Densification of activated sludge for better settling performances: experimental characterization in batch column and model parameters calibration

In the conventional activated sludge (CAS) process, bacteria and wastewater are in contact (mixture) in a reactor to reduce the amount of organic material and other nutrients such as nitrogen and phosphorus. This biomass (activated sludge) grows and forms biological flocs that must be separated from treated water. This is usually performed by means of gravity in a secondary settling tank (SST). SSTs must complete three main functions:

• effluent clarification by producing an effluent with low total suspended solids (TSS) concentration,

• sludge storage during peak flows,

• sludge recycling back to the biological reactor: this requires a certain level of sludge thickening.

As a consequence, activated sludge sedimentation performance within the SST governs effluent quality directly in terms of TSS and indirectly as it will affect the biomass retention within the system (through recycling), thus affecting the biokinetic processes occurring in the biological reactor (Torfs & Maere 2015). Also, the intensification of this process would require increasing the mixed liquor suspended solids (MLSS) concentration.

In this context, sludge settleability appears as the limiting factor for SST design and operation, potentially making SSTs the bottleneck of this process (Ekama 1997). For instance, to handle high wet weather flows, deeper clarifiers have to be designed: this allows extended possibility to store sludge during peak flows without significant degradation of effluent quality (Jimenez et al. 2008). However, this does not impact the ability to maintain sufficient MLSS within aeration tank.

Sludge settleability can be impacted by several factors related to the present microorganisms (Jenkins et al. 2003). As an example, an increase in phosphorus-accumulating organisms (PAOs) in flocs has been attributed to improved settleability in CAS processes (Schuler et al. 2001). A good settleability is generally the consequence of an equilibrium between floc-forming and filamentous bacteria. The latter are long, thread-like shaped microorganisms, which are generally present in the activated sludge suspension. Their overabundance leads to sludge bulking, which is a common problem in activated sludge (Krhutková et al. 2002; Martins et al. 2004). It affects sludge settleability and compaction, yielding a high sludge volume index (SVI) and low TSS concentration in the return activated sludge (RAS) and waste activated sludge (WAS). In terms of design and operation, the following issues arise (Environmental Protection Agency 1987):

• limitation the acceptable surface overflow rate (SOR) and solids loading rate (SLR) of the SST,

• poor WAS dewaterability,

• hydraulic overloading of WAS handling systems,

• difficulties to maintain the required MLSS concentration in the aeration tank.

The causes of bulking sludge include mainly: low dissolved oxygen concentration in the aerated tank, nitrogen or phosphorus limitation, septic wastewater (presence of sulfide), low pH, presence of grease and oil, low F/M loading. Due to more and more stringent regulations, a large portion of aerobic activated sludge processes operates at very low F/M loading (extended aeration, oxidation ditches), making them likely to experience such issues. In winter conditions, the potential dilution with rainwater (combined sewer) worsens this problem.

Methods for bulking control include curative solutions (e.g. chlorination) or adapting operation (e.g. control of dissolved oxygen, control of influent septicity). It is also possible to implement an anoxic selector upstream the aeration basin (Shao & Jenkins 1988; Pujol & Canler 1994; Tampus et al. 2004).

To go further and tackle the settleability limitation of suspended growth processes, it is possible to select fast-settling micro-organisms and improve notably sludge settleability. In this respect, aerobic granular sludge (AGS) reactors gained an increasing popularity since the two last decades (Beun et al. 1999; de Kreuk & Van Loosdrecht 2004; Wan et al. 2009; Filali et al. 2012). However, this technology is mainly suitable for new plants or requires a substantial investment to retrofit existing facilities: indeed, it is operating as a sequencing batch reactor (SBR) and necessitates several biological reactors operating in parallel.

In this context, densification of continuous flow activated sludge has been proposed recently (Bott et al. 2018; Ford 2018; Avila et al. 2021). The implementation of the process remains relatively easy and can be carried out with little or no civil engineering work. The densification process is based on the implementation of a biological selection by upstream (anaerobic) selector(s) providing feast/famine conditions (Sun et al. 2021) to trigger granule growth coupled with an external gravimetric selection to retain and enrich granular biomass in the system. The selective wasting of flocs is physically carried out on the WAS stream by hydro-cyclones (Sturm & De Clippeleir 2020). Sludge with superior settling characteristics is recovered by underflow from the hydrocyclones to be sent back to the head of the biological reactor, while the sludge with low settling characteristics goes overflow and is evacuated towards the sludge line, constituting the new WAS of biological system.

