RATE-CONTROLLED

SEPARATIONS

PHILLIP C. W ANKA T
Schoo/ of Chemica/ Engineering, Purdue University,
West Lafayette, Indiana, USA

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

First edition 1990
Reprinted 1994

© 1994 Springer Science+Business Media Dordrecht

Originally published by Chapman & Hall in 1994
Softcover reprint of the hardcover 1st edition 1994

ISBN 978-94-010-4585-8 ISBN 978-94-011-1342-7 (eBook)

DOI 10.1007/978-94-011-1342-7
Apart from any fair dealing for the purposes of research or private study, or
criticism or review, as permitted under the UK Copyright Designs and Patents
Act, 1988, this publication may not be reproduced, stored, or transmitted, in
any form or by any means, without the prior permission in writing of the
publishers, or in the case of reprographic reproduction only in accordance with
the terms of the licences issued by the Copyright Licensing Agency in the UK,
or in accordance with the terms of licences issued by the appropriate
Reproduction Rights Organization outside the UK. Enquiries concerning
reproduction outside the terms stated here should be sent to the publishers at the
Glasgow address printed on this page.
The publisher makes no representation, express or implied, with regard to the
accuracy of the information contained in this book and cannot accept any legal
responsibility or liability for any errors or omissions that may be made.

A Catalogue record for this book is available from the British Library
Library of Congress Cataloging-in-Publication Data available
® Printed on acid-free text paper, manufactured with ANSI/NISO Z39.48-
1992 and ANIS/NISO Z39.48-1984 (Permanence of Paper)
 Part III
MEMBRANES
 NOMENCLATURE CHAPTERS 12 AND 13

a,a' tenns in quadratic equation

a constant for diffusivity, Eq. (12-75)

a constant for osmotic pressure, Eqs. (12-18b,c)

A area of membrane, m2 or cm2

Acs cross-sectional area for flow, m2 or cm 2

Am area of a membrane in ED stack, m2

b constant, Eq. (12-64)

b, b' tenns in quadratic equation

c, c' tenns in quadratic equation

c concentration. Either weight or molar or equivalents (ED) gIL,

glcc, gmoles/L, g equivalents/L

Cb bulk concentration

cg gel concentration

cin inlet concentration

em . concentration in membrane

COUI outlet concentration

Cp penneate concentration

Czrn solute concentration in membrane

,
C2rn solute concentration in membranes on high pressure side
"
c2rn solute concentration in membrane on low pressure side

CZp solute concentration in penneate

Cw wall concentration

623
 624

CZ w solute concentration at wall

heat capacity liquid, caVgOC

C;,V heat capacity vapor, cal/gOC

d,D diameter, cm or m

dcq Equivalent diameter, Eq. (13-12).

~ particle diameter in gel layer, m

D Diffusivity, cm zIs

solvent diffusivity in membrane, cmzls

solute diffusivity in membrane, cm zIs

parameter for diffusivity, Eq. (12-75)

energy, joules

Fanning friction factor, Eq. (13-lOb,c)

friction factors, Eqs. (13-86) and (13-87)

flow rate, gmoles/time or g/time

inlet flow rate, kg molesls or m3 Is

outlet flow rate, kg molesls or m 3 Is

permeate flow rate, cm 3 Is, m 3 Is or kg molesls

FRecycle recycle flow rate

F Faraday. 96,500 amp'sec/equiv

h half height between parallel plates, em

H hydrodynamic permeability, Eq. (13-96)

~Hf latent heat of freezing, Eq. (12-19)

i current density, amp/m z

 625

I current, amps

jn Colburnj-factor, Eq. (13-10a,b)

J asy asymptotic volumetric flux, Eq. (13-19)

J icit initial flux

J,olv, J 1 volumetric solvent flux, cm3 /cm 2s or L/m2 day

J:01v , J~ mass solvent flux, g/cm2s

J so1ute , J 2 solute flux. g/cm 2s

hdif solute flux due to diffusion

k mass transfer coefficient, cm/s

k!mute feed mass transfer coefficient

k~ute dialysate mass transfer coefficients

kT total mass transfer coefficient in dialysis

KA solubility parameter, Eq. (12-15)

Kg gel permeability. Eq. (12-38)

~lute solute permeability

Ksolv solvent permeability

K} =Cl m/c1p' solvent distribution coefficient

K2 solute distribution coefficient, Eq. (13-78)

L length, cm or m

Lu, L;j phenomenological coefficients, Section 13.6

m exponent for flux decay, Eq. (13-19)

m mobility, Section 13.6

M =Cw/Cb' concentration polarization modulus

626

MW molecular weight
n total moles solute, Section 12.5.

n number of tubes or cells

n constant in Eq. (12-18c)

N number of molecules or particles

Rate of transfer, cm3 /s

pressure, atm, bar, psi, mmHg, kPa

high pressure

low pressure

permeate pressure

effective pressure driving force across membrane, Eq. (13-94)

partial pressure

partial pressure on liquid side for pervaporation

P2 partial pressure on vapor side for pervaporation

p permeability of gas, cm3 (STP) cm/cm 2 s (cm Hg)

molar flow rates on high pressure and permeate sides

permeate molar flow rate

charge,coloumbs

volumetric flow rate, cm3/s or L/hr

dialysate volumetric flow rate
feed volumetric flow rate

r radial direction

R radius, cm

R resistance per cell, ohms

 627

R retention

R gas constant
RO inherent retention = 1 - c;,/cw

Rapp apparent retention = 1 - Cp/Cb

Renergy gas constant in energy units, (Example 13-1)

~ pore radius, cm

~ress gas constant in pressure units, (Example 13-1)

Re Reynolds number =d Ub p/~

S* Henry's law constant, Eq. (12-76)

Sc Schmidt number =~ p/D

t time, s or days

fg gel layer thickness, cm

tm membrane thickness, cm

T absolute temperature, K
T f , T; freezing temperature of solution and pure solvent, K

Tin inlet temperature

Tout outlet temperature

Tref reference temperature

u velocity in x direction, cm/s

u center of mass velocity of pore fluid

llb bulk velocity, cm/s

llm inlet average velocity, em/s

v velocity in y or r direction, cm/s

v·1 partial molar volume, Eq. (13-63)

628

velocity through wall, cm/s

partial molar volume of solvent, gmoles/cm 3

V.olvcnt partial molar volume of solvent, gmoles/cm 3

v voltage, volts

v volume, cm3 or L

VP vapor pressure

w width, cm

x axial distance, cm

x distance into membrane, cm

x liquid mole fraction

X-l force on component i, Section 13.6

Y direction perpendicular to parallel plates

Y gas or vapor mole fraction

Y' y/h
gas mole fraction on high pressure side

Yin inlet mole fraction

Yout outlet mole fraction

permeate mole fraction

mole fraction on high pressure side

Y2 mole fraction on permeate side

Y2.0 eqs. (13-44a,c)

external force on component i, Section 13-6

z feed mole fraction

 629

Greek Letters

n selectivity, KsolvlK.olute in RO

nAB selectivity, P A/PB, in gas permeation

,
nAB selectivity, Eq. (12-81), in pervaporation

'Yw fluid shear rate, Eqs. (13-15) and (13-17)

a boundary layer thickness, em

£ porosity

11m manifold efficiency, ED

115 semipermeability membrane efficiency, ED

11w water transfer efficiency, ED

a cut = Fp/Fin

6rot total cut in recycle system, Fp/Fnew

A latent heat of vaporization, cal/g or cal/gmole

A tortuosity

J..l viscosity, poise or pascal-sec.

J..li chemical potential of species i

v kinematic viscosity = Wp
~ efficiency of current utilization, Eq. (12-67)

~ v! h x/3Um D 2, Section 13.5

7t osmotic pressure, atm

7t Pi,3.141592654 ....
P density, g/cm 3
A

P molar density, gmoles/cm 3 or kgmoles/m 3

ro angular velocity, radians/s
 Chapter 12

INTRODUCTION TO MEMBRANE SEPARATIONS

A variety of membrane separation processes such as reverse osmosis,

ultrafiltration, gas permeation, and electrodialysis are becoming increasingly
popular in@industry. This popularity is due to the simplicity of the processes,
the gentle nature of the separation (high temperatures and phase changes are
not required), the usual low energy requirements, and the often low capital and
operating costs. In this chapter all of the membrane separation processes will
be introduced. Simple equations for understanding and designing the processes
will be presented. The emphasis will be on understanding the physical nature
of the processes. Chapter 13 builds on the basics presented in this chapter and
presents more detailed analyses of the processes occurring in membrane
separators.

12.1. BASIC CONCEPTS

A membrane is a physical barrier between two fluids. In over 99% of the cases
of current industrial interest the membrane is made from a polymer. This poly-
mer is cast or spun or extruded to form a continuous film without holes in the
desired geometry. The simplest geometry used for membrane separators is the
stirred cell shown in Figure 12-1. The fluid enters into the upper chamber. Part
of the feed stream is transferred through the membrane to the lower chamber.
With a perfect membrane only species A will transfer through the membrane
while species B will remain in the upper chamber. Either species may be the
desired product. The stirrer is used to improve mass transfer rates and prevent
the buildup of species B at the membrane surface.

630
 631

~
~

,- I
~
-
Membran e
t +
Cp
III

Figure 12-1. Stirred cell membrane system

The flux through the membrane can be written as

Flux = Transfer rate (12-1)

Transfer area
= Permeability
Thickness
(Dri'
vmg 'Force)

Flux is defined either as J = (Volume)/(area)(time) or as J' =

(mass)/(area)(time). Obviously, these two fluxes are related by J' = Jpp where
Pp is the permeate density. The flux depends linearly on both the permeability
(a constant which indicates how "tight" the membrane is) and the driving
force. The driving force depends upon the type of separation. For instance, in
gas permeation the driving force is the pressure difference across the mem-
brane. The membrane can separate two gases because the permeabilities are
very different Note that, in general, both gases do transfer through the mem-
brane, but one of them transfers at a much higher rate. Thus, most membrane
separators are not equilibrium processes but are rate processes. The flux also
depends inversely upon the thickness of the membrane. The thinner the mem-
brane the higher the flux.

High fluxes are usually very desirable since the transfer area required
will be lessened. Thus we want a high driving force. In the cases of gas per-
meation, reverse osmosis (used for purifying water) and ultrafiltration (used for
concentrating large molecules) this means a high pressure drop across the
membrane. The membrane must be mechanically strong to withstand these
high pressures. Mechanical strength requires a thick membrane. However,
632

a Feed Phase
Thin
Skin

Porous
Support
Layer

Permeate

Figure 12-2. Asymmetric membrane. a. Schematic. b. Scanning electron

microscope photograph. (Warashina et ai, 1985). Part b
reprinted with permission from Chern. Tech, 558 (1985).
Copyright 1985, American Chemical Society.
 633

Eq.(12-1) shows that thick membranes will have low fluxes and hence large
are$ will be required to achieve the desired transfer rate. A fundamental prob-
lem which stopped the commercial use of membranes for many years was how
to make a membrane which was thin enough to have high fluxes while at the
same time it was mechanically strong.

The solution to this problem was found in the late 1950's and early
1960's by Sidney Loeb and S. Sourirajan (Loeb and Sourirajan, 1960, 1963).
Their solution was to make a membrane which was asymmetric or anisotropic.
This type of membrane is illustrated in Figure 12-2. The membrane was cast to
have a very dense thin layer or skin on one side. This layer was typically 0.1 to
1.0 microns thick. The thin skin does the actual separation of the species.
Thus in Eq.(12-1) the thickness term is very small. The thin skin is supported
by a much thicker porous layer. This layer provides the required mechanical
strength, but it is so porous that no separation occurs in this layer. These mem-
branes can be operated at pressure drops in excess of 1000 psig. The Loeb-
Sourirajan membrane was made from cellulose acetate for desalination of
water by reverse osmosis. Methods for casting these membranes are reviewed
by Kesting (1985) and Soltanieh and Gill (1981). The use of asymmetric mem-
branes has been extended to a variety of other polymers and to other membrane
separation methods. The history of membrane separations is discussed by
Lonsdale (1982). Not all membrane separators use asymmetric membranes.
The exceptions will be discussed in the individual sections.

A variety of different polymers have been used commercially for mem-

brane systems. The polymers used can be classified into three categories: rub-
bery polymers, glassy polymers, and ion exchange membranes. Rubbery poly-
mers such as silicone rubber operate above the glass transition temperature, Tg ,
of the polymer. As a generalization, rubbery polymers have high fluxes of
organics and low flux of water. Glassy polymers such as polycarbonate operate
below Tg • These polymers can be either amorphous or partially crystalline.
Glassy polymers are selective for water. Ion exchange membranes are essen-
tially ion exchange resins (see Chapter 9) made in membrane form. Ion
exchange membranes are very selective for water.

The most common structures are shown in Figure 12-3 (Kesting, 1985;
634

b c
(-C- y- )
\ C=N n

(-S02~r@-o~~?)
H ~ cf H n

f
) -CF2,C-) n
(-(CF2 CF2,

to-poj
O-(C~ yF )y-OC~C~S02F g
CF3

CH 3 n

Figure 12-3. Polymers used for membrane separators. a. Cellulose (D,UF).

b. Polyacrylonitrile (D,UF). c. Poly sulfone (GP,RO,UP,ED). d.
Polybenzimidazolone (RO). e. Polyamide (from lcrephlhalic
acid m-arnino benzarnide and m-phenylene diamine) (RO,UF).
f. Nation (Donnan dialysis). g. Silicone (GP). Key: GP = Gas
Permeation, D = Dialysis, RO = Reverse Osmosis, UP =
Ultrafiltration, ED = Electrodialysis
 635

Pusch and Walch. 1982; Soltanieh and Gill. 1981; Ward et al., 1985). Addi-
tional examples of polymers used for membranes are discussed by Kesting
(1985). Lloyd and Meluch (1985). Finken (1985). Pusch and Walch (1982), and
Ward et al., (1985). For example. Figure 12-3 shows only one of the possible
polyamides (Kesting, 1985; Pusch and Walch, 1982). The cellulose (Figure
12-3a) and cellulose acetate membranes were the firstcommercialIy successful
membranes. Although still commonly used, considerable research has gone
into developing membranes with better chemical and thermal resistances.
Polysulfone (Figure 12-3c) and Nafion (Figure 12-3f) have outstanding resis-
tances to chemicals. Polystyrene cross-linked with divinylbenzene (Figures 9-1
a and b) also has excellent chemical resistance and is used as a backing for
composite membranes and as the basis for electrodialysis membranes.

With a high flux rate species B which does not transfer through the mem-
brane may build up to a high concentration at the surface of thel membrane.
This is called concentration polarization and is illustrated in Figure 12-4.
Since the transfer rate of species B through the membrane is low, the permeate
concentration is low, cp «Cb' Species A (which might be the solvent) has a
high flux through the membrane. Thus. there must be a convective flow
towards the membrane. This convective flow carries B to the membrane sur-
face. Since the flux of B is low. the concentration of B will build up at the
membrane wall. Thus, a concentration gradient is produced. This buildup of B

I
+ Back Diffusion
Convective Flow

I
---:--- cb
-~·I

y=8

Membrane
Figure 12-4. Build-up of retained species at membrane wall (concentration
polarization).
636

is counteracted by diffusion into the bulk fluid. If diffusion rates are high as is
true in most gas systems, then the concentration buildup will be modest and
concentration polarization will not be a problem.

In liquid systems diffusivities are small and Cw > Cb. Concentration

polarization is detrimental for several reasons. The high Cw value means that
any liquid that leaks through the membrane will be at a high concentration of B
and thus will increase Cpo The high wall concentration also increases the
osmotic pressure (see Section 12.4.) which reduces the driving force for
transfer of solvent High concentrations of solute can also cause gelling or
fouling at the membrane surface which will drastically decrease flux (see Sec-
tion 12.5.). To avoid concentration polarization we can use stirring, turbulence,
high shear rates, or very thin channels.

U.2. MEMBRANE MODULES

The general design requirements for any membrane system can now be listed.
1. Thin active layer of membrane.
2. High permeability for species A and low permeability for
species B.
3. Stable membrane with long service life.
4. Mechanically strong.
5. Large surface area of membrane in small volume.
6. Concentration polarization eliminated or at least controlled.
7. Easy to clean if necessary.
8. Inexpensive to build.
9. Low operating cost
This list of desirable design conditions is general. Items specific to each of the
membrane devices will be discussed later. A historical perspective on mem-
brane modules with many early advertisements is given by Kremen (1986).
Currently, spiral wound and hollow fiber systems have most of the market.

The simplest geometry for a membrane separator is the stirred cell shown
in Figure 12-1. Stirred cells are commonly used in laboratories, but the mem-
brane surface area is small. Thus for large scale commercial applications other
 637

a b
Porous Tube

O
------------r-.
I \
( \
I I
I I
\ I

- - -
\ I
- - - - - - ~ ...'
Membrane Coating
(Inside or Outside)
----..:: Spacers

Reten~
c
... :~

Permeate

Permeate channel

r Potting IIIt Permeate

d

Feed--(]lilililililililllll: ·
L Hollow fibers
Distribution
Figure 12-5. Geometries of commercial membrane systems. a. Plale-and-
frame. b. Tubular. c. Spiral wound. d. Hollow fiber.

geometries are required. The most commonly used commercial systems are
shown in Figure 12-5. The plate-and-frame system is similar to a plate-and-
frame filter press with a series of flat membrane sheets. The area per volume
ratio is significantly higher than stirred cells but lower than spiral wound or
hollow fiber systems. The sheets can be quite close together to reduce concen-
tration polarization. The major advantage of plate-and-frame systems is they
 638

can be taken apart for cleaning if necessary. Their major application has been
in the food industry where fouling is a major problem. The plate-and-frame
geometry is also used for electrodialysis.

Cylindrical geometries are also used. In the tubular system shown in

Figure 12-5b the membrane is coated on the inside or outside of the porous
support tube. Typical inside diameters are in the range from 1 cm to 1 inch. A
bundle of tubes can be packaged together as in a shell-and-tube heat exchanger,
but the area/volume ratio is low. The main advantage of tubular systems is
they can be mechanically cleaned by forcing a sponge ball through the tubes.
Thus tubular systems compete with plate-and-frame systems for dirty or foul-
ing applications. Design details of these systems are given by Stana (1977).

The spiral wound system shown in Figure 12-5c is a way of increasing

the area/volume ratio of flat plat devices. Two feed channels, two membranes,
and a permeate channel are layered and wrapped around a central porous tube
several times. Spiral wound systems made from cellulose acetate are com-
monly used for reverse osmosis. This design is a good method for achieving
large area/volume (about 300 f~ per ft3 ) with membrane materials which can-
not easily be extruded into hollow fibers. Concentration polarization is con-
trolled by using thin channels and wrapping the membrane around plastic net-
ting which induces turbulence. Unfortunately, these systems cannot be used for
treating very dirty or fouling systems since they are not easy to clean. Design
details are given by Kremen (1977). Spiral wound is one of the two most
popular configurations.

The hollow fiber system shown in Figures 12-5d and 12-6 is also one of
the two most popular designs. The fibers are extruded from polymers such as
Nylon or polysulfone. A typical reverse osmosis hollow fiber would have a
inner diameter of 42 microns, an outer diameter of 85 microns, and a 0.1 to 1.0
micron skin on the outer surface (Applegate, 1984). For gas permeation the
fibers are made somewhat larger to reduce the pressure drop inside the tubes.
For ultrafiltration the fibers are much larger (500 to 1100 micron i.d.) and the
skin is on the inside of the fiber. These numbers indicate that a considerable
amount of engineering design can be done with the fibers. The fibers can have
fairly thick walls and with flow radially inward large pressure gradients can be
 639

- 1126mm------·---1

-+ Solution outlet
tS9-m;-
diameter

Case

Hollow-liber membrane

Tube.eat

Figure 12-6. Details of hollow-fiber module. (Warashina et ai, 1985).

Reprinted with permission from Chern. Tech., 558 (1985).
Copyright 1985, American Chemical Society.

handled. The major advantage of the hollow fiber systems is that a huge
number of fibers (4.5x106 ) can be packed inside a 25.4 cm (10 inch) diameter
cartridge. Thus the area/volume ratio is very high (about 5000 ft2 per ft3) and
costs are kept down. A major disadvantage of hollow fibers is particulate
solids can permanently clog the tiny fibers. Thus dirty fluids must be
thoroughly cleaned before being sent to the hollow fiber separator. Details of
hollow fiber systems are given by Orofino (1977) and Caracciolo et al., (1977).

All of the basic systems are usually built in a few basic modular sizes. A
"module" is a complete unit such as a hollow fiber separator capable of pro-
cessing a given amount of fluid with a given separation. The details of a hol-
low fiber module used for ultrafiltration are shown in Figure 12-6 (Warashina
et al., 1985). A single module is often not sufficient for the required flow rate
and/or separation. Thus modules are cascaded in a variety of ways to produce
the desired separation. One advantage of modular construction is the modules
640

$ - -~
a b
OOOOOOOIOUI.

,OOOOOOtOOlOL ~:~: 1 001 OOOOfOOOOOO

FIN

YIN

C d

~'I~iQt~ ~ ~ -- i-- ~
! ~

-- ---
e f

Figure 12-7. Idealized flow patterns for flow inside module. a. Completely
mixed. b. Mixed on feed (high pressure side). c. Mixed on per-
meate (low pressure) side. d. Cross-flow. e. Co-current. f.
countercurrent.

can be constructed in a plant instead of being field constructed. This is usually

significantly cheaper. One disadvantage of modular construction is that there is
probably very little economy of scale.

The flow patterns inside the module have a large effect on the outlet con-
centrations and flow rates of the high and low pressure products. Several ideal-
ized flow patterns are shown in Figure 12-7. The completely mixed case, Fig-
ure 12-7a, is the easiest to analyze but the least advantageous for separation.
The system which is mixed on the feed side, Figure 12-7b, is close to the flow
pattern which occurs for most hollow fiber systems for gas permeation and
reverse osmosis. Figure 12-7c shows the idealized flow pattern of a hollow
fiber system for ultrafiltration where the feed is inside the fibers. The solution
is well-mixed on the permeate side. Plate-and-frame and spiral wound systems
can have a modification of the cross-flow system, Figure 12-7d, or they could
 641

be modifications of the co-current flow system, Figure 12-7e or the counter-

current flow system, Figure 12-7f. The countercurrent flow pattern will be
most advantageous and has been used to design experimental fractionation sys-
tems.
Once the flow pattern is known external mass balances can be combined
with Eq. (12-1) to solve for the outlet concentrations and flow rates. For a
completely mixed system Eq. (12-1) is the same everywhere. Mass balances
for perfectly mixed systems will be illustrated throughout this chapter. These
are the easiest balances to solve, and since perfectly mixed gives the least
separation the design will be conservative. For the other flow patterns the driv-
ing force varies throughout the module and an integration is required. These
external mass balances are developed in Chapter 13.
Cascade schemes are shown in Figure 12-8. When the capacity of a sin-
gle module is insufficient, the obvious solution is to connect two or more
modules in parallel (Figure 12-8a). If a single module is unable to remove the
desired amount of permeable species A, the series arrangement shown in Fig-
ure 12-8b can be used. The parallel-series arrangement is useful when both
high flow rates and high recoveries of A are desired. Note in Figure 12-8c that
the number of modules in parallel can be decreased as more fluid is removed.
A commonly used alternative is the recycle system in Figure 12-8d. This sys-
tem keeps flow rates high to decrease concentration polarization and it allows
for a high recovery of the permeable species. Recycle systems are commonly
used in batch operation to process liquids. Recycle systems can also be put in
parallel to increase the capacity. When the selectivity of the membrane is too
low to produce a pure enough permeate, the permeate in series arrangement
shown in Figure 12-8e can be used. The engineer would prefer to obtain a
better membrane instead of staging the permeate since permeate staging
requires an extra pump or compressor. Thus operating costs will be high. Note
that a recycle stream will often be used for the stream that does not pass
through the membrane. If both streams need to be put in series the counter-
current cascade shown in Figure 12-8f can be used. This countercurrent cas-
cade is similar to other countercurrent cascades discussed in this book except
that additional energy must be input before each stage. This occurs because the
rate process is unable to reuse the energy from stage-to-stage. Because of this,
642

l:
~ ~
fA
~B d41r'
fA

1
e
~~
~ Recycle

~ C A
~
Figure 12-8. Cascades for membrane separators. a. Parallel. b. Series. c.
Parallel-series. d. Recycle. e. Permeate in series. f. Counter-
current fractionation. g. Batch. Species A is more permeable
and species B is less permeable.

operating costs will be high and the use of this countercurrent cascade in a
membrane process will be unusual. In the batch operation shown in Figure
12-8g recovery in a single pass is insufficient, and operation is continued until
enough of the more permeable species has been recovered. Batch systems are
common in applications such as food processing where frequent cleaning is
required. Modifications of these cascades used for gas permeation are shown
in Figure 12-11.

In the remainder of this chapter the different membrane separation sys-

tems will be considered individually. The presentation starts with gas permea-
tion since in many ways it is the simplest of the membrane processes. Then the
liquid systems reverse osmosis, ultrafiltration and dialysis are considered.
Electrodialysis uses electrical forces to separate liquids containing ions, and
 643

thus this process introduces new concepts. Following this, pervaporation which
involves vaporizing a liquid at the membrane surface will be considered.
Finally, the experimental liquid-membrane systems will be briefly discussed

12.3. GAS PERMEATION

In gas permeation two or more gas species are separated based on different per-
meabilities in a membrane. Although any of the devices shown in Figure 12-4
could be used, hollow fiber and spiral wound systems are used commercially.
The hollow fiber systems have the advantage of very large surface to volume
ratios. The hollow fibers can have an i.d. up to 200 microns in diameter. The
larger the i.d. the lower the pressure drop inside the tubes ( a surprisingly
important design consideration), but the lower the area/volume ratio. The wall
thickness can be from 25 to 250 microns depending upon the pressures which
must be withstood (Henis and Tripodi, 1983). The skin is from 0.1 to 1.0
microns thick and is formed on the outside of the hollow fiber. Thus flow is
radially inward since this allows the fiber to withstand much higher pressures.
Shell-and-tube separators similar to Figure 12-6 are used. The hollow fiber
systems are currently being used commercially for recovery of hydrogen and
carbon dioxide, and nitrogen generation from air.

The spiral wound configuration is also used commercially for carbon

dioxide recovery using cellulose acetate membranes. This configuration has
the advantages of a lower pressure drop in the direction of flow, and the tech-
nology to build the system is less complex. However, the area/volume ratio is
much lower than in hollow fiber systems. A variety of both hollow fiber and
spiral wound applications are discussed by Spillman (1989).

The basic flux equation written in words in Eq. (12-1) becomes

(12-2)

where h is the volumetric flux, Fp,A is the steady state volumetric transfer rate
of species A through the membrane, A is the membrane area, PAis the permea-
bility of the membrane for species A. the driving force L\p is the change in the
 644

partial pressure of the species across the membrane, and lm is the thickness of
the membrane skin. Obviously, consistent units must be used. The usual units
for permeability are cm3 [S1P)crn/cm2 s (cm Hg). Since the partial pressure is
the mole fraction times the total pressure, PA = YAPtot, Eq. (12-2) can also be
written as

J A = Fp,A = PAw!'Yh,A - PpYp,A) (12-3)

A lm

1n this equation Pp is the total permeate pressure, Ph is pressure on high pres-

sure side, Yh,A is mole fraction A on high pressure side and Yp,A is the mole
ftaction A in the permeate. This equation often has to be applied point-by-
point since Y and p can vary. With a large pressure difference across the mem-
brane, it is possible to have Yp,A>Yout,A and still have transfer of this species
through the membrane. In order to obtain separation of different gases the per-
meability of the membrane must be significantly different for the two gases.
The selectivity of the membrane

(12-4)

should be greater than 20 or preferably 40.

For the completely mixed system in Figure 12-7a the external mass bal-
ance at steady state is

(12-5)
FinYin = FoutYout + Fpyp
where F is the molar flow rate and y is the mole fraction. This mass balance is
often written as

(12-6)
Yin =(1- 0) Yout + 0 yp
where the cut 0 is the permeate flow rate divided by the feed flow rate.

(12-7)
 645

The cut is usually one of the design parameters. With a specified, we can solve
Eq. (12-6) for Yom
Yin - a yp (12-8)
Yout = 1- a

For gas permeators concentration polarization is usually not a problem

since diffusivities are high and can probably be neglected. Thus the flux equa-
tions apply to bulk: concentrations since the wall concentration equals the bulk
concentration at each point The external mass balance, Eq.(12-5), and the
flux, Eq. (12-3), can be solved simultaneously for the completely mixed separa-
tor shown in Figure I2-6a (Hwang and Karnmermeyer, 1975). For a binary gas
Eq. (12-3) can be written for both species A and B. In molar units the result for
component A is,

(I2-9a)

while for component B,

A

(12-9b)
F p (1 - yp ) = P B1m
ApB [PH(1- Yom) - pdI - yp) ]

where PA and PB are the molar densities at the permeate conditions, and A is
the area for mass transfer. Dividing Eq. (12-9a) by (12-9b), we obtain

yp PA
A Yout - [~ 1 yp
(12-10)
-"--- = ClAB -,..- I:'H
----~---''-----
1- yp PB PL
(1 - Yout) - - (1 - Yp)
PH

Substituting in Eq. (12-8) we have

,..
Yp PA (12-11)
--=ClAB -A-

I- Yp PB 1 _ Yin - a yp _ PL (1 _ Y )
I-a PH p
646

Once fractions are cleared, Eq. (12-11) becomes

(12-12)

where

a= [_9 + PL
1-9 PH
1[a PB~A
AB -1]
(12-13a)

PA
b=(I-aAB)- A [
PL
-+--+-- ---
PH
9
1-9
Yin
1-9
1 1
1-9
(12-13b)
PB
A

1
A

PA (12-13c)
c=aAB -A- Yin
[ 1-9
PB
and then

-b± ~bz-4ac (12-14)

yp =
2a

The root which gives yp between zero and one is used. For ideal gases
PA/PB = 1 and the equations simplify. Alternate manipulations of these equa-
tions are shown in Example 12-1.

The penneate concentration increases as the membrane selectivity

increases, the pressure ratio PH/PI.. increases, or the cut 9 decreases. These
effects are illustrated in Section 13.4. We will see that the completely mixed
module provides the worst separation and countercurrent is best.

Example 12-1. Gas Penneation.

We wish to remove COz contaminating CJ4. At 35°C and PH = 20 atm

the penneability of cellulose acetate membrane is (Chern et ai, 1985) P co.
= 15 x 10-10 and PCH,. = 0.48 x 10-10 [cc (STP) cm]/[cmzs cm Hg]. The
 647

unit 10-10 [CC (STP) cm]/[cm 2 scm Hg] is defined as 1 Barrer. The gas is
perfectly mixed on both sides of the membrane. The high pressure feed
gas is 90 mole % C~ and 10 mole % CO2 , The permeate pressure is 1.1
atm, and the cut e = 0.5. Determine the membrane selectivity, Yp,eo" the
CO2 flux and the Cf4 flux if Lm = 1 J..UTI.

Solution

..
SeIecUVlty, Pea, 15
a.co..~ = - - = - - = 31.25. The permeate mole frac-
Pea. 0.48
tion Yp = Yp,eo, can be determined from Eqs. (12-12) to (12.14). From
Eq. (12-13),

a= r 0.5
L1- 0.5
+ Q]
20
(31.25 - 1) = 31.91

b~ (- 30.25) [;~ + 1+ ~:! ]- 2 ~ -39.96

~
c (31.25) [~:! ] ~ 6.25
Where we have assumed PAIPB = 1.0. Then from Eq. (12-14)

_ 39.96± ..J1596.8 - (4) (31.91) (6.25)

yp - 63.82

Since O:s; yp :s; 1, use the minus sign. yp = 0.183. Now using mass bal-
ance Eq. (12-8)

Yout
= 0.1 - 0.5 (0.183)
0.5
=0.0168
648

The CO2 flux can be found from Eq. (12-3).

15 x 1010 [cc (S1P) cm 1

co. -
cm2 scm Hg
J - ---~:-----........
(1~) (10-4 cm/1 ~)
[
76cm Hg
1 aun
1
x [(20) (0.0168) - (1.1) (0.183)]aun = l.535 x 10-4 cc S21P
cm s

where Yh,CO. = Yout since the module is well mixed.

Jeu. - [ 0.481~~0- 10
1(76) [20 (0.9832) - (1.1) (0.817)1 - 6.85 x 10-'

Notes: l.) Most of the CO2 is removed, yet permeate gas is still
mainly C~. This is important since driving force for CO2
transfer remains positive.

2.) Since the module is perfectly mixed, Yh,A = Yout in Eq.

(12-3).

3.) JCll. > J co• despite much higher CO2 permeability. This
occurs because of the much higher driving force for C~

4.) Appropriate conversion factors are required.

5.) Concentration polarization (see next section) is assumed to

be negligible.

6.) Note that transfer of CO2 is "uphill" in the sense that CO2
mole fraction is higher on the permeate side. The pressure
drop allows this direction of mass transfer since
Pin,co. > Pp,co •.

7.) Since other flow configurations such as counter-current

will give higher Yp and lower Yout. this design is conserva-
tive.
 649

A large literature has developed on the subject of membrane penneabilities

(Chern et at., 1985; Finken, 1985; Hwang and Karnmenneyer, 1975; Kesting,
1985; Koros and Chern, 1987; Rogers. 1985; Stannett et at., 1979; Stern, 1976;
Stem and Frisch. 1981). A simplified picture following Baker and Blume
(1986) will be presented here. The penneability is the product of the solubility
of the gas in the membrane and the diffusivity of the gas in the membrane.
(12-15)
PA =KAD A

The solubility parameter K can be considered an Henry's law constant which

links the concentration of a species in the membrane to the partial pressure in
the adjacent gas. For simple gases the solubility increases as the gas diameter
increases because the gas becomes easier to condense. This is illustrated in
Figure 12-9a where the Lennard-Jones collision diameter is used as a measure
of the size of the gas. As the diameter of the penneate molecule increases, the
diffusivity nonnally decreases becauses large molecules interact more with the
polymer chains. This effect is illustrated in Figure 12-9b. Figures 12-9a and b
are for natural rubber membranes. The values of K and D depend upon the
polymer, but the general behavior will be similar to that shown in these figures.
The penneability is the product of K and D. This product is shown in Figure
12-9c for several different polymers. The diffusivities of most gases in rubbery
polymers are approximately the same; thus, selectivity comes from solubility
differences. Ll glassy polymers both diffusivity and solubility differences are
important in detennining selectivity. Listings of penneabilities for several
gases in a variety of polymers are given by Koros and Chern (1987), Pusch and
Walch (1982), Stannett et at. (1979), and Stem (1986).

Figure 12-9c helps to explain why purification of hydrogen from carbon

monoxide, nitrogen or methane; and separation of carbon dioxide are currently
the major industrial successes. Hydrogen and carbon dioxide have significantly
greater penneabilities than the gases they are separated from since H2 has a
high diffusivity and CO2 has a high solubility. Development of membranes for
other separations is more difficult since more specific interactions between the
penneant and the polymer are probably required to achieve high selectivities.
One way to do this is to use "carriers" which will complex with the desired gas
to help transport it across the membrane (Baker and Blume, 1986).
650

,. 140
,.. 15 150 ..
E%
U E
III
120
'C02

EU ~C02 ......
~ ~

r
-, 100

,
C E
....
~ lOr tH2 100 III U
u' u%
Q
i!
=
~. 8 80
iIII
'u
-I
\ U = il-;' 60
§~ ~~ r:
1U
5 \
\
50 e.!!
i =
tEEu
Q, U
40
I!
c

~~H'
C Co CH.
0
'iii .!!!. ::.
=
= H. 02 JiI.::: 2 .. E
.~!
20
C OJ. H4 0 Cu
3.5 4.0 4.5
III ..
%CII 0
0'2.5 3.0 u
C
0 2.5 3.0 3.5 4.0 4.5
Lennard-Jones collision diameter (i) !
'-'
Lennard-Jones cOllision diameter (i)

a b

CO CH.

,
He H. O. N. C02
10,000
~-~---

Silicone
1,000 rubber /
.-- - 'V
!

Natural f
rubber I
.... ~ I.
)... '\/ L
Poly->(" \.
c
Co
styrene~ \ -- /
/I
:::. Cellulose _"- \ 1/
acetate '\ -/
"J
0.1
o 2.5 3.0 3.5 4.0 4.5
Lennard-Jones collision diameter (i)

Figure 12-9. Penneation of gases in polymer membranes (Baker and Blume,

1986). a. Diffusivity and solubility in natural rubber. b. Per-
meability in natural rubber. c. Penneabilities in four polymers.
Reprinted with pennission from Chern. Tech., 232 (1986).
Copyright 1986, American Chemical Society.
 651

Obtaining a membrane with high permeability is not sufficient to have an

economical commercial separator. In addition, we must be able to form the
membrane into a geometry which will be able to withstand fairly high pres-
sures, will have high area/volume ratios, and will have a thin active skin with
no holes. Spiral wound cellulose membranes and hollow fiber poly sulfone
membranes satisfy the first two requirements. The third requirement is much
more difficult. The thinner the skin the more likely there will be imperfections
or pores which are much larger than the gas molecules being separated. The
gas flows through these pores by convective flow driven by the pressure gra-
dient across the membrane. Unfortunately, this gas does not undergo the
separation mechanism and the concentration of this gas will be the same as that
on the high pressure side of the membrane. Although significant strides have
been made in producing more perfect membrane skins, this approach has not
yet made thin skinned membranes which are perfect enough for hydrogen
purification. A surface porosity as low as 1~ will lower the separation factor
of H2 compared to CO for poly sulfone membranes from 40 to approximately 5
(Henis and Tripodi, 1983). This value is not high enough to obtain pure hydro-
gen.

