Vous êtes sur la page 1sur 15

iournalof

MEMBRANE
SCIENCE
ELSEVIER Journ~ of Membrane Science 129 (1997) 221-235

Crossflow microfiltration of oily water


Jeffrey Mueller 1, Yanwei Cen 2, Robert H. Davis*
Department of Chemical Engineering, University of Colorado at Boulder, Boulder, CO 80309-0424, USA

Received 19 July 1996; received in revised form 12 November 1996; accepted 14 November 1996

Abstract

Two a-alumina ceramic membranes (0.2 and 0.8 jam pore sizes) and a surface-modified polyacrylonitrile membrane
(0.1 jam pore size) were tested with an oily water, containing various concentrations (250-1000 ppm) of heavy crude oil
droplets of 1-10 ~tm diameter. Significant fouling and flux decline were observed. Typical final flux values (at the end of
experiments with 2 h of filtration) for membranes at 250ppm oil in the feed are ~30--40kg m -2 h -1. Increased oil
concentrations in the feed decreased the final flux, whereas the crossflow rate, transmembrane pressure, and temperature
appeared to have relatively little effect on the final flux. In all cases, the permeate was of very high quality, containing <6 ppm
total hydrocarbons. The addition of suspended solids increased the final membrane flux by one order of magnitude. It is
thought that the suspended solids adsorb the oil, break up the oil layer, and act as a dynamic or secondary membrane which
reduces fouling of the underlying primary membrane. Resistance models were used to characterize the type of fouling that
occurs. Both the 0.2 jam and the 0.8 jam ceramic membranes appeared to exhibit internal fouling followed by external fouling,
whereas external fouling characterized the behavior of the 0.1 jam polymer membrane from the beginning of filtration.
Examination of the external fouling layer showed a very thin hydrophobic oil layer adsorbed to the membrane surface. This oil
layer made the membrane surface hydrophobic, as demonstrated by increased water-contact angles. The oil layer proved
resistant to removal by hydrodynamic (shear) methods. By extracting the oil layer with tetrachloroethylene, followed by IR
analysis, its average thickness at the end of a 2 h experiment under typical conditions was determined to be 60 ~tm for the
0.2 jam ceramic membrane and 30 jam for the 0.1 jam polymer membrane. These measured amounts of oil associated with the
membrane at the end of the experiments are in good agreement with those determined from a simple mass balance, in which it
is assumed that all of the oil associated with the permeate collected is retained on or in the membrane, indicating that the
tangential flow did not sweep the rejected oil layer to the filter exit.

Keywords: Fouling; Water treatment; Produced water; Microfiltration

1. Introduction

*Corresponding author. Tel.: (303) 492-7314; fax: (303) 492- Wastewaters containing dispersed oils and sus-
4341, e-mail: davisr@spot.colorado.edu.
p e n d e d particles are p r o d u c e d f r o m diverse industrial
tPresent address: Harrison-Western Environmental, 1208 Quail
Street, Lakewood, CO 80215, USA. sources, such as general metal-working, f o o d proces-
2present address: Chusei, Inc., 12500 Bay Area Boulevard, sing, transportation, and gas and oil production. Waste
Pasadena, TX 77507, USA. streams f r o m onshore and offshore oil and gas opera-

0376-7388/97/$17.00 © 1997 Elsevier Science B.V. All rights reserved.


PII S0376-7388(96)00344-4
222 J. Mueller et al./Journal of Membrane Science 129 (1997) 221-235

tions are among the largest sources of oily waste- tegies and pretreatments, and hydrodynamic techni-
waters. The volume of oily waste or 'produced' water ques [8]. However, specific fouling mechanisms and
generated annually in the US onshore oil and gas reduction strategies during microfiltration of produced
operations is estimated at 33 billion barrels [1]. These water are not well understood.
produced wastewaters cannot be reinjected to oil wells Although there is relatively little understanding of
or discharged to receiving waters, because they con- fouling mechanisms for oily wastewaters in microfil-
tain high concentrations of oil, grease, and suspended tration processes, some results have been reported for
particles. Typical composition ranges include 50- ultrafiltration (UF) membranes used for the removal of
1000 ppm total oil and grease, and 50-350 ppm total lubricating and cutting oils used in the metal industry.
suspended solids [2,3]. Oil and particulate concentra- Bhattacharyya et al. [11] observed internal and exter-
tions >10 ppm in water can plug injection wells, foul nal fouling during UF of a lubricating oil-nonionic
equipment, or form heat-insulating films. Moreover, detergent-water solution through noncellulosic, tub-
regulations require that maximum total oil and grease ular membranes. They noted that membrane fouling
concentrations in discharge waters be 5-40 ppm, with and cleaning requirements depend on the type of oily-
a typical requirement of 10-15 ppm [4]. Thus, sub- water systems and membranes. Lee et al. [12] studied
stantial removal of small oil droplets and particles concentration polarization and fouling during UF of a
from produced waters is required prior to reuse or soluble oil-surfactant-water emulsion through a poly-
discharge. meric membrane in a stirred filtration cell. They found
Since conventional-treatment technologies such as that fouling was due to adsorption of oil on the
gravity separators and coalescer plates cannot meet membrane structure. Lipp et al. [13] studied UF of
the high purity requirements for discharge or reinjec- soluble oil-in-water emulsions through a batch cell
tion, new or improved technologies have been inves- containing various polymeric and cellulosic mem-
tigated. Among them, membrane microfiltration (MF) branes. They stated that fouling followed a gel-polar-
has been used in a growing number of cases for ized, film-model behavior, with the oil droplets
successful treatment of oily wastewaters [5-7]. Sys- coalescing into the surface-fouling oil film. Koltunie-
tems have already been pilot-tested on offshore oil wicz et al. [14] analyzed and compared the perfor-
platforms and onshore facilities [3,8-10]. These tests mance of various ceramic and polymeric membranes
have shown that membrane systems can be successful operated in crossflow and dead-end filtration modes
in the field for treating produced water. during microfiltration of a n-dodecane/water emul-
Distinct advantages of membrane technology for sion. They observed that the membranes initially
treatment of produced water include reduced sludge, fouled internally (described by a pore-blocking
high quality of permeate, and the possibility of total model), and then external fouling began to dominate
recycle water systems. When considering these advan- (described by a cake-filtration model). Finally, Bha-
tages along with the small space requirements, mod- numurthy et al. [15] studied UF of cutting oil with
erate capital costs, and ease of operation, membrane polysulfone membranes. They presented the results of
technology provides a very competitive alternative to changing various operating variables, including the
conventional technologies. increase in flux observed at higher crossflow velocities
Although MF membranes can successfully treat due to washing away of the polarized layer. They also
produced waters, they experience a decline in perme- proposed a model for the prediction of the permeate
ate throughput or flux as a result of fouling. This flux flux which incorporates droplet coalescence and shear
decline is due to the adsorption and accumulation of rate.
rejected oil, suspended solids, and other components Our study was focused on understanding and char-
of produced water on the membrane surface (external acterization of the type of fouling occurring during
fouling) or in the membrane pores (internal fouling). MF of a model-produced water, containing crude-oil
This fouling can be irreversible or resistant to clean- droplets and, in some cases, suspended solids. Experi-
ing, making the original flux unrecoverable. Fouling mental results on flux decline and permeate quality for
can be reduced through the use of different or surface- ceramic and polymeric membranes are reported for a
modified membrane materials, various operating stra- variety of process- operating conditions. Simple mod-
J. MueUer et al./Journal of Membrane Science 129 (1997) 221-235 223

els have also been employed to help analyze the ment was calibrated for the Hueneme heavy crude oil
membrane-fouling process. emulsion used in the experiments.
Oil-droplet size distributions for the feed emulsions
were determined using a Coulter Multisizer. A hema-
2. Materials and methods cytometer slide and optical microscope were also used
to confirm these measurements. Results showed that
2.1. M o d e l - p r o d u c e d w a t e r a n d c h a r a c t e r i z a t i o n most of the droplets were 1-10 ~tm in diameter, with a
number-averaged droplet diameter of ,-~2~tm and
Synthetic-produced water was developed for use in volume-averaged droplet diameter of ~ 4 ~tm
this study. A heavy crude oil (API 12 weight, density (Fig. 1). Natural surfactants in the heavy crude oil
0.972 g cm-3), supplied by UNOCAL from the Hue- stabilized the emulsion sufficiently for the purpose of
neme field in California, was added at various con- these experiments, although some of the oil gradually
centrations to tap water. A blender (Osterizer Model came out of solution and formed a ring at the water
890-28M) mixed the oil and water at high shear rates level in the reservoir.
for ~ 2 min. Feed-emulsion samples were also centrifuged for
An Horiba (Model OCMA-220) oil-content analy- 60min at 1850rpm in 100ml pear-shaped, glass
zer was used to determine the total oil and grease centrifuge tubes using an International Portable Cen-
content of the feed and the permeate. The Horiba trifuge Model IPC-2. Under these conditions, all
analyzer uses tetrachloroethylene to extract oil and droplets >0.5 ~tm in diameter are predicted to be
grease from water, and then measures absorbance at a removed. Starting with a feed concentration of
single IR wavelength, 2930 cm -1 [16]. The instru- 250 ppm total hydrocarbon, the concentration remain-

1600 2.5
N u m b e r Distribution
1400

2.0 A
1200 E
I i tvo,o.oo,= , =,oo Q.
i._

J~ Q.
E 1000 O
:3 1.5 E
t--
m 800 O
>
o
I,,.
a 600 1.0 --~
Q.
2
a
400
0.5
200

0.0
0 5 10 15 20 25 30

Droplet size (ixm)

Fig. 1. Oil-dropletnumber (solid curve) and volumeversus diameter (shaded histogram) size distributions for the heavycrude oil-in-water
emulsion.
224 J. Mueller et al./Journal of Membrane Science 129 (1997) 221-235

ing in solution after centrifugation was ~20 ppm. The at pH levels typical of produced water (pH 6.6-7.8)
remaining concentration is due to soluble hydrocar- [17]. The PAN membrane has a nominal pore diameter
bons and finely dispersed, submicron-sized droplets. of 0.1 ~tm, and its surface layer is 150 ~tm thick. It also
The viscosity of the crude oil was measured using a has a tubular geometry, with an i.d. of 2.1 cm, a useful
rotating viscometer (Brookfield Model DV-1 with the length of 27.9 cm, and a surface area of 185 cm 2.
Brookfield Model 74R temperature controller). The
measured viscosity of the heavy crude oil is 2.3. Experimental apparatus
5.3 g cm -1 s -1 at 40°C, and 135 g c m -1 s -1 at 23°C.
Fig. 2 shows the apparatus constructed for the
2.2. Membranes experiments. It is capable of operating in a variety
of flow patterns, temperatures, and pressures. A gear
Two a-alumina ceramic membranes marketed by pump (Liquiflow Model CF8M) circulated the feed
the Membralox division of US Filter, and a surface- emulsion from the jacketed, temperature-controlled
modified polyacrylonitrile (PAN) membrane marketed glass-feed reservoir through the vertically aligned
by Zenon Environmental, were used. The two ceramic membrane unit. The retentate was returned to the feed
membranes are of 0.2 and 0.8 ~tm nominal pore dia- reservoir. A diaphragm valve was used to control the
meters, and each has 35% porosity and an asymmetric pressure in the system. An electronic balance inter-
surface layer which is 4-5 lam thick. The ceramic faced with a personal computer was used to collect
membranes have a tubular geometry, with an i.d. of permeate mass versus time data, and the computer
0.7 cm, a useful length of 20.6 cm, and a surface area controls a regeneration pump simultaneously which
of 45.3 cm 2. The membranes carried a negative charge maintains a constant liquid level in the feed vessel.

retentate line
~? _| ~?,

1
diaphragm
membrane ] valve
module
backflush line
Ii

permeate backflush
. make-up line line reservoir
jacketed
feed vessel
permeate
vessel | ,
level
, , control
I---q ~ [1circuit
--~

temperature make-up microbalance I


computer I
controller water
reservoir
Fig. 2. Produced water experiment apparatus.
J. MueUer et al./Journal of Membrane Science 129 (1997) 221-235 225

The feed solution was changed every 30 rain to pre- numbers are 2500 and 7600, respectively, for the
vent possible depletion of oil concentration due to ceramic and polymer membranes.
coalescence and deposition on the side of the reservoir. The apparatus containing the ceramic membranes
A backflush line was included to aid membrane clean- was cleaned after each experiment with tap water, then
ing and for future studies of backflushing and back- with 0.2 wt% sodium hydroxide, and finally with 1.0%
pulsing. nitric acid at 40°C. For the PAN membrane, tap water,
then an anionic caustic detergent (pH 12.1, Zenon),
2.4. Experimental procedures and finally citric acid (pH 1.4), were used after each
experiment. After the wash cycle, the PAN membrane
The experiments were repeated three times for each was removed and soaked in the anionic caustic deter-
set of conditions. Tap water was run through the gent followed by water for several hours as a final
membrane for the first 1500 s, followed by the cleaning step. The method of using an alkaline-acidic-
crude-oil emulsion for 7200 s, and then by a second alkaline wash cycle was found to be the most effective
water flux for an additional 1300 s. A single ceramic cleaning method by us and by others [18] for similar
membrane was reused (after cleaning) for all experi- situations involving organic fouling.
ments for the corresponding membrane type. One
polymeric membrane was used for the first three sets 2.5. Determination o f foulant amount
(11-13) of conditions, and a second one was used for
the other two sets (14 and 15). Experimental condi- The fouling layers were too thin to be determined
tions used in testing the three membranes are given in accurately by weighing the fouled and unfouled mem-
Table 1. Baseline or 'benchmark' conditions were branes. Instead, organic solvent was used to extract the
10psig (69.4kPa) transmembrane pressure, 40°C, oily foulants from the membranes. After an experi-
2 5 0 p p m heavy crude oil in the feed, and a ment, the membrane was removed from its module
0.24 m s - I mean crossflow velocity. The Reynolds and immersed in tetrachloroethylene. The oil was

Table 1
Experimental conditionsused in testing three microflltrationmembranes
Experiment number Description Temperature Pressure Flow-rate Velocity Coil.feed
(°C) (psig) (ml min- 1) (m s- 1) (ppm)
0.8/am ceramic
1 Baseline 40 10 549 0.24 250
2 High zS,P 40 20 549 0.24 250
3 High flow 40 10 2094 0.91 250
4 High conc. 40 10 549 0.24 1000

0.2 ~tm ceramic


5 Baseline 40 10 549 0.24 250
6 High AP 40 20 549 0.24 250
7 High flow 40 10 2094 0.91 250
8 High conc. 40 10 549 0.24 1000
9 SS 40 10 549 0.24 250
10 SS, high flow 40 10 2094 0.91 250

0.1 ~tm PAN


11 Baseline 40 10 4929 0.24 250
12 High Ap 40 20 5036 0.24 250
13 Low T 24 10 4929 0.24 250
14 High conc. 40 10 4929 0.24 1000
15 SS 40 10 4929 0.24 250
226 J. Mueller et al./Journal of Membrane Science 129 (1997) 221-235

partitioned into the known volume of solvent, and the 3.1. Fouling of the 0.8 gm ceramic membrane with
solvent was placed in the Horiba analyzer for the produced water
measurement of its oil content.
The 0 . 8 g m ceramic membrane experienced a
severe flux decline during the baseline experiments
3. Experimental results and discussion at 10 psig transmembrane pressure, 40°C, 250 ppm
heavy crude oil, and a 0.24 m s -1 mean crossflow
Water-flux experiments were run on each mem- velocity (see Fig. 4 and Table 2). A nearly steady flux
brane before testing began. Results at 10 psig trans- of only ,-~30 kg m -2 h -1 was achieved after 7200 s of
membrane pressure are shown in Fig. 3. The 0.8 gm oil-emulsion filtration. The second water flux, run
ceramic membrane experienced a significant water- after fouling by the oil emulsion, showed no recovery
flux decline of 50% in 1.5 h, while the PAN membrane of the flux. This implies that the fouling layer is
experienced a 35% water-flux decline, and the 0.2 ~tm strongly adsorbed and resistant to shear. However,
ceramic membrane experienced an 18% water-flux the permeate quality was excellent, with < 1 ppm total
decline, during the same time. The water-flux decline hydrocarbons.
may be attributed to the small particles in the tap After completing the baseline experiments, the
water, any residual oil which still remains in the effects of three operating variables were investigated:
apparatus, and the formation of microscopic bubbles transmembrane pressure, oil concentration, and cross-
on the permeate side of each membrane due to the flow velocity. The results shown in Table 2 are the
depressurization which occurs as water passes through mean values plus and minus one standard deviation.
the membrane. The greater water-flux decline for the The curves in Fig. 4, and in the subsequent figures, are
0.8 gm ceramic membrane may be due, at least in part, the averages of the flux versus time data for the
to its faster permeate flow carrying more micro-con- repeated experiments. The initial flux values in
taminants to the membrane. Table 2 are those observed at the end of the 1500 s

1600

1400

1200

~1000
~ 0.8~I.Lm ceramic membrane
E
800
x

,T 600
0.2 pm ceramic membrane

4O0

200 0.1 pm PAN membrane

0 I I I I I

0 1000 2000 3000 4000 5000 6000


Time (seconds)
Fig. 3. Comparison of water flux decline at 40°C and 10 psig transmembrane pressure for the three different membranes
J. Mueller et al./Journal of Membrane Science 129 (1997) 221-235 227

2500

2000
, High Pressure

1500 L H
igh Crossflow Velocity

lOOO
,-r
~ne
5O0 • . •

0 2000 4000 6000 8000 10000 12000


Time (seconds)
Fig. 4. Flux-decline curves for the 0.8 ktm c~-alumina ceramic membrane and heavy crude oil-in-water emulsion at various conditions.

Table 2
Summary of results for the three tested microfiltration membranes - the results are shown as the average of three repetitions, plus and minus
one standard deviation, with +, - and o representing positive, negative, and insignificant differences from baseline results at the 90%
confidence level

Experiment number Initial flux Final flux Flux change Coil, ee~a Con, p~rmeate Oil removal
(kg m -2 h -1) (kg m -2 h 1) (%) (ppm) (ppm) (%)
0.8 ~tm ceramic
1 6784-51 334-6 -95.1 250 0.35:0.2 99.9
2 998+500 40±22 (o) -95.6 250 0.64-0.3 (o) 99.7
3 8004-128 46+6 (+) -94.2 250 1.8+0.5 (+) 99.3
4 471+15 2 6 i l l (o) -94.5 1000 1.4d:0.2 (+) 99.4

0.2 ~tm ceramic


5 525+50 424-19 -91.9 250 3.7+1.9 98.5
6 6574-156 21±2 (o) -96.8 250 5.1+0.8 (o) 98.0
7 2114-19 324-13 (o) -84.8 250 4.64-0.2 (o) 98.2
8 3014-52 25±6 (o) -91.7 1000 5.84-1.0 (+) 99.4
9 305 312 +1.6 250 5 98.0
10 281 577 +105.4 250 5.4 98.5

0.1 ktm PAN


11 2044-27 344-3 (o) -83.3 250 1.84-0.6 99.3
12 2874-30 32+3 (o) -88.9 250 0.54-0.2 ( - ) 99.8
13 1635:5 314-3 (o) -81.0 250 2.84-1.4 (o) 98.9
14 4384-30 6.94-2 ( - ) -98.4 1000 0.94-0.3 (o) 99.9
15 455±30 226 -50.3 250 0.9 99.7
228 J. Mueller et al./Journal of Membrane Science 129 (1997) 221-235

800

¢-
700

600

500
~ ressure

High Oil Concentration


400

300

u. High Crossflow Velocity


200

100

0 I I I I I

0 2000 4000 6000 8000 10000 12000

Time (seconds)

Fig. 5. Flux decline curves for the 0.2 I.tm alpha-alumina ceramic membrane and heavy crude oil-in-water emulsion at various condition

periods of pure-water flux; the large variations of the nificantly different at the 90% confidence level. The
initial flux indicate that the cleaning procedures do not higher crossflow velocity ( 0 . 9 1 m s -1) did not
yield a consistent state of the membrane. The higher improve the final flux. In each of these tests, the
pressure gave a higher initial flux, but the final flux permeate oil concentration remained below 6 ppm.
after oil-emulsion filtration (~40 kg m -2 h -1) is not Fig. 5 shows that the initial membrane permeability
significantly higher than that of the baseline experi- changed after the baseline experiments. An irreversi-
ment. The higher crossflow velocity (0.91 m s -1) ble flux decline is the reason for the difference
resulted in a small increase in the final flux, whereas between the starting water fluxes for the baseline
the increased oil concentration (1000 ppm) resulted in experiments and for the other experiments. The clean-
a slightly reduced final flux. In each of these tests, the ing procedure was unable to remove all of the foulants.
permeate oil concentration remained below 2 ppm. Stronger acid, base, and detergent washes also failed
to fully recover the initial clean-membrane water flux.
3.2. Fouling of the 0.2 gm ceramic membrane with However, the starting water fluxes or state of the
produced water membrane do not appear to significantly affect the
final fluxes for the oil emulsion.
The 0 . 2 g m ceramic membrane experienced a
severe flux decline during the baseline experiment 3.3. Fouling of the 0.1 ~am PAN membrane with
(Table 2 and Fig. 5). A nearly steady flux of produced water
,-~40 kg m -2 h -1 was achieved during filtration of
the oil emulsion, and no recovery of flux was observed The PAN membrane experienced a large flux
when the feed was switched back to water. A higher decline during the baseline experiments (Table 2
transmembrane pressure (20 psig) resulted in the most and Fig. 6). A nearly steady flux of ~30 kg m -2 h -1
fouling and the lowest final-flux value. The higher oil was achieved by the end of the 7200 s of oil-emulsion
concentration (1000 ppm) also resulted in a low final filtration. Again, the second water flux, run after
flux. In all cases, the final flux values are not sig- fouling by the oil emulsion, gave no recovery of flux.
J. Mueller et al./Journal of Membrane Science 129 (1997) 221-235 229

600

500
High Concentration

" ~ 400
gh Pressure
E
300
,-I
X

i, 200

1O0

0
0 2000 4000 6000 8000 10000 12000

Time (seconds)
Fig. 6. Flux-decline curves for the 0.1 ~tm PAN membrane and heavy crude oil-in-water emulsion at various conditions.

The final flux values were not changed significantly by ever, this membrane exhibited the quickest fouling
increasing the transmembrane pressure or lowering after the oil emulsion was introduced into the mem-
the temperature. However, the PAN membrane fouled brane system. Its total transmembrane flux dropped
very badly at a higher feed-oil concentration below that of the 0.2 ~tm ceramic membrane after only
(1000 ppm), and a thick external oil layer was evident a few minutes of oil-emulsion filtration. The flux
upon examination of the membrane. In all cases, the decline for the 0.2 ~tm ceramic membrane was much
permeate contained <3 ppm total hydrocarbons. From slower. Even though the three membranes have sig-
the initial flux values in Table 2, and noting that one nificantly different initial fluxes, the final fluxes after
membrane was used for experiments 11-13 and a 2 h of oil-emulsion filtration are remarkably similar
second one for experiments 14 and 15, it is apparent (30-40 kg m -2 h -1) and much lower than the clean-
that the variations in the initial membrane permeabil- membrane fluxes. This indicates that the primary
ity between the five different membranes used in the resistance is provided by the fouling layer formed
five different sets of experiments is greater than the on or in the membrane, and not by the membrane
variations in the initial membrane permeability of a itself. The fact that the permeate concentration does
given membrane reused after cleaning for each repeti- not generally increase with membrane pore size pro-
tion in a given set of experiments. vides further support for the filtration being controlled
by the fouling layer rather than the membrane.
3.4. Comparison of results for the three different
membranes 3.5. Effects of suspended solids

The results for the three different membranes under In order to simulate produced water, containing
baseline conditions are shown in Fig. 7. The 0.8 jam suspended solids as well as oil droplets, 250 ppm
ceramic membrane has the highest original water flux diatomaceous earth was added to tap water along with
(1600 kg m -2 h-l), due to its high permeability. How- the 250 ppm heavy crude oil. The resulting suspension
230 J. Mueller et al./Journal of Membrane Science 129 (1997) 221-235

1400

1200

1000
ceramic membrane
800

600
E
~ . . ~ ~ 0 .'1~ llm PAN membrane
400

200 ~ ~ . . ' ~ g m ceramicmembrane


II I II
I I , I i'_l I I
0 I
0 2000 4000 6000 8000 10000 12000
Time(sec)
Fig. 7. Comparisonofbaseline fluxes(10psig ~ansmembranepressure, 250ppm heavy oil, 40°C, and 0.24 m s -1) ~ r thet~eedifferent
membranes.

1000

900

800

70o nd Oil
600

500
g ,

E
300

200

100 ~

0
0 2000 4000 6000 8000 10000 12000

Time (seconds)
Fig. 8. Flux-decline curve for the ceramic membrane at baseline and high crossflow-velocity conditions with suspended solids for the heavy
crude oil-in-water emulsion at 250 ppm oil and 250 ppm diatomaceous earth.
J. Mueller et aL /Journal of Membrane Science 129 (1997) 221-235 231

500

\
200'--
F '-------

lOO

o I I I I I
0 2000 4000 6000 8000 10000 12000

TIME (seconds)

Fig. 9. Flux-decline curve for the PAN membrane at baseline conditions with suspended solids for the heavy crude oil-in-water emulsion at
250 ppm oil and 250 ppm diatomaceous earth.

was used with the 0.2 ~tm ceramic and 0.1 ~tm poly- 4. Modeling and characterization of membrane
meric membranes. The results for the 0.2 ktm ceramic fouling
membrane at 0.24 and 0.91 m s -1 crossflow velocities
are shown in Fig. 8. Even though the diaphragm valve In order to better understand the nature and mechan-
was specially designed to resist clogging, some sus- isms of membrane fouling by oily water, several
pended solids accumulated there and in other parts of methods of analysis and characterization were
the apparatus; this caused a jump in the transmem- employed. These include analyzing the flux-decline
brane pressure and flux upon switching from water data as total resistance versus time, comparing pre-
feed to emulsion feed. The transmembrane pressure dictions and measurements of the oil-fouling layer
then stabilized and relatively little flux decline was thickness, and comparing the water-contact angles for
observed. The final flux values are one order of the clean and the fouled membranes.
magnitude higher than the final flux in the absence
of suspended solids. 4.1. Total resistance versus time analysis
The results for the 0.1 ktm PAN membrane with
suspended solids are shown in Fig. 9. Again, the A resistance model is used to describe the transport
suspended solids reduced the membrane fouling. of solvent through the membrane due to transmem-
The final flux with suspended solids is about seven brane pressure, as described by Darcy's law:
times that of the baseline final flux in the absence of Ap Ap
suspended solids. J - - vpR~ - - v p ( g m -~-Re) (1)
It is thought that the suspended solids enhance the
flux in two ways. First, they act as a dynamic mem- where J is the permeate mass flux (mass of permeate
brane layer which prevents oil from fouling the mem- per unit time per unit membrane surface area), A P is
brane internally. Second, they adsorb the rejected oil the transmembrane pressure, Vp is the permeate kine-
droplets and break up the continuous coalesced oil matic viscosity, and g t is the total resistance to flow.
layer that otherwise may form on the membrane sur- The permeate viscosity may be taken as that of pure
face. water (6.6x 10 -6 m 2 s -1 at 40°C). The resistance to
232 J. Mueller et al./Journal of Membrane Science 129 (1997) 221-235

flow may be further described as the sum of individual times [20], and experimental results shown in the
resistances from the internally fouled membrane, R m , next section indicate that sweeping of the rejected
and from the external fouling or cake layer, R e . The oil layer to the filter exit was not significant for the
experimental data are used to determine whether experiments described here. This results in the follow-
internal or external fouling is dominant, as described ing expression for the total resistance as a function
previously [19]. of time [19]:
In the case of internal fouling, Rm increases with
Rt = R0(1 + 4 t K c ~ Q2)0.5 (4)
time. Flux decline due to internal fouling has been
described by two different models, a standard block- where K c ~ is a constant proportional to Re. Graphing
ing model and a pore blocking model. The standard resistance versus time, using this equation, will yield a
blocking model assumes that the number of pores curve that is concave down. Thus, the resistance
remains constant, but that each pore becomes con- increases with time with a decreasing slope for exter-
stricted due to deposition of the foulant on the pore nal fouling, and with an increasing slope for internal
walls in proportion with the permeate volume. This fouling This provides a simple means for identifying
results in the following expression for the resistance as the fouling mechanism.
a function of time [19]: The total resistance versus time curves for the three
membranes at the baseline-experiment conditions are
Rm = R0(1 + KSBMQ0 t) 2 (2)
shown in Fig. 10. The curve for the PAN membrane is
where Ro is the initial membrane resistance, Qo is the concave down, starting immediately after the first
initial volumetric permeate flow rate, KsBM is a con- water flux. This indicates that external fouling dom-
stant, and t is time. Plotting resistance versus time inates throughout the period of filtration of oily water.
using this equation will yield a curve that is concave For the ceramic membranes, the total resistance curves
up. are concave up for a short period after the start of oily-
The pore-blocking model assumes that the pore size water filtration, before becoming concave down. This
remains constant, but that the number of pores per indicates that the ceramic membranes initially foul
membrane unit surface area decreases due to blocking internally and later external fouling begins to dom-
or plugging, with the number of pores that become inate. It may be that the large pore sizes and perme-
plugged increasing in proportion with the permeate abilities of the ceramic membranes allowed for
volume. This results in an exponential decline in the penetration of some of the smallest drops into the
flux, or an exponential increase in the resistance [19]: membrane pores before they became sufficiently con-
stricted or plugged that external fouling became domi-
Rm = Ro e (KPBMt) (3)
nant, whereas the lower pore size and permeability of
where gpB M is a constant. The resistance versus time the polymeric membrane led to immediate external
curve is again concave up. fouling. During microfiltration of protein solutions,
The decline in the flux may also be due to external Tracey and Davis [ 19] also observed immediate exter-
fouling. Assuming only external fouling, R m remains nal fouling for small pore sizes, and internal followed
constant at its initial value, Ro. The external resistance, by external fouling for large pore sizes.
Re, is the product of a specific cake resistance, ke, and
a cake thickness, 6. In the cake filtration model, k~ is 4.2. Analysis of the fouling layer
assumed constant and ~ increases in proportion with
the permeate volume due to the deposition of foulants Examination of the membranes after selected base-
on the top surface of the growing cake layer. This line experiments showed very thin oil layers adsorbed
model was developed for dead-end filtration; in cross- to the membrane surfaces. Apparently, a phase inver-
flow filtration, the rejected foulants may also be swept sion occurred at the membrane surface during filtra-
to the filter exit rather than continuing to accumulate tion, with oil as the continuous phase and water as the
on the membrane surface. However, a recent theory discrete phase which passed through the oil layer in
predicts that cake build-up for crossflow filtration small drops or globules. The oil layers made the
is described by dead-end filtration theory for short membrane surfaces hydrophobic. For the 0.2 ~m cera-
.L Mueller et al./Journal of Membrane Science 129 (1997) 221-235 233

