Improved Sulfide Mitigation in Sewers Through On-Line Control Of

Water Research 135 (2018) 302e310

Contents lists available at ScienceDirect

Water Research

journal homepage: www.elsevier.com/locate/watres

Improved sulﬁde mitigation in sewers through on-line control of ferrous salt dosing

Ramon Ganigue a, b, Guangming Jiang a, Yiqi Liu a, c, Keshab Sharma a, Yue-Cong Wang d, * Jose Gonzalez d, Tung Nguyen d, Zhiguo Yuan a, a Advanced Water Management Centre, Building 60, Research Road, The University of Queensland, St. Lucia, Brisbane QLD 4072, Australia b Center for Microbial Ecology and Technology (CMET), Ghent University, Coupure Links 653, 9000 Ghent, Belgium c School of Automation Science & Engineering, South China University of Technology, Guangzhou, 510640, PR China d Sydney Water Corporation, 1 Smith St, Parramatta, NSW 2150, Australia article info abstract

Article history: Water utilities worldwide spend annually billions of dollars to control sulfide-induced corrosion in Received 9 October 2017 sewers. Iron salts chemically oxidize and/or precipitate dissolved sulfide in sewage and are especially Received in revised form used in medium- and large-size sewers. Iron salt dosing rates are defined ad hoc, ignoring variation in 16 January 2018 sewage flows and sulfide levels. This often results in iron overdosing or poor sulfide control. Online Accepted 8 February 2018 dosing control can adjust the chemical dosing rates to current (and future) state of the sewer system, Available online 13 February 2018 allowing high-precision, stable and cost-effective sulfide control. In this paper, we report a novel and robust online control strategy for the dosing of ferrous salt in sewers. The control considers the fluc- