Desalination - Elsevier Publishing Company, Amsterdam - Printed in The Netherlands

THE STRUCTURE OF CELLULOSE ACETATE MEMBRANES FOR

REVERSE OSMOSIS. PART I . MEMBRANES PREPARED FROM A
DIOXAN BASED DOPE
G . J . GITTENS, P. A. HITCHCOCK, D . C . SAMMON' AND G . E . WAKLEY
Royal Military College of Science. Shrivenham, Wills. (England)
(Received July 30, 1970)

SUMMARY

A study was made of the effect of casting conditions and heat treatment on
the structure of cellulose acetate membranes cast from a dioxan based solution .
It was concluded that optimum performance of such membranes is associated with
subtle changes in the morphology of the upper layers (in contact with the salt
solution) of the 3-layer structure . The formation of the membrane structure during
casting is discussed . The non-reproducibility of the system is compatible with
the instability of the dope, the complexity of the morphology and its sensitivity
to evaporation conditions during casting .

INTRODUCTION

Cellulose acetate is so far the best membrane material for most applications
of reverse osmosis . Various casting conditions, cellulose acetate solutions and heat
treatments have been studied but little has been reported on the relationship
between these parameters and morphology of the final membrane . The general
structure of such systems has been reported by Riley . Gardner and Merten (l, 2) .
Their work only reported on a single membrane and unfortunately used a technique
of staining with osmic acid which can introduce artefacts into the electron micro-
scope image (3) . We are making a study of the effect of membrane preparation
technique on morphology and correlating this with desalination properties .

EXPERIMENTAL

Membrane preparation and performance

The conventional casting procedure as suggested by Loeb and Sourirajan (4)
was used . Films were cast on a glass plate with a doctor blade from a solution of

' Atomic Energy Research Establishment, Harwetl, Didcot, Berks . (England) .

Desalination, 8 (1970) 369-391



370 G . J . GI7TENS et al.

cellulose acetate (Eastman 398-10) in dioxan and methanol . (15-35-50 parts by

weight respectively) . Evaporation was carried out in a forced draught for the times
specified_ followed by dipping in iced water and then heat treatment in water . The
details of casting and subsequent heat treatment are given in Table I . The dioxan/
methanol solvent system is being studied as part of a general programme investi-
gating the effect of casting solvent on structure ; the more usual acetone/formamide
system has also been studied and gives very similar results which will be reported
later.

TABLE I

DETAILS OF MEMBRANES EXAMINED

Leading (L) Trailing (T)

Afem- Evap- Heat Flow rate RF Flow rare RF Evaporation
bracer oration treatment USG!ft' d USGlft 2 d conditions
tiate
(sees)

I UT 5 None 127 1 .14 125 1 .18

I HT 5 lOm/80° 50 4 .14 64 2 .52 Fan on
2 15 10m/80° 21 .2 18 .5 23 .t 18 .5 "Full Cool"
3 60 torn/8o° 0.30 17 .4 0.30 20.9 r A
4 5 5m/75' 63 .7 3 .78 85 .0 1 .98
5 10 5m/75° 56 .6 3 .78 59.0 3 .46
6 15 5m/75c 37 .8 5 .91 42.5 5 .32
7 20 5mJ75` 40.1 4.75 33.0 7 .39
8 30 5m175' 38 .9 4.87 32.3 6.14
9 60 5m175 ° - - - -
10 5 2Om175` 49 .6 3 .67 47.2 5 .38
11 10 20m/75° 31 .9 8.96 34 .2 11 .0
s
12 15 20m/75 3 16 .0 11 .0 25 .7 12 .9
13 20 20mf5' 11 .6 11 .6 15 .1 11 .6 n
14 5 IOm/80° 37.8 9.68 41 .3 10.5
12.7
r
i
15 10 10m180° 35 .4 8.96 33 .0
127 I
16 IS 10m180° 24 .3 24 .2 29 .5
17 20 IOM/80' 24 .8 12 .7 21 .2 24.2
18 30 10m/80° 18 .2 17.2 16.5 17.2 I
19 60 lOm/80° 4 .0 21 .5 4.7 21 .5

Test conditions! 60D psi 0 .05 N NaCl

Typical test duration four hours .

Samples of cast film were taken from the leading (L) and trailing (T) edges
of the casting plate, performance tested in a reverse osmosis cell and examined by
electron microscopy . The fan conditions refer to the forced air draught, at room
temperature (20` ± 1 0 ) and room humidity, over the surface of the evaporating
solution . The air velocities at full cool and half cool respectively were 253 and
218 cm/sec .

