Full dynamic control of dairy wastewater treatment by aerobic granular sludge using electric conductivity and oxygen uptake rate

Within the food industry, the dairy sector is one of the largest producers of industrial wastewater in Europe (Kolev Slavov 2017). On average, the industry produces 2.5 times more wastewater than dairy products. Biological processes represent a central step in dairy wastewater treatment, which could be further optimized by adopting the novel aerobic granular sludge (AGS) technology (Schwarzenbeck et al. 2005; Wichern et al. 2008; Silva et al. 2023). AGS relies on the formation of microbial self-immobilized aggregates, resulting in a very compact and dense structure. Therefore, it has better settleability, larger biomass concentration, higher resilience to toxicity, capability to operate at higher organic and shock-loading rates, and the possibility to biodegrade organic carbon and remove nutrients simultaneously, with less sludge production and a lower footprint (Campo et al. 2020). Although there is still debate on which factor is the key driver for stable granulation, de Kreuk & van Loosdrecht (2004) showed the decisive role of slow-growing organisms and more specifically of phosphate (PAO) and glycogen (GAO)-accumulating organisms. The selection of GAO and PAO, through an anaerobic feast/aerobic famine regime, leads to the formation of dense granular sludge, because the uptake and conversion rate of the substrates by the granules is lower than the transport rate into the granules (van Dijk et al. 2022). Other researchers stress the importance of a hydraulic (settling) pressure to get rid of the bad settling floccular sludge fraction, while critical organic loading rates, shear stress, and so on, have all been proposed as equally relevant (e.g. Franca et al. 2018). Furthermore, AGS produces a valuable gel-forming biopolymer referred to as alginate-like extracellular polysaccharides (ALE), responsible for the strength, elasticity, and hydrophobicity of AGS (Schambeck et al. 2020).

PAOs act in the enhanced biological phosphorus removal (EBPR) process, operated under alternating anaerobic and aerobic conditions. Under anaerobic feast conditions, PAOs use the energy from hydrolyzing phosphodiester bonds of stored polyphosphates to convert and store volatile fatty acids (VFAs) as polyhydroxyalkanoates (PHAs). In this phase, PAOs release the orthophosphate (PO₄-P) in the bulk liquid (Izadi et al. 2020). Under aerobic famine conditions, the stored PHA is oxidized, and the energy released is mainly used for biomass production, PO₄-P absorption, and polyphosphate synthesis. More PO₄-P is absorbed in the aerobic step than released in the anaerobic step, leading to a net decrease in the liquid PO₄-P concentration. GAOs, on the other hand, have a similar metabolism, i.e. VFA conversion to PHA under anaerobic feast conditions followed by PHA oxidation under aerobic famine conditions, but without the associated P cycling.

EBPR is a promising process for treating the PO₄-P-rich dairy wastewater, with P concentrations 5–20 times higher than in municipal wastewater (Kushwaha et al. 2011). An efficient EBPR process could meet stringent effluent qualities (1–2 mg P L⁻¹) if a careful design and operation are practiced (Ahn & Park 2008). Industrial application of EBPR however is not widespread. A recent survey among 90 of the 354 industrial biological wastewater treatment plants in Flanders showed that, of the 39 companies requiring P removal, none applied EBPR but instead removed P by precipitation after chemical treatment with ferric iron chloride or poly aluminium chloride (Cornelissen et al. 2018). Possible reasons are the perceived complexity and the lack of knowledge of the EBPR process, the ease of the chemical alternative, and the low cost of chemicals. A specific concern with the EBPR process is how it handles wastewater variability, which could be addressed by implementing a dynamic control strategy that measures the variability in wastewater composition and adjusts the operation of the treatment accordingly. While specific nutrient analysers and sensors are available on the market today, they require high investment and maintenance costs (Olsson 2012; Zhang et al. 2022). As a result, they are barely applied in industry (Cornelissen et al. 2018). An alternative approach is the use of low-cost and robust indirect sensors. Control strategies for biological nitrogen (N) removal using pH, redox potential, and dissolved oxygen (DO) sensors have been developed and extensively tested (e.g. Yang et al. 2010; Ribeiro et al. 2017), but much less so for P removal. In EBPR systems, conductivity sensors could be used, as changes in the conductivity are strongly related to the dynamics of PO₄-P uptake and release. In the anaerobic step, the conductivity increases because of the release of PO₄-P ions in the bulk liquid (Maurer & Gujer 1995). The conductivity profile can thus be used to dynamically control the anaerobic step in an EBPR AGS system. Short-term studies showed stable N and P removal by AGS with varying influent concentrations (Kishida et al. 2008). Long-term studies have verified the effectiveness of conductivity-based control while treating P-rich synthetic wastewater with variable composition (De Vleeschauwer et al. 2019). To the best of our knowledge, no industrial application of a conductivity-based dynamic control has been reported.

The objectives of this study are therefore to determine the applicability and effectiveness of a sensor-based dynamic control strategy for an EBPR AGS reactor treating real variable dairy wastewater.

Two parallel AGS SBRs were operated using an anaerobic/aerobic reaction sequence for 92 days. SBR-STA was operated statically with a fixed length of the different steps in the SBR cycle. SBR-DYN, on the other hand, was operated dynamically using conductivity-based dynamic control to adapt the length of the anaerobic step, and oxygen uptake rate (OUR) based control to adapt the length of the aerobic step. The reactors were seeded with activated sludge from a local sewage treatment plant with confirmed EBPR activity (Dockx et al. 2021). The sludge retention time in both reactors was kept at 30 days throughout the study.

The reactors had a diameter of 24 cm and a height of 40 cm, corresponding to a height to diameter ratio (H/D) of 1.66. The volume exchange rate (VER) was 8%, with a working volume of 12.6 L and an influent volume per cycle of 1 L. The reactor was equipped with a peristaltic pump for influent feeding (Watson Marlow^®), a mixer (Heidolph^® RZR2020), a discharge valve (Eriks RX ER10.X33.S00), and an aeration system consisting of an aeration pump (koi flow 60, Aquadistri China^®) and a 13 cm aeration disc (Aquadistri China^®). Furthermore, the reactors were equipped with a luminescent dissolved oxygen (LDO) sensor (Hang Lange^®) and a conductivity sensor (Hang Lange^®). The DO concentration in the aerobic step was controlled using an on/off aeration control strategy. The aeration pump was activated when the DO concentration decreased below 1 mg L⁻¹ and deactivated when the concentration increased above 2 mg L⁻¹. This aeration control allowed for the calculation of the OUR, as described by Dobbeleers et al. (2017). All sensor values were logged, and archived at the end of each cycle.

The wastewater originated from a local full-scale plant treating dairy wastewater. The wastewater treatment consists of chemically enhanced primary treatment, using ferric chloride, in a dissolved air flotation unit to remove suspended solids and P, followed by a conventional activated sludge plant performing biological carbon and N removal. The raw wastewater was collected before the DAF unit, on a weekly basis, and kept in a fridge at 4 °C to minimize degradation. Before use, the wastewater was chemically pre-treated using ferric chloride (at 100 mg L⁻¹ Fe) to remove the suspended solids (De Vleeschauwer et al. 2021). The average total and filtered chemical oxygen demand (COD) and suspended solids concentration in the raw wastewater were 2,158 ± 714 mg L⁻¹, 1,096 ± 515 mg L⁻¹, and 792 ± 300 mg L⁻¹, respectively. The variable composition of the wastewater, after pre-treatment, is shown in Table 1. Note that PO₄-P was added to the chemically pre-treated influent to reflect the average concentration in the raw influent.

CV1, CV2, and CV4 were 20 s, 30, and 30, respectively. The slope setpoint CV3 needed to be adjusted regularly to achieve effective endpoint detection, probably due to the variable composition of the industrial wastewater. CV3 varied between 0.8 and 3 mS (cm d)⁻¹.

The aerobic step in SBR-DYN was controlled dynamically based on the OUR value. When three consecutive OUR values were below a biomass-specific OUR (sOUR) of 2 mg (g_MLVSSh)⁻¹, the aerobic step was terminated (De Vleeschauwer et al. 2019).

COD, PO₄-P, NH₄-N, NO₃-N, and dissolved organic carbon (DOC) were analysed using standard methods, as described by De Vleeschauwer et al. (2019). All the analyses except total COD were performed on prefiltered samples (VWR^® glass microfibers filter 693, particle retention: 1.2). The mixed liquor suspended solids concentration (MLSS), the mixed liquor volatile suspended solids concentration (MLVSS), and the sludge volume index (SVI) were measured according to the standard methods (APHA 1998). The sludge particle size distribution (DV10, DV50, and DV90) was measured with a Malvern Mastersizer 3000, as described by Caluwé et al. (2017).

The alginate-like extracellular polymer (ALE) content was measured by extracting the ALE at high temperature (80 °C) in a sodium carbonate solution (Na₂CO₃) (Felz et al. 2016). Sludge sampled from the reactors at the end of a cycle was centrifuged at 3,500 rotations per minute (rpm) for 10 min. Approximately 3 g of centrifuged sludge was resuspended in a sodium carbonate solution (0.5% w/v Na₂CO₃) and mixed at 400 rpm and 80 °C for 35 min. Afterwards, the mixture was centrifuged, and the supernatant, with the dissolved ALE, was collected. The pH of the supernatant was then adjusted to 2.2 ± 0.05 using a 1 M HCl solution to obtain the ALE in its acidic form. The solution was centrifuged, and the pellet, which is ALE in its acidic form, was retained. The pellet was then resuspended in demineralized water and centrifuged twice to remove any ions. Finally, the pellet was dried overnight at 105 °C and weighed.

De Vleeschauwer et al. (2019) reported the successful application of a sensor-based control strategy to adapt the length of the anaerobic and aerobic reaction steps in an AGS SBR treating P-rich synthetic wastewater with COD/P values ranging from 21 to 116. This study investigated the applicability of this control strategy when treating real industrial wastewater from the dairy industry with variable COD content and COD/P ratios ranging from 27 to 151 (Table 1).

The changes in the conductivity profile were better detected by using the MCS that showed a low and constant value when PO₄-P release and DOC uptake were finished (Figure 3(b)). The increase in conductivity is associated with the release of PO₄-P and the additional release of potassium and magnesium ions (Aguado et al. 2006). In studies using synthetic wastewater, a clear break in the conductivity profile and a sharp drop in the MCS are observed at the end of PO₄-P release, which is related to the complete uptake of the simple carbon substrate applied, mostly acetate or propionate only (e.g. Kishida et al. 2008; Weissbrodt et al. 2014; De Vleeschauwer et al. 2019). In contrast, in the present study, we observed a more gradual change in the conductivity and MCS profiles (Figure 3(a) and 3(b)) which is probably related to the more complex composition of the real dairy wastewater (Meunier et al. 2016). In addition, the value for CV3, which is the MCS setpoint (Figure 2), had to be regularly adjusted to allow successful endpoint detection when a new influent batch was treated. This was not the case when treating VFA-based synthetic influent even when the COD and PO₄-P concentrations varied significantly (De Vleeschauwer et al. 2019). In future research, more robust endpoint detection methods could be explored e.g. methods that are based on pattern recognition rather than on absolute setpoints to avoid this issue (e.g. Marsili-Libelli 2006).

The dynamic control of the aerobic reaction step, based on the OUR, is considered an established method and will not be discussed in detail. In brief, the aerobic reaction step was successfully terminated when three consecutive sOUR values were recorded below an endogenous respiration threshold value of 2 mg (g_MLVSS h)⁻¹ as extensively described in earlier research (e.g. Dries 2016; De Vleeschauwer et al. 2019, 2021).

The correlation between the duration of the anaerobic and aerobic steps and the influent COD concentration was statistically significant (Figure 4; p < 0.0005) indicating that the variation in the influent COD significantly contributes to the variation in the duration of both steps. This is explained by the specific metabolism of PAO in an EBPR system, where the uptake of COD from the liquid phase (and the subsequent intra-cellular storage) is driven by the release of PO₄-P which, in turn, is related to the conductivity profile. The close relation between COD uptake and PO₄-P release is expressed in the typical molar PO₄-P release to carbon uptake ratio of about 0.5 Pmol Cmol⁻¹ recorded in PAO-enriched systems fed with VFAs (Diaz et al. 2022). The dynamic control strategy thus ensures that all the assimilable carbon is taken up by PAO and/or GAO during the anaerobic reaction step, and prevents leakage of COD to the subsequent aerobic phase. The average DOC uptake during the anaerobic reaction step was 94 ± 4 and 92 ± 5% in SBR-DYN and SBR-STA, respectively. These uptake values are in the same range as previously observed in AGS systems fed with VFAs (De Vleeschauwer et al. 2019; Dockx et al. 2021). The near-complete substrate uptake during the anaerobic step promotes PAO (or GAO) over ordinary heterotrophic organisms and filamentous bacteria, and is key to the microbial selection strategy applied for granulation (Haaksman et al. 2020, 2023).

Table 2 shows the difference in the reactor performance parameters between both reactors. The effluent COD and PO₄-P concentrations were well below the Flemish discharge limits for surface water (125 and 2 mg L⁻¹, respectively). We should mention that, although the total phosphorus concentration (TP) was not measured, it is expected to be equally low as the average effluent suspended solids concentrations was below 10 mg L⁻¹ for both reactors (data not shown). There was no statistical difference in PO₄-P removal efficiency between SBR-STA and SBR-DYN. In contrast, the COD removal efficiency was statistically higher in SBR-DYN. No clear reason for this difference was found, but more importantly, it is operationally not significant, as both the COD and PO₄-P removal efficiencies are high, and in range with values reported in previous works for AGS and conventional SBR systems treating dairy wastewater (Arrojo et al. 2004; Kushwaha et al. 2011; Bumbac et al. 2015; Meunier et al. 2016). With respect to N, full NH₄-N removal and almost complete N removal were observed in both reactors (results not shown).

Although granular sludge was clearly present in both reactors (Figure 7), full granulation was not obtained, and flocks were still present at the end. This hybrid floccular-granular sludge morphology was observed previously in AGS reactors treating real industrial wastewater (De Vleeschauwer et al. 2021; Caluwé et al. 2022) and can be explained by the granulation strategy applied. The main granulation driver applied in the present study was the implementation of an anaerobic feast/aerobic famine reactor operation. This reactor operation leads to the enrichment of slow-growing PAO and/or GAO that first store COD (during the anaerobic feast phase) and then consume the stored COD (during the aerobic famine phase). This strategy relates to the so-called microbial selection approach which is considered to be a main granulation mechanism (van Dijk et al. 2022). However, to achieve full granulation, additional mechanisms may have to be applied. These mechanisms are physical (or hydraulic) selection by selectively wasting the bad settling floccular sludge fraction, and selective feeding of the larger granule by plug-flow feeding from the bottom of the reactor. In the present study, no selective sludge wasting was applied and the influent was pulse fed into the completely mixed reactor, making microbial selection the only mechanism used. Although specific microbial analyses were not performed, the observed P release and P uptake dynamics clearly indicate PAO activity in the reactors. The average molar P release to C uptake ratio Pmol Cmol⁻¹ was 0.34. This ratio is lower than the value of approximately 0.5 found in fully enriched PAO systems, indicating that besides PAO also GAO were enriched in the AGS reactors (López-Vázquez et al. 2007; Weissbrodt et al. 2014). As mentioned above, both PAO and GAO are key organisms in aerobic granulation (Haaksman et al. 2023).

The current study showed, for the first time, the successful application of a sensor-based dynamic control strategy in AGS performing EBPR from real industrial wastewater. The control strategy uses common, robust, and low-cost sensors such as DO and electrical conductivity to adapt the duration of the aerobic and anaerobic reaction steps depending on the variable nature of the influent wastewater and the activity of the (granular) activated sludge. This results in a dynamic reactor operation with significant efficiency gains in the AGS process, such as lower cycle times and higher sludge loading rates without impairing the effluent quality nor the sludge properties. These results also extend the existing potential of indirect control strategies to full biological N and P removal processes, which may be of great assistance to the operators and designers of industrial installations for treatment optimization. Finally, the current study also demonstrated stable EBPR from variable dairy wastewater, holding great promise for the application of this sustainable P removal technology in industry.

This work was financed by the Industrial Research Fund (IOF) of the University of Antwerp.

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

The authors declare there is no conflict.