Two methods are used in commercial separators to circumvent this prob-

lem. For CO2 purification cellulose acetate membranes are used with a fairly
thick skin. This reduces the permeability, but with CO2 this is less critical
since CO2 has a very high permeability (see Figure 12-9c). The second solu-
tion used for H2 purification is to coat an asymmetric polysulfone membrane
which inherently has a high separation factor with a layer of silicone rubber
(Henis and Tripodi, 1983, Finken, 1985; Schell, 1985) and Stem (1986). The
silicone rubber layer serves to seal the defects. These resistance model (RM)
membranes are illustrated schematically in Figure 12-10. The silicone rubber
has a very high permeability (but low selectivity). The permeabilities [cm 3
(STP) cm/cm 2 s cm Hg] of silicone rubber are 5.2 x 10-8 and 2.5 x 10-8 for H2
and CO, respectively. The polysulfone has permeabilities of 1.2 x 10-9 and
3.0 x 10-11 for H2 and CO, respectively (Finken, 1985). The RM membrane
will have its permeability reduced slightly, but has a selectivity> 30 which is
high enough for commercial H2 purification. Note in Figure 12-10 that it is
difficult to determine exact values for 1m for either layer. Experimentally, it is
652

Silicone Rubber
Defect

}-Thin Skin

Polysulfone
Support Layer

Figure 12-10. Resistance model (RM) gas permeation membrane.

easiest to measure the ratio Pltm and to not separate the tenns. Other commer-
cial applications are discussed by Schell (1985) and Stem (1986).

A first cut at design of a gas penneator can be made once the penneabili-
ties, pressure drop, thickness of the active layer, and area per volume of the
spiral wound or hollow fiber system are known. To detennine the amounts and
compositions of both high and low pressme products the flow patterns inside
the separator and the product cut must be specified. These details are discussed
in Chapter 13.

The single stage systems shown in Figures 12-5 to 12-7 are often not
optimal and modifications of the cascades shown in Figure 12-8 are often used.
Three cascades used commercially for gas penneation are shown in Figure
12-11 (Spillman, 1989). The two-stage process with recycle shown in Figure
12-11a is used to improve recovery of the slower penneating species at the cost
of additional compression and membrane area. If the concentration of the per-
meating species is very high, the "premembrane" shown in Figure 12-11b can
be useful for bulk removal. According to Eq. (12-3) a low penneate pressure
will decrease membrane area or increase product recovered; however, if
recompression is required a low penneate pressme increases recompression
costs. One compromise is the two stage process shown in Figure 12-11c where
the penneate pressures are different. Since gas compression is done in stages,
the two penneate pressures are chosen to match the compression scheme.

Currently. gas penneators are speciality items which would be ordered

from the manufacturer. The penneators are often ordered as a turn-key device.
That is, the purchaser specifies required operating conditions including the inlet
 653

