Densified Sludge Model Sensitivity Analysis and Pilot Data Calibration

Background
The densified sludge process is a popular method by which to intensify secondary wastewater treatment (Daigger et al., 2023; Winkler et al., 2018). While batch processes are more widely utilized to date, continuous flow processes using hydraulic cyclones have generated significant interest (Bauhs et al., 2024). Unfortunately, there is limited understanding of good modeling practices for hydrocyclone-based densified systems. Process models take different approaches to model densified sludge behavior and selective retention by cyclones. One method is to define the fractions of state variables retained by cyclones, but these fractions are poorly understood and difficult to measure (Uri et al., 2017). Alternatively, a zero-dimensional densified sludge (DS) model has been proposed to streamline cyclone modeling by differentiating granule biomass from floc biomass, allowing for selective retention of granule mass fraction by the cyclone (Baeton et al., 2018). Granular biomass half-saturation coefficients are multiplied by a diffusion resistance (DR) term, simulating restriction of access to substrate within granules and subsequent reduction in granule organism growth kinetics (Baeton et al., 2018). However, there remains limited documentation regarding parameter sensitivity and calibration of this approach.

To help provide an improved practical understanding of the application of both the state variable and DS model approaches to hydrocyclone simulation, this paper undertakes (1) a sensitivity analysis of key input parameters for both approaches and (2) a calibration of both approaches to long-term pilot data.

Methods
Sensitivity analyses were conducted using the A2O plant + Cyclone simulation available in the SUMO library following the installation of the DS model addon (Dynamita, SUMO22) (Figure 1). A key benefit of densification is the elevation of nitrification capacity without capital expansion (Bauhs et al., 2024). For this reason, key performance indicators related to nitrogen removal were used as dependent variables during the sensitivity analysis. These include densified fraction of total biomass (FGranule), fraction of nitrifying biomass, and effluent quality.

Several parameters were evaluated in the DS model sensitivity analysis (Table 1). The variation range for each parameter was typically calculated using a change of +/-25% & 50% from the default value, though alterations were made as appropriate (Baeten et al., 2018; Cosenza et al., 2013). Further details regarding model setup and simulation will be included in the final paper, as will the parameters used for sensitivity analysis of the state-variable cyclone approach and their respective ranges.

Results
DR and mass split to cyclone underflow emerged as particularly impactful calibration parameters. The mass split to underflow is an operational parameter which sets the fraction of flocculant suspended solids retained by the cyclone. This abstract focuses on the impact of these parameters, though the final paper will include a deeper analysis.

Diffusion Resistance. FGranule is inversely correlated with DR, whilst maintaining the same fraction of granule mass retention by the cyclone (Figure 2a). As DR for all substrates increases, FGranule decreases due to the simulated inhibition of granule biomass. High DR conditions resulted in marginally reduced nitrification but an overall lower effluent total nitrogen (Figure 2b), potentially due to increased anoxic biomass within granules. At a DR of 1.0, ordinary heterotrophic organisms (OHO) were the most abundant granular organism group, as one would expect when the 'diffusion handicap' is removed (Figure 2c). Phosphorus accumulating organisms (PAO) preferentially granulate and were insensitive to DR, though granule ammonia oxidizing bacteria (AOB) and nitrite oxidizing bacteria (NOB) were nearly eliminated under high DR conditions (Figure 2d). It is crucial to note that individual substrate DR can be calibrated in the model using field data or literature values (Baeten et al., 2018); the sensitivity of individual substrate DR will be included in the final paper.

Mass Split. FGranule is positively correlated with the mass split to underflow since more solids, and thus more granules, are retained within the system (Figure 3a). As the mass fraction retained by the cyclone increased, MLSS and SRT also increased, with total SRT increasing from 14 days (-50%) to 39 days (+50%). Nitrification was maximized at high granule retention (Figure 3b) however, wasting would need to be appropriately adjusted to yield an operable MLSS. Interestingly, AOB and NOB composition in the granules remained relatively unchanged. Granular biomass was maximized with increased mass retention, for OHO and PAO in particular. While PAO growth was insensitive to DR, their growth is directly correlated to mass split to underflow (Figure 3c and d).

Next Steps. Results from the sensitivity analysis of the state variable model will be compared to the results of the DS model analysis. Finally, a DS model will be calibrated to hydrocyclone field data, and calibrated parameters will be compared to hydrocyclone vendor-supplied recommended values.

The learnings presented in this paper will provide practitioners insight into the use and sensitivity of two hydrocyclone process models, as well as practical good modeling practices for calibration.