Desalination, 8 (1970) 369-391



STRUCTURE OF CELLULOSE ACETATE MEMBRANES FOR RO 371

Electron microscopy
Samples were freeze dried to obtain anhydrous films, and then stored in
a vacuum desiccator . Samples were impregnated with resin (Shell Epon 812) .
under vacuum, and then cast into blocks_ Sections were microtomed from the block
using a diamond knife and floated on to water . The average section thickness was
about 0 .1 pm . Sections were collected on grids and examined with an AEI EM 6G
electron microscope . High contrast plates were used for the micrographs and
final magnification checked using standard particles .
The detail that can be seen in cellulose acetate membranes is limited because
of the poor contrast obtainable with these organic materials . The contrast in
biological materials is normally enhanced by staining with a heavy-metal-containing
stain . We have found that the use of osmic acid solution as a stain (1) causes
hydrolysis of the cellulose acetate and enlargement of the pores with no enhance-
ment of the contrast of the bulk material and was thus not used . No stain was used
in this work .
The ultimate features seen by electron microscopy are limited by the magni-
fication which can be achieved, without loss of contrast, and the grain of the photo-
graphic material . We feel confident of identification of features 10 nm in size .

RESULTS

The product flow rate and salt rejection factor (Initial concentration/Product
concentration) of the membranes studied are given in Table I and Figs .7, 9, 11 .

Type of structures present and their representation

A study of the electron micrographs at higher magnifications than reproduced

ACTIVE SURFACE (FRONT OR TOP)

A

0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 G 0 0 0 0 0
C
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0

NON-ACTIVE RWACE (BAa OR REAR)

Fig. 1. Representation of membrane structure .
Desalination, 8 (1970) 369-391
 Fig . 2. Electron micrographs showing typical 3-layer structure . Sample membrane 12T
Magnification x 5,400.

Desalination. 8 (1970) 369-391

STRUCTURE OF CELLULOSE ACETATE MEN3RANES FOR RO 373

Fig. 3. Typical surface layers A, Band C'. Sample membrane 12T. Magnification :. 18,000.

here . suggest that the membranes can conveniently be divided into three regions
(Fig .!) . Electron micrographs that illustrate these three regions are shown in
Figs . 2 and 3 .
A. This is the "active layer" and appears as a dark band in the micrographs .
It does not generally have pores > 10 nm and does not appear to have pores > 3 nm
in most cases. These identifiable pores would be considered as defects .
B. This is the "sub-layer" immediately below the active surface appearing
as a lighter band on the micrographs . Layer B has pores < 10 nm plus a few isolated
pores > 10 not. The transition between this layer and the A or C layers is sharp .
Fig . 3 is a typical micrograph illustrating the A, B and C layer junctions .
C. Layer C has pores > 50 nut and is the grossly porous region of the mem-
brane . The pore size of the matrix material in this region would be c 10 nm and
probably be of the same order as that in the B layer . This matrix material encloses
or surrounds the obvious pores of the C layer .
Two types of large pore structure are discernible in the C layer :
a. "0" type . This is an open network of pores seen in the electron micro-

Desalination, 8 (1970) 369-391

374 G . 3 . GITTENS et at.

Fig. 4. Typical •'0" type structure in the C layer. Sample membrane t1T. Magnification
18.000 .

graphs as a lace-work pattern with darker vacant space . interpenetrated by epoxy

resin in the sections . Fig . 4 shows a typical example of such a structure . An
analysis of the variation in pore diameter in this material and other membranes
showed little variation perpendicular to the membrane surface but tendency to
increase in size parallel to the membrane surface . Typical sizes near the B-C
junction were 250 ttm diameter. Near the back edge pores were 275 nm diameter
perpendicular to the surface and 400 nm diameter parallel to the surface suggesting
compression of pores during use . The pore density remains fairly constant at 70%,
throughout the C layer .
b. "P" type . This is a closed pore structure . seen in the micrographs as
discrete pores surrounded by areas of membrane matrix . Fig. 5 is typical of this
structure . The size of pores tends to increase from B-C junction to the back surface
(100 to 250 rm diameter) with a slight flattening in the direction parallel to the
surface (350 nm diameter). The pare density does not alter significantly across
the depth of C and is about 45% .

Desalination. 8 (1970) 369-391

STRUCTURE OF CELLULOSE ACETATE MEMBRANES FOR RO 375

Fig. 5 . Typical •' p" type structure in the C layer . Sample membrane 2 HT . Magnification
18 .000.

The flattening into oblate spheroids is not so marked in the P type structure
as in the 0 type. This is consistent with the more rigid nature of the membrane
matrix in this structure, due to the thickness of the interpore regions compared
with that in the 0 structure .
Very often, as will he seen from Tables 11, 111 and V, Oand P structures occur
together in the same membrane and very often differentiation is indistinct . This
probably arises for three reasons :
(1) Heterogeneity in the casting solution .
(2) When phase separation occurs, a wide variation in structure results,
depending on the concentration of cellulose acetate at the point of separation .
our cissification is that for the extremes .
(3) The thickness of sections used in the electron microscope is of the order
of 100 nm . It is possible in such sections to get at least two overlapping pores in
the section thickness . Thus, a high pore density P structure would have the
appearance of an 0 structure as well as the same practical effect .
The physical effect of these different structures in the C layer is that, in
general, the P type would have higher hydraulic resistance and greater rigidity .

