6

Process Design and Optimization

Robert Rautenbach

INTRODUCTION-MASS TRANSPORT AT THE MEMBRANE SURFACE

The Local Mass Transport

It is understood that the economical success of any membrane process de-
pends primarily on the quality of the membrane, specifically on flux, selectivity
and service lifetime. Consideration of only the transport mechanisms in mem-
branes, however, will in general, lead to an overestimation of the specific per-
meation rates in membrane processes. Formation of a concentration boundary
layer in front of the membrane surface or within the porous support structure
reduces the permeation rate and, in most cases, the product quality as well.
For reverse osmosis, Figure 6.1 shows how a concentration boundary layer
(concentration polarization) forms as a result of membrane selectivity. At steady
state conditions, the retained components must be transported back into the
bulk of the liquid. As laminar flow is present near the membrane surface, this
backflow is of diffusive nature, i.e., is based on a concentration gradient. At
steady state conditions, the concentration profile is calculated from a mass bal-
ance as
‘I,/ k

(1) ‘2-‘3 = e
9
c1-c3

where k is the mass transfer coefficient which can be assumed, at least with a
good approximation, to be independent of the permeate flux V,. For this rea-
son, the analogy between heat and mass transfer is valid and k can be calculated
using the well-known heat transfer equations. This has been tested experimen-
tally for a number of cases.’

349
350 Handbook of Industrial Membrane Technology

a
n-r3 mass transport by
convection
permeate

diffusive mass transport

c3 F membrane

Figure 6.1: Concentration polarization.

Figures 6.2 and 6.3 indicate the order of magnitude of concentration polar-
ization for laminar and turbulent flows through tubular membranes. The dia-
grams illustrate the dependence of the concentration boundary layer on flow
conditions along the membrane (Re) and on the permeation flux (Pe,).

II

=-
v,.d
D i

100 1000
Reynolds number Re
Figure 6.2: Concentration polarization in laminar flow.
 Process Design and Optimization 351

1
“I
OS VI7I.d ! !!
- Pex-fj-
- Sh= ~.02JRe7/8!k”L
400 .\

Reynolds number Re

Figure 6.3: Concentration polarization in turbulent flow.

According to Figures 6.2 and 6.3, turbulent flow is more advantageous. Fur-
thermore, these diagrams explain why, in the case of gas permeation, in contrast
to separation processes in the liquid state, the resistance due to a boundary layer
can be neglected: the diffusion coefficients of gases, for example 02/Nz are lo4
times larger than those of dissolved components in liquids such as NaCI/l$O
which, in combination with the permeation rates reported in literature, leads
to Pe numbers of less than unity.

Influence of the Asymmetric Structure of Membranes

While it is possible to enhance mass transfer in the boundary layer by im-
proving the flow conditions, this is impossible if a concentration profile exists
within the porous support structure.
For reverse osmosis, Figure 6.4 shows the predicted concentration profile
in the porous structure of an asymmetric membrane. According to Figure 6.4,
in liquid systems, the concentration c3, which affects permeate flux and product
quality, cannot be influenced significantly by the concentration of the liquid on
the product side. Conditions are nearly always favorable for the permeate to flow
unhindered. Even in the extreme case of local concentration on the permeate
side being equal to that of the liquid on the high pressure side of the membrane
(c, = c,), the error made in calculating the solvent flux V, and the salt flux J,
is only 2% or 5% respectively, if the effect of cs is neglected.
Again, the situation is different in the case of gas permeation. Here, the lo-
cal concentration affects the (local) permeate fluxes of the mixture components
and, for this reason, the type of flow for gas permeation on the permeate side is
352 Handbook of industrial Membrane Technology

much more important than for liquid separation processes such as reverse os-
mosis. Figure 6.5 shows theoretical results of Blaisdell and Kammermeyer.2
Mol fraction of oxygen is plotted against the yield during fractionation of air
for different flow configurations and cases of mixing. The diagram demonstrates
that (1) countercurrent flow provides the best results and (2) marked differences
exist between the perfect mixing and plug flow.

feed ~membrone-----_tcpermeate

porous
subslucture

membrane thickness

Figure 6.4: Concentration profile in porous substructure of asymmetric mem-

branes.

0.29, t 1 1 I
I

0.2A - complete mixing

stage-cut 0

Figure 6.5: Effect of flow configuration in gas permeation.

 Process Design and Optimization 353

Change of Conditions Along the Membrane

As a consequence of the permeate flux, the pressure gradient as well as the
mean velocity and concentration will vary along the membrane surface-in prin-
ciple at the feed side as well as at the product side. Depending on module de-
sign, this has to be taken into account for one or both sides. If, for example, the
membrane is of tubular design with feed-flow inside, the following has to be ex-
pected along the tube-axis:

(1) Decrease of the transmembrane pressure-difference as a conse-

quence of the drop in pressure due to friction ( + decrease of per-
meate flux).
(2) Increase of the mean concentration of the feed solution ( + in-
crease of osmotic pressure + decrease of permeate flux and in-
crease of salt flux).
(3) Decrease of the mean velocity ( + increase of concentration polar-
ization + decrease of permeate flux and increase of salt flux).

For these reasons, the overall permeation rate and the product quality of a
membrane channel have to be calculated by integrating over the length elements
employing, in addition to the equations discussed above,

(1) the mass balance

(2) the material balance
(3) the energy balance.

The equations necessary for the calculation of tubular RO-membrane chan-

nels (flow of permeate unhindered, no influence of the porous support-layer) are:

mass transfer at the membrane surface

concentration polarisation e I evw’,,

I 3

equation for mass transfer Sh = y = ShlRe, Sc)

mass transfer in the membrane

permeate flux V, = A,IAp-blcz-c,)) ; AP ‘P, -P,,,
salt flux 3s = WC,-c,)

salt concentration C) = Ir
VW
balance for a length element
LAX
mass balance v,*,., = K..- - (j VW*,

material balance (salt) ‘jr.“., CI,~.( =TX,n.Cl,n- 9 .V,,. C,,,

x,d LAX.‘&
energy balance(p=constl pm+,= &[p”+~pT,,~(1-5 l-~.Pn

- + P %:,,.I - APV
364 Handbook of industrial Membrane Technology

This set of equations has to be solved numerically.

Figure 6.6 discusses the results of such a numerical calculation of a long tu-
bular RO-membrane channel for seawater desalination. It clearly demonstrates
the decrease of permeate (mass) flux J, and, at the same time, the increase of
local and mean product concentration cs and &.

0
‘0 20 LO 60 80 100
Number of tubes

Figure 6.6: Calculation of a tubular module.

MODULE CONCEPTS AND DESIGN

The module, i.e., the industrial configuration of membranes, has to meet a

variety of requirements which are sometimes even contradictory. Points of major
importance are: (1) flow conditions along the membranes, (2) ratio of membrane
area to pressurized vessel volume, (3) price of module and (4) the possibility of
cleaning the membrane.
Depending on the process application, one or the other of these require-
ments is of primary importance and, for this reason, a number of different mod-
ules have been designed. The most important designs are the plate and frame,
the spiral wound and the hollow fiber module. Optimization procedure and
some of its results will be discussed for only one module configuration, the hol-
low fiber module-and for only one application-RD.

The Hollow Fiber Module

Hollow fiber modules contain very fine fibers forming asymmetric or sym-
metric membranes and capable of withstanding pressure differences (high-pres-
 Process Design and Optimization 357

50- 10

bar l$ $!

-6 0 20 LO 60 60 100 cm 120
fibre length L

Figure 6.6: Permeation behaviour of hollow fibers.

fibre length L

Figure 6.9: Concentration profiles in hollow fibers.

358 Handbook of Industrial Membrane Technology

In theory, larger fiber diameters should increase the flux, i.e., the perform-
ance of a single fiber. However, at the same time, the membrane area per unit
volume decreases. For some simplifying assumptions, it can be easily discussed
how fiber diameter and fiber length have to be chosen in order to maximize the
“yield” per unit volume of the bundle.
Neglecting (1) the radial pressure losses in the feed and (2) the influence of
osmotic pressure on flux, the specific yield of the fiber bundle is given by:

$2$&d&.
v
4(1-E)
d
3 3

(8) tanh (HLI

l+Hsls*tanh (HL)

For a fixed ratio of 2 = 2, Equation 8 is discussed in Figure 6.10.

I

I I I I I 1
0 20 LO 60 80 plm 100
fibre diameter di

Figure 6.10: Optimum design of hollow fibers.

According to Figure 6.10, an optimal fiber length exists for a given fiber
diameter. Figure 6.8 clearly indicates that hollow fiber modules for reverse os-
mosis should be made as short as possible. It should be noticed that a simple
correlation between yield of the single fiber and yield of the bundle can be de-
rived in any case only if the concentration and the pressure in the shell (outside
of the bundle) are considered to be constant! In principle, the concentration of
the feed increases between entrance and outlet while the pressure of the feed
 Process Design and Optimization 359

decreases. A numerical solution taking this into account can be achieved if the
bundle is considered as a continuum containing sinks (Figure 6.11). In this case,
the following equations have to be solved:

mass balance (pL = const):

(9) = -& (r V, (t-11 =

material balance:

energy balance:

dp, (r) U-EIZ ; (r)

(11)
-a-i=-= - K0 nFtv] Ej 1

K, has to be varied according to the fiber arrangement. K, = 30

is valid for fibers in parallel.

concentrate

fiber bundle

on

permeate 1
Q39C30 1

feed
solution
Qla, ClorlPISX

Figure 6.11: Balance element in the fiber bundle.

In the case of reverse osmosis, the relative flow configuration does not af-
fect the performance to any large extent. As already indicated, the situation is
quite different for gas permeation. While countercurrent flow improves the sepa-
ration efficiency of hollow fiber modules in reverse osmosis only slightly, as far
 360 Handbook of Industrial Membrane Technology

as gas permeation is concerned, countercurrent flow produces the best results. A

consequent application of the countercurrent principle led to the concept of
Hwang,4 i.e., the membrane column (Figure 6.12). This column consists of a rec-
tification and a stripping section and permits binary mixture to be fractionated
into its components. For a given geometry of the membrane fractionating col-
umn (length of each section, membrane area per unit length), the product qual-
ity is determined by the flow rate of the gas being recirculated.

product 1
(permeate I
4

i
I enriching section
I
I
I
feed

I
I stripping section

I
I
I

product 2
(concentrate 1

Figure 6.12: Membrane column for gas separation.

The modules employed in a membrane fractionating column must be dif-

ferent from those used in cascades: in hollow fiber modules, for example, the
fibers cannot be U-shaped and must be open at both ends. In Reference 5, a
module is described based on this principle.

CASCADES

There are cases where it is impossible to fractionate a given mixture in one

 Process Design and Optimization 361

stage to the desired degree of purity; the separation process will consist of sev-
eral stages. These stages can be realized in several ways, for example, by employ-
ing the countercurrent flow principle as much as possible within one unit like in
packed towers for absorption or rectification or by combining definite separa-
tion units in the form of a cascade. The best known examples of this case are
fractionating towers employing trays.
In fractionating towers, the trays are still incorporated in one vessel, i.e.,
part of one unit. In other cases, for example, cascades of hydrocyclones, gas
centrifuges and-last not least-membrane modules, the unit itself will be the stage
of the cascade.
In membrane technology, processes consisting of more than one stage are
known in gas permeation and for the production of boiler feed water from sea-
water. Furthermore, cascades will be necessary if organic mixtures are to be sep-
arated by membrane processes only.
The best known example of a membrane cascade consisting of a high num-
ber of stages is the Oak-Ridge plant for enriching U-235 in the gaseous phase
(employing porous membranes).

Definitions
Separation Unit. The elements of a separation process. Separation units of
a separation process could be, for example, a gas centrifuge, a membrane mod-
ule, a tray of distillation column or the evaporator of a multiple effect plant. It
should be kept in mind, however, that in plate-and-frame modules as well as in
modules of the spiral-wound-type usually every block or pressure-vessel contains
more than one separation unit!
Stage. The element of a cascade will be named “stage” and the stage itself
might contain several separation units connected in series and/or in parallel. Sep-
aration units are considered part of a stage as long as they are connected only at
the feed and/or retentate side!
The elements of a stage can be arranged according to Figure 6.13 (this is
sensible with respect to mass transfer in the boundary layer because of the de-
creasing volume flow at the feed side along the “axis” of a stage), in a “tapered”
fashion or in a “squared-off” fashion.

Figure 6.13: “Tapered” module arrangement.

362 Handbook of Industrial Membrane Technology

P(Product1 , yp

Rnlxn
.

,,bed& RN ‘,“N

(Retentatel

Figure 6.14: Cascade without reflux.

Cascade. A combination of stages where the permeate of a stage will be

the feed of the next stage.
Yield. Yield is defined as the ratio of a certain (key) component in the
product and in the feed:

(12)

Selectivity. Selectivity in general is defined as the quotient of ratios of con-

veniently chosen concentration measures. Convenient measures for a concentra-
tion are, for example, the partial pressure pi and the molar fraction xi or a mass
fraction Wi. Accordingly, the selectivity is defined as:

PA’PB 1 Prod = ‘A”BI Prod

(13) s, 5 /-
PA PBI Feed ‘A’“B 1 Feed

or

WA”JB 1 Prod
(14) s2 = -i--7?-
A B1 Feed
 Process Design and Optimization 363

For the permeation of ideal gases, where

.
n.
1 = Qi (PI xi - P2 Yi )

describes the transport mechanism (in sorption-diffusion) membranes, Equation

13 is reduced to

(15)
s=-=;
QA
,O<ol<l
QB

(in cases, where psvi << pr Xi and, consequently, ni - Qiprxi is valid).

Another entity which is often employed because of its simplicity is defined
as:

(16). x =-'AI Permeate = yA

x
xAj Retentate A
or

xn =
YAPI module p ermeate, stage n =-Yn
'AR1 module retentate, stage n xn

if characterizing the selectivity of the entire module.

“Cut-Rate” (Split-Factor). The cut-rate, sometimes called “split-factor”,
is defined as the ratio of permeate flow and feed flow. There will be, however,
a difference between the cut-rate defined for molar flows and for mass flow.

(17)

or

(18)

Operating Line and “Equilibrium” Curve. Both terms are of importance for
the graphical solution of a separation problem, i.e., for the graphical determina-
tion of the number of stages of a cascade. This method has been developed for
the design of distillation columns by MacCabe and Thiele and should be well
known. For all cases, the operating line represents the mass and material bal-
ances. In distillation, the equilibrium curve represents the thermodynamical va-
por/liquid equilibrium. For an ideal binary system, the equilibrium curve can be
calculated from Raoult’s law and the saturation-pressure curves of the pure com-
ponents of the mixture. In all other cases, however, for example, for all mem-
brane processes, the equilibrium curve does not represent a thermodynamical
equilibrium at all but will represent the separation characteristics of the module
or that of the stage.
364 Handbook of Industrial Membrane Technology

CascadesWithout Reflux
Such cascades are only sensible in cases where the retentate is practically
worthless. Norsk-Hydro, for example, is operating an electrolysis-heavy-water-
production-plant where the deuterium is separated in a cascade without reflux
(at least in the lower part of the cascade). Figure 6.14 illustrates the principle
of a cascade without reflux.

Reflux Cascades
The inevitable losses of product with the retentate of cascades without re-
flux can be avoided by recycling the retentate according to Figure 6.15. This ad-
vantage, however, has to be weighed against the higher capital costs of reflux
cascades. Only a detailed calculation of the specific separating costs will lead to
a sound decision on whether a separation process should operate with or without
reflux. In principle, there are four possible modes of operation:

(I) Constant reflux R/P

(2) Constant cut rate and variable reflux
(3) Variable cut rate and variable reflux
(4) “Ideal” cascade operation (x, _ t,R = x, + 1,p).

Here, only the first two cases will be discussed because with modern computers
at hand, for these two cases only a graphical solution will be of value.
Constant Reflux. In general, the operating line of a cascade follows from a
mass/molar balance and a material balance between the top and a certain stage n
of the cascade. For constant reflux R = const resp. R/P, = v = const.

the molar balance:

P - Rn - p, = 0
n+l
and the material balance:

(20) P
yn+l -
Rnxn P, y, = g
n+l
result in:
V 1
(21) Yn+l = v+l % G-i y1

The operating line is a straight line with the slope --$

If the “equilibrium curve” is known, the number of stages necessary to

achieve a certain product quality can be easily found according to Figure 6.16
by a step-by-step graphical procedure.
Constant Cut Rate, Variable Reflux. When the molar balance:

(22) P - Rn - P, = 0
n+l
is developed for several stages, a pattern can be noticed if the (constant) cut
rate is introduced:
 Process Design and Optimization 365

Rn,xn

RN-l

Pm,Ym
Rm-loxm-l_, -5--

--

RM-l ‘“M-l

Figure 6.15: Refl ux cascade.

Figure 6.16: Graphical evaluation of the number of stages for a reflux cascade
(enriching section), operated with constant reflux.
366 Handbook of industrial Membrane Technology

(23) = P, + R, = P, (l+y)
p2
Here, y = R/P is used instead of 0 = P/F. 8 and y are related by the molar bal-
ance of a stage: y = (1 - 0)/e.

= P, + R2 = P+l+Y (l+y))
p3

pn = P, Cl+ Y +y2 + . . . Y”“)

(24) 1 -y”
‘n =pl -1 -y

With the material balance:

Rn pl Yl
(25) y n+l =Gxn +p,,1

the operating line follows:

n+l
(26)
_ Y -Y X 1-Y
yn+ 1 n+l n +n+r Yl
1-Y 1 -Y

Quite obviously, the slope of the operating line differs from stage to stage! The
number of stages follows according to Figure 6.17. It should be noticed that all
operating lineswill intersect at y1 = x1.

Figure 6.17: Graphical evaluation of the number of stages for a reflux cascade
(enriching section), operated with constant cut-rate, variable reflux.

Once the necessary number of stages is known, the product rate PI/F can be
calculated according to:

(27)
 Process Design and Optimization 367

The “Equilibrium Curve”

As mentioned above, the “equilibrium curve” in the case of membrane cas-
cades is not at all a thermodynamic equilibrium but will represent the separation
characteristics of the module or of the stage. This separation characteristic is in
general a function of:

(1) the selectivity of the membrane

(2) the fluid dynamics in the module
(3) the driving force
(4) the concentration level (because the membrane selectivity is or
might be concentration dependent)
(5) the flow pattern in the module, for example, co- or countercurrent
flow.

With respect to flow pattern, five different patterns are possible:

(1) ~~z;r?$e mixing at both sides of the membrane (Weller Steiner

(2) Cross-plug flow, i.e., plug flow at the feed side and permeate flow
orthogonal to the membrane without mixing (Weller Steiner Case
2 6 Naylor-Baker’)’
(3) Parallel plug flow, cocurrent flow (Blaisdell-Case’)
(4) Parallel plug flow, countercurrent flow
(5) Partial (incomplete) mixing of feed and permeate (Breuer Case’ ).

Here, only the first case will be discussed. A detailed discussion of the other
cases can be found elsewhere.lO~‘l
Complete Mixing of Feed and Permeate. In this case (Figure 6.18), the ma-
terial balances for a binary mixture are:

P* Y * = Q, (P, x, - ?2 y*)

(29) P,(l- Y n) = Q, (p*(l-x, > - p2(1- Yn))

Combining of both equations and introducing 6 = ps/pr and o = QB/QA leads to:

n
(30)
;‘-I- = L
-‘pn o ( Ynx
-.
n
-1 -(l-
yn c^yJ6

As Equation 30 indicates, the equilibrium curve in this case is influenced by the

ideal separation factor So and the pressure ratio r6.
For large pressure ratios, the equilibrium curve simplifies to:

(31)
).n
 368 Handbook of Industrial Membrane Technology

T pnl Yn

Figure 6.18: Complete mixing.

a = 0.2
---
--- a =0.05 I
0 0.5 xn 1.0

Figure 6.19: Separation characteristics (complete mixing).

Membrane Column
The membrane column can be an alternative to membrane cascades.12-14
The membrane column consequently utilizes the countercurrent flow principle
(Figure 6.20) and, for this reason, is restricted to cases where concentration pol-
arization is negligible. (It has been suggested that the membrane-column concept
be applied to the separation of liquid mixtures. However, this concept must fail
here-at least for asymmetric membranes because of the concentration polariza-
tion within the support layer.) Membrane columns utilize the maximal driving
forces because of the countercurrent flow at both sides of the membrane. The
preferably permeating component is depleted on the high-pressure-side in the
direction of the flow. The preferably permeating component can be withdrawn
at the top of the column, the slower permeating component of the mixture at
the bottom. At present, all modules for membrane columns described in the lit-
erature employ hollow fiber membranes.
 Process Design and Optimization 369

Top product
(Permeate)
4

i Enriching section
I
I
I
Feed

I
!
I Stripping section
I
I
I

Bottom product
(Retentate)

Figure 6.20: Membrane column.

The design equations for a membrane column are:

(1) the transport equation for material transport in the membrane

(2) mass, material and energy balances (pressure losses), formulated
for the differential element of the membrane channel.

Material transport in the membrane:

.
(32) n. = Qi- (Xi PI - Yi P2)
1

Mass (molar) balance (M = constant) combined with Equation 32:

.
370 Handbook of Industrial Membrane Technology

Material balance:

d’x. * dxi -Ind RT Qi

(34) --$ + =+ x.
P2
--y.
1
dt PlDijd ==D nd2 C Pl l
ij

- xi (xi-yi !5 (,_pz,
i Qi Pl I
For a membrane column, it is essential that axial backmixing by diffusion
(dispersion) be negligible compared to convective transport. In this case, Equa-
tion 34 is reduced to:
d
(35) ~ (xi ~) = TTd Qi (xi P, - yi P2)

Combining Equation 35 with Equation 32 (transport in the membrane) leads to:

dx . 77 d QiPl
(36) -a-$ =
9

Energy-balance: For the calculation of pressure losses, it must be taken

into account-besides the compressibility of gases-that hollow fibers are elasti-
cally deformed (widened) by the higher pressure inside the fiber. According to
Hwang and Thorman,” in the case of a deforming fiber, the pressure losses can
be calculated by:

lbK, nh RI' 8 Re, 2

(37) 2 _ --
.I (t Re;d)-
nd
Pl

P”,,.d z PV d
with Re, =
: rl ; KeZ Iz2 2 rl

= 8 (1+0.75*Ree - 0.0107 Rei + 0.0125 Re:)

Kt
K2 = 1 l 0.056 Rew - O.OlSS*Re~

The set of Equations 32 f 37 has to be solved numerically. For the special case
of total reflux, however, an analytical solution for the concentration profile along
the membrane column can be derived if the pressure losses are neglected. In the
case of total reflux, for every section of the column (Figure 6.21):

(38) x. = y.
1 1

is valid.
 Process Design and Optimization 371

ki

Figure 6.21: Membrane column, material balance around a fiber end (total
reflux).

With the boundary conditions:

(39) z = 0: xi = xi a ; i = ka

Equation 36 can be integrated:

l+Oj/Qi Qj /Oi
h x, l-Oj/Qi

&t I
l-Xi, l-Qj/Ql

a lo. -AT Xi l-xi,

(40) z=- -1
Xia l-Xi
nd Qj p,(l-2) 1

-.- 1
l-Oj/Qi

Qj/Qi
l-Oj/Oi
Xia

+1_xla 1
( \-_;;
Xi

--Xia

)
l-Xi,

l-Xi

I-d,,Qi
-1 11

(41) ti = ti, (2)

The separation process is determined (among other parameters) by the re-

flux ratio. Similar to fractionating by distillation, the calculation with total re-
flux leads to the minimal necessary column length. For a given column length,
on the other hand, Equations 40 and 41 indicate the maximum possible purities
of top and bottom product. For these reasons, a discussion of Equations 40 and
41 is useful for a first estimate of the column length and the operating conditions.
Column length depends on the recycled molar flow rate f&. Figure 6.22
indicates that a specific molar flow rate (molar flow per fiber) as low as possible
should be chosen. The lower limit is set by the condition that diffusive axial
backmixing must be negligible with respect to the convective mass transport. On
the other hand, the volume-specific yield (yield per unit-volume of fiber bundle)
should be high.
 372 Handbook of industrial Membrane Technology

10
0.1 1.0 21) m
Fiber (Column) length L

Figure 6.22: Membrane column, influence of reflux (recycled molar flow rate)
on column length.

In general, the optimization of a membrane column must consider molar flow

rate &, reflux ratio and the pressure losses. The mutual dependencies, which
have been only shortly discussed here, are the reason why hollow fiber modules
for the membrane column are relatively long (L/d,-20,000) compared to hole
low fiber modules for the separation of liquid mixtures.
Figure 6.23 shows the concentration profile of a membrane column as cal-
culated by Equations 40 and 41 (total reflux). The concentration profile is very
similar to the profile calculated by a numerical solution of the set of Equations
37 for the separation of C0JN2 (Figure 6.24), taking into account the pressure
losses and a finite reflux ratio. Figures 6.23 and 6.24. clearly indicate that a
membrane column is especially effective in the stripping section. For the enrich-
ing section, especially for high concentrations of the preferably permeating com-
ponent, the necessary membrane area and the reflux ratio (pressure losses) are
increasing rapidly. It should be emphasized that a membrane column represents
a multi-stage process just as a reflux cascade. The question, whether a membrane
column or a reflux cascade is superior cannot be answered in general but must
be evaluated for every individual case. It seems doubtful, however, whether
multistage membrane processes will be often employed. According to our opin-
ion”, only one stage or, at the most, two-stage cascades seem to be competitive
compared to proven processes like pressure swing adsorption, absorption, or
low-temperature distillation.
 Process Design and Optimization 373

I I

0.8 ------ ---

Pl = 10 bar

0 0.2 04 0.6 0.8 m 1.0

Fiber (Co(umnI length L

Figure 6.23: Calculated concentration profile of a membrane column. total

reflux.

Feed = 52.6 % CO2

0 Experiment

- Theory ( Thorman 1
I I I I
0 1 2 3 4 m
Length

Figure 6.24: Concentration profile of a membrane column, comparison be-

tween theory and experiment.13
374 Handbook of Industrial Membrane Technology

PROCESSES

Any application of membrane processes will be ultimately determined with

respect to economic reasons, when compared to proven separation processes
such as distillation. Strictly speaking, membrane processes will not replace con-
ventional processes but, in the main, a combination of membrane processesand
conventional processes will be the optimal solution. This trend can be noticed at
least partly in the following examples.

Seawater Desalination by RO

At present, nearly all seawater desalination plants are distillation plants.

Since low pressure steam can be utilized for heating, all major plants are com-
bined with power plants-the heat input section of the desalination units replac-
ing the condenser in the Rankine cycle. By this combination, the primary en-
ergy consumption of seawater desalination can be kept Iowl which is one of
the reasons that thermal processes will always have their place in seawater de-
salination although the relatively rigid coupling of power and water production
is sometimes disadvantageous.
Modern distillation plants have unit capacities of about 25,000 t/d. The
energy consumption is approximately 95 KWh/to distillate in the form of sat-
urated steam of a pressure of about 2.2 bar and 4 to 5 KWh/to distillate in form
of power for the pump drives. All distillation processes produce a distillate con-
taining only 20 to 30 ppm salts (total dissolved solids). Reverse osmosis (RO)
will be the alternative to distillation. The modules which are successfully em-
ployed are of the “spiral wound” or the “hollow fiber” configuration. Under
favorable conditions, the economic service lifetime of the modules resp. the
membranes is about 5 years and warranties for such a figure are given by manu-
facturers.
RO plants are, however, much more sensitive to insufficient feedwater pre-
treatment than thermal plants. For this reason, it is accepted technology to use
well water from wells drilled at the seashore rather than surface water as in
thermal desalination.
As an example of the present state of the art in RO-seawater desalination,
some data of the Malta facility shall be discussed.” The entire plant, consisting
of 10 trains of hollow fiber modules (BlO Permeators of DuPont), produces
20,000 t/d fresh water at a specific power consumption of 6.3 KWh/to fresh
water. This very low registered figure includes the power consumption for the
feed pretreatment and is a direct consequence of the installed energy recovery
system (Guinard integrated turbo pumps). Figure 6.25 shows the process flow
sheet of the plant. The feed pretreatment is simple: sand-filtration (well water)
followed by a settling tank, acid dosing and cartridge filtration. Feedwater tem-
perature is not a critical factor in this case (in general, the manufacturer limits
the maximum temperature to about 30°C despite the fact that the membrane
itself can withstand higher temperatures). Table 6.1 summarizes the major op
erating data of the plant; Table 6.2 is a breakdown of cost.
In general, the following measures have to be considered for feed pretreat-
ment of RO seawater plants:‘s

l pH-adjustment by acid dosing against carbonate scaling

a chlorine dosing against marine life in the plant
 Design and Optimization
Process 375

0 polyphosphate dosing against calcium sulfate scaling

0 cartridge filtration.

Verlical

-‘:
Well,
(15.3)

Feed
Break
Tank

Feed Boosf
Pumps
,---_A (‘+‘I 1 Product Storage
Tank
t5*11
I
- Caurtlc

t -
I
1
I
I
t + t

R.O. SYSTEMS
(10)

BRINE TO SEA
37.000 Cum/d

Figure 6.25: 20,000 cu-m/d seawater R.O. plant, Ghar Lapsi, Malta-process
flow.

Table 6.1: Malta Seawater RO Plant Initial Operating Data

ACTUAL DESIGN
ITEM UNITS DATAa BASIS
FEEDWATER
Total Dissolved Solids 36,500 39,200
Temperature rr;* 17
pressure bar 70 ;;
Recovery % 33 3.5
Silt Density Indexb 2-4 o-3

PRODUCT
Flowrate m'/d 18,600C 20,000
Total Dissolved Solids pp* 380 500

CONSUMABLES
Electricity KWh/m' 6.30 6.12
Sulfuric Acid pp* 6.6 6.6
Caustic Soda pp* 3.2 3.2
Chlorine pp* 1.0 1.0
I-
a 13 March 1983 (400 Operating Hours)
b Before cartridge filters
' 80% of permeators installed
376 Handbook of Industrial Membrane Technology

Table 6.2: Production Cost Summary Malta Seawater RO Plant

S/m’ S/Kgal
Electricity 0.43 1.62
Permeators 0.11 0.43
Labor L Overhead 0.05 0.20
Spare Parts 0.04 0.14
Chemicals 0.01 0.03
Filters 0.01 0.04
Total Operating 0.65 2.46
Total Investment
costs RO-plant 1.250.000
Civil works 5.000.000

Total production 1.08 4.07

Furthermore, a flushing and cleaning system is necessary. For membranes

which are sensitive to oxidizing agents, active charcoal filtration and/or sodium
bisulfite dosing is mandatory if chlorine has been used in the pretreatment. Note
that the product of any one-stage RO seawater plant contains about 300 to
500 ppm total dissolved solids, i.e., much more than the distillate from thermal
desalination plants. Higher product purities-for chemical processesor for boilers-
can only be achieved by a (2-stage) RO cascade. The second stage can be equipped
with a different type of membrane (high-flux) than the first stage and certainly
must be operated at a different transmembrane pressure difference. The module
arrangement in each stage, i.e., the number of modules in series and/or in par-
allel is determined for a given type of module by the desired yield of the stage
and a necessity to maintain optimal flow conditions in the modules (concen-
tration-polarization). In general, these requirements lead to a so-called “tapered
design” (see Figure 6.13) In any case, the designer faces on optimizing problem
which can be solved numerically by a relatively simple computer program based
on the general flow sheet indicated in Figure 6.26. One result of numerical cal-
culations which seems to be generally valid is that recycling pumps (ZP in Fig
ure 6.26) for the improvement of the flow conditions in the modules and/or
pumps between the module-banks of a single stage (DEP in Figure 6.26) are
not economical. More details about seawater desalination by RO can be found
in the literature, especially in the proceedings of the IDEA conferences and of
the symposia “Fresh water from the Sea”,1912o

Total Desalination of Brackish Water

When desalting brackish water far inland, the discharge of the concentrated
brine may pose a serious problem. For the case where neither deep-well-injection
nor a discharge into a river or canal is possible, a hybrid plant consisting of a re-
verse osmosis process as a first stage and a special crystallization process as a sec-
ond stage seems to be feasible (Figure 6.27).
 Process Design and Optimization 377

SEAWATER

CONCENTRATE
HDl

m. STAGE

FRESH WATER

Figure 6.26: General flow sheet of an m--stage RO cascade.

1 Cry s’ar’ion]

Solid waste

Figure 6.27: Total desalination of brackish water by a combination of RO and

crystallization.

Studies on three real brackish waters of different sources indicated that for
the RO stage the optimum pressure will be between 40 and 50 bar. If modules
of the spiral wound type are employed, for example ROGA 4160 HR. the
Reynolds number, Re = h.v&,, at the inlet should be about 80. Although the
378 Handbook of Industrial Membrane Technology

Reynolds number has a significant influence on the plant structure, i.e., on the
ratio of modules in series and in parallel, its influence on the minimum specific
cost of permeate is nil in a wide range (Figure 6.28).‘l Furthermore, the calcula-
tions clearly indicate that internal recirculations and/or repressurizing pumps be-
tween the module banks of a single stage will not pay off (see Figure 6.26). For
spiral wound modules (Figure 6.29) as well as for hollow fiber modules (Figure
6.30). the lowest costs are obtained without recirculation (z = 0) and without
repressurizing pumps (in Figures 6.29 and 6.30). Costs due to frictional losses
and additional pumps are higher than the possible gain in permeate-flux by
improved flow-conditions.
The second stage of the hybrid-process is essentially a crystallization capable
of crystallizing all solids dissolved in the feed resp. the brine of the RO stage.
Parts of these substances tend to heavy scaling and, therefore, the choice of the
crystallization equipment is limited. A proven solution is, of course, the agitated
thin film evaporator. However, this type of equipment must lead to high specific
treatment costs for two reasons:

(1) An agitated thin-film evaporator is costly and limited iv: size.

(2) Because of its high capital cost, it is in general out of the WeStiOn
to combine agitated thin film evaporators to an energy SaVing mul-
tiple effect evaporation process.

5 IO 15
Number of modules in series

Figure 6.28: Influence of flow conditions in the modules on the specific cost
of RO (tapered arrangement).
 Process Design and Optimization 379

without
repressurizing
--- with I

Roga 4160
Re = 120 I

Module design : 4 /2/l

_-
I LO 60 bar
Pressure Ap

Figure 6.29: The influence of brine recirculation within each bank of modules
(r$++Jti~ = z > 0) on specific cost of RO (spiral-wound).
380 Handbook of Industrial Membrane Technology

1.;

DrJ
t

1.1

f;;
00

IS

0.9 I

- without
repressurizing
--- with I
Permasep B9
Re = 50000
Module design: Z/l/O
0.8
20 40 bar 60
Pressure Ap
Figure 6.30: Influence of brine recirculation within each bank of modules
(mRec/mF = z > 0) on specific cost of RO (hollow fiber).

Low treatment costs for the brine can only be expected if the heat transfer sur-
faces can be designed as tube bundles and if energy saving concepts like MSF,
ME or VC can be applied.
Promising crystallization processes like the seeding process and a modified
MSF process, employing an inexpensive oil as heat carrier, are discussed in Ref-
erence 22.
 Process Design and Optimization 381

One of the most important variables for the hybrid process is the concentra-
tion of the RO brine. In principle, the specific treatment costs of each stage
must increase with increasing concentration of the RO brine. For the RO stage,
this is mainly due to the rising osmotic pressure. For the crystallization step not
concentration per se, but the fact that a rising brine concentration in the RO
stage is equivalent to a decrease in feed flow for the crystallization step, results
in smaller but specifically more costly crystallization units. However, because
the specific treatment cost of the crystallization is in any case higher by an order
of magnitude than the specific treatment cost of the RO step, it is easily under-
stood that the optimum of the hybrid plant is determined by the maximum con-
centration which can be achieved with the RO modules! This result is demon-
strated by Figure 6.31 comparing specific costs of different process combina-
tions.

Capacity RO

ATF: Agitated Ihin Eilm Evaporator

VTE: Vertical Tube Evaporator
MSF: Multi Stage Elash

Figure 6.31: Water cost of alternative hybrid plants,

 382 Handbook of Industrial Membrane Technology

In any case, the specific costs of fresh water for a 2-stage hybrid process
will be 3 or even 4 times higher than for a single-stage RO process. This is a di-
rect consequence of the specific costs of the second stage, regardless of whether
a VTE-seeding process, a modified MSF-process or vapor compression is consid-
ered. (VTE = Vertical Tube Evaporator, here in combination with a seeding tech-
nique, MSF = Multi-Stage-Flash, ATFE/VC = Agitated-Thin-Film-Evaporator in
combination with Vapor Compression). Because of these high treatment costs,
the chances for such a hybrid process are very limited with respect to brackish
water desalination. The situation seems to be different for wastewater treatment
and it is not surprising, that a process as described has been realized for the treat-
ment of the effluents of a power plant (San Juan, New Mexico, USA).
The process is designed to handle 10.000 m3/d wastewater from the flue
gas scrubber and the cooling towers (Figure 6.32). The mechanically pretreated
effluents are fed to a RO unit, designed for a recovery rate of 80%. The concen-
trate of the RO unit is fed to the crystallization stage, designed as vertical tube
evaporator combined with vapor compression. Scaling in the evaporator is pre-
vented by the application of a (calcium-carbonate) seeding technique.

Cleonlng Feed waler COllVENTIOt4AL

of flue gas treatment PRETREATMENT EVAPORATION
llue dust removal
Calcium- Addatives vopour -
__..^^_a^ I I
__---L___,
I

Permeate

Figure 6.32: Treatment of power-plant wastewater; combination of RO and

crystallization.

Automotive Industry-Combination of UF, Cross Flow Filtration and Evaporation

for Recycling of Detergents and Process Water
The final step in the production of automotive parts such as pistons, axles,
etc. is usually a rinse. The spent wash water of such a process contains, in addi-
tion to the detergents, cutting oil and suspended matter and it is impossible to
discharge it into the canalization without further treatment. Fractionating the
wastewater into a concentrate containing the oil and the suspended particles and
into a premeate containing practically no oil is possible by UF.
 Process Design and Optimization 383

The detergent concentration of the permeate and, consequently, its biologi-

cal oxygen demand (BOD) is high, leading to further treatment costs if discharged
into the canalization. For two obvious reasons, the question arises whether such
a permeate could be recycled:

(1) a further treatment or further treatment costs would be obsolete and

(2) the recycling would lead to substantial savings in process water and
in detergents.

The savings in detergents are of prime importance and it must be the aim of
process development to maximize the detergent concentration in the permeate
without increasing the oil-concentration to an intolerable amount.
In on-line experiments with a plant effluent containing 2% detergents,
1% oil and 0.8% suspended solids (metal, dirt, etc.), tubular modules equipped
with UF membranes of different pore-sizes and a microfiltration membrane have
been tested. (The experiments were carried out in the gear and axle production
of Daimler Benz AG.) As expected, the MF membrane produced the highest flux
for detergents. Because of th relatively low retention capacity for oil, however,
the use of MF membranes is limited to a maximum oil concentration of about
10% in the concentrate. In this case, the oil concentration of the permeate is
about 50 mg/S?, contrary to the UF membranes, where the oil concentration
of the permeate is independent of the oil concentration of the concentrate up
to the phase inversion concentration of about 41%.
For ultrafiltration as well as microfiltration the permeate flux is controlled
by a gel layer-not by the membrane itself. This must be concluded from Figure
6.33 because the permeate flux for steady state conditions is almost exactly the
same for the three tested membranes (for comparison, the data of Figure 6.33
have to be corrected for the same temperature and the same flow conditions).

@a Membrane
XT, 30m/s, 33 bnr

0 10 20 30 50 60 min

Figure 6.33: Gel-layer controlled flux in UF and MF (MF-membranes with and

without backflushing).
384 Handbook of Industrial Membrane Technology

Quite often, the formation of a gel layer is reversible; the gel layer can be
removed by cleaning. In every case, periodic cleaning and a higher average flux
have to be weighed against the time for cleaning. Microfiltration is advantageous
compared to UF because microfiltration membranes are available in the form of
tubes withstanding outer or inner pressure of several bars which can be easily
cleaned by reversing the flux for a short time. Despite the advantages with re-
spect to flux and yield of detergents, microfiltration alone will not be a solution
for the described effluent-treatment problem since the retention capacity for
oil at higher feed concentrations is too low. The optimal process is a combina-
tion of microfiltration and ultrafiltration according to Figure 6.34.24

MICROFILTRATION ULTRAFILTRATION
11.6 m3/d 1.16m3/d
FEED F F=2,Om2,Ap=2bor .F=1.6m2,Ap=3.3bor
1 % Oil 10% Oil
T=38OC, Y =S.Sm/s T=36*C, Y =3m/s

TO COMBUSTION 0.12 t-n31 d

1 EVAPORATION 1-1
OR REFINERY 90 % Oil

Figure 6.34: Flow sheet for a combination of micro- and ultrafiltration for the
treatment of oily wastewater.

The recovery of detergents of such a process has been determined to an av-

erage of 85%. It should be noted, that the permeabilities of the individual com-
ponents of a detergent differ and that the upgrading of the recycled permeate
must be based on the selectivity of the process.
The limiting oil-concentration for the first stage, microfiltration, is deter-
mined by the tolerable concentration of oil in the recycled product. Since the
oil retention capacity of the ultrafiltration stage is independent of feed concen-
tration, the final concentration of the retentate is solely determined by the
phase inversion “oil/water -+ water/oil”. At this concentration, the overall water
recovery of the process is above 97.5%. However, the retentate of the process
still contains too much water if incineration or refining is considered. Therefore,
an evaporation step must be included in the process. Almost certainly evapora-
tion will not be economical for the small capacities indicated in Figure 6.34 and
a central evaporation station for the tentates from several production lines
should be considered.

Galvanic Industry-Treatment of Effluents

In the galvanic industry, a treatment of the effluents is mandatory. A
process which is not only capable of water treatment but, at the same time, of
reclaiming metals which would otherwise be lost would be most favorable. RO
is such a process. In the U.S., more than 200 RO plants are installed in the gal-
 Process Design and Optimization 385

vanic industry and, according to reported data, these plants are very profitable
because of the reclaimed metals.
Figure 6.35 is a flow sheet of a galvanic production line completed by RO.
The modules are of the spiral wound type and the membrane material is cellu-
lose acetate in this case.25

Evoporat~on losses
Ltquid 0.04llmm l0.3at/mm
Orogglng losses 0,04llmIn
Delonlrlng wa!er
I I
0.3allm8fl
Golvoniz!ng balh
2aoooomgII
9.Sml O,oLl/mln

I c 1 1 * I_

1027mgll - 407mgll 37mgll

,,1?04
Ilmin
LOBrngll

1703 ltmin 11,361Imm

9663mgA 14300mgll

Reverse 0srnos15 ’ Reverse 05nlos~*

5.6aIlmln
575mg11

5,6alImm I
zadoomg/I

Figure 6.35: Flow sheet of nickel plating, inclbding RO for wastewater treatment.

By recycling of the permeate into the first stage of the rinsing cascade and
recycling the concentrate into the galvanic bath, the recycling is complete; the
evaporation losses of the plant are balanced by the deionized water necessary in
the last stage of the rinsing cascade. Depending on the permeate concentration
which, in turn, depends on membrane retention and on the required recovery
rate of the RO unit, the permeate must be added to the first or second stage of
the rinsing cascade. Table 6.3 shows the (average) composition of the galvanizing
bath; Figtire 6.36 shows flux and selectivity (retention rate) of the modules.

Table 6.3: Composition of Watts-nickel bath

concentration
total nickel ratio 82 g/l
nickel sulfate 6 H20 255 g/l
nickel chloride 6 HZ0 105 g/l
boric acid 45 g/l
brightening agent not analyzed
pH- value 4.3
temperature 60 =C
386 Handbook of Industrial Membrane Technology

80

60

50
0 200 LOO 600 800 1000 1200 1400 h
Time

I
-ii

600

200 LOO 600 800 1000 12

Time

Figure 6.36: Treatment of nickel-plating effluents by RO, module flux and

selectivity.

Figure 6.36 indicates the high retention capacity of the membranes for nickel
and solubles whereas the retention for boric acid is not quite satisfactory (no
explanation could be found for the concentration fluctuations in time).
Payback time of the RO units is given as 8.1 years for single shift operation,
2.4 years for a two shift operation and 1.4 years for a three shift operation.
An analysis of the RO units installed in the galvanic industry demonstrates
 Process Design and Optimization 387

that presently RO units are mainly installed in the nickel plating industry for
two reasons:

(1) the pH value of Watts nickel baths is between 3 and 5 permitting

the installation of cellulose acetate membranes (cellulose acetate,
still being a very common membrane material). It should be men-
tioned that more recently synthetic membranes have been devel-
oped (i.e., the PA 300 of UOP, San Diego, the NS 100 of Abcor
Inc. Wilmington or the PBIL of Teijin Ltd., Tokyo) which proved
to be resistant to chromium acid, copper cyanide and zinc cy-
anide baths at extreme pH values.
(2) the temperature of the Watts nickel bath is 50” to 70%. At this
temperature, the evaporation losses of the bath are such that no
additional evaporators for further concentration of the RO con-
centrate are necessary.

Gas Permeation
With few exceptions, gas permeation on a technical scale employs mem-
branes of the sorption diffusion type. In this case, the flux of a permeating com-
ponent is proportional to the difference of the partial pressures at both sides of
the membrane.

Ji = Qi (PlXi - P2 Yi)

at least as long as the conditions of the gasesare well above the critical point. On
a commercial basis, gas separation started with the recovery of Hz from the bleed
of (high pressure) synthesis loops, employing in most cases a composite mem-
brane “silicon/polysulfone” in the form of hollow fibers.
Asymmetric phase-inversion membranes like the membranes employed in
reverse osmosis are difficult to prepare as gas permeation is much more sensitive
to micropores than RO due to the much higher diffusion coefficients of gases.
For the same reason, the composite membrane differs from RO composite
membranes: in gas permeation, the top layer of the asymmetric support struc-
ture is responsible for the separation while it is the sole duty of the coating to
plug the micropores. Consequently, the material of the coating chosen (silicone)
has a high permeability but a low selectivity while the membrane material (poly-
sulfone) has a high selectivity (and a much lower permeability).
Figure 6.37 shows the flow sheet for the recovery of Hz from the bleed of
the synthesis loop of ammonia plants. In ammonia synthesis, a recycling of the
unreacted components is mandatory because of its low equilibrium conversion
rate. In all technical processes, however, a bleed is necessary because otherwise
the concentration of impurities would increase to an intolerable level. Normally,
the bleed is utilized for heating purposes in the reforming stage. Here, the bleed
is fed to a conventional separation unit first-a scrubber for recovering ammonia.
Behind the scrubber, the modules for the recovery of Hz are arranged in a “one-
stage-two unit” form. The first unit, consisting of 8 hollow fiber modules (total
feed capacity 3,800 Nm3/h) is operated with a transmembrane pressure differ-
ence of 60 bars, the permeate leaving at a pressure of about 70 bars. At this pres-
sure, the permeate can be fed to the second stage of the synthesis feed compres-
sor.
388 Handbook of Industrial Membrane Technology

05 9Sbor
Retentate 4 ’

Water
,

I
I

i 25bar
I
-1,
Feed

AMMONIA -
SYNTHESIS

I
+ Ammonia

Figure 6.37: Flow sheet for the recovery of Hz from the synthesis loop of an
ammonia plant.
 Process Design and Optimization 389

The retentate of the first unit is fed to the second unit (and for this reason,
it cannot be considered a two stage cascade), the permeate leaving at 25 bars as
an additional feed to the first stage of the compressor. The retentate is utilized
for heating pur?osesz6
Since flux and selectivity increase with increasing transmembrane pressure
difference and the modules can tolerate pressure differences of about 166 bars,
the reader might question why the first unit is operated with a transmembrane
Pressure difference of only 60 bars. The reason is that the Hz recovery system has
been added to an existing plant and that the first stage of the synthesis compres-
sor would not accept the permeate flux of both units.
Figure 6.38 shows another example of Hz recovery; here from the tail gas of
a high pressure synthesis (UOP Butamer process). In this case, the tail gas leaving
the conventional fractionating system contains about 70% HZ. The membrane
unit placed behind the fractionating system recovers about 90% with a Purity of
>96%.27 This process seems to be interesting for two reasons:

(I) spiral-wound modules are employed with dry asymmetric phase in-
version (cellulose-acetate) membranes.

(2) the feed contains small amounts of HCI.

ISOEiJTANE
PRODUCT GAS

Figure6.38: Hydrogen recovery from the tail gas of the UOP Butamer proc-
27
ess.

Figure 6.3927 shows a two-stage cascade for the purification and dehydra-
tion of sour gases, mainly removing CO2 and H2S. Again, spiral wound modules
with asymmetric cellulose acetate membranes are employed. It should be noted,
that in this case as in all other cases discussed here, no compressors had to be
installed. This is the main reason why these applications show excellent payback
times.
390 Handbook of Industrial Membrane Technology

TO AMINE UNIT -

FEED @

8
N
TO SULFUR RECOVERY
UNIT

VOLUME FLUX
I Nm3/h I 17110 29830 12625 16872 12712 L130

MOLAR FLUX
[kmol /h I 763 1331 563 753 566 10L

PRESSURE
lbarl 66,5 66,5 65,5 18,2 17,2 1,35

Figure 6.39: Purification of sour gas by a two-stage stripping cascade.27

Pervaporation
Pervaporation differs from all other membrane processes because of the
phase change of the permeate. Masstransport across the membrane is not achieved
by elevated pressures at the feed side as in RO and in gas permeation but by low-
ering the activity of the permeating components at the permeate side. This can
be achieved by applying either a vacuum at the permeate side or by sweeping
the permeate side with a carrier like air or Hz0 vapor (Figure 6.40). Phase
change occurs because the partial pressure of the permeating components is
lower than the corresponding saturation pressure.
392 Handbook of Industrial Membrane Technology

Figure 6.41 shows that the selectivity decreases with increasing pressure at
the permeate side. For a ratio of p/p; = 1 finally, the separation characteristic
is identical to the thermodynamic equilibrium curve (for 19= constant). Pervapo-
ration will always be a process which is relatively expensive compared to other
membrane processes for two reasons:

(1) The modules must be designed for a low pressure drop at the per-
meate side despite the increasing volume of the permeate due to
the phase change since the principle of pervaporation is very sensi-
tive to such pressure losses.
(2) The process requires heat transfer surfaces because the heat of
evaporation necessary for the phase change of the permeate must
be supplied to the process (and must be rejected by condensation
of the permeate).

The necessary heat-flux

;I = Z Ji (Ahvi + c A91
piv
is usually drawn from the feed and will inevitably lead to temperature gradients-
orthogonally to the membrane as well as in the dirction of flow. Especially in
pervaporation of water mixtures with water as the preferably permeating com-
ponent, heat transfer between the bulk of the liquid and the membrane surface
(permeate-side) can become the rate-controlling step.
Figure 6.42 shows the temperature drop orthogonally to the membrane as
a function of heat flux for water/isopropanol flowing in a rectangular channel
(laminar flow, Reh = 498, dh = 4 mm). According to Figure 6.42, the correlation
between measurements and calculations (straight line) based on

is fairly good. ” Quite often the heat-flux is small and, therefore, the tempera-
ture drop orthogonally to the membrane can be neglected. In any case, the tem-
perature decrease in the direction of flow has to be taken into account.
Figure 6.43 shows, for the pervaporation of benzene/cyclohexane and for
symmetric PE-membranes, the temperature decrease along the membrane. The
temperature distribution has been calculated numerically assuming laminar flow
in a rectangular channel with membranes at only one side. Due to the laminar
flow, heat transfer orthogonally to the membrane is only achieved by heat con-
duction and, for this reason, the channel height should be small. The dotted lines
in Figure 6.43 are valid for external heating of the channel wall opposite to the
membrane. Such a design is expensive. In most cases, it will be the far better so-
lution to achieve a quasi-isothermal operation by a combination of modules and
heat exchangers connected in series.
 Process Design and Optimization 393

5-

‘C

I-

3- --

Membrane Mcmbrone Srm I

2-

l-

o’
I
I
I
I
ICAl
I
1
1 0
I
I
I

1
I

1 1500 2000 2500 3000 W/m2

Heat FLUX 4

Figure 6.42: Temperature difference orthogonally to the membrane between

bulk of feed and permeate.

I I /

BenzenelCyclohexane
PE -Membranes
Rear= SO,wi,=O.S,3,=L
- ile, :o
---
% :180Wlm2
-q ’ ’ ’ I I I I I I -
I I II
10’ lo2 mm IO3
Channel Length L

Figure 6.43: Temperature distribution of the feed along a membrane channel.

394 Handbook of Industrial Membrane Technology

The major field of application for pervaporation will be either the separa-
tion of organic components with almost identical boiling characteristics or azeo-
tropic mixtures. In most cases, the desired product quality cannot be achieved in
a single step-either a combination of different processes or a “multi-stage” proc-
ess will be necessary. In this case, a reflux-cascade has to be designed.
Figure 6.44 shows some results of a numerical optimization of pervapo-
ration cascades.32 Assumptions: (I) the selectivity is constant and equal for all
stages; (2) the fluxes are small enough to neglect concentration polarization.

10

6,

Number of Stages

a 10 20
Selectlvlly Sn

Figure 6.44: Optimisation of a reflux cascade for pervaporation: influence of

selectivity on total permeate flow and number of stages.

Mostly, cascades are designed for either constant reflux, or constant cut
rate or as a so-called “ideal cascade”. Here, a different optimization procedure
has been chosen: with respect to costs, the permeate flows are of prime impor-
tance and, therefore, the calculation was aimed at minimizing the sum of all per-
meate flows in the cascade.
Figure 6.44 clearly indicates that for a decreasing selectivity of the module
the specific process costs will increase for two reasons:

(1) the number of stages increases and

(2) the membrane area increases because the internal permeate (reflux)
flows increase.

There already exist applications of pervaporation on a technical scale. A

 Process Design and Optimization 395

hybrid process (Figure 6.45) for the production of pure alcohol, combining
distillation and pervaporation, proved superior to the conventional approach
(extractive distillation) with respect to specific energy consumption (1 kg/kg
ing to Figure 6.45, the alcohol concentration is increased from about 80% to
99.5% by only one pervaporation stage.33 It should be noticed, that for this
range of concentrations membranes for which water is the preferable permeating
component must be used.

1 r--l I I
Ethanol

__ _ 99,a %
I
r

I I
Vacuum

waste Permeate

Feed

Figure 6.45: Hybrid-plant for production of pure alcohol.

Although the high separation potential of permeation has been demonstrated

for a number of systems in laboratory experiments,30 pervaporation seems to be
economical only in cases where high product purities are required-and in com-
bination with a conventional separation process. This statement is based at pres-
ent only on a detailed analysis of the separation of benzene/cyclohexane and
corroborated by the abovementioned production of pure alcohol, but there can
be no doubt that the results are valid for many other systems as well.
The fractionating of a 50% benzene-cyclohexane mixture into products of
98% purity is normally achieved by extractive distillation with furfural as a car-
rier. If pervaporation is to be employed instead, at best a three stage cascade
with the refluxes indicated in Figure 6.46 is needed, where each stage consists
of a combination of membrane-modules and heat exchangers. In a numerical
calculation, the temperature decrease “along” each module is a free parameter.
In Figure 6.46, this temperature decrease is set equal for each module (7.5 K).
This leads directly to the necessary number of heat exchangers and membrane
units per stage and to the necessary heat transfer area.
The flow-sheet indicates a major difficulty of the process: if the membrane
selectivity is high, fluxes will vary markedly with the feed concentration of the
396 Handbook of Industrial Membrane Technology

preferably permeating component. Permeate fluxes might differ by a factor of

10 between the first and the last stage if the same membrane is employed
throughout the cascade. Compared to extractive distillation, a pervaporation
cascade will not be competitive (Table 6.4); investment costs as well as operating
costs are much higher. Especially noted should be the high costs for condensa-
tion-a direct consequence of the low pressure (150 mbar) at the permeate side.
(Entrance temperature of the cooling water was assumed to be 2O’C.)
The situation is different if product purities of about 99.5% are required. In
this case, the energy consumption of the extractive distillation is very high be-
cause of the high reflux ratio (in the extractive distillation tower, the effective-
ness of the carrier is rapidly decreasing in the section above the feed tray for the
carrier and as a consequence, benzene will “enrich” in this section).

COOLING WATER -‘t’T

HEATING

Figure 6.46: Flow sheet of an optimized cascade for the separation of a 50%
benzene-cyclohexane mixture, product quality 98% benzene and cyclohexane.
(a19 = 7.5 K per membrane unit).

A hybrid process according to Figure 6.47 with a one-state pervaporation

for final purification of the top product of, the extractive distillation shows a
cost advantage of 20%, compared to pure extractive distillation, even for very
conservative assumptions. The calculations are based on the following assump-
tions:
 Process Design and Optimization 397

(1) Capacity NF = 40 kmol/h

(2) Equimolar feed of benzene (1) - cyclohexane (2)

(3) Product purities xzl A 0.992 and x12 = 0.995

(4) Effective thickness of the membrane = 8 pm

The optimal feed concentration of the pervaporation unit depends on car-

rier flow rate, reflux ratio and number of theoretical trays of the extractive dis-
tillation. Retentate concentration and cut rate of the pervaporation stage follow
from the requested product quality xl1 + x3, GO.008. For the design of the per-
vaporation stage, the worst case has been assumed that only benzene and no fur-
fural (3) will pervaporate. The major factor for the cost reduction is the much
lower energy consumption of the hybrid process of 1.18 t/h heating steam
against 1.7 t/h for the conventional process.

Table 6.4: Extractive Distillation vs. Pervaporation-Cost Analysis

DE1
spec. costs

3::::::::1.,,

J sieve-tray column 6,11 I __

2
u module -_ 4.85
e, evaporator/heat exchanger 4,43 2,67
2
condenser 1.41 6.33
5
: Pomps 0.45 0,72
z piping 1,86 2.19

membranes (Dbl 100/m* 1 __ 43,65

z
.rl steam 18.81 14.44
+
2:: cooling water 1 ,38 12.69
E.;: power 0,09 0.04
ou

total costs 34.54 87,58

Assumptions
1) membrane service lifetime 20 months
2) cost of piping = 15 % of cost of main items
3) linear amortization within 5 years,
10 % interest
4) Pervaporation: $)u = 75 “C
9v = 3 K per membrane unit (optimal A81
= 8 urn .
%i
LIW
reflux ratio in stage 1: i = 0.75
= 150 mbar lo
p2
extractive distillation: reflux ratio
column I: vI = 2.5, column II: VII = 1
398 Handbook of Industrial Membrane Technology

QC
l
Benzene
x,rr = 0,995
=:

Figure 6.47: Flowsheet of a hybrid-process “extractive distillation-pervapo-

ration” for the production of 99.2% cyclohexane.

According to Table 6.5, a further reduction of the specific separation costs

is feasible if the effective membrane thickness can be reduced, for example, by
employing “optimized” asymmetric membranes.31 It should be noted, that the
optimal feed concentration of the pervaporation unit is not only determined by
the design of the extractive distillation but by the effective membrane thickness
(or specific membrane costs) as well. With increasing permeate fluxes, or lower
specific membrane costs, the optimal feed concentration is shifted to higher
benzene concentrations.

Table 6.5: Hybrid-Process, Influence of Membrane Thickness on Process Costs

and Optimal Intermediate Concentration xlDl of the Hybrid Process

Feed Concentration Product Concentration . bMa L costs [mm 1

Xla X2a x3. xlW x2w xsw [urn) PV +' Dist. Total

0.0062 0,991l 0.0027 0,OOSZ 0,992O 0.0028 8 10.55 63.96 74.51

0.0071 0,9899 0.0030 0,0048 0,992O 0,0032 4 Il.41 53,39 64,80

0.0100 0.9865 0.0035 0,0039 0.9920 0,0041 2 14.25 41.18 55.43

Extractive Distillation 0.0055 0.9920 0.0025 I 93.65

+’ Reference-Stream: Retentate + Distillate of Column II

 Process Design and Optimization 399

REFERENCES

1. Sherwood, T.K., Pigford, R.L., Wilke, C.R., Mass Transfer, McGraw-Hill

Kogakusha, Ltd., Tokyo, 167/169 (1975).
2. Blaisdell, C.T., Kammermeyer, K.,Chem. Eng. Sci., 28, 1249 (1973).
3. Gill, W.N. and Bansal, B., AIChE J., 19, 823/831 (1973).
4. Hwang, S.T. and Thorman, J.M., AlChE J., 26,558/566 (1980).
5. Kammermeyer, K., Chem.-/ng,-Tech., 48, 672/675 (1976).
6. Weller, S. and Steiner, W.A., J. Appl. Phys., 21, S. 279 (1950).
7. Naylor, R.W. and Backer, P.O., AlChE J., 1, S. 95 (1955).
8. Blaisdell, C.T. and Kammermeyer, K., Chem. Eng. Sci., 28, S. 1249 (1973).
9. Breuer, M.E. and Kammermeyer, K., Seperation Science, 2, 319 (1967).
10. Hwang, S.-T. and Kammermeyer, K., “Membrane in Separations”, Tech-
niques of Chemistry, Volume VII, John Wiley & Sons (1965).
11. Walawender, W.P. and Stern, S.A., Separation Science, 7, 5, S. 553-584
(1972).
12. Thorman, J.M., Engineering Aspects of Capillary Gas Permeators and the
Continuous Membrane Column., Dissertation, University of Iowa (1979).
13. Hwang, S.T. and Thorman, T.M., AlChE J., 26,4, 5581566 (1980).
14. Hwang, S.T. and Ghalchi, S., J. Membr. Sci., 11, 187/198 (1982).
15. Rautenbach, R. and Dahm, W!, Chem. Eng. Technol. 10,256-261 (1987).
16. Rautenbach, R. and Arzt, B., The Arabian Journal for Science and Engi-
neering, Vol. 8, No. 3 (1983).
17. Andrews, W.T. and Berman, R.A., “The Malta Seawater RO Facility”
presented at the First World Congress on Desalination and Water Re-
Use, Florence (Italy) 23-27 (May 1983).
18. GVC, Seawater Desalination-Water Pretreatment and Conditioning, Edi-
tor: VDI-Gesellschaft Verfahrenstechnik und Chemie-ingenieurwesen,
Diisseldorf (FRG).
19. First World Congress on Desalination and Water Reuse, Florence (Italy)
(May 1983).
20. Proceedings of the 7th International Symposium on Fresh Water from the
Sea, Amsterdam, 23-26 (September 1980).
21. Henne, K.-H., Ph.D. Thesis, RWTH Aachen (1981).
22. Rautenbach, R. and Henne, K.-H., “Total Desalination of Brackish Water,”
Proceedings of the 7th International Symposium on Fresh Water from
the Sea, Amsterdam, Vol. 1, 163-171 (1980).
23. Paul, H., Schock, G. and Rautenbach, R., Maschinenmarkr, 89, S. 1785-
1988 (1983).
24. Rautenbach, R., Schock, G. and Paul, H., Ferre, Seifen, Ansrrichm., 86, 1,
l-46 (1984).
25. McNulty, K.J. and Goldsmith, A.Z., Gollan, US-NTIS-Rep. PB 266919
(Feb. 1977).
26. Monsanto-Communications, Symposium in Scheveningen (NL) 24.03.
(1982).
27. Schell, W.J. and Houston, C.D., Use of Spiral-Wound Gas Permearors for
Purification and Recovery ( 1982).
28. Heisler, E.G., Science, 124, 77/78 (1956).
400 Handbook of Industrial Membrane Technology

29. Binning, R.C. and James, F.E., Petroleum Refiner, 37, 5, 214-216 (1958).
30. Aptel, P., Cuny, J., Challard, N. and Neel, J., J. Membr. Sci., 1, 271-287
(1976).
31. Rautenbach, R. and Albrecht, R., On the Behavior of Asymmetric Mem-
branes in Pervaporation, European Workshop on Pervaporation, Nancy,
21-22.9 (1982).
32. Albrecht, R., Pervaporation-BeitrSge zur Verfahrensentwicklung, Ph.D.
Thesis, RWTH Aachen (FRG) (1983).
33. Ballweg, A.H., Briischke, H.A., Schneider, W., Tusel, G.F., et. al., Per-
vaporation Membranes-An Economical Method to Replace Conventional
Distillation and Rectification Columns in Ethanol Distilleries, 5th Int.
Symp. on Alcohol Fuel Techn., Auckland, New Zealand (1982).