At wastewater treatment plants (WWTPs), the aerobic conversion processes in the bioreactor are driven by the presence of dissolved oxygen (DO). Within these conversion processes, the oxygen transfer is a rate limiting step as well as being the largest energy consumer. Despite this high importance, WWTP models often lack detail on the aeration part. An extensive measurement campaign with off-gas tests was performed at the WWTP of Eindhoven to provide more information on the performance and behaviour of the aeration system. A high spatial and temporal variability in the oxygen transfer efficiency was observed. Applying this gathered system knowledge in the aeration model resulted in an improved prediction of the DO concentrations. Moreover, an important consequence of this was that ammonium predictions could be improved by resetting the ammonium half-saturation index for autotrophs to its default value. This again proves the importance of balancing sub-models with respect to the need for model calibration as well as model predictive power.
At wastewater treatment plants (WWTPs), the conversion processes in the bioreactor are driven by the presence of an electron acceptor, i.e. dissolved oxygen (DO) in the case of aerobic processes (e.g. aerobic heterotrophic growth and nitrification). Within these conversion processes, the gas–liquid oxygen transfer has been reported as a rate limiting step (Ardern & Lockett 1914; Garcia-Ochoa & Gomez 2009). In addition, aeration is the largest energy consumer at WWTPs (Reardon 1995; Tchobanoglous et al. 2004; Devisscher et al. 2006; Zahreddine et al. 2010). Hence, a sound description of this process is vital in the cost-effective optimization of the aerobic biokinetic processes.
From a modelling perspective, a good description of the gas–liquid and liquid–gas transfer is crucial. Benedetti et al. (2008) performed a global sensitivity analysis on biochemical, design and operational parameters with the benchmark simulation model no. 2 (Jeppsson et al. 2007; Nopens et al. 2010) illustrating that the oxygen transfer coefficient is the second most sensitive parameter in the model. In a modelling study comparing different model choices on a full scale WWTP, Maere et al. (2008) and Cierkens et al. (2012) show that a good estimation of the aeration flow rate (and hence of the oxygen input), leads to excellent prediction of oxygen consumption and concentrations in the activated sludge process. Further modelling work on improving the aeration model has led to even better prediction of the oxygen consumption and concentrations, consequently, improving the prediction of ammonium (NH4) removal, which omits the need for unnecessary calibration of the biokinetic model (personal communication, Cierkens).
In order to estimate the oxygen transfer, correlations exist that link the kLa to the airflow rate (Gillot & Héduit 2000) or the gas flow velocity (Dold & Fairlamb 2001; Gillot et al. 2005; Rosso et al. 2005; Gillot & Héduit 2008).
Several attempts have been made to predict the α-factor, assigning its observed variation to either surfactant concentrations (Rosso & Stenstrom 2006), mixed liquor suspended solids (MLSS) concentration (Germain et al. 2007; Racault et al. 2011), sludge age (Rosso et al. 2005; Gillot & Héduit 2008; Henkel et al. 2011) and viscosity (Fabiyi & Novak 2008). Most of the observed correlations have in common that the studied factors affect the shape of the bubbles and the gas flow at the interface. Also, the physical properties of the liquid, together with its flowing regime, can influence the kLa as coalescence or breakage can vary the bubble size distribution and, thus, the available area for gas transfer, i.e. the a part of the oxygen transfer coefficient (Fayolle et al. 2010). Viscosity, in particular, was observed to affect the shape of a bubble plume and, thus, increasing the probability that a bubble will collide with a neighbouring one (Ratkovich et al. 2013).
The β-factor is the correction for the process water oxygen saturation concentration. This factor is governed by the same corrections (e.g. temperature and salt concentration) as the Henry coefficient, by which the saturation concentration is determined in combination with the mole fraction of oxygen in the gas phase.
The θ-factor, also known as geometric temperature correction coefficient, is used to relate mass transfer coefficients to a standard temperature (Equation (4)). Because of lack of a proper appraisement, a value of 1.024 should be used unless differently specified and strongly supported by consultants and manufacturers. Moreover, it is recommended to limit the temperature correction for deviations smaller than 10 °C, although it is well known that corrections over a wider range are often needed (Stenstrom & Gilbert 1981).
To date, WWTP models consist of highly detailed biokinetic models (Barker & Dold 1997; Henze et al. 2000) but often lack detail of other critical processes, such as aeration. This paper is addressing (1) the need for higher detail in modelling of the aeration process, hereby using off-gas measurements for more detailed insight in the process and (2) the impact on the calibration of the model. Based on an extensive measurement campaign, the applied aeration model, which is a combination of several models available in literature, is calibrated. The effect of this improved calibration on the overall model calibration is shown.
MATERIALS AND METHODS
Aeration system at the WWTP of Eindhoven
Within the 7 m deep biological tanks, aeration is provided in the outer ring by plate aerators divided in two sections: a continuously active summer package and a winter package (used as additional capacity under extreme conditions and when T is low leading to a reduced nitrification rate). The summer package comprises 504 plate aerators, totalling a surface of 1,063 m2, whereas the winter package comprises 84 plate aerators totalling a surface of 177 m2.
The airflow of the summer package is continuously controlled by an NH4-DO cascade feedback controller. On the other hand, the winter package is only switched on in case the capacity of the summer package proves not to be sufficient (mainly during storm events) and is running on full capacity when active.
Modelling of aeration
Because the aeration model is not able to capture all the variability observed in the measurement campaign, the model is also evaluated using the measured αSOTE as input to the whole plant model.
The floating hood was equipped with an LDO probe (Hach-Lange) and DO data were acquired in order to correct for variable DO gradients during the oxygen transfer process and relate the efficiency results to standard conditions with Equation (4). Data of DO and oxygen content in the off-gas were acquired with a data acquisition card (DAQ-card USB-6341, National Instruments, USA) using a graphical user interface developed in LabView (National Instruments, USA). Adjustments for CO2 content in the off-gas were performed in a post-processing step when both the data from the off-gas analyser and the X-Stream were available.
The manner of calculating the saturation concentration may influence the results of αSOTE. In particular, the pressure correction and the applied effective saturation depth vary depending on the source, either 50% (Tchobanoglous et al. 2004), 22–44% (US EPA 1989) or 33% (Gillot et al. 2005) of the total submergence depth. The application of 33% of the total submergence depth would result in slightly higher αSOTE values and, as such, lead to the same conclusion as the applied 44%.
On a side note, but worthy of mention, is the difference in oxygen concentration between the SCADA system and the hood measurements. The difference amounts up to 0.5 mg/L and is due to both the distance between the plant sensor and the hood and to the heterogeneity caused by mixing patterns. It emphasizes the importance of (1) having a good location for the sensor measurements and (2) an appropriate hydraulic mixing model (Rehman et al. 2015).
As expected, the NH4 was even further depleted (dashed line) as compared to the previous simulations (not shown) and, therefore, more distant from the measured NH4 (dots) (Figure 6, right). These observations may lead to consider a re-evaluation of the model parameters linked to NH4 depletion, in particular the ammonium half-saturation index for autotrophs (KNH,ANO), which was previously fixed (0.05 mg/L) to match the SCADA measurements. Resetting the half-saturation index to its default value (1.0 mg/L) improves the model predictions for ammonium (full line) considerably (Figure 6, right). This shows again the importance of sufficient model complexity of sub-models which otherwise forces the modeller to calibrate biokinetic parameters.
The aeration model applied is certainly an improvement over state of the art aeration models applied in WWTP models and provided good simulation results for the case of Eindhoven (Cierkens et al. 2012). However, the new measurement campaign shed light on the real evolution of the α-factor in the aerated tanks and it was demonstrated that applying this more realistic α-factor improves the simulation results.
However, the model is not able to describe all the observed variations in αSOTE (Figure 5). Some of the variations in the data can be well described, whereas other days, the variations cannot be explained with the model at all. Important to note is that, for the good fit periods, the parameter values had to be changed each day. This demonstrates that the model could still be improved as, on the one hand, it obviously misses some key elements and, on the other hand, it shows some features that are doubtful. First, the fact that α-factor is linked to the sludge age (SRT) is questionable as the SRT is a global WWTP characteristic and aeration (oxygen transfer and the α-factor) is a local parameter influenced by the local physics of the tank. Actually, the US EPA (1989) mentions the change in α-factor over the length of the aeration tank.
The apparent correlation between SRT and the α-factor can be explained by the link of SRT with some of the influential factors. First, SRT is directly linked to the MLSS concentrations, which is considered as having an impact on the oxygen transfer, and also to the sludge composition and morphology. The combination of sludge concentration, composition and morphology can be related to changes in viscosity which are known to influence the gas–liquid transfer (Fabiyi & Novak 2008; Ratkovich et al. 2013). In addition, the floc volume is one of the drivers for changing α-factors mentioned by Henkel et al. (2011). The SRT is also linked to the capacity of biodegradation of, e.g. surface active agents, although that should only be really differing with very low SRTs. The surface active agents, however, are known to impact oxygen transfer (Rosso & Stenstrom 2006). Finally, the need for recalibration, in this study, suggests that the SRT is probably not the only predictor for the variations in the α-factor.
Many of the factors (surfactants, viscosity, floc volume) mentioned above, have an impact on the formation and shape of the air bubbles. With regards to this, another suggested future improvement in the modelling of the aeration process would be the evaluation of the drivers and the impact of bubble coalescence. In this evaluation, bubble column tests in combination with modelling frameworks such as population balance models (Nopens et al. 2015) will have a definite added value.
These observations prove the need for a recalibration of the aeration model. Moreover, after applying this more realistic oxygen input, ammonium predictions could be improved by resetting the ammonium half-saturation index for autotrophs to its default value. In a previous calibration exercise, this ammonium half-saturation index was lowered in order to enable the model to predict the lower ammonium concentrations measured. Apparently, the aeration model, calibrated on a single off-gas test, was not able to predict a sufficiently high oxygen input, which was then, incorrectly, corrected for by increasing the oxygen consumption for nitrification through the ammonium half-saturation index.
The application of the aeration model is clearly an advancement over the state of the art description of the aeration process in WWTP models. During an extensive measurement campaign with off-gas tests performed at the WWTP of Eindhoven a high variability was observed in the OTE. Applying this gathered system knowledge in the aeration model and after re-adaptation of the ammonium half-saturation index for autotrophs, the model proved, based on the airflow rates, to give good predictions for the oxygen and, consequently, ammonium concentrations at the WWTP of Eindhoven. As such, it was shown that the effect of applying the correct α-factors omits the need for calibration of the biokinetic model. This again proves the importance of balancing the complexity or detail in the different sub-models which is preferred over parameter calibration with respect to model predictive power.