In a previous work at a full-scale experimental facility operating densification for 18 months, Roche et al. (2022) investigated the sludge granulation dynamics that eventually led to establish a hybrid granule-floc ‘densified’ biomass. This previous work focused on biomass density, granules inventory and biomass particles size distribution. The aim of the present study is to characterize the settleability of such ‘densified’ activated sludge sampled from the same full-scale experiment. Batch settling tests were performed to identify the sedimentation regimes occurring and to derive 1D settling model parameters that can be used for benchmarking simulations.

The trial plant from which the control CAS sludge and the test ‘densified’ sludge were sampled featured a secondary biological stage in Contact-Anaerobic-Oxidation ditch configuration spread over 4 identical lines (total capacity 400,000 Population-Equivalents). The biological treatment lines are independent from one another, without any mixing of the activated sludge upstream of the clarifiers or at the recirculation (RAS). This feature allowed a comparison of two perfectly independent parallel lines (one densified, the other left intact) and under the same operating conditions (temperature, load, flow). The underlying objective was to be able to explicitly compare the differences in settling performance and SVI behaviour between ‘densified’ biomass (test sludge) and conventional biomass (control sludge). The plant already had a biological selection in place with 3 distinct anaerobic zones in series (contact zone followed by 2 anaerobic compartments in series), which provided a good feast-famine basis to grow aerobic granules on. The gravimetric selective wasting technology used in this study (inDENSE™, NEWport GmbH, Austria) facilitates the progressive bio-augmentation of organisms adopting the densest forms (dense aggregates, small aerobic granules), because they are preferentially retained in the system, to the detriment of the light forms (pin flocs, filamentous bacteria), which are now primarily extracted from the biological system. This is carried out by the means of hydrocyclones (Ford 2018). This selective wasting, combined with the appropriate biological selection, results in the establishment of a biomass densification in the biological system, the settling characteristics of which are then increased as compared to CAS.

Figure 1 shows this implementation, which consists of abandoning the current WAS line (1) and adding RAS pumping (2) to the hydrocyclone battery (3). The latter is generally installed onto or near to the biological tank. A return line to the head of the biological reactor is established for the retained dense sludge (4) as well as a new wasting line (5) to the sludge thickening stage for the light sludge. This wasting line can operate by gravity or by pumping.

Further details about the physical and biological set-up of the full-scale densification process trial are given in Roche et al. (2022).

An experimental settling column made of transparent material was used to allow visual observation. Its dimensions are 40 cm in diameter and 45 cm water level for a working volume of 56.5 L. The sludge was collected on site from the RAS line for both test and control biomass. It was immediately poured in the column. It was stirred with a metallic propeller to maintain a homogeneous suspension before settling measurements.

Initial MLSS for test and control sludges were 6.85 and 8.43 g·L⁻¹ respectively, as they were both sampled from recirculation streams. Clarified water from the outlet was used to perform the necessary dilutions to cover a range of MLSS concentrations.

Two ultrasonic transducers with different frequencies monitored the settling process (Figure 2). The transducers were placed at the sludge suspension surface to perform vertical measurements. The transducers were connected to a UB-Lab ultrasonic profiler (Ubertone, Schiltigheim, France): backscattering signal intensity, velocity and data quality profiles were recorded. The sludge suspensions were stirred for about 15 minutes before each measurement. The intensity of the backscattered signal was monitored: the observation of a log-linear attenuation profile ensured suspension's homogeneity. The stirring was then stopped to evaluate the settling process.

For more details about the ultrasonic measurements, one can refer to Locatelli et al. (2015) and François et al. (2016).

Diluted sludge volume index (dSVI) was measured five times a week as per the European standard method (EN 14702-1:2006 2006), using 1 L class A graduated cylinders (VWR^® ref 621-3839, France). Settling profiles in the same type of cylinders were also established once a week to determine hindered settling velocity (also called zone settling velocity (ZSV)) following the method described in Van Loosdrecht et al. (2016) (except for the cylinder diameter). These two tests were performed at a normalized MLSS concentration of 2 g·L⁻¹. This enabled to dynamically compare the settling behaviour of the two types of sludge while setting aside the MLSS concentration impact.

