The argument for putting a biofilm process in front of a membrane is easy to make. Membrane fouling in an MBR is usually attributed to the sludge. A moving bed biofilm reactor keeps most of its biomass on plastic carriers, so the water reaching the membrane carries hundreds of milligrams of solids per litre instead of several grams. Less material arriving at the membrane should mean less material stuck to it.
That expectation ran through a line of biofilm-MBR work at NTNU from the mid-2000s, although the published literature was already mixed: some studies favoured the biofilm configuration and others the activated sludge one. I built the pilots described here partly to settle that with direct comparisons. In the two side-by-side comparisons, the biofilm membrane fouled faster in one and at a similar rate in the other. In a third study, with no activated sludge reference, fouling stayed in the biofilm range even when suspended solids were pushed down to a level no activated sludge plant would ever see.
The likely reason is more interesting than the expectation. The MBBR itself is a sound process. What the biofilm configuration removes, along with the sludge, appears to be something that was protecting the membrane.
Three studies, one question
The work comprised two journal papers and a conference abstract. They were not designed as a series, which matters later, but they were all asking a version of the same question: what happens to membrane fouling when you take most of the suspended biomass away?
| Parallel AS-MBR and BF-MBR | Two-stage BF-MBR | CLSM comparison | |
|---|---|---|---|
| Published | Desalination and Water Treatment, 2012 | Water Science & Technology, 2012 | NAMS conference abstract, 2012 |
| Feed | Semi-synthetic municipal, high soluble COD | Real municipal sewage, three loads | Real municipal sewage |
| Biology | AS at about 5 g/L SS; MBBR with K1 carriers held below 1.1 g/L by an external settler | Fixed bed for pre-denitrification ahead of an aerobic MBBR, with recycle between them | AS and biofilm reactors in parallel |
| Membrane | Flat sheet MF, 0.4 µm, 10.3 L/m²h, with relaxation | Hollow fibre UF, 0.05 µm, 8.8–13.2 L/m²h, no relaxation or backwash | Flat sheet, 10 L/m²h, no relaxation in the reported example |
| Main result | BF fouled consistently faster; rate tracked filtered COD | SS below 60 mg/L, yet fouling still followed filtered COD and load | Similar fouling rates at low load, but different fouling layers |
The pattern
Put the fouling rates on one axis and the suspended solids in the membrane tank on the other, and the expected relationship does not appear. Solids span more than two orders of magnitude. Fouling does not follow them down.
In the parallel study the activated sludge membrane barely moved over five-to-six-day runs, while the biofilm membrane fouled between roughly 1 and 2.4 mbar/h depending on the run. In the two-stage system, run at a higher flux of 13.2 L/m²h for these tests, the recycle between the fixed bed and the MBBR pulled suspended solids down from about 125 mg/L to below 25 mg/L. Fouling halved, but it halved to a rate in the same range as the single-stage biofilm reactor, with a small fraction of its solids.
The most telling detail is at the low end of that curve. Going from R = 2 to R = 3, suspended solids fell a little further. Filtered COD did not fall, and neither did the fouling rate. And when the organic load was raised, fouling climbed steeply even though the solids at the membrane never exceeded 60 mg/L. At high load, and at the lower flux of 8.8 L/m²h, the membrane reached the 0.3 bar limit on the fourth day.
Below a certain point, suspended solids stopped being the variable that mattered.
What the membrane was actually seeing
In every study the variable that did track fouling was the organic material passing a 1.2 µm or 0.45 µm filter: filtered COD, a crude measure of soluble microbial products and colloids together. In the parallel study the biofilm reactor carried more of it on average at both filter sizes, though the error bars overlapped, and within the five biofilm runs the fouling rate rose almost linearly with it. In the two-stage study it moved with fouling across recycle ratios and across loads.
The other indicators pointed the same way. The biofilm mixed liquor had a capillary suction time more than twice that of the activated sludge, settled poorly, and had a slightly more negative zeta potential, all consistent with a finer, more stable colloidal suspension. Under the microscope the activated sludge formed much larger flocs, with what appeared to be more filamentous bacteria; the biofilm suspension was small scattered flocs.
Then came a result that seemed to contradict the rest. In the number-weighted particle size distribution, the activated sludge was if anything more weighted toward submicron particles than the biofilm suspension, not less. The fine material was there in both systems. The explanation proposed in 2012 was that what differed was what it met on the way to the membrane.
Activated sludge, about 5 g/L
Proposed: large flocs arrive quickly and build a thick, porous cake. Fine particles are caught in the cake before they reach the membrane, and the cake is loose enough for air scouring to remove.
CLSM, low load: the biocake grew about twice as fast, had its lowest porosity 5–10 µm above the membrane, and a higher density of carbohydrate and protein.
Biofilm, below 1 g/L
Proposed: too few large particles to form a protective cake. Colloids and soluble polymers reach the membrane surface directly and block pores, which scouring does little about.
CLSM, low load: a thinner layer after one and two weeks, with higher cell density, less carbohydrate and a porosity that decreased toward the membrane, lowest right at the surface.
The confocal work is where this became partly visible rather than inferred, with one caveat: it ran on real sewage at low load, not on the semi-synthetic feed of the parallel study, so it shows the structures rather than proving the mechanism for that study. At low loading the two membranes fouled at a similar rate while carrying layers that grew at rates differing by a factor of about two. Thickness alone evidently did not set the resistance; structure, including where the densest material sat, plausibly did. A drinking water pilot pointed to a similar lesson about the position of polysaccharide, described on biofilm.no, from a completely different direction.
Putting the solids back
If the suspended solids were protecting the membrane, adding them back should help. In one further run the settling tank on the biofilm pilot was disconnected and solids were allowed to accumulate, rising from about 0.4 to about 3.4 g/L over two weeks.
The pressure record split into three stages. For the first week the transmembrane pressure climbed steadily. Around the point where solids passed about 1.76 g/L it turned, fell for two and a half days, and then resumed rising at a much lower rate, of the same order of magnitude as the activated sludge pilot. Filtered COD, normalised capillary suction time and zeta potential levelled off around the same time, and the mean size of the submicron fraction trended upward, with considerable scatter.
The conclusion drawn in 2012 was that a biofilm MBR can match an activated sludge MBR if it is run with enough suspended solids, which moves it toward an IFAS configuration: carriers and suspended sludge in the same tank. As a direction that still looks right. As a number, 1.76 g/L deserves the caution set out below.
Others found the same thing, earlier
The clearest precedent came from Chung-Hak Lee's group in Seoul. In 2001 they built an attached growth reactor specifically to take suspended biomass out of the fouling picture, and found the opposite of what they intended: the attached growth membrane fouled about seven times faster than the suspended growth one. They explained it the same way, as a dynamic membrane formed by suspended solids that the attached system lacked. I cited that paper in 2012 and should have leaned on it harder.
There is one difference worth keeping. Lee and colleagues reported that the soluble fractions of their two reactors had similar characteristics, and concluded that filtration performance depended on the suspended solids concentration. In our parallel study the filtered COD differed on average. The two are not necessarily in conflict. Depending on how a soluble fraction is defined, a 1.2 µm filtrate can include colloids that it would not, and the colloidal fraction may be where the difference lives.
The same group went on to study membrane-coupled moving bed reactors and found fouling there governed more by the physical action of the moving media than by the chemistry of the liquid, and, citing earlier Norwegian work, they noted the poor settling of MBBR sludge that our settling and capillary suction observations also suggested. The same group's later confocal work on bio-cake structure and the two-stage pressure rise, and their demonstration that quorum sensing is associated with membrane biofouling, took the subject well beyond a question of solids.
Closer to home, Åhl, Leiknes and Ødegaard had already shown in 2006 that a moving bed membrane reactor fouled faster at high organic load, and that the membrane reactor held a higher fraction of submicron colloidal particles, which they took to be the dominant contribution to fouling. Ivanovic and colleagues then treated submicron particles as a primary foulant in biofilm MBRs and reduced fouling by adding a flocculation zone beneath the membrane module. Both fit what the two-stage load tests showed, and both are consistent with the possibility that the difference between the parallel study and the low-load confocal study was partly a matter of load. The feeds also differed, so load cannot be isolated.
The hybrid direction has since been tested repeatedly. Khan and colleagues, for example, added sponge carriers to a conventional MBR and got longer filtration cycles, with lower cake and pore-blocking resistance. They attributed the more porous cake to the hybrid biomass and to fewer filamentous bacteria. That is carriers added to sludge, not sludge removed in favour of carriers, and the distinction is the whole point.
What I would say differently now
The correlation rests on five points. The fit between filtered COD and fouling rate in the parallel study had an R² of 0.91, from five runs. That is a direction, not a model. Filtered COD also rises with load, so it may be standing in for several things that rise with it.
The 1.76 g/L threshold came from one run. Solids and time increased together in that run, so a change in the membrane or the biology over two weeks cannot be separated from the change in solids. Run-to-run variation was also large: even the first week of that run fouled more slowly than most of the five earlier biofilm runs. The value belongs to that pilot, that feed and that membrane.
One indicator improved partly by arithmetic. Capillary suction time normalised by suspended solids peaked on day three and then fell while solids kept rising. Dividing by a number that is increasing will produce a falling curve on its own. The raw values should have been shown alongside.
The submicron size trend may be an intensity-weighted mean. The instrument used light scattering, which usually reports a mean weighted heavily toward larger particles. If so, a rising mean diameter can mean fewer small particles or a few more large ones. The paper read it as the first.
The two-stage improvement had two causes that moved together. Raising the recycle ratio lowered suspended solids and filtered COD at the same time. The paper reported both, but framed the benefit of the process mainly around low solids. Pre-denitrification consumes readily degradable dissolved carbon, and that could explain much of the fouling reduction. The R = 2 to R = 3 step, where solids fell slightly and filtered COD and fouling did not, suggests the dissolved and colloidal fraction was doing the work.
"Irreversible fouling" was inferred, not measured. No resistance-in-series analysis separated cake from pore blocking, and chemical cleaning restored permeability fully between tests. The mechanism is plausible and consistent with later work, but the pilots did not demonstrate it.
The pilots cannot be laid on top of each other. Different membranes, pore sizes, fluxes, feeds and cleaning practices. The chart above is honest only about orders of magnitude. The two-stage pilot also ran without relaxation or backwash, which would be expected to raise its rates relative to a relaxed membrane.
The external settler had a pump. Mixed liquor in the parallel biofilm pilot was pumped through a settling tank at 20 mL/min and the overflow returned. Pump shear can break flocs, and broken flocs mean more fines. I cannot say how much of the biofilm system's colloidal load came from that loop, and in 2012 the question was not asked.
The feed favoured this outcome. About 80% of the COD in the semi-synthetic feed of the parallel study passed a 0.45 µm filter. A feed that soluble is likely to push the balance toward soluble and colloidal foulants. Real sewage carries far more particulate material.
The runs were short. The parallel runs lasted five to six days. Negligible fouling over that period says little about how the activated sludge membrane would behave over months.
"Confirm" was the wrong verb. The conclusions described the results as confirming earlier findings. They were consistent with them. With one feed, one set of pilots and a handful of runs, that is as far as the evidence went.
None of this overturns the main finding. Every correction above points the same way: toward the colloidal and dissolved fraction as the thing that matters, and away from suspended solids as a proxy for fouling potential.
If you are putting a membrane behind a biofilm process
Do not read low suspended solids as low fouling potential. At the low concentrations typical downstream of biofilm processes, suspended solids may tell you little about how the membrane will behave. The absence of a protective cake is itself a risk.
Measure the fraction that reaches the membrane. Filtered COD at two pore sizes is cheap and was informative here. Submicron particle counts and simple filterability tests add to it. Watch these through load changes, because load is where the biofilm configuration was most sensitive.
Treat the colloids, not the solids. Options that act on the fraction that fouled here include a flocculation zone ahead of the membrane, an anoxic stage that consumes dissolved carbon, and flocculant or coagulant additives. Removing still more suspended solids may not help, and may remove the protection.
Consider keeping some sludge. A hybrid with carriers and a moderate suspended solids concentration keeps the biofilm's benefits and gives the membrane a cake to hide behind. Where that concentration lies depends on your membrane and your water, and has to be found on site.
Mind the hydraulics. Pumped loops shear flocs. In the two-stage pilot a recycle ratio above four detached biofilm from the fixed bed, washed particles forward and markedly worsened fouling; the paper reported this without showing the data. Every recycle is also a source of fines.
Plan the cleaning you will actually need. A membrane downstream of a biofilm process may foul mostly at its surface and pores, where air scouring is weak. Relaxation, backwashing and maintenance cleaning are not optional because the solids look low.
Sources
The work behind this page
- Sun, C., Leiknes, T., Fredriksen, R. H. & Riviere, E. (2012). Comparison of membrane filtration performance between biofilm-MBR and activated sludge-MBR. Desalination and Water Treatment, 48(1–3), 285–293. doi:10.1080/19443994.2012.698824
- Sun, C. & Leiknes, T. (2012). Two-stage biofilm-MBR for nitrogen removal and enhanced membrane performance. Water Science & Technology, 66(3), 588–593. doi:10.2166/wst.2012.211
- Leiknes, T. & Sun, C. (2012). Assessment of biofouling behavior and properties in activated sludge vs. biofilm MBR using confocal laser scanning microscopy (CLSM). North American Membrane Society annual meeting, New Orleans, 11 June 2012. Conference abstract.
Other work cited
- Åhl, R. M., Leiknes, T. & Ødegaard, H. (2006). Tracking particle size distributions in a moving bed biofilm membrane reactor for treatment of municipal wastewater. Water Science & Technology, 53(7), 33–42. doi:10.2166/wst.2006.205
- Hwang, B.-K., Lee, W.-N., Yeon, K.-M., Park, P.-K., Lee, C.-H., Chang, I.-S., Drews, A. & Kraume, M. (2008). Correlating TMP increases with microbial characteristics in the bio-cake on the membrane surface in a membrane bioreactor. Environmental Science & Technology, 42(11), 3963–3968. doi:10.1021/es7029784
- Hwang, B.-K., Lee, C.-H., Chang, I.-S., Drews, A. & Field, R. (2012). Membrane bioreactor: TMP rise and characterization of bio-cake structure using CLSM-image analysis. Journal of Membrane Science, 419–420, 33–41. doi:10.1016/j.memsci.2012.06.031
- Ivanovic, I., Leiknes, T. & Ødegaard, H. (2008). Fouling control by reduction of submicron particles in a BF-MBR with an integrated flocculation zone in the membrane reactor. Separation Science and Technology, 43(7), 1871–1883. doi:10.1080/01496390801974704
- Ivanovic, I. & Leiknes, T. (2012). The biofilm membrane bioreactor (BF-MBR): a review. Desalination and Water Treatment, 37, 288–295.
- Khan, S. J., Zohaib-Ur-Rehman, Visvanathan, C. & Jegatheesan, V. (2012). Influence of biofilm carriers on membrane fouling propensity in moving biofilm membrane bioreactor. Bioresource Technology.
- Lee, J., Ahn, W.-Y. & Lee, C.-H. (2001). Comparison of the filtration characteristics between attached and suspended growth microorganisms in submerged membrane bioreactor. Water Research, 35(10), 2435–2445. doi:10.1016/S0043-1354(00)00524-8
- Lee, W.-N., Kang, I.-J. & Lee, C.-H. (2006). Factors affecting filtration characteristics in membrane-coupled moving bed biofilm reactor. Water Research, 40(9), 1827–1835.
- Yeon, K.-M., Cheong, W.-S., Oh, H.-S., Lee, W.-N., Hwang, B.-K., Lee, C.-H., Beyenal, H. & Lewandowski, Z. (2009). Quorum sensing: a new biofouling control paradigm in a membrane bioreactor for advanced wastewater treatment. Environmental Science & Technology, 43(2), 380–385. doi:10.1021/es8019275
The pilot data were published in 2012 and describe specific pilots, feeds, membranes and cleaning regimes. They illustrate mechanisms and are not design values. Figures on this page are re-read from the published papers and abstract; small reading errors are possible.