Iron oxide adsorption

Transcription

1 FILTRATION Iron oxide adsorption and UF to remove NOM and control fouling Yujung Chang and Mark M. Benjamin M This study applied a novel single-fiber membrane module to study fouling of ultrafiltration (UF) membranes by natural organic matter (NOM) molecules, and used a combined iron oxide UF process to reduce that fouling. Addition of iron oxide particles to a UF system can significantly increase NOM removal efficiency, because NOM molecules that would otherwise pass through the membrane sorb to the oxides and are retained. Furthermore, despite the fact that iron oxide particles cannot selectively adsorb foulant molecules, their presence in the system can reduce the membrane fouling dramatically. This effect is mediated through two processes: a decrease in the NOM concentration in the circulation loop because of sorption and the formation of iron oxide cake layer deposits on the membrane surface. In such a system, the condensed layer of NOM forms on top of the cake, protecting the underlying membrane surface. Addition of iron oxide particles to a UF system can significantly increase NOM removal efficiency, slowing membrane fouling. embrane technologies, including microfiltration (MF), ultrafiltration (UF), nanofiltration (NF), and reverse osmosis (RO), have recently been extensively investigated for water treatment. In addition to removing virtually all particles larger than the nominal pore size, some membrane processes (e.g., NF and RO) reject significant amounts of soluble species and are therefore prospective technologies for removing natural organic matter (NOM). 1 3 However, these membranes are also more easily fouled by NOM than are membranes that have larger pores. 4,5 Thus, both the NOM that passes through the membrane and the NOM that is rejected by the membrane are problematic, because of possible downstream disinfection by-product formation and membrane fouling, respectively. 74 JOURNAL AWWA

2 UF is a low-pressure membrane process used primarily for particulate removal in drinking water treatment. Because the pores in ultrafilters are usually in the.1.1-µm size range, NOM removal by such filters is usually slight, typically 5 to 25 percent. 6,7 Even so, ultrafilters often suffer significant fouling when used to treat drinking water supplies. This article describes an integrated system that combines UF with iron oxide adsorption to accomplish good NOM removal with minimal fouling. The authors also present a novel experimental setup that can be used to obtain direct visual information about the membrane fouling process. FIGURE 1 Epoxy resin Specific goals of the research were to (1) develop and demonstrate use of a single-fiber membrane (SFM) module to study the mechanisms and kinetics of membrane fouling and the effectiveness of different membrane cleaning schemes, (2) evaluate the performance of a combined adsorption UF process for NOM removal, (3)study the effect of the deposition of adsorbent particles on the hydraulic resistance across the membrane and the interaction of the membrane with NOM, (4) identify the membrane fouling potential of different fractions of NOM, and (5) evaluate the fouling reduction that can be accomplished by removing a portion of the NOM from solution via adsorption onto particulate adsorbents. Background Behavior of NOM in membrane systems. In pressure-driven membrane systems, water is driven through the membrane in response to a cross-membrane pressure gradient. NOM molecules and other contaminants are brought to the membrane surface by this flow. When a portion of the NOM is rejected by the membrane, it remains near the membrane surface, so its concentration is higher near the surface than in bulk solution. This phenomenon is called concentration polarization (CP), and the region where it occurs is called the CP layer. The increased NOM concentration in the CP layer Schematic diagram of a single-fiber membrane module T-connector Permeate Membrane fiber Surface analysis Ultrafilters often suffer significant fouling when used to treat drinking water supplies. Inlet for backwash water Transparent vinyl plastic ;;; increases the likelihood that NOM molecules will foul the membrane by adsorbing in the pores or on the membrane surface or by depositing as a thin film on the surface. 4,8 11 The thickness of the CP layer and the contaminant concentration in that region can be reduced by inducing a flow and a corresponding shear force along the membrane surface; this operation is called cross-flow filtration. The adsorption of NOM onto membrane surfaces is controlled by the chemical characteristics of the NOM molecules and the membrane surface. Such adsorption is often at least partially irreversible. Factors favoring adsorption of solutes onto membranes include hydrophobicity of the solute (which encourages its exclusion from the aqueous phase and its accumulation adjacent to hydrophobic membranes), specific chemical interactions between the solute and the membrane, and attractive electrostatic interactions. For instance, Matthiason 4 reported that less bovine serum albumin (BSA) protein adsorbs on cellulose acetate than on polysulfone UF membranes and that the amount of BSA adsorbed on either membrane is proportional to the BSA concentration in solution and is strongly ph dependent. Jucker and Clark 11 also reported that the adsorption of humic substances on membrane surfaces increased as the ph decreased. NOM molecules are highly heterogeneous, and most such molecules probably contain both hydrophilic and hydrophobic parts. It is generally observed that fouling by NOM is more severe on hydrophobic than on hydrophilic membranes. 12 If the solubility of the NOM is exceeded in the CP layer, the NOM may form a new phase on the mem- DECEMBER

3 FIGURE 2 A Static mode Membrane module ;;;;;;; B Circulating mode Three operational modes used in experiments Lumen of membrane fiber Effluent ;;;;;;; Effluent C Adsorption UF mode ;;;;;;; ;;;;;;; Effluent Influent Influent Adsorbent particles Influent brane surface, thereby increasing the hydraulic resistance across the membrane and limiting the permeate flux. Although the exact form and composition of the condensed, NOM-containing phase cannot be predicted, it is thought to be a gel. (This article presents evidence for such a form and refers to the phase as a gel.) Unfortunately, the solubility limit for NOM is not well established and undoubtedly depends on the nature of the NOM as well as other components of solution. As a result, the extent of such fouling is difficult to predict. Lahoussine-Turcaud et al 9 reported that UF of a humic acid solution (1 mg/l dissolved organic carbon [DOC]) using hollow-fiber UF membranes with a pore size of 1 nm caused the flux to decrease by 55 percent during 3 min of treatment. They suggested that backtransport of humics away from the mem- FIGURE 3 brane (by diffusion) was not great enough 25 to overcome convective transport toward the membrane, so a layer of humic material accumulated near the membrane sur- 2 face. They also observed a 25 percent 15 decrease in flux within the first 5 min of operation and suggested that this decrease is due to an operationally irreversible deposition of humic material on the surface of the hollow fiber and in the membrane pores. 5 Controlling membrane fouling by adding adsorbent particles. Because NOM can cause irreversible membrane fouling, the removal of NOM upstream of a membrane unit, e.g., by adding metalbased coagulants, powdered activated carbon (PAC), or both to the water, is likely to reduce fouling and prolong membrane life. If the resulting suspension is then treated in a membrane process, the particulates in the feedwater (including both the adsorbent particles and microorganisms) can be rejected almost completely. 7,13,14 Adsorption membrane filtration combinations can also dramatically reduce the space and residence time requirements for solid liquid separation, because adsorption and solid separation are accomplished simultaneously in the compact membrane module. Although coagulation by Al(III) or Fe(III) salts is able to remove some NOM, such a process does not necessarily prevent membrane fouling and may exacerbate it. 14 On the other hand, several researchers 5,13,15,16 have reported that the addition of PAC to the influent in MF or UF processes significantly increases NOM removal efficiency and reduces membrane fouling, although there has been at least one report of a system in which PAC particles formed a cake layer on the membrane that increased the overall hydraulic resistance. 7 PAC regeneration is difficult and expensive, necessitating disposal of the PAC sludge. 17 Jacangelo et al 7 reported that the residual adsorbable NOM from granular activated carbon membrane processes could still cause an appreciable decrease in permeate flux. Similar results were reported by Lahoussine-Turcaud et al; 9 even when as Typical permeate flux decline in static operation mode Operation Time h TOC 4.5 mg/l, ph 7., pressure 9 kpa (13 psi) 76 JOURNAL AWWA

4 much as 7 percent of the NOM was removed from a surface water by coagulation with ferric chloride, the remaining 3 percent of the NOM reduced the permeate flux by 63 percent during 3 min of operation. Thus, although cross-flow operation, addition of alum, ferric chloride, or PAC, and frequent backwashing may slow the fouling process, the growth of NOM film on the UF membrane surface during long-term operation remains problematic. The adsorbents used in this study were iron oxide particles (IOPs) prepared in a way that minimized their tendency to foul the membranes. The major potential advantages of using IOPs over PAC in an adsorption UF process are that iron oxide is cheaper than PAC and that NOM can be desorbed from the surface of iron oxides simply by adjusting the ph, 18 thereby regenerating the adsorbent. SFM module. In efforts to understand membrane fouling, surface analysis techniques such as scanning electron microscopy (SEM) have been used to provide a visual image of the membrane surface and the fouling layer. Because the membrane sampling procedure for such analyses is destructive, the conventional multiple-fiber membrane module has an inherent limitation the analyses can only be done either at the beginning of the test (using a fresh membrane sample) or at the end of the test (using a fouled membrane). Analysis of membrane fouling at different stages requires using membrane samples from different modules, increasing the cost of the study and generating ambiguity about the results. In order to obtain semicontinuous information about flux decline and the physical state of the membrane surface on the same membrane fiber, an SFM module was designed (Figure 1) and used in this study. With this system, a piece of membrane can be cut off the fiber for analysis at any time and the rest of the fiber replaced in a shorter module to continue the operation. As a result, information about the fouling history and the effectiveness of different membrane cleaning schemes can be obtained from the same membrane fiber. Along with information about the permeate flux decline or pressure buildup, this technique can be used to diagnose the cause of changes in membrane performance. Evidence presented elsewhere19 indicates that the sampling process does not alter the membrane performance (the specific flux is identical less than 2 percent of deviation before and after removing a section for analysis) and that the foulant is evenly distributed across the length of the membrane fiber, so that a membrane sample taken from either end of the module is representative of the entire membrane. Another benefit of using an SFM module is that the water sample consumption can be minimized. DECEMBER

5 FIGURE 4 FIGURE 5 Fraction of Original Flux percent 4.7 mg/l DOC Raw water Schematic diagram of serial filtration test and three membrane modules in series First run ;;;; Second run ;;;;; 4. mg/l DOC Run First Second Third Influent TOC mg/l Third run Because of the high production rate of the UF process (on the order of L/m 2 /h), the consumption of water is relatively large for a bench-scale study. For example, a test using a 5-mm (2-in.) module that contains 2 fibers may require 5 7 L/d of water sample. Because filtration tests might last from several days to several months, this consumption rate requires either extensive storage space or frequent sampling trips. To overcome this difficulty, 4. mg/l DOC 4. mg/l DOC Permeate flux curves for the three membranes in serial filtration test Time h Static mode, ph 7., pressure 9 kpa (13 psi) some researchers 5 recycle both permeate and concentrate in a container when conducting UF tests in the laboratory. But because the amount of foulant in the container is expected to decrease steadily (as foulant accumulates in or on the membrane), this approach might not reflect realistic operating conditions. With the SFM module, the consumption of water can be reduced to a small fraction of that required for multiple-fiber modules. SEM of membrane surfaces. SEM has been used to provide visual information about fouled and unfouled membranes by several researchers For example, Fane et al 23 examined colloids in the fouling layer of a microfilter and found that at the isoelectric point of the particles, fouling occurred only at the surface of the membrane, whereas both surface and internal deposition occurred when the solution ph drifted away from the isoelectric point of the colloids. The beam energy of conventional SEM is so strong, however, that it can damage membrane surfaces when an image of higher magnification is attempted. In this study, field emission SEM (FESEM), which uses much lower beam energy, was used to provide images of the internal surface and the cross section of the membrane. Material and methods Water samples. The NOM-laden water used for this study was collected from Lake Pleasant, a peat lake 24 km (15 mi) north of the University of Washington at Seattle. The lake water has a ph between 7.5 and 8., has low ionic strength and algal content, and is highly colored because of the presence of humic substances. The DOC concentration of the lake water measured approximately 2 3 mg/l during the course of study. Tests using this water were designed to investigate the behavior of soluble NOM in the adsorption UF process, so particulate matter was removed from the samples by 78 JOURNAL AWWA

6 After 48 h of operation, SEM images from the serial filtration test show the first membrane (top left), second membrane (top right), and third membrane (bottom). prefiltration through sequential 25-µm and 1-µm-over-.45-µm nylon filters* immediately after sampling. Samples were then stored at 4 o C. Adsorption tests using a nonionic commercial resin suggest that about 6 to 7 percent of the NOM in Lake Pleasant water is humic substances. Samples were diluted to the desired NOM concentration with low-organic water, adjusted to the desired ph using NaOH or HCl, and equilibrated with air for about 1 h to bring the solution to ambient temperature before being used in UF tests. Dilution water, prepared by passing deionized water through an RO membrane and a cartridge containing activated carbon, contained.15 mg/l total organic carbon (TOC).** All other chemicals used in the tests were reagent grade. In one test, water samples were taken from Lake Washington in Seattle and from Judy Reservoir (the water supply source for Mt. Vernon, Wash.) for evaluation of the NOM-removal efficiency using the combined adsorption UF process. Lake Washington has a water quality similar to most source waters of municipal utilities in the Pacific Northwest; UV absorbance at 254 nm is about.5/cm, and the DOC is about 3 mg/l. The Judy Reservoir water contains 3 to 4 mg/l DOC, on average, and has a UV absorbance at 254 nm of approximately.13/cm. Iron oxide particles. This study investigated two types of iron oxide particles unheated iron oxide particles (UHIOPs) and heated iron oxide particles (HIOPs). UHIOPs. UHIOPs were generated by neutralizing a ferric nitrate (Fe(NO 3 )3 9H 2 O) solution using 1 N NaOH. The precipitated particles were rinsed three times with low-organic water to remove excess sodium and nitrate ions. HIOPs. HIOPs were obtained by heating a thick slurry of UHIOPs at 11 o for 14 h. The heated particles dried out during the heating process and were resuspended in distilled water after the heating. The heated particles ranged in size from.5 to 5 µm. UF membrane. The UF membrane used in the study was a hollow-fiber membrane. The membrane is made of cellulose acetate derivatives and has a nominal pore size of 1 nm. The inside diameter of the membrane fiber is.93 mm, and the thickness of the wall is about.4 mm. Most of the membrane wall is merely supporting material; the thickness of the membrane layer itself is less than 1 µm. Although these membranes are typically operated with a permeate flux of about L/m 2 /h, a target flux rate of 15 L/m 2 /h at a pressure differential of.9 bar (13 psi) was used in most experiments reported here, in order to stress the system. SFM module. The SFM module used in this work is illustrated in Figure 1. To construct a module, a 4-mm (16-in.) single membrane fiber was inserted inside soft, transparent vinyl plastic tubing, and nylon T-connectors were attached to both ends. The gap between the membrane fiber and the T-connector was sealed by rapid-curing epoxy resin (curing time less than 5 min). To sample a portion of the membrane for analysis, a section of the module was cut off, another T-connector was inserted in the cut end of the module, and the connection was resealed with epoxy resin. The sampling and reconstruction were accomplished within an hour. Because drying can constrict the membrane pores, the membrane was kept moisturized at all times, except for the section used for epoxy sealing. Preparation of samples for FESEM analysis. Membrane samples were prepared for FESEM analysis according to the method used by Kim et al. 21 The samples were rinsed three times with low-organic water and then immersed in 3 percent glutaralde- *CALYX nylon filter cartridge, Micron Separations Inc., Westboro, Mass. XAD-8, Rohm and Haas, Philadelphia, Pa. Ametek Corp., Pittsburgh, Pa. Ion X changer adsorber, Cole Parmer, Niles, Ill. **OI 7 carbon analyzer, OI Corp., College Station, Texas Spectronic 1, Milton Roy Co., Rochester, N.Y. Purified, J.T. Baker Chemical Co., Phillipsburg, N.J. Lyonnaise des Eaux Dumez, Le Pecq, France DECEMBER

7 FIGURE 6 Original Flux percent FIGURE 7 Fraction of Original Permeate Flux percent FIGURE 8 Percentage of Original Flux percent Permeate flux curves for six membranes receiving different fractions of NOM as influent percent 75 percent 66 percent 55 percent 2 percent 14 percent Operation Time h TOC 4.5 mg/l, ph Effect of ph on UF membrane fouling ph 5 ph 7 ph Operation Time h Cross-flow mode, TOC 4.7 mg/l, tangential velocity.7 m/s (2.3 fps), cycle length 1 h Effect of loop-flushing frequency on permeate flux Control run no flushing Control run (NOM-free water) no flushing 3 s/1 min 1 min/h Time h Static mode, source filtered Lake Pleasant, TOC 4.5 mg/l, ph 7.; data points represent the ratio of the permeate flux at the beginning of each cycle to the original flux. hyde in 25 mm sodium phosphate buffer (ph 7.2) for 15 min. This fixation process prevented the organic samples from being deformed during the drying process. Postfixation using 2 percent osmium tetroxide in phosphate buffer was also used for some samples, but this extra step did not enhance the image contrast significantly, and its use was abandoned. The samples were then frozen in liquid nitrogen and broken by applying bending stress. Freezing the membranes before breaking them generated membrane pieces with sharp edges ideal for microscopic observation. The membrane specimens were immobilized on Al specimen mounts using Ag paste and were allowed to dry in ambient conditions for 24 h. They were then coated with 2-nm Au/Pd alloy using a coater* operating at a current of 2 ma for 2 min. A high-resolution, low-damage electron microscope was used to analyze the morphology of the sample. Membrane system operation. The membrane systems were operated in three different modes in this study: static, circulating, and combined (Figure 2). The static mode (Figure 2, part A) was essentially dead-end filtration. In circulating mode (Figure 2, part B), the membrane was operated in a conventional cross-flow configuration, with a tangential velocity of.7 m/s (2 fps) supplied by a circulation pump. The water flow inside the membrane fiber during a filtration cycle was assumed to be laminar because the Reynolds number (n Re ) of the flow under normal operational conditions was only about 65. The volume of the circulation loop was approximately 8 ml. The average transmembrane pressure (the average of the 8 JOURNAL AWWA

8 pressures at the inlet and outlet of the module) of the system was controlled at around 1 bar (13 psi), so the permeate flow rate was determined by the hydraulic resistance across the membrane. Because no waste was discharged from the circulation loop during the operating cycle, the concentration of nonpermeable DOC inside the loop increased with time. The membrane was backwashed at the end of each operational cycle (usually 1 h) for 2.5 min. The backwash pressure was controlled at around 2.5 bar (6 psi) as suggested by the membrane manufacturer. To prevent biological growth in the system, 5 mg/l of free chlorine was added to the backwash water. In addition to the regular backwashing, a backward (from-outlet-to-inlet) tangential flush with the same tangential velocity as that in the regular cycle was applied to the system during the first and last 3 s of the backwashing step. In the combined UF adsorption process (Figure 2, part C), an operation cycle started with a 1-min period during which a slurry of IOPs was injected in circulating mode. In a typical test, the adsorbent dose was 17 mg/l Fe of water treated. This dosage was chosen based on the amount of iron oxides needed to The UF membrane used for NOM filtration is shown before (left) and after backwashing (right). remove almost all the adsorbable NOM from solution in batch adsorption isotherm tests. Although this dosge was greater than that used in conventional coagulation (1 mg/l), the presumption was that the IOPs could be regenerated and reused, so the generation of sludge from adsorbent addition would be small. During the treatment portion of the cycle, the permeate passing through the membrane wall was collected, and the nonpermeable NOM and adsorbent particles stayed in the loop, with some of the particles forming a thin cake on the membrane surface. The membrane was backwashed at the end of each operational cycle, and tangential flow was applied to the system to flush out the NOM-laden HIOPs and the IOP clumps that broke loose from the membrane surface during backwashing. FIGURE 9 Original Flux percent Effect of backwashing frequency on permeate flux Control run no flushing or backwashing 3 s/1 min Control run (NOM-free water) no flushing or backwashing 2.5 min/h Time h Static mode, source filtered Lake Pleasant water, TOC 4.5 mg/l, ph 7.; data points represent the ratio of the permeate flux at the beginning of each cycle to the original flux. Results and discussion Operation without adsorbent addition: formation of the NOM gel layer. UF membranes are designed primarily for particulate removal. Nevertheless, even though they reject only a small portion of NOM molecules, the presence of NOM in the water may still cause significant fouling. Figure 3 shows a typical profile of permeate flux versus time for a run in static operation mode. The water used in this test was Lake Pleasant water diluted to 4.5 mg/l TOC. Over a 2-h period, the permeate flux decreased from 23 to 5 *Hummer V Sputter, Ametek Ltd., Alexandria, Va. JEOL JSM 63F, UEOL USA, Peabody, Calif. Sodium hypochlorite, Aldrich Chemical Co., Milwaukee, Wis. DECEMBER

9 FIGURE 1 Fraction of Original Flux percent TABLE Operation Time h DOC 4.5 mg/l, ph 7., HIOPs 15 mg/l Fe Treatment Type Effect of addition of adsorbent particles on permeate flux Clean water NOM only IOPs + NOM NOM removal efficiencies for treatment of three natural waters with the combined adsorption UF process Lake Pleasant Lake Washington Mt. Vernon L/m 2 /h. SEM images of these membrane samples (page 77) showed the growth of an NOM gel layer up to 1.5 µm thick over the course of the experiment. The top left photomicrograph on page 77 shows a clean membrane. The porous material at the bottom is the supporting layer; only the dense layer on the top is the membrane itself. After 3 min of operation (top right photomicrograph, page 77), the flux had declined by 15 percent. Some scattered organics were deposited on the membrane surface, but the amount of organic deposit was insufficient to form a complete gel film. The lower left photomicrograph on page 77 shows the same membrane after 1.5 h of operation, by which time the flux had declined by 3 percent. By this time, a continuous NOM film has formed on top of the membrane. The thickness of the film increased to.53 mm after 9 h of operation (middle right photomicrograph, page 77) when the flux had declined by 65 percent, and to 1.5 µm (lower right photomicrograph, page 77) by the end of the 2- h operation, when the flux was only 21 percent of its initial value. During the course of the test, NOM removal efficiency increased from approximately 21 to 28 percent. This series of SEM pictures suggests that the formation of the gel layer started with the deposition of NOM on the membrane surface by either adsorption or precipitation and that the deposits then gradually grew to form a continuous film that covered percent UF only UF + HIOPs (17 mg/l) the entire surface, including the membrane pores. These SEM images provide strong evidence of the formation of a gel layer on the membrane surface and suggest a correlation between the thickness of the layer and permeate flux decline, although it is clear that flux decline begins long before a continuous film forms. Fouling potential of selective subgroups of NOM molecules. NOM molecular size and fouling potential. The NOM molecules that foul the membrane must be retained by it. However, the fraction of potential foulant molecules that are retained by the membrane could range from negligible to nearly percent. To test the hypothesis that the molecules responsible for fouling are very efficiently retained, a single batch of water was passed through three new membrane modules in series (Figure 4), and the TOC removal and permeate flux in each membrane were monitored. The first membrane removed about 15 percent of the organics (the TOC concentration decreased from 4.7 to 4. mg/l); the second and third membranes did not remove detectable amounts of NOM. The NOM removal accomplished in the first membrane is typical for UF membranes, and the absence of any NOM removal in the second and third membranes is consistent with the idea that the membranes act as a physical barrier to passage of molecules larger than the pore size. Because most of the large molecules were removed in the first membrane, almost no molecules larger than the pore size entered the second or third membrane module. Figure 5 shows the flux through each of these membranes as a function of operation time. The flux through the first membrane decreased quite rapidly. Surprisingly, however, the flux through the second and third membranes declined almost as fast as that through the first. Although the initial flux decline in the latter two runs may have been slightly slower than in the first, this difference can reasonably be attributed to the lower TOC in the influent to the latter systems. Eventually, the flux through all three membranes reached the same level. This result suggests that many of the molecules that can pass through the membrane can also, under almost identical circumstances, be retained and cause significant fouling. Fouling of UF membranes by organic 82 JOURNAL AWWA

10 In an SEM image of the UF membrane used in the combined HIOP-UF system (top left), the thickness of the IOP cake layer is approximately µm before backwashing. After backwashing (top right), the thickness of the IOP cake layer is approximately 15 2 µm. The bottom photomicrograph shows the area where the residual IOP cake layer was completely removed. molecules that are smaller than the pores has been reported by Crozes et al. 24 They reported that when solutions containing 15 mg/l of tannic acid or dextran were treated by UF, adsorption of these small organic molecules decreased the permeate flux by 8 percent within 4 and 2 h, respectively. SEM photographs of the membranes used in these tests (page 79) show that some NOM deposits on the membrane surface in all three systems, even though the NOM removal could not be detected as a decrease in TOC concentration in the last two runs. Estimating the mass of TOC that would be required to form a continuous film layer explains this apparent anomaly. For instance, for the membrane tested (with a length of 4 mm [16 in.]), a continuous 1-µm-thick film covering the entire membrane surface could be generated if only.37 mg/l of TOC was removed from the amount of water (8 L) treated in these tests. These calculations assume that the film has a density of 1 g/cm 3 and a water content of 5 percent and that NOM is 5 percent carbon by weight. Correlation between affinity of NOM molecules for HIOPs and their fouling potential. Metal oxides interact selectively with NOM molecules with higher molecular weights, 25 and those molecules might be more hydrophobic, on average, than smaller NOM molecules. Therefore, because HIOPs are able to remove a significant amount of NOM from water, it seemed possible that they selectively remove NOM molecules that contribute disproportionately to the fouling of UF membranes. To test this hypothesis, the authors conducted a series of experiments using six water samples containing NOM with different affinities for HIOPs. The waters used in these experiments were prepared by adding different amounts of HIOPs to a batch of raw Lake Pleasant water that contained 3 mg/l TOC. When the HIOPs were added to the water, between and 86 percent of the NOM in the water sorbed onto the particles, leaving 4. 3 mg/l TOC in solution. The water samples that contained more than 4.5 mg/l TOC were then diluted to 4.5 mg/l TOC (the sample that had been exposed to the highest HIOP concentration contained only 4. mg/l TOC). Thus, this procedure generated several samples with equivalent NOM concentrations, but with the NOM molecules that had the greatest affinity for HIOPs removed from some of the samples. The water sample containing the entire pool of NOM molecules (prepared by dilution of raw water without exposure to HIOPs) was referred to as the percent sample. The water from which the most strongly adsorbing 25 percent of the NOM molecules was removed prior to adjusting the sample concentration to 4.5 mg/l was referred to as the 75 percent sample, meaning that 75 percent of the initial pool of molecules remains in the water. If sorption to HIOPs does indeed selectively remove the foulant molecules, the fouling tendency of the samples would have increased as the percentage of the original NOM in the sample increased. When the six samples were tested individually in UF experiments in static mode, the NOM removal efficiencies were similar (ranging from 19 to 22 percent), and the waters had virtually identical fouling tendencies (Figure 6). Thus, the affinity of NOM molecules for HIOPs is apparently not correlated with their fouling properties, at least for the membrane tested. This result is consistent with the results reported by Lahoussine-Turcaud et al 9 that even when up to 7 percent of the NOM was removed from the influent by sorption onto ferric coagulants, the NOM remaining in solution could cause significant fouling. Effect of ph on NOM membrane fouling. Solution ph may affect the sizes of NOM molecules and their chemical characteristics, e.g., hydrophobicity and charge, and it is therefore expected to affect the foul- DECEMBER

11 FIGURE 11 Fraction of Original Permeate Flux percent Permeate flux decline using influents with different TOC concentrations.9 mg/l TOC 4.5 mg/l TOC Operation Time h ing potential of NOM molecules. The fouling potential of diluted Lake Pleasant water (with a TOC of 4.7 mg/l) was tested at ph values of 5, 7, and 9 in conventional cross-flow mode using a tangential velocity of.7 m/s. Although the NOM rejection efficiency by UF membranes was slightly better at ph 7 and 9 (23 percent in both cases) than that at ph 5 (19 percent), the profiles of permeate flux reduction in these three tests were very similar (Figure 7). Thus, although the ph affected the efficiency of NOM rejection slightly (an effect that might be attributed to reduced molecular size at lower ph), it had no effect on the membrane fouling potential under the conditions tested. A reduction in the fouling potential of NOM might be expected at lower ph because less NOM is rejected by the membrane. On the other hand, at lower ph, the NOM molecules are more fully protonated and therefore more hydrophobic and more strongly attracted to the membrane surface. As noted earlier, however, significant fouling can occur even with minimal NOM rejection. Therefore, although the difference in NOM removal at the three ph values tested might affect the gross rate of film growth slightly, its minimal effect on overall flux is not surprising. Thus, tests investigating serial filtration, HIOP sorption, and solution ph all failed to identify a subgroup of NOM molecules that was selectively responsible for fouling the UF membranes. Although such a subgroup of molecules might exist, it is apparently not possible to separate it from the nonfouling or less-fouling molecules by passage through a UF membrane or by exposure to HIOPs. Interfering with formation of the NOM gel layer. The selective removal of the molecules responsible for formation of an NOM gel on the membrane surface appeared to be impractical, at least by the methods investigated. For this reason, the focus of the study was shifted to alternative methods for preventing the formation of the gel layer. Operating strategies that were investigated included using loop flushing, backwashing, or both to clean the membranes after a period of operation and adding NOMadsorbing particles to the loop. Loop flushing. When a portion of the NOM molecules is rejected by the UF membranes, the concentration of NOM in the circulation loop increases. The fouling process may be accelerated by the increased NOM concentration in the bulk solution, the CP layer, or both. Frequent flushing of the concentrated solution out of the circulation loop may reduce the NOM concentration in the CP layer and therefore prevent formation of the gel layer This test investigated the effect of loop flushing on the permeate flux, using loop flushing frequencies of 1 min every hour or 3 s every 1 min. The results from these two tests, along with those for two control runs, are shown in Figure 8. One of the control runs used the same operating conditions without any loop flushing, and the other used low-organic water as influent. The system with more frequent loop flushing experienced much less fouling than the one with a lower flushing frequency. In fact, the flux in the run with the 1-min flushing cycle was virtually identical to that in the control run using low-organic water. NOM molecules may be imagined as fouling the membranes in two ways: molecules may adsorb on the wall inside the pores and gradually plug the pores, or they may adsorb on the membrane surface and grow a film that bridges over the pores. Loop flushing might reduce fouling by the second mechanism, but backwashing would be required to interfere with the first. Because the run with the 1-min flushing cycle completely prevented fouling (compared with the clean water control) without use of backwashing, internal clogging of the pores was apparently not an important fouling mechanism in this system. Rather, it appears that the formation of a gel layer that bridges over the pores was the primary fouling mechanism and that the loop flushing washed away the NOM near the membrane surface before the gel layer formed. However, if loop flushing is not frequent enough to prevent the formation of a gel, as was apparently the case in the run with a 1-h cycle, the tangential flush may not be able to remove the gel layer. Because the tangential velocities during loop flushing and during the regular operational cycle were 84 JOURNAL AWWA

