or computational routines is
required. In the present work, the necessary environment was
provided by process simulation packages such as Aspen Plus, Aspen
HYSIS and COFE through the CO interface standards.
The OSN unit can be part of a complete process flowsheet.
Therefore, during solution the model may have to be evaluated
at initial conditions far away from the solution. Additionally, if the
model is not robust enough, the UO may request thermodynamic
properties at conditions such as mole fractions which do not sum-
up to unity for a single phase, negative concentrations, tempera-
tures or pressures. These would all cause failure in the simulation.
Non-unique solutions are possible too. Therefore care was taken in
order to accurately define the initial conditions and to ensure that
all functions accessing thermodynamic and physical properties
from the property package are evaluated at feasible conditions.
3.2. Integration of OSN Matlab UOs in AspenONE and COFE
The OSN Designer (c.f. Fig. 3) is implemented using the
modelling framework described in Section 3.1 in different PMEs
using MATLAB as a modelling tool and variety of thermophysical
properties engines and databases in accordance with the meth-
odologies presented in Section 2. To the author's knowledge, this is
the first work reporting implementation of MATLAB models in the
Aspen Plus environment. They were solved using the nonlinear
solver fsolve and the implicit method ode15s in Matlab. In the case
of the non-steady state operations, the initial conditions for ode15s
were computed as solutions of the steady state at initial conditions
using fsolve. As illustrated in Fig. 3, the created UOs were plugged
in the environment of three different PMEs and tested with
property packages provided by four thermophysical properties
servers using the CO interface standards. The results obtained
when using different PMEs were consistent. However, Aspen Plus
was employed in all example simulations discussed in this work.
In the non-steady unit operations, the feed, permeate and
retentate streams are time averaged with respect to mass so that
their molar flow rates and fractions were defined as follows:
Z operation with fresh extract as diluent.
1 t
JP ¼ J ðtÞdt
t 0 P
Rt
J ðtÞxP;i dt
xP;i ¼ 0R tP
0 J P ðtÞdt
Fig. 3. A schematic representation of the process simulation elements used in the OSN Designer and their interactions.
xR;i ¼ xEND
R;i
J R ¼ NM END =t ð10Þ
R
This effectively represents the situation in a process diagram
in which the membrane module is connected to a holding tank,
which facilitates the interfacing of the non-steady UOs with the
steady-state flowsheet environments.
Any user defined input model parameters can be set or any
output parameters can be monitored through the graphical user
interface of the UO.
4. Example simulations
4.1. The SIMPLE model. Evaporation versus OSN
Several recent publications claimed that OSN can be superior to
conventional evaporation in processes involving thermally suscep-
tible products or high boiling solvents (Peshev et al., 2011; Geens
et al., 2007; Vandezande et al., 2008; Darvishmanesh et al., 2011).
Some studies have focused on exploring the possibilities for
integration of OSN into existing technologies for herbal extracts
processing (Peshev et al., 2011; Tylkowski et al., 2010; Peev et al.,
2011; Sereewatthanawut et al., 2011). In all these applications,
the potential benefits from OSN implementation were seen in
terms of milder operating conditions, extracts fractionation or
selective removal of undesired components, costs reduction, etc.
Nevertheless, there is a lack of studies providing quantitative
measures for these benefits. Darvishmanesh et al. (2011) make a
comparison between the energy consumption during steady-state
solvent recovery by means of evaporation and OSN using Aspen
Plus. However, the simulation was performed using single set of
parameters, and did not investigate the impact of the process
parameters on the economic viability of the two alternatives. The
possibility for utilization of the heat of condensation was not
considered. A very simple modelling approach was embedded in
the nanofiltration unit operation, which did not permit the effect
of the process parameters to be accounted.
In contrast, the OSN Designer implements predictive mem-
brane transport models, and each of them can be employed
depending on the available data for the system studied. It also
can efficiently simulate non-steady state OSN UOs in combination
with steady-state built-in operations. To illustrate this we have
simulated the process proposed by Peshev et al. (2011) for
concentration of antioxidant ethanolic rosemary extracts by CVS
The Aspen Plus flow diagrams for the two analyzed processes
are shown in Fig. 4. The nanofiltration diagram comprises a high
pressure pump (PUMP1), which operates at the same volumetric
flow rate as the permeate stream. A second pump recirculates
 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987
Fig. 4. Aspen Plus process diagrams for concentration of ethanolic rosemary extracts by semi-batch constant volume diafiltration (a) and evaporation under reduced
pressure (b).
the retentate solution through the OSN module. Therefore, the
required net work of the pump (W2) will depend on the volu-
metric flow rate and the pressure drop in the module. For a typical
2.5 in. spiral-wound module a reasonable estimate for the flow
rate and pressure drop per unit area are 600 L h  1 and 0.25 bar m2
respectively (Silva et al., 2010; Ohya and Taniguchi, 1975). These
values were used in all simulations.