Membrane
~~~~~~--~8 E
Feed

Membrane

Corr4:lress

Permeate

b
Bulk Removal Ptrification

Permeate

c 60% Hydrogen
20% Nitrogen _ _ _ _ ....-,
20% Hydrogen

20% Inerts r-
,--_ _ _ - , 4;~%N:~r~r~:n
2,000 psi

PURGE GAS FUEL GAS

1,000 psi

90% Hydrogen 88% Hydrogen

6% Nitrogen 70/0 Nitrogen
4% Inerts 50/0 Inerts

86% Hydrogen Recovery

Figure 12-11. Cascades for gas permeation (Spillman, 1989). a. Two stage
with recycle LO improve recovery of the slower permeating
species. b. Three stage showing addition of a "premembrane"
for bulk removal. c. Two stage with different permeate pres-
sures. Reprinted with permission from Chern. Eng. Prog ..
85(1),41 (1989). Copyright 1989, Amer. Inst. of Chern. Eng.
 654

gas composition and flow rates, temperature and pressure, and the required
outlet gas composition for both high and low pressure gases. The manufacturer
then designs and builds a system which he guareentees will meet the required
specifications. Unfortunately, this arrangement is not as full-proof to the pur-
chaser as it sounds. If the actual operating conditions are different than
specified, the guareentee may be voided. In addition, the supplier will tend to
over-design to be sure he meets the specifications. Finally, tum-key units tend
to be more expensive than systems where the buyer does the design. It is best
if the buyer knows as much as possible about the separator even when he is
buying a tum-key system. This allows the buyer to write a more advantageous
contract and to do a better job bargaining. Additional factors which can be
important in design are discussed in Chapter 13.
The economics of gas permeation separation compared to competing
processes is discussed in detail by Spillman (1989). He concludes:

1. Membranes are efficient concentrators, but are not efficient in producing

100% purity.

2. Only optimized systems including multi-stage processes such as those

shown in Figure 12-11 should be used in economic comparisons.

3. Recompression costs are very important and must be included in the

analysis.

4. Since the technology is changing very rapidly, economic analyses can

become rapidly outdated.

5. Hybrid systems (membranes plus another separation) are often advanta-

geous. This occurs since membranes work best at low concentrations of
the permeating species while systems such as absorption work best at high
concentration. Examples of hybrid systems for air separation are given by
Beaver et al (1988) and Gienger and Ray (1988).

12.4. REVERSE OSMOSIS (RO)

Reverse osmosis (RO) is a process which reverses the normal direction of

osmosis by increasing the pressure of the concentrated stream. Obviously, this
 655

definition is useless unless we know what osmosis is. Thus we need to digress
slightly into the thennodynamics of membranes.

Osmosis is the flow of solvent through a semipenneable membrane from

the less concentrated to the more concentrated region. A "semipenneable
membrane" is a membrane which allows solvent to pass but completely
prevents the flow of certain solutes or ions in solution. The osmotic flow is a
natural occurrence as the system tends to come to equilibrium and equalize
chemical potentials. The osmotic flow can be decreased by applying a pressure
to the concentrated solution. The higher the applied pressure the less the
osmotic flow. When the flow stops, the applied pressure is the osmotic pres-
sure which is given the symbol1t. At higher pressures flow will be reversed
and we will transfer solvent from the concentrated to the dilute solution. This
is reverse osmosis.

As long as the membrane is perfectly semipermeable, the osmotic pres-

sure is a thennodynamic property of the solution and is independent of the
membrane properties. The osmotic pressure can be measured directly,
although it is often much more convenient to estimate it from other thenno-
dynamic quantities. Osmotic equilibrium requires that the chemical potentials
of the solvent on both sides of the membrane are equal.

(12-16)
Jll.solvent = Jl2,solvent

Note that the equilibrium condition is on solvent only. Since solute does not
pass through the membrane, solute ~s not in equilibrium. Thermodynamic
arguments can now be used to derive equations which allow calculation of the
osmotic pressure (for example, see Reid, 1966). Assuming that the liquid is
incompressible and that activities can be estimated from vapor pressure meas-
urements, the result is

ltVsolvent
VPpure solvent
= TIT In [ VP .
1 (12-17)
soluuoo

where vsolvent is the partial molar volume of the solvent Equation (12-17) is
usually quite accurate, but is not always convenient to use since it requires
656

Table 12-1. Osmotic pressure of aqueous sucrose

solutions at 30°C (Reid, 1966).

7t in atmospheres

c, gmoles/liter Van'tHoff Eq. (12-17) Experimental

Eq. (12-18a) Data
0.991 20.3 26.8 27.2
1.646 30.3 47.3 47.5
2.366 39.0 72.6 72.5
3.263 47.8 107.6 105.9
4.108 54.2 143.3 144.0
5.332 61.5 199.0 204.3

vapor pressure data of solutions. A simpler result is obtained by assuming that

the solution is dilute and that Raoult' s law is valid. This result is the van' t Hoff
equation,

(12-18a)
7t=cRT

where c is the molar concentration of solute in solution (moles/volume).

Although very convenient, the van't Hoff equation is often incorrect. This is
illustrated in Table 12-1 which presents the calculated and experimental values
for the osmotic pressure of aqueous sucrose solutions (Reid, 1966). For this
system Eq. (12-17) is quite accurate while the van't Hoff equation seriously
underestimates the osmotic pressure. If the solution dissociates or associates
in solution Eq. (12-17) remains valid, but Eq. (12-18a) should not be used.
Although the van't Hoff equation may be invalid it is often useful to assume a
linear dependence on concentration

(12-18b)
7t=ac

where a is an empirically determined constant, and c may be in any desired

units. For macromolecules the osmotic pressure increases much faster than
 657

linearly, and can often be represented as (Wijmans et ai., 1984),

(12-18c)

where a is a constant and n> 1. Often n is approximately 2.

The osmotic pressure of solutions is not widely tabulated and vapor pres-
sure data may not be available. Fortunately, the osmotic pressure can be
related to freezing point depression which is easy to measure and is widly tabu-
lated (e.g. in Handbook of Chemistry and Physics). The osmotic pressure is
related to freezing point depression by (Reid, 1966),

(12-19)

where T; is the freezing point temperature of pure solvent, Tf is the freezing

point temperature of solution of the desired mole fraction and .1 Hf is the latent
heat of freezing of the solvent. The ratio ~lRenergy is the ratio of gas con-
stants in pressure and energy units. This conversion is required to obtain 1t in
pressure units (see Problem 12-F2). Note that both Eqs. (12-18a) and (12-19)
predict 1t depends linearly on T. Use of Eq. (12-19) is illustrated in Example
13-1.

The flow in reverse osmosis is driven by pressure differences. Although

concentration polarization is usually important in reverse osmosis (see Figure
12-4), for the moment we will assume that the solutions are perfectly mixed so
that there is no concentration buildup. Then Cw = Cb (This development follows
that of Blatt et ai, 1970). Equation (12-1) becomes

K.olv
Jso1v = --(.1p - .11t)
(12-20a)
1m

where K.olv is the permeability of the membrane to the solvent (usually water),
.1p is the pressure drop across the membrane, and .11t is the osmotic pressure
difference across the membrane. Note that the pressure driving force is
reduced by the osmotic pressure difference. As the solution becomes more
658

----L)- -IT I . ··l.111-

3.0
(a)
1 - ;
i
-~ - .. -
I
Reference IS 800 pSlg, 25·C, 30% recovery i
, and 30,000 ppm NaCI feed

I
2. To find productivity at any other concentration i ! I
---
-~I--
.. __
and pressure, multiply data in Table I by the ~T-~
J~_
._.

l.-
appropriate factor
..J.--
---
2.0

--
II:
I
~
·. V V
~ ~ -~

..
N()CI leeO c.-- l--
.-- -
l-- V
f-

--
>-
I-

....--V
:;
t .\,'!l~: ....
.- ... I
---
.. V
...
\O,~ ...-::.
~
o k:
V k-
~
~k
o ..
II: . r:: j:; . '

..- i-"'""
1.0
p. .
Q.
..
~ :. I
~~
009911\
L.;..rl ... 'l:,
00 . ;::;...,.-- V
~., .... . " ,. . .. 11\
I--
:.;- ~ ~
·
..

--
...
L--I-
,. t'
'!)o,oo~j...--
• I • •

..
"-j • 1" "

~.
V~ r-
k ~ l-- ttO~
>-:: ...
...
.. "1'
· t~ •• , t
V
o
400 500 600 700 800

PRESSURE, psig

(b)

-= =
_. --,+ . . . ' ....... -,,:, ---'-,. .
~.:: :-::-: HI. Reference is 800 psig,' 30·/~ recovery
~

,
. ... . .. land 30,000 ppm NaCI feed. i-- h---,' .~.~. --·'~--lf~· I~
~I'-=:~: I~2. To find salt passage at any other can- I-+Lt::t=--.:We-:-- •::=;
~ 2.0 1---., -;w .J centration and recovery, multiply data - '. - 40 r'jt:-t:-:::.:. ..
t; ~= ~ ~ in Table 1 by the appropriate factor. ~f-W+. ~:.. Jyt::~ --:-.
~ :::.... ::-.: h ++T ::tixf-:;:;=:-:- _ . ::::.f:-'-"' -+j- ~,Itil: :-:.I.::-::;;II::r:::
-./
~ t:::""-R::- !+H=t-:H-'FH+t~ ~¥ - ;:;- : +,(--= .:: j/': 3oo4=lr :-":::
~_ ~ -+-4-+1-+-+-+
r=;=1---t+.:...J.
. ,.. ~. fEf,,+t. ;-~f=I '.~" -I--~.-. V_'
:::. .
_
:
Q. 1--. !=h.

:.::e::-:~~~jrr: ':-:'t=F '.~c::t= ". h+T1 It+==--.::±

o .... -- 1. .. - - - L.. tn:: ,f+. 1+4+ i Lg+- . !~~ . . ' .
o 10,000 20,000 30,000 40,000
FEED CONCENTRATION, ppm NoCI
 659

concentrated, 1t increases and higher pressures are required to achieve accept-

able fluxes.

Real membranes are partially permeable. Thus there will be some solute
flux through the membrane. Neglecting concentration polarization, the flux of
solute is
(12-21)

where Ksolutc is the permeability of the membrane to the solute, and c p is the
permeate concentration. Conservation of mass requires that

(12-22)

which relates the two flux equations. Experimentally the rejection coefficient

(12-23)

is often reponed. Combining Eqs. (12-20a) to (12-23), we obtain the rejection

coefficient RO with no concentration polarization

RO = Ct.(~p - ~1t) (12-24a)

1+ Ct.(~p - ~1t)

Figure 12-12. Reverse osmosis results (Caracciolo et ai, 1977). a. Effect of

pressure and feed concentration on production from a DuPont
B 10 hollow fiber RO system for desalinating water. Produc-
tivity at reference conditions (referred to as Table 1) is 1500
gpd. b. Effect of feed concentration and recovery on salt pas-
sage for a DuPont B 10 hollow fiber RO system for removing
NaCI from water. Salt passage at reference (referred to as
Table 1) is 1.5 wt%. Reprinted with permission from S.
Sourirajan (Ed.), Reverse Osmosis and Synthetic Membranes
(1977). Copyright 1977, National Research Council of
Canada, Onawa.
660

where a is the selectivity,

(12-24b)
a=KsolvlKsolute

For a perfectly impermeable membrane the selectivity is infinite and R= 1. For

membranes with finite selectivity some increase in the observed rejection
coefficient can be obtained by increasing ~p. This also increases the solvent
flux. This pressure effect is illustrated in Figure 12-12a (Caracciolo et ai,
1977) for a hollow fiber membrane.

Concentration polarization is usually important in reverse osmosis.

Equation (12-20) remains valid except we must now calculate ~1t using the
concentrations across the membrane. Referring to Figure 12-4,
~1t=1t(cw)-1t(cp). Since cw>Cb, M increases and solvent flux decreases as con-
centration polarization increases. Equation (12-21) changes since the bulk fluid
concentration Cb is replaced by the wall concentration Cw. This is usually writ-
ten as
(12-25)

where M is the polarization modulus defined as

(12-26)

As M increases the solute flux increases since any solution which leaks through
the membrane is more concentrated.

The solvent flux, Eq. (12-20a), can be written in terms of M if we assume

that 1t is proportional to concentration (Eq. (12-18b». This makes the osmotic
pressure difference

and Eq. (12-20a) becomes

(12-27)
 661

Note that this predicts that solvent flux decreases as M increases. Combining
Eqs. (12-22), (12-23), (12-25) and (12-27), we obtain an expression for the
rejection coefficient when there is concentration polarization.

(12-28)

As expected R decreases as M increases except when the selectivity ex is

infinite. Equation (12-28) reduces to Eq. (12-24a) when M=1. Additional
manipulations of these equations are illustrated in Example 12-2.

Example 12-2. Reverse Osmosis of Na2C03

An experimental RO membrane is being used to produce a waste stream

which is Na2C03 solution. The pure water flux at 30°C is measured as
J so1v = 1500 L/m2 day when hop =25 abn. With the Na2C03 solution,
operation is at 30°C with Ph = 50 abn and PI.. = 1 abn. Both permeate
and feed sides of the membrane are assumed to be perfectly mixed.

a. For an ideal membrane (R = 1) find Jso1v when M = 1 and when M

= 1.2.

b. For a real membrane we measure cp = 0.005 when M = 1. Find

J so1v when M = 1 and when M = 1.2.

Solution
A. Define. The problem is clearly defined in the problem statement.
B. and C. Explore and Plan. Although the definition is clear, the path
to solution is not. We obviously need to determine osmotic pressure.
Equation (12-17) is accurate. Vapor pressure data (perry and Green,
1984), p. 3-73) at 30°C is:

wt% Na2 C03 0 5% 10% 15% 20% 25% 30%

VP, mm Hg 31.82 31.2 30.4 29.6 28.8 27.8 26.4

662

From the pure water flux we can find Ksolv/t.n from Eq. (12-20a). Then
for part a J so1v is obtained from Eq. (12-20a) when M = 1.0 and from
Eq. (12-27) when M = 1.2. For the real membrane with M = I, cp can
be found from Eq. (12-23). Then Jso1v is determined from Eq. (12-20a).
We can also use the measured value of R to find a = K.olvlK.olute from
Eq. (12-24a). This gives us K.olute/tn,. This value is needed when M =
1.2. Combining Eqs. (12-22), (12-25) and (12-27) we can solve for Cp.
Once Cp is known Eq. (12-27) is used to find Jso1v .

D. Do It. Eq. (12-17), X = -RT-

vwater
In [ VP
VPwater
.
solution
1
(MW)w 18.016 glgmole
V water = Pw = 0.996547 glee = 18.095 cc/gmole

For 5 wt % solution at 30°C (Cb =0.05),

[ 82.057 cc atm ] (303.16K)

gmole K [ 31.824]
x= In =27.2 atm
(18.095 ee/gmole) 31.2

Assuming linear dependence, Eq. (12-18b),

c
x = 0.5 (27.2) = 544.5c, or a = 544.5

Part a. Ideal (R = 1). For pure water Eq. (12-20a) becomes

K.olv _ J so1v _ 1500 _ 60 L/( 2 da

- -------- ,m yatm )
t.n Ap 25

When M = I, Ap = 50 - 1 =49 atm, Ax = 27.2 atm (x (c p ) = 0). Note

Ax is the same everywhere since the membrane module is perfectly
mixed.

Jsolv = - - (Ap - Ax) = (60) (49 - 27.2) = 1306.6 LIm day

Ksolv 2
tn,
 663

Ksolv
When M = 1.2, J solv =- - [~p - M 1t (Cb)]
fro
J so1v :;;:; (60) [49 - (1.2) (27.2)] =979.9 L/m2 day
Pan b. Real Membrane. For M = 1 we observe c p = 0.005. Rearrang-
ing Eq. (12-23a), we obtain

R =1- .2 = 1 - 0.005 = 0.9

Cb 0.05

For a linear dependence of osmotic pressure,

1t (cp ) = (0.1) 1t (5%) = 2.72 attn

Equation (12-20a) can be written as,

(12-20b)

This equation is valid everywhere since Cb and cp are constants.

Jso1v = 60 [49 - (27.2 - 2.72)] = 1469.9 L/m2 day

Note that solvent flux of the real membrane is greater than that of the
ideal membrane since ~1t is decreased.

Rearranging Eq. (12-24a) to solve for a,

Ksolv R (12-24c)
a = Ksolute = -:-----:---:-:---=:-:-
(~p - ~1t) (1 - R)

Then, a = (24.~)~0.1) = 0.367

Ksolute = Ksolv/tm = ~ = 163.32 L
fro a 0.367 m2 day
664

For M = 1.2 we need to solve Eqs. (12-22), (12-2S) and (12-27) simul-
taneously. The resulting equation is

(12-29)

where a is defined in Eq. (12-18b). Equation (12-29) is easily solved

with the quadratic formula. The three constants for the quadratic for-
mula are

a Ksolv = (S44.4) (60) = 32.669

t.n
and

K.olv ( A
- - LlP-
M 1t (» Ksolute
Cb + - -
t.n t.n
= 60 [49 - (1.2) (27.2)] + 163.3 = 1143.2
and

K.olv
- - - M Cb = - (163.32) (1.2) (O.OS) =- 9.8
1m

From the quadratic formula,

- 1143.2 ± ....j(1143.2)2 - 4(32,669) (- 9.8)

cp = 2 (32,699)

To have JX>sitive cp must use + sign. cp = 0.00712. Then from Eq.

(12-27) and (12-15),

J I01v = 60 {49 - [(1.2) (27.2) - (0.00712) (S44.5)]} = 1212.S

JlOlute = (163.3) [(1.2) (O.OS) - 0.00712] = 8.64

 665

E. Check. Can use Eq. (12-22) as one check. For Part b with M = 1.2
(the most difficult problem),

8.64 = Jso1ute = cp Jso1v = (0.00712) (1212.5) = 8.63 which is OK.

F. Generalize. There are several parts of this detailed example:

1. Note that the units on the gas constant R controls the units of 1t.

2. The value K.olv/lm determined from data can be used under dif-
ferent conditions.

3. As M increases Jsolv decreases.

4. Note that Jsolv increased as R decreased since permeate osmotic

pressure increased.

5. K.oluteltm calculated from observed retention can be used under

different conditions.

6. Calculation of cp by simultaneous solution of mass balances when

R < 1.0 and M > 1.0 using previously determined values of
K.olute/tm and Kaolvltm, differed from the method used in Example
12-1 since different information was given.

7. If we do not have perfect mixing, Cb and c;, will both vary. Thus
1t (Cb) and 1t (c;,) vary which means the flux varies as we go down
the membrane. Mass balances for other situations are illustrated in
Section 13.3.

8. If the cut e = Fp/Fin is specified, then the required feed concentra-

tion can be determined from the analogue to Eq. (12-8),

(12-30)

9. This system was a continuous, steady state process. Batch opera-

tion is considered in Eqs. (12-43) to (12-46).
 666

Determination of the polarization modulus M is important When rejec-

tion is complete (R=l), a steady state, one-dimensional mass balance on the
feed side of the membrane is (see Figure 12-4),

(12-31)

This equation is a statement that the movement of solute to the membrane by

convection (the bulk flow through the membrane) is equal to the movement of
solute away by diffusion. At the membrane wall (y=O) c equals cw , and in the
bulk fluid which starts at y=O, c equals Cb' With constant diffusivity the solu-
tion to Eq. (12-31) is

(12-32a)

Since D/o = k, the mass transfer coefficient, this result is often written as

M=-=exp
CW
Cb
-S01V
- [J 1
k
(12-32b)

Although considerably over-simplified and difficult to use without a value for

film thickness 0, or mass transfer coefficient k, these results do show that con-
centration polarization increases as:

1. Solvent flux increases. Thus highly permeable membranes have

more of a problem with concentration polarization.

2. The boundary layer thickness 0 increases. The boundary layer

thickness can be decreased by decreasing the channel diameter or
width or by promoting turbulence.

3. The diffusivity increases. This shows that large solutes will have
worse concentration polarization.

More detailed theories are discussed in Chapter 13.

 667

The effect of feed concentration on the productivity of an hollow fiber

RO permeator was shown in Figure 12-12a (Caracciolo et ai, 1977). This con-
centration effect is due to concentration polarization and the increased osmotic
pressure at higher feed concentrations. Salt passage as a function of feed con-
centration and the recovery (which is the same as cut) is shown in Figure 12-
12b (Caracciolo et ai, 1977). As feed concentration or recovery increase, the
salt passage increases. This would be expected, but the nonlinear increase
illustrated is a surprise. This effect is due to the worsening of concentration
JX>larization at increasing concentrations. Typical rejection coefficients for a
variety of membranes are listed in Table 12-2 (Pusch and Walch, 1982). More
detailed tables for a variety of inorganic and organic compounds are compiled
by Eisenberg and Middlebrooks (1986).

The spiral wound and hollow fiber configurations are commonly used for
RO. The common membrane polymers include cellulose acetate, polyamides
(see Figure 12-3), and a composite membrane developed by FilmTec. The
composite membrane is a crosslinked polyamide barrier layer supported by a
microporous polysulfone coating on a JX>lyester web. This complex membrane
has high flux, high salt rejection, a pH range from 4 to 11, and a wide range of
temperature stability. Other membranes are listed in Figure 12-3 and Table
12-2.
The skin of the RO membranes have pores which are several Angstroms
or larger in diameter. These pores are larger than the hydrated ions, yet the
ions are excluded. The separation is clearly not a sieving effect. The current
physical model of the separation effect can be summarized as follows (Kesting,
1985). The water molecules are preferentially attracted to the hydrophilic
membrane surface. An ordered water sheath of thickness t (- ~) exists at the
membrane surface and repels ions. This water sheath pours into pores. If the
pore is less than 2t (- 10~) in size the ordered water sheath will be continuous
and ions will be excluded. In the occasional large pores (imperfections) with
diameter > 2t, the ordered water sheath does not fill the pore. In the center of
these pores ordinary water plus ions pass. The observed rejection coefficient is
the sum of the almost pure water from the small pores and the contaminated
water from the larger pores. These membranes are often called "solution-
diffusion" membranes.
 0\
0\
00

Table 12-2. Rejection coefficients and fluxes for a variety of RO membranes (Cadotte et al, 1988; Pusch
and Walch, 1982). HF = hollow fiber, PF =plate and frame, SW = Spiral wound.

Membrane Module Manufacturer R Flux Test

% L/m2 day Conditions
Cellulose-3 acetate HF DowChem 97 200 28 bar, 0.3% NaCI
pH 6-8, 35°C
Cellulose-2.5 acetate PF DDS ~99 ~400 40 bar, 1% NaCl
pH 2-8, 30°C
Cellulose-2.5 acetate SW Millipore 90 14 bar, 0.1 % NaCI
Cellulose 2-acetate HF Toyobo 93 30 bar, 0.2% NaCI
pH 3-7, 30°C
Cellulose acetate Tube UOP 97 450 40 bar, 0.5 % NaCI
pH 3-7, 45°C
Cellulose-2.8 acetate SW UOP ~99 400 70 bar, 3.5% NaCl
pH 4-7.5, 35°C
Aromatic Polyamide (B9) HF DuPont 95 50 28 bar, 0.15% NaCl
pH4-11,35°C
Aromatic Polyamide (B 10) HF DuPont 98.5 40 56 bar, 3.5% NaCI
pH 5-9, 35°C
Composite PF FilmTec 98 1800 40 bar, 1% NaCI
$60o C
Composite, polyamide SW Film Tech 99.5 20 gfd 1.4 :rv1Pa, dilute NaCI
FT30
Polyfuran, composite SW Osmonics 99 200 56 bar, pH 0.5-10.5
75°C
Composite SW Toray 99.7 700 400 bar, 0.25% NaCI
pH 4-12, 40°C
Composite SW UOP 99.2 600 63 bar, 3.5% NaCI
pH 2-12, 45°C
Composite, polyamide SW Film Tech 50 (NaCI) 20gfd 0.7 :rv1Pa, dilute
XP45 (Nanofilter) 97.5 (MgS04 )
Composite, polyamide SW Film Tech 75 (NaCI) 20gfd 0.4 MPa, dilute
NF70 (Nanofilter) 97.5 (MgS04 )

0-
0-
\0
670

As shown in Figure 12-12a the required pressure depends on the feed

concentration and also on the recovery of water. For brackish water (1500
mg/L dissolved solids) 1t is about 104 kPa (15psi) and a typical pressure range
is from 2,760 to 3, 137 kPa (400 to 600 psig). For a typical seawater of 35,000
mg/L dissolved solids the osmotic pressure is about 2,415 kPa (350 psi). Since
a 50% recovery will approximately double the osmotic pressure, operating
pressures from 5,515 to 6895 kPa (800 to 1000 psig) are often used for seawa-
ter desalination (Applegate, 1984).

The effect of recovery on salt passage was illustrated in Figure 12-12b.

Increasing the recovery decreases the average water flux since the average salt
concentration on the high pressure side of the membrane is raised. This, in
tum, increases the osmotic pressure and decreases the flux. Typical recoveries
are about 33% for seawater and up to 90% for low salinity brackish water
(Kesting, 1985). In under the sink units for treating tap water recoveries as low
as 10% are used. These low recoveries decrease the purchase price of the unit,
but raise the operating cost since more water is thrown away. However, the
average consumer is much more aware of the initial purchase cost than the
water costs. Calculating the effect of recovery or the product concentrations
requires a knowledge of the flow geometry, the degree of concentration polari-
zation, and the external mass balances. Details of the calculation procedures
are in Chapter 13.
The reverse osmosis module needs a variety of supporting equipment to
function properly (Applegate, 1984; Cheremisinoff and LaMendola, 1983;
Eisenberg and Middlebrooks, 1986; Hwang and Kammermeyer, 1975; Porter,
1979; and Sourirajan, 1977). This is illustrated in Figure 12-13 where a water
treatment plant is illustrated. The raw water must be stored to avoid temporary
interruptions in supply. If required, chemicals will be added to adjust the pH,
control biological growth, prevent calcium carbonate or other scaling, or aid in
coagulation and settling. For very dirty waters a settling basin and/or a prefilter
would be added to Figure 12-13. The water is then sent to a micron sized filter
to remove particulates which could clog the membrane. Then a high pressure
pump is used to boost the pressure. The pressure control valve is used for pres-
sure control and the high pressure switch is for safety in case a valve is inad-
vertently closed or a pipe becomes clogged. A parallel-series arrangement of
 671

Permeate

Product
Storage
F

Pressure Bypass
Control
Valve Waste Water
to Disposal

Figure 12-13. Complete RO system showing auxiliary equipment

penneators with bypass lines is shown in Figure 12-13. The bypass lines are
used to mix in raw water if less pure water is required. The water which does
not pass through the membranes is more concentrated in the dissolved solids.
Usually, this stream is a waste stream which must be disposed of. The product
water is sent to a storage tank where additional chemicals may be added. For
example, if the water is for drinking chlorine or ozone will be added here.
Chlorine is detrimental to the membranes and thus is added after the mem-
branes unless it is required to stop biological growth. When treating highly
fouling materials such as food products, a cleaning cycle must be added. This
requires shut-down of the equipment, and it requires quite a bit of hardware in
addition to that shown in Figure 12-13. Details of cleaning cycles and the
chemicals used for cleaning are given by Eisenberg and Middlebrooks (1986).
Unfortunately, this can greatly increase the costs.

Membranes have a finite lifetime which depends upon the operating con-
ditions. Membrane replacement is usually treated as an operating cost. In
addition, there is usually a long tenn flux decay caused by membrane compac-
tion and fouling of the membrane. To take this into account the initial flux of
the membrane is usually not used in the design calculations. Instead the flux
after a few months of operation is used since the additional decay after this
time is quite slow. Obviously, the membrane must be protected from chemical
upsets since the membranes can be destroyed quite quickly.

There are, of course, other ways to purify water such as distillation,

adsorption, ion exchange, and electrodialysis. The method used depends upon
 672

the economics of the particular case. The economics of RO are often favor-
able, and RO is used very extensively for treating brackish water, seawater, and
tap water. One of the largest RO units is at Ras Abu Janjur, Bahrain, which
produces 12.7 million gallons of potable water per day from brackish well
water. The system consists of seven trains of hollow-fiber RO units with 2100
membrane modules per train. RO applications and economics are discussed by
Belfort (1984), Cheremisinoff and LaMendola (1983), Applegate (1984), and
Eisenberg and Middlebrooks (1986).

Reverse osmosis has also been used for separation of organics from
water. Although this is not yet a common industrial separation, it may be
important in particular cases. For example, dilute £-caprolactam solutions are
concentrated commercially using RO. If the organic recovered dissolves poly-
mers or if the module needs to be steam sterilized, ceramic membranes can be
used (Hsieh, 1988). Data is presented by Eisenberg and Middlebrooks (1986),
Leeper (1986), and Sourirajan (1977).

Nanofiltration is a variant of reverse osmosis using membranes which

pass small solutes but retain somewhat larger molecules such as sucrose.
Nanofiltration membranes would typically have NaCI rejections from 20 to
70% and organic cutoffs in the 200 to 500 MW range. (Cadotte et ai, 1988)
This cutoff range corresponds to molecules with a diameter of approximately
10 ~ which is one nanometer. These membranes are thus significantly looser
than RO membranes but tighter than ultrafiltration membranes (minimum cut-
off about 10,000 MW). The nanofiltration membranes operate at pressures in
the range 0.4 MPa to 0.7 MPa which is from 1/4 to 1/2 a seawater RO mem-
brane. Some results are given in Table 12-2. The rejection mechanism is the
same as for RO, and the membranes can be thought of as loose RO membranes.

There are many applications where the ability to discriminate between

salts and moderate molecular weight organics is useful. For instance,
nanofiltration membranes can desalt whey or sugars. Nanofiltration mem-
branes can also be made with 50% rejection of NaCI and 97.5% rejection of
MgS04 • These then are an alternative to water softening by ion exchange (see
Section 9.5). Nanofiltration is a new development which can be expected to
grow in the future.
 673

Osmotic concentration can be considered a variant of osmosis. In

osmotic concentration a dilute stream we wish to concentrate is separated from
a concentrated salt solution by a reverse osmosis membrane such as cellulose
acetate. Due to osmosis water migrates from the dilute stream to the concen-
trated salt stream. This concentrates the dilute stream and simultaneously
dilutes the salt stream. The energy for this process comes from the energy
required to evaporate water to reconcentrate the salt solution. If a waste brine
stream (say from an RO plant) is available, than the energy required for con-
centration may be practically free.

12.5. ULTRAFILTRATION (UF)

Ultrafiltration (UP) is used for retaining larger solutes than in RO, but many of
the operating characteristics are very similar to RO. The osmotic pressure in
UP is invariably quite low since the molar concentration of the high molecular
weight molecules separated by UP is quite low even when the weight concen-
trations are high. Because a large osmotic pressure does not have to be over-
come, much lower operating pressures are used in UP than in RO. Typical
operating pressures are in the range from 10 to 100 psig.
The membranes used in UP are usually asymmetric polymers used in one
of the geometries shown in Figure 12-5. A variety of polymers such a cellulose
acetate, aromatic polyamide, polysulfone, polyvinyl chloride, polyacrylonitrile
and polycarbonate have been used to cast membranes (see Figure 12-3).
Polysulfone has a wide temperature tolerance (to 93° C) and a wide pH range
(0.5 to 13) and is probably the most commonly used membrane material
(Applegate, 1984). The membranes are formed so that the dense skin is
microscopically porous. Thus the separation mechanism can be considered (in
a somewhat over-simplified form) to be sieving of the molecules. The mem-
branes can be tailor made to fix the desired pore size distribution and thus pro-
vide for different molecular weight cutoffs. The molecular weight cutoff is
loosely defined as the molecular weight below which species begin to pass
through the membrane. The values for molecular weight cutoffs can range
from 1000 to 80,000. Values for a variety of membranes are listed in Table
12-3. Molecular weight cutoffs should be employed with discretion since siev-
 0\
--.I
~

Table 12-3. Molecular weight cutoffs and fluxes ofUF systems. (pusch and Walch, 1982; and
Warashina et ai., 1985). HF:;: hollow fiber, PF:;: plate and frame, SW:;: spiral wound

Membrane Module Manufacturer MW Flux Test

L/m2 day Conditions

Cellulose 2-aeetate SW Abeor IS,OOO 6000 3.S bar, SOoC,

pH 3-7.S
Polyvinylidene fluoride SW Abeor 10,000/20,000 7S00 3.S bar, 90°C,
pH 2-9
Cellulose acetate PF DDS 6000 to 6S,000 2-7000 S bar, SO°C,
pH 2-8
Polysulfone PF DDS 8000/20,000/ 1-7000 3 bar, 80°C,
6S,000 pH 0-14
Polysulfone PF Dorr-Oliver 10,000 10,000 2 bar, 70°C,
pH 2-12
Polysulfone SW Millipore 80,000 2000 5 bar, 32°C,
pH 3-11
Polysulfone PF Osrnonics 1000/20,000 4000/12,000 3.5 bar, 93°C,
pH 0.5-13
Cellulose acetate Tube Patterson Candy 1000-20,000 pH 3-6, 30-S0°C
Co-poly acrylonitrile/ PF Rhone-Poulenc 20,000 7000 2 bar, 40°C
Methallyl sulfonate pH 1-10
Polyacrylonitrile HF Asahi Chern. 13,000 6380 to 19,100 3 kg/crn 2 , 25C
pH 2-10
Polyacrylonitrile HF AsahiChern. 6,000 14,900 3 kg/crn 2 , 25C
pH 1-14

0'1
-.I
VI
 676

Approximate molecular dlometer, A

08

.: 06
>
c
Q)

~ 04

02
, Nominal molecular welCJht
exclusion of membrane

Molecular welCJht
Figure 12-14. Solute retention on Amicon Diaflo membranes (porter, 1979).
Reprinted with permission from P.A. Schweitzer (Ed.), Hand-
book of Separation Techniques for Chemical Engineers (1979).
Copyright 1979, McGraw-Hill.

ing depends on size and shape of the hydrated molecule, and molecular weight
is only a rough guide to size and shape. In addition, there is considerable reten-
tion of molecules of lower molecular weight. Roughly, a sharp cut-off UP
membrane will have R = 0 when the molecular weight is 1(2 the molecular
weight which has R = 0.9. Diffuse cutoff membranes have a much broader
range of fractionation. Typical retention data are shown in Figure 12-14
(porter, 1979). A comparison of Tables 12-2 and 12-3 is interesting. Note that
the UP membranes with their much larger pores have considerably larger fluxes
at lower pressures than the RO membranes. Details of UP membranes are dis-
cussed by Blatt (1976), Cooper(1979), Kesting (1985), Lloyd (1985), Porter
(1979), and Pusch and Walch (1982).

Concentration polarization, gel formation and fouling are extremely

important in UP and need to be included in any detailed analysis. We will first
look at the equations ignoring these effects, and then include them. Solvent
flux without concentration polarization can be written as Eq. (12-lOa or b).
 677

For the large molecules separated by UP the molar concentration will be quite
low even if the weight concentration is high. Thus, the osmotic pressure differ-
ence ~1t will be small and is usually ignored. The resulting solvent flux equa-
tion is,
Kso1v
J I01v = - - ~p
(12-33)
tm
For small and intennediate macromolecules (5000 or 100,000 daltons) it may
be necessary to include the osmotic pressure tenn. This is discussed further in
Chapter 13.

For sieve type membranes flow of both solvent and solute is in pores.
Let R represent the fraction of solvent flux carried by pores which exclude the
solute. Then l-R of the solvent flux is in pores which do not exclude solute.
With no concentration polarization the liquid passing through these pores will
have a solute concentration Cb. This gives a solute flux of
(12-34)

In addition, Eq. (12-22) must remain valid since it represents a mass balance.
Combining Eqs. (12-22), (12-33) and (12-34) and solving for R, we obtain
Eq.(12-13). Thus in sieve type membranes the solute rejection can also be con-
sidered as the fraction of solvent flux in pores which reject solute molecules.

This analysis predicts that without concentration polarization the solvent

flux is proportional to ~p and is independent of the feed concentration. In addi-
tion, the rejection coefficient should be constant These results are a reasonable
first approximation, but are drastically oversimplified. R is usually constant for
very dilute systems, but may decrease for more concentrated systems. Some
pores may pass solute molecules slowly so that the concentration in the pores is
less than Cb, but the cut-off is not perfect.

Concentration polarization can cause an increase in solute concentration

as shown in Figure 12-4, or it can cause a gel or cake layer to fonn. Assum-
ing that there is no gel or cake layer fonned, the solute flux equation becomes

(12-35)
J I01uIe =MCb (1 - R) J I01v
678

y=8

Figure 12-15. Ultrafiltration system with gel layer at membrane wall.

where M is again the concentration polarization mooulus, Eq. (12-26). Solving

Eqs. (12-22) and (12-35) simultaneously, we obtain the apparent rejection
coefficient Rapp '

cp (12-36)
R
app
= 1- -
Cb
=1- M (1 - R)

Unless R=1 (all pores completely reject solute), concentration polarization

decreases the apparent rejection coefficient. When M=I, R.pp=R. This non-
gelling concentration polarization is predicted to have no effect on the solvent
flux since osmotic pressure is negligible.

Unfortunately, many of the .materials which the engineer desires to

ultrafilter (paint pigments, foods, polymers, minerals, and so forth) form gels,
cakes or slimes at the wall. This gel cake or slime can form at concentrations
cg below 1 wt% for polysaccharides to 50 vol% for suspensions of polymer lat-
tices (Blatt et al, 1970). The concentration profile for UP with gel formation is
shown in Figure 12-15. This gel layer acts as a porous solid which causes an
additional resistance to flow. The solvent flux is now,

l\p
J lalV = --.-!;....-- (12-37)
_tm_+_ls
Kaalv Kg
 679

where tg is the thickness of the gel layer and Kg is the permeability of the gel
layer. Since 19 can vary throughout a run, Eq. (12-37) is usually not useful for
predictions. Often, the gel layer will have far more resistance to flow than the
membrane and the gel will control the solvent flux. In these cases it does not
matter what type of membrane is used as long as it retains the solute and causes
a gel to form. The value of Kg can be estimated from the Carman-Kozeny
equation (Blatt et ai, 1970).

(12-38)

Note that the gel permeability will be very sensitive to the particle diameter ~
in the gel layer. If e=O.5 and the viscosity is 1.0 cp, then Kg = 3xlO-ll for one
micron particles and Kg = 2x10- 16 for 30 Angstrom particles. Thus a one
micron layer of these gels with a pressure drop of 100 psi across the gel would
have solvent fluxes of 200 and 0.0013 cm 3 /cm 2 sec, respectively. Obviously, a
gel layer can have a major effect on the solvent flux.

A simplified steady state, one-dimensional mass balance is still given by

Eq. (12-31). Now the concentration at the wall (y =-tg) is Cw =c g • The boun-
dary conditions for the mass balance (see Figure 12-15) are at y = 0, C =c g and
at y =0, C =Cb. The solution to this is

(12-37)

Since cg and Cb are constants, the solvent flux is the variable. This is different
from Eq. (12-32) where Cw was unknown. The solvent flux is set by the rate at
which the solute can diffuse back from the gel layer to the bulk fluid. Increas-
ing the pressure drop will cause a brief temporary increase in the solvent flux.
However, during this period of increased flux transfer of solute to the gel layer
is greater than the rate of transfer of solute back into the fluid. The gel layer
must then become thicker. According to Eq. (12-37), J I01v will then decrease.
This decrease in J I01v continues until the gel layer is thick enough so that the
solvent flux has obtained the value predicted by Eq. (12-29). In other words,
680

the engineer has no control over the flux rate except by increasing the dif-
fusivity (increase temperature) or decreasing ~ (stir or use thin channels). For
high molecular weight polymers diffusivities are quite low and the solvent flux
is low if gels form. The mass transfer coefficient k can be estimated for many
flow geometries. This analysis is done in Chapter 13.

Example 12-3. Protein Ultrafiltration.

We wish to ultrafilter a 2 wt % aqueous solution of casein at 25°C. The

casein gels with cg = 0.14 .. We do two experiments as follows:

Experiment a: Run pure water and obtain Jsolv = 7000 L/m2 day with
~p= 5 atm.

Experiment b: Run casein solution at ~p = 5 atm and obtain J501v =

233 L/m 2day, R = 1.
a. Calculate the mass transfer coefficient k and Kg/tg for Experiment b.
b. If we increase ~p to 7 atm, determine the peak value of Jsolv , the
final value of J solv, and the final value of Kg/tg.

Solution

a. From the pure water data we can find Ksolv/1m from Eq. (12-33).

K.olv = J so1v = 7000 = 1400 L

1m ~p 5 m2 day atm

From the casein data and Eq. (12-39),

~. n3 L
k = In (Cg/Cb) = In (0.14/0.02) = 119.74 m2 day

Rearranging Eq. (12-37) and using the casein data,

-tg = -~p - -1m- = - 5 - -1- = 0.0207

Kg J so1v Ksolv 233 1400
 681

Thus, Kg/tg =48.20 L/m2 day atm

b. If we increase AP to 7 atm, we will have a momentary increase in

water flux since tg does not increase immediately. Using Kg/tg found in
part a and Eq. (12-37) we have,

Jso1v = ___ 7_ _ _ = 326.2 L/m2 day

1 1
--+--
1400 48.2

Mter a short period of time, tg increases and J so1v is given by Eq. (12-
39). Since k, c g and Cb are unchanged from part a, J so1v = 233. Solving
for tg/kg

-tg = -Ap - -1m- = - 7 - -1- =0.0293

Kg Jsalv K.olv 233 1400

and Kg/tg = 34.096. Thus, tg has increased by 40.8%.

Note that in both parts the resistance of the gel, fglKg, is significantly
greater than the membrane resistance, tmlKm. This is commonly the
case when proteins are ultrafiltered.

Although simple and very appealing, there are significant problems with
the gel layer model (Fane, 1986; Porter, 1979; Wijmans et aI., 1984). The con-
centration cg found by extrapolation may depend upon equipment geometry.
The measured cg may be too low and independent experiments show no gel-
ling. In addition, c g may be estimated as > 100% which is physically impossi-
ble. The mass balance, Eq. (12-31) does not include convection in the axial, x,
direction and the boundary condition at y = 5 is hypothetical. However, the
model is useful for correlating data. Modifications of this model are discussed
in Chapter 13.

The presence of a gel layer may have conflicting effects on the rejection
of solutes. If the gel layer is very viscous, but remains fluid so that molecules
682

diffuse in the layer, solute molecules will continue to pass through the mem-
brane. Since Cw>Cb, the fluid passing through pores which do not reject solute
will be concentrated. This will cause the apparent rejection coefficient to drop.
This situation is similar to concentration polarization without a gel layer. The
second case occurs when the gel layer has a solid-like nature and solute
molecules are essentially immobile. Now the rejection coefficient can increase
since the gel layer traps the solute molecules and very few of them can diffuse
into the pores.

The fonnation of a gel layer also has a large effect on the separation of
mixtures containing several solutes of widely different size. Suppose we wish
to separate a low molecular weight compound from a very high molecular
weight polymer. When run separately, the low molecular weight compound
has R=O while the polymer has R=l. What happens when the mixture is
ultrafiltered? Our first guess is probably to use superposition and assume each
molecule is unaffected by the presence of the other molecule. If no gel layer is
fonned, this assumption is probably close to true. Then the small molecules
have R=O and the polymer has R=1. The separation is easy. If a gel layer
fonns, the gel serves as a membrane which has different retention characteris-
tics than the original membrane. The gel layer may retain the small molecules.
If this happens both solutes may have retentions near 1.0 and separation will
not be obtained. Thus gel layers change the fundamental nature of UF.

Once fonned, the gel layer may not come off easily. In this case the
membrane is said to be fouled. Fouling often occurs when the solute is
adsorbed to the membrane or when charges OR the membrane attract colloidal
particles of opposite charge. The entire UF installation is often controlled by
the presence of gel layers and fouling. Fouling can also occur from precipita-
tion of salts, and is discussed in detail by Fane (1986) and Potts et al. (1981).

In non-fouling or moderate fouling applications hollow fiber and spiral

wound configurations are commonly used. The hollow fibers have a 0.1
micron skin on the inside of the membrane. The fibers are from 500 to 1,100
microns inside diameter. Since pressures used in UF are much lower than in
RO the membranes do not have to withstand very high pressures. Thus the
high pressure liquid can be on the inside of the membrane. This has the advan-
 683

tage of reducing concentration polarization since velocities are higher and

boundary layers thinner than for systems with the high pressure liquid on the
outside of the fibers. The systems can be cleaned by washing with high velo-
city clean water or with chemical solutions. In many applications a dilute caus-
tic wash is a very effective cleaning agent if the membrane can withstand the
high pH. Hollow fiber membranes can also be cleaned by back-washing with a
clean liquid. Back-washing forces liquid through the porous support and then
through the skin. Any particles attached to the skin will tend to be washed off
into the bulk fluid.

In highly fouling applications plate-and-frame and tubular membranes

are often used. In food plants the membranes are cleaned daily and sometimes
every few hours. Obviously, the cleaning cycle becomes very important in
controlling costs. Biotechnology applications of UP with an emphasis on pro-
tein recovery are discussed by Belter et al (1988).

All of the cascade arrangements shown in Figure 12-8 can be used.

When gel formation or fouling are severe, the cascade design as well as the
module design are controlled by the concentration polarization. Designs which
keep liquid velocities on the feed side of the membrane high are favored.
These are the parallel-series and recycle cascades shown in Figures 12-8 c and
d. In particular, the recycle system is often used since it is readily adaptable to
the batch operation shown in Figure 12-8g which is required when cleaning
must be frequent. Recycle systems are often cascaded in series to increase the
recovery of the permeate or in parallel to increase capacity.

The batch system shown in Figure 12-8g is easily analyzed for simple
cases. A mass balance on the entire system (tank + pipes + UP module) is

dV (12-40)
-=-AJ
dt 1
SOY

Where V(t) is the volume of liquid in the tank + pipes + UP module. If osmotic
pressure is negligible and no gel forms, Jso1v is given by Eq. (12-33). Then

dV A Ksolv .1p (12-41)

-=-----
dt t.n
684

The initial condition is V =V 0 for t =O. Assuming that ~p is kept constant, we

obtain the solution to Eq. (12-41).

(12-42)

If a gel forms then J. olv is a constant given by Eq. (12-39). Now Eq.
(12-40) becomes
dV (12-43)
- = -A k In(c ICb)
dt g

Since R = 1, the total moles of solute in the system, n, must remain constant

(12-44)
n = c(t) V(t) = constant

Then Eq. (12-43) becomes

dV (12-45)
""(it = - A k In (cg VIn)

The initial condition remains V = V 0 when t = O. The solution to Eq. (12-45) is

+ 1
(3)(3!)
[(In V)3 - (In V )3] + ...
o = -A kt In (cg/n)
(12-46)

which may be inconvenient to use since convergence may be slow.

If osmotic pressure is important, 1t is proportional to concentration, and

no gel forms; then Eq. (12-27) should be used for Jso1v with 1t ('1» = 0('1> = 0
since R = 1) (Belter et ai, 1988). Assuming 1t is linear with respect to concen-
tration, 1t =ac. Since R = 1, Eq. (12-44) remains valid. Then,

(12-47)
 685

Substituting Eq. (12-47) into Eq. (12-40), we obtain

dV
dt
= _ A K.olv l\p
1m
[1 - [al\pMn 1~V 1 (12-48)

with the initial condition V = V0 when t = 0. The solution to Eq. (12-48) is (see
Problem 12-C4),

V _ V _ a Mn In aMn
vo - - -
l\p
[ V _ a Mn
1= _ A
K.olv
l\
P t (12-49)
o l\p 1m
l\p

Note that when a = 0, Eq. (12-49) reduces to Eq. (12-42). Equation (12-49) is
also valid for batch operation of RO when R = 1.

Ultrafiltration and reverse osmosis systems are often used together. For
example, cheese whey is the left over liquid from making cheese. The whey is
approximately 1% protein, 5% lactose (milk: sugar), 1% salts, and the
remainder is water. The whey has a very high Biological Oxygen Demand
(BOD) and is expensive to dispose of properly. One processing route which
utilizes all of the components of whey is shown in Figure 12-16. The protein is
recovered by UP. This protein concentrate is quite valuable and can either be
dried and sold or it can be added wet to cottage cheese or ice cream. The per-
meate from the UP system is sent to RO where the sugar and salts are concen-
trated by removing water. This is desirable since it makes both the downstream
processing steps smaller and cheaper. The concentrated solution can then be
fermented with yeast to produce a dilute ethanol solution. The ethanol is
recovered by distillation or other separation scheme. The waste liquor contain-
ing about 3% salts is sent to evaporators to recover the salts which can be used
as fertilizer. Note that the UP system is first. The UP system serves to protect
the RO system from particulates and protein which can cause fouling and gel
formation. Both UP and RO systems would be cleaned at least once a day.

Somewhat more complex combined UP and RO systems can be used for

 686
Protein
Concentrate

F
r====t-- Permeate
(Water)

Storage

Fermentation
Tanks ''----95wt%
Ethanol

Disti Ilation
Water Vapor

Evaporators
L---=r- Stea m

Conc.
Salt
Solution
Figure 12-16. Simplified flow diagram for whey processing.

processing fruit juices (Stana, 1977). In concentration of fruit juices the pro-
cessor wants to recover all of the pulp, the sugar, and the volatile flavor
ingredients. The final product may be from 20 to 25% pulp and 20 to 25%
sugars. This product is very viscous and has a very high osmotic pressure.
(Open a can of frozen orange juice and let it melt. This is the desired final pro-
duct) A high recovery of sugar is desired; thus, the permeate leaving the RO
system should be close to zero concentration. A 50 wt% solution of sugar has a
very high osmotic pressure (see Table 12-1), and any leakage through the
membrane will cause significant contamination of the waste stream. Thus a
single RO module will be difficult to use and the membrane arrangement used
in Figure 12-16 for whey processing is not a good scheme. Instead, the pro-
cessing scheme shown in Figure 12-17 can be used (Stana, 1977). Note that
 687

Concentrate

40-50%
Pulp

Raw
Juice
UF

Recycle 10-15% sugar

Figure 12-17. Scheme for processing fruit juices. (Stana, 1977).

the UF system is again placed first to protect the RO systems. Since the UF
membrane does not retain the sugar, the sugar concentration is the same in the
permeate and the high pressure products. Thus, although 7t is significant, d7t is
approximately zero and the usual UF equations can be used. The first RO sys-
tem produces the desired low concentration waste water, but the sugar solution
is concentrated to only 25%. This keeps M reasonable. The second RO sys-
tem produces the desired sugar solution, but the permeate is a 10 to 15% solu-
tion which is recycled to the first RO system. This keeps d7t modest in the
second RO system also. The pulp and sugar are then mixed to produce the
desired final product.

Other applications of UF include removal of colloidal matter in water

treatment, pigment recovery in painting operations, concentration of oil emul-
sions, depyrogenating pharmaceutical process water, and recovery of dyes.
Applegate (1984), Eykamp and Steen (1987), and Klinkowski (1978) discuss
these applications.

12.6. DIALYSIS

Dialysis is a process where small molecules diffuse through a membrane

because of a concentration driving force. Large molecules are excluded. The
process differs from UF since the solute movement in dialysis is caused by dif-
fusion while in UF the major flux is that of solvent caused by pressure gra-
dients. In UF the small molecules are carried along with the solvent flux. In
dialysis the flux of small molecules is independent of the solvent flux and may
688

be in the opposite direction from the solvent flux. Dialysis also usually does
Rot use asymmetric membranes, but instead uses homogeneous membranes
without skins.
The major commercial use of dialysis is hemodialysis for removal of
waste products such as urea and creatine from patients who have had kidney
failure. This application is commonly known as the artificial kidney even
though hemodialysis is not a good mimic of the way the kidney functions (Col-
ton and Lowrie, 1981; Ward et ai, 1985, Lysaght et ai, 1986). Other commer-
cial applications include the recovery of caustic in the manufacture of rayon,
salt removal in pharmaceuticals manufacturing, and recovery of spent acid in
the metal industry. Except in medical applications, the engineer is more likely
to be involved with RO, UP, or electrodialysis than with dialysis. Both plate-
and-frame and hollow fiber dialyzers are available (Lacey, 1979; Klein et ai,
1987).
As small solutes pass through the microporous membrane, concentration
gradients will develop on both sides of the membrane. This is illustrated in
Figure 12-18. The flux of the permeable solutes through the membrane can be
written as
Ksolute (12-50)
J solute = (c m h - cm '!)
lm .

where 'K.oIute is the permeability of the membrane to the nonexcluded solute.

As shown in Figure 12-18, the concentration driving force is the difference in
solute concentrations across the membrane. If both sides are very well mixed
the membrane concentrations will be the same as the bulk concentrations. In
most systems there will be a resistance to mass transfer on both sides of the
membrane and concentration profiles will build up. When this occurs, the
values of Cm.h and cm,! will not be known. An overall driving force is more
convenient The usual mass transfer expression is (Hwang and Kammermeyer,
1975)
(12-51)

where Cb and Cd are the bulk fluid concentrations on the feed and dialysate
sides, and k~lute is the total or overall mass transfer coefficient. If the solvent is
 689
Flux of small
--l-+- molecules
- - - c b Small molecules
Dialysate
Side
Feed Side

- - - - - Macromolecules
Membrane
Figure 12-18. Concentration gradients in dialysis.

stagnant, the overall mass transfer coefficient can be found from a sum of resis-
tances.

1 1m 1 1 (12-52)
--- +--+--
k;olute - Ksolute k~olute k~olute

where k!olule and k~lute are the mass transfer coefficients on the feed and
dialysate sides of the membrane. The individual mass transfer coefficients can
be estimated from standard correlations such as those in Chapter 13. The three
terms on the right hand side of Eq. (12-52) are often the same order of magni-
tude.

As illustrated in Figure 12-18 there can be concentration polarization of

macromolecules if the solute flux is high enough. In some cases this may be
severe enough to form a gel which can then interfere or stop the transfer of
small molecules. The concentration polarization can be controlled by using
thin channels or turbulence.

Solvent flux (see Eq. (12-27» can be in either direction in Figure 12-18.
If the pressures on both sides of the membrane are equal, the solvent will flow
from the dialysate to the feed side because of osmosis. This osmotic flow can
be drastically increased by the concentration polarization of macromolecules,
and in turn will tend to reduce the concentration polarization. Unfortunately,
the osmotic flow also decreases the flux of the small solutes since the convec-
tive flow is in the opposite direction to the solute diffusive flux. The solvent
690

flow can be stopped or reversed by applying a positive pressure on the feed side
of the membrane. Dialyzers are often operated with essentially no solvent flux,
or with a small flux of solvent into the dialysate.

Any of the flow patterns shown in Figure 12-7 can be used in a dialyzer.
Hemodialyzers are usually countercurrent If the volumetric flow rates of
liquids on the feed <IF and dialysate sides Qd and the overall mass transfer
coefficient kT can be assumed to be constant, a variety of steady-state flow case
can be solved (Michaels, 1966). The correct average concentration difference
in Eq. (12-51) is the logarithmic mean of the inlet and outlet concentration.

(12-53)

where 1 and 2 are the two ends of the dialyses. For example, if countercurrent
flow is used,
(12-54a)

(12-54b)

Solving Eqs. (12-51), (12-53) and (12-54) with the mass balance
(12-55)

where NA is the amount transferred

(12-56)

we can obtain the result for a counter-current dialyszer.

(12-57)
 691

In the derivation of Eq. (12-57) the volumetric flow rates Qp and Qd were
assumed to be constant. At equilibrium, cP,out = Cd,in' Thus the denominator on
the LHS of Eq. (12-57) is the maximum possible concentration change in the
feed stream while the numerator is the actual concentration change. The LHS
of Eq. (12-57) is thus the fraction of maximum solute removal which is
attained.

Results for other geometries are also easily obtained Michaels (1966).
For parallel flow of feed and dialysate the result is,

I-exp -~ [1+ ~1 (12-58)

=
1 + Qp/Qd

If the dialysate is completely mixed and the feed is in plug flow, the
result is

(12-59)

The counter-current system is superior. Dialysis is also used in batch processes

(Klein et ai, 1987).

Membranes used include cellulose, cellulose acetate, and polyacryloni-

trile, but cellulose has been most common. For hemodialysis (dialysis of
blood) isotropic cellulose hydrogels with a molecular weight cutoff of approxi-
mately 1000 are the most common membranes. The membranes are fonned by
the cuprammonium process and the membranes are homogeneous and do not
have a skin. These membranes easily pass the low molecular weight waste pro-
ducts such as urea (MW=60) and creatinine (MW=I13), but do not pass high
molecular weight proteins which are desirable. In addition, the membranes will
 692

pass amino acids and salts which it is desirable to retain. This latter problem is
solved by using a dialysate fluid which is a physiological electrolyte solution
containing the appropriate concentrations of the salts. Since salt concentrations
are the same on both sides of the membrane, there will be no transfer of the
salts. In hemodialysis a positive pressure is applied to the feed (blood) side in
order to remove some water (which can be considered the solvent). Equipment
used in medical applications must satisfy a large number of constraints such as
preventing blood clotting. Details of hemodialysis are discussed by Colton and
Lowrie (1981), Lysaght et al (1986), and Ward et al (1985). Other membrane
separations used in artificial organs are discussed by Lysaght et al (1986).
In the usual dialysis process exclusion of large solutes is based on steric
effects. In ion exchange dialysis anion exchange membranes are used. The
membranes are the same as those used in electrodialysis (see the next section).
These membranes will exclude cations (ions with positive charge), but the
exclusion of H+ cations is poor (see Chapter 9 for a discussion of Ion
Exchange). Thus, if a solution of acid and metal salts is on one side of the
membrane and water or a dilute acid solution is on the other side of the mem-
brane, the anions (ions of negative charge) readily transfer through the mem-
brane. To keep electroneutrality cations must also transfer. Since the only
cation which can transfer through the membrane is H+, the acid is separated
from the metal ion. This is illustrated in Figure 12-19a. Ion exchange dialysis
is used commercially to recover metal ions in the plating industry. The systems
use plate-and-frame arrangements similar to those used in electrodialysis
except no current is required. Water and feed solutions are alternated in the
different compartments. Other geometries could also be used.
Donnan dialysis is another variant using ion exchange membranes. The
stack utilizes all cation or anion exchange membranes. For purposes of this
explanation we will assume that all of the membranes are cation exchange
membranes since our purpose is to concentrate a valuable cation. A dilute
solution containing the valuable cation (for example Cu++) is circulated in the
odd-numbered compartments. In the even-numbered compartments a concen-
trated solution of cheap acid is circulated. One cell is illustrated in Figure 12-
19b. The W ions will transfer through the membranes due to the concentration
driving force. To keep electroneutrality, either the anion must transfer through
 693

a Dilute b dilute conc.

Acid
* +
! CuS0 4 acid

l
Metal
Salt

Anion conc.

l
+ dilute CUS04
H+
H2 S0 4

Metal
Salt Acid
t \ Cation

\
Exchange

~ Membrane

* Anion
Exchange
Membrane

Figure 12-19. Special dialysis systems. a. Ion exchange dialysis. b. Donnan

dialysis.

the cation exchange membrane (which it cannot do because of Donnan

exclusion), or Cu++ must transfer through the membrane in a direction opposite
its concentration driving force. Thus the valuable cation (Cu++) is concentrated
at the price of diluting the acid. The copper can be concentrated from 30 ppm
Cu S04 to as high as 4000 ppm CUS04' The cost of the separation is the dilu-
tion of the sulfuric acid.

12.7. ELECTRODIALYSIS (ED)

Electrodialysis is dialysis in the presence of an electrical field. The driving

force for electrodialysis is the electrical potential difference. This driving force
pushes cations through cation exchange membranes and anions through anion
exchange membranes. The arrangement of membranes shown schematically in
Figure 12-20 is used. Cation and anion exchange membranes are alternated. In
694
Desalted solution

NaCI solution to be treated

A= anion - permeable membrane
C = cation - permeable membrane

Figure 12-20. Schematic of electrodialysis.

the presence of the electrical field the cations will migrate towards the cathode
and anions towards the anode. The cations can pass through the cation
exchange membranes but not through the anion exchange membranes. The
anions can pass through the anion exchange membranes but not through the
cation exchange membranes. Because of the alternation of the cation and anion
exchange membranes the result is to concentrate ions in the odd numbered
compartments and to dilute the ions in the even numbered compartments in
Figure 12-20. In this way a desalted water and a concentrated brine are
formed. In commercial systems from 100 to 600 unit cells (pairs of mem-
branes) are in a stack. Reviews of electrodialysis are given by Applegate
(1984), Hwang and Kammermeyer (1975), Klein et al (1985), Komgold
(1984), Lacey (1979), Mintz (1963), Solt (1976), and Spiegler (1984).

Ion exchange membranes can be thought of as ion exchange resins which

are formed as membranes. The same polymers used for ion exchange resins
(see Chapter 9) are used for some of the commercially available ion exchange
membranes. The most common cation exchange membrane has been a sul-
fonated polystyrene cross linked with divinylbenzene. The same basic polys-
tyrene backbone is used for anion exchange with a CH2~(CH3h group cou-
pled to the benzene ring. Other membranes such as Nafion (Figure 12-3g) are
 695

used in special cases. Ion exchange membranes are discussed in detail else-
where (Kesting, 1985; Flett, 1983; Klein et al, 1987; Komgold, 1984; Lacey,
1979; Pusch and Walch, 1982; Solt, 1976; Strathmann, 1985). Lists of sup-
pliers and membrane properties are given by Komgold (1984), Spiegler (1984),
and Strathmann (1985).

The usual reaction at the cathode is

(12-60a)

The OIr fonned must be neutralized in the cathode compartment. The

cathode can be made from tantalum, niobuim, or titanium all of which must be
coated with platinum. In the anode the usual reaction is

(12-60b)

while if Cl- is present the reaction will be

(12-60c)

The anode can be made from the same materials as the cathode or from stain-
less steel. The anode and cathode compartments are separated from the stack
by membranes. These compartments are vented to remove the gases fonned.

Since the stack is in series, the current flow through each membrane and
each compartment must be the same. The reason for using a large number of
unit cells is to use this current many times and thus produce more desalted
water. The current is carried by the ions. The fraction of current carried by an
ionic species is the "transference number". This fraction can be different for
the different ions since current is the product of the charge times the velocity
times the number of ions transferred. Small ions such as W will carry more
current than large ions such as Cl- since the H+ have a higher velocity. In a
solution such as KCI where the ions are approximately the same size and thus
have the same velocity, the transference numbers will be approximately equal.
 696
cations onions

I I
I Well Mixed 1

I Zone I
1 I
1 1 1 1

18 81 18 81
1 1
Ion Concentration
I I
I I
Cation Anion
Exchange Exchange
Membrane Membrane

Figure 12-21. Concentration polarization in electrodialysis.

In solution in the comparonents both ions carry current. In the mem-

branes one ion is excluded; thus, the other ion must carry all of the current.
The transference number of the cation in the cation exchange membrane is
essentially 1.0 while its transference number is essentially zero in the anion
exchange membrane. Suppose that K+ is carrying half the current in the com-
paronent. Suddenly, in the cation exchange membrane K+ must carry all of the
current. This requires that more K+ be transferred through the membrane than
is transferred in solution. Referring to Figure 12-21, we see that the cations are
diluted on the right hand side of the cation exchange membrane and concen-
trated on the left hand side of the membrane. Since concentration gradients
have been developed, there will be diffusion of the cation towards the cation
exchange membrane on the right hand side of this membrane (see Figure 12-
21). This diffusion provides part of the cations needed to carry the current
through the membrane. On the other side of the membrane the cations diffuse
away from the membrane. Since electroneutrality must be satisfied, the total
ion concentration will be diluted when the cation concentration is diluted and
concentrated when the cation is concentrated Exactly the same phenomena
occurs at the anion exchange membrane where the anion transfers all of the
current. This picture of concentration polarization is oversimplified since it
does not include osmotic flow of water and other effects (Hwan¥ and Kammer-
meyer, 1975).
 697

Table 12-4. Basic electrical equations and units

V(volts) = I(amps) R(ohms) (12-61a)

I(amps) = Q(charge in coloumbs) /t(s) (l2-61b)

E(joules) = V(volts) I(amps) t(s) (l2-61c)

F = 96 500 coloumb = 96 500 amp sec (l2-6ld)

, equivalent ' equivalent

Concentration polarization has important ramifications for the operation

of electrodialysis. If one increases the current density (current per area), then
the rate of electrical transport also increases. This will require that the dif-
fusive transport must also increase. This requires an increase in the concentra-
tion gradient and the ionic concentrations at the membrane surface will
decrease. If these concentrations approach zero at the membrane, the ions will
be unable to carry all the current. This point is called the limiting current den-
sity. The resistance at the membrane surfaces will be high and energy require-
ments will increase. At this point water will be ionized and W and OH- will
transfer through the membranes to carry the current. This changes the pH in
the compartments, and represents energy which is not used to separate the salts
from the water. The changes in pH can cause precipitation of salts. The
transfer of OH- through the anion exchange membrane can cause dimensional
changes in the membrane. Because of these adverse effects, operation is con-
ducted at a current density below the limiting current density. This upper limit
on the current density puts a lower limit on the area of membranes which must
be used.

A brief review of electrical equations and units is given in Table 12-4.

The energy consumption in electrodialysis can be calculated by combining Eqs.
(12-61a) and (12-61c) for n cells.

(12-62)
E=J2nRt
698

where E is the energy consumption in joules, I is the current in amps, n is the

number of unit cells in the stack, R is the resistance per unit cell in ohms, and t
is the time in second. The resistance in the dilute and concentrated parts of the
cell differs. Thus, R is the average cell resistance. Alternately, we can calcu-
late E from,
(12-63)

if each chamber is completely mixed. The resistance is usually approximately

inversely proportional to concentration,
(12-64)
R=b/c

where b is a constant, and c is in equivalents/liter. After some algebraic mani-

pulation, we obtain

(12-65)

where R(CF) is the resistance per cell determined at the feed concentration. For
example, if we recover a product water which is one half as concentrated as the
feed this result is

The current requirements can be determined from the required concentra-

tion changes.
(12-66)

where F is Faraday's constant which is 96,500 coulombs/equivalent, Q is the

volumetric flow rate of solution in liters/sec, and ~ is the current utilization.
The concentration difference must be in gram equivalents per liter. The most
important term in this equation is the current utilization~. The current utiliza-
tion is directly related to the number of cells n in the stack and the efficiency of
utilization of the current.
(12-67)
~ = nll sllw 11m = n (Elec. Eff.)
 699

where n is the number of unit cells, 11, is the efficiency due to the semipennea-
bility of the membrane (co-ions transfer through a membrane that they should
be excluded from), l1w is the efficiency related to water transfer through the
membrane, and 11m is the efficiency since some of the current invariably leaks
through the manifold holding the membranes. All of the efficiencies are less
than 1.0 and the overall electrical efficiency is often about 0.9. However, n
often varies from 100 to 600. Thus, ~ will usually be significantly greater than
1.0. An important exception to this occurs as the feed concentration becomes
high (3 to 5 mol/liter). As this happens 11, and hence Elec. Eff. approach zero
since the coions are not strongly excluded from the membrane. At these high
concentrations ~ becomes small and energy requirements become too large for
electrodialysis to be economical.

The membrane area A for the cation and anion membrane can be
estimated from

(12-68)

where i is the current density in amp/cm 2 • The membrane area A is related to

the area of each individual membrane Am by

(12-69)
A=nAm

The area Am is set by the dimensions of the plate-and-frame module which are
standardized by the manufacturer. Substituting in

(12-70)

plus Eqs. (12-67) and (12-69) into (12-68), we obtain

(12-71)
n=
1115 11m l1w

Mintz (1963) presents a more detailed analysis.

700

The major commercial application of electrodialysis has been for desalt-

ing brackish water (1000 to 5000 mgIL of total dissolved salt) to produce pot-
able water (less than 500 mgIL). ED is also used for producing salt from sea-
water in Japan, for desalting cheese whey, and for recovering metals from plat-
ing baths. In water treatment ED appears to be competitive with RO and ion
exchange in the range from 1000 to 3000 mgIL IDS. Typical ED systems
operate with current densities from 6 to 20 milliamp/cm 2 • The cell would typi-
cally be 40 cm wide by 100 cm long and 0.1 cm thick (Mintz, 1963). The sys-
tem would be designed for 50% demineralization. One major advantage of ED
is that significantly less pretreatment is usually required than will be required
for RO or ion exchange. The raw water does need to be filtered before the ED
system, and the water may need to be treated chemically.

Many ED plants use polarity reversal to reduce scaling caused by precip-

itation of salts. Polarity reversal involves periodically switching the polarity of
the electrodes so that the cathode becomes the anode and vice versa. This
switches the direction of ion transfer. The pmpose of this is to prevent scaling
by causing precipitated salts to dissolve when the concentration drops after the
direction of ion transfer is reversed. Unfortunately, when the polarity is
reversed both streams will have a high salt concentration and must be sent to
waste for a minute or two. In a commercial ED system using polarity reversal
approximately 13% of the production is lost. Thus the polarity reversal system
must be larger than a normal ED system. Since scaling can also be prevented
by adding acid, the economics of polarity reversal depends upon acid costs.
Since more than 500 of the 1000 commercial ED plants use polarity reversal, it
appears to be economical in many cases (Applegate, 1984).

Example 12-4. Electrodialysis of Brackish Water

a A laboratory electrodialysis unit is available to determine cell

resistance and electrical efficiency for desalting a NaCI solution.
The following experiment is done: Q = 30 L/min, n = 50, CF = 20
gIL, Cdil = 0.8 gIL. We measure I = 21.5 amp and E = 60,670
joules in one minute. Determine the resistance per unit cell at eF,
and the electrical efficiency.
 701

h. Assuming that R and electrical efficiency are unchanged for a

commercial unit, design a system for Q = 300 L/min going from Cr
= 2.0 gIL to Cdil = 0.49 gIL if ~ = 0.4 m2 and i = 20 amp/m2.
Find I, n, and E in 1 minute.

Solution
a Rearranging Eqs. (12-66) and (12-67),

(EI Eff) = F Q (CF - Cdil)

ec.. NI

Since MWNaC1 = 58.45,

- g [ 1 equiv ]_ eq
CF - Cdil - (1.2 L) 58.45 g - 0.0205 L

96,500 am.: s 30 ':;n ][~i: ][0.0205 i ]

(E1ec. Eff.) = -"-____ .L.....>~_~~_~"___ _____"_

(50) (21.5 amps)

= 0.920
Assuming that cell resistance is inversely proportional to concentration
Eq. (12-65) is valid. Rearranging,

R(CF)~ t~n [2:F~ - [c: r]

R C _ 60,670 joules
( F) - (60 s) (21.5 amp)2 (50)
2(0.8) -
[ 2.0
[.2.!]2]
2.0

= 0.028 ohm/unit cell

702

Note that the units work since

joule = (volt) (amp) (s) = (amp)2 (ohm) (s)

b. We can use Eq. (12-70) to find I.

Then from Eq. (12-71),

[96,500 amp sec] 300 ~ 1 min] [2.0 - 0.49 e q ]

eq mm 60 sec 58.45 L
n=------------~----~~----~~--------~
(8 amp) (0.92)

where 118 11m 11n = elee. eff. = 0.92. n = 1693.6 or 1694.

From Eq. (12-65),

(8)2 (1694) (60) (0.028)

E= = 423,604 joules
2 (0.49) _ [0.49]2
2.0 2.0

Since there are 300 Ljmin, this is 1412 joules/L = 337.5 callL.

Notes: 1.) One of the hardest parts of ED calculations is the use of

unfamiliar units. Thus it is important to cheek units and
refer to Table 12-4.

2.) Equations (12-62) to (12-66) assume that electrode resis-

tance is small compared to nR. If n is small, this may not
be true. n = 50 is probably large enough.

3.) If the cell design and size or spacer design are changed,
R and Elec. Eff. may not be the same for laboratory and
commercial units. Also, if the high concentrations are
very different 115 may be different in the two units.
 703

12.8. PERVAPORATION

Pervaporation is a membrane process where there is liquid on one side of the

membrane and a vapor on the other side. The liquid diffuses through the mem-
brane and vaporizes. This is illustrated in Figure 12-22. The partial pressure in
the vapor phase is usually reduced by pulling a vacuum or using a sweep gas.
Any of the modular geometries shown in Figure 12-5 could be used although
the hollow fiber and spiral wound configurations are most likely. Unlike the
other membrane processes discussed up to now, pervaporation is not currently
a well established separation method. However, the method has considerable
promise and commercial units are available (Sander and Soukup, 1988).

We usually want the least concentrated species to penneate preferentially

through the membrane. This reduces the required membrane area and the load
on the vacuum pump. For removal of trace organics rubbery polymers such as
polyether block amide (PEBAX) are used. For removal of water from azeo-
tropes such as the ethanol-water azeotrope either glassy polymers or ion
exchange membranes are used.
Theoretically, pervaporation is the most complex membrane separation
process. On the liquid side concentration polarization may be important partic-
ularly for the removal of trace organics from water. The membranes operate
by a solution-diffusion mechanism. Since the polymer is highly swollen on the

VP Vacuum

2
---p

Figure 12-22. Profiles for pervaporation.

 704

liquid side and dry on the vapor side, the diffusivity varies markedly across the
membrane. The evaporation at the vapor side may have a major effect on the
observed selectivity. For the separation of trace organics from water, the rela-
tive volatility of the trace organic is often very high even though the boiling
point of the organic is higher than water's boiling point. This occurs because
these are very nonideal solutions with high activities. Finally, because of eva-
poration the process is nonisothermal. The temperature change affects solubil-
ity, diffusivity, and evaporation. Currently, a complete model including all
these effects is not available.

Pervaporation must be analyzed by using a concentration dependent dif-

fusivity. For the one dimensional system shown in Figure 12-22 the flux is

(12-72)

Since flux is proportional to dcJdx, the steady state form of Fick's second law
becomes,

(12-73)

Boundary conditions are

(12-74)
'1 = '1,1 at x=O , Ci = Ci.2 at x=tm

Concentration polarization on the liquid side may be important, and must be

estimated to calculate Ci,l' To solve this set of equations a functional form for
the diffusivity as a function of all solutes is required.

Unfortunately, solution of these coupled equations is very difficult and

has only been done analytically for a pure fluid (Li and Long, 1969; Hwang and
Kammermeyer, 1975). This has been done for an exponential dependence of
diffusivity on concentration which is appropriate for polymer membranes.

(12-75)
D(c) =Do exp (ac)
 705

The parameters Do and a are functions of temperature and the properties of the
polymer.

Ifequilibrium follows a form of Henry's law, at the wall

(12-76)

where S* is the Henry's law constant and is a form of solubility. Then the flux
is

(12-77)

Note that Eq. (12-77) does not fit the standard form given in Eq. (12-1). If we
force fit pervaporation into the form of Eq. (12-1) using ~P as the driving force,
we obtain a "permeability" P of

J1m
P= - = -
Do
[exp(aS "Pl)-exp(aS "P2)]
(12-78)
~P Mp
This result shows that the permeability is a function of the partial pressures, in
addition to the diffusivity and the solubility. Since both solubilty and dif-
fusivity usually increase as the temperature increases, the permeability and per-
meation rate increase with temperature increases. The selectivity between dif-
ferent species may decrease as temperature increases, although for commercial
alcohol-water pervaporation there is little decrease in selectivity (Sander and
Soukup, 1988).

For multicomponent systems the logical first attempt is to assume super-

position. Then the partial pressures become

(12-79)
Pz = YiP

The flux and concentration profiles are assumed to be unchanged by the pres-
ence of the other diffusing components. Unfortunately, superposition is seldom
valid. The presence of a second component affects the permeation rate of the
 706

30

25

.s
u
20

...."
c:
-2 15
"<-
"
0-
Il
til
o 10

o
Weight fraction of benzene in liquid
o = permeation 40° e • =equilibrium 40 0 e
t:l. = permeation 5Qoe • = equilibrium 50° C
o = permeation 60 0 e • = equilibrium 60 0 e

Figure 12-23. Separation factors for separation of benzene and isopropyl

alcohol mixtures through polyethylene by pervaporation. (Car-
ter and Jagannadhaswamy, 1964). Reprinted with permission
from Symposium on The Less Common Means of Separation,
Institution of Chemical Engineers, London, 1984.

first component. Analysis becomes significantly more complicated and is

beyond the scope of this book. The selectivity can be a complex function of
concentration and temperature. This is illustrated in Figure 12-23 (Carter and
Jagannadhaswamy, 1964). Note that selectivity can increase or decrease as
mole fraction changes and as temperature increases depending on the system.
The selectivity can be significantly higher than the vapor-liquid equilibrium
selectivity.

If selectivity and flux are known from laboratory data, we can design the
 707

Figure 12-24. One pass pervaporation system.

pervaporation system without knowing the details of the diffusion. Design

does depend upon the details of the flow geometry. The simplest case, com-
pletely mixed liquid and vapor (Figure 12-7a), will be considered here. Hwang
and Kammermeyer (1975) can be consulted for the analysis for other flow
geometries.

A schematic of a pervaporation system is shown in Figure 12-24 for a

binary separation. Fin, FL and Fp are flow rates in moles/hr while z, X out and yp
arc mole fractions of the more permeable component. Defining the cut e as
0= Fp/Fin, the mass balances will be similar to Eqs. (12-5) to (12-7). The
result is

-
z-Oyp (12-80)
Xout - I-a

Since a is usually determined based on energy balance considerations (see Eqs.

(12-88) to (12-90b), Eq. (12-80) is one equation with two unknowns.

An additional equation can be obtained from selectivity data, a~

(12-81)

Note that this definition is similar to the definition of relative volatility in

vapor-liquid equilibrium although pervaporation is not an equilibrium system.
Solving for y, we obtain

(12-82)
708

Equations (12-81) and (12-82) apply to local values at the same location of the
membrane. For a perfectly mixed system

(12-83)
yP =Y , Xout = X

and Eq. (12-82) becomes

(12-84)
yP = 1 + (ClAB
, - 1) Xout

The global character of this expression for perfectly mixed systems greatly
simplifies the remaining analysis.

Equations (12-80) and (12-84) can be combined. Upon clearing fractions

this becomes

(12-85)

This equation can obviously be solved for yp with the quadratic formula (see
Example 12-5). Equation (12-85) is linear in z and 9; thus, if yp and 9 are
known

(12-86)

or if yp and z are known

Cl~ (Cl~ 1) z Yp - Yp
9= (12-87)
Z- -
, 2
(ClAB - 1) (Yp - yp)

These equations are obviously valid and easy to use if Cl~ is constant.
However, Figure 12-23 shows that Cl~ is a function of liquid mole fraction x.
The equations are still valid, but must be used in a trial-and-error fashion. For
 709

example, if z and e are given we can proceed as follows:

1. Guess Xout.guess

2. Detennine (l~ (Xo.lt ) from data.

3. Calculate yp from the solution of Eq. (12-85).

4. Calculate Xout,calc from Eq. (12-80).

5. Check: Is Xout.guess = Xout,ca1c? If not, return to step 1.

The value of the cut is often controlled by the energy balance. For the system
shown in Figure 12-24 the energy balance is,

where A is the latent heat of vaporization. With thennal equilibrium in a com-

pletely mixed system, Tp = Tout. Since the reference temperature is arbitrary,
we can pick T ref = Tp = Tout. Then, the energy balance simplies to

(12-89)

which is

(12-90a)

or

(12-90b)

The high temperature, Tin, is limited by the stability of the membrane. The low
temperature, Tout = T p' is limited by the need to have a vapor on the penneate
side. Thus, if T p is decreased a very low pressure may be required. Since
latent heats are significantly greater than specific heats, the amount of energy
710

Recycle

Permeate Recycle

Product I Product 2

Figure 12-25. Pervaporation system coupled with distillation.

required to vaporize the permeate will not be available in the feed unless the
permeate rate is low. For removal of trace organics permeate rates will be low
and sufficient energy is usually available in the feed. For breaking azeotropes
conentrations are usually significantly higher and heat effects become
important. A recycle stream and/or interstage heaters (see Figure 12-25) are
used when necessary to provide sufficient sensible heat for the vaporization.
Additional heat input into the pervaporation unit is provided by the hot distil-
late streams. The value of e per pass is low, but e..r for the entire unit shown in
Figure 12-25 can be high. It is desireable to keep the temperature drop quite
modest (5 to lO°C) since fluxes are significantly higher at higher temperatures
(Sander and Soukup, 1988). Thus a large number of stages or high recycle rate
is desirable. The use of these equations for a one-pass system is illustrated in
Example 12-5. Recycle systems are left to Problem 12-C3.

What are the possible advantages of pervaporation as compared to other

membrane processes? In pervaporation the driving force .1p can be increased
by decreasing the downstream pressure by drawing a vacuum or using a sweep
gas. In liquid permeation similar increases in the driving force usually require
high upstream pressures. To use gas permeation the entire vapor stream must
be vaporized instead of just the permeate as in pervaporation. If the feed is
already vaporized, gas permeation will be a competing process. Pervaporation
 711

often has much larger selectivities but lower fluxes than RO. For trace oganics
the high activity of the organic gives high relative volatilities and thus both
high selectivities and high fluxes. In addition, pervaporation will not be limited
by osmotic pressure.

Pervaporation has been extensively studied as a method of breaking

azeotropes, and the method is now available commercially for the ethanol-
water azeotrope (Sander and Soukup, 1988). Leeper (1986) has collected per-
vaporation data for separations of alcohols from water and from a variety of
organics. Some of this data is shown in Table 12-5. It is useful to extract as
much information as possible from this data. For example, the fluxes in Table
12-5 are considerably less than those in Tables 12-2 and 12-3 for RO and UP.
Fluxes and selectivities in Table 12-5 vary significantly. In general, high selec-
tivity membranes will have low fluxes and high flux membranes will have low
selectivities. The cellophane data shows that additives can have a major effect
on the selectivities. The acetone water data shows that downstream pressure
has little effect as long as a vapor is present. Although not shown in Table 12-
5, upstream pressure usually increases flux only slightly and has little effect on
selectivity. Flux usually increases and selectivity usually decreases as tempera-
ture increases although there are exceptions (eg. the cellophane data). Leeper
(1986) also presents RO data for the same separations. Reverse osmosis results
show higher fluxes and lower selectivities.

In breaking azeotropes pervaporation will usually be coupled with distil-

lation columns. One way of doing this is shown in Figure 12-25 (Goldblatt and
Gooding, 1986; Leeper, 1986; Sander and Soukup, 1988). Two or more stages
are used in series. The liquid is heated between the stages. Stages are shown
operating under a vacuum since this technique is more common than using a
sweep gas. Efficient permeate condensation is important since it greatly
reduces the load on the vacuum pump. The vaccum pump normally represents
the biggest energy cost. Refrigeration systems are used commercially to reduce
the load on the vacuum pump (Sander and Soukup, 1988). The permeate recy-
cle is required since permeate is unlikely to be pure. The method shown in Fig-
ure 12-25 could be used for separation of isopropyl alcohol and water. Optimi-
zation of the system shown in Figure 12-25 will be quite difficult.
 -....J
N

Table 12-5. Pervaporation data (Leeper, 1986).

Separation Membrane Cone. A T, of P2, Selectivity Flux

Problem in feed (wt%) mmHg (lAB Ib/ft2/hr

Ethanol (B) Cellophane 50 77 0.94 0.21

from water (A) Cellophane 50 113 2.0 0.43
Cellophane 50(1) 77 24.6 0.08
Cellophane 5(Ji2) 77 0.63 0.12
Cellulose Acetate 50 68 0.1 2.0 0.22
Polyvinylaleohol 10-50 104 0.1 220-75
Polyvinylalcohol 50-90 104 0.1 -75
Polyacrylonitrite 9-94 68 0.1 16-35 0.0001
Polysulfone 50 68 0.1 332 0.001
 Polysulfone, Asymmetric 50 68 0.1 3.0 0.025
Polysulfone, Composite 50 68 0.1 19.0 0.008
Aromatic polyamide 4.5 90-130 0.5 0.59-0.36 1-2
Aromatic polyamide 18.7 70-130 0.5 0.56-0.36 0.3-2
Silicone rubber 8-12 86-140 0.14 0.02-0.05
Silicone rubber 45-46 86-140 0.23-0.22 0.04-0.10
Silicone rubber 81-87 86-140 0.67-0.71 0.04-0.16
n-Butanol(B) Cellophane 10-24 140 20 10-15 0.2-0.4
from Water (A) Cellulose 2.5 acetate 85 140 20 45 0.2

Acetone (A) Polypropylene 45 230 10 2.5 0.094

from Water (B) Polypropy lene 45 230 49 2.8 0.094
Polypropylene 45 230 488 2.8 0.092

(1) Includes 2 wt% sodium citrate. (2) Includes 2 wt% benzoic acid.

-...J
W
714

Distillation column 1 is used to approach the azeotrope concentration

while column 2 is used to distill on the other side of the azeotrope. In some
cases the retentate stream may be of high enough purity that the second distilla-
tion column is not required. This is the case for breaking the ethanol water
azeotrope (Leeper, 1986; Sander and Soukup, 1988). The commercial ethanol
water pervaporator uses a composite poly (vinyl alcohol) membrane which has
a very high selectivity for water. The permeate recycle stream is still required
for this simplified ftowsheet Of course, there are alternatives for separating the
ethanol water azeotrope such as azeotropic distillation, extraction, adsorption,
and gas permeation. The choice will depend on the economics of the particular
case. Sander and Soukup (1988) state that capital costs were higher for perva-
poration than the use of azeotropic distillation, but utility costs were about 1/3
those of azeotropic distillation.

Example 12-5. Pervaporation

We wish to remove water from n-butanol using a cellulose 2.5 acetate

membrane. Feed 90 mole % n-butanol. Feed is at 60°C. Selectivity,
a~ = 43; Flux = 0.2Ib/ff hr. Maximum permissible temperature drop
is 30°C. Find the cut e and the permeate and liquid mole fractions. If
feed rate is 100 lb/hr, find the membrane area.
Data: A.B = 141.6 cal/g, Aw =9.72 kcal/gmole
cp,B (45°C) = 0.625 cal/gOC, Cpw (45°C) = 1.0 cal/gOC
MWB =74.12, MWw = 18.016

Solution

We can find the cut a from Eq. (12-90b). First, we need consistent
units.

For 90 mole % butanol, A. =(0.1) 9.72 + (0.9) 10.5 =10.4.

 715

Since liquid heat capacities are close to linear, cp was estimated at

T.vg = [Tin + Toot J12 =(60 + 30)12 =45°C

c
p.B
= 0.625 caV °C [
g
cal
1000 kca1
1[74.12
gmole
g 1= 0.046 kcal
gmole °C

Then, Cp,F = (0.1) (0.0180) + (0.9) (0.046) = 0.0435.

Eq. (12-83b) gives,

[Tin - Tout J Cp.F _

(30°C) [0.0435 kal
gmoleOC
1
/.. - 10.4 kcal/gmole
= 0.125

Penneate mole fraction can be found by solving Eq. (12-85).

where a = (a~ - 1) e = (43 - 1) (0.125) = 5.25

b = - [1 - e + (a~ - 1) z + a~ e ] = - [0.875 + (42) (0.1) +
(43) (0.125)] =- 10.45
C = a~z =(43) (0.1) =4.3

Note that z = 0.1 refers to water mole fraction since a' = a~. Then yp
is also water mole fraction.

yp = 10.45 ± 4.35
10.5 0581 mo Ie fracllon
=. . water

where the minus sign is used since yp < 1.0.

716

From Eq. (12-80),

= z - e yP = 0.1 - (0.125)(0.581) = 0.031.

x w•out 1- e (1 - 0.125)

l00lb/hr
Area = Feed Rate/Flux = 2 = 500 ft2
0.2Ib/hr ft

Notes: 1. e is small despite the significant temperatme drop. Thus,

recycle is often required.

2. yp is increased significantly, but liquid product still contains

3% water. A cascade or recycle system is probably needed.

3. This separation can also be done by RO, distillation, adsorp-

tion and extraction. The choice will depend on economics.

4. In this example A should be calculated at X<xlt and at the

vacuum pressure. Both these corrections will be small.

There are two alternatives to pervaporation in addition to RO. One is to

do gas permeation with the vapor stream. For reasons that are not completely
understood, this method gives selectivities and fluxes which are significantly
less than those observed for pervaporation. The second alternative is perstrac-
tion which is liquid permeation into a carrier liquid on the permeate side. Per-
vaporation has been more extensively studied than perstraction since recovery
of permeate is normally easier from a vapor than from a liquid. In perstraction
the permeate is usually recovered from the high boiling carrier liquid by distil-
lation. Perstraction may be advantageous in some cases since a vacuum pump
is not required.

12.9. LIQUID MEMBRANES

A liquid membrane has a layer of liquid which serves as the separation medium
instead of the solid polymer used in the other membrane methods discussed
 717

a b ~ Feed Phase
o 00 0
Feed Phase o 00 0 Recelvlng
..
0 0 00 0
o 0 0 Phase
0
Liquid 0

Membrane
o
Liquid Exploded
View
Membrane
Receiving Phase

Figure 12-26. Liquid membrane systems. a Supported liquid membranes. b.

Emulsion liquid membranes. c. Facilitated transport.

previously. Two types of liquid membrane systems have been extensively stu-
died. In supported liquid membranes the liquid is held in a porous matrix
which serves to support the liquid (Danesi, 1984-85; Noble et ai, 1986; Ward,
1970; Noble and Way, 1987; Way et ai, 1982, 1985). The equipment and
operating procedures for supported liquid membranes are very similar to those
for the other membrane separations discussed in this chapter. In emulsion
liquid membranes a double emulsion is formed with the receiving liquid encap-
sulated by an immiscible material which is also immiscible with the outer fluid
(Cahn and Li, 1976; Li, 1971; Marr and Kopp, 1982; Noble et ai, 1986; Way et
ai, 1982). The emulsion liquid membrane systems are operated in a manner
very similar to liquid-liquid extraction. Liquid membrane systems have been
an area of considerable research, but only waste water treatment and gas sen-
sors for ion selective electrodes are commercial (Noble and Way, 1987). In the
future these separation techniques may become important as commercial
separation devices.

Supported liquid membranes are shown schematically in Figure 12-26a.

The liquid membrane phase is held in a porous matrix which is usually a poly-
mer. The liquid membrane itself must be chosen to be immiscible with both
718

the feed phase and the receiving phase. The feed and receiving phases can be
essentially the same liquid and in many cases will be an aqueous solution. The
liquid membrane serves to separate the feed and receiving phases. In order to
increase the capacity of the receiving phase, it will often contain a chemical
which will react with the diffusing solute. The supported liquid membrane sys-
tems can use one of the geometries shown in Figure 12-5. One of the major
problems with this type of liquid membrane is bleeding which is loss of the
liquid membrane as it dissolves into the feed and receiving phases. Supported
liquid membrane systems have apparently not been commercialized yet.
The emulsion liquid membrane system is shown schematically in Figure
12-26b. The system is a double emulsion with the receiving phase distributed
within the liquid membrane drops which are dispersed within the feed phase.
The emulsion type systems are fairly simple to make and have a very large sur-
face area. Normal extraction equipment is used for the contacting. Mter the
extraction, the liquid membrane usually must be broken to recover the solute.
The receiving phase usually contains a chemical to react with the solute to
increase the capacity. Emulsion type systems have been extensively designed
and piloted for a variety of commercial processes such as copper recovery,
phenol removal from waste water, and hydrocarbon separations.
Both types of liquid membrane systems can use facilitated transport
(Cussler, 1971; Goddard, 1977; Noble et ai, 1989). Facilitated transport is
shown schematically in Figure 12-26c. A carrier contained within the liquid
membrane reacts with the solute and then diffuses across the membrane. In the
receiving phase the carrier-solute complex is broken and carrier diffuses into
the receiving phase. The carrier then diffuses back across the membrane to
pick up another "load". This mechanism is similar to the carrying of oxygen in
blood by hemoglobin. In the second mechanism shown in Figure 12-26c a
second molecule of A is back-transported across the membrane. The back-
transported molecule supplies the energy for the process (Cussler, 1971).
Examples of this process are using W to supply energy to separate cations.
Facilitated transport is of interest since it can provide much faster mass transfer
rates with higher selectivities than the liquid membrane alone, and because
facilitated transport appears to be important in transfer across cell membranes
in living systems. In terms of a solution-diffusion membrane, facilitated tran-
 719

sport involves increasing the solubility and/or the diffusion rate of the solute.
Noble et al (1989) list a large number of systems which have been developed
for facilitated transport. Note that these facilitated transport chemical systems
can often be used to advantage in other separation processes such as absorption
or extraction, and these alternatives may be cheaper than liquid membranes.

12.10. SUMMARY -OBJECTIVES

At the end of this chapter you should be able to meet the following objectives.

1. Explain and describe the following membrane separation

processes: Gas permeation, reverse osmosis, ultrafiltration,
dialysis, electrodialysis, pervaporation, and liquid membranes.
Clearly delineate between the different processes.

2. Determine the flux for the different membrane separation tech-

niques.

3. Describe concentration polarization, explain what can be done to

reduce it, and estimate the effect of concentration polarization on
flux and rejection coefficients.

4. Describe gelling and fouling, and estimate the effects on flux and
rejection. Explain why flux is determined by back-diffusion when
a gel layer exists.

5. Explain the effects of flow geometry qualitatively. Do this quanti-

tatively for perfectly mixed systems.

REFERENCES

Applegate, L.E., "Membrane separation processes," Chern. Eng., 91 (12) 164

(June 11, 1984).
720

Baker, R.W. and I. Blume, "Permselective membranes separate gases," Chem-

tech. 16. 232 (1986).

Beaver, E.R., P.V. Bhat, and D.S. Sarcia, "Integration of membranes with other
air separations technologies," AIChE Symp. Ser .• 84 (261), 113 (1988).

Belfort, G. (Ed), Synthetic Membrane Processes. Fundamentals and Water

Applications, Academic Press, Orlando, FL, 1984.

Belter, P.A., E.L. Cussler, and W.-S. Hu, Bioseparations. Downstream Pro-
cessingfor Biotechnology, Wiley-Intersciences, 1988, Chapt 9.

Blaisdell, C.T. and K. Kammermeyer, "Countercurrent and co-current gas

separation," Chem. Eng. Sci., 28. 1249 (1973).

Blatt, W.F., A. Dravid, A.S. Michaels, and L. Nelson, "Solute polarizatio,' and
cake formation in membrane ultrafiltration: Causes, consequences and COl trol
techniques," in J.E. Flinn (Ed.), Membrane Science and Technology, Plenum
Press, NY, 1970,47-97.

Blatt, W.F., "Principles and practice of ultrafiltration," in P. Meares (Ed),

Membrane Separation Processes, Elsevier, Amsterdam, NY, 1976, Chapt. 3.

Cadotte, J., R. Forester, M. Kim, R. Petersen, and T. Stocker, "Nanofiltration

membranes broaden the use of membrane separation technology," Desalina-
tion, 70, 77 (1988).

Cabn, R.P. and N.N. Li, "Hydrocarbon separation by liquid membrane

processes," In P. Meares (Ed.), Membrane Separation Processes, Elsevier,
Amsterdam, 1976, Chapt.9.

Caracciolo, V.P., N.W. Rosenblatt and V J. Tomsic, "Du Pont's hollow fiber
membranes," in S. Sourirajan (Ed.), Reverse Osmosis and Synthetic Mem-
branes, Ottawa, Canada, National Research Council of Canada, 1977, Chapt.
16.
 721

Carter, J.W. and B. Jagannadhaswamy, "Liquid separations using polymer

films," Proceedings Symposium on the Less Common Means of Separation,
Institution Chemical Engineers, London, 1964,35-42.

Cheremisinoff, P.N. and A.R. LaMendola, "Industrial applications of reverse

osmosis," in N.P. Cheremisinoff and D.S. Azbel (Eds.), Liquid Filtration, Ann
Arbor Science, 1983, Chapt. 14.

Chern, R.T., WJ. Koros, H.B. Hopfenberg, and V.T. Stannett, "Material selec-
tion for membrane-based gas separations," in D.R. Lloyd (Ed.), Materials Sci-
ence of Synthetic Membranes, Am. Chern. Soc., Washington, DC, 1985, Chapt.
2.

Colton, c.K. and E.G. Lowrie, "Hemodialysis: Physical principles and techni-
cal considerations," The Kidney, Vol. 2, 2nd ed., Philadelphia, W.B. Saunders
Co. 1981,2425-2489.

Cooper, A.R. (Ed.), Ultrafiltration Membranes and Applications, Plenum Press,

NY,1979.

Cussler, EL., "Membranes which pump," AIChE Journal, 17, 1300 (1971).

Danesi, P.R., "Separation of metal species by supported liquid membranes,

Separ. Sci. Technol., 19, 857 (1984-85).

Eisenberg, T.N. and EJ. Middlebrooks, Reverse Osmosis Treatment of Drink-

ing Water, Butterworths, Boston, 1986.

Eykamp, W. and J. Steen, "Ultrafiltration and reverse osmosis," in R.W.

Rousseau (Ed.), Handbook of Separation Process Technology, Wiley-
Interscience, New York, 1987, Chapt 18.

Fane, A.G., "Ultrafiltration: Factors influencing flux and rejection", in RJ.

Wakeman (Ed.), Progress in Filtration and Separation 4, Elsevier, NY,
101-180 (1986).
 722

Flett, D.S. (Ed), Ion Exchange Membranes, Ellis Horwood Ltd., Chichester,
England. 1983.

Finken, H., "Asymmetric membranes for gas separations," in D.R. Lloyd (Ed.),
Materials Science of Synthetic Membranes, Amer. Chern. Soc., Washington,
D.C., 1985, Chapt. 11.

Gienger, J.K. and R.J. Ray, "Membrane-based hybrid processes," AIChE Symp.
Ser.,84 (261), 168 (1988).

Goldblatt, M.E. and C.E. Gooding, "An engineering analysis of membrane-

aided distillation",AIChE Symposium Series, 82 (248) 51 (1986).

Goddard, J.D., "Further applications of carrier-mediated transport theory - A

survey," Chern. Eng. Sci., 32,795 (1977).

Henis, J. M.S. and M.K. Tripodi, "The developing technology of gas separating
membranes," Science, 220 (4592), 11 (April 1, 1983).

Hsieh, H.P., "Inorganic membranes," AIChE Symp. Ser., 84 (261), 1 (1988).

Hwang, S.-T. and K. Kammermeyer, Membranes in Separations, Wiley, NY,

1975.

Kesting, R.E., Synthetic Polymeric Membranes. A Structural Perspective, 2nd

ed., Wiley, NY, 1985.

Klein, E., R.A. Ward, and R.E. Lacey, "Membrane processes-Dialysis and elec-
trodialysis", in R.W. Rousseau (Ed.), Handbook of Separation Process Tech-
nology, Wiley, New York, 1987, Chapt. 21.

Klinkowski, P.R., "Ultrafiltration: an emerging unit-operation," Chem. Eng.,

85 (11) 165 (May 8, 1978).

Komgold, E., "Electrodialysis - Membranes and mass transpon", in G. Belfon

 723

(Ed.), Synthetic Membrane Processes. Fundamentals and Water Applications,

Academic Press, Orlando, FL 1984, Chapt. 6.

Korns, W J. and RT. Chern, "Separation of gaseous mixtures using polymer

membranes," in RW. Rousseau (Ed.), Handbook of Separation Process Tech-
nology, Wiley-Interscience, NEw York, 1987, Chapl20.

Kremen, S.S., "Membrane systems", in M.B. Chenoweth (Ed), Synthetic Mem-

branes, MMI Press, Midland, MI, 1986,53-87.

Kremen, S.S., "Technology and engineering of ROGA spiral-wound reverse

osmosis membrane modules," in S. Sourirajan (Ed.), Reverse Osmosis and Syn-
thetic Membranes, Ottawa, Canada, National Research Council of Canada,
1977, Chapt. 17.

Lacey, RE., "Dialysis and electrodialysis," in P.A. Schweitzer (Ed.), Hand-

book of Separation Techniques for Chemical Engineers, McGraw-Hill, NY,
1979, Section 1.14.

Leeper, S.A., "Membrane separation in the production of alcohol fuels by fer-

mentation," in W.C. McGregor (Ed.), Membrane Separations in Biotechnology,
Marcel Dekker, NY, 1986, Chapt. 8.

Li, N.N. and RB. Long, "Permeation through plastic films," AIChE Journal,
15, 73 (1969).

Li, N.N., "Separation of hydrocarbons by liquid membrane permeation," Ind.

Eng. Chern. Process Des. Develop., 10, 215 (1971).

Lloyd, D.R (Ed.), Materials Science of Synthetic Membranes, Amer. Chern.

Soc.," DC, 1985.

Lloyd, D.R and T.B. Meluch, "Selection and evaluation of membrane materi-
als for liquid separations," in D.R. Lloyd (Ed.), Materials Science of Synthetic
Membranes, Amer. Chern. Soc., Washington, DC, 1985, Chapt. 3.
724

Loeb, S. and S. Sourirajan, "Seawater demineralization by means of an osmotic

membrane," Advan. Chem. Ser., 38,117 (1963).

Loeb, S. and S. Souriraj an , "Seawater demineralization by means of a semi-

permeable membrane," University of California at Los Angeles Engineering
Report No. 60-60, 1960.

Lonsdale, H.K., "The growth of membranes technology," J. Membrane Sci., 10,

81 (1982).

Lysaght, MJ., D.R. Boggs, and M.H. Taim isto , "Membranes in artificial
organs", in M.B. Chenoweth (Ed), Synthetic Membranes, MMI Press, Mid-
land, MI, 1986,100-117.

Marr, R and A. Kopp, "Liquid membrane technology - A survey of

phenomena, mechanisms and models," Internat. Chem. Eng., 22 (1),44 (1982).

Michaels, A.S., "Operating parameters and performance criteria for hemodi-

alyzers and other membrane-separation devices", Trans. Amer. Artij. Intern.
Organs, 12,387 (1966).

Mintz, M.S., "Electrodialysis, principles of process design," Ind. Chern., 55(6)

18 (June 1963).

Noble, RD., C.A. Koval, and JJ. Pellegrino, "Facilitated transport membrane
systems," Chern. Engr. Prog., 85 (3), 58 (March 1989).

Noble, R.D. and J.D. Way (Eds.), Liquid Membranes, Theory and Applications,
ACS Symp. Ser., No. 347, ACS, Washington, D.C., 1987.

Noble, RD., J.D. Way and A.L. Bunge, "Liquid membranes," in Y. Marcus
(Ed.), Ion Exchange and Solvent Extractions, vol. 10, Marcel Dekker, NY.

Orofino, T.A., "Technology of hollow fiber reverse osmosis systems," in S.

 725

Sourirajan (Ed.), Reverse Osmosis and Synthetic Membranes, Ottawa, Canada,

National Research Council of Canada, 1977, Chapt 15.

Porter, M.C., "Membrane filtration," in Schweitzer, P.A. (Ed.), Handbook 0/

Separation Techniques/or Chemical Engineers, McGraw-Hill, NY, 1979, Sec-
tion 2.1.

Potts, D.E., R.C. Ahlert and S.S. Wang, "A critical review of fouling of reverse
osmosis membranes," Desalination, 36,235 (1981).

Pusch, W. and A. Walch, "Synthetic membranes-preparation, structure, and

applications," Angew Chern. Int. Ed. Engl., 21,660 (1982).

Reid, C.E., "Principles of reverse osmosis," in U. Merten (Ed.), Desalination by

Reverse Osmosis, MIT Press, Cambridge, MA 1966, Chapt 1.

Rogers, C.E., "Permeation of gases and vapours in polymers," in J. Comyn

(Ed.), Polymer Permeability, Elsevier, London, 1985, Chapt. 2.

Sander, U. and P. Soukup, "Design and operation of a pervaporation plant for

ethanol dehydration," J. Memb. Sci., 36,463 (1988).

Schell, W J., "Commercial applications for gas permeation membrane sys-

tems," J. Memb. Sci., 22, 217 (1985).

Solt, G.S., "Electrodialysis," in P. Meares (Ed.), Membrane Separation

Processes, Elsevier, Amsterdam, NY, 1976, Chapt. 6.

Soltanieh, M. and W N. Gill, "Review of reverse osmosis membranes and tran-

sport models," Chern. Eng. Commun., 12, 279 (1981).

Sourirajan, S. (Ed.), Reverse Osmosis and Synthetic Membranes. Theory-

Technology - Engineering, National Research Council of Canada, Ottawa,
Canada, 1977.
726

Spiegler, K.S., "Electrodialysis", in RH. Perry and D. Green (Eds.), Perry's

Chemical Engineer's Handbook, 6th ed., McGraw-Hill, N.Y., 1984, 17-37 to
17-47.

Spillman, RW., "Economics of gas separation membranes", Chern. Eng.

Prog .. 85(1), 41 (Jan. 1989).

Stana, RR, "Westinghouse membrane systems," in S. Sourirajan (Ed.),

Reverse Osmosis and Synthetic Membranes. Ottawa, Canada, National
Research Council of Canada, 1977, Chapt 18.

Stannett, V.T., W J. Koros, D.R. Paul, H.K. Lonsdale, and RW. Baker,
"Recent advances in membrane science and technology," Adv. Polymer Sci ..
32.69 (1979).

Stem, S.A., "New developments in membrane processes for gas separations",

in M.B. Chenoweth (Ed.) , Synthetic Membranes, MMI Press, Midland, MI,
1986, 1-37.

Stem, S.A. and H.L. Frisch, "The selective permeation of gases through poly-
mers," Ann. Rev. Mater. Sci., 11, 523 (1981).

Stem, S.A., "The separation of gases by selective permeation," in P. Meares

(Ed.), Membrane Separation Processes. Elsevier, NY, 1976, Chapt. 8.

Strathmann, H., "Membrane separation processes," J. Memb. Sci .• 9, 121

(1981).

Strathmann, H., "Electrodialysis and its application in the chemical process

industry," Separ. Purific. Methods, 14, 41 (1985).

Warashina, T., Y. Hashino and T. Kobayashi, "Hollow-fiber ultrafiltration,"

Chern. Tech .• 15, 558 (1985).

Ward, R.A., P.W. Feldhoff and E. Klein, "Membrane materials for therapeutic
 727

applications," in D.R. Lloyd (Ed), Materials Science of Synthetic Membranes,

Amer. Chern. Soc., Washington, DC, 1985, Chapt. 5.

Ward, WJ., "Analytical and experimental studies of facilitated transport,"

AIChE Journal, 16,405 (1970).

Way, J.D., RD. Noble and B.R Bateman, "Selection of supports for immobil-
ized liquid membranes," in Lloyd, D.R (Ed.), Materials Science of Synthetic
Membranes, ACS, Washington, DC, 1985, Chapt 6.

Way, J.D., RD. Noble, T.M. Flynn and E.D. Sloan, "Liquid membrane tran-
sport: A survey," J. Memb. Sci., 12, 247 (1982).

Wijmans, J.O., S. Nakao, and c.A. Smolders, "Flux limitation in ultrafiltration:

Osmotic pressure model and gel layer model," J. Memb. Sci., 20, 115 (1984).

HOMEWORK

A. Discussion Problems

AI. How does polymer structure affect selectivity?

A2. What are the advantages and disadvantages of each module shown in
Figure 12-5?

A3. In pervaporation the permeate side is usually under a vacuum. Explain

why. Would pulling a vacuum on the permeate side of a gas permea-
tion system be useful? Explain why.

A4. In gas permeation too high a selectivity can be detrimental and may
actually limit the removal of the permeating gas from the feed. Explain
this phenomenon.

A5. How can RO and UF complement each other?

A6. Contrast concentration polarization without a gel to concentration polar-

ization with a gel.
728

A7. Contrast how concentration polarization develops in RO and in ED.

A8. Figures 12-4 and 12-15 show no concentration polarization on the per-
meate side. Figures 12-18 and 12-21 show mass transfer on both sides
of the membrane. Explain these differences.

A9. Why is ED more economical for separation of relatively dilute solutions

than for concentrated solutions?

AlO. Sketch a pervaporation system to break the ethanol-water azeotype

when column 2 in Figure 12-25 is not required. Assume water is the
preferred permeate. Why might a recycle stream be required for the
permeator?

A 11. Develop your key relations chart for this chapter.

B. Generation of Alternatives
B 1. The RM membrane is a clever and commercially successful solution to
the problem of how to make essentially defect free gas permeation
membranes. Brainstorm at least 5 other solutions to this problem.

C. Derivations
Cl. Show that Eq. (12-32a) is a solution to Eq. (12-31) and the appropriate
boundary conditions.

C2. Show that Eq. (12-39) is a solution to Eq. (12-31) and the appropriate
boundary conditions.

C3. A simple pervaporation unit with a recycle stream is shown below.

Assume that the following variables are known: Fnew , Xnew , T new , R, aT,
TR , Cp,in, cp,new, cp,R, a', A, Tout. The module is perfectly mixed and
operates at thermal equilibrium.

a. Calculate a

b. Assuming a is known, solve for yp in terms of known variables.

 729

C4. Derive Eq. (12-49) from Eq. (12-48).

C5. Derive Eq. (12-46) from Eq. (12-45).

D. Single-Answer Problems

Dl. Estimate the permeabilities for Hz and CO of a RM membrane for gas

permeation if both membranes are perfect The top layer is 1J.U1l sil-
icone rubber. The second layer is polysulfone with a 1.2J.U1l thin skin.
Estimate aH..co. Data are in the text How will surface pores affect a?

D2. Data for a hollow fiber RO module from Dow Chemical is given in
Table 12-2. Details of how the test were done are not complete.
a. If there was no concentration polarization, calculate Ksolv/tM for this
system.
b. IfM = 2 during the experiment, determine K.olv/tM'
You can use the van't Hoff equation.

D3. An ultrafiltration membrane is first tested in a stirred cell where there is

no concentration polarization. The experimental values obtained in the
table are obtained. Then the same membrane is used in a spiral wound
module where there is concentration polarization. Values listed are
obtained. Assume membrane thickness and permeabilities are the same.
Calculate the expected solvent flux and the concentration polarization
modulus in the spiral wound membrane system. No gel layer forms.

Cb(conc. cp(conc. Jso1v•

solute) solute) L/mz day dP, bars
Stirred Cell 10 gIL 0.25 gIL 7000 3.5
Spiral Wound 6g/L 1.0 gIL 4.0
730

04. Ultrafiltration of a dextran solution gives the following data

J, mVcm2 min 0.055 0.045 0.030 0.018

wt. frac dextran 0.01 0.02 0.05 0.10

Estimate the weight fraction at which dextran gels.

D5. A countercurrent dialyzer is using an experimental membrane to

remove urea from an aqueous feed. The entering feed contains lOgIL
urea and outlet should be 19/1iter. Feed rate is 1 L/min. Inlet dialysate
is pure water and flow rate is 3 L/min.

a. Determine the overall transfer rate NA in g/min.

b. Determine required value of kT A.

c. If A = 100 m2 calculate kT.

D6. We wish to desalinate 400 liters/min of a brad"ish water containing 1.5
gIL salt to 0.5 gIL salt in an ED system. The system has 200 cells. The
efficiencies are estimated as: TI. = 0.98, Tlw = 0.99, TIM = 0.99. The
average resistance of each cell is 0.03 ohms. Determine,

a. The current in amps and the voltage in volts.

b. The energy required in cal/liter water.

c. The power in kilowatts.

D7. An experimental RO cellulose-acetate membrane with an active layer 6

microns thick achieved a flux of 0.33 x 10-4 g/cm 2 sec for a 40 atmo-
sphere pressure difference with 0.1 N NaCl in water (n=4.9 atm). What
was the water permeability and list the units if:

b) This membrane had RO = 0.96. What was the salt permeability for
 731

D8. We are ultrafiltering a flexible chain polymer. In a stirred cell with no

concentration polarization RO = 0.996 and K.ruvltm = 2000 L/(m2 day
atm). In a hollow fiber system gelling occurs at cg = 80 gIL. The gel
layer appears to be completely mobile. In an experiment where a gel
layer formed with Cb = 26 gIL and .1p = 6 atm we find Jso1v = 82.83
L/m2 day.

a In the stirred cell find Jso1v and Jsolule if .1p = 5 atm and Cb =20g/L.
b. In the hollow fiber system find J so1v , RaP!' and J so1uIe if .1p = 5 atm
and Cb = 2Og/L. Other conditions are the same as in previous
experiment

D9. In Example 12-1 we obtained Yout = 0.0168. Suppose we want Yout =

0.010. What value of e is required? Other specifications are the same
as in Example 12-1.

D1O. A cellulose acetate membrane is separating a CO2-N2 mixture. The

feed gas is 40 mole % N2 and 60 mole % CO 2, The permeate pressure
is 770 cm Hg while the feed pressure is 10,000 cm Hg. The effective
membrane thickness is 1.5 J..U11. The membrane module can be assum~
to be completely mixed on both sides. If the cut e = FpIFin = 0.55,
determine Yout and yp' Assume permeate is an ideal gas. Permeability
data from Figure 12-9 gives PN, - 0.3 x 10-10 and Peo, - 20 x 10-10

Dl1. We are separating oxygen and nitrogen by gas permeation with a sil-
icone rubber membrane. The membrane module is perfectly mixed.

°
Inlet gas is 21 mole % 02' We desire a permeate product which is 27
mole % 2 , Membrane has a selectivity (lAB =P AIPB = 2.1. Pressure
ratio is PL/PH = 0.35. Treat the gases as ideal gases. What cut e must
be used?

D12. Blood plasma proteins are known to gel at a concentration of cg = 0.2

weight fraction. The proteins can be treated as particles approximately
30 x 10-10 meters diameter. With pure water the membrane has a flux
of 2.5 x 10-5 m 3 /m 2 sec at a transmembrane pressure drop of 68.96 kPa.
At an operating pressure drop of 103.44 kPa with a plasma solution the
732

flux is 0.416 x 10-5 m3 /m 2 sec. Assume the gel layer has a porosity of E
=0.5.

a) What is the gel thickness at steady state?

b) If pressure is doubled, what is the gel thickness at steady state?

Note: Watch your units on viscosity. Assume jJ. =~.ter at 25°C.

D13. We are separating oxygen and nitrogen with a composite membrane,

ao. r.. = Po.lPN• = 5.4. We have set up a membrane cascade of per-
fectly mixed modulus to produce a final permeate product of yp = 0.95
oxygen. In the last module PIiPH = 0.25 and e =0.60. What must be
the inlet mole fraction to this last module (feed contains only O2 and
N 2 )?

F. Problems Requiring Other Resources

FI. Estimate the osmotic pressure of acetic acid, CH3 COOH, in a 25 wt %

aqueous solution at 25°C.

a. Use van't Hoff Eq.

b. Use Eq. (12-17).

c. Use Eq. (12-19).

Data is in Perry's and the Handbook of Chemistry and Physics.

F2. (Derivation). It is unusual to have a ratio of gas constants in an equa-

tion such as in Eq. (12-19). Show in the derivation (see Reid, 1966)
how this occurs. Note that Reid (1966) does not show this since he is
not concerned with units.
 Chapter 13

DETAILED THEORIES FOR MEMBRANE SEPARATIONS

The basic concepts, definitions, and theories for a variety of membrane separa-
tors were presented in Chapter 12. In order to cover the entire area of mem-
brane separations while at the same time keeping that chapter a reasonable
length some of the more involved analyses were not included. These more
involved analyses will be discussed in this chapter. Chapter 12 is a prerequisite
for this material.

The organization of this chapter is to divide the membrane separator into

three parts. The first part (Sections 13.1, 13.2, and 13.5) is the transfer from the
bulk fluid to the membrane wall. This part discusses concentration polariza-
tion, gel formation and fouling. The second part (Sections 13.3 and 13.4) con-
siders the overall flow pattern in the membrane module such as those illustrated
in Figure 12-6. The third part (Section 13.6) treats the transport of solvent and
solute inside the membrane.

Concentration polarization was discussed in Chapter 12 and a simple

theory was presented there. However, the equations were left in terms of the
boundary layer thickness B or the mass transfer coefficient k, and no mention
was made of how to determine B or k. The determination of k using correla-
tions will be done in Section 13.1. The gel model for UP was also discussed in
Chapter 12. A more complete analysis will be done in Section 13.2.

The bulk flow patterns in the membrane modules have a large effect on
separation. In Chapter 12 fluid on both sides of the membrane was assumed to
be perfectly mixed. In Section 13.3 perfectly mixed modules with concentra-
tion polarization are studied. Countercurrent flow is studied in Section 13.4.

The transfer phenomena inside the membrane was essentially ignored in

733
734

Chapter 12. In Section 13.6 an irreversible thennodynamics analysis of this

transfer is developed. The solution-diffusion model for transfer inside of RO
membranes and a frictional model appropriate for transfer inside of porous UF
membranes will be presented.

13.1. APPROXIMATE ANALYSIS OF CONCENTRATION POLARIZA·

TION

In this section a somewhat more detailed model of concentration polarization

than that presented in Eqs. (12-25) to (12-32) will be developed. The analysis
in this section uses mass transfer correlations to find k.

13.1.1. Basic Equations.

The usual picture of concentration polarization was shown in Figure 12-4. The
solute mass balance was given by Eq. (12-31) when the rejection is 1.0 and cp
= O. When the rejection is not 1.0. the mass balance can be written as:

dc (13-1)
]solv cp +]solv c +D -
dy
=a

where the first tenn represents solute which passes through the membrane. The
boundary conditions are

(13-2a)
c=cw at y=O

(13-2b)
c = Cb at y= 0

Since Cp =(l-RO)cw • the solution to this set of equations is

Cw exp (Jsolvo/D) (13-3a)

M=-=---------
Cb RO + (l-RO) exp (Jsolv B/D)
 735

which reduces to Eq. (12-32a) when RO = 1.0. In this result RO = 1 - cp/cw is

the intrinsic rejection in the absence of concentration polarization.

To use this result we must be able to estimate o.

Since 0 controls the
boundary condition. This can be done by assuming that 0 is the same as the 0
for a semi-impermeable wall which has R=1. This is a reasonable estimate for
low flux membranes since the solute flux through the membrane is much less
than the solute flux parallel to the membrane. Then the mass transfer
coefficient to the wall is

(13-4)
k=D/o

This mass transfer coefficient k can be calculated from mass transfer correla-
tions for different flow regimes. This is the subject of Section 13.1.2. Substi-
tuting Eq. (13-4) into (13-3a), we obtain

Cw exp Osolvlk) (13-3b)

M=-=---------
Cb RO + (1 - RO) exp Osolvlk)

JI01v is given by Eq. (12-27). Note that Equations (12-27) and (13-3b) are valid
at each point of the membrane since, in general, Cb, c w, cP ' J solv , and M vary
throughout the membrane module. For a mixed system Cb = COul is constant and
hence Cw, cP ' M, and Jsolv are constant. When RO = 1, simultaneous solution of
Eqs. (12-27) and (13-3b) gives J so1v and M. If RO < 1, we must simultaneously
solve Eqs. (12-22), (12-25), (12-27) and (13-3b). For high flux membranes
Eqs. (13-4) and (13-5) are both suspect.

Equations (13-3), (13-4) and the expressions for k are convenient to use
when Jsolv is constant and specified. If Jso1v is unknown, these equations must
be solved simultaneously with Eq. (12-27). Unfortunately, the polarization
modulus M is required to calculate J so1v in Eq. (12-27) and Jso1v is required to
determine M in Eq. (13-3). Thus an iterative trial-and-error solution is often
required Fortunately, if Jso1vlk «1.0 an approximate solution can be
developed (Rao and Sirkar, 1978). This solution is particularly simple for ideal
semi-impermeable membranes where RO = 1. Assume that the osmotic pres-
sure is a linear function of concentration following Eq. (12-18b) which is rea-
 736

sonable for dilute solutions. When J101v/k « 1.0, the exponential tenn in Eqs.
(12-32a) and (13-3) can be expanded as

cw J 1 JIOIV
, (13-5)
M =-=exp - - - 1 +--
( so v )
Cb k k

where we have inserted RO=1. Setting cp = a (R0 = 1) and 7t(c w ) = ac w in Eq.

(12-27), we obtain

(13-6)

Equations (13-5) and (13-6) can now be solved simultaneously for J so1v '

Ksolv
--(~P-acb)
J lmv = -1m- - - - - (13-7)
1 Ksolvacb
+ 1m k

When using this equation, the assumption that JI01v/k« 1.0 should be checked.
Equation (13-7) is useful when Cb is known. If Cb is not known, we need to
include mass balances (see Section 13.3.2.).

IfRO < 1 and/or (JlOlvlk) is not very small, the result from Eq. (13-7) can
be used as the first guess for J101v ' Then Eq. (13-3b) can be solved for M. Now
the solvent flux equation with RO < I, Eq. (12-27), can be solved simultane-
ously with Eqs. (12-22) and (12-25). This value of J solv can be used in Eq.
(13-3b) and the procedure can be continued until there is convergence.

Once M and JI01v have been detennined the required membrane area and
module length can be found. An external mass balance for the module is

(13-8)
J lmv A Cp + (F - J I01v A)cout =F Cp

In this equation F is the volumetric feed rate to the module (e.g. liter/hr) and
cout is the concentrated product on the feed side of the membrane. Since cout is
 737

usually given, Eq. (13-5) can be solved for A. For tubular or hollow fiber sys-
tems,

(13-9)
A = (xLd)n
where n is the number of tubes of diameter d and length L.

13.1.2. Mass Transfer Correlations

For turbulent flow the Nernst film theory can be used to estimate the value of k.
Figure 12-4 can again be used. The boundary layer is assumed to contain all of
the concentration gradient while the bulk flow is turbulent and well mixed. The
mass transfer coefficient in turbulent flow can be written in terms of the
Chilton-Colburn j factor.

. SC-2/3
k = UbJD (13-lOa)

where Sc = WpD is the Schmidt number. In round tubes

(13-lOb)

where f is the Fanning friction factor which in turbulent flow can be estimated
from
(13-1Oc)
f = 0.0791 Re-1/4

where Re = dUbp/1J. is the Reynolds number. This allow estimation of k and o.

An alternate method is to use the standard correlation for mass transfer in
round pipes in turbulent flow (Blatt et al., 1970; Porter, 1979).

(13-11)
k = 0.023 ~ ReO. 83 Sc 1/3

Fully developed turbulent flow on tubes will certainly occur for Re > 20,000
and in UP devices appears to occur at Re = 2000 (porter, 1979). These results
can be substituted into Eq. (13-3) to estimate the concentration polarization
modulus in turbulent flow. Note that concentration polarization in turbulent
738

flow does not depend upon the distance down the tube. For turbulent flow
between parallel sheets the same correlation can be used except the equivalent
diameter of the channel deq should be used

deq = 4 cross-sectional area = 4 2hw (13-12)

wetted perimeter 2w + 2h

where 2h is the gap between the sheets and w is the width.

For turbulent flow in a stirred vessel (Blatt et al., 1970),

0.33 [ 2] 0.75
k=0.0443 ~;
[ ] 00: (13-13)

00
where d is the vessel diameter in cm, is the stirrer speed in radians/sec, v is
the kinematic viscosity in cm 2 /sec, and D is the diffusivity in cm 2 /sec. This
form is useful for laboratory stirred tanks, but the stirred tank configuration is
unlikely to be used on a large scale. An example calculation is part of Example
13-2.
Turbulent flow can be important in tubular (Figure 12-5b), stirred tanks
and plate-and-frame (Figure 12-5a) modules. Qualitatively, Eq. (l3-3b) shows
that M decreases if the mass transfer coefficient k increases. This can be
achieved with high velocities ~, high diffusivities, and low viscosities.
Operating at high temperature increases D and decreases)J.. Increasing Jsolv
increases concentration polarization. Thus high flux membranes have more
concentration polarization.

For laminar flow conditions the average mass transfer coefficient can be
estimated from (Blatt et al. 1970; Porter, 1979)

(13-14)

where Yw is the fluid shear rate at the membrane surface and L is the length of
the flow channel. The fluid shear rate in laminar flow in a round tube is

4~ (13-15)
Yw=T
 739

and thus the average mass transfer coefficient in round tubes is

k= 1.295 [ U~ 2]1/3 (13-16)

For flow between parallel plates with a spacing of 2h

3 Ub (13-17)
YW=-h-

and the mass transfer coefficient is

(13-18)

These laminar flow correlations are used when the concentration polarization
layer is thin, which holds when the axial distance is much less than the entrance
length (see Example 13-2). These correlations be used when a gel does not
form to estimate Jso1v and Musing Eqs, (13-3b) and (13-7), or when a gel forms
(Section 13-2).

Unfortunately, these correlations do not cover many applications which

are commercially significant. RO and gas permeation in hollow fibers usually
have the feed on the outside. The appropriate mass transfer coefficient then
depends on the flow on the shell side of the module. Spiral wound systems are
not just parallel plates wound in a spiral; instead, the spacers are designed to
promote mass transfer. For both these cases experimental data is required.
Fortunately, the manufacturers have much of the required data.

Example 13-1. Concentration Polarization in Turbulent Flow

A 0.4 gmole/L NaCI solution in water is being purified by RO with a

composite polymer membrane. The membrane has J501v = 750 L/m 2 day
with pure water and M' = 70 atm. With a 0.4 gmole/L solution, R O =
0.995 when M = 1 and M> = 70. An existing tubular module with 1 cm
740

id tubes is being used. The average fluid velocity Ub = 1300 cm/sec and
oM> =50 atm. The cut S =0.10. Operation is at 18°C.
Find the average concentration polarization modulus M, the average
solvent flux Jaolv , Couto and cpo
Data:
at 18°C: DNaC! (0.4 gmoles/L) = 1.17 x 10-5 cm 2 /s
DNaC! (0.8 gmoles/L) = 1.19 x 10-5 cm2 /s
(Sherwood et aI. 1975, p. 37)

From Handbook Chemistry and Physics (p. 0-175 of 48th Ed):

Freezing point depression: T; - Tf = 1. 19°C for 0.346 gmole/L
= 1.49°C for 0.435 gmole/L
MIr,w = 1436 ca1Igmole, T;,w =O°C =273.16 K

For 2 wt% NaCI solutions: p = 1.01442 (l0°C), p = 1.01112 (25°C)

4 wt% NaCI: p = 1.0292 (l0°C), p = 1.0253 (25°C)
(perry and Green, 1984, p. 3-83)

Viscosities at 18°C: J.lw = 1.1 cp = 0.011 poise

Jl (25 wt% NaCI) = 2.41 cp
(perry and Green, 1984, p. 3-252)

Solution

A. Define. In problem statement

B and C. Explore and Plan. In order to determine the mass transfer
coefficient from the correlations we need the physical properties. Obvi-
ously, some interpolation will be needed. The osmotic pressure can be
determined from the freezing point depression data with Eq. (12-19). If
the flow is turbulent we will use Eq. (13-11) to find k. Since R is close
to 1.0, we will first assume R = 1 and cp =O. Then from Eq. (12-30).

Cout =cin/(I-S) =0.4/0.9 =0.444 gmoles/L.

The average Cb = (Cin + crut )/2 = 0.422 gmoles/L. Assuming Jso1vik «
1, we can determine J. olv from Eq. (13-7). Now M can be found from
 741

Eq. (13-3b) and R and hence S, from the solution to Eq. (12-29). We
can then check that Coo.ll is not changed significantly.

D. Do It In a tube with turbulent flow, radial mixing is fairly uniform

(high k), but axial mixing is incomplete. Thus Cb (and 7t) vary from em
»'
(7t (Cin» to Cout (7t (caut Average conditions are at Cb. Thus, we will
do all calculations at Cb = 0.422 gmoles/L.
1. Calculate 7t. Eq. (12-19) is

= (1.45) (1436) (291.16) [82.057] = 18.69 attn

(18.05) (271.71) (273.16) 1.9872

where T; - Tf is estimated at 0.422 gmole/L by linear interpolation

Then Tf = T; - 1.45 =2.73.16 - 1.45 =371.71 K

T = 18°C =291.16 K

cc attn]
The factor [82.057 gmole K / [ 1.9872
gmole Cal]
K .
converts two different

values of the gas constant. The values ~ and Renc::rgy depend on the
units of .1~, v.olvent, and the desired pressure units for 7t.

Since osmotic pressure data for NaCI is readily available (perry and
Green, 1984, p. 17-23), the purpose of this calculation is illustrative.
Interpolating between concentrations and extrapolating to 18°C using
the tabulated data, we find 7t = 18.57 which is a 0.6% difference.
Assuming 7t is linear with concentration, a = 7t/c = 18.69/0.422 = 44.29
atm/gmole/L.
742

2. Calculate k.
A 0.422 gmole/L solution is about 2.4 wt% (same Table as Freez-
ing point depression). Linearly interpolating, p(WOe) - 1.0174 and p
(25°C) - 1.0140. Then doing another linear interpolation p(l8°C) -
1.0156 glee. [This is not the best way to estimate density, but since the
differences are small it will not cause much error].
Viscosity of a mixture can be estimated from (Reid et ai, 1977, p. 462)

where the Xi are mole fractions. Here component 1 is pure water and
component 2 is 25 wt% solution. From the same table which gives
freezing point depression:

Water. 998.2 gIL = 55.4 gmoles/L

0.422 gmoles salt/L: has - 990.83g water/L = 55.0 gmoIe water/L
25 wt% has 5.086 mole NaCl/L and - 891.5 gwater/L = 49.48 gmoles/L
of water.
Assuming volumes add, we need 0.422/5.086 = 0.083 fraction of 25
wt% and 0.917 fraction of pure water.

In /Jmix =(0.917) In 0.011 + (0.083) In 0.024 =- 4.445

/Jmix = 0.0117 poise

Then, Sc = WpD = 0.0117/(1.0156)(1.17 x 10-5 ) = 984.6

Re = D Ub pill = (1)(1300)(1.0156) = 112,844

0.0117
which is clearly tubulent.
From Eq. (13-11)
 743

3. Solvent flux.

From the pure water data:

Ksolv = Jso1v = 750 = 10.71 L/m2 day atm

trn Ap 70

= 1.24 x 10-5 cm/s atm

Assuming J 501v lk« 1.0, we can use Eq. (13-7),

J -
(1.24 x 10-5 ) (50 - (44.29) (0.422» = 0.000386 cm/s
101v - (1.24 x 10-5 ) (44.29) (0.422)
1+ 0.0418

Which is 333.48 2/m2day

Check: Jso1vlk = 0.000386/0.0418 =0.0092« 1.0.

4. Solute.

Now use actual RO = 0.995 in Eq. (13-3b)

M= exp (Jsolvlk) = exp (0.0092) = 1.009

RO + (l-RO) exp (Jsolvlk) 0.995 + 0.005 exp (0.0092)

which is very modest due to the highly turbulent system.

The value of Ksolute /1m can be obtained from the salt rejection data with
M = 1. From Eq. (l2-24C),

Ksolv RO 0.995
0.= = =- - - - - -----
Ksolute (AP- An) (l-RO) (70 - (44.29) (0.4» (1- 0.995)

0.= 3.806

Th Ksolute = Ksolv/1m = 1.24 x 10-5 = 3 25 10~ /

en 1m 0. 3.806 . x cm s.
744

Now the solution to Eq. (12-29) gives~. Constants for the quadratic
formula are

a Ksolv = (44.29) (1.24 x 10-5) = 0.000549

1m

- - up- M 1t (»
Ksolv ( A
Cb +
Ksolute
1m 1m

= (1.24 x 10-5) [50 - (1.009) (18.69)] + 3.26 x 1O~ = 0.000389

Ksolute M Cb =- (3.26 x 1O~) (1.009) (0.422) = -1.387 x 1~

1m

From quadratic formula,

_ - 0.000389 ± ~(0.000389? - 4(0.000549) (- 1.387 x 1O~)

cp = 2 (0.000549)

Use the plus sign to obtain a positive value ofcp •

~ = 0.0039 gmoles/L

Note that, as expected, c p is quite low. This is an average value of cp

since average values for Cb and 1t were used.

E. Check. The solute flux J so1ute = cp Jsolvent = 1.515 x 1O~ which is

small compared to Jsolvent. Thus ignoring this flux in Eq. (13-7) is rea-
sonable. Recalculation of Coot using Eq. (12-30) gives

_ em - e cp = 0.4 - (0.1) (0.0092) = 0443.

Coot - 1- e 0.9 .

where Cp = ~ was used. This is essentially unchanged from our initial

guess based on Cp = O.
 745

F. Generalizations. 1. In this example average values of M. J101v • and

S, were found. Since concentration varies along the tube, an exact cal-
culation requires an integration along the length of the tube. Usually,
calculation of average conditions will be accurate enough for both tur-
bulent and laminar flow.
2. Much of the effort in this example was involved with determining
physical properties. This is often true.
3. Section 13.3.2 develops equations for completely mixed modules
with concentration polarization.

13.2. GEL FORMATION AND FOULING

Gel formation was discussed in Chapter 12 following Eqs. (12-37) and (12-39).
In this section more details of the calculation procedures will be given. The
formation of a gel depends mainly upon the solute type and concentration
although the membrane characteristics and hydrodynamics also affect gel for-
mation (Blatt et ai, 1970; Fane, 1986). Rigid chain, solvated macromolecules
such as polysaccharides can gel at concentrations well below 1 wt %. Flexible
chain, linear macromolecules often gel in the range of 2 to 5 wt%, while highly
structured spheroidal macromolecules such as proteins and nucleic acids gel in
the range from 10 to 30 wt %. Colloids can also form gels. Submicron pig-
ments or minerals gel in the range from 5 to 25 vol % while polymer lattices
require from 50 to 60 vol %.

For membranes with R=I, the solute mass balance was given in Eq. (12-
31) and the solution was given in Eq. (12-39),

cg (12-39)
J I01v = k In-
Cb

The mass transfer coefficient can be estimated from the mass transfer theories
discussed in Section 13.1.2. Experimental UP data can be used to easily esti-
mate k and cg • If J solv is determined at a series of bulk concentrations, Cb. a
plot of J solv versus In Cb will have a slope of -k and an intercept on the concen-
tration axis of In cg (see Problem 12-D4). The resulting k and cg values are
 746

then useful for correlating data. Changes in conditions such as velocity or tem-
perature will change the mass transfer coefficient and can be correlated with
Eqs. (13-11), (13-13), (13-16) or (13-18).

Should one employ a turbulent or a laminar flow system? For a given

volumetric flow rate the mass transfer rate can be maximized by maximizing
the shear rate. For tubular and plate-and-frame systems the flow can always be
made turbulent by recycling liquid. This will give high k values but also results
in high pressure drops along the length of the module. If the value of k/.1p is
found for both turbulent and laminar flow, one finds that this ratio is much
larger for laminar flow. (See problem 13-Cl). Thus, when energy costs are
included in the calculation laminar flow is usually preferable. When UP is
done in hollow fibers the feed flows inside the fibers, and flow almost has to be
laminar. It is fortunate that laminar flow is desirable.

Although very appealing the gel formation theory often does not agree
with experimental data (Fane, 1986; Le and Howell, 1984; Porter, 1979; Wij-
mans et ai, 1984; Zydney and Colton, 1986). For instance, the model does not
include any affect of the membrane, velocity, or feed concentration on cg , but
experimental results show such effects. Since cg is usually much less than the
solubility limit, it is difficult to determine fundamentally exactly what cg is
measuring. Solute-solute and solute-membrane interactions are important
experimentally, but are not included in the model. Experiments with colloids
show fluxes which can be one or two orders of magnitude greater than expected
(porter, 1979). (This is very helpful and helps explain why the earliest com-
mercial applications of UP were for processing colloids, but it is a major prob-
lem for the theory.) Finally, the theory predicts a limiting flux which is
independent of.1p while experiments often show a.1p dependence on the limit-
ing flux.

Many attempts have been made to explain these problems. For example,
in medium molecular weight applications the osmotic pressure can be impor-
tant and should be included in the model (Fane, 1986; Wijmans et ai, 1984).
Unfortunately, models with osmotic pressure or with gels have very similar
predictions and it is difficult to determine which is correcL For macro-
molecules> 100,000 daltons the osmotic pressure cannot have much influence.
 747

Porter (1979) suggested that the colloid data can be explained by the tubular
pinch effect. This is the tendency of small particles flowing in small rubes to
migrate away from the tube walls. This effect will increase the mass transfer of
particles away from the wall and will increase the solvent flux. Although this
effect definitely exists, it does not appear to produce a large enough effect to
completely explain the data (Le and Howell, 1984). Other explanations for
high fluxes with particles are the particle diffusivity is much higher than
Brownian diffusivity because it is caused by particle-particle interaction
(Romero and Davis, 1988; Zydney and Colton, 1986). Other models which
include variable diffusivity or cake aging or adsorption or leaky pores all seem
to be able to explain only part of the phenomena. The mass balance, Eq. (12-
31), does not include axial convection of solute and the boundary condition c =
a
Cb at y = is hypothetical. Shen and Probstein (1977) and Trettin and Doshi
(1980) improved the model by including convection in the x direction (e.g. see
Eq. (13-48» and writing the boundary condition as c = Cb at Y = 00. They also
allowed for a concentration dependent diffusviity. Their results for the limiting
flux agreed better with experimental data for protein UP than Eq. (12-39). At
the current time gelling in UP must be considered only partially explained.
Future research will eventually unravel the complexities involved. The simple
gel formation model still serves as a simple physical picture and as an excellent
method for correlating data.
Fouling can occur in both RO and UP although it is often worse in UP
since dirtier streams are often processed. Fouling is a plugging or coating on
(external) or in (internal) the membrane which is partially irreversible. That is,
in normal operation the fouled membrane will stay fouled until a separate
cleaning step is employed. After clean-up part or all of the original flux will be
recovered. Fouling is complex because it can be caused by many different
phenomena such as precipitation, adsorption, electrostatic attraction, biological
growth, chemical reaction or polymerization.
In RO fouling can be classified as inorganic, particulate, or biological
(potts et aI, 1981). Inorganic fouling occurs when compounds such as CaC03 ,
Caso 4 , MgC03 , and silica or iron precipitate out onto the membrane. Since
the purpose of RO is to remove water and since the concentration of salts is
highest at the membrane wall, it should not be surprising that precipitation can
 748

be a problem. The higher the water recovery the more likely precipitation is to
be a problem. Fortunately, precipitation can often be controlled by adding sul-
furic acid to adjust the pH in the range 4 to 6, and by adding compounds which
bind calcium (e.g. Calgon). Fouling from particulates can be caused by parti-
culates in the feed. This is relatively easy to control by prefiltering the feed. A
more serious problem is the presence of both inorganic and organic colloids.
Methods for solving these problems will be very specific for each case since
the chemistry will differ from case to case. Some membranes such as cellulose
acetate are susceptible to biological growth. This can be controlled by operat-
ing in the pH range from 4 to 6, removing dissolved oxygen, or chlorinating.

All of the above types of fouling can also occur in UF. In addition, since
macromolecules are often ultrafiltered, polymer precipitates or gels may form
which can be attached to the membrane by adsorption or electrostatic forces.
The polymer may also react to form a cross-linked gel layer. Generally speak-
ing, the higher the molecular weight of the solute the worse fouling will be.
Modules should be designed to have no stagnant regions since fouling is invari-
ably worse in these locations. Separate cleaning steps are often employed.
These include pulsing clean water, back-flowing clean water through the mem-
brane, mechanical cleaning with sponge balls or other methods, and chemical
cleaning such as a caustic wash. Fouling may also be reduced by changing the
membrane properties. For instance, electrostatic attraction can usually be
reduced by giving the membrane a negative charge since fouling colloids are
often negatively charged.

Flux decline in both RO and UP often follows an exponential decline and

then levels off to a low asymptotic value.

(13-19)

where Jso1v(t), Jinit , and Jasy are the fluxes at time t, initially and asymptotically.
The time t is often measured in days, and JlO1v(day 1) = Jinito The empirical
constant m is negative. An m = -0.1 corresponds to about a 45% decline in
flux in one year while an m = -0.03 corresponds to about a 16% flux decline in
one year. If the membrane is chemically or mechanically cleaned, much of the
original flux can be restored. The decline will then start again from this
 749

restored flux. Every time the membrane is cleaned some flux is. pennanently
lost. Thus the membranes will have a finite life. This lifetime can vary from a
few months to several years depending on the service conditions. Membrane
replacement should be included as an operating cost. Membrane systems are
usually designed with an average flux. For a parallel cascade membranes can
be replace in a staggered pattern so that some membranes are always fresh and
others are near the end of their useful life.

13.3. COMPLETELY MIXED SYSTEMS WITH CONCENTRATION

POLARIZA nON

In Chapter 12 the mass balances for completely mixed systems were con-
sidered. In this section we will first develop the external mass balances for
reverse osmosis and ultrafiltration with concentration polarization. The possi-
ble occurrence of a gel will be included in Section 13.3.2.

13.3.1. Completely Mixed RO and UF With Concentration Polarization

and No Gel Formation

In reverse osmosis and ultrafiltration concentration polarization must be

included in the analysis. This will be done for a completely mixed flow case
using the approximation developed by Rao and Sirkar (1978) in Eq. (13-5). In
RO and UP rejection is quite high and it makes more sense to write equations
in tenns of rejections instead of selectivity as used in gas penneation. The
development in this section will include osmotic pressure and is valid for RO
and for UP when a gel does not fonn.

The balances will be done for the system shown schematically in Figure
13-1. The overall external mass balance is

(13-20)
Fin = Fout + Fp

while the external solute balance is

(13-21)
Fin em = Fout Coot + Fpcp
750

~N,CIN
Figure 13-1. Schematic of completely mixed RO or UF with concentration
polarization.

The volumetric permeate flowrate can be determined from the flux as Fp = JA.
For a perfectly mixed container Coot = Cb' Equations (13-20) and (13-21) can
be solved to find Couto

(13-22)

which is Eq. (12-30). Note that these equations are essentially the same as Eqs.
(12-5) to (12-8) since the external balances are unaffected by what happens
inside the system. The permeate concentration is related to the wall concentra-
tion
(13-23)

where RO is the intrinsic rejection which will be assumed to be constant Com-

bining Eqs. (13-23) and (13-22), we obtain

(13-24)

The wall concentration can be removed from Eqs. (13-23) and (13-24) by sub-
stituting in Eq. (13-3b). The results are

(13-25)
 751

and, after some algebra,

= ------------
em
Cb (13-26)
8(I-RO) exp (Jsolvlk)
(1-8) + - - - - - - -
RO + (l-RO) exp (JlOlvlk)

Note that if RO =1.0, cp=O and Cotlt=cuJ(I-8). When RO =1, the flow patterns
and solvent flux are unimportant for determining concentrations.

For R O <1 Eqs. (13-25) and (13-26) are very useful if Jso1v is known.
Unfortunately, J so1v is usually not known. The solvent flux must now be deter-
mined from the flux Eq. (12-27). We will assume that the osmotic pressure
depends linearly on concentration so that Eq. (12-18b) is valid. Combining
Eqs. (12-27), (12-18b) and (13-23), we have

(13-27)

The wall concentration can be removed using Eq. (13-3b), and this result can
be solved simultaneously with Eq. (13-26). The equations are nonlinear and a
closed form solution cannot be obtained. If the exponential term is approxi-
mated as in Eq. (13-5), an algebraic solution can be obtained. The wall con-
centration becomes

(13-28)

which gives

(13-29)

After some algebra the bulk concentration is

(13-30)
752

These two equations can now be solved simultaneously. After some manipula-
tions the following quadratic equation results.

(13-31)
a' Jtolv + b' Jso1v + C' = 0

where

(13-32a)

(13-32b)

, _ Kaolv aRc (1 ORO) K.olv (13-32c)

--.:- up
A
c - --.:- cin - -

After determining k from the appropriate mass transfer correlation, J solv can be
found from the quadratic formula.

- b' ± ""/b'2 - 4a'c' (13-33)

J so1v = 2a'

Once Jso1v is known, Cp and Cb can be calculated from Eqs. (13-25) and (13-26).
If gel formation is possible, calculate Cw from Eq. (13-28). If Cw < c g a gel
won't form and this solution is correct. If Cw > cg , a gel forms and this solution
is not correct (see Section 13.3.2). Also, this solution is based on Jso1vlk < < 1
and this restriction should be checked. If the assumption is invalid, we need to
solve Eqs. (13-3b), (13-26) and (13-27) simultaneously. Equation (13-33) can
be used for a first guess.

IfRo=1 the solvent flux simplifies to

(13-34)
for R=1
 753

It is interesting to compare this result to the solvent flux without concentration

polarization and with RO=1. With no concentration polarization Cw = Cb while
cp = O. Solving Eqs. (12-18b), (13-6) and (13-24) simultaneously for these
conditions we obtain,

(13-35)

Then the ratio of solvent fluxes with RO=1 is

J solv 1
-----=--=--------
w. COllC pol.

J101v w/o conc. 1 + em a Kaolv

pol.
forRo = 1 (13-36)

k(1-9)tro

The flux without concentration polarization is always higher, and thus is an

upper limit to the actual flux. As the mass transfer coefficient becomes very
large the two solvent fluxes approach each other.

The flow pattern used in RO and UP is usually not completely mixed.

However, stirred tanks and systems with a very large recycle (see Problem 13-
06) are often close to perfectly mixed. At higher concentrations 7t is usually
not linearly dependent upon concentration. In addition, the pressure drop on
the high pressure side decreases PH and thus decreases ~p (Hwang and Kam-
menneyer, 1975). Complete design calculations are qUite complex and require
computer solutions. These design calculations are discussed by Harris et al
(1976), and Sourirajan (1970,1977).

13.3.2. Completely Mixed UF with Gel Formation

If a gel forms, the analysis in Section 13.3.1 needs to be modified. Equations
(13-23) and (13-24) become

(13-37)

(13-38)
754

where we assume that the solute is mobile in the gel and does not change RO.
Since cg is known, Cp and Cb can be calculated immediately.

When a gel is formed Eq. (13-3b) becomes

(13-39)

which reduces to Eq. (12-39) when RO = 1.0. We can determine k from the
appropriate correlation and then calculate l solv • For laminar flow Eqs. (13-16)
and (13-18) are convenient for first estimates of the average value of k.

If c g is high, cin is low, or 9 is low, we may not know in advance if a gel

will form. In this case, we can calculate Cw using the method in Section 13.3.1.
A gel will form if Cw > c g , and Eqs. (13-37) to (13-39) should be used. This
procedure will be illustrated in Example 13-2.

Example 13-2. Concentration Polarization in Completely Mixed UP

A 10 cm diameter stirred tank with a mixer operating at 4000 rpm is

being used to ultrafilter albumin. With pure water and ~p = 5.0 atm, the
flux is 2500 L/m2 day. RO = 0.992. Operation of the stirred cell will be
at 18°C with ~p = 4.2 atm.
Data: Albumin (porter, 1979), D =6)( 10-7 cm2/s, cg =0.45 (wt. frac),
MWalbumin = 70,000.
Water (perry and Green, 1984): J..lw = 0.011 poise, Pw = 0.9986 g/cm 3 •
Assume v = wp is approximately constant.

Find Cp, Cb and J I01v for feed concentrations of 0.04 and 0.20 wt frac.
albumin ifa = 0.5.

Solution
We will first assume a gel does not form and use Eqs. (13-32) and
(13-33) to find J 101v • Then we will find Ct, from Eq. (13-26) and Cw from
 755

Eq. (13-28). If C w < cg , this solution is correct If Cw > cg we need to

use Eqs. (13-37) to (13-39). The mass transfer coefficient k for a stirred
tank can be found from Eq. (13-13). To use this equation,

v = IJ/pw = 0.011/0.9986 g/cm 3 = 0.011 cm2/s

which we will (for now) assume to be constant.

OJ = 4000 rpm I 21t radians I 1 min I = 418.8 radians/s

revolution 60s

Then from Eq. (13-13),

k = (0.0443) [6 x 1~0-7] [ 0.011 ] 0.33 [(418.8) (10)2]0.75

6 x 10-7 0.011

= 0.00594 cm/s

From the pure water flux we can find k,.olv/tm

k,.olv J solv 2500 2

-- = -- = -5 0 = 500 L/m day atm = 0'(lOO578 cm/s atm
1m ~p .

To estimate a in Eq. (12-18b) we will use Eq. (12-18a). Since the solu-
tion will be approximately doubled in concentration, use c = 8 wt%
albumin. Then for 1 liter have approximately: 960 g water = 53.3
gmoles water and 83.5 g albumin = 0.00119 gmoles albumin.
Thus

1t= [0.001~9gmOle] [11itef] [82.057ccatm] (291.16K)

liter 1000 cc gmole k

= 0.0285 atm

from Eq. (12-18a). We can use Cin as 0.08 wt. frac in Eq. (12-18b).
756

Then
1t 0.0285 atrn
a =- = 0 .08 wt fraC =0.356 atrn/wt frac.
c

Since 1t/Ap = (0.0285/4.2) x 100 = 0.68%, we can ignore osmotic pres-

sure for the 4 wt% feed and set a = O. Remember that the van't Hoff
Eq. (12-18a) is highly unreliable; however, the basic conclusion that
osmotic pressure can be ignored is probably valid.

Now, we can detennine J so1v from Eq. (13-33), assuming Jso1v/k « 1.

From Eq. (13-32),
a' = (I-RO)/k = 0.008/0.00594 = 1.347

b' = 1 - (0.5) (0.992) - (0.0005789) [ 0~!4 right)(4.2) + 0 =0.5007

c' = 0 - [1 - (0.5) (0.992)] (0.000578) (4.2) = - 0.001224

From Eq. (13-33)

J _ - 0.5007 ± -/(0.5007)2 - (4) (1.347) (-0.001224)

solv - (2)(1.347)

= 0.00243 cm/s
where the positive sign is used to make J ao1v positive. To check the
assumption calculate

Jsolv/k = 0.00243/0.00594 = 0.409

There will be some error in the calculation (see Notes) and a second
tria' would be required for accurate results.

From Eq. (13-16) we find Cb

Cb =
em = 1.9763 Cin
0.5 + 0.5 (0.008) exp (0.409)
0.992 + (0.008) exp (0.409)
 757

while from Eq. (13-38)

(1.9763) em (1.409)
Cw :::: 0.992 + (0.008) (1.409) :::: 2.775 Cin

and from Eq. (13-23)

c p :::: Cw (1 - RO) :::: (2.775 Cin) (0.008) :::: 0.0222 cin

Note that since osmotic pressure is negligible, em does not affect the
calculation of J so1v and the other concentrations can be detennined as
functions of inlet concentration.

4 wt% feed: Cb:::: 0.0791, Cw :::: 0.111 < c g , c p =0.00089

20 wt% feed: Cw :::: 0.555 > cg

Thus, for the 20 wt% solution we need to use the solution when a gel
forms. With a gel formed we use Eqs. (13-37) to (13-39).

Cp :::: (0.45) (0.008) :::: 0.0036

:::: 0.2 - (0.5) (0.45) (0.008) :::: 0 3964

Cb (0.5) .

J so1v ::::
(0.992) (0.45) / (0.3964)
(0.00594) In [ 1 _ (0.008) (0.45) / (0.3964)
1:::: 0.00076 cm/s

Notes:
A. There are several approximations involved in this solution.

1. The worst approximation is v:::: v water in Eq. (13-13). The

albumin solutions will be significantly more viscuous, v will
758

be higher, and k will be lower than predicted. We need

viscosity data.
2. Jso1v/k is really too large to use the approximation that
exp (Jsolv/k) = 1 + Jso1v/k, and R :#:- 1.0. Using calculated
J.olv/k = 0.409, exp (Jsolv/k) = 1.505 which is 6.8% greater
than 1 + Jao1v/k = 1.409. Another trial should be done. For
UP this approximation will often be suspect.

3. Ignoring osmotic pressure in reasonable for the 4 wt% feed.

For the 20 wt% feed osmotic pressure is unimportant
because a gel fonns.
4. When the gel fanned, we assumed R was unchanged. We
need data to check this assumption. If the gel is rigid, R may
become 1.0.

B. The fonnation of the gel reduces Jsolv by a factor of 3.

C. Even with vigorous stirring M = Cw/Cb =2.775/1.9763 = 1.4 which

is significant.

13.4. COUNTERCURRENT GAS PERMEATION WITH NO CONCEN-

TRATION POLARIZATION

The effect of flow patterns in gas penneators is discussed in great detail by

Hwang and Kammenneyer (1975). We will follow their analysis for the coun-
tercurrent gas penneator shown schematically in Figure 13-2. Consider the

1-----1

~ L _____
:::::f1..J

~
Figure 13-2. Schematic of countercurrent gas penneator.
 759

separation of a binary gas stream. The differential mass balances for

components A and B are

(13-40a)

"
-d [ql (l-Yl) ] = -d [<U(l-Y2)] = P~B [(l-Yl)PH - PL (l-Y2)] (13-40b)

In these equations ql and qz are the flow rates in moles/hour on the feed and
permeate sides of the membrane, and Yl and Y2 are the mole fractions on the
feed and permeate sides of the membrane, respectively. A is the membrane
area PA and PB are the permeabilities of the membrane. The second equation
p,
in each set comes from the flux Eq. (12-3). The molar density, is required to
convert the volumetric flux JAto a molar flux. These two mass balances can be
rearranged. Adding Eqs. (13-40a) and (13-40b) we obtain

(13-41)

Rearranging Eq. (13-40b) we have a second equation.

(13-42)

A third equation is also required. This can be obtained from the overall
and component mass balances for the mass balance envelope shown in Figure
13-2.

(13-43a)

(13-43b)
760

Removing the unknown flowrate 'l2 and solving for Y2, we obtain

(13-43c)

Equations (13-41), (13-42), and (13-43c) are solved simultaneously. The boun-
dary conditions are:

(13-44a)

(13-44b)

In Eq. (13-44a) Y2.0 is the penneate mole fraction corresponding to Yout. This
penneate mole fraction can be obtained by taking the ratio of flux Eq. (12-3)
written for components A and B at the outlet. This result is

PL
Yout - PH Y2.0
Y2.0 PA (13-44c)
- - = aAB - --------
l-Y2.0 PB PL
l-Yout - - (I-Y2.o)
PH

Note that Eq. (13-44c) is essentially the same as Eq. (12-10) for a perfectly
mixed system.

Obviously, a numerical solution of Eqs. (13-41), (13-42), and (13-43c) is

required. This was done by Blaisdell and Kammenneyer (1973) for oxygen
and nitrogen penneation through a silicone rubber membrane. Results are
shown in Figure 13-3. The results in Figure 13-3 are the same for all flow
configurations at 9 = O. If we desire a concentrated oxygen product, operation
should be at low cut (very little penneate product). If a concentrated nitrogen
product is desired, a high 9 with countercurrent flow should be used. The
=
selectivity, a 2.05, used in these calculations is low. The best currently avail-
able oxygen membrane has a selectivity of about 7.5. This membrane is used
for N2 production. The countercurrent flow arrangement is best since the aver-
age driving force is highest while completely mixed is worst since the average
driving force is lowest (see Problem 13-C4). Balances for flow arrangements
 II
0.29

"'~'
0.28 \':0:,
, '.,
.... ',
,
'\ ...... ',
0.27 \. .....
', '"
0.14
'\,".. ,, ,
~ '\, U
<l>
ClJ '~~
'"
"..... ,,
0.26 "\, ". ~
E
ClJ ,,', ".
a.
,"
\', .••.. c
<l>
" " 012
OJ)
c \ '..... ,
ClJ
" 0.25
OJ) ~ 0.10
>-
x ". c
0
'. \.'<-'." \ , 0
c \ ':,.... \ .;::;
0 u
.;::; 0.24 Countercurrent - - - Countercurrent
u
.' \ ".'\;" ,\ .::'" 008[
.::'" ---- Cross-flow " '';;.,'. \ - - - - Cross-flow
\
........ Cocurrent " 'i:. \ 0.06 ....•..... Cocurrent
0.23
,,
___ ._ Constant yp '. \ •...'\;." \ \ \ ----- Constantyp
_ •. _ Well mixed
",:\.\ ---- .. Well mixed
0.22
'\.\
, \\' '" 0.02
~-'\,~-~
0.21
I I I I
0
°T0 0.2 04 06 08 10
,~
Cut, !:I Cut
(a) (b)
Figure 13-3. Product concentration for air separation (ex =2.05, prJPH = 0.359, YF = 0.209). a. Permeate con-
centrations. b. Low pressure product concentrations. (Blaisdell and Kammermeyer, 1973).
Reprinted with permission from Chern. Eng. Sci., 28, 1249 (1973). Copyright 1973, Pergamon
--.J
Press. 0\
 762

other than completely mixed and countercurrent are developed by Hwang and
Kammermeyer (1975).

13.5. EXACT SOLUTION FOR CONCENTRATION POLARIZATION

IN LAMINAR FLOW

In laminar flow both concentration and velocity boundary layers will form.
Since diffifusivities are low, the concentration boundary layer will be thinner at
the beginning of the tube. Both boundary layers become thicker as one
progresses down the tube. If the tube is long enough (or the channel is thin
enough) both boundary layers will eventualy fill the tube. Thus, one should
expect two solutions for laminar flow: an entrance region solution and a fully
developed far-downstream solution. The solutions have been generated for the
case where the velocity profile is fully developed immediately, but the concen-
tration profile is not. The solute was assumed to be completely rejected. Solu-
tions for both parallel plates and round tubes are available (Sherwood et ai,
1965.
The geometry for RO between two parallel sheets is shown in Figure
13-4a For laminar flow the velocity must first be calculated from the following
Navier-Stokes equations,

(13-45a)

(13-45b)

and the continuity equation

(13-45c)

The boundary conditions are: a) no slip at the wall,

(13-46a)
u=O at y=±h
 763

a
t t t t t
yLx tv .. u th
th -
• tv w
• • •
b

2R
\

Figure 13-4. Geometries for concentration polarization analysis. a. Parallel

plates, b. Round tubes.

b) v equals solvent withdrawal rate at wall,

(13-46b)
v= Vw at y=±h

and c) symmetry at the centerline.

au (13-46c)
v=-=O
ay at y=O

The flow solution obtained by Berman (1953) is

-u= -
Uin
3 [ 1 -Vw
2
--x (1-y')
Urn h
1 - h- [ 2-7y'2_7y'4 J
2 [ 1 -Vw
420 v
1(13-47a)
and
v y' ,z vwy ,2 I(j (13-47b)
- = - (3 - y ) - - - (2 - 3y - y )
Vw 2 280 v
764

where Ujn is the average fluid velocity at the channel inlet, and y' = y{h. This
solution reduces to the usual solution for laminar flow between flat plates when
vw =0.
These solutions for velocity can be inserted in the steady state mass bal-
ance for solute. When rejection is perfect, R = I, this balance is,

a [uc - D ax
ax ac] + ay
a [vc - D ac] = 0
ay
(13-48)

The solute boundary conditions are: a) concentration equals Cin at inlet to RO

system,
(13-49a)
c=cin at x=O

b) at the wall the flux of solute to the wall equals the backward difussion of salt

c Vw
ac
= D ay at y= ± h
(13-49b)

and c) symmetry at the centerline.

(13-49c)
~=O at y=O
oy

The solution to Eqs. (13-47) to (13-49) is difficult; however, the equa-

tions can be simplified (Sherwood et al., 1965). For typical RO solvent fluxes,
the terms divided by 420 v and 280 v in Eqs. (13-47) can be ignored. Also,
since the axial concentration gradient will be small, the axial dispersion term,
- Doc/ox, can be neglected in Eq. (13-48). The solution for the entire range of
variables was obtained numerically by Sherwood et al. (1965), and is shown in
Figure 13-5.

Sherwood and his coworkers also obtained asymptotic solutions for cer-
tain regions. The solutions are written in terms of the dimensionless variable

(13-50)
 765

l:::,. a =0.0677
o a =0.27
o a =0.50
10.0

M-I

o.I 1..-----'L-...J'---'-J'-'---'--...L-L-'-'---'----L-.<-L-.l----1.----1.--'--I...J..-"----"-,--'-'

0.001 0.01 0.1 1.0 10.0 100.0

~ =8L/3a 2

Figure 13-5. Solutions for laminar Bow between parallel plates (Sherwood
et ai, 1965). Reprinted with permission from Ind. Eng. Chem.
Fundam .. 4, 113 (1965). Copyright 1965, Amer. Chern. Soc.

Very close to the inlet (~ < < I),

(13-51)
M = 1 + 1.536(~)1!3

For larger values of ~ the approximate solution for the entrance region is

(13-52)
M = 6 + ~ - 5 exp(--.J'fj3)

Very far downstream ( large ~ ) the boundary layer will completely fill the
tube. Once this occurs M will be constant.

(13-53)

The locations of validity of these equations are shown in Figure 13-5.

To determine if the far downstream solution Eq. (13-53) or the entrance
region solution Eq. (13-52) should be used, calculate M with both equations.
The smaller M should be used. Very close to the channel entrance, ~ < < I, use
the larger value of M calculated from Eqs. (13-51) and (13-52).
 766

A similar but more complicated analysis can be done for laminar flow in
round tubes. The geometry is shown in Figure 13-4b. The Navier-Stokes
equations were solved by Yuan and Finkelstein (1956). Sherwood et ai. (1965)
used this velocity solution in the solute mass balance. The approximate solu-
tion for the entrance region for v! x R/(4llm D2) s 0.02 is

(13-54)

and for v! x R/(4 Uin D2) > 0.02

M= v3XR2
w + 6 - 5 exp [v3
- ( w XR 2 )112 1 (13-55)
4 Uin D 12 llm D

Note that the concentration polarization depends upon the axial distance x.

Far downstream the boundary layer will completely fill the tube. Once
this occurs M will be constant. This far-downstream solution was obtained
numerically and was presented in graphical form. This result is shown in Fig-
ure 13-6 (Sherwood. et ai, 1965). To determine if the entrance region solu-
tions, Eqs. (13-54) or (13-55), or the far-downstream solution (Figure 13-6)
should be used. calculate M by both methods. The smaller M should be used.

If the average polarization modulus M is desired, this can be estimated

by doing the calculation at x = L/2. If either the solvent velocity through the
tube wall Vw or the solvent flux J so1v are given, the calculation of M(x) or M is
straightforward for both geometries. However, if M, J so1v and Vw are unknown
the problem is trial-and-error. This involves guessing M and then calculating
Jso1v from Eq. (12-27) or Eq. (13-6) if R=1. Then calculate M from the
appropriate solution. If this M does not agree with the M used to calculate
Jsolv' the procedure is repeated.

Although usually quite accurate, these results for both turbulent and lam-
inar flow are approximate for a variety of reasons. The no-slip boundary condi-
tion at the wall is not strictly true for a porous solid since there can be lateral
fluid movement in the pores (Fane, 1986). Fluid properties P,1l and v and
 767
100
80
60 -

40
0

20

M-I 10
8
6

0.02 0.04 0.06 0.10 0.2 0.4 0.6 0.8 1.0

..b
vwR
Figure 13-6. "Far-downstream" solution for round tubes. (Sherwood et ai,
1965). Reprinted with permission from Ind. Eng. Chern. Fun-
dam .. 4, 113 (1965). Copyright 1965, Arner. Chern. Soc.

solute diffusivity vary if the polarization modulus M is large. Mass transfer

coefficients are correlated for systems with impermeable walls, but will be
increased by the flux through the walls. Since concentration polarization
increases as the concentration boundary layer develops, the solvent fiux and
hence Vw usually decreases with increasing x. Including this variation is
significantly more complex than the constant Vw case presented here (see
Matthiasson and Sivik, 1980, for references).

Example 13-3. Concentration Polarization in Laminar Flow

Estimate the average concentration polarization modulus M for a dilute

solution which is being concentrated in an RO system. The properties
768

are D = 5.2 x 10-7 cm 2 /sec, P = 1.01 glee, J.L = 0.0095 cm 2/sec, CL = 0.99
gH 2 0/ml (solvent concentration). The operation is inside round hollow
fibers which are 0.01 cm in radius. Bulk v~locity at the inletis 3.2
cm/sec. Flux rate is 1.8 x 10-4 glsec cm 2 • The totaIlength of each hol-
low fiber is 50 cm. Assume RO = 1.0 and Ppc:rmeate = 1.0.
If the hollow fibers are increased to 500 cm long, estimate the average
concentration polarization.

Solution

First, we need to check the Reynold's number.

R = DUb = 2R Ub = 2(0.1)(3.2) = 6.74

e v v (0.0095)

Flow is laminar. For average value of M use x = L/2 = 25 cm.

Wall velocity,

vw _- - - -_ 1.8 x 10-4 -
flux _ 1. 80 x 10-4 Cm/S
Pw 1.00

and parameter for abscissa of Figure 13-3 is

_D_ =
Vw R
1
5. 2x 0- 7
(1.8 x 10 ) (0.01)
=0.2888

Far downstream result from Figure 13-6, M-l - 3.0 or M = 4. Also,

need entrance region solution. Parameter value

v! x R ] = (1.8 x 1Q-4)3 (25) (0.01) = 0.421

[
4 llm D2 (4) (3.2) (5.2 x 10-7 )2

From Eq. (13-55),

0.421
M=0.421+6-5exp [- [- 3 - ]'h] =2.98
Since entrance region result is smaller, use that value of M.
 769

For L = 500, x = L/2 = 250 cm. Since abscissa in Figure 13-6 is

unchanged, Mfar downstream =4.0.

For entrance region

v3
w
R
X
= (0.421) -250 = 4.21
4lljn D2 25

M = 4.21 + 6 ~ 5 exp [ ~ [ 4.~1 r] = 8.68

Since the far downstream solution is smaller, M =4 when L =500 cm.

Note that this method of calculating an average using x = L(2 is reason-

able if all of the system is in the entrance region or most of the system is
far downstream. The value of M calculated is always somewhat too
large which makes this design approach conservative. A more accurate
prediction can be made by calculating M at several points and then
integrating numerically to find an average.

As a check on the solution for L = 50 we can use Eqs. (13-27) and (13-
3b).

k = 1.295 [ u~~2]1/3 = 1.295 [ (3.2lo~~·~ ~;~-7)2]1/3 = 0.0001555 cm/s

For R O = 1, M = exp (Jsolvlk)

J 501v -- 1.8 x 10-4 g/s 3cm
2 _ 18 10-4 I
-. x cm s
(1.00 g/cm )

M=ex [ 1.8 X lO-4]=3.18

P 1.555 x 10-4

% Dif = 3.1~~~.98 x 100 = 6.7%

 770

Note that 0 = D/k = 5.2 x 10-7 /1.555 x 10-4 = 3.34 x 10-3 cm < R.
If we apply this correlation when L = 500,

k =0.00007218, M =12.108

and 0 = 0.0072 which is close R = 0.01. This closeness of Rand 0 is a

signal to not use this mass ttansfer correlation.

13.6. TRANSPORT INSIDE THE MEMBRANE

In this section two models for ttansport inside the membrane will be developed.
The structure of irreversible thermodynamics will be used for these models and
will be briefly presented first Then the solution-diffusion model appropriate
for RO membranes will be developed in a simplified form. Finally, a frictional
model appropriate for UP membranes which have small pores will be dis-
cussed.

13.6.1. Basic Irreversible Thermodynamics

Irreversible thermodynamics has been extensively applied to study the ttansport

inside the membrane (Hwang and Kammermeyer, 1975; Johnson et ai, 1966,
Lakshminarayanaiah, 1969; Merten, 1966; Soltanieh and Gill, 1981). The flux
of species i through the membrane, Ji is

(13-56)
Ji = Flux i = Ln (Force on i) + L ~j (Force on j)
Thus the flux of species i is proportional both to the force on that species and to
the force on other species. This is usually written as

J.=T··X+
1 ~ 1
~ T··X
~ '-'iJ J
(13-57)
jU,.l)

where Xi is the force on species i, Ln is the phenomenological coefficient for

the movement of species i caused by the force on that species, and Lij is the
 771

phenomenological coefficient for the movement of species i caused by the

force on another species j.

A thennodynamic analysis based on the continuous increase in entropy

can be used to show that
(13-58)

and
(13-59)

Equation (13-59) states that a species will move in the direction of the force
applied on it Equation (13-58) requires that ~j =0 if Lii = O. Thus, if a com-
ponent will not move through a membrane due to forces acting directly on the
component then it will not pass through the membrane at all. Thennodynami-
cally, ~j can be either positive or negative since Eq. (13-58) only limits the
absolute magnitude of the coupling coefficients L ij . In practice all known Lij
are ~ O.
One other condition on the phenomenological" coefficients is Onsager's
reciprocal relationship.
(13-60)
~j = Lji
This states that coupling of two species is the same regardless of which is
experiencing the force and which flux we are considering. This relationship is
based on microscopic reversibility.

The validity of these general statements concerning the phenomenologi-

cal coefficients requires the appropriate choices of the forces Xi' Under isoth-
ennal conditions the forces are

(13-61)

where Ili is the chemical potential of species i and Yi is the external force on
component i. The external force in ion exchange membranes is an electrical
force. In porous membranes where separation is based on sieving the Yi are
frictional forces between the component and the membrane.
772

The chemical potential term can be expanded in terms of concentration

gradients and pressure gradients.

grad Ili = [ dlli

dC' 1 grad Ci + Vi grad p
(13-62)
1 p,T

where Ci is the concentration of species i in weight per volume and Vi is the

partial molar volume of species i,

(13-63)

where N is the number of molecules or particles. Equation (13-62) can be

integrated for the solvent (which is usually called species 1).

Remembering that the osmotic pressure difference across a membrane is the

pressure difference that exists when.1J.!l = 0, then Eq. (13-64) becomes

(13-65)

Substituting this result into Eq. (13-64), we obtain

(13-66)

where we have assumed that Vl is constant so that we could do the last integra-
tion in Eq. (13-64). Note that &t is from one membrane interface to the other.
That is, from c'm to C"m in Figure 13-7.

It will be our purpose to first determine the appropriate terms for the
external forces, to then arrange the equations in a form which allows experi-
mental determination of the coefficients Ln and ~j' and finally to predict the
 773

phenomenological coefficients from other properties such as diffusivities and

solubilities.

13.6.2. Solution-Diffusion Membranes (RO and Gas Permeation)

=
Assume that there is no coupling of fluxes (Lij 0), and that external forces are
unimportant (Yi = 0). This assumes that convective flow of water or carrier gas
through pores or pinholes is negligible. This is appropriate for solution-
diffusion membranes used in reverse osmosis and gas permeation. Combining
Eqs. (13-57), (13-61), and (13-66), we obtain

(13-67)

where we have approximated grad J..li as ~Itm. J' is the mass flux in units such
as g/cm 2 s, and J'solv = J solv Psolv. This result agrees with Eq. (12-20a) if we
relate

_ Kao1v Psolv
L 11- (13-68)
VI

The minus sign in Eq. (13-67) comes in because the +x direction shown in Fig-
ure 13-7 is in the direction of osmotic flow for the irreversible thermodynamic
argument. The solvent density appears because mass and volumetric fluxes are

MEMBRANE
SUPPORT

e"2m

Figure 13-7. Schematic of concentration profiles for RO membrane.

 774

being compared. Since.rut refers to the difference from concentration c'm to

c"m, concentration polarization and solubilities are included. If 1t is propor-
tional to concentration, Eq. (13-67) will have the same form as (12-27). Thus,
after a lot of work we find that the solvent flux is given by the same equation
we have been using all along. The advantage of the irreversible thermodynam-
ics development is we can now relate the phenomenological coefficients and
the fluxes to more fundamental quantities.

Assume that sorption and desorption at the membrane surface are rapid
so that diffusion controls. Then Lu can be related to the concentration within
the membrane Cim and the mobility IDjm .

(13-69)

The mobility is the velocity per unit force on one mole of particles. In dilute
solutions the mobility can be related to the diffusivity.

(13-70)
mim =D 1m /RT

where R is the gas constant and D 1m is the solvent diffusivity within the mem-
brane. Since the diffusivity and the solvent concentration are often approxi-
mately constant, Lll is often approximately constant. Now the solvent flux is

(13-71)
]'solv =

Remember that .rut refers to the osmotic pressure difference across the mem-
brane. The solvent permeability can be estimated by comparing Eqs. (13-71) to
(13-67) and (13-68)

(13-72)

This allows estimation of permeabilities, and if permeability is determined

experimentally it allows extrapolation to other conditions. This result also
shows that Kso1v will be approximately constant in dilute solutions where the
 775

solvent concentration and diffusivity are approximately constant. In concen-

trated solutions Ksolv is not constant (Soltanieh and Gill, 1981).

The flux of solute (species 2) can also be estimated. In RO systems

external forces and the coupling of fluxes can usually be ignored. Then com-
bining Eqs. (13-57), (13-61) and (13-62) gives

(13-73)

where J 2 is a mass flux which is the same as in Chapter 12. Since the last term
is usually negligible, we obtain

(13-74)

The phenomenological coefficient Lzl can again be related to the mobility and
for dilute solutions to the diffusivity.

D'}m (13-75)
~2 = m:an C']m = - - '2m
RT
Since c'}m can change significantly in the membrane, L22 may not be constant.
For dilute solutions

[ ~l
dC:an T,p
RT (13-76)

Combining Eqs. (13-74) to (13-76), we have

(13-77)

where we have approximated dc'}m/dx as AC'}m/tm. In this equation AC'}m is the

concentration difference within the membrane. This is illustrated in Figure
13-7.
 776

The relationship between the solute concentration in the membrane and

in solution depends upon the solubility. We will assume that the membrane
surfaces are in equilibrium with the solution and that the distribution coefficient
is constant.
c~ C"2m (13-78)
Kz = - - = - -
CZ w cp

Substituting this into Eq. (13-77), we have

D 2m K z D 2m Kz (13-79)
Jsolute = ---(czw-cZp) = - (M CZb - cp )
t.n t.n
This result agrees with Eq. (12-25) and shows that
(13-80)

Note that this result also agrees with Eq. (12-15) used for gas permeation which
is also a solution-diffusion process.

It is interesting to use these results to study the change in salt concentra-

tion across a RO membrane. The salt concentration on the permeate side, cZ,p
can be calculated from

CZp
[
g salt
cm 3
1 = J'
J, [~l
[g water 1
c
Ip
[g water
cm 3
1 (13-81)

101v cmz s

Then from Eqs. (13-71), (13-79) and (13-81), concentration ratio is

(13-82)

Assuming that cZw » CZp and defining Kl = Clm I Cl p as the water distribution
coefficient in the membrane, we obtain

(13-83)
 777

From Eq. (13-83) we can estimate the properties required to obtain a given con-
centration ratio. Note that the concentration ratio in RO membranes is propor-
tional to (Ap - Ax). More separation is achieved by increasing Ap.

Example 13-4. Water Desalination.

What membrane properties are required to desalinate sea water to

potable water using RO?

Solution

A. Define. This problem is not well defined Typical operating condi-

tions are Ap = 100 attn with an average brine concentration of 5 wt %.
Potable water is 500 ppm or 0.05 wt %. We wish to find the required
value of Dim Kl / D2m K 2 •
B and C. Explore and Plan. We can use Eq. (13-83) to estimate the
required properties. We will assume concentration polarization is negli-
gible. The osmotic pressure of 5 wt % brine is about 37 attn while the
osmotic pressure of 0.05 wt% is about 0.4 atm. The value of vdRT is

l
~= 1(8.095 cc/gmole ::.:: 0.000752 atm-l
RT 82.057 (cc atm)
gmoleK
1
(293.16 K)

Solving Eq. (13-83)

D. Do It. If concentration polarization is negligible,

CZw c2b
-=-=100
c2p c2p

Ap - A1t = 100 - (37 - 0.4) = 63.4 attn

 778

and

DIm Kl =100 [ 1 ] [_1_] = 2100

D 2m K2 0.000752 63.4

For typical RO membranes Kl is slightly less than 1.0 and

D 1m =D water in membrane - 2.5 to 16 x 10-7 cm 2fs for high selectivity cel-
lulose acetate. (Soltanieh and Gill, 1981). Thus, a typical required salt
permeability is
DIm Kl
D2m K2 = 2100 = 1.2 to 7.6 x 10-10 cm 2fs

E. Check. These results agree with typical values of K2 in cellulose

acetate membranes from 0.017 to 0.039 and typical values of D 1m from
0.8 to 5 x 10-9 cm 2fs (Soltanieh and Gill, 1981).

F. Generalization. Obviously, these results can be changed for other

operating conditions and other membranes. For less selective cellulose
acetate membranes D 1m is higher. Values of diffusivities and distribu-
tion coefficients for a variety of membranes are tabulated by Soltanieh
and Gill (1981). If concentration polarization is important, .11t and
Czw = M C2b increase. Both these effects increase the value of
DIm KIf D2m K 2 . When RO is used to produce Ultrapure water,.11t is
negligible and significantly higher values of C2.w f cz.P can be obtained
(see Problem 13-D8).

13.6.3. Frictional Model For Porous Membranes (UF)

In ultrafiltration separation appears to be based mainly on a sieving mechanism

and not on solution and diffusion. The frictional model assumes that solvent
flow through the membrane is mainly laminar convective flow. Thus the sol-
vent flow can be estimated as Poiseuille flow. The solute flux depends upon the
convective flow (that is coupled to the solvent flow), diffusion, and friction. In
this model coupling between fluxes and external forces (friction) is important.
 779

The solute flux is due to the sum of contributions from the convective
flow and diffusion.

(13-84)
Jz = c2m U + J2djf

where u is the center-of-mass velocity of the pore fluid. The term c2rn u
represents -~1 dJlddx. The diffusive flux can be written as LzZX 2 • Lz2 can
again be related to the mobility, and X z is the force from Eq. (13-61). Thus,

aJl2] dC2m
J2di[ = m2 c 2rn [- [ aC2m p,T ~ + Y2
1 (13-85)

where Y2 is the frictional force on the solute due to friction against the mem-
brane. This external force can be related to the solute velocity within the pores
as

(13-86)

where f23 is a friction factor between the solute (2) and the membrane (3). The
mobility mz can also be related to a friction factor except now it is the friction
between the solute and solvent

(13-87)
mz = l/fz1 = D21 / RT

where (2) is solute and (1) is solvent. Combining these equations and solving
for J 2, we obtain

(13-88)

For dilute solutions Eq. (13-76) is valid and we have

(13-89)
780

As u approaches zero this gives us diffusion with a drag factor applied. The
solute flux can also be related to the solvent flow as
(13-90)
J 2 = E U c2p

where E is the porosity of the membrane and ~p is the permeate concentration

in the fluid. Assuming that the distribution coefficient given by Eq. (13-78) is
constant for the porous membrane, Eqs. (13-89) and (13-90) can be combined
and integrated to find C2. The integration is from x=O where c'2m=K2C2w to
X= - lm). where c2m"=K2c2p. ). is the tortuosity which takes into account the
fact that the pores are not straight, ). > 1. The result after some manipulation is

(13-91)

which allows one to predict the permeate concentration once the solvent velo-
city is known. For high negative velocities Eq. (13-91) simplifies to,

C2w K2 f21 (13-92)

~=
(f23 + f2d E
The solvent velocity due to Poiseuille flow in the pores without any fric-
tion would be,
dp .1Pm (13-93)
E u=-H - --H--
dx lm).

where H is the hydrodynamic permeability of the membrane and .1Pm is the

effective pressure driving force.
(13-94)

The terms .11t' and .11t" are the osmotic pressure differences at the two
membrane-liquid interfaces.

~lt' =1t [ c':-J- 1t(c,.)

(13-95a)

~1t" = c'~2m J-
1t [ 1t(c2p)
(13-95b)
 781

The hydrodynamic permeability in Poiseuille flow can be estimated from

£R2 (13-96)
H=--P
8J.l
where ~ is the pore radius and J.l is viscosity.

In addition to the pressure gradient there is a frictional force which must

be added to the pressure gradient. Then,

c2m 1
[ddxP+ Y E(MWh (13-97)
u = -.!!. 2
E

where Y2 is the force per mole of pore fluid given by Eq. (13-86), and
C7m/e(MWh is the molar concentration of solute in the pore fluid. The product
of these is the force per volume of the pore fluid. Combining Eqs. (13-86),
(13-92), and (13-97), we obtain

u =- H
E
[.11mpA. +
m fn J2
E(MW)2
1 (13-98)

The solute flux J2 is given by Eq. (13-90). If this equation is substituted into
Eq. (13-98) and the result is solved for u, we obtain

u =- H
E
[ H
1 + (MW)
~23
2E
c2p
j.11mpmA. (13-99)

Note that u is negative because of the chosen direction for x. Solution of Eqs.
(13-91) and (13-99) simultaneously allows prediction of the solvent velocity
and the permeate concentration through the membrane. (See Problem 13-C3).
In Eq. (13-91) C2w is considered to be known since it can be calculated from the
concentration polarization analysis. The solute flux can then be determined
from Eq. (13-90). The solvent flux can be found from Eq. (13-81) which after
substituting in Eq. (13-90), is
(13-100)
J'lOlv = EU Cl p
782

This model appears to be a reasonable physical model for sieve type

membranes with intrinsic rejections in the range 0 < RO < 1. The theory agrees
qualitatively with experimental data. Unfortunately, predicting the necessary
friction factors, f21 and f23 , and the tortuosity A is difficult Thus this model
should probably be considered a qualitative model which helps to explain the
transport processes inside sieving type membranes. It can serve as a guide to
qualitatively predict or explain the effect of various variables and to help extra-
polate data.

13.7. SUMMARY -OBJECTIVES

At the end of this chapter you should be able to meet the following objectives.

1. Predict the polarization modulus for both laminar and turbulent flow
when no gel layer is formed using mass transfer correlations.

2. Predict the mass transfer coefficient and the solvent flux for UP when a
gel forms. Determine whether or not a gel does form. Discuss the problems
with the gel formation theory.

3. Explain the mass balances for a completely mixed RO systems with con-
centration polarization. Estimate the solvent flux for this system.

4. Explain the mass balances for countercurrent gas permeation and explain
qualitatively why countercurrent operation is superior to other flow patterns.

5. Use the exact solutions for laminar flow to predict the concentration
polarization modulus in tubes or between flat plates.

6. Discuss the application of irreversible thermodynamics to models of

transport inside of membranes, and explain qualitatively both the solution-
diffusion and the frictional models.
 783

REFERENCES

Berman, A.S., "Laminar flow in channels with porous walls," J. Appl. Phys.,
24, 1232 (1953).

Blaisdell, C.T. and K. Kammermeyer, "Countercurrent and co-current gas

separation," Chern. Eng. Sci., 28, 1249 (1973).

Blatt, W.P., A. Dravid, A.S. Michaels, and L. Nelson, "Solute polarization and
cake formation in membrane ultrafiltration: Causes, consequences and control
techniques," in J.E. Flinn (Ed.), Membrane Science and Technology, Plenum
Press, NY, 1970,47-97.

Fane, A.G., "Ultrafiltration: Factors influencing flux and rejection," in RJ.

Wakeman (Ed.) Progress in Filtration and Separation 4, Elsevier, NY, 101-
180,1986.

Harris, FL., G.B. Humphreys and K.S. Spiegler, "Reverse osmosis

(hyperfiltration) in water desalination," in P. Meares (Ed.) Membrane Separa-
tion Processes, Elsevier, Amsterdam, 1976, Chapt. 4.

Hwang, S.-T. and K. Kammermeyer, Membranes in Separations, Wiley, NY,

1975.

Johnson, J.S., L. Dresner and K.A. Kraus, "Hyperfiltration (Reverse Osmosis),"

in K.S. Spiegler (Ed.), Principles of Desalination, Academic Press, NY, 1966,
345-439.

Lakshrninarayanaiah, N. Transport Phenomena in Membranes, Academic

Press, NY 1969.

Le, M.S. and I.A. Howell, "Alternative model for ultrafiltration," Chern. Eng.
Res. Des., 62,373 (1984).

Matthiasson, E. and B. Sivik, "Concentration polarization and fouling", Desali-

nation, 35, 59 (1980).
784

Merten, U. "Transport properties of osmotic membranes," in U. Merten (Ed.),

Desalination by Reverse Osmosis. MIT Press, Cambridge, MA, 1966, 15-54.

Porter, M.C., "Membrane filtration," in Schweitzer, P.A. (Ed.), Handbook 0/

Separation Techniques/or Chemical Engineers. McGraw-Hill, NY, 1979, Sec-
tion 2.1.

Potts, DE., RC. Ahlert and S.S. Wang, "A critical review of fouling of reverse
osmosis membranes," Desalination. 36. 235 (1981).

Rao, G. and K.K. Sirkar, "Explicit flux expressions in tubular reverse osmosis
desalination," Desalination. 27. 99 (1978).

Romero, C.A. and RH. Davis, "Global model of crossflow microfiltration

based on hydrodynamic particle diffusion," J. Membrane Sci .. 39, 157 (1988).

Shen, J.J.S. and RF. Probstein, "On the prediction of limiting flux in laminar
ultrafiltration of macromolecular solutions," Ind. Eng. Chern. Fundarn., 16,459
(1977).

Sherwood, T.K., P.L.T. Brian, RE. Fisher and L. Dresner, "Salt concentration
at phase boundaries in desalination by reverse osmosis," Ind. Eng. Chern. Fun-
darn .• 4. 113 (1965).

Soltanieh, M. and W.N. Gill, "Review of reverse osmosis membranes and tran-
sport models," Chern. Eng. Commun .• 12. 279 (1981).

Sourirajan, S., Reverse Osmosis. Logos Press, London 1970.

Sourirajan, S. (Ed.), Reverse Osmosis and Synthetic Membranes. Theory -

Technology - Engineering. National Research Council of Canada, Ottawa,
Canada, 1977.

Trettin, D.R. and M.R Doshi, "Limiting flux in ultrafiltratio of macromolecular

solutions," Chern. Eng. Commun .• 4, 507 (1980).
 785

Wijmans, J.G., S. Nakao, and C.A. Smolders, "Flux limitation in ultrafiltration:

Osmotic pressure model and gel layer model," J. Memb. Sci .• 20. 115 (1984).

Yuan, S.W. and A.B. Finkelstein, "Laminar pipe flow with injection and suc-
tion through a porous wall," Transactions ASME. 78.719 (1956).

Zydney, A.L. and C.K. Colton, "A concentration polarization model for the
filtrate flux in cross-flow microfiltration of particulate suspensions," Chern.
Eng. Commun., 47, 1 (1986).

HOMEWORK

A. Discussion Problems

AI. UF systems will almost always show concentration polarization. Under

what conditions would you expect this to lead to gel formation? When
would fouling be likely?

A2. Why are there multiple solutions for concentration polarization in lam-
inar flow?

A3. Theoretical analyses of turbulent flow are much more complicated than
laminar flow analysis. However, the concentration polarization analysis
for turbulent flow is simpler than the exact analysis for laminar flow.
Explain this paradox.

A4. Qualitatively, why is a countercurrent flow pattern superior to com-

pletely mixed?

A5. How can you determine if the solution obtained with Eqs. (13-3b) and
(13-16) or (13-18) is reasonable?
 786

A6. On one page develop your key relations chart for this chapter. This will
be a challenge because you will have room for only the more important
equations.

C. Derivations

C1. Derive the ratio of k/.1p for both laminar and turbulent flow. Compare
these results.

C2. Derive Eq. (12-57) for a countercurrent dialyzer.

C3. If a constant b < < 1, then exp (b) - l+b. Using this approximation,
simplify Eq. (13-91) when (- u lmAf21IRn < <1. Then solve Eqs. (13-
91) and (13-99) simultaneously to determine c2p' Note u < O. Note that
this assumption of small b will often not be valid.

C4. Prove that the lowest driving force for a countercurrent gas permeation
module equals the driving force for a completely mixed module.

D. Single-Answer Problems

Dl. We are doing an ultrafiltration experiment with a high molecular weight

solute in water. The bulk concentration is 0.005 weight fraction. The
osmotic pressure can be ignored. A 1.5 cm id tubular membrane is
used. Feed to each tube is 100 ml/sec. The membrane has a pure water
flux of 10.5 x 10-4 g/sec cm 2 at a pressure of 38 atmospheres. The
inherent rejection coefficient (M=I) for the membrane is RO = 0.972.
Under the conditions of this experiment (M > 1) the measured rejection
coefficient is R = 0.85. Estimate the solute diffusivity.
Data: p = 1.0 g/cm 3 , J.l =0.0105 poise, no gel is formed.

D2. We are ultrafiltering a dextran solution in a laminar thin-channel sys-

tem. The channel consists of two flat plates 0.1 cm apart. Fluid flows at
3.0 cm/min and is recycled through a 20 cm length of the channel. The
dextran is known to gel at -30 wt %. The solvent flux is measured as a
 787

function of dextran concentration. Kinematic viscosity v = 8.96 cm 2 /s.

For the data given in problem 12-04 estimate the dextran diffusivity.

D3. We wish to ultrafilter a 2 wt % aqueous solution of casein at 25°C. The

cg for casein is approximately 14 wt %. The channel is to be 25 cm
long and is a rectangular slit 0.025 cm deep. The bulk velocity is 50
cm/sec. The pressure drop across the membrane is 40 psi and casein
diffusivity is approximately 10-7 cm 2 /sec.

a. Predict the water flux at steady state.

b. If RO = 0.99 (without concentration polarization) and assuming

that the casein is mobile in the gel, estimate R when a gel layer
fOnTIs.

04. Estimate the average concentration polarization modulus for a 4 wt%

NaCI solution. D = 1.61 x 10-6 cm 2/sec. Ph = 1.025 glcc, Sc = 560,
CL = 0.987 gH2 0/ml. Operation is in round tubes 0.05 cm in radius, 50
cm long, and with a bulk velocity at the inlet o( 2.0 cm/sec. Flux rate is
3 x 10-4 glsec cm2 • Use the exact solution for laminar flow.

D5. We wish to do a reverse osmosis experiment on a 0.98 % salt solution.

The osmotic pressure is 6.958 atmospheres and 1t = be is valid. Opera-
tion is at 45 atmospheres. For pure distilled water a flux of 8.5 x 10-4
glsec cm2 is observed. Salt retention without concentration polarization
is 0.998. Tubular membranes 2 cm in diameter are used. Each tube is
80 cm long. Feed to each tube is 1.0 glsec. Including concentration
polarization find the average water flux.
Physical properties: D = 1.6 x 10-6 cm 2 /sec, P = 1.01 glcc, Sc = 560,
CL = 0.99g H2 0/ml (solvent concentration)

06. We are ultrafiltering a polymer in a hollow fiber system. The hollow

fibers are 100 cm long and 0.1 cm id. The polymer gels at cg = 0.015
wt frac. Inlet velocity Um = 3.5 cm/s and em = 0.001 wt frac. Flux rate
before gelling occurs is 3.2 X 10-4 cm 3/cm2 s. RO = 1.0.
a. Estimate the distance down the tube at which the polymer starts to
gel (± 0.1 cm).
788

b. What is the flux after gelling occurs?

DATA: p = 1.01 glee (assume constant)

1.1 = 0.03 poise (assume constant)
D = 5 X 10-7 cmz/s

D7. An experimental gas permeation system with flat plates is being tested.
Rejection, R O = 0.96. The flow path is 100 cm long and the plates are
0.02 cm apart. The high pressure side is at ten atmospheres and the low
pressure side is at one atmosphere. The total gas flux through the mem-
brane is J = 1.0 x 10-3 cc (SzTP). The inlet gas velocity is 100 cm/s.
cm s
Operation is at 25°C and the system is an ideal gas. Find the average
concentration polarization modulus.

DATA: Average MW on high pressure side = 40

D = 0.42 cm3 /s
1.1 = 1.4 x 10-4 poise

D8. We wish to use a cellulose acetate membrane to purify tap water for an
electronics planL The tap water contains 120 ppm IDS (total dissolved
salts) and we wish a product which is 1 ppm or less . .1p = 110 atm.

a Can we do this with existing membranes?

b. If we can do it, what cut a should be used if the RO module is per-
fectly mixed? Set c p = 1 ppm

c. Predict J:01v and J.olut.e if 1m = 1.2 x 10-4 cm for part b.

d. If M = 2.0, set cp = 1 ppm and find a, J~v and Jso1ut.e.

DATA: Operate at 18°C. Assume properties of water (see Example
13-1). Use p. 17-23 in Perry and Green (1984) to estimate 1t. Assume
salts can be treated as if they are NaCI. Kz = 0.025, D 2m =
=
2.5 x 10-9 cm3/s, D w 1.0 x 1O~ cmz/s, Kl 0.98. =
Ignore concentration polarization in parts a, b, and c.
 789

D9. Merten (1966) reports values for transfer of sucrose through a cello-
phane membrane at 25°C. At infinite dilution, E = 0.67 and

If- UE =0.9 x 10-4 cm/s at L1Pm = 1 atm, determine:

b. ~
Note: Watch your units.
 Part IV

SELECTION AND SEQUENCING

 NOMENCLATURE CHAPTER 14

linear equilibrium constant for adsorption

bottoms flow rate of melted crystals

heat capacity bulk adsorbent

C;.sorb heat capacity sorbate

distillate rate

capacity factor in chromatography Eq. (7-32a)

K.olv membrane permeability

L column length. m

LMI'Z length of mass transfer zone. m

M concentration polarization modulus

rate adsorbate product

p pressure
p permeate rate

<kvg average adsorbed phase loading

<kat adsorbate loading at saturation

Q heat supplied or removed

heat for condenser and melter

heat for reboiler

heat required for regeneration

R gas constant

LID. external reflux ratio

external reflux ratio for melter
membrane thickness

793
794

T temperature, K

To surroundings temperature, K

Tc cold (refrigeration) temperature

TRG temperature regeneration gas is supplied at

Ts steam temperature

Tim supply temperature for melter

Weq equivalent work

WmID minimum reversible work

W rev reversible work

WRO equivalent work for RO

~j mole fraction component i in phase j

Yi mole fraction in vapor

Greek

CX;j separation factor, Eq. (14-1)

'Yi activity coefficient

.6.H ads heat of adsorption

power cycle efficiency

Carnot cycle efficiency for refrigeration

latent heat of vaporization

latent heat of freezing

factor defined in Eq. (14-21)

1t osmotic pressure

factor for excess regeneration gas

 Chapter 14

SELECTION AND SEQUENCING OF SEPARATIONS

Throughout this book we have been discussing mainly known results - we now
move into the unknown area of selection and sequencing of separations. This
is an area which has been extensively studied for distillation and to a lesser
extent for extractive and azeotropic distillation - all equilibrium staged
processes. Very little research has been done for the rate controlled separations
which are the subject of this book. In this chapter we will discuss what has
been done, and extend by analogy the results obtained for equilibrium staged
processes to rate controlled processes. The results obtained are logical, but are
not proven; thus, they should be used carefully.

We will first consider the general characteristics of separation problems.

Then, an overview of separation methods will be presented. The purpose of
this overview is to aid in the selection of separation methods which will work.
Energy criteria for evaluating separation processes will be presented. After a
brief introduction to the strategy of process design and synthesis, heuristics
(rules of thumb) for developing separation schemes will be discussed. Evolu-
tionary methods which are used to improve the initial sequences will be briefly
introduced. Finally, needed research will be outlined.

14.1. GENERAL CHARACTERISTICS OF THE SEPARA nON PROB-

LEM

There are so many possible types of separation problems that it is hard to be

specific about your immediate problem which becomes the separation problem.

795
796

However, there are a few abstract ideas which are useful in looking at the
separation problem.

1. Invariably, there are several (n) components present and the

initial problem is multicomponent Often only one or two of
these components are valuable.

2. Most real problems will require several methods in series.

Often, one or more separation methods capable of separating
most or all of the components are readily available. There is a
strong temptation to immediately use these methods. Resist
this temptation.

3. The separation can be split into n - 1 binary pairs. These

binary pairs are not unique since they depend upon the physi-
cal properties used by different separation methods (see
Section 14.2). For any given set of n - 1 binary pairs there
will be one separation which is most difficult This most
difficult separation may be different for different choices of
separation methods.
4. If we know how to do all of the separations then the problem
becomes one of choosing the appropriate set of separation
methods. If we do not know how to do all of the separations,
then the initial problem is to find some set of separation
methods which will do the separation. The next problem is
detennining the optimum set of separation methods.
5. The appropriate separation method depends upon where you
are in the life cycle of the product If you are racing to be first
on the market and the product can demand a high price, then
the series of separation processes which can be put on stream
fastest has a definite edge. For more mature products the cost
of the separation methods becomes much more important and
different separation processes may be optimum.
6. The scale of operation is also important. A process which is
economic on a small scale may not be economic on a large
 797

scale and vice versa. For example, cyrogenic distillation of

air is currently the most economic air separation method at
very large scale, but either adsorption or membrane processes
or more economic at small scale.

7. The concentration of the feed stream affects the separation

costs. The effect of the naturally occuring concentration on

IOr----------------------------------------------,----~

Vitamin 8-1: •

\\ined gold •
Gold from
Penicillin _
sea water

o
• Bromme from sea water
•
Sulfur from
stack gas (cost)
-,

Wat~r from ~J v.'at~r

-4 0 12 14
-log (.:on,~ntration. mole fraction)

Figure 14-1. Relation between pure product cost and raw material concen-
tration (Sherwood et ai, 1975). Reprinted with permission
from T.K. Sherwood, RL. Pigford and C.R. Wilke, Mass
Transfer, McGraw-Hill, 1975. Copyright 1975, McGraw-Hill.
798

price is illustrated in Figure 14-1 (Sherwood et al, 1975). In

general, products which are dilut.e cost more to separate.
Gold is an interesting example of this. Gold can be recovered
from sea water, but this is not currently economic since mined
gold is more concentrated and hence is cheaper.

8. Every industry has its favorite separation methods. Since

these methods are known to work within that industry, there is
a stong predisposition to use those methods.

9. The criteria used to select a separation method for research

and development (either academic or industrial) are different
than the criteria used to select separation methods to solve a
problem. The focus of this chapter is on solving separation
problems. Bravo et al (1986) and Fair (1987) discuss the
results of a study to evaluate separation processes as candi-
dates for further research with respect to significant energy
savings.

14.2 OVERVIEW OF SEPARATION METHODS

The purpose of this section is to provide an overview of separation methods

and a very preliminary guide on methods of selecting separation methods
which will work.

There are a variety of ways to classify separation methods. One

approach is shown in Figure 14-2. This book has been concerned with diffu-
sional separation processes where the separation is at a molecular level.
Mechanical separation processes involve the separation of bulk phases. Real
problems will often require both mechanical and diffusional separation
methods. The mechanical separation processes are covered in detail elsewhere
(Evans, 1980; Jacobs and Penney, 1987; Ludwig, 1979; Perry and Green, 1984;
Schweitzer, 1979).
 799

Chemical
Mechamcal DiffusIOnal Modifications

Centrifuge D,L isomer sep.

Cyclone Ester production
Decanter
Demlster

Electrostatic Separator

T
Emulsion Separator
Heterogeneous
Expression Homogeneous
Filtration Mass Spectrometer
Flotation
High-gradient Magnetic
Impingement Separator
~ Pressure Diffusion
Thermal Diffusion
Ultracentrifuge
Magnetic Electrophoresis
Sedimentation
Scrubber
Non-Equilibrium Equilibrium
E IEF
ITP
Sink-float

Membrane or Barrier Non·Membrane Gas-Solid Liquid-Solid Gas-Liquid

Kinetic Adsorption Adsorption Absorption

Electrodialysis Adsorption
Freeze Drying Molecular Sieve Distillation

!
Pervaporation Molecular
Molecular Sieve Clathration Azeotropic
Gas permeation Distillation
Sublim/Desublim Crystallization Extractive
RO
Precipitation Flash
UF
Zone Melting Reactive
Gas Diffusion
Ion Exchange Steam
Chromatography
Ion Exc'~s'on Vacuum
Affinity
Leaching Evaporation
Capillary
Washing Foam Frac!.
Electrochromatography
Drying Solids Stripping
GLC
Liquid-Liquid
GSC
Extraction
Ion Ex.
Dual Temp. Exchange
LLC
Liquid Membrane
HPLC
LSC 3-phase
PC Dialysis
SEC
TLC

Figure 14-2. Classification of separations.

800

The third major category, chemical modification, involves doing a reac-

tion which simplifies the separation. In the production of esters and MTBE in
distillation columns the reaction and the separation are done simultaneously. In
D, L amino acid separations one of the isomers first reacts to fonn an ester, and
then the remaining isomer in the acid fonn is easily separated from the ester.

The diffusional separation processes can be further subdivided into

homogeneous (one phase present) and heterogeneous (two or more phases
present). Electrophoretic processes (Chapter 11) are considered to be homo-
geneous since the actual separation occurs in a single phase-the addition of a
second phase as a anticonvective medium is optional. The heterogeneous
separation processes have been further subdivided into equilibrium and none-
quilibrium separations. Nonequilibrium processes without a membrane or bar-
rier present are relatively Tare. Kinetic adsorption was discussed in Chapters 6
and 8. The nonequilibrium processes with a membrane include all the mem-
brane processes discussed in Chapters 12 and 13. The equilibrium processes
have been subdivided based on the phases present except for chromatography
which is listed separately. I have included chromatography as an equilibrium
process since the major reasons for separation in these methods are equilibrium
differences even though the column may never be at eqUilibrium. The equili-
brium processes also include crystallization methods (Chapters 2 to 5), adsorp-
tion and ion exchange (Chapters 6 to to), and the equilibrium staged processes
such as distillation, extraction and absorption which are covered in many other
books. The most commonly used diffusional separations are the equilibrium
processes.

Figure 14-2 has over 70 separation methods listed, and it does not
include all methods. In addition most methods have many variants which are
not listed, and many methods can be done with a variety of solvents, adsor-
bents, additives, or membranes. The result of all these variations is there are
literally thousands of separation methods available if one includes all the varia-
tions. The engineer's job is to select methods which will solve the problem
economically.

There are many other approaches to classifying separation methods

which may be useful in selecting the appropriate methods. Since separations
 801

are based on physical property differences, a classification based on the physi-

cal property differences can be very useful. One way of doing this is shown in
Table 14-1 where the sizes of species being separated is also indicated. Some
separations are listed more than once in Table 14-1 since different property
differences can be used. For example, absorption can be due to vapor pressure
differences (Physical absorption) or due to reversible or irreversible chemical
reaction.

The separations listed in Table 14-1 are also marked as to whether they
use an energy (E) or mass (M) separating agent. Separations using an energy
separating agent use heat (e.g. crystallization) or pumping energy (e.g. gas per-
meation) to effect the separation. Separations using a mass separating agent
add an additional species such as a solvent or purge gas to the mixture to be
separated. Note that the insoluble membrane, adsorbent or packing material is
not considered to be a mass separating agent. However, the solvent or carrier
gas in chromatography and the purge gas or displacer in adsorption is a mass
separating agent The disadvantage of adding a mass separating agent is that
usually the mass separating agent has to be removed before the product can be
used. Thus, a separation scheme using an energy separation agent is usually
required in addition to the mass separating agent processes. Separations such
as precipitation or adsorption can use either an energy or a mass separating
agent depending on how they are operated. Separations such as electrochroma-
tography use both energy and mass separating agents. Gradient chromatogra-
phy systems have two or more mass separating agents.

The separation factor obtainable with the different separations listed in

Figure 14-2 and Table 14-1 is a very important first indication of the feasiblity
of a given separation method. For equilibrium processes the inherent separa-
tion factor is defined as (King, 1980)

(14-1)

where Xi,1 is the mole fraction of component i in phase 1. Phases 1 and 2 are
assumed to be in equilibrium. For distillation Cli,j is the relative volatility. For
 00
Table 14-1. Property Differences for Separations otv
(King, 1980; Null, 1987; Rudd et al., 1973)

Property
Differences Molecules Macromolecule Particles

GLC(M)
flash dist. (E)
vapor distillation (E)
pressure absorption (M)
(reI. volatility) sublimation/desublimation (E)
stripping (M)
evaporation (E)
high vacuum distillation (E)

reI. volatity azeotropic dist. (M+E)

+ chern. interact extractive dist. (M+E)
reactive dist (M+E)

distribution extraction (M)

2 liquids liquid membranes (M)
f- two-aqueous phase extraction (M) -t
-LLC(M)-
melting point crystallization (E)
zone melting (E)

solubility crystallization
absorption/stripping (M)
leaching (M)
precipitation (E or M)

solubility + gas permeation (E)

diffusi vity reverse osmosis (E)
pervaporation (E)
liquid membrane (M)

molecular sieves (M) filtration!

size & shape f- size exclusion chromatography (M) -7 demister
clathration (M) centrifuge
(filtering)
f- ultrafiltration (M) -7 micro filtration (M)

00
ow
 00
Property ~
Differences Molecules Macromolecule Particles

electric ~ ion exchange (M) ~

charge ~ ion exchange chromatography (M) ~

ion exclusion (M)

electric charge ~ electrodialysis (E) ~

& mobility ~ electrophoresis (E) ~

~ isotachophoresis (E) ~

electric mobility ~ immunoelectrophoresis (M+E) ~

& chern. rxn ~ affinity electrophoresis (M+E) ~

surface GSC (M)

adsorption ~ TLC(M) ~

f- LC (M) ~

adsorption (M or E)
~ foam fractionation (M) ~

adsorption and shape molecular sieves (M)

adsorption & charge f- paper chromatography (M) ~

adsorption & diffusivity carbon sieves (M)

hydrophobic interaction f- hydrophobic chromatography (M) ---7

rate of evaporation molecular distillation (E)

diffusivity gas diffusion (M + E)

electric mobility & f- electrochromatography (M + E) ---7

SIZe (w. SEC)

electric mobility f- electrochromatography (M) ---7

& adsorption (w. adsorption)

reversible absorption/stripping (M)

chern. rxn liquid membrane (M)
extraction (M)
dual-temperature exchange (M)
f- affinity chromatography (M) ---7

irreversible chern. rxn absorption (M)

00
oVI
 00
Property 0
0'\
Differences Molecules Macromolecule Particles

density & size ultracentrifuge (E) settling

centrifuge
(sedimentation)
cyclone
sink-float
decanter

high gradient
magnetic mag. separation magnetic separator

surface wettability floatation

electrical electrostatic
conductivity separator

thermal diffusivity thermal diffusion (E)

isoelectric point (- isolectric focusing (M+E) ~

molecular wt. (- SDS electrophore~is (M+E) ~

 807

linear adsorption or chromatographic systems ~,j is the ratio of equilibrium

constants.

~ k~ (14-2)
~.=-=-,
J AJ k.J

For eutectic crystallization systems in the concentration range where a pure

species is in equilibrium with the eutectic the inherent separation factor is
infinite.

For non-equilibrium processes the inherent separation factor can always

be determined experimentally. The inherent separation factor is again defined
by Eq. (14-1), but Xi, 1 is now the mole fraction of species i in product 1, and the
two product streams are not in equilibrium with each other. In certain cases
where simplifying assumptions can be made ~j can be estimated. For exam-
ple, for reverse osmosis irreversible thermodynamics arguments were used in
Chapter 13 to estimate the change in salt concentration across a membrane. If
Eq. (13-83) is combined with Eq. (14-1) and we assume that the water distribu-
tion coefficient Kw = Kl is the same on both sides of the membrane, we obtain

cwp / cwwaste Dw m Kw Vw (~p - ~1t) (14-3)

aws= ' , =
, cg,p / cs,wutc D S,In Ks RT

where terms are defined in Chapter 13.

The value of the separation factor required before the separation method
is of interest depends on several factors. The ease of multi staging is very
important. In this context multistaging means that the separation acheived can
be multiplied many times and the separation agent can be reused. A packed
bed chromatograph and a packed bed distillation column are both multistage
systems even though no physical stages are present Equilibrium processes
such as distillation, adsorption and chromatography can easily be operated with
many stages, and the separation agent can be reused many times with low addi-
tional capital investment. Thus the separation acheived with a low separation
 808

factor can easily be multiplied many times in an economical fashion. Other

equilibrium processes such as crystallization can be staged, but the staging is
not nearly as convenient Membrane processes can be staged (operated in
series), but the low pressure product needs to be repressurized after each stage
and a new module is needed. Thus staging is not easy. The easier staging, the
lower the inherent separation factor can be. As a rough guide the ease of stag-
ing is

adsorption, chromatography> distillation> crystallization> membrane (14-4)

In addition to ease of staging we prefer energy separation processes com-

pared to mass separation agent processes since an additional separation is not
automatically required. This choice favors distillation, crystallization and
membrane processes.

Ultimately, the choice of separation method usually depends upon

economics (occassionally government regulations control the choice). Energy
use is important and the low energy use of the membrane processes is one of
their most appealing features; however, energy is not the only important factor.
Keller (1982) and Ho and Keller (1987) show that for distillation the per year
cost of capital is equal or greater than the energy cost. This is significant since
distillation is normally considered to be a low capital, high energy process.
Thus, capital costs must be included. Energy criteria for selecting separation
processes are discussed in more detail in Section 14.3. Since capital costs are
important, simple systems are favored This favors distillation, absorption, and
membranes. Very complex processes such as displacement adsorption and
electrophoresis tend to be used only when necessary. Some rough guidelines
for chosing processes are given in Section 14.5.

Distillation keeps appearing in the list of possible alternatives. Distilla-

tion is an energy separating agent process which is easy to stage and simple to
construct The design principles are well understood and many companies will
bid to design and build distillation systems. Distillation has good economy of
scale and can be operated in very large sizes. Two pure products can be
recovered, and the principles for sequencing columns have been extensively
 809

studied. Although energy use may be high, the energy is usually supplied by
low pressure steam which is often very cheap. The net result is that if distilla-
tion can be used it should be considered as one of the alternatives. Guidelines
for when distillation cannot be used are discussed in Section 14.5.

Example 14-1. Generation of Possible Separation Processes

Generate possible separations to separate ethanol from water.

Solution

A. Define. This problem was purposely ill-defined. The choice of

separation process will depend upon:

a. The purpose of the separation. Do we want pure ethanol or

the azeotrope, or is it a pollution control problem?

b. The feed concentration.

c. The scale of operation.

d. Presence of other components.

To further define the problem asswne a fermentation process is

used to produce approximately an 8 wt% ethanol in water mixture.
We want a water waste stream which can be discarded and pure
(99.9%) ethanol. Both small and large plants should be con-
sidered. For now, the trace contaminants such as glycerol will be
ignored.

B. and C. Explore and Plan. There are obviously a variety of

ways to proceed. For this example we will consider the physical
property differences listed in Table 14-1 to develop possible
separation methods.
810

D. Do It We will list the physical property and the separation

methods that result

1. Relative Volatility: Vapor-liquid equilibrium data is readily

available (e.g. Seader. 1984; Wankat. 1988). At low concen-
trations the relative volatility is quite high (- 12); however.
there is an azeotrope at 0.894 mole fraction ethanol at 1 atm
pressure. Thus distillation can produce pure water and the
azeotrope. and is a commercial method. The azeotrope con-
tains less water as the pressure is decreased (Seader. 1984).
At 70 mm Hg the azeotrope disappears. Thus distillation at
70 mm Hg could be used. but does not appear to be economi-
cally feasible (Black. 1980).

2. Relative volatility and chemical interaction. Azeotropic distil-

lation is commonly used to break the azeotrope. A hydrocar-
bon entrainer such as pentane or benzene can be used to form
a heterogeneous. ternary azeotrope overhead (e.g.• Black.
1980; Seader. 1984; Wankat, 1988). Extractive distillation
can also be used (Black, 1980). but unlike azeotropic distilla-
tion is not a commercial process.

3. Distribution between two liquids. Equilibrium data is avail-

able in several sources such as Leeper and Wankat (1982);
and Roddy (1981). Liquid-liquid extraction can be used to
either extract the ethanol from the azeotrope (Leeper and
Wankat. 1982) or extract ethanol from the feed. The first may
be economical while the second does not appear to be
economically feasible.

4. Melting point. Data is available in the Handbook of Chemis-

try and Physics and Lange's Handbook of Chemistry. Addi-
tion of ethanol to water causes a significant freezing point
depression (-51.3°C at 71.9 wt% ethanol). Freezing can be
used for removing water at low ethanol concentrations
although a countercurrent wash column (see Chapter 5) will
 811

be needed. This method might be useful for breaking the

azeotrope, but more data is needed.

5 Solubility and diffusivity. Membranes can be fabricated

which preferentially pass either water or ethanol. Some data
was given in Chapter 12. Reverse osmosis can be used to
preconcentrate a dilute feed, but by itself cannot produce pure
ethanol. Pervaporation units are available commercially for
breaking the azeotrope (see Chapter 12).

6. Surface adsorption. Adsorbents will show preference for

either ethanol or water. Activated carbon preferentially
adsorbs ethanol and is used for ethanol recovery from air
while activated alumina, com grits, starches and molecular
sieves prefer water. Cornmeal is used commercially for
breaking the azeotrope (Ladisch, et al., 1984). Operation is in
the vapor phase.

7. Adsorption and shape. 3A molecular sieves exclude ethanol

and adsorb water (Garg and Ausikaitis, 1983). This method is
used commercially for breaking the azeotrope operating with
a vapor.

8. Rate of evaporation. Water will evaporate more rapidly and

thus molecular distillation could be used, but will be very
expensive.

9. Chromatography. Both GLC and LLC work with a variety of

packings and are used for analysis. They are likely to be
expensive, but could be considered for breaking the azeo-
trope.

E. Check. Data can be checked from other sources, but this is

probably not necessary for an initial list like this.

F. Generalization. This list is not inclusive. It certainly does not

include all possible solvents, entrainers and adsorbents. A detailed
 812

literature search will generate additional infonnation. This list

does show that there are a variety of possibilities several of which
compete commerica11y. Also, use of an organized list such as
Table 14-1 is a good method to generate possible separation
schemes.

14.3. ENERGY CRITERIA

Energy requirements for separation processes are discussed in many sources

(Ho and Keller, 1987; Keller, 1982; King. 1980; Null. 1980; Robinson and Gil-
liland, 1950). In this section we will first discuss minimum work requirements
for separations. and then compare the equivalent work required for some
separation processes. Before starting. we will reiterate that energy use is only
one of the criteria used in selecting a separation process.

14.3.1. Minimum Work

The minimum reversible work required for the complete separation of a mix-
ture of ideal gases is (Keller. 1982)

(14-5)

If the feed, PF. and product, pp. pressures are equal this result simplifies to
c
(14-6)
Wmin,T,p =-RT 1: (Yi,F In Yi,P)
i=l

For the separation of a mixture of nonidealliquids the result is

(14-7)
Wmin,T,p = -RT ~ [Xi.F In (Yi Xi.F)]
1=1

These results are easily derived by postulating a reversible process. One

way of doing this is to postulate the existence of perfect semipenneable mem-
branes which will pass one component but none of the other components
 813

}----..- I at T,pp

2 at T, Pp
Feed _

PF ,T, Yi,F

1------< } - - - . n at T, Pp

\Membranes

Figure 14-3. Reversible membrane separation process for calculating

minimum work.

(Keller, 1982). This process is illustrated in Figure 14-3. After passing

through the membranes, the gases are completely pure but are at reduced pres-
sures.
(14-8)
Pi = Yi,F PF

Each of these gas streams is compressed isothermally to the product pressure

pp. For an ideal gas the reversible work required to isothermally compress one
mole from Pinitial to Pfinal is

(14-9)
W rev =RT In (Pfinal / Pinitial)
Since Plinal = Pp and Pi,initial = Pi in Eq. (14-8), the reversible work for one gas is

(14-10)
W~v,i = RT In (Pp / Yi,F PF)
The total reversible (and hence minimum) work for the separation is the sum of
Wrev,i for all components. The result is Eq. (14-5). More detailed develop-
ments which include partial separation and nonisothermal operation are given
by King (1980).

The minimum work required for any separation is easily determined.

This is illustrated in Example 14-2.
814

Example 14-2. Minimum Work of Separation

We wish to separate a liquid mixture of ethylbenzene and styrene

at 110°C. Determine Wmin,T,p for mole fractions ethylbenzene of
0, 10,20, ... , 100 in the feed.

Solution

Since the liquid can be assumed to be ideal, ~ = 1. Then Eq. (14-

7) becomes
(14-11)
Wmin,T,p = -RT [XE In XE + (I-XE) In (I-XE)]

for this ideal binary mixture. For this problem,

R = 1.9872 cal/gmole K, T = 383.16 K.
As an example, for a 0.1 mole fraction ethylbenzene feed

W min ,T,p(O.1O) = -(1.9872)(383.16)[0.1 In 0.1+0.9 In 0.9]

= 247.52 cal/gmole feed
The remaining results are sirniliar and are listed in the Table.

W'nun,T ,p Wmin,prod
xE,F (cal/gmole feed) (cal/gmole EB prod)

0 0
0.1 247.52 2475.2
0.2 381.01 1905.1
0.3 465.12 1550.4
0.4 512.44 1281.1
0.5 527.77 1055.5
0.6 512.44 854.1
0.7 465.12 664.5
0.8 381.01 476.3
0.9 247.52 275.0
1.0 0 0
 815

Note that the minimum work required to separate a 0.0 and a 1.0
mole fraction feed are both zero since these feeds are already pure.
The minimum work required for separation per mole of feed is
symmetric around a feed mole fraction of 0.5. For nonidealliquids
this will not be true. The third column in the table is of consider-
able interest since it shows the minimum energy required per mole
of ethylbenzene product. As expected, this value decreaBl!s as the
feed becomes more concentrated in ethylbenzene.

The energy requirements calculated in Example 14-2 s.~m quite reason-

able. The minimum work of separation is easily calculated and gives a simple
first approach to compare the difficulty of different separations. Unfortunately,
the fractional efficiency (WminIWactJJal) of most separation processes ranges
from a few percent to about twenty percent (Keller, 1982). Since distillation is
a benchmark process which other processes are compared to, it is interesting to
look at the maximum efficiency of distillation processes. This is shown in Fig-
ure 14-4 (Ho and Keller, 1987) for a simple distillation column producing two
pure products at minimum reflux. Actual columns must operate above the
minimum reflux ratio and the real efficiency will be 5 to 20% lower than the
values shown in Figure 14-4 (Ho and Keller, 1987).

Further research is required on the minimum energy requirements for

some of the more complex processes. In addition, research is required on how
the minimum energy requirements can be incorporated into a search for the
best separation method for a given problem.

14.3.2. Equivalent Work Requirements

Equations (14-5) to (14-11) determine the minimum mechanical work require-

ment. Many separations use heat instead of mechanical work as the energy
source. In order to compare different processes we will calculate the
equivalent work which is the work which could have been obtained by a rever-
sible heat engine. This equivalent work can be calculated from (Null, 1980)

Weq=Q [ T-T
TO 1Ep (14-12)
816

.-
Q)
Q)
0.
E 6

-
0
u
o~
>, c-
uo
c'';::;
.!!:! ~ 4
.5,? ro
:t;c.
Q)Q)
VI
E
~
E 2
'xro
~

o 0.2 0.4 0.6 0.8 1.0

Mole fraction of more volatile component
in feed

Figure 144. Maximum energy efficiency for simple distillation (Ho and
Keller, 1987). Reprinted with permission from Y.A. Liu, H.A.
McGee and W.R. Epperly (Eds.), Recent Developments in
Chemical Process and Plant Design, Wiley-Interscience, 1987.
Copyright 1987, Wiley-Interscience.

where Q is the heat supplied at absolute temperature T and the surroundings are
at To. The efficiency Ep is the power cycle efficiency compared to a Carnot
cycle. From this definition of equivalent work Null (1980) derives approximate
equations for the energy requirements of several commercial separation
processes. The following development is based on Null's (1980) development.

Consider first a simple distillation column with one feed, a condenser,

and a reOOiler. The approximate energy requirements are

(14-13)

where ~ and Qc are the reOOiler and condenser heat loads, respectively; Dis
the distillate product rate; A. is the latent heat of vaporization of the distillate
product; and RD = LID is the external reflux ratio. Combining Eqs. (14-12) and
 817

",0
"
~Q

.,
>
~ ..,
c
"-
0'
Q)

~ N

100 200 300 400 500 600

(37.8) (93.3) (149) (204) (260) (316)

T s' F (el
Figure 14-5. Equivalent work requirement for normal distillation. To =
37.8°C (Null, 1980). Reprinted with permission from Chern.
Eng. Prog., 76(8),42 (1980). Copyright 1980, American Insti-
tute of Chemical Engineers.

(14-13), we can determine the equivalent work required for simple distillation.

(14-14)

where Ts is the temperature of the stearn used to supply the heat. Note that Ts
> Treboiler since some temperature difference is necessary for heat transfer. To
is the temperature of the cooling water used. The predicted equivalent work
values normalized by DA. are given in Figure 14-5 (Null, 1980). Null (1980)
also presents results for somewhat more complicated distillation columns.
 818

Although the relative volatility does not appear explicitly in Eq. (14-14) or Fig-
ure 14-5, a. is an important parameter in determining Weq since a. controls the
required external reflux ratio. The value of Weq for distillation can now be
compared to the results for other processes.

Melt crystallization is analogous to distillation with the solid crystals tak-

ing the place of the liquid in distillation, and the melt taking the place of the
vapor. Processes were shown in Figures 5-4, 5-5, and 5-6. Consider the center
feed system shown in Figure 5-5. In the melter the approximate amount of
energy required to melt all the crystals is

(14-15)

where Be is the bottoms product of melted crystals, At- is the latent heat of
freezing, and RM is the external reflux ratio for the melter. If the heat is sup-
plied to the me Iter at temperature TIM where TIM> TM• then the equivalent
work for the heated end of the crystallizer is

(TsM - To) (14-16)

Weq.M = Be Ep (RM + 1) At- - . T
.. --
sM

where To is the ambient temperature of the surroundings. At the cooled end of

the crystallizer approximately the same amount of heat is removed. Usually.
refrigeration will be required to form the crystals for reflux. The refrigeration
is required at temperature Te. Then the equivalent work for cooling is

(14-17)

where ER is the Carnot cycle efficiency of the refrigeration system. The total
equivalent work for the melt crystallization process is the sum of Eqs. (14-16)
and (14-17).

W =B (R +1)'L[Ep(TSM-To)+(To-Te)] (14-18)
eq.cry e M "'f T1M T0 E
R
 819

Based on energy considerations only, we can now compare crystalliza-

tion to simple distillation by comparing the equivalent work terms from Eqs.
(14-18) and (14-14). Crystallization will be preferred on an energy basis if
Weq.ery < Weq.diA which occurs if

Null (1980) shows some step-by-step simplifications. If Be = D (same amount

of product), T. = TaM (normally Ts > TIM), and cooling can be done with nor-
mal cooling water (Te = To), then Eq. (14-19) becomes

(14-20)

Since latent heats of freezing are typically about 1/5 of the latent heat of vapor-
ization, Eqs. (14-19) and (14-20) often show an energy advantage for crystalli-
zation. However, Fair (1987) notes, "The specific energy advantages of cry-
stallization have not been a factor in its selection for commercial use." These
results do show the potential of crystallization particularly if energy costs
escalate.

Temperature swing adsorption of gases is another process which is

amenable to this type of analysis. Assuming that the column is adiabatic and
that the energy required to heat the column itself is small, we obtain a
simplified energy balance for the energy required for regeneration.

where Cp•sorb and Cp,B are the heat capacities of the sorbate and the bulk adsor-
bent in kJ/kgmole K and kJ/kg adsorbent K, respectively, and 'lavg is the aver-
age adsorbed phase loading at the end of the feed step. For a symmetric mass
transfer zone,

(14-22)
<lavg = <laat (L - 0.5 LMfZ) / L
 820

'l'in Eq. (14-21) is a factor to account for the use of excess regeneration gas,
Tads is the average bed temperature at the end of the feed step, Trcg is the aver-
age temperature of the bed after the regeneration step, Mlads is the heat of
adsorption in kJ/kgmole, Nads is the average rate the adsorbate product is pro-
duced in kgmole/hr. Then; is the total regeneration heat required per kg mole
adsorbed.

Combining Eqs. (14-12) and (14-21), we obtain the equivalent work

required

(14-23)

where TRG is the supply temperature of the regeneration gas, TRG > Treg. This
result can be compared to Eq. (14-14) for distillation. Adsorption uses less
equivalent work when

(14-24)

Usually, MladJI > A., and hence C;/A »1. Adsorption typically adsorbs the least
volatile componenL Thus, NadJI/D corresponds to BID if either adsorption or
distillation can do a separation. Adsorption will be favored if BID is small
which means that a small amount of nonvolatile material is to be removed.
Adsorption is also favored if distillation is difficult and a large reflux ratio is
required. Null's (1980) numerical results are shown in Figure 14-6.

Reverse osmosis directly uses mechanical energy and the work required
is easily calculated. The mechanical work required is the volumetric flow rate
times the pressure increase. Although the entire feed stream is pressurized, part
of the mechanical work can be recovered from the high pressure waste stream.
Only the permeate product is completely degraded. Assuming that the permeate
pressure and feed pressures (before pressurizing) are the same, we obtain

(14-25)

where P is the volumetric flow rate of permeate product and ~p is the pressure
 821

"...0
0

-
N
G)

110
>

-
co
u. co
(/)

..... -
c
.2
Q.
:!
N

0
(/) 0

".c
c(
110
(J
.c co
~
G) .,
>
0
..a N
co
0.1
0
0
Ix: 500 600
200 300 400
(93.3) (149) (204) (260) (316)

Ts ,F(C)
Figure 14-6. Comparison of equivalent work requirements for temperature
swing adsorption with regeneration at 316°C versus simple dis-
tillation (Null, 1980). Reprinted with pennission from Chern.
Eng. Prog .• 76(8), 42 (1980). Copyright 1980, American Insti-
tute of Chemical Engineers.

drop across the membrane. We have also assumed that the efficiency of energy
recovery from the high pressure waste stream is 100%. Obviously, ~p must be
greater than the osmotic pressure difference. From Eq. (12-27) ~p can be
related to the volumetric solvent flux Jso1v = PIA and the osmotic pressure
difference

Ap __
LJ.
PIA
~--'-~- + M 1t (Cb) - 1t (cp) (14-26)
K so1v 11m

where 1t is the osmotic pressure, A is the membrane area, and M is the concen-
tration polarization modulus. Combining these equations, we obtain

(14-27)
822

The work required is inversely proportional to the permeability of the mem-

brane which is a rate constant Thus, the work requirements depend upon the
membrane rate properties in addition to physical properties. This differs from
previous results obtained for equilibrium processes where rate properties do not
enter into the equivalent work expressions. Note that one must be careful with
units in Eqs. (14-25) and (14-17). In SI units WRO will be in J/S.

With reasonable values for fluxes and permeabilities the work calculated
from Eqs. (14-25) or (14-27) is very low when compared to the equilibrium
processes which involve a phase change (note that there is no latent heat term
in these equations). Thus, from an energy point of view reverse osmosis and
the other membrane separations are very appealing. However, RO is not a
complete separation process since the reject stream is a concentrated solute in
solvent, and cannot be pme solute. The concentration of the reject stream is
limited by the osmotic pressure which increases rapidly at high concentrations.
If a complete separation of solute and solvent is required then RO can be
advantageously coupled with another separation method in a hybrid process.

Further research is required to determine the equivalent work require-

ments for more complex separation processes. Also, methods for using the
equivalent work results to aid in the selection of the appropriate separation pro-
cess are needed.

14.4. STRATEGY OF PROCESS DESIGN AND SYNTHESIS

The appropriate strategy to use for process design and synthesis has been
extensively studied recently (Douglas, 1988; Rudd et al, 1973; Westerberg,
1982). We will follow Westerberg (1982) and present some guidelines without
examples. Detailed examples are given in these references, and in Blumberg
(1988) for liquid-liquid extraction.

Westerberg (1982) presents five guidelines for attacking design prob-

lems.

1. Evolve from simple to complex calculations.

Use approximate calculation methods initially. So many changes
 823

will be made in the design that the use of complex and time-
consuming detailed calculation methods is unecessary. Douglas
(1988) discusses short-cut calculation methods for equilibrium
staged processes. Detailed calculation methods should be used in
the later stages of design.

2. Quickly develop a complete feasible solution.

This is a depth first approach and requires that one avoid back-
tracking to improve the design of subproblems. The first feasible
design should be considered to be a learning experience. It shows
where the easy and where the difficult steps are. It may even show
that the entire project probably cannot be done economically with
existing technology. It serves as a guide for later backtracking to
improve steps. Douglas (1988) considers this step in detail.

3. Develop approximate criteria for screening among alternatives.

It is not possible to determine if one has a good design until one
has a design. It is also impossible to evaluate the criteria which
will be used to assess the design without a completed design. To
avoid this dilemma one should use aproximate criteria and heuris-
tics (see Section 14.5.) to obtain an initial design.

4. Alternate between global and detailed solutions.

A global approach is needed to keep the overall goal of the design
in mind. This goal is then partitioned into subgoals which are
further subdivided into additional subgoals until one obtains
subgoals which can be designed without further partitioning. This
is known as a top-down approach. Once the lowest level subgoals
have been identified, detailed design using approximate, short-cut
methods is done. Initially, the purpose of this design is to find
those subgoals which preclude obtaining an economic solution.
This is called a bottom-up approach. If any impossible subgoals
are found one returns to the global design and makes appropriate
changes. Some sort of strategy such as heuristics or an evolution-
ary approach (see Section 14.6.) should be used for generating the
next global design.
824

5. When trying to prove that concepts will not wolk, use optimistic conditions.
Most initial design concepts will not work. During the screening
process the designer's job is to prove things will not work or can-
not be economic. This type of comparison is convincing only
when optimistic guesses are made (e.g. make the optimistic
guesses that there is no entrainment in a crystallization process or
that there is no concentration polarization in a membrane separa-
tion). If one were conservative in making the guesses, the failure
of the design might be due to the guess and one does not have a
clear reason for rejecting the design.

14.5 HEURISTICS

Heuristics are rules of thumb which allow us to make preliminary choices in

complicated situations without doing a large number of detailed calculations.
The use of heuristics for general selection of separation processes has been
considered by Douglas (1988), Hendry et al (1973), Keller (1982), Kelly
(1987), King (1980), Liu (1987), Nishida et al (1981), and Rudd et al (1973)
among others. The heuristics have been most highly developed for distillation.
Some work on heuristics has been done for modifications of distillation such as
azeotropic and extractive distillation, and for absorption and extraction. Only
the heuristics for sequencing distillation systems have been rigorously tested.
The other heuristics appear to be valid, and have received some modest testing.

One problem with heuristics is that different heuristics often conflict with
each other. Then the proper choice of heuristic is unclear. All of the heuristics
should be preceeded with the words, .. IT all things are equal ... " Liu (1987) has
rank ordered heuristics for distillation and modified distillation systems.
In this section a large number of possible heuristics for the selection and
sequencing of separations will be listed and the reasoning for the heuristic will
be explained. These heuristics were developed either by generalizing the
heuristics developed for the equilibrium staged separations or from comments
in the literature. Except for the distillation heuristics, these heuristics have not
been rigorously tested and should always be checked with calculations.
 825

14.5.1. G. General and Flowsheet Heuristics

10. Always generate more than one alternative separation method or

sequence.
Although not usually explicitly stated, this idea is inherent in the ideas of pro-
cess synthesis and good problem solving practice.

20. In developing a flow sheet select the separation schemes first before
adjusting temperatures or flow rates.
This heuristic is stated by Rudd et al (1973) and Kelly (1987). Compared to
temperature and flow rate adjustment separations are much more expensive and
require more energy. Thus, the separation problems should receive priority.

30. Reduce the separation load by splitting streams, blending or mixing as

appropriate.
This heuristic applies to the development of flow sheets and was stated by
Rudd et al (1973) and Kelly (1987).

40. Favor schemes which give the smallest product sel

This heuristic states that components which will eventually be in the same pro-
duct should not be separated. This heuristic was listed by Kelly (1987), King
(1980), Liu (1987), Nath and Motard (1981) and Thompson and King (1972)
and others. Nath and Motard (1981) consider this the primary heuristic while
Liu (1987) ranks it third. If extended to chromatography it implies that com-
ponents which will be lumped together as a waste stream should not be
separated.

50. Do not produce purer products than required.

This can be considered as a corollary of heuristic 40.

6G. Do not remix partially separated streams.

If components are ultimately to be separated, any partial separation should be
retained since remixing is an increase in entropy. Although not explicitly
stated elsewhere, this heuristic is a generalization of common practices such as
the use of two-feed distillation columns, "super-loading" in adsorption
columns, and segregated recycle in chromatography.
 826

7G. Consider using the standard separation methods of the industry.

These methods may not be optimum. but they will probably give a working
base case. This heuristic is often standard operating procedure. but should
always be used in conjuction with heuristic #lG.

8G. Favor separation methods which are known within the company.
The separation methods which the company is familiar with will require the
least design and development time. These methods will also be easiest to sell
to management. Heuristic #lG should not be forgotten when applying this
heuristic.

9G. For new products look for separations which can be put on stream quickly.
With new products the company which gets to the market first often captures a
large proportion of the market Thus. separation schemes which will take a
long time to develop are not favored for the first plant Known separation
processes will be favored (See heuristics 7G and 8G).

1OG. Consider hybrid plants when one separation method by itself is not ideal
for a given separation problem.
Hybrid plants use two or more separations for the same problem. An example
would be concentrating a stream by reverse osmosis before doing evaporation
even though evaporation by itself could do the separation. Hybrid plants can
utilize the strength of a separation in the concentration range where it is best,
and avoid weaknesses.

14.5.2. D. Design Heuristics

ID. Favor separations using energy separation agents.

These separations do not add anything to the mixture to be separated. This
heuristic is explicitly stated by Liu (1987) and is first in his ordered list. This
heuristic leads to heuristic IS (favor distillation).

2D. If a mass separating agent is used remove it early and do not use a mass
separating agent to recover a mass separating agent.
 827

This heuristic is commonly stated (Kelly, 1987; Liu, 1987; Nath and Motard,
1981; Rudd et aI, 1973). It probably does not strictly apply when solutes must
be dissolved in a solvent so that they can be processed. Rudd et al (1973) also
suggest considering the use of a chemical already present in the process even if
it is not optimum based on its properties alone.

3D. Prefer processes with only one and not two mass separating agents.
The use of an additional mass separating agent such as a gradient in chroma-
tography makes the downstream recovery more difficult However, Mazsaroff
and Regnier (1986) note that gradients are required for elution of
macromolecules when separation is due to surface forces.

4D. Avoid excursions in temperature and pressure.

Both very high and very low temperatures should be avoided, but when neces-
sary go high not low (Kelly, 1987; King 1980; Liu, 1987; Rudd et aI, 1973).
This implies that vacuum distillation and distillation with refrigeration are less
likely to be economical. It also favors PSA over VSA. This heuristic also
shows a preference for liquid chromatography instead of gas chromatography
since LC operates at lower temperatures. Both pervaporation (which uses a
vacuum) and low temperature crystallization are contraindicated.

5D. Do the cheapest separation next.

This heuristic was developed by Thompson and King (1972) and is also used
by Douglas (1988). The reasoning is that every separation method used
reduces the volume of material to be processed; thus the more expensive
separations will process less material and will cost less. Although the use of
this heuristic alone may not give optimum flowsheets, it does appear to gen-
erate reasonably good flowsheets quickly. However, this heuristic was tested
only with distillation type separations. Heuristics 7S and 8S for adsorption
and chromatography follow this heuristic.

6D. Avoid solids and if they are present remove them early.
Solids cause processing problems. Separations involving solids (e.g. adsorp-
tion and chromatography) are difficult to operate as countercurrent processes.
828

70. When a solid material needs to be processed consider putting the sample
into solution and removing insolubles to be a separation step.
This heuristic is explicitly stated for chromatography by McDonald and
Bidlingmeyer (1987), but can be extended to any process such as extraction,
crystallization, or adsorption where solids are first dissolved.

80. Consider side withdrawals fer sloppy splits.

Sloppy splits are problems where the product does not have to be of high purity.
This heuristic was developed by Tedder and Rudd (1978) for distillation, but
can be extended to other separations. For example, a side withdrawal for
column switching of chromatography (see Figure 7-16) is roughly analogous to
a side withdrawal in distillation. Side withdrawals would be relatively easy to
use in fractional extraction or crystallization. A single cascade with a side
withdrawal will be quite inexpensive when the side withdrawal can produce the
desired purity.

14.5.3. C. Component and Composition Heuristics

IC. Remove corrosive, hazardous, and unstable compounds early.

This heuristic is commonly stated (Douglas, 1988; Kelly, 1987; Liu, 1987;
Rudd et ai, 1973). Corrosive materials often require expensive materials of
construction, and overall cost can be reduced by removing these components
early. Hazardous materials may require special conditions and also cost more
to process. Unstable compounds decompose less if they are removed early. In
particular, reactive monomers and proteases should be removed early.

2C. Do the difficult separations last.

The difficult separations will require the longest column. It usually makes
sense to do these separations with no other compounds present unless these
other components change the equilibrium and make the separation easier. This
heuristic has been commonly presented for both equilibrium stage separations
(Douglas, 1988; Kelly, 1987; Liu, 1987; Rudd et ai, 1973) and for chromatog-
raphy (Guiochon and Colin, 1986; McDonald and Bidlingmeyer, 1987; Wan-
kat, 1986). Nath and Motard (1981) state that the "easiest separation should be
done first", which is essentially the inverse.
 829

3C. Do high recovery and high purity separations last.

This heuristic is closely related to heuristic 2C and is based on the same reason·
ing.

4C. Remove most plentiful component next.

This heuristic was developed for distillation (Douglas,1988; Kelly, 1987; Liu,
1987; Rudd et al, 1973) but is probably applicable to other separations as well.
Use of this heuristic reduces the load on downstream processes.

5C. If component compositions are similiar, favor 50/50 splits or equal size
parts.
This was also derived for distillation and serves to balance the distillation
column (Douglas, 1988; Kelly, 1987; Liu, 1987; Rudd et aI, 1973). Its use for
other separations is not clear.

6C. In organic systems remove water early.

In organic systems water often causes significant problems due to immiscibility
(in distillation), freezing out (in crystallization), or strong adsorption (in
adsorption). Thus, removing the water early even though it may be only a trace
component is often good practice.

14.5.4. S. Separation Specific Heuristics

IS. Favor distillation unless a < <Xcritical.

This heuristic is commonly stated, but the value of Clcritical varies. Null (1987)
uses Ucntical = 1.05, Nath and Motard (1981) and Douglas (1988) use Ucritical =
1.10, for comparison with adsorption Ruthven (1984) uses Clcritical = 1.25 while
Keller (1982) uses a range from 1.2 to 1.5.

2S. In distillation remove products as distillates.

This heuristic is commonly stated (Douglas, 1988; Kelly, 1987; Rudd et aI,
1973). The reason is that most degradation products have high boiling points
and will appear in the bottoms. Thus, distillate products are more likely to be
pure. This heuristic can be extended to other separations (see Problem 14·
All).
830

3S. If the product is sold as a solid the last purification step should probably be
crystallization.
Crystallization can be used both to pnxluce crystals with the desired size distri-
bution and as a final polishing step.

4S. Start considering adsorption instead of distillation when have:

a. (l < 1.2 to 1.5.

b. An azeotrope forms.

c. lightgases

d. a.ds is very high

e. feed is dilute « 10 to 25% of product). This is particularly important if
the product is the heavier component

f. product is thennally unstable

g. there are two sets of compounds with boiling point overlap. Note that
extraction is also commonly used in this case.
This set of heuristics is from Keller (1982).

5S. Consider temperature swing adsorption if adsorbate concentration in the

feed is < 5 mole%.
Keller (1982) notes that energy costs in TSA can be very high and they
increase as the adsorbate concentration in the feed increases.

6S. Consider displacement adsorption only when a.ds > 2 and (ldist < 1.2 to
1.3.
Keller (1982) notes that displacement adsorption processes are quite complex,
and the process should be considered as a last resort.

7S. In adsorption and chromatography consider methods with cheap pac kings
first
This heuristic is related to heuristic 5D. For adsorption this heuristic favors
activated carbon. For chromatography it favors plain silica, activated
alumina, or ion exchange chromatography systems.
 831

8S. In adsorption and chromatography use expensive packings last.

This is common practice particularly with HPLC and affinity chromatography
where the packing can be very expensive. Using the expensive packing last
means that the solution is cleaner and is less likely to foul the packing.

9S. Chromatography is expensive - use it only when necessary.

This heuristic is stated by McDonald and Bidlingmeyer (1987).

lOS. In large scale liquid chromatography always optimize the solvent sta-
tionary phase system.
McDonald and Bidlingmeyer (1987) note that the selectivity ex is the most
important parameter, and increasing ex by optimization is well worth the
effort. However, Kopaciewicz and Regnier (1983) note that solvents need to
be compatible for chromatography columns in series or an extra separation
will be required between columns. Solvent selection is also very important in
absorption (Keller, 1982), azeotropic and extractive distillation (Van Winkle,
1967; Wankat, 1988), and extraction ( Blumberg, 1988; King, 1981).

lIS. Consider large-scale gas-liquid chromatography instead of distillation if

ex < 1.10. If ex> 1.2 prefer distillation or vacuum distillation.
This heuristic was stated by Bonmati et al (1984). However, note that liquid
chromatography will also be a competing process and LC has been more suc-
cessful for large-scale applications.

12S. Do ultrafiltration or microfiltration before reverse osmosis.

This heuristic is implied by many process flow sheets. The reasons are that
UP and microfiltration are better able to handle feeds containing particulates,
and they have has higher fluxes. Thus the UP or microfiltration system will
protect the RO system and will decrease the volume which has to be pro-
cessed by the RO system.

13S. Consider RO for preconcentration of dilute aqueous streams in a hybrid

process.
This is one of the classical uses of hybrid processes and is often done with RO
followed by evaporation.
832

14S. If hydrogen or carbon dioxide need to be purified, consider gas permea-

tion.
With currently available membranes these two gases have the highest per-
meabilities, and these applications have now been widely commercialized.

As was noted at the beginning of this section many of these heuristics have
not been rigorously tested. In addition, an ordering of these heuristics for
nondistillation applications is needed. This implies that considerable research
is needed. In the meantime the heuristics must be used with caution.

Example 14-3. Use of Heuristics

Use heuristics to generate preliminary separation fiowsheets for

Example 14-1.

Solution

Following heuristic IG, we will generate several sequences.

Heuristics 2G to 6G, 8G and 9G are not applicable. Heuristic 8G
suggests:
Distillation to produce waste water and azeotrope (also supported
by ID, IS and 2S).
Adsorption or azeotropic distillation for breaking the azeotrope.

Heuristics 9G and 13S suggest preconcentrating feed with RD.

Heuristic 2D suggests removing mass separating agent early. This
can easily be done in the azeotropic distillation since the azeotrope
is heterogeneous.
Heuristics 3D and 65 suggest adsorption should be either tempera-
ture or pressure swing.
Heuristics 5D and 2C lead to distillation first, followed by breaking
the azeotrope.
Heuristics 70, 8D, IC and 3C are not applicable.
Heuristic 4C suggests removing water first.
Heuristic 2S is satisfied by distillation taking the azeotrope as over-
head, but is not satisfied by the usual azeotropic distillation.
 833

Heuristic 5S suggests TSA may have high energy costs for break-
ing the azeotrope. An alternative such as PSA should be con-
sidered.
Heuristic 9S probably rules out chromatography.
Heuristic 12S would be applicable if solids were present in the
feed.

The net result are the flow sheets shown in Figure 14-7. Figure
l4-7a is a fairly standard distillation column followed by azeotro-
pic distillation. Reverse osmosis is optional as a preconcentrator
(Note: because of the unusual shape of the VLE curve this precon-
centration has little effect on the reflux ratio RD' Then according
to Eq. (14-13) 0& is not changed much by preconcentration.
Although suggested by the heuristics RO is probably not useful for
ethanol-water and is not shown on the other flow sheets. Precon-
centration by RO can be useful for other systems.) Figure l4-7b
uses distillation followed by liquid phase adsorption. This requires
a drain step, evaporation of liquid, and then thermal regeneration.
Since energy requirements will be quite high the vapor phase
adsorption alternative shown in Figure 14-7c should be considered.
The adsorption can be done as TSA or PSA. TSA is done com-
mercially, but from heuristic 5S we should also consider PSA.
This will require operating the distillation column under pressure.
This may be advantageous since a smaller diameter distillation
column can be used although equilibrium is less favorable.

Other known processes can be substituted for breaking the

azeotropes. Examples are extraction and pervaporation. Heuristic
4D suggests pervaporation may not be advantageous; however,
since it is commercially available it should be considered.

The heuristics give several reasonable flowsheets. Genera-

tion of a final flowsheet will require further economic comparis-
ons. Note that in plants there are also solids and trace components
to be removed.
834

a
Water + EtOH

Azeotropic
Distillation

EtOH

Water

b EtOH

Drain
Adsorption
F

c EtO
nT
PSA or TSA
Regeneration

Go.

F
 835

14.6. EVOLUTIONARY METHODS

Evolutionary methods are used to systematically modify an existing flow sheet

to improve it. Although evolutionary methods do not guarentee that the
optimum flowsheet will be generated, they are excellent insurance against
catastrophic failure. Evolutionary methods are discussed by Liu (1987), Nath
and Motard (1981), Nishida et al (1981), Seader and Westerberg (1977), and
Stephanopoulos and Westerberg (1977) among others.

To use an evolutionary method one must first generate an initial

sequence. The initial flowsheet is important. The better this flowsheet is, the
easier it will be to apply evolutionary methods and find a result close to the
optimum. Generation of the initial flowsheet is usually done using heuristics
(Liu, 1987; Nath and Motard, 1981; Nishida et ai, 1981; and Seader and
Westerberg, 1977). This is one reason that heuristics have a prominent place in
this chapter. An alternative to using heuristics is to use the solution to a simi-
liar problem solved in the past. Conversion of flow sheets for similar liquid-
liquid extraction problems is discussed by Blumberg (1988).

The second step is to develop the evolutionary rules. Local evolutionary

methods look only at separation problems which are neighbors (Liu, 1987;
Seader and Westerberg, 1977; Stephanopoulos and Westerberg, 1976). The
local methods have the advantage of reducing the number of calculations which
have to be done, but at the price of limiting the search field. The two evolu-
tionary rules used for the local evolutionary method are (Seader and Wester-
berg, 1977):

1. Interchange the relative position of two neighboring subprob-

lems.

2. Substitute an alternate separation method.

Figure 14-7. Possible ftowsheets for Example 14-3. a Distillation followed

byazeotropic distillation with optional RD. b. Distillation fol-
lowed by liquid phase adsorption. c. Distillation followed by
vapor phase adsorption (TSA or PSA).
836

Application of the rules requires a strategy. The strategy developed by

Seader and Westerbel'g (1977) consisted of the following:

1. Generate each possible neighbor using the two rules. This will
result in a lot of possibilities.

2. Keep a neighbor as a candidate if the heuristic used to choose

the original separation did not clearly discriminate between the two
choices. or ifa competing heuristic would have selected the neigh-
bor.

3. Evaluate the neighboring flowsheets until a locally "best"

flowsheet is found

4. Consider neighbors which are plausible. but were not included

in strategy item 2.

It should be evident that this method can involve a lot of work. Thus, one
needs to use short-cut estimation methods in the evaluation of flowsheets.

A global evolutionary method was developed by Nath and Motard (1981)

and was used by Liu (1987). Unfortunately, this development included only
distillation and extractive distillation; however, the method can probably be
expanded to include other separation processes. Nath and Motard (1981) used
heuristics in the following order:

1 (4G). Favor the smallest product set

2 (IS). Favor distillation.
3 (Inverse of 2C). Easiest separation first.
4 (2D). Do not use a mass separating agent to recover a mass
separating agent
5. (IS). Separations with a < aunn are unacceptable.
6 (4D). Set operating pressure close to ambient.
7. Use user set split fractions.
=
8. Use lID 1.3 (L/D)min.
 837

These heuristics are used in the order listed to generate the initial fiowsheet.
The following evolutionary rules are then applied:

1. Challenge Heuristic 1 (smallest product set).

To obtain the smallest product set one may have to use a mass
separating agent Relaxing heuristic 1 may result in a fiowsheet
without the mass separating agent, and this may be a better
fiowsheet

2. Challenge neighboring separations if the difficulty of separation

is about equal or if refrigeration is required for condensation of
distillate.
Nath and Motard (1981) define a Coefficient of Difficulty of
Separation to aid in the application of this rule.

3. Challenge heuristic 2 (favor distillation).

This rule allows one to consider mass separating agent processes.

4. Examine neighbors to see if removal of a mass separating agent

should be delayed.
Heuristic rule 3 (easiest separation) will almost always force the
removal of the mass separating agent next If the mass separating
agent can be used in more than one separation it is advantageous to
leave it in the process longer.

5. Challenge heuristic 3 (easiest separation) if the following

separation is much more difficult than the one being looked at.
Removing a component may make the next separation very
difficult because of composition effects on the equilibrium parame-
ters.

Liu (1987) notes that rules 1,3, and 4 may be very valuable in developing less
expensive designs.
These evolutionary rules can be applied in a variety of ways. The stra-
tegy chosen by Nath and Motard (1981) makes this a global method. Evolu-
tionary rule 1 is done first. Then rules 2 to 5 are treated equally. When a
838

change proves to be advantageous, the downstream process is eliminated and

regenerated. All parts of the process are checked with rules 2 to 5. Nath and
Motard (1981) give examples, but they include only distillation and extractive
distillation.

The evolutionary approach or some other systematic approach to improv-

ing ftowsheets are very useful in developing better ftowsheets. Unfortunately,
the evolutionary approaches need to be developed and validated for separations
other than distillation and extractive distillation. This requires more research.

14.7. SUMMARY - OBJECTIVES

At the end of this chapter you should be able to meet the following objectives.

1. Discuss the type of separation problems which occur.

2. Develop tables of properties to compare separation alternatives for a given

problem.

3. Use energy considerations to compare separation alternatives.

4. Discuss and use different strategies for developing ftowsheets.

5. Use heuristics to develop an initial flowsheet Discuss the rationale for each
heuristic.

6. Discuss and use evolutionary methods.

7. List and discuss the areas which need further research in selecting a separa-
tion method.

REFERENCES

Bonmati, R. t G. Chapelet-Letourneax, and G. Guiochon, "Gas chromatography:

A new industrial process of separation. Application to essential oils", Separ.
Sci. Technol.,19, 113 (1984).
 839

Black, C., "Distillation modelling of ethanol recovery and dehydration

processes for ethanol and gasohol," Chern. Eng. Prog .. 76 (9), 78 (1980).

Blumberg, R., Liquid-Liquid Extraction, Academic Press, London, 1988.

Bravo, J.L., J.R. Fair, J.L. Humphrey, CL. Martin, A.E. Seibert and S. Joshi,
Fluid Mixture Separation Technologies for Cost Reduction and Process
Improvement, Noyes Publications, Park Ridge, N.J., 1986.

Douglas, J.M., Conceputal Design of Chemical Processes, McGraw-Hill, New

York, 1988.

Evans, FL., Jr., Equipment Design Handbook for Refineries and Chemical
Plants, 2nd ed., Gulf Pub. Co., Houston, TX, 1980.

Fair, J.R., "Energy-efficient separation process design", in Y.A. Liu, H.A.

McGee and W.R. Epperly (Eds.), Recent Developments in Chemical Process
and Plant Design, Wiley-Interscience, New York, 1987, Chapt. 3.

Garg, D.R. and J.P. Ausikaitis, "Molecular sieve dehydration cycle for high
water content streams," Chern. Eng. Prog .• 79 (4) 60 (1983).

Guiochon, G. and H. Colin, "Theoretical concepts and optimization in prepara-

tive scale liquid chromatography", Chromatography Forum. 21, (Sept.-Oct,
1986).

Hendry, J.E., D.P. Rudd, and J.D. Seader, "Synthesis in the design of chemical
processes," AIChE Journal. 19, 1 (1973).

Ho, F.-G. and G.E. Keller, II, "Process integration", in Y.A. Liu, H.A. McGee,
and W.R. Epperly (Eds.), Recent Developments in Chemical Process and Plant
Design, Wiley-Interscience, New York, 1987, Chapt 4.

Jacobs, L.J., Jr., and W.R. Penney, "Phase Segregation", in R.W. Rousseau
 840

(Ed.), Handbook of Separation Process Technology, Wiley-Interscience, New

York, 1987, Chapt 3.

Keller, G.B., n, Adsorption, Gas Absorption, and Liquid-Liquid Extraction:

Selecting a Process and Conserving Energy, Manual 9 in Industrial Energy-
Conservation, The MIT Press, Cambridge, MA, 1982.

Kelly, R.M., "General Processing Considerations:' in R.W. Rousseau (Ed),

Handbook of Separation Process Technology, Wiley-Interscience, New York,
1987, Chapt 4.

King, CJ., Separation Processes, 2nd ed., McGraw-Hill, New York, 1980.

Kopaciewicz, W. and F.E. Regnier, "A system for coupled multiple-column

separation of proteins", Anal. Biochem.129, 472 (1983).

Ladisch, M.R., M. Voloch, J. Hong, P. Bienkowski and G.T. Tsao "Commer-

cial adsorber for dehydrating ethanol vapors," Ind. Eng. Chem. Process Des.
Develop., 23, 437 (1984).

Leeper, S. and P.C. Wankat, "Gasohol production by extraction of ethanol from

water using gasoline as solvent," Ind. Eng. Chem. Process Des. Develop., 21,
331 (1982).

Liu, Y.A., "Process Synthesis: Some simple and practical developments," in

Y.A. Liu, H.A. McGee, Jr., W.R. Epperly (Eds.), Recent Developments in
Chemical Process and Plant Design, Wiley-Interscience, New York, 1987,
Chapt6.

Ludwig, E.E., Applied Process Design/or Chemical and Petrochemical Plants,

2nd ed., Gulk Pub. Co., Houston, TX, 1979.

Mazsaroff, 1. and F.B. Regnier, "An economic analysis of performance in

preparative chromatography ofproteins",l. Liqd. Chromatogr. 9,2563 (1986).
 841

McDonald, P.D. and B.A. Bidlingmeyer, "Strategies for successful preparative

liquid chromatography," in B.A. Bidlingmeyer (Ed), Preparative Liquid
Chromatography, Elsevier, Amsterdam, 1987.

Nath, R. and R.L. Mothard, "Evolutionary synthesis of separation processes",

AIChE Journal, 27, 578 (1981).

Nishida. N., G. Stephanopoulos and A.W. Westerberg, "A review of process

synthesis," AIChE Journal, 27, 321 (1981).

Null, H.R., "Energy 'eCOnomy in separation processes," Chem. Engr. Prog., 76

(8),42 (Aug. 1980).

Null, H.R. "Selection of a separation process," in R.W. Rousseau (Ed.)., Hand-

book of Separation Process Technology, Wiley-Interscience, New York, 1987,
Chapt 22.

Perry, R.H. and D.W. Green (Eds.), Chemical Engineers' Handbook, 6th ed.,
McGraw-Hill, New York, 1984.

Robinson, C.S. and E.R. Gilliland, Elements of Fractional Distillation, 4th ed.,
McGraw-Hill, New York, 1950.

Roddy, J.W., "Distribution of ethanol-water mixtures to organic liquids," Ind.

Eng. Chem. Process Des. Develop., 20,104 (1981).

Rudd, D.F., G.J. Powers, and J.J. Siirola, Process Synthesis, Prentice-Hall,
Englewood Cliffs, NJ., 1973.

Ruthven, D.M., Principles of Adsorption and Adsorption Processes, Wiley,

New York, 1984.

Schweitzer, P.A. (Ed.), Handbook of Separation Techniques for Chemical

Engineers, McGraw-Hill, New York, 1979.
842

Seader, J.D., "Distillation," in RH. Perry and D.W. Green (Eds.), Perry's
Chemical Engineers' Handbook, 6th ed., McGraw-Hill, New York, 1984, Sec-
tion 13.

Seader, J.D. and AW. Westerberg, "A combined heuristic and evolutionary
strategy for synthesis of simple separation sequences", AIChE Journal, 23, 951
(1977).

Sherwood, T.K., RL. Pigford and C.R. Wilke, Mass Transfer, McGraw-Hill,
New York, 1975, Chapt. 1.

Stephanopoulos, G. and A W. Westerberg, "Studies in process synthesis - II",

Chem. Eng. Sci., 31, 195 (1976).

Tedder, D.W. and D.F. Rudd, "Parametric studies in industrial distillation. Part
I. Design Comparisons", AIChE Journal, 24,303 (1978).

Thompson, RW. and C.J. King, "Systematic synthesis of separation schemes,"

AIChE Journal, 18,941, (1972).

Van Winkle, M., Distillation, McGraw-Hill, New York. 1967.

Wankat, P.C., Large-Scale Adsorption and Chromatography, CRC Press, Boca

Raton, FL, 1986.

Wankat, P.C. Equilibrium-Staged Separations, Elsevier, New Yorle, 1988.

Westerberg, AW., "Design research. Both theory and strategy", Chem. Eng.
Educ., 16, 12 (1982) and 16,62 (1982).

HOMEWORK

A Discussion Problems

A I.Define the following terms

a. Minimum work
 843

b. Equivalent work

c. Heuristic

d. Evolutionary method

e. Local evolution

f. Global evolution

g. Separation factor

h. Diffusional separations

i. Mechanical separations

j. Hybrid process

k. Neighbor
j. Top-down!Bottom-up

A2. When one is developing new separation processes, distillation is often con-
sidered "the enemy". Explain why.
A3.Heuristics can be misleading. For example, heuristic 4D could be mislead-
ing when high pressure gas streams are processed. Explain why one prob-
ably wants to do the separation at high pressure.

A4.Based on an equivalent work analysis crystallization is a very attractive

separation process. Why isn't crystallization used more often?

A5.Explain the differences and similarities between local and global evolution.

A6.Expand on the idea of using flowsheets for similar problems to generate an

initial flowsheet How similar must the problems be? What are some
sources of flowsheets?

A7.0n Figure 14-1 "Water from sea water", vitamin B-12, and "gold from sea
water" all falloff the empirical straight line. Hypothesize why each of
these three is different
844

A8.Why is adsorption favored compared to distillation when a small amount of

nonvolatile material is to be removed?

A9.List and discuss the areas requiring additional research.

AlO. What are "optimistic conditions" for comparison of the following separa-
tion processes?

a. zone melting

b. UP
c. ED
d. LC for protein purification

All. Extend heuristic 2S to the following separations:

a. crystallization

b. adsorption

c. membrane separations

A12. Westerberg (1982) suggests that theories evolve from simple to complex.
Explain how one can do this for

a. Chromatography

b. Reverse osmosis

c. Solution crystallization

A13. Compare depth-first with breadth-first approaches. Why should both

depth and breadth be included? List and discuss the areas requiring addi-
tional research.

A14. Develop your key relations chart for this chapter.

B. Generation of Alternatives

B1. Hybrid processs allow the designer to stretch his/her imagination. Mem-
 845

brane processes in particular seem to be very useful in hybrid processes.

Develop two novel hybrid processes involving membranes.

B2. Suppose you work for a company whose business is selling processes.
Brainstorm and then pare down a list of separations which should be
explored further by R&D. Explain your final list of 5 items.

B3. Generate possible separations to recover acetic acid from a dilute aqueous
solution.

B4. Use heuristics to generate preliminary separation flowsheets for Problem

14-B3.

C. Derivations

C1.Develop a local evolutionary process which uses Nath and Motard's (1981)
evolutionary rules.

C2.Derive Eq. (14-19).

C3.Derive Eq. (14-14).

C4.Many of the heuristics in this chapter were developed from material in pre-
vious chapters. Generate two additional S heuristics based on information
in previous chapters.

D. Problems

D1.We wish to separate a gas mixture of ethylbenzene and styrene at 110°C.

Determine Wmin,T for mole fractions of ethylbenzene of 0,10,20, ... ,100 in
the feed. Ratio Pp/PF = 1.5.

D2.Determine the equivalent work requirement for a melt crystallization sys-

tem purifying p-xylene from a mixture of para and meta xylenes. The feed
is 40 mole % p-xylene, crystal product is 99.9 mole % p-xylene. Operation
of the crystallizer is very slightly above the eutectic temperature (221 K).
846

Assume the melt product is essentially at the eutectic composition (13 mole
% p-xylene). Pressure is 101.3 kPa Determine the ideal equivalent work
(ep = eR = 1) for melt crystallization if a lOoC approach temperature is used
in the melter and the crystallizer. The ambient temperature To is 18°C.
Melter reflux ratio RM = 0.5. Use a feed rate of 100 kglhr as a basis.

Xylene Data (King, 1980, p. 740): MW = 106.16, Ap = 340 kJ/kg, AM =

343 kJ/kg, Tboiling,p = 411.8 K., Tboiling,m = 412.6 K, Tr,M = 225.4 K, Tr,p =
286.6 K,

(perry and Green, 1984, p. 3-123) Ar,M =26.045 calIg, Ar,p =38.526 cal/g
D3.Find the equivalent work in calI(kg product) required for the RO system in
=
Example 12-2b for M 1.2. Watch your units. Use a water density of 1
kg/liter.

F. Problems Requiring Other Resources

Fl. Different chromatographic methods used for protein purification have dif-
ferent solvent requirements. Read Kopaciewicz and Regnier (1983) and
develop heurispcs for sequencing chromatography columns for protein
purification.

F2. Extend Nath and Motard's (1981) "CoefficienL of Difficulty of Separation"

to center-feed melt crystallization of solid solutions.