A Virtual Full-scale Ozonation Plant for Micropollutant Removal: How to Reduce (Eliminate?) Piloting Efforts

ABSTRACT
The AMOZONE kinetic ozonation model was applied for assessing the suitability of ozonation for the removal of micropollutants (MPs) from the effluent of three full-scale Water Resource Recovery Facilities (WRRFs). The model was calibrated on the real water matrix. Different control strategies for MPs removal and bromate (BrO3) formation were tested. 11 MPs were considered from the list of guide MPs from the current Dutch rules. Results unraveled the complex dynamics behind O3 and HO* reactions. The virtual full-scale experiments revealed the potential MPs removal, the extent of production of BrO3 for the three virtual installations, and the influence of the upstream WRRF dynamics. OpEx and CapEx of the virtual full-scale installations were also assessed (the total cost varied between 9.0 and 10.7 cEUR/m³). No onsite piloting or MPs analyses were needed and innovative controls could be tested.
INTRODUCTION | BACKGROUND
Properly sizing a water treatment installation is of crucial importance, especially when the MPs removal must be completed with expensive technologies such as ozonation. Mechanistic models are nowadays supporting important decisions in design, optimization, revamping, and scale-up, and can be a powerful tool to save time and resources from long and expensive pilot testing. In addition to this, mechanistic models provide in-depth understanding of the complex net of reactions occurring in every point of an installation, impossible to study in real-life testing. In this work the potential MPs removal, the BrO3 formation risk, and the costs of three ozone installations for the WRRFs effluent were assessed and key questions were answered. This was possible by building and operating three virtual full-scale plants with the novel AMOZONE model.
METHODOLOGY
The AMOZONE model [1] mechanistically describes the complex net of reactions leading to the formation of BrO3 (inter alia: [2], [3]), hydroxyl radical (HO*) formation and scavenging, O3 reaction with the organic and inorganic water matrix, and a potentially limitless list of MPs which can be integrated.
Model calibration and validation
Several 24h secondary effluent composite samples of the Soerendonk, Eindhoven and Hapert WRRFs (operated by Waterboard De Dommel (WSDD), Netherlands) were provided, along with one month of high-resolution online data, and weekly lab analysis of the three effluents quality. Dedicated ozonation batch tests were performed in replicates on the samples at Ghent University lab in Kortrijk (Belgium) and IUTA lab (Germany) to assess reaction kinetics at two O3 dosages. During the batch testing, O3 and UVA254 decay, BrO3 production, and HO* concentrations (probe component) were measured as function of time. The results bench experiments were used to calibrate and validate the AMOZONE model.
Full/scale plant operational scenario analysis The model layout of the virtual O3 installation was conceived to ensure the necessary retention time for the O3 reactions to occur and for O3 to deplete completely (i.e. no final residual in effluent). Simulations of different control strategies for O3 dosage were run: i) O3/DOC proportional; ii) delta-UVA proportional; iii) MPs removal-based). Results were compared in terms of O3 demand, BrO3 and costs. The MPs removal-based control is a very novel control strategy as it implies the direct control of the O3 dose on MPs at a specific removal (not feasible without the model as MPs cannot be measured in real-time). The simulations also considered the contribution by the upstream biological treatment on MPs removal. Finally, an overview of CapEx and OpEx was provided in terms of volume of treated water (cEUR/m3) and yearly costs (kEUR/y). All the scenarios were run for 1 month of real plant operation. The online and offline data collected by WSDD were processed and used as model input to observe the effect of real dynamics on the different control scenarios.
RESULTS AND DISCUSSION
Model calibration and validation An extensive model calibration and validation was performed on the results of the batch experiments. The results of the model calibration for the three effluents is presented (Figure 1, Figure 2 and Figure 3). A cross-project validation of the model prediction capabilities was made comparing the results from this study with the results of the European CWPharma project [4]. In Figure 4, can be observed how the measured and simulated data of the first Swedish full-scale effluent ozonation plant (the Linköping WRRF, operated by Tekniska verken in Linköping) compare with the results of the Dutch plant of this study.
Dynamic simulations Knowing the detailed dynamics of the upstream WRRF helped in assessing its potential effect on the performance of the ozonation plant. In particular, the fluctuations in NH4 concentration, resulting from the combined effect of the influent dynamics and the WRRF airflow controller action, have a relevant effect on BrO3 formation (Figure 5). On the other hand, rain events and fluctuations in the water matrix (e.g. DOC, UVA) can also affect the removal of MPs (Figure 6). Scenarios analysis Figure 7 shows the peak demand of O3 for each of the scenarios. For this study, the peak demands resulted from the control strategy based on the highest O3/DOC. For all the scenarios, the BrO3 formation was evaluated and compared among the three different WRRFs (Figure 8). The BrO3 production is rather similar in the three plants due to similarities in the secondary effluents characteristics, even though the bromide concentration in Soerendonk (up to 400 µg/L) is higher than the other two plants (< 100 µg/L). The higher BrO3 is observed with O3/DOC = 1.0 gO3/gDOC, while in the other scenarios, the formation can be minimized due to the lower O3 dose applied. The peak in BrO3 in the delta-UVA controller in the Soerendonk case is mainly due to controller instability leading to moments with an applied dose higher than necessary. Finally, a cost calculation and a quantitative comparison was performed in terms of Total Costs (OpEx + CapEx) based on the model outcomes (Figure 9). The differences highlighted depend on the control strategy used (operational) or the plant size (capital). In fact, small-scale plants such as Soerendonk and Hapert have proportionally higher CapEx as compared to Eindhoven.
CONCLUSIONS
For the first time, an ozone installation was modelled in three WRRFs prior to building any pilot or full-scale installation, and a large amount of information was extracted concerning the water matrix reactions with O3 and HO*, the formation of BrO3, the removal of MPs, and cost analysis. The virtual O3 installations showed that the BrO3 formation risk in all the three plants is rather low. The calculated OpEx varied between 1.8 and 3.4 cEUR/m³ treated. The total cost varied between 9.0 and 10.7 cEUR/m³. A smart control based on the removal of MPs can potentially save around 10% OpEx compared to scenarios that are conventionally used.