With:

• ⁠: Mixture center of mass velocity (convective velocity), m·s⁻¹

• ⁠: Continuous phase (water) velocity, m·s⁻¹

• ⁠: Dispersed phase velocity, m·s⁻¹

• ⁠: Drift velocity (hindered settling), m·s⁻¹

• ⁠: Continuous phase (water) volume fraction, Dimensionless

• ⁠: Dispersed phase (sludge) volume fraction, Dimensionless

• ⁠: Continuous phase (water) density, kg·m⁻³

• ⁠: Dispersed phase (sludge) density, kg·m⁻³

• ⁠: Turbulent diffusion coefficient, m²·s⁻²

• t: Time, s

• ⁠: compression function, m²·s⁻²

With:

• X: sludge concentration, kg TSS·m⁻³

• ⁠: Maximum settling velocity, m·h⁻¹

• ⁠: Hindered settling parameter, m³·kg⁻¹

With:

• g: Gravity acceleration, 9.81 m·s⁻²

• is the derivative of the effective solid stress which primitive is expressed as in Torfs et al. (2017):

  

  (4)

With:

• ⁠: Effective solid stress parameter, m²·s⁻²

In this work, simulations were performed in 1D using a mesh of the batch settling column with 90 cells (layers). Full description of the solver is provided in Valle Medina & Laurent (2020).

The identification of the best set of parameters for the settling model was performed using Design Analysis Kit for Optimization and Terascale Applications (DAKOTA, https://dakota.sandia.gov). This open-source (GNU LGPL license) toolkit provides an interface between simulation codes such as OpenFOAM® and a variety of iterative systems analysis methods, including optimization, uncertainty quantification, etc.

In general, DAKOTA® is linked to OpenFOAM to perform parametric study, e.g. to find the optimal velocity or for geometry/shape optimization (Daymo et al. 2019) and geometry parametrization. In this study, the novelty comes when DAKOTA is coupled to OpenFOAM to make a parameter estimation of a model.

Estimation consists in finding the optimal values for the parameters that can minimize a cost function or Quantity of Interest. In the present case, the cost function, often described as the sum of the squared errors between the observed data and the simulated data, is minimized with respect to the parameters.

A non-linear squares method is used to minimize the objective function. In such method, the model equation is fitted to the experimental data rather than transforming it into a linear form (Sagnella 1985).

Coupling DAKOTA and OpenFOAM mainly consists in running an OpenFOAM solver in order to get a response or output, in which later, DAKOTA will read to analyze the response. If the response has not a desirable value then DAKOTA will set new conditions for a new simulation in OpenFOAM until a convergence criterion is reached. When coupling DAKOTA to another software, an external analysis driver (it can be a shell code) will link both. DAKOTA calls OpenFOAM through it and acts like a black box simulator to calculate the objective function and estimate the parameters.

The basic OpenFOAM case must contain: the initial and boundary conditions, mesh information, constant values, time controls and tools to extract the information of interest (simulated data). After setting the optimization method and the initial guess of the parameters, DAKOTA will call OpenFOAM to run a simulation with the initial guess of the parameters. Then, OpenFOAM will sample the points of interest in the mesh. With the additional shell codes, the simulated settling curve is extracted and the values are compared to the experimental data. The latter values are then used by DAKOTA to obtain the cost function and estimates the new parameter set if needed.

Here, parameter estimation was thus performed using the non-linear squares (NLS) method of DAKOTA. This is a deterministic calibration where parameters are adjusted in order that simulation data points are the closest to the experimental data points. The objective function is then the sum of squared errors (SSE) between experimental and simulated batch settling curves. The NL2Sol algorithm in DAKOTA, uses the Gauss-Newton gradient method to minimize the cost function.

The quality/accuracy of model calibration was evaluated by the Nash Sutcliffe efficiency (NSE). This evaluates the quality of the estimation by determining the relative magnitude of the residual variance compared to the measured data variance (Moriasi et al. 2015).