EJfecr of hear treatment and evaporation tine in casting on morphology

The morphology of the membranes examined is detailed in Tables 11, Ill,
IV, V . Tables II, 111, IV contain data for membranes at 3 different heat treatments .
Table V gives results on membranes unheat-treated (U), heat-treated (H) and
tested (T).

Desalination, 8 (1970) 369-391



TABLE II

MORPHOLOGY OF MEMIRANES JIEAT•TRRATEO 5 MIN AT 75'

Membrane Eraporvdon Thickness of layers tea;

rime (sees) T cal Structure of C Remarks

4L $ 0 12 .5 28 40 P Large interconnecting
pores at back
4L 5 0 2 .5 40 42 0 B variable thickness
4T 5 0 7 .5 35 42 .5 P much less dense
than 4L
5L 10 0 6 30 36 1' B variable thickness
ST 10 0 10 68 78 P top 0 back B variable thickness
Very porous
6L 15 0,02 22 18 40 CID layer ill-defined A indistinct and non-
due to very brittle coherent
area, Low density I'.
6T IS 0.10 20 90 110 0 tending to P at A irregular with many defects
rear
7L 20 0,7 4 35 39 0 A varies 0.2 to .08 tent with
defects
7T 20 0 .25 16 24 40 0 A layer even
8L 30 0 .3-2 .0 4-20 9-34 30-40 Mixture Pj0/B B variable and brittle struc-
tures. Gross porous defects
in A
8T 30 0,12 4-10 50 60 0 tending to P II variable with large pores
at rear approaching close to surface
a 9L 60 1 .2 top 35 36 Absent A layer at top and bottom
IA rest
v
0 9T 0h top 28 29 Absent A layer at top and bottom
w 0.4 rear
rn
p
W


'TAIILE III

MOI4PtIOLOOY Cl: MFMPIIANLS llEATfltLAlEt) 20 MIN Al 75 c

(fsvpiira ;ion Thickness o/Ixnyr .i (pin)

Menihrnnc tine (sees) A Ii C' Structure uj• C' Rrnmrks

IOL 5 0.15 12 43 55

lOT 5 0 16 54 70 Porous areas to top of B

IlL 10 0 56 25 81 Some pores tip to 1St) nni in H

lIT 10 0 .6 IS 40 Layers variable in width

12L IS 0 .3 12 22 Some porous areas in B layer

12T IS 0 .26 2 41 Less porous than DL in B
13L 20 0,09 27 13 C low pore density
B layer contains pores 50 rim
131' 20 025 30 tO 40 B layer conlairis SO am pores
 TABLE IV W
.1
r_
MOItPIIOLIXIY OF MLMfRANE511t :AT-TNEATEO 10 MIN AT 80'C

Eavrporariorr T/uc4rretr of lai¢•cr (lun)

MentLrtnre trine (sees) .-A It C ---- Total diraewre of C Rernor6 .s

14L 5 5 20 25 O tending to Pat rear U layer thickness variable

141 5 1 .5 50 51 o all through
15L Ill 0 14 38 52 P all through
151 10 t) 19 19 38 P all through B layer brittle towards
C layer
16L 15 0 .10 16 39 55 o at centre 19 pm thick B layer brittle towards
changing to I' at rear C layer
Two quite distinct layers A layer variable in thick-
ness
161 IS 0 6 40 41, Mixed 0(P tending to Pat rear
I7i. 20 0-0 .4 25 t0 35 P structure but very Brittle layer very aeccn-
dense membranes tuated at B(C junction
171 20 0 .3 20 14 34 P structure mixed with some U liritde at UIC junction
181 30 0 37 6 41 I' only B layer alt along hack edge
19L 60 0 .25 40 25 42 P but very few A variable in thickness
Brittle towards back or
It layer
191 0 .06 35 35 Non existent Brittle towards rear
Many defects in A
a
0 n
W

aJ
W
a'
b
t
b A
TABLE Y -I
x
11
El{rt'1(HIII .At RI AtMflrx113lt .S1INCFONnti}Rl'11ou%iY 11

LI-(IparaIuur 3Tricknes n) k ; r rr"nl

;1ft,ubrauv hint' (.,rrs) 71-rat,n nr :I C n(Gr Rentmrks

lU Uncunxl 0,2 IS 24 tT \Vhcre distortions occur sur-

face gets a Ihickcning of ae-
ISe buyer
IUF 5 Uncured IS IS Indktirrct brittle 0 structures collapsed
and tested appctranet
Cured 0 .4 C 24 30 0 tc, .flog to I' in
regions of support p
grids r~
2U IS Uncured 0 .4 45 48 P . . O ' P Poles less dense towards rear '=
cs
ID11 ;IItlS tear x
214 15 Cured (1A4 49 51 0 Pores opened up in C layer
2FCf 15 Cured 0,04 42 48 i' Whole structure appears to
and tested open Out although total k
membrane shrinks x
3U 6(Uncured 0 16 IG C'omptetcly homogeneous
311 60 Cured 0 17 17 Completely homogeneous
CO
3W' 50 Cured 20 21) Completely homogeneous
and 1cuclF

w
'C
a


380 c . 1 . GSrItNS et at

2.0 40

30
E
.
a
0

20 u
a a'

