Realization of a Digital Twin of an Operating 2.200 m3/h O3/H2O2 and UV/H2O2 Advanced Surface Water Treatment Plant

SUMMARY A digital twin of a 2.200 m3/h advanced surface water treatment installation including O3/H2O2 and UV/H2O2 AOP has been developed. This mechanistic virtual model mirrors the actual plant and allows running scenarios for the most economical way of micro pollutant reduction while minimizing bromate formation. The model was set up based on years of on-site research and piloting from the water supplier Dunea and water treatment solution provider Xylem for that specific water source. The advanced modelling company AM-Team then created and calibrated a Sumo-based mechanistic model on the pilot results with further validation based on the installed 2.200 m3/h full-scale plant. The digital twin allows Dunea to quickly and safely simulate the operation of the physical plant with changes in water quality and plant control. INTRODUCTION Dunea supplies about 1,3 million people in The Netherlands' capital The Hague and the surrounding regions with ca. 9.000 m3/h of drinking water. For 2.200 m3/h of the total supply, Dunea decided to go even beyond the ambitious goal of realizing a full advanced O3/H2O2 and UV/H2O2 treatment in bromide-rich surface water with up to 80% reduction of micro pollutants. After the installation of the full-scale advanced treatment plant in 2018 including a 3.5 kg/h ozone generator and two duty LPHO UV-reactors (285 kW in total) (both technologies supplied by Wedeco, a Xylem brand), AM-Team was contracted to realize a digital model of the installation, including the complex chemical reactions of O3/H2O2 and downstream UV/H2O2 Advanced Oxidation Processes. METHODS General Both the O3/H2O2 and the UV/H2O2 model are composed of core models and model extensions. The core model parts focus on consumption and production rates of oxidative species and predict OH production and consumption rates, O3 transfer and consumption rate (for O3 process only), H2O2 consumption rates and UV254 decrease. The model extensions relate to bromate formation (only when O3 is involved) and Trace Organic Contaminant (TrOC) removal. The basic model AMOZONE The AMOZONE model is a detailed mechanistic kinetic model for the prediction of TrOCs removal and BrO3 formation including many chemical reactions derived from years of experience and sound literature. Some inputs are needed to feed the model which solves the partial differential equations. Figure 1: General AMOZONE model structure Calibration and validation O3/H2O2 The AMOZONE model calibration and validation for the O3/H2O2 process occurred on the pilot installation reproducing the experiments in steady state conditions. Validation was then extended on the full-scale plant data to reproduce specific experimental tests. Given a known O3 dose for the specific experiment, the first step consisted in the calibration of the O3 consumption by the organic matter present in the water matrix. The second step consisted in the calibration of the net HO

* production (net of the scavenging). The last step was to verify the production of bromate. The resulting calibrated model was validated against additional data coming from the pilot experiments and external O3/H2O2 projects. UV/H2O2 Maintaining the O3/H2O2 parameters, the UV/H2O2 model was then calibrated on the experimental data coming from the pilot installation and on the full-scale installation. In this phase the only calibrated parameter was the constant to convert the UV dose used onsite (J/m2) into the UV light intensity needed in the model (Ein/m3/d). The model results were validated with the data from the operation of the full-scale plant. CONCLUSIONS The AMOZONE model was calibrated and extensively validated for both the pilot and the full-scale installations using a large set of experimental data provided by Dunea. Already at the preliminary calibration stage was observed a good performance of the model as compared to the experimental results, which resulted in minor changes of the parameters values at the calibration. The O3/H2O2 reactions were improved in parallel projects and included in the present work. These model improvements allowed to exactly reproduce the reduced overall exposure to O3, at increased H2O2 dose and its effect on all TrOCs and bromate. The UV/H2O2 reactions were implemented and a dedicated UV process unit was developed in Sumo. The modelled removal aligned with the experimental data catching the differences among the TrOCs in terms of relative removal. Both the O3 AOP and the serial AOPs performances were correctly reproduced confirming the solidity of the mechanistic model behind. A final validation of the full-scale plant model was performed using a time series of almost 4 years. During this time, different UV, H2O2, and O3 dosage strategies were tested measuring bromate regularly and also often TrOCs removal. The dynamic influent characteristics from the online sensors and lab analysis, and the operational settings (i.e. UV, H2O2, and O3 dosages) were given as input to the model. This allowed to perform an accurate dynamic prediction of bromate measurements during the whole period, including TrOCs relative removal. With the calibrated and extensively validated full-scale model, it was possible to run dynamic What-if scenarios to test the effect of different dosage strategies during the whole period.