The dSVI was measured before and after the implementation of the hydrocyclone based densification process on the experimental line (start of the trial on 8th October 2019). Figure 3 shows the seasonal repartition of dSVI using box plots, respectively observed for the control line (conventional sludge all along test period, Figure 3(a)) and the experimental line (densification process starts after the dashed vertical line, Figure 3(b)). While the CAS control sludge suffers the usual sinusoidal dSVI variation pattern, with bulking periods recurrently occurring during low-temperature high dilution seasons (Knoop & Kunst 1998; Noutsopoulos et al. 2007; Jones & Schuler 2010), the test sludge displayed a steady flat behavior throughout the first winter. Then the start of aerobic granulation triggered a reduction of the dSVI value during the spring transition period, to finally reach the steady state stabilized nearly constant dSVI value of 41±6.8 mL·g⁻¹ over the last year of the experiment (as opposed to the conventional control sludge displaying larger variation of 101±26 mL·g⁻¹ over the same period). Over the last year of the experiment, the test sludge displayed a hybrid granule-floc biomass constituted of 39±8% of aerobic granules (AG) measured as MLSS by wet sieving at 200 μm. AGs were confirmed under microscope. The dSVI value were positively correlated to the mass percentage of biomass under granular form. Over the same period, the control sludge, displayed only 12± 3% of its biomass with size >200 μm, mainly constituted of debris and fibers with only few occurrences of AG confirmed under the microscope. Detailed results of granulated inventory dynamics and correlations can be seen in Roche et al. (2022).

The ZSV acquired from cylinder tests at a normalized MLSS concentration of 2 g·L⁻¹ (Figure 4) followed an inverse pattern compared to the dSVI. ZSV evolution on Figure 4 reflects the intrinsic sludge settling characteristics evolution. For the densified sludge, it maintained around 4 m·h⁻¹ over the first winter. Then, it gradually increased as granulation triggered over the next spring transition period to finally reach steady-state above 6 m·h⁻¹. Meanwhile, the control sludge suffered recurrent winter-spring bulking, with ZSV degradation down to 1.5 m·h⁻¹ followed by a period of improvement up to 4–5 m·h⁻¹ at the end of summer. Both dSVI and ZSV evolutions substantiate the year-round stability of the test sludge settling characteristics, which clearly contrasts with the highly fluctuating profile of the CAS.

dSVI and normalized ZSV are useful to get a quick overview of activated sludge settleability. However, to consider the improved settling properties of test sludge in the future design and operation of SSTs, further insight into the involved settling regimes is necessary. Therefore, the following sections present results obtained during batch tests performed in a larger settling column equipped with an ultrasonic profiler (sampling period identified in red on Figure 3).

The sludge blanket produces a strong ultrasonic backscattering echo. Thus, a precise recording of the sludge blanket height (SBH) with time is possible. The time evolution of the backscattered amplitude can be observed on Figure 5. The upper part shows the amplitude profile as a function of the distance to the transducer along the recording time. One can clearly see the sludge blanket evolution. The second part of Figure 5 shows the amplitude variation over time at a given distance, 0.2 m from the transducer. At the beginning of the recording a strong signal is observed due to the particles present in the homogeneous sludge suspension. The peak corresponds to the blanket: after its passing, the amplitude decreases drastically as the water is clear above the blanket.

The initial concentrations for test and control CAS sludges are 6.85 and 8.43 g·L⁻¹ respectively, with AG proportions (wet sieving mass percentage >200 μm) being 12 and 41% respectively (Roche et al. 2022). The fresh sludge samples exhibited dSVI, measured in graduated cylinders, of 31 and 126 mL·g TSS⁻¹ respectively. Using batch settling column data, dSVI were 55 and 163 mL·g TSS⁻¹ for test and control sludge respectively. This difference highlights the impact of measuring vessel as well as the sampling location on the results: for graduated cylinder and batch column, samples were taken from aeration tank and recirculation respectively. After processing the ultrasonic data, the batch settling curves are obtained (Figure 6 and Supplementary Information). They show the evolution of the SBH for each analyzed concentration. The linear part of the curves corresponds to the zone sedimentation (hindered settling velocity ⁠).

Control sludge batch settling curves were obtained for different initial concentrations (Figure 6(a)). For a first concentration of 8.43 g·L⁻¹ (data not shown), as well visually as acoustically, no sludge blanket was observable. This sludge was left to settle overnight, and no settling velocity variation was observed: one can conclude that the sludge is already in permanent contact and consequently in advanced stage of compression.

As expected, lower the sludge concentration, higher the value: hindered settling velocity for a concentration of 5.49 g·L⁻¹ was 17 times smaller than the one observed for a concentration of 1.78 g·L⁻¹. For the sludge with an initial concentration of 5.49 g·L⁻¹, the linear part of the curve is hardly detectable: this suggests that conventional sludge above this concentration is already in compression phase.

The theory of Kynch (1952) indicates that for a higher sludge concentration, the sedimentation velocity will decrease. This can still be confirmed for the test sludge. However, this sludge settles very fast for all assessed concentrations.

If one considers control and test sludges with similar concentrations (3.78 and 3.51 g/L respectively) at the same settling time (1000s), the test sludge blanket is much lower (12 cm) in comparison to the control sludge blanket (31 cm). (Figure 6b). This indicates that the densified process clarifier can provide sludge storage to a greater extent in case of high hydraulic loading events.

It turns out that the sedimentation velocity for the test sludge at 6.85 g·L⁻¹ concentration is approximately equal to the one of control sludge at a concentration of 1.78 g·L⁻¹.

This potentially allows an equivalent settling velocity at much higher MLSS concentration in the aeration tank.

The UB-Lab ultrasonic profiler, in complement to the profile of the backscattering amplitude, also records the velocity profile based on Doppler shift measurement.

Two transducers with different frequencies were used. Their sensitivity is correlated to the backscattering particles size: as higher the frequency as smaller the detected particles. As confirmed by the data quality, the different frequencies are not pertinent for a global velocity exploration over the whole sludge volume. The lowest frequency, 1.97 MHz will be sensitive to rough particles with radius close to 120 μm. Such kind of particles will rather appear under the sludge blanket: thus, this frequency will be significant below the blanket. On the contrary, the second frequency 7.5 MHz is optimal for the observation of particles with a radius close to 30 μm which will be present in the clear water above the sludge blanket.

Negative velocity values stand for particles moving away from the transducer which is the case when they are settling to the bottom of the tank. Figure 7 shows the recorded velocity profiles for comparable concentrations of control activated sludge and test sludge, 3.78 g·L⁻¹ and 3.51 g·L⁻¹ respectively. The velocity profiles at eight different settling times are shown, on the left for test sludge, on the right for control. Their values confirm the extrapolated settling velocity obtained through the sludge blanket height measurements.

In general, the test sludge velocity profiles show higher velocity values than CAS velocities. For both types of sludge, the velocities observed below the interface of the sludge and clear water are fluctuating less than the ones observed above. This is an expected behavior due to the permanent contact of the sludge particles within the blanket which blocks their free movement in different directions. On the contrary, the particles in the clear water settle more independently as suggested by the important velocity fluctuations recorded there.

At a frequency of 1.9757 MHz (below the interface) and after a minute of settling (Figure 7), the test sludge particles velocities along the column are higher than the velocities of the control sludge at the same time. Indeed, the test sludge particles can attain maximum velocities of 40 m·h⁻¹ in certain regions of the column against 23 m·h⁻¹ for the control. Hence, the CAS particles are for a long time in permanent contact, which causes low settling velocity values creating a very homogeneous settling velocity within the sludge blanket. Interestingly, the velocity profile at 2 min for test sludge indicates positive velocity values: the particles are transitionally coming back to the surface. This is likely due to significant density currents induced by the fast movements of particles.

The calibration of the sedimentation model parameters (Valle Medina & Laurent 2020) is presented in Table 1. In order to challenge the identifiability of settling parameters, two calibration methods were performed:

• Method #1: Vesilind parameters (Equation (2)) were first derived from the analysis of batch settling curves for all the tested TSS concentrations: determination of relationship (Figure 8). Then, parameters of the compression function were calibrated using DAKOTA toolkit combined with OpenFOAM solver.

• Method #2: both Vesilind and compression parameters were calibrated using DAKOTA toolkit combined with OpenFOAM solver.

For DAKOTA based calibration, only batch settling curves obtained at 3.78 g·L⁻¹ and 3.51 g·L⁻¹ respectively for control and test sludge had been used. These concentrations are close enough to enable the sound comparison of both type of sludges, and these also are typical MLSS concentrations practiced in aeration tanks. The calibration quality was evaluated using the NSE statistical criteria, for which values close to 1 indicate accurate correlation. The NSE values (Table 1) substantiate consistent correlations, except for Method #1 with the densified sludge, for which the value is below 0.8.

Figure 8 compares both sludges Vesilind (Equation (2)) parameters calibration for tests carried out in graduated cylinders (from classical dSVI trials following (EN 14702-1:2006 2006)) and in the 40 cm diameter settling column. For test sludge, both protocols yield similar parameters values and provided a good fit. However, for control sludge, graduated cylinder test results could not be predicted accurately by Vesilind function, especially for TSS above 2 g·L⁻¹. As settling velocity is much higher in the case of test sludge, wall effects in the 1 L graduated cylinder have a smaller effect in this case. Here, it must be stressed that all settling tests in this study were carried out without stirring, which can reduce these wall effects occurring in small vessels (Bye & Dold 1998).

From an operational perspective, this suggests that derivation of test sludge ZSV by the classical dSVI tests performed in graduated cylinders is reliable enough to be extrapolated in larger scale devices, such as full-scale clarifiers. Also, the rather constant dSVI obtained at steady state (last 4 months on Figure 3(b)) with test sludge suggests that parameter setting for the 1D settling model could be kept constant for simulating plant operation throughout the year.

Considering the parameters from batch settling column (Table 1), settling properties can be compared for both sludges. In general, the hindered sedimentation velocity of the test sludge is much higher than the one of control sludge at any concentration (Figure 8): if one takes similar concentrations as for example 5.3 g·L⁻¹ for test sludge and 5.49 g·L⁻¹ for control sludge, one can observe that the velocities are 3.43 and 0.16 m·h⁻¹ respectively. If the values of (theoretical maximum speed of sedimentation) are of the same order of magnitude, the exponential parameter is much lower for the test sludge: this confirms an increased hindered settling velocity for the test sludge, especially for the highest TSS concentrations.

Comparing calibration methods #1 and #2, it must be stressed that different parameters sets were obtained, especially for hindered settling parameters and (Table 1). For control sludge, despite different values, Vesilind's function predicts experimental ZSVs with similar quality as there is correlation between these two parameters in terms of identifiability (Figure 9) (Torfs 2015). However, for test sludge, Vesilind's parameters deriving from method #1 yields a less good prediction of SBH over time (Figure 9). In fact, it can be observed on Figure 8 that ZSV from Vesilind's function at the tested concentration of 3.51 g·L⁻¹ is higher than the experimental one. Conversely, using Vesilind's parameters from method #2 yields an underestimation of ZSV for the range of tested concentrations. As test sludge is settling very fast and entering the compression regime very quickly, it is more difficult to decorrelate hindered settling from compression parameters. Test sludge probably has a closer behavior to primary sludge as there is very little zone settling (Laurent et al. 2022).

Investigating the use of a power-law function for hindered settling and using method #1 would help solving this identifiability issue, as suggested by Torfs et al. (2017).

The critical sludge concentration (X_crit) corresponds to the limit beyond which the sludge enters in compression: sedimentation then corresponds to the flocs compaction under the effect of their own weight with occurrence of a solid stress (quantified by the parameter λ).

From results presented in Table 1, it appears that control sludge is entering compression regime at an X_crit concentration of about 4 g·L⁻¹. This is consistent with the observation of batch settling curves in Figure 6: the slope for conventional sludge with X₀=5.49 g·L⁻¹ is virtually inexistant as the control sludge is already in compression regime. Furthermore, this critical concentration is of the same order of magnitude than the MLSS concentrations typically encountered in CAS aeration tanks.

In the case of test sludge, the slowing down of sedimentation linked to this phenomenon will appear for a much higher X_crit >7 g·L⁻¹.

In the model, the parameter allows calculation of the solid stress that sludge flocs exert when they get in permanent contact. Higher the value, higher the solid stress (Equation (4)). For the test sludge, value is lower than the one of control sludge. This might be explained by the more regular and compact morphology of the test sludge which creates a sludge bed structure which is less resistant to the sludge weight compared to CAS flocs. Some pictures of densified sludge morphology from this study can be seen in Roche et al. (2022).

To summarize, test sludge enters compression regime for a higher concentration and will exert a lower resistance to compaction. This is the effect of sludge structural properties (more compact flocs, granulation process) allowing a greater and faster compaction of the sludge (Roche et al. 2022), which therefore can thicken to a greater extent than CAS sludge for a given retention time in the clarifier.

Figure 10 presents simulated (with method #2 parameters) and experimental settling velocity profiles, for the two types of sludge at different times. After 5 minutes of sedimentation, one can observe that the velocities are still higher for the test sludge, despite its higher concentration in the sludge blanket. Control sludge still undergoes hindered settling, as confirmed by the simulated constant velocities for height above 0.25 m. At the same location, experimental velocity displays some fluctuations as already observed in past studies: even in the hindered settling regime, initial particle size distribution and flocculation state still impact settling behavior (Locatelli et al. 2015; Torfs et al. 2015). At the same time, the test sludge is already fully in the compression regime, as shown by the gradually decreasing settling velocity from the top of sludge blanket. After 10 minutes, both sludges are in compression regime. It is noteworthy that the model, which has been calibrated on SBH data only, is able to predict settling velocities with a good accuracy.