05 10

O O
p 20 30 40 50 60
EVAPORATION TIME fknl

Fig . 6. Variation of morphology with evaporation time for membranes heat treated for
5 min, at 75`C. A , B - -'-, Speculatiae . . . L-heading edge, T-trailing edge.

80 e

v 60
N
a
R
a
3t 40
O
V

20 2

10 20 30
EV'APORAT0
.V TIME 15ecs)

Fig. 7. Variation of performance with evaporation time for membranes heat treated for
5 min. a t 75 'C. Flow rate Rejection factor---.

Desalination, 8 (1970) 369-391



STRUCTURE OF CELLULOSE ACETATE MEMBRANES FOR RO 381

The variations of morphology with evaporation time are given graphically

in Figs. 6.8 and 10 . Regions of the graphs where data are missing are indicated by
dotted lines drawn with reference to the general shape of the carves . Where a
range of values is indicated this was the observed variation in the samplesexamined,
generally in the order of 10 sections from one piece of membranee

06

e
a 04

5 to t5 20 25
eVCPORATtO TIME ttecss

Fig. 8. Variation of morphology with craporation time for membranes heat treated for
20 min . at 7YC. A , n ---, Speculati~c . . . L-leading edge, T---trailing edge .

I6

z
0

5 tO 45 20 25
EVAPORATION TIME tsn,)
Fig. 9. Variation of performance with evaporation time for membranes heat treated for
20 min at 75 - C. Flow rate , Rejection factor - --.

Desalination, 8 (1970) 369-331



3S2 G . l . GITTENS et al.

l ty f t
fi
I, , t

10 20 30 40 50
..e .veT vs¢. nm : :-e v4v EVAPORATION TIME (Sees)

A- x n-$
B- S s- •
Fig . 10. Variation of morphology with esaporation time for membranes hcal treated for
IOmin-atSOC . .A----,8---,Spcculati~c . . .L leading edge, T-trailing edge .

40 40

U
V
4
2
C 0

z 20 20 W
P W
Cc

t0 10

ID 20 30 40 50 60
EVAPORATION TIME (Sets)

Fig . It . Variation of performance with evaporation time for membranes heat treated for
10 minat SOt. Flow rate , Rejection factor---.

DISCUSSION

Genera/ observations
All membranes show a large degree of heterogeneity in their features . This
appears to b: due to two factors :

Desalination, 8 (1970) 369-391



STRUCTURE OF CELLULOSE ACETATE MEMBRANES FOR RO 383

Firstly, the casting solution itself was very unstable with respect to phase
separation : slight lowering of temperature causing precipitation and coagulation
of the cellulose acetate .
The second factor is the sensi'ivity of morphology to evaporation time during
cistlnLt .
The latter explains the results 8L and 171 (Tables Ii and IV) where the A
layer thickness appears to vary randomly . These membranes have evaporation
times where the A layer thickness is changing very rapidly with evaporation time .
These two factors mean that minor changes in the solution concentration, solution
temperature and rates of evaporation from the surface of the casting solution
could easily cause these rapid changes in morphology .
These factors not only show themselves in layer thickness but also in the
structure of the C layer which varies from O to P type, sometimes both types
coexisting with large areas of denser membrane of B like structwic . The general
tendency is far the C layer to be O type in the early stages of evaporation tending
to P as evaporation in_reases. There is also a gradual decrease in pore volume as
evaporation time increase until the C layer contains no large pores .
The rear edge of the membrane always shows a coherent B layer of the order
of 0.1 -.0.25 pm thick : (Fig. 4) very rarely do pores open out into the supporting
epoxy resin_ This suggests that when water penetrates into the rear of the membrane
or when evaporation occurs from this surface, precipitation of membrane material
occurs, as with the upper surface at low evaporation time, to produce a fairly
porous B layer .