The evaporation process (Fig. 4b) must be carried out at reduced
pressure to avoid thermal degradation of the active components
in the extract. The maximum allowable temperature was 30 1C,
which implies absolute pressure as low as 0.1 bar. The degree of
concentration was controlled by the value of the vapour fraction in
the evaporator block. The vapors leaving the evaporator were
condensed at reduced pressure, created by the vacuum pump. The
latter was simulated by a compressor block available in Aspen
Plus. The feed stream for the vacuum pump, AIRIN, was predefined
by a custom Matlab UO. AIRIN is composed of air and has molar
flow rate equal to 1% of the flow rate of the VAPOUR stream
leaving the evaporator, which is an estimation of the amount of
the non-condensed gasses and vapours as well as any leaks in the
system. For each run, the heat duties of the evaporator (Q1),
condenser (Q2) and the required net work of the vacuum pump
(WVAC) were recorded and used to evaluate the specific energy
consumption in the system (SEC). In both diagrams, all pump
efficiencies were set to 0.7.
A challenge to the implementation of multicomponent mix-
tures such as natural extracts in simulation environments is the
fact that the exact composition is not always known. However,
detailed data for the structure of each component is required as to
define the respective property package and method. In Peshev
et al. (2011) a modelling solution containing rosmarinic acid has
been used as a substitute for real extract in order to streamline the
selection of the most appropriate membrane. Moreover, it has
been illustrated that in each nanofiltration fraction the RA to dry
solids ratio (w/w) remains constant. Accordingly, there is a unique
correlation between dry solids content and rosemarinic acid in
the extract fractions and also between dry solids content in the
extracts and their antioxidant activity. In the present work this fact
was exploited by the assumption that the rosemary extract
consists of pure rosmarinic acid with the same mass concentra-
tion. The effective osmotic coefficient, required to evaluate the
SIMPLE model was calculated from Eq. (8) (Table 1), using a single
pair of experimental data for permeate flux and DS concentration
981
Table 2
Process and model parameters for rosemary extract concentration.
Lp, φ, – ΔP, p, T, K R, – k, CF, RA/DS Membrane
m Pa  1 s  1 bar bar m s  1 kg m3 ratio, –
1.34  10  12 3.69 40 41 298 0.992 inf 11 0.0107 Duramem
200
in the retentate during extract nanofiltration. All other parameters
used in the simulations are taken from Peshev et al. (2011), and are
shown in Table 2.
Rosmarinic acid was introduced as a user defined component in
Aspen Plus using its molecular structure. Further, its physical and
thermodynamic properties were estimated by NIST's ThermoData
Engine embedded in the simulation environment. The property
method and models set used was RK-ASPEN. All missing para-
meters were simultaneously estimated by Aspen Plus's property
estimate system during the first simulation run.
Fig. 5 illustrates that OSN Designer with the SIMPLE membrane
transport model has excellent predictive power for both the experi-
mental data and the analytical solution for the retentate concentra-
tion of RA reported in Peshev et al. (2011). However, the simple
analytical solution is based on an assumption of constant permeate
flux and is not capable of describing the experimentally observed
(and predicted by the OSN Designer) flux decline with increasing the
time of operation (respectively retentate concentration).
The overall cost of a technology is comprised of capital and
operation costs. The former requires detailed specification of the
equipment used and was not investigated in the present study.
The energy consumption is a major component of the operating
costs, with increasing importance in recent years. Therefore the
comparison between the two operations was made based on the
specific energy consumption defined as
Power ðheat and=or mechanicalÞ; W
SEC ¼ ; J m3 ð11Þ
Feed flow rate; m3 s  1
This definition of SEC makes it independent of the feed flow
rate, initial feed volume and membrane area. It is determined
solely by the degree of concentration achieved (vapor fraction or
number of diafiltration volumes permeated).
982 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987
Fig. 5. Data for retentate concentration of rosemarinic acid and instantaneous
permeate flux during single stage constant volume diafiltration with fresh extract
as diluent – experimental data, analytical solution and OSN Designer predictions.
Fig. 6. Comparison of the energy consumption for processing of fresh extract at
specified dry solids content in the concentrate/retentate (CR,DS ¼ 32.5 kg m  3),
using evaporation and OSN.