12 identical, the shear stress on the membrane wall (estimated to be about 6 N/m 2 ) was the same in both instances. Therefore, although the shear force might have The photomicrograph at left shows the membrane surface after being run in a conventional cross-flow configuration for two days. The photomicrograph at right shows the membrane after 16 days in a conventional cross-flow configuration using pretreated, low-organic influent (.9 mg/l TOC) with a 1-h backwash cycle. reduced membrane fouling during the regular filtration cycle, it was probably not responsible for the improvement in system performance prevention of fouling observed when frequent loop flushing was applied. Rather, two other factors may account for the ability of frequent flushing to prevent fouling. First, loop flushing reduces the NOM concentration in the loop and therefore reduces the mass flux of NOM convected toward the membrane surface. Second, lower bulk concentration accelerates the back-diffusion of NOM from the CP layer to bulk solution, decreasing the NOM concentration in the CP layer. Backwashing. Backwashing is the cleaning scheme most commonly used in MF and UF processes. By inducing flow from the permeate to the concentrate side of the membrane, backwashing can, at least in theory, unplug clogged membrane pores and disrupt the foulant layer. This is a relatively easy and effective method to restore the permeability of membranes when particulates are the primary foulant. In a test using the same operating schemes as in the previous test, backwashing was applied to the membrane during the loop flushing. Results from these tests as well as the earlier test are shown in Figure 9. For the runs with a 1-min flush cycle (with flushing and backwashing for 3 s at the same time), the additional backwashing did not improve performance significantly, because no gel layer formed on the surface. However, for the runs with a 1-h flush cycle (with a 2.5-min backwashing and a 1-min flushing), the permeate flux was substantially improved when backwashing was applied. Comparison of the SEM images of the membrane surface before and after backwashing (page 81) indicated that backwashing removed a portion of the deposited NOM and made the gel layer more irregular and presumably more porous. Nonetheless, the fact that a significant amount of NOM gel remained on the membrane surface after backwashing suggests that backwashing only cleans the membrane pores and removes the deposited NOM directly above the pores. The backwash water may break through the NOM gel layer and create some porosity, but it is unable to lift the entire gel layer. This residual NOM may cause fouling problems in the long term, because the void space generated in the film by backwashing might collapse during pressurized operation in subsequent runs. Addition of HIOPs. The behavior of heated and unheated oxide particles in the UF system was compared with a control system (no particles added) in a 13-day test using low-organic water and with hourly backwashing. These tests were run in constant-flux mode, so fouling is indicated by a buildup of crossmembrane pressure. Because no NOM was present, these tests evaluated the fouling potential of the IOPs by themselves. No significant buildup of the system pressure occurred in the tests with either no particles or HIOPs present in the system. When UHIOPs were added, however, the pressure increased significantly (from.9 to 1.7 bar [13 to 25 psi]). The increase of transmembrane pressure was partially due to the restriction of the flow channel caused by the deposited oxide cake layer. The reduced flow channel increased the tangential velocity inside the membrane fiber and the frictional head loss for flow through the fiber. The net effect was manifested as an increase in pressure at the entrance to the fiber. In some runs, the flow-channel restriction led to operational failures. Subsequent backwashing reduced the cross-membrane pressure needed to maintain the desired flux by.2.4 bar (3 6 psi) in the system with UHIOPs but did not return it to its clean-membrane value; in other words, the fouling was physically irreversible. Even after thorough chemical cleaning using the cleaning reagents (a surfactant and an iron reductant) supplied by the membrane manufacturer, the fouling in the system with UHIOPs could not be completely reversed. Although the cause of membrane fouling in the system receiving UHIOPs is not entirely clear, the particles must have been responsible in some way. Most of the membrane surface was still covered by brown-colored material after thorough backwashing and chemical cleaning, and SEM analysis showed a layer of iron oxides on the membrane surface, which DECEMBER

13 FIGURE 12 Fraction of Original Permeate Flux percent Permeate flux decline using membranes with a precoated cake layer of preloaded IOPs Tangential velocity = Tangential velocity =.7 m/s (2.3 fps) Operation Time h may block the surface pores. Also, although most of the UHIOPs were expected to be larger than the membrane pores (~1 nm), it is possible that trace amounts of dissolved iron adsorbed inside the pores, creating nuclei for subsequent growth of iron oxides and gradually blocking the pores. Table 1 lists NOM removal efficiencies for the three natural waters tested using the combined process. The addition of HIOPs dramatically increased the TOC removal efficiencies of the systems, from between 1 and 2 percent in the absence of HIOPs to between 5 and 85 percent with them. Furthermore, membrane flux deteriorated minimally during these tests. Three sets of experiments were conducted for about one month each to further illustrate the mitigation of membrane fouling by adding HIOPs. In these tests, the permeate flux in the system with HIOPs was essentially identical to that in the control run using low-organic water and was substantially higher than in the run without adsorbents (Figure 1). SEM images of these membranes (top left, photomicrograph, page 83) showed that a cake layer approximately µm thick formed on the membrane surface during operation with the HIOPs. Because this cake layer did not reduce flux compared with the control run, the hydraulic conductivity of the oxide cake layer must have been higher than that of the membrane, i.e., the oxide cake layer did not contribute significant hydraulic resistance to the permeate flow. When the membrane was backwashed, many of the HIOPs were flushed out of the system as aggregated clumps. After backwashing, the thickness of the cake layer was reduced to about 15 2 µm over most of the surface and was completely removed from patches of the surface, as shown in the top right photomicrograph on page 83. These patches contained no visible deposited or precipitated NOM (bottom photomicrograph, page 83), suggesting that no NOM gel layer grew on the membrane surface when oxide particles were added. Mechanisms of fouling prevention by the addition of HIOPs. The previous experiment showed that the growth of an NOM gel layer on the UF membrane surface was prevented by the addition of HIOPs. In view of the relatively thick layer of HIOPs that deposited on the membrane surface, scouring of the membrane surface by adsorbent particles, which has been suggested by researchers 14 studying different systems, cannot account for the results in this study. The authors next examined two other possible mechanisms that could have led to the observed behavior. Reduction of TOC concentration. The first hypothesis tested was that the approximately 8 percent reduction in the TOC concentration in the bulk solution (by sorption) prevented the TOC concentration in the CP layer from exceeding the NOM solubility. A lower NOM concentration in bulk solution leads to lower concentration in the CP layer for two reasons: the mass flux of NOM toward the membrane surface is lower, and the back-transport of NOM molecules from the CP layer to bulk solution is higher, because the concentration gradient between the CP layer and the bulk solution is greater (given the same CP layer concentration). To test this hypothesis, a batch of water with a TOC concentration of 4.5 mg/l and a ph of 7 was pretreated with HIOPs in a beaker to remove 8 percent of the TOC. The treated water was then passed through a UF membrane module operated in conventional cross-flow filtration configuration as described earlier. The system was operated for two weeks. A control run using whole influent (4.5 mg/l TOC) was conducted in the same way. Membrane samples were taken for FESEM analysis after two days of operation and at the end of two weeks. There was no significant flux decline during the first two days of operation (Figure 11) and the membrane surface was still relatively clean, as shown by the photomicrograph at left on page 85. Although some patches of NOM deposited on the membrane surface, the amount of NOM deposited was not enough to establish a continuous fouling layer. After two weeks of operation, however, the flux had declined by 3 percent, and there was a nearly continuous layer of NOM film on the membrane surface (the photomicrograph at right on page 85). This film- 86 JOURNAL AWWA

14 growth behavior was consistent with the SEM images (page 77) obtained during earlier experiments. The fact that the flux declined faster in the control run (using 4.5 mg/l TOC) suggests that the concentration in the bulk solution is an important factor controlling fouling kinetics. Interaction of the HIOP cake layer with the NOM gel layer. The second hypothesis tested was that the oxide cake layer prevents gel formation by interacting physically or chemically with the CP layer. To test this hypothesis, adsorbent HIOPs used in previous experiments that were partially saturated with NOM were added to the loop in place of fresh HIOPs. These particles were circulated in the system for 1 h to preform an oxide cake layer on the membrane surface. The membrane was rinsed with NOM-free water to remove unattached HIOPs from the system and was then used in a static (deadend) mode filtration test for 24 h using an influent containing 4.5 mg/l of TOC. Another membrane fiber precoated with HIOPs in the same way was operated in a similar manner, except a tangential velocity of.7 m/s was applied. Within 24 h, the permeate flux decreased by 6 percent in the test with tangential velocity and 75 percent in the test without tangential velocity (Figure 12). This result suggests that even in the presence of a pre-formed HIOP cake layer, NOM can still cause permeate flux decline; i.e., the mere presence of a layer of oxide particles on the membrane does not prevent fouling. Furthermore, the benefits of tangential flow were slight in the absence of a cleaning scheme that flushed the particles and accumulated NOM out of the system. FESEM images (page 87) clarified why fouling occurred in these systems: NOM accumulated on the upper layer of the HIOP cake, just as it does on the membrane surface if there is no such cake layer. The NOM deposit mixed with the HIOPs and gradually blocked the pores in the upper part of the cake. During backwashing, a portion of this NOM was undoubtedly washed out of the system, along with some of the HIOPs. On the other hand, the photomicrograph (on the right on page 87) of the membrane surface taken at areas where the HIOPs were removed during the sample preparation step showed that NOM did not accumulate on the membrane surface. Thus, the cake did serve as a protective layer in that it prevented the membrane surface from being fouled by NOM, although in doing so the cake layer itself became fouled and led to flux reduction. SEM images taken at different points on the same membrane sample show the precoated IOP layer after it has been fouled by NOM. Summary and conclusions In this research, an SFM module was used to monitor changes in the morphology of deposits near a UF membrane surface. Considering the information obtained by subsampling the membrane at various stages of operation and the additional benefits of using the module (such as better control of hydraulic conditions in the test fiber and minimal water consumption), the SFM module provides a propitious approach for bench-scale membrane research. NOM fouling of UF membranes started with the formation of NOM patches on the membrane surface. This process, which might be caused by either surface adsorption or precipitation, was accelerated by the buildup of NOM near the surface because of concentration polarization. The NOM patches gradually expanded and eventually formed a continuous gel film that bridged across the pores and covered the entire membrane surface. As the thickness of the film grew, the permeate flux through the membrane steadily decreased. The tendency of NOM molecules to foul the membrane surface was not closely related to their size, affinities for iron oxides, or solution ph, and UF membranes could be fouled by NOM even if a negligible amount of NOM was retained by the membrane. The NOM-induced fouling occurred primarily on the membrane surface. No evidence that relates permeate flux decline to fouling in the pores was found. NOM-induced fouling of UF membranes can be reduced or eliminated by very frequent and brief loop flushing, by frequent loop backwashing, or by addition of a cake-forming NOM adsorbent. Although the optimum frequency for loop flushing and backwashing depends on other operational conditions, use of an appropriate cleaning frequency is clearly important. In this study, loop flushing and backwashing partially cleaned the membrane surface, but they did not affect NOM removal. By contrast, addition of HIOPs dramatically reduced NOM accumulation on the membrane surface while simultaneously and dramatically increasing NOM removal. The addition of HIOPs to a UF process can significantly increase NOM removal efficiency during treat- DECEMBER