Performance of an aerobic granular sludge membrane filtration in a full-scale industrial plant

According to Eurostat (2022), Belgium is ranked in the top six European countries with water scarcity issues for the years 1994–2017 due to high water exploitation. Climate change worsens this problem since available freshwater resources are decreasing. Cornelissen et al. (2018) reported that approximately one company in four of the 90 companies surveyed, reuse part of their treated wastewater in Flanders, thus, showing that the majority of companies select to discharge their treated effluent. Furthermore, conventional activated sludge system (CASS) and sequencing batch reactor (SBR) systems account for about 70% of biological treatment systems, even with the separation problems that are often encountered.

Membrane bioreactors (MBRs) combine the biological treatment process with excellent separation. As a result, the obtained effluent has a high potential for reuse. Depending on the respective application, reverse osmosis or other advanced wastewater treatment technologies could follow. In addition, MBRs occupy limited space. Despite MBRs’ advantages, their popularity for broad applications has increased only recently, because both the investment and operating costs are quite high. The main cause of high operating costs is the mitigation of the fouling phenomenon; fouling can lead to a loss of permeability and/or a sharp increase of transmembrane pressure (TMP) (Le-Clech et al. 2006).

Membrane fouling is the result of the interaction between the membrane and sludge liquor. The deposition of biosolids on the surface will lead to the formation of a cake layer. Fortunately, this fraction of fouling is reversible and can be mitigated most of the time through physical cleaning. Another fraction of fouling is the pore blocking of membranes. This is irreversible but can be removed most of the time through chemical cleaning (Chang et al. 2002). According to Drews (2010), the factors affecting membrane fouling are the membrane characteristics, the sludge characteristics, and the operational conditions of the membrane and biological unit. Concerning the membrane properties, increasing the hydrophilicity of the membrane surface (Kumar & Ismail 2015) or modifying the membrane surface geometry to improve the hydrodynamics (Ilyas et al. 2022) could lead to lower fouling rates (FRs). From an operational point of view, the operational fluxes could be lowered, but then larger membrane areas are necessary to achieve the desired permeate volume. Another solution could be to increase the cross-flow velocity, however, this requires more energy (Chang et al. 2002).

Regarding the impact of biomass characteristics, it is reported that the bio flocculation level of the biomass significantly affects the fouling rate (FR) and filtration performance (Van den Broeck et al. 2010). Bio flocculation is an indicator of the biomass condition and can be characterized by the extracellular polymeric substance (EPS) content, the hydrophobicity level, the particle size, the morphological structure, or the presence of specific microorganisms. Stronger relations between bio flocculation and filtration performance have been found in some of the above-mentioned biomass properties, such as hydrophobicity. There are also studies with conflicting results about the impact of biomass properties on sludge filtration. For instance, the effect of the EPS content is questionable (Van De Staey et al. 2015). Additionally, there is very little research addressing the impact of the metabolism of microorganisms on filtration performance. Corsino et al. (2022) found that activated sludge enriched in polyhydroxyalkanoate (PHA)-storing microorganisms by the feast/famine regime presented lower fouling, however, microorganism identification was not performed.

An innovative fouling mitigation strategy, related to the morphology and the microbial community of the biomass, is the coupling of aerobic granular sludge with an MBR (AGSMBR). The rigid and compact structure of aerobic granules affects the filtration performance in a positive way (less fouling) (Wang et al. 2013; Liebana et al. 2018; Campo et al. 2021; Stes et al. 2021). The reason for this is probably that aerobic granular sludge (AGS) is more hydrophobic compared with conventional activated sludge (Liu et al. 2002).

An effective feeding strategy for aerobic granulation on a full-scale system was presented by Caluwé et al. (2022) for the treatment of industrial wastewater. The granulation strategy consisted of the introduction of a slow anaerobic mixed feeding step in the process operation of existing SBRs. As a result, microbial selection of microorganisms that store carbon anaerobically such as glycogen-accumulating microorganisms (GAOs) will be achieved. The selection of these microorganisms is of high importance in the full-scale AGS process, because it improves its stability (de Kreuk & van Loosdrecht 2004; van Dijk et al. 2022). To the best of our knowledge, no studies that evaluate the membrane filtration performance of full-scale AGS systems currently exist.

The first goal of this study was the quantification of the filtration performance of a pilot-scale ultrafiltration (UF) unit coupled to a full-scale AGS reactor treating industrial wastewater. The second goal was the evaluation of the effect of long-term UF on the characteristics of AGS.

The biological treatment step in the full-scale installation consists of two parallel SBRs, Bio1 and Bio2, daily treating together about 110 m³ of water. The main difference between the two reactors is the aeration system: Bio1 is equipped with fine bubble disc diffusers, while Bio2 is equipped with air tube diffusers. The wastewater originates from a tank truck cleaning (TTC) company, mainly cleaning trucks transporting chocolate and beer. The wastewater is poor in nutrients; therefore, the biological treatment is aimed at carbon removal only. The SBR cycles consisted of 4 h anaerobic feeding, 16 h aerobic reaction, 2 h settling (during the start-up), and 2 h discharge. A more detailed description of the full-scale installation can be found in Caluwé et al. (2022). The most important feature of the installation is the feeding strategy, which involves an anaerobic feast/aerobic famine regime to select slow-growing carbon-storing organisms which lead to granulation (de Kreuk & van Loosdrecht 2004).

For comparison purposes, biomass samples were collected from another company with similar activities. The treatment process also includes pre-treatment with dissolved air flotation and biological treatment with a continuously aerated SBR (Bio3).

A pilot installation containing four UF membranes in series was connected with one of the two full-scale SBRs. Each membrane had a surface area of 27.2 m². One membrane was standing, which means the actual total membrane area used for filtration was 81.6 m². It was decided to couple Bio1 with the UF installation first because Bio1 presented a more stable performance compared to Bio2, with respect to settleability, sludge morphology, and anaerobic uptake capability. The pilot was operated for approximately 10 h per day, hereby producing approximately 50 m³ permeate. The feed stream for the membranes was pumped out from the reactor during the aeration phase, first entering a roto-sieve to remove large solid particles such as fibers. The desired permeate flow was about 6 m³·h⁻¹. To preserve stable permeate flow, the cross-flow velocity (CFV) was increased from 2 to 2.5 m·s⁻¹ when necessary. Biopulses, backwashing, and chemical backwashing were applied as fouling mitigation measures (check the supplementary material for the process cycle). The UF filtration test consisted of three stages. In stage I (approximately 2 months), the membrane unit was connected to Bio1. Then, in stage II (approximately 1 week), it was decided to couple the installation with Bio2 to compare the two systems. After switching back to Bio1, filtration continued for 11 more days (stage III).

Sludge samples taken from the two SBRs were delivered frequently and analyzed in the lab for: filterability, settleability, anaerobic substrate storage capability, morphology, particle size, and microorganism abundance.

The sludge mixed liquor suspended (volatile) solids (ML(V)SS) and the sludge volume index (SVI) were measured according to the standard methods (APHA 1998). Particle size distribution was measured with A Malvern Mastersizer 2000 Ver. 6.00. Sludge morphology was observed using an Olympus B310 light microscope with phase contrast illumination (Ph1) and a total magnification of 2 times. Microbial community identification was performed with 16S rRNA amplicon sequencing as described by Tsertou et al. (2022).

Dissolved organic carbon (DOC) uptake (%) during the anaerobic period (120 min) was calculated as described by Tsertou et al. (2022). The sludge was always collected at the end of an aeration phase, to ensure that the sludge was in the endogenous phase. 400 mL of the sampled sludge was transferred to a beaker on a magnetic stirring plate. Then 100 mL of wastewater was added. During the test, mixed liquor samples were taken at three time points: (1) before adding the wastewater, (2) after adding the wastewater, and (3) after 2 h of anaerobic mixing. The samples were filtered over a glass microfiber filter (particle retention 1.2 μm). DOC was then analyzed using a Sievers innovox TOC analyser.

Two clean water filtrations (CWF) were performed, before (CWF1) and after (CWF2) the sludge filtration test. A slightly modified clean water protocol, based on the one described by Stes et al. (2021), consisted of 10 min of filtration at a flux of 40 L·m⁻²·h⁻¹ (0.011 × 10⁻³ m³·m⁻²·s⁻¹). When sludge filtration was over, the cake layer on the membrane surface was rinsed with demineralized water. Assuming that the resistance of the membrane due to the cake layer (Rc) is zero, then both the net loss of permeability and the net increase of membrane resistance could be attributed to pore blocking.

R indicates total resistance or Rcwf2 (m⁻¹); Rm indicates membrane resistance or Rcwf1 (m⁻¹); Rf indicates fouling resistance (m⁻¹); Rp indicates pore blocking resistance (m⁻¹); Rc indicates cake layer resistance (m⁻¹); μ indicates viscosity for demineralized water (=0.001 Pa. s).

The granular sludge of the two full-scale SBRs was cultivated by an anaerobic mixed feeding step followed by an extended aerated step without selective discharge of slow-settling sludge (Caluwé et al. 2022). The formed granular sludge presented a compact structure with excellent settleability (SVI 5 equal to 56 and 68 mL·g⁻¹ for Bio1 and Bio2, respectively), large particle size (dv50 equal to 158.62 and 144.68 μm for Bio1 and Bio2, respectively), and a high percentage of read abundancies of glycogen-accumulating microorganisms (GAO) (29.8 and 18.9% for Bio1 and Bio2, respectively). It can be said that the granular sludge of Bio1 performed slightly better than Bio2.

The stability of the AGS was hindered by a biomass overloading episode. The average COD was 5,741 ± 639 mg·L⁻¹, whereas the average COD before and after the overloading episode was 4,303 ± 160 and 3,869 ± 355 mg·L⁻¹, respectively. The fact that industrial wastewater often presents large variations could account for these episodes. The measured COD in the wastewater on DAY 34 was 7,736 mg·L⁻¹. The loading of biomass increased up to 73% within only 6 days. Furthermore, the average COD values in the influent remained relatively high for the next 20 days.

According to Stes et al. (2021), the critical flux is the flux at which the FR remains below 0.5 mbar·min⁻¹. Based on this definition, no critical flux was reached for Bio1. The threshold value of 0.5 mbar·min⁻¹ was never exceeded even at the highest applied flux (55 L·m⁻²·h⁻¹). For Bio2 on the other hand, the critical flux was between 30 and 35 L·m⁻²·h⁻¹, as the FRs were 0.49 and 0.78 mbar·min⁻¹, respectively. To better understand the filtration performance of the cultivated sludge (Bio1 and Bio2) under the feast (anaerobic)/famine (aerobic) regime, a sample from a company (Bio3) with similar activities was collected, measured, and compared. The feeding strategy of the biological reactor is the main distinction between the two companies. The biological treatment (Bio3) includes only aerobic feeding, settling, and discharge. The particle size of Bio3 (dv50) was equal to 70.4 μm, and its biomass structure is illustrated in Figure 8. Both the FR and the permeability of Bio3 were consistently worse than Bio1 and Bio2. Bio3 presented higher FRs at all applied fluxes; with the critical FR being already obtained at a flux of 15 L·m⁻²·h⁻¹. In practice, the filtration performance of Bio3 indicates that the compensation of permeability loss during long-term operation will require more chemical cleaning of the membranes, as well as larger membrane surfaces (Chang et al. 2002).

The membrane resistance due to pore blocking for Bio3 was 2.5 times larger than Bio2, and 7.5 times larger when compared to Bio1 (Rp_Bio1 was 0.06 × 10¹² m⁻¹, Rp_Bio2 was 0.18 × 10¹² m⁻¹, Rp_Bio3 was 0.45 × 10¹² m⁻¹).

The monitoring of different sludge characteristics from both SBRs took place to gain insight into the impact of filtration on sludge.

Figure 5 shows the settling performance of the sludge in the two full-scale SBRs. The average sludge concentrations were 7.4 and 6.1 g·L⁻¹ for Bio1 and Bio2, respectively. The biomass settled very well in both reactors during the first half of stage I (ongoing filtration on Bio1). The SVI value was approximately 33 mL·g⁻¹ for both on DAY 0. An increase of the SVI30, reaching close to 100 mL·g⁻¹, was observed on DAY 41 for Bio2 and on DAY 50 for Bio1. The deterioration was attributed to the overloading of the two bioreactors. Although the settleability of Bio1 and Bio2 evolved similarly during stage I, showing a good performance initially, it was followed by a worsening in the settleability; Bio1 seemed to withstand the overloading episode better. The highest SVI30 value was 105 mL·g⁻¹ for Bio1 and 211 mL·g⁻¹ for Bio2. Furthermore, Bio1 presented an improvement after DAY 88, but no change was observed for Bio2. This evolution indicates that the filtration process was not responsible for the changing settleability.

Figure 6 shows the anaerobic uptake ability of the microorganisms. Anaerobic uptake batch tests are a useful method to confirm that the microorganisms with a slow-growing metabolism have been enriched. Before the overloading episode, the average DOC uptake was above 80% for both Bio1 and Bio2. These values of uptake were satisfactory for industrial wastewater, as complete uptake is impossible. The consequences from the overloading episode were first detected in Bio2 (around DAY 42) and later in Bio1 (DAY 56) during stage I (ongoing filtration on Bio1). The delayed deterioration of anaerobic carbon uptake percentage in Bio1 proves that Bio1 contained better quality AGS in terms of DOC removal. Furthermore, the disturbance of the anaerobic uptake in both reactors indicates that the filtration process did not affect the metabolism of the slow-growing microorganisms (e.g. GAOs). The overloading episode was overcome once the restoration of the anaerobic metabolism (average percentage of anaerobic DOC uptake was above 90% for Bio1 and above 80% for Bio2) occurred.

Figure 8 shows the changes in sludge morphology during the operation of the UF pilot. In both reactors, the sludge consisted of compact granules on DAY 0. Bio1 contained well-shaped spherical aggregates. Bio2 contained more irregular aggregates, with filamentous bacteria extending from them. The biomass morphology changed significantly over time for both reactors. The morphology was consistently in line with the particle size distribution (Figure 4). Smaller granules popped up until DAY 42 for Bio Ι; it is believed that until that day, only the filtration operation affected both the morphology and particles' size. Later (until DAY 63), more deflocculated sludge (loose structures without specific demarcation) dominated. Both the filtration and the overloading episode could have affected the morphology during that period. On DAY 90 (7 days after the UF installation was shut down), there was a small improvement in the morphology of Bio1. Furthermore, the sludge seemed to have recovered from the overloading episode (SVI30_Bio1 was 51 ml·g⁻¹, and the anaerobic carbon uptake percentage was 94%). The microscopic examination for Bio2 showed that the biomass morphology did not change a lot during the first 40 days. The overloading episode drastically affected the biomass morphology; on DAY 55, the excessive presence of filaments appeared, and the granules almost disappeared. Both the biomass morphology and SVI30 of Bio2 had not recovered yet by DAY 90. The lower recovery rate of Bio2 in comparison to Bio1 confirmed that Bio1 consisted of better quality biomass.

In this study, we show that the mitigation of fouling could be achieved when the filtered biomass is well-formed AGS. There was a convergence of lab and full-scale results for membrane filtration performance. Despite the criticism about the interpretation of lab-scale filtration results to predict full-scale performance (Drews 2010), the results confirmed that the sludge quality affects the permeability in both cases. However, it is believed that the filtration performance should be evaluated over an extended period of time. Furthermore, reporting data from full-scale installations, especially for industrial wastewater cases, should be encouraged more. Building a repository of full-scale data will facilitate the subjective comparison of results of filtration performances of companies with similar activities at a full-scale level.

In this reported case, the cultivation of the AGS in an SBR took place first (Caluwé et al. 2022), with the side stream MBR being integrated later on. The SBR is considered an excellent configuration for direct cultivation of AGS because it is easy to apply the feast (anaerobic)/famine (aerobic) regime, while at the same time, it occupies minimal space. The aerobic granules were obtained under minimum washout conditions, which makes the granulation process slower (Liebana et al. 2018). Furthermore, the absence of selective wasting resulted in smaller granules (particle size < 200 μm) (van Dijk et al. 2022). The aerobic granular membrane bioreactor (AGMBR) was operated in batch mode during this study. The filtration started when the anaerobic feeding was completed, then followed by 2–3 h of aeration to ensure that the non-stored organic substrate was degraded below 100 mg·L⁻¹. This action is considered auxiliary for the membranes. The leaking of a part of the substrate into the aerobic phase is inevitable (Tsertou et al. 2022) since the wastewater consists of both easily and slowly biodegradable COD. During the anaerobic phase, the easily biodegradable substrate was taken up and stored as biopolymers (e.g. PHAs stored by GAOs in this study) (Figure 7(a)). However, part of the slowly biodegradable substrate will flow into the aerobic phase, which is known as the extended feast phase (van Dijk et al. 2022).

Small granules with excellent settling properties were obtained in a full-scale system, with a minimum washout of poor settling biomass. The aerobic granules of this study were cultivated in a 24 h cycle under an SRT of 30 days, slow anaerobic feeding (4 h), a long settling time (30–120 min), low OLR (the average OLR was 0.54 ± 0.02 kg COD·m⁻³·day⁻¹), and a DO between 1 and 4 mg·L⁻¹. In this study, the overloading episode had the greatest impact on the morphology of the granules. The instability of aerobic granules to higher applied OLR has been reported extensively (Adav et al. 2009). Aerobic granules' formation has been achieved under different OLR values (low, medium, and high) (Iorhemen & Liu 2021). Under low OLR (lower than 2 kg COD·m⁻³·day⁻¹), smaller granules are formed. The small size results in the absence of an oxygen gradient within the granules. By avoiding anaerobic conditions inside the granules, due to the efficient penetration of the oxygen into the core of the granules, the disintegration of the granules is avoided, resulting in more stable granulation (Long et al. 2015). Although the low OLR is in favor of the stability of the granules, it negatively impacts the time needed for granulation (Liu & Tay 2015). This study showed the intolerance of the cultivated aerobic granules to the sudden increase of OLR (from 0.56 to 0.97 kg COD·m⁻³·day⁻¹). The sudden OLR increase (73%) disrupted the granular sludge, and the compact, well-shaped granules were replaced by loose structures. The disruption in the sludge morphology was worse for Bio2, where there was an overgrowth of filamentous bacteria. It is believed that the higher OLR resulted in a more inadequate dissolved oxygen concentration in Bio2. Since the only difference between the two full-scale reactors was the aeration system, it seems that the air tube diffusers of Bio2 are less efficient for oxygen transfer in comparison with the fine bubble disc diffusers of Bio1. All in all, the results show that if the OLR increase is not gradual, the structure of the granules will be affected negatively.

Many studies have investigated the filtration performance of conventional activated sludge and AGS, and proven the superiority of AGS (Li et al. 2005; Liebana et al. 2018; Stes et al. 2021). Li et al. (2005) reported a permeability loss of up to 50% in the conventional floccular sludge MBR, compared to aerobic granular MBR. Stes et al. (2021) reported a reduced FR for AGS in comparison with floccular sludge. However, the majority of such studies were conducted under different experimental conditions, different experimental set-ups (submerged or side stream membranes), and different filtration protocols (TMP applied protocols or flux applied protocols), which limits the comparison between them. Nevertheless, most of the studies attribute the better filtration performance of AGS to its dense structure. In this study, the filtration performance (Figure 3) can also be related to sludge morphology (Figure 8). Bio1 consisted of well-shaped granules on DAY 42. Bio2 consisted of more irregular granules on DAY 42, with more filaments present. The biological community analysis (Figure 7) showed more filaments for Bio1 compared to Bio2 on DAY 42, which is not depicted in the morphological examination. However, it should be highlighted that Candidatus Villigracilis, which was more abundant in Bio1, is often located inside the flocs/granules (MiDAS Field Guide). On the other hand, Kouleothrix, which was more abundant in Bio2, has been connected with bulking episodes (Tsertou et al. 2022). This could explain why Bio2 presented worse settleability than Bio1, while both reactors had similar profiles of GAOs.

Differences in filtration between Bio1 and Bio2 were observed in the pilot- (Figure 2(a)) and lab-scale (Figure 3) systems. In the pilot scale, when the UF MBR was coupled with Bio2, a sharp drop in permeability was noted. The permeability was more than double in Bio1. This confirmed that Bio1 had a better quality of granules than Bio2. The results of Stes et al. (2021) validate the lab-scale results of this study. In that study, the FR for the AGS did not exceed 0.1 bar·min⁻¹ when high fluxes were applied. Furthermore, the FR for the floccular sludge already exceeded 0.5 mbar·min⁻¹ at fluxes between 10 and 15 L·m⁻²·h⁻¹. Additionally, the comparison between the two studies showed that the less well-performing sludge in this study was better than the floccular sludge in the study of Stes et al., since the critical flux was identified at higher fluxes (30–35 vs. 10–15 L·m⁻²·h⁻¹). This indicates less fouling for the sludge cultivated with the anaerobic (feast)/aerobic (famine) strategy in a full-scale SBR.

The maintenance of granular stability is one of the biggest challenges in AGMBR. The shock overload during stage I had an impact on the sludge settleability, the morphology, the ability for anaerobic uptake, and the microbial community (especially GAO read abundancies) in both Bio1 and Bio2. Industrial wastewaters present more variable compositions compared to municipal wastewaters. In essence, the microorganisms will experience overloading episodes. An overloading episode unsettles the feast (anaerobic)/famine (aerobic) system (Tsertou et al. 2022). Then, more substrate will be available in the aerobic phase, which gives an advantage to filaments and ordinary heterotrophic microorganisms in the competition with GAOs. It seems that the GAOs did not directly affect the permeability of a full-scale system. The permeability was very satisfying at the end of stage I, while at the same time, the GAOs reached their lower read abundancies in Bio1. On the other hand, GAOs affect filtration performance indirectly by contributing to strong granule formation even when they are present at very low abundances. However, a long-term overloading episode could be a serious issue for two reasons. Firstly, the recovery of the system would take longer. Secondly, the increased OLR will also lead to increased sEPS, and more particularly the sPS (soluble polysaccharides) fraction, which is considered the main cause of membrane fouling (Thanh et al. 2008).

The reuse of treated water is imperative as the supply–demand relationship of water has been disrupted. Furthermore, future solutions should have a limited footprint. The AGSMBR technology addresses both of these points and presents extra advantages compared to conventional MBR, including less fouling. In this study, we show the granular sludges' filterability performance in a full-scale WWTP. Aerobic granules were cultivated in a full-scale SBR system under the feast (anaerobic)/famine (aerobic) regime and minimum washout conditions. In practice, an SBR can be easily coupled with an UF installation in order to be transformed into a side-stream MBR. In addition, maintaining the anaerobic/aerobic feeding strategy, which is a necessary condition to stimulate the slow-growing microorganisms (GAO), is also feasible in this coupling. In that way, the novel configuration of an AGSMBR can be achieved. The critical fluxes for AGS were lower in comparison with conventional activated sludge, indicating the possibility of higher applied fluxes and by extension, a lower required membrane surface area. Further research about long-term filtration operations at full scale, as well as the stability of AGS in overloading episodes, is recommended.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.