Effect of hear treatment and resting on morphology

The series of results in Table V demonstrate that the onl obvious effect of
heat treatment and testing is to decrease the size of the C layer without any signifi-
cant effect on the B layer .
If the membrane is tested without preaous heat treatment, the structure
completely collapses (I UT) . demonstrating that heat treatment causes a strength-
ening of the formed structure . This is further confirmed by comparing the results
in Tables II . Ill and IV . As the heat treatment is taken to higher temperatures or
longer times the membrane material becomes harder, as evidenced by the brittle
appearance of the cut sections for membranes in Table V . Temperature 180") has
a greater effect than increasing time at the lower temperature . although the reverse
is true of water flow rate .
Tables II, III and IV show that heat treatment affects the nature of the C
layer . In Table II. the variability of the C layer structure continues up to 30s
evaporation, while Tables Ill and IV demonstrate that with more heat treatment
the structure is predominantly P after 15s_ This suggests that O type may change
to P type on heat treatment . This could possibly occur by the collapse of some of
the O pores to produce discrete P pores . It would be expected that the P structure

Oesatinction. S (1970) 369-391



384 6 . 1 . 611T£rs et al.

with its lower pore density would have a lower hydraulic permeability . Thus this
could contribute to the decreased flow rates after heat treatment .
There is some evidence that using membranes has a similar effect to heat
treatment (2H and 2HT) . This has direct relevance to the changes which occur
with time under pressure, since annealing of the structure gives rise to compaction
and would gradually reduce the flow rate through the membrane . Any compaction
of the B layer which is apparently less porous than the C layer, would decrease
flow rate more significantly than merely compression of the grossly porous C
layer_
The variation in the absolute value of the thickness of the A layer in the
three sets of data at different heat treatments (Tables II, Ill and IV) is not con-
sidered significant. These membranes were cast on different days and variations
from day to day may have caused this . This difference may also be due to the
inherent variations in the membranes and the particular samples studied . Ti .e very
thick A layer of Fig. 6 merely being an example of what might have been found
with the other heat treatments had the particular samples been examined .
Wasilewsl.i ( .'•) showed, using microwave adsorption techniques, that on
heat treatment there was a rotational movement of the segments of polymer
chains within the forming active layer. This gal a rise to a change from intramolec-
ular hydrogen bonding of molecules as predominantly occurs in the 'as cast"
state, to intermolecular hydrogen bonding in the heat treated and salt rejecting
states . This rearrangement occurs most rapidly between 70 and 85°C . The re-
arrangement effectively immobilises segments and demonstrates the importance
of hydrogen bonds in the membrane structure . Such a rearrangement can only
occur when the polymer chains or segments are suRiciently close together that
bonding is likely to occur . This will be most likely in the relatively dense surface
layer and in the inter-pore matrix of the sub-layers . Thus, although a shrinkage
and stiffening of the whole membrane is observed on heat treatment, there does
nort seem to be a corresponding shrinkage of the open pores at the rear and most
of the effect must occur in the surface or matrix material of the sub-surface layers .
This is some justification for the idea that the solvent depleted layer is more porous
in membranes with an active layer than membranes without, otherwise the flew
rate would drop much further in the case of membranes with an active layer.
These considerations mean that the major structural changes on heat treatment will
occur over distances below the limit of our technique and hence these subtle
changes would not be seen .

Effect of evaporation time

The most noticeable effect of increase in evaporation time at the casting
stage, besides that of changes in the C layer structure, is the very great changes
which occur in the thicknesses of the A . B and C layers . The variation in thickness
of the A and B layers, adjacent to the concentrated salt solution, is of great

Desalination, 8 (1970) 369-391

STRUCTURE OF CELLULOSE ACETATE MEMBRANI3 FOR RO

interest since these layers are considered to control the water flow rate and the
salt rejecting properties of the membrane .
The smoothed variations in thickness of these two layers are plotted in Figs . 6.
8 and 10, From a comparison of the behaviour of the three sets of data and the
leading (L) and trailing (T) edge data of the individual sets, the dotted lines were
drawn to include regions where data were missing . Such a treatment is fully
justified since it has been observed that heat treatment has no effect on B layer
thickness and sufficient experimental points are available to indicate the general
trend of results . The thickness of C does not show similar regular variations, due
to variation in the total thickness of the membrane during casting. It has been
assumed that the thicknesses of B and A are independent of the total membrane
thickness except where B extends the full thickness of the membrane at the longer
evaporation times .
Figs. 6, 8 and 10 demonstrate the very rapid variations in structure which
occur even with small differences in evaporation time . These rapid variations
account for the heterogeneous nature of the structure of some membranes even
from the same sample . It will be appreciated that small changes in the evaporation
conditions over the surface of the membrane could result in the present or absence
of the A layer, e.g. 17 L They also account for the fact that the maxima in the case
of Fig. 6 AT and Fig. 8 BT and the minima Fig. 10 AL and Fig . to &L were missed
by the experimental points ; these occurring in regions where membranes were not
produced .
A fact which adds confirmati o dependence of the rejection factor on the
thickness of A and B layers and also gives confidence in the general features
observed, is the relationship between the position of maxima in thickness for the
leading and trailing edge results, and the shape of the rejection factor curves . In
every case except one the order of L and T maxima are the samv for individual
sets of data. L before T . Fig . 6 and Fig . 10 and T before L. Fig .S. The exception is
the Fig. 6 BT results and suggests that salt rejection is very largely dependent on A .
The absolute position of maxima and their height is, of course, open to
insufficient results have been obtained . Unless a large increase in the
number of measurements were made . very little increase in the precision of the
location of maxima could be made, due to the very rapid variation in values and
the variation withinn samples_
The fact that a reversal of the effect of trailing edge is obtained with one set
of results and the closeness of the ma,dma in the morphology curves suggests that
the effect of leading and trailing edge may not be very significant, and that varia-
tions over quite small distances, due to slight changes in evaporation are more
significant .

DemIkarion, 8 (1970) 369-31 •€

386 c . .2, cmExs et al.

Origin of the variation in thickness of A and B layers, and the difference in structure
of the A . B and C Layers
Fig .12 is a generalised representation of the variations in thickness involved
and will be used as a basis for discussion .

FLOW uTta11 TnIWiESS a +

sutaioM $ VIICKME55 s +
r1CT0i1(An

Fig. 12 . Generalised representation of morphology and properties. --- etaporation

produced component of .4_

Many factors must be considered :

a . The concentration of solvent in the system at which separation of solid
cellulose acetate occurs_
b. The permeability of layers produced by separation of solid cellulose
acetate to solvent, as vapour or liquid .
c . The effect of water on the cellulose acetate solution at various solvent
concentrations (degrees of evaporation) .
d . Permeability to water of layers of cellulose acetate separated from solution
by solvent evaporation or precipitation on contact with water .
e . The variation in concentration gradient in the membrane with evaporation
time, and the effect of separated layers on this gradient .
The solvent in all discussion will be considered to be the dioxan/methanol
system . In evaporation, the methanol will be removed faster than the dioxan . but
as the ratio remaining at any one time is not known this factor will be ignored .
Papkov et al. (5) showed that cellulose acetate can yield true thermodynami-
cally stable homogeneous solutions which obey the phase rule ant which assume
metastable character in specific temperature and composition ranges, separating
spontaneously into two phases .
Tarakanov et at. (6) made a study of the cellulose acetate (54 .5% acetyl)/
diethylene glycol ethyl system which has an upper consolute temperature of 60°C .

Desalination, 8 (1970) 369-391

STRUCTURE OF CELLULOSE ACETATE MEMBRANES FOR RO 387

They demonstrated the formation of dilicrenl types of gel structure on cooling

hot solutions depending on the composition of the original. With polymer contents
less than 10°• , they suggested that on two-phase separation drops of concentrated
polymer solution separate, which partly coalesce forming a three-dimensional
network composed of interpenetrating and interlocked particles of the new polymer
phase . The spaces between the structural elements are appreciably interconnected,
type O structure . Wnen more concentrated polymer solutions separate into two
phases, a cellular structure is produced . This is because the drops of polymer
solution which are formed cannot diffuse to form large globules because of the
higher viscosity of the system and instead join up with nearest neighbours to form
a network . In this system one gets a large number of isolated partially connected
pores, type P structure.
The separation of cellulose acetate solutions into two phases leads to the
formation of the porous dispersed precipitated gel structures . The separation is
brought about by evaporation of the volatile solvent, addition of a non-sohent
such as water or the change in temperature which causes the system to move into
the two-phase region . It must be borne in mind that ve :y often the system will be
in metastable equilibrium and will thus be a homogeneous one-phase system but
very susceptible to thc influence of temperature or concentration/non-solvent
effects . If rapid precipitation occurs due to the virtually instantaneous transforma-
tion from a metastable to a stable precipitated structure then the grain size would
tend to be very small gi'ing rise to structures such as are seen in the A and B
layer . This rapid precipitation would also give rise to sharp boundaries between
layers as is observed . Rapid precipitation could be caused either by nucleation
of a weakly metastable situation by precipitation of polymer on contact with a
non-solvent, or purely thermodynamic by moving into a very metastable situation
due to rapid decrease in temperature on immersion in iced water . Thus the A . B
and C layers would be produced by two different mechanisms . A and B by rapid
precipitation and C by slower phase separation, where grain growth is possible,
into type O or P structures .
When the cellulose acetate dope is spread on a glass plate, evaporation takes
place from its surface . This results in the formation of a layer depleted of solvent,
with increasing solvent concentration as we move away from the surface . Due to the
surface active properties of the polymer molecule, there is a predisposition for
molecules to align parallel with the surface, rather than forming agglomerates . The
solvent concentration gradient is maintained during this process by solvent diffusing
from the rear of the membrane, with a gradually decreasing solvent concentration
on the surface . Depending on the length of time this evaporation has proceeded
and the speed at which it proceeds, different phenomena can occur before or
after immersion of this solvent depleted system into iced water .
Minimal evaporation of volatile components (a) Fig . 12 . The solvent depleted
layer is thin and the concentration of solvent is 3ulhcient to prevent phase separa-
Desalination, 8 (1970) 369-391