With

• SS_BS maximal allowable sludge concentration in RAS in (g·L⁻¹)

• dSVI diluted sludge volume index (mL·g⁻¹)

• t_Th sludge thickening time expressed in (h)

From the SBH measurements and initial concentrations (Figure 6), the average suspended solids concentration within the sludge blanket was thus compared in Figure 11 to the ATV achievable concentration from Equation (5) as well as the bottom sludge concentration computed from the 1D simulation of the batch column. One has to remember that ATV function was developed for estimating sludge concentration in the context of a continuous clarifier, not a batch column.

Nevertheless, regardless of initial concentration, the test sludge attains an average of 10g·L⁻¹ in less than 5 minutes whereas the control sludge requires more than 2 hours to reach such level of concentration. The faster thickening of test sludge also means much higher achievable concentrations, in the range 25–30g·L⁻¹, for a reasonable sludge retention time of 1–2 hours in the clarifier. The comparison with the ATV function shows an increased safety margin for the test sludge, particularly for longer thickening times. Logically, the model also predicts that bottom sludge concentration is much higher in the case of test activated sludge, even at early times of simulation.

In this study, settling properties of conventional (control) and ‘densified’ (test) sludge from a full-scale pilot facility were characterized with two levels of details at two timescales: the dSVI and ZSV were followed for more than 2 years and detailed batch-settling tests were performed in winter 2020–2021 for deriving more precise settling parameters.

The hydrocyclone selective wasting enabled avoiding the bulking phenomenon and resulted in an enhanced and much more stable settling behavior of the test ‘densified’ sludge throughout the year (dSVI below 50 mL·g⁻¹ TSS).

The velocity measurements performed with the ultrasonic transducers proved that test sludge settles faster than control sludge particles. This non-invasive method allows to measure the velocities of the particles at different heights within the column. Such information is useful to validate settling model parameters.

The calibration of 1D settling model parameters was achieved for both types of sludges. The identifiability of hindered settling function is more challenging in the case of test sludge due to the very fast ZSV and the higher importance of compression. Parameters values clearly demonstrate the improved settling properties of test sludge: the hindered settling velocities are always much higher for it. Furthermore, the test sludge enters the compression regime with concentrations almost twice as high as for control sludge. Based on the settleability parameters investigated, the combination of hydrocyclones and biological selection could have led to granulation/densification. Considering the stable dSVI observed over a nearly year-round period, the parameters derived from the batch experiment could be used as fixed parameter to model densified sludge settling behaviour regardless of seasonality.

The improved settling properties of ‘densified’ test sludge, especially its enhanced thickening properties (bottom concentration predicted in the batch column up to 25 g·L⁻¹), opens perspectives for extending the range of WRRF design and easier operation:

• RAS rate optimizations: operating with a reduced recirculation rate without necessarily raising the SBH. The RAS rate was optimized on the test line from 157% to 50 and 30% respectively for peak dry weather flow and storm conditions, providing an approximate saving of 1,000 MWh/year on recirculation electrical expenditure when extrapolated to the whole plant (e.g. if the 4 lines were equipped). The limited RAS flow also reduces the SLR that the clarifier has to deal with, which helps to secure operation. The lower RAS flow also results in an extended hydraulic retention time in biological tanks and a less diluted influent in the first compartment, which provides a higher substrate gradient along the plant profile, which is favorable to sustain granulation.

• Secondary clarifier resilience under adverse peak flow conditions thanks to an extended sludge storage possibility.

• Flexibility in operation, with the possibility to temporarily store in the aeration tank without risk for the clarification, notably in case of events where the sludge concentration in the aeration tank can drift up (extraction failure, problem on the sludge evacuation).

• The possibility to increase moderately the biomass inventory in an existing plant without jeopardizing the clarification stage, as the sludge would still settle quicker than SOR even at higher concentration. This could help to accommodate moderate increase of load capacity without having to build new volumes, provided that subsequent supplemental oxygen demand can be met with the aeration system; or help to maintain nitrification under adverse cold-water conditions.

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

The authors declare there is no conflict.