14

12
0.1 gm PAN membrane

,-
04
10
o

¢ 8
tj
c

._¢ 6 2 gm ceramic membrane

4
O
I-
0.8 gm ceramic membrane
2

0 2000 4000 6000 8000 10000 12000


Time (seconds)
Fig. 10. Total resistance versus time curves for the three different membranes at baseline conditions.

mic membrane, the water droplet-membrane surface brane, respectively, corresponding to consolidated
contact angle increased from 50 ° to 60 ° after fouling thicknesses of 66 and 32 gm. The good agreement
the membrane under baseline conditions. The contact between the measured and predicted values indicate
angle increased from 50 ° to 70 ° for the PAN mem- that the rejected oil remained associated with the
brane. membrane and did not experience any significant flow
Soaking the membranes in tetrachloroethylene to to the filter exit under the action of shear forces from
remove the oil, followed by IR analysis, yielded oil the imposed axial flow. This result is substantiated by
masses of 0.274-0.05 g for the 0.2 gm ceramic mem- the earlier observation that no flux recovery was
brane and 0.54±0.14 g for the 0.1 gm PAN mem- observed during the second pure-water flux, which
brane. If it is assumed that the absorbed oil formed followed the 7200 s period of oily-water filtration. The
uniform, consolidated layers, then these amounts cor- stagnant nature of the absorbed oil layer is related, at
respond to thicknesses of 60±10 and 3 0 ± 1 0 g m , least in part, to its high viscosity and small thickness.
respectively. These results may be compared with a In addition, any oil deposited internally within the
simple model which assumes that all of the permeate membrane pores would not be able to flow.
that passes through the membrane deposits its oil on or
within the membrane, and that no oil passes through
the membrane or flows along its surface to the filter 5. Concluding remarks
exit. This situation is analogous to dead-end filtration
with complete rejection. Using the measured permeate Each of the ceramic and polymeric microfiltration
volumes collected during the 7200 s of oily-water membranes always produced a high-quality permeate,
filtration, and the known feed concentration of containing <6 ppm total hydrocarbons, starting with a
250 ppm oil, this yields predicted absorbed oil masses synthetic-produced water, containing 250-1000 ppm
of 0.29 and 0.57 g under baseline conditions for the heavy crude oil. Since centrifugation tests indicated
0.2 ~tm ceramic membrane and the 0.1 gm PAN mem- that ~20 ppm of the oil in the feed was in the form of
234 J. Mueller et aL /Journal of Membrane Science 129 (1997) 221-235

submicron-sized droplets or soluble hydrocarbons, the Acknowledgements


results indicate that the fouled membranes are able to
remove a substantial portion of these finely dispersed This research was supported by the US Bureau of
or dissolved compounds. Typical final-flux values Reclamation, the US Environmental Protection
measured after 2 h of filtration for 250 p p m oil in Agency, and the Center for Separations Using Thin
the feed are of the range ~ 3 0 - 4 0 kg m -2 h - t , more Films at the University of Colorado, Boulder. The
than a factor of 10 lower than the initial flux. Increased authors wish to thank Jennifer Bieker (supported by
feed concentrations decreased the final flux, but cross- the Colorado Commission on Higher Education) and
flow rate, temperature, and transmembrane pressure Nikki Sikes (supported by the National Science Foun-
had relatively little effect on the final flux. The addi- dation's Research Experiences for Undergraduates
tion of suspended solids to the feed significantly Program) for their help with the experiments, and
increased the final flux. It is thought that the suspended Jim Howard for determining the hydrocarbon content
solids adsorb the oil, break up the oil layer, and act as a after centrifugation. Appreciation is also extended to
dynamic membrane which reduces the fouling of the Cliff Juengst of Unocal for providing oil samples and
underlying membrane. Steve Kelley of N R E L for viscosity determination.
Total resistance versus time curves from the flux-
decline data were used to identify fouling mechan-
isms. The 0.2 and 0 . 8 ~ m ceramic membranes
appeared to exhibit internal followed by external References
fouling, while external fouling appeared to dominate
[1] R.L. Arscott, New directions in environmental protection in
the behavior of the 0.1 ~tm PAN membrane from the oil and gas operations, J. Pet. Technol., April (1989) 336-342.
start. [2] J. Taylor, R. Larson and B. Scherer, Treatment of Offshore
The fouling-layer thicknesses measured experimen- Produced Water: An Effective Membrane Process, reprint of
tally are in good agreement with those predicted by a paper presented at EnviromnentalNorthern Seas International
mass balance based on the assumption that all of the Conference and Exhibition, Norway, August 1991.
[3] J.P. Ray and ER. Engelhardt (Eds.), Produced Water, Plenum
oil originally associated with the volume of permeate Press, New York, 1992.
collected remained on or in the membrane. The good [4] A. Bevis, The treatment of oily water by coalescing, Filtr.
agreement indicates that the rejected oil did not flow Sep., July/August (1992) 295-301.
along the membrane to the filter exit, presumably [5] R.H. Bhave and H.L. Fleming, Removal of oily contaminants
because the heavy crude oil is extremely viscous or in wastewater with microporous alumina membranes, New
Membrane Materials, AIChE Symp. Ser. No. 261, 84 (1988)
sticky. Furthermore, no increase in flux was observed 19-27.
after the oily-water feed was replaced with pure-water [6] A. Benedek, E A. Tonelli, Innovative technologies for
feed. treatment of oily wastewater, Iron and Steel Engineer, June
Comparing the behavior of the 0.1 ~tm PAN mem- (1992) 44-50.
brane and the 0.2 ~tm ceramic membrane under base- [7] Koch Membrane Systems, Oil Recycler Increases Efficiency
with WastewaterTreatment Technology,Hazmat World, May
line conditions leads to an interesting observation. The 1992.
0.2 ~tm ceramic membrane is more permeable and [8] A. Zaidi, K. Simms, S. Kok and R. Nelson, Recent advances
exhibits a higher flux than does the 0.1 ~tm polymeric in the application of membrane technology for the removal of
membrane. As a result, twice as much rejected oil per oil and suspended solids from produced water, in J.P. Ray and
unit membrane surface area is deposited on it during ER. Engelhart (Eds.), Produced Water, Plenum Press, New
York, 1992.
the 7200 s of oily-water filtration. However, the final [9] R.J. Lahiere and K.P. Goodboy,Ceramic membrane treatment
total resistance is lower for the ceramic membrane of petrochemical wastewater, Environ. Progr., 12 (1993) 86-
than that for the polymeric membrane. This implies 89.
that the specific resistance of the rejected oil from the [10] J.C. Yang and J.P. Meyer, Industry's Field Experience with
polymer membrane is more than two-fold higher than Membrane Filtration Technology, Report prepared by Mara-
thon Oil Company and ARCO Oil and Gas Company, Piano,
that from the ceramic membrane. Apparently, the Texas, June 1991.
morphology of the fouling layer is strongly affected [11] D. Bhattacharyya, A.B. Jumawan and R.B. Grieves, Ultra-
by the membrane material and morphology. filtration characteristics of oil--detergent-water systems:
J. Mueller et al./Journal of Membrane Science 129 (1997) 221-235 235

Membrane fouling mechanisms, Sep. Sci. Tech., 14 (1979) [16] Horiba Oil Content Analyzer, OCMA-220, Product bulletin,
529-549. 1994.
[12] S. Lee, Y. Aurelle and H. Roques, Concentration polarization, [17] K.M. Simms, S.A. Zaidi and C.M. Tam, Ongoing Evaluation
membrane fouling and cleaning in ultrafiltration of soluble of MF and UF Membranes for Produced Water Treatment,
oil, J. Membrane Sci., 19 (1984) 23-38. Produced Water Seminar, American Filtration Society, Texas
[13] P. Lipp, C.H. Lee, A.G. Fane and C.J.D. Fell, A fundamental Chapter, January 1992.
study of the ultrafiltration of oil-water emulsions, J. [18] J. Lindau and A-S. Jonsson, Cleaning of ultrafiltration
Membrane Sci., 36 (1988) 161-177. membranes after treatment of oil waste water, J. Membrane
[14] A.B. Koltuniewicz, R.W. Field and T.C. Arnot, Crossflow and Sci., 87 (1994) 71-78.
dead-end microfiltration of oily-water emulsion. Part I: [19] E.M. Tracey and R.H. Davis, Protein fouling of track-etched
Experimental study and analysis of flux decline, J. Membrane polycarbonate microfiltration membranes, J. Colloid Interface
Sci., 102 (1995) 193. Sci., 167 (1994) 104-116.
[15] T. Bhanumurthy, S.K. Gautam and S. Basu, Separation and [20] R.H. Davis, Modeling of fouling of crossflow microfiltration
recycling of oily industrial effluents: A study with cutting oil, membranes, Sep. Purif. Methods, 21 (1992) 75-126.
ITT Bombay, personal communication, 1994.

Vous aimerez peut-être aussi