388 c. : . GtTIENS et alt

non before the material is immersed into water . A fairly fast interchange of solvent
and water then occurs with the formation of a B layer . The thickness of B in the
absence of any other surface layer is dependent on evaporation time. The B layer
which results would be fairly open-pored . When water penetrates past the solvent
depleted layer . slower phase separation occurs, producing from the dilute polymer
solution a very open O type structure of the C layer .
Short evaporation time ((b) Fig .l2) : The width of the solvent depleted layer
before immersion is large (thicker than for (a)). approaching the maximum of the
B curve . The concentration of solvent at the air-solution interface is now much
less and the system is in a one-phase metastable condition within the two-phase
region_ immersion in iced water causes a rapid phase separation and precipitation
of cellulose acetate which, combined with the orientation of molecules at the
interface gives rise to the "active" or A layer . The A layer limits the flow of water
to the adjacent solvent depleted layers allowing subsequent phase separation in the
B layer to occur more slowly, producing a more open layer than that formed
without the presence of the A layer .
There is no gross change in taster throughput of the final membrane on the
formation of the A layer, suggesting that minimal evaporation produces a thick
small pore layer, while short evaporation produces a thin less porous layer and a
thick more porous layer .
Longer evaporation time ((c) Fig . 12) : As the concentration of solvent at the
air-cellulose acetate solution interface decreases phase separation and formation of
an A layer occur before immersion in water . When this A layer is present the rate
of evaporation of solvent from the membrane is reduced enough to allow diffusion
of solvent from the rear layers to be faster than evaporation . This results in a
redistribution of solvent, an increase in solvent concentration in the B region
and thus a reduction in thickness of the B layer . Eventually as evaporation proceeds .
B increases again as more solvent is removed and finally extends the full thickness
of the membrane . ((c) Fig . 12) thus corresponds to the maximum B layer thickness,
which occurs just after the formation of the first A layer by water immersion .
On immersion in iced water, more A layer is produced by the intervention of
water into the metastable layer, below the evaporation formed A layer, and
diffusion through this layer produces the B layer . The maximum thickness of A
occurs at (d) Fig .l2, after which it decreases due to the thickness of the A layer,
produced by evaporation, now effectively preventing permeation of solvent and
formation of A by pr.cipitation . This point also corresponds to about the minimum
in B . This minima is a result of the combination of the redistribution of solvent
after skin formation and the poor penetration of water into the structure, opposed
by the decrease in solvent concentration in the cellulose acetate solution due to
increased evaporation time. At (e) Fig .l2 the membrane structure is probably
completely formed by evaporation or diffusion of solvent from the cellulose acetate
solution into water, without the intervention of the water precipitation step . B

Desalination, 8 (1970) 309-391



STRUCTURE OF CELLULOSE ACETATE MEMBRANES FOR RO 389

increases in size until it is the total thickness of the membrane . Any C layer
present would be formed by lifting of the membrane from the casting plate and
water penetration from the rear. At and after this point, the boundary between
A and B layers becomes more diffuse and this contributes to the apparent decrease
in A . Some membranes still show evidence of an A layer at long evaporations
(9L, 9T, 19L, 19T) and even A layers along both edges . This latter is consistent
with the idea of skin formation limiting evaporation and/or solvent diffusion
from membrane into water and water diffusion into the membrane .
The evaporation produced component of the A layer could be represented
by the dotted line in Fig . 12 joining up with the dashed extension of the A tine
where evidence is doubtful

The function of die layers in membrane performance

The optimum value of the rejection factor is reached at or just belbre the
maximum value of A. It is reasonable to assume that A is the controlling factor
in salt rejection, although in the very early stages of evaporation B may show
salt rejection. Plotting experimental layer ihicknesses against salt and water fluxes
suggests that certainly in the later stages of evaporation increased B layer thickness
has no effect on salt flux only reducing water flux. The technique of examin~tian
cannot distinguish a layer on the surface less than 10 nm thick . A layer at this
value or less could be responsible for the rejection observed at very low evaporation
times but would probably be subject to many defects. The fact that optimum salt
rejection occurs about the maximum thickness of A is probably because the
minimum number of continuous defects will occur, per unit area of A, at this
value. The salt rejection properties of the membrane after formation of the active
layer are superior to membranes without this layer, but will of coarse be more
subject to surface faults, since rupture of this thin layer exposes a more porous
layer than present in membranes without surface layers. The thickness of A must
thus be optimised since it also contributesalong with the thickness of B to decreasing ;
the flow rate . The fact that the maximum thickness of A occurs about the minimum
thickness of B means that the increase is hydraulic permeability due to the decease
in B is compensated by the thickness of A . The steady fall in flow-rate with
evaporation time suggests that the total flow rate is goverr :ed by the sum of the
permeabilities of the two layers . The ratio of the thickness of the two layers is
about 20 :!, B :A at the maximum thickness of A, and suggests that permeabilities
may also be in this ratio . Absolute evaluation cannot be made since the actual
ihicknesses are not known, particularly at maxima and minima, with any certainty,
and also the permeability of B may be different before and after the formation of
the A layer. It does seem that taking into account this difference in permeabilities,
the sum of permeability x thickness for the two layers would give fair agreement
with the shape of the flow-rate curve . Since B does contribute to the flow rate,
speculation is aroused about its behaviour on long-term compression and corn-

Desalination, 5 (1970) 369-391



3 90 G . I. GITltNS et af.
paction and the part which it will play in the long-term performance of membranes .
It will probably be this factor which will largely determine the decrease in flow rate
with time of the membrane.
The best membranes would be produced by minimizing ti- thickness of
any sub-layer formed, but at the same time improving the coherency of the top
layer.
From the point of view of hydraulic resistance oldie membrane, we can thus
consider three regions : (a) the active layer of permeability x per unit thickness ;
(b) the sub-layer of permeability say 20 x per unit thickness and (c) the very
porous backing of permeability 100 x per unit thickness- The pore walls them-
selves are probably porous on the < 10 nm scale, while the open pores have sizes
from SO to 1000 nm . The sub-layer would have matrix pores < 10 nm probably
down to 2 nm . This would account for the results of the small angle X-ray results
(3) which show that most membranes have an average pore size around $-9 nm
which is little influenced by heat treatment or usage and consists of predominately
spherical pores.

Mechanism offailure (effect of use)

Two types of failure occur : (1) due to specific defects and (2) due to the
inherent properties of the cellulose acetate .
(1) That due to defects occurs due to the heterogeneity of the structure.
Electron microscopy is the only technique capable of seeing these changes in the
active layer because of the thinness of this layer. Electron microscopy for this
type of defect suffers from the disadvantage that time does not allow enough
samples to be taken to ascertain the frequency of defects and thus observations
which are made are, for the most part, circumstantial. This means that general
conclusions on specific defects are difficult to draw, although chance observation
demonstrates their presence. The large variations in thickness of the active layer
from zero to very thick and often a saw tooth variation, mean that at some places
the active layer is virtually nonexistent . At these points, the sub-layer may of
course be of greater density to compensate, but if not, considerably higher through-
puts and less salt rejection will occur .
To reduce the salt rejection by 1 from say 99 .9% to 99% these defects
would need to affect 1 % of the surface, assuming similar flow rates through both
regions . Flow rates through these regions may be faster if the underlying structure
is porous.
If flow rates were faster in defect regions, it is possible that mechanical and
chemical (hydrolysis) effects on the membrane would be larger ; defects providing
nucleation points for chemical and mechanical breakdown . Reduction of the
acetate Content of the membrane by 2% would cause a doubling of flow rates .
(2) The inherent properties of the membrane are the main reason for the
progressive failure of membrane, the symptoms of which are a steady drop of

Desalination, 8 (1970) 364-391



STRUCTERE OF CELLULOSE ACETATE MEMBRANES FOR RD 391

water flow. It is observed that membrane thickness is reduced during life suggesting
a compaction of the cellulose acetate . It is believed, by us, though information is
difficult to obtain even by electron microscopy, that the main cause of increased
hydraulic resistance is closure of already quite small pores < 10 nm in the cellulose
acetate immediately below the active layer .

CONCLUSION

It appears that the preparation of asymmetric membranes is a subttecombi-

nation of phase equilibria and diffusion rates, and that the properties of such
membranes are a complex function of their morphology . This suggests that more
attention should be placed on the interrelations of variables in the casting process
if a reproducible cellulose acetate membrane is to be produced .

ACKNOWLEDGEMENTS

We would like to acknowledge the work of Mr. D . Warren in preparation

of the membranes and the Ministry of Defence for supporting the work .

REFERENCES

1 . R. L. Ran . 10. GARDNER AND IL MERTEN, Science. 143 (19611.401 .

2. R . 1. . RIEy. U . MEi UN AND 1 . O. GARDNLa, Desalination, 1 (1966) 1 .
3. G . J. GtrrENS, Results to be published .
4. S.LAEBANDS.SOURIRAfAN, iJnirersiet4California atLos Angeles, Department ofEngineering
Reps . No . 6640, July 1960.
5 . S . F. FAPROV, V_ A . KARCIN AND Z. A . R000QIN, Zh. Tb. Khim., M (1937) 156, 607, 793 :
13 (1939) 206.
6 . O . G . TARAkANDV, A . 1 . DtaNINA, N. A, ABATUROVA AND 1 . N . VLODAVnS, Dokiady, And.
Nook. S.S.S.R., 18011965) 164.
7 . S . W*saiws a, Brit. J. Appl. Physics, 17 (1966) 675.

Desalination, 8 (1910) 369-391