Obviously the SEC with evaporation is an order of magnitude
higher than when nanofiltration is considered (Fig. 6). Even though
in practice energy of condensation is often utilized, recovery is never
complete while the efficiency strongly depends on the operating
conditions. In this particular case, utilization of the heat of con-
densation was not considered because of the low temperature of the
vapours (30 1C).
It is clear from Fig. 7 that in the range of dry solids content from
11 to around 130 kg m  3 the OSN process is efficient. However,
to achieve a higher degree of concentration using only OSN is
infeasible due to the sharp increase in the required membrane
area for processing of 100 L h  1 fresh extract – Fig. 7. This can be
explained by the increasing osmotic pressure difference across
the membrane. The above calculations are valid only if no crystal-
lization occurs in the retentate. Due to solubility issues, any of the
extract constituents may crystallize before the osmotic equili-
brium is reached. If further concentration is required, then hybrid
OSN/crystallization or OSN/evaporation scheme may be more
efficient.
Evidently the SIMPLE model in its formulation, embedded in
the OSN Designer, can produce reliable results when used for
Fig. 7. Specific energy consumption for production of concentrate with dry solids
concentration CR,DS and the corresponding required membrane area for processing
of 100 L h  1 of fresh extract.
extrapolation of existing experimental data. Laboratory scale data
for the system investigated can be used for process design and
optimization.
4.2. Solution diffusion or pore-flow?
While the constant rejection approach can be useful in some
cases as illustrated in Section 4.1., in other processes, particularly
non-steady state ones with significant variations in the concentra-
tion, transmembrane pressure, temperature, etc., a predictive
model for rejection and flux is required. As discussed earlier, the
choice between PF and SD is still controversial. To the knowledge
of the authors, there is only one study (Silva and Livingston, 2010),
which directly compares the capability of PF and SD models to
describe the membrane transport of solvent–solute mixtures in
the OSN.
In this study, the ability of each model to predict flux and rejection
for a mixture of dimethyl methylsuccinate (DMMS) in methanol
was evaluated.
A comparison of the OSN Designer predictions for rejection
and flux with the calculated and experimental data available in
Silva and Livingston (2006) was carried out. The experimental data
were obtained in a steady state cross-flow system and modeled by
means of the solution diffusion (SD) (Peeva et al., 2004) and pore
flow (PF) (Bowen and Welfoot, 2002) models for the membrane
transport mechanism. Thus, the steady-state UO was used with
both SD and PF models to generate our predictions. So as to ensure
accurate comparisons, the same values for the process parameters
and mass transfer coefficients as in Silva and Livingston (2006)
were assumed. The latter are listed in Table 3.
Since DMMS is not included in the Aspen databanks, it was
defined as a new component, by setting its structure in accordance
with the so called “general method”, which requires only the
structural formula of the compound as an input. All its physical,
transport and thermodynamic properties were estimated using
the Aspen Plus Property Estimate System based on the Dortmund
Modified UNIFAC Method (DMD-UNIF).
Figs. 8 and 9 show predictions on permeate flux and rejection
respectively, normalized to the experimental values. It is clear that
there is good agreement between the predictions made by means of
OSN Designer and the calculated and the experimental data in Silva
and Livingston (2006) at low solute concentrations and for both
 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987
Table 3
Process and model parameters for nanofiltration of DMMS–methanol mixture.
Operating conditions and model parameters
ΔP, bar p, bar T, K P1, mol m  2 s  1 P2, mol m  2 s  1
30 31 303 17.5 0.21
Mass transfer coefficient as a function of retentate concentration of DMMS
CR, wt% 5 7.5 12.3 13.8
k  105, m s  1 1.49 1.46 1.39 1.37
Fig. 8. Simulated permeate fluxes, normalized with respect to the experimental data.
Fig. 9. Simulated rejections, normalized with respect to the experimental data.
membrane transport models. This proves the adequacy of the
numerical routines and more generally the feasibility of the model-
ling framework. However when using the PF model at retentate
concentrations higher than 28 wt%, there is significant discrepancy
between all the modelling predictions and the experimental data.
This discrepancy can be explained by the fact that in the Hagen–
Poiseuille equation used for the volumetric permeate flux in the OSN
Designer the liquid is considered pure solvent (with viscosity of
pure solvent). This is an approximation, since in reality there is a
concentration gradient along the pore, and as a result a viscosity
gradient too. In order to obtain the analytical solution for the
rejection (Bowen and Welfoot, 2002) an assumption for constant
viscosity along the pore is required. Moreover, the Stokes–Einstein
(used in Silva and Livingston (2006)) and Wilke–Chang (used by
Aspen's PES) equations are intended to estimate the diffusivity
coefficients at infinite dilution and may not be reliable at solute
concentrations higher than 28 wt%. Therefore we believe that in its
present form the PF model should be used with dilute solutions
983
Str, m rp, m r2, m Membrane
5.3  10  15 9.1  10  10 4.4  10  10 Starmem 122
15.5 28 33.2
1.33 1.12 1.04
only. Thus, in the OSN Designer all the transport properties inside
the membrane pores were estimated assuming pure solvent or
infinite dilution. It is not clear what assumption has been made in
this respect in Silva and Livingston (2006). In contrast the SD model
disregards any convective flow and in its form embedded in the OSN
Designer, does not explicitly require diffusivities. This makes it
more adequate at higher retentate solute concentrations as evident
from Figs. 8 and 9. However, experimental determination of the
permeant-membrane specific permeability coefficient is necessary.
The small deviations when using the SD model can be attributed to
the fact that in Silva and Livingston (2006) the required physical and
transport properties were determined empirically or taken from the
literature, while in the Aspen unit operations all these properties are
estimated based on the molecular structures using the group
contribution methods and semiempirical models embedded in
Aspen's PES.
4.3. Effect of the thermodynamic nonideality on the process
performance
Shortly after the appearance of organic solvent nanofiltration
membranes, their potential use in the homogeneous phase trans-
fer catalyst recovery from the post-reaction organic phase was
explored (Luthra et al., 2002). Since phase transfer catalytic
reactions are most often carried out in batch mode, downstream
processing by nanofiltration would be most efficient in batch or
semi-batch concentration mode. The appropriate membrane shall
completely retain the catalyst and allow the reaction products to
pass through unhindered. Luthra et al. (2002) has demonstrated
that Starmem™ 122 membrane has good performance in recover-
ing Tetraoctylammonium bromide (TOABr) as a phase transfer
catalyst from reaction mixture in toluene. A following study on the
same model system has shown that the solution diffusion model
can be used to predict membrane transport (Peeva et al., 2004).
The binary mixture of TOABr and toluene has also shown osmotic
pressure significantly lower than that predicted by the van't Hoff
equation during nanofiltration, which has been explained by its
thermodynamic nonideality (Peeva et al., 2004; Stamatialis et al.,
2006).
From engineering point of view it is important to make a
motivated choice of operating mode, as well as to estimate the
effect of the thermodynamic nonideality on the overall process
performance. Knowledge of these effects can help to indicate
the necessary level of model complexity to be used, as well as to
explore the ways to benefit from them in the process design.
While TOABr is available in the NISTV73 database of Aspen,
its records are insufficient for our purposes. It was inefficient to
use the Aspen's property estimate system to estimate the missing
parameters. Most of the activity coefficient methods rely on group
contribution approaches, which cannot handle ionic compounds
such as TOABr. An alternative property method available in Aspen
Plus, which relies on theoretical quantum chemical and statistical
methods and therefore can handle any type of molecular structure,
is the COSMO SAC method. The COSMO SAC model for both TOABr
984 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987
and Toluene was specified by their s-profiles and molecular
volumes. They were computed using the COSMO-RS method,
implemented in the COSMOthermX program. Toluene was avail-
able in the TZVP-parametrized C30_1301 database developed by
COSMOlogic and supplied with the COSMOthermX software. So as
to parametrize TOABr at the same computational level, a simula-
tion sequence was developed. The initial structure was optimized
as an ion pair in vacuum using the MM2 molecular mechanics
method available in ChemBio 3D software. The optimized struc-
ture was further imported into a Gaussian program and subjected
to a twofold optimization – first at PM6 semi-empirical level in
vacuum and finally at BVP86/TZVP/DGA1 level considering solvent
effects through the conductor-like polarizable continuum model
taking water as a solvent. The optimized structure was used for
COSMO calculation in order to obtain the charge screening on the
molecular surface and the respective.cosmo file. This calculation
was performed in Gaussian, using the BVP86 density functional,
the TZVP basis set, DGA1 fitting set and the SCRF COSMORS
method. In this case, the TOABr structure was parameterized at a
computational level compatible with the TZVP C30_1301 database.
The created.cosmo file contains all the required information for
evaluation of the COSMO-RS method on TOABR in COSMOthermX.
For TOABr, in addition to the s-profile and molecular volume, the
predicted using COSMOthermX normal boiling point as well as its
molecular weight were required to fulfill the minimum input for
Aspen's COSMO SAC method. The calculated s-profile and the
s-surface for the TOABr are shown in Fig. 10.
All other properties of the pure components and mixtures used
in the simulations were estimated by Aspen's PES.
As is pointed out in the explanatory notes provided by COSMO-
logic GmbH (COSMOlogic, 2013), there are three approaches to
simulate ionic compounds using COSMOtherm and namely assume
an ion pair as a single molecule, characterized by a single.cosmo file;
combine the.cosmo files of both the cation and anion in a meta-file;
consider distinct ions with equal concentrations in the system
having separate.cosmo files. Amongst these, as described above we
used the first alternative when implementing TOABr in Aspen Plus
environment. The molecular model of an ion paired structure has
already been successfully used for modelling of ionic liquids using
COSMORS methodology (Navas et al., 2011). However, in this case,
the structure of the ionic pair is fixed and despite it representing the
most stable conformation, it may not adequately represent the
electrostatic interactions and the size, shape and free volume in
the real liquid system. This will reflect the accuracy in the prediction
for the residual and combinatorial terms of the activity coefficients
respectively. Using a meta-file with the individual s-profiles and the
s-surfaces for both ions would take into account interactions of two
distinct ions, but the combinatorial term of the pair will be used.
In spite of usually being much smaller than the residual term, the
Fig. 10. TOABr sigma surface, (a), and profile, (b), used for parametrization of the Aspen Properties COSMO SAC method.
combinatorial term may also have a significant contribution to the
activity coefficient as shown in Soares (2011), and needs attention.
Therefore in the present work the.cosmo files for the tetraoctylam-
monium cation and bromide anion were also generated as in the
case of the ionic pair. They were used to compute the activity
coefficient profiles in their ternary mixture with toluene using
COSMOthermX software. The resulting activity coefficient profiles
were then used with the SD membrane transport model of the OSN
Designer.
The data in Fig. 11 demonstrate that the COSMORS method with
an assumption of distinct ions most accurately describe the experi-
mental data for the toluene activity coefficients at freezing point.
However, there is a slight deviation from the predictions reported by
Peeva et al. (2004) based on the modified UNIFAC method. Therefore
both profiles were used in the following simulations.
Fig. 12 shows a comparison of the steady state predicted by
using the UO of the OSN Designer with experimental and calcu-
lated data published in Peeva et al. (2004). The permeability
coefficients and mass transfer coefficient, necessary to evaluate
the steady-state UO in OSN Designer are taken from Peeva et al.
(2004) and listed in Table 4.
Due to the almost complete observed and predicted rejections for
TOABr (Fig. 13), the contribution of the TOABr permeate flux
equation (Eq. (7); Table 1) and therefore its activity coefficients
can be neglected. In accordance with this, both permeate and
retentate side TOABr activity coefficients can reasonably be approxi-
mated by unity. Moreover, since the solute concentration in the
permeate is negligible, the toluene activity coefficient there can also
be approximated by unity. Therefore, the only activity coefficient
component which can potentially influence the process performance
Fig. 11. Activity coefficient profiles for toluene in binary mixtures with TOABr –
data from Peeva et al. (2004) and calculated using the COSMO-RS method.
 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987 985

time required to achieve a certain concentration or volume

reduction. The lowest and highest TOABr concentrations (0.05 M
and 0.7 M) in the following simulations correspond to typical
concentrations in the post-reaction mixture and saturation in
toluene respectively (Luthra et al., 2002).
Fig. 14 shows the predicted process time for concentration of a
typical post-reaction mixture up to near saturated with respect to
TOABr (0.65 M), using CVS. Three cases were assumed respectively
for the activity coefficients: unity; calculated by modified UNIFAC;
and COSMORS methods. Obviously, the thermodynamic non-
ideality has a significant effect, resulting in almost half the time
for processing. It is also seen that despite the slightly different
profiles for the permeate fluxes, shown in Fig. 12, when using one
or the other activity coefficient method, the effect on the overall
process performance is negligible. Both methods for activity coeffi-
cient prediction give similar results at process level.
An important aspect of the process design is the choice of
Fig. 12. Experimental data and model predictions for the permeate flux during
operational mode. The following simulation results demonstrate
steady state nanofiltration of 0.21 M (0.023 mole fraction) solution of TOABr in
toluene. SD based process model used for all simulations. the potential benefits from running the process in batch concen-
tration or constant volume semi-batch mode. Fig. 15 shows a
comparison of the predicted time necessary to obtain a specific
Table 4 degree of concentration of 0.05 M diluted solution of TOABr in
Process and model parameters for nanofiltration of DMMS–toluene mixture. toluene while using both BC and CVS UOs. The data demonstrate
that using BC instead of CVS can result in a significant saving of
P1, P2, T, K ΔP, p, k, m s  1 Area, Membrane
time only for very high concentration in a single stage. However,
mol m  2 s  1 mol m  2 s  1 bar bar m2
due to the 13 times decrease of the volume during a batch process,
1.1 3  10  5 303 30 31 1.1  10  5 1 Starmem a large initial volume needs to be fed and pressurized in the
122 system, which in turn may significantly increase the capital costs
for equipment. From that point of view the CVS process may be
more feasible, particularly when larger volumes have to be
processed in a production cycle.
The above discussion clearly demonstrates that the thermo-
dynamic nonideality of the solutions can have a significant effect
on the process performance. Special attention shall be given to any
nanofiltration systems involving solutions of organic salts where
due to the presence of strong electrostatic interactions, deviations
from the ideal solution lows are common. The motivated choice of
operational mode can be crucial for the efficiency.

5. Conclusions

OSN Designer is a process modelling and simulation tool, which

can streamline OSN process design. It implements a state of the art

Fig. 13. Experimental data and model predictions for the rejection during steady
state nanofiltration of a 0.21 M (0.023 mole fraction) solution of TOABr in toluene.
SD based process model used for all simulations.

is the toluene activity coefficient in the retentate. As expected, the

OSN Designer prediction with modified UNIFAC activity coefficients
coincides with the modelling predictions and experimental data
published in Peeva et al. (2004). Using COSMORS activity coefficients
results in slightly lower and higher permeate fluxes as compared to
the experimental ones at low and high transmembrane pressures
respectively. The assumption of all activity coefficients being unity
results in general underestimation of the permeate flux.
It is interesting to evaluate the impact of these thermodynamic
non-ideality effects on the overall performance of a non-steady
state process. The filtration time is a substantial part of the cycle of
a batch or semi-batch production process, and thus an important Fig. 14. Simulated process time required to produce 10 L of 0.65 M solution of
factor which has an impact on the production cost. Accordingly, TOABr in toluene by means of CVS, starting with 0.05 M solution and using 1 m2 of
the results on TOABr concentration were represented in terms of membrane area at 30 bar.
986 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987

C concentration (kg m  3)
D12 diffusivity of solute in solvent (m2 s  1)
d diameter (m)
J total volumetric flux (m3 m  2 s  1)
^J molar flow rate (mol s  1)
Kc hindrance factor for convection (dimensionless)
Kd hindrance factor for diffusion (dimensionless)
k mass transfer coefficient (m s  1)
LP permeability (m3 m  2 s  1 Pa  1)
l membrane thickness (m)
N number of permeated diafiltration volumes (dimension-
Rt
less); N ¼ 0 ^J P vP dt=V 0;R
NM number of moles, (mol)
P membrane permeability (mol m  2 s  1)
ΔP transmembrane pressure (Pa)
Fig. 15. Process time and retentate concentration of TOABr at different times Pe modified Peclet number (dimensionless)
volume reduction (BC) or diafiltration volumes permeated (CVS). In all cases: initial p applied pressure (Pa)
concentration  0.05 M; concentrate volume – 10 L; model parameters – Table 4. R rejection coefficient (dimensionless)
R^ ideal gas constant (J mol  1 K  1)
2
modelling framework in different PMEs. All numerical routines were Str membrane structural parameter (m); Str ¼ ðεdPore =321Þ
programmed in MATLAB and evaluated in several PMEs such as Aspen T temperature (K)
Plus, HYSYS and COFE. Practically unconstrained exchange of data was t time (s)
achieved between different data bases and thermophysical engines V volume (m3)
such as Aspen Properties, HYSYS COM Thermo, COSMOthermCO and v molar volume (m3 mol  1)
TEA(CAPE-OPEN 1.x). This removes the necessity for experimental x mole fraction (dimensionless)
properties determination in the initial stage of process development. Y solute function defined by Eq. (6) (dimensionless)
OSN UOs for batch concentration, constant volume semi-batch, and
steady state filtration were embedded in the OSN Designer. The Greek symbols
developed software and methodologies provided flexibility in terms
of membrane transport or process models selection, choice of thermo- γ activity coefficient (dimensionless)
dynamic and physical property databases and engines, as well as the ε porosity (dimensionless)
possibility to use any molecular structures in the simulations. λ ratio of solute radius to pore radius (dimensionless)
The results from three example simulations based on the OSN μ viscosity (Pa s)
Designer tool in the environment of Aspen Plus addressed the Φ solute steric partition coefficient into a cylindrical pore
following important topics: (dimensionless)
φ osmotic coefficient (dimensionless)

energy efficiency of OSN compared to evaporation. The OSN CVS
process for concentration of ethanolic antioxidant rosemary
Subscripts
extracts can result in considerable energy savings compared to
conventional evaporation under reduced pressure. However, for
0 initial
dry solids content higher than a certain limit, due to thermo-
1 solvent
dynamic restrictions on the nanofiltration process, a hybrid OSN/
2 solute
crystallization or OSN/evaporation scheme was suggested,

eff effective
use of PF or SD membrane transport model. Steady state nanofil-
F feed
tration simulations for a mixture of DMMS in methanol have
M feed side liquid phase, at the membrane–liquid interface
demonstrated that the PF model is applicable in diluted solutions
P permeate
only. The SD model is also adequate at higher retentate solute
Pore pore
concentrations,

R retentate; feed side liquid phase, in the bulk
effect of the thermodynamic non-ideality on the overall pro-
cess performance. The simulation results for BC and CVS
concentration for recovery of TOABr phase transfer catalyst in Superscripts
toluene have revealed that the thermodynamic non-ideality of
the binary mixture has a significant effect, resulting in almost END at the end
half the time for processing, and should be taken into account t at time t
during process design.
Abbreviations
In all three case studies there was sufficient agreement
between OSN Designer predictions and experimental data. BC batch concentration
CAPE computer aided process engineering
CO CAPE-OPEN
Nomenclature COSMO conductor-like screening model
CVS constant volume semi-batch
List of symbols DAE differential algebraic equations
DMMS dimethyl methylsuccinate
Area membrane area (m2) DS dry solids
 D. Peshev, A.G. Livingston / Chemical Engineering Science 104 (2013) 975–987 987

GCM group contribution methods Kovács, Z., Discacciati, M., Samhaber, W., 2009. Modeling of batch and semi-batch
NIST National institute of standards and technology membrane filtration processes. J. Membr. Sci. 327, 164–173.
Luthra, S.S., Yang, X., Freitas dos Santos, L.M., White, L.S., Livingston, A.G., 2002.
OSN organic solvent nanofiltration Homogeneous phase transfer catalyst recovery and re-use using solvent
PES property estimate system resistant membranes. J. Membr. Sci. 201, 65–75.
PF pore flow Navas, A., Ortega, J., Palomar, J., Diaz, C., Vreekamp, R., 2011. COSMO-RS analysis on
mixing properties obtained for the systems 1-butyl-X-methylpyridinium tetra-
PME process modelling environment
fluoroborate [X¼ 2,3,4] and 1,ω-dibromoalkanes [ω ¼1–6]. Phys. Chem. Chem.
RA rosmarinic acid Phys. 13, 7751–7759.
SCD screening charge density Ohya, H., Taniguchi, Y., 1975. An analysis of reverse osmotic characteristics of ROGA-
SD solution diffusion 400 spiral-wound modules. Desalination 16, 359–373.
Paul, D.R., 2004. Reformulation of the solution–diffusion theory of reverse osmosis.
SEC specific energy consumption J. Membr. Sci. 241, 371–386.
SS steady-state Peev, G., Penchev, P., Peshev, D., Angelov, G., 2011. Solvent extraction of rosmarinic
TOABr tetraoctylammonium bromide acid from lemon balm and concentration of extracts by nanofiltration: effect of
UO unit operation plant pre-treatment by supercritical carbon dioxide. Chem. Eng. Res. Des. 89,
2236–2243.
Peeva, L.G., Gibbins, E., Luthra, S.S., White, L.S., Stateva, R.P., Livingston, A.G., 2004.
Effect of concentration polarisation and osmotic pressure on flux in organic
solvent nanofiltration. J. Membr. Sci. 236, 121–136.
Acknowledgement Peshev, D., Peeva, L.G., Peev, G., Baptista, I.I.R., Boam, A.T., 2011. Application of
organic solvent nanofiltration for concentration of antioxidant extracts of
rosemary (Rosmarinus officiallis L.). Chem. Eng. Res. Des. 89, 318–327.
The research leading to these results has received funding from Pons, M., 2010. How to make use of CAPE-OPEN? AIChE Annual Meeting, Salt Lake
the European Community's Seventh Framework Programme under City, Utah.
grant agreement MemTide 238291 FP7-PEOPLE-ITN-2008. Santos, J.L.C., Geraldes, V., Velizarov, S., Crespo, J.G., 2007. Investigation of flow
patterns and mass transfer in membrane module channels filled with flow-
aligned spacers using computational fluid dynamics (CFD). J. Membr. Sci. 305,
References 103–117.
Sereewatthanawut, I., Baptista, I.I.R., Boam, A.T., Hodgson, A., Livingston, A.G., 2011.
Nanofiltration process for the nutritional enrichment and refining of rice bran
AmsterCHEM, 2013. 〈http://www.amsterchem.com/〉 (accessed 13.06.13.). oil. J. Food Eng. 102, 16–24.
AspenTech, 2013. 〈http://www.aspentech.com/products/aspen-properties.aspx〉 Sereewatthanawut, I., Lim, F.W., Bhole, Y.S., Ormerod, D., Horvath, A., Boam, A.T.,
(accessed 13.06.13.).
Livingston, A.G., 2010. Demonstration of molecular purification in polar aprotic
Assi, A.H. (Ed.), 2011. Engineering Education and Research Using MATLAB. InTech,
solvents by organic solvent nanofiltration. Org. Process Res. Dev. 14, 600–611.
Rijeka, Croatia.
Silva, P., Livingston, A.G., 2006. Effect of solute concentration and mass transfer
Bowen, W.R., Welfoot, J.S., 2002. Modelling the performance of membrane
limitations on transport in organic solvent nanofiltration – partially rejected
nanofiltration – critical assessment and model development. Chem. Eng. Sci.
solute. J. Membr. Sci. 280, 889–898.
57, 1121–1137.
Silva, P., Peeva, L.G., Livingston, A.G., 2010. Organic solvent nanofiltration (OSN)
CO LaN, 2013. 〈http://www.colan.org/〉 (accessed 13.06.13.).
with spiral-wound membrane elements – highly rejected solute system.
COSMOlogic, 2013. 〈http://www.cosmologic.de〉 (accessed 13.06.13.).
J. Membr. Sci. 349, 167–174.
Darvishmanesh, S., Degreve, J., Van der Bruggen, B., 2010. Mechanisms of solute
Soares, R.P., 2011. The combinatorial term for COSMO-based activity coefficient
rejection in solvent resistant nanofiltration: the effect of solvent on solute
rejection. Phys. Chem. Chem. Phys. 12, 13333–13342. models. Ind. Eng. Chem. Res. 50, 3060–3063.
Darvishmanesh, S., Firoozpour, L., Vanneste, J., Luis, P., Degr'eve, J., Van Der Stamatialis, D.F., Stafie, N., Buadu, K., Hempenius, M., Wessling, M., 2006. Observa-
Bruggen, B., 2011. Performance of solvent resistant nanofiltration membranes tions on the permeation performance of solvent resistant nanofiltration
for purification of residual solvent in the pharmaceutical industry: experiments membranes. J. Membr. Sci. 279, 424–433.
and simulation. Green Chem. 13, 3476–3483. Tylkowski, B., Trusheva, B., Bankova, V., Giamberini, M., Peev, G., Nikolova, A., 2010.
Eckert, F., Klamt, A., 2002. Fast solvent screening via quantum chemistry: COSMO- Extraction of biologically active compounds from propolis and concentration of
RS approach. AIChE J. 48, 369–385. extract by nanofiltration. J. Membr. Sci. 348, 124–130.
Ferro, V.R., Ruiz, E., Tobaias, M., Palomar, J.F., 2012. Integration of COSMO-based Van Baten, J., Szczepanski, R., 2011. A thermodynamic equilibrium reactor model as
methodologies into commercial process simulators: separation and purification a CAPE-OPEN unit operation. Comput. Chem. Eng. 35, 1251–1256.
of reuterin. AIChE J. 58, 3404–3415. Vandezande, P., Gevers, L.E.M., Vankelekom, I.F.J., 2008. Solvent resistant nanofil-
Fierro, D., Boschetti-de-Fierro, A., Abetz, V., 2012. The solution–diffusion with tration: separation on molecular level (critical review). Chem. Soc. Rev. 37,
imperfections model as a method to understand organic solvent nanofiltration 365–405.
of multicomponent systems. J. Membr. Sci. 413–414, 91–101. Vanneste, J., Ormerod, D., Theys, G., Van Gool, D., Van Camp, B., Darvishmanesh, S.,
Geens, J., De Witte, B., Van Der Bruggen, B., 2007. Removal of API's (active Van der Bruggen, B., 2013. Towards high resolution membrane-based pharma-
pharmaceutical ingredients) from organic solvents by nanofiltration. Sep. Sci. ceutical separations. J. Chem. Technol. Biotechnol. 88, 98–108.
Technol. 42, 2435–2449. Wang, S., Sandler, S.I., 2007. Refinement of COSMO-SAC and the applications.
Hesse, L., Micovic, J., Schmidt, P., Gorak, A., Sadowski, G., 2013. Modelling of Ind. Eng. Chem. Res. 46, 7275–7288.
organic-solvent flux through a polyimide membrane. J. Membr. Sci. 428, Wijmans, J.G., Baker, R.W., 1995. The solution diffusion model: a review. J. Membr.
554–561. Sci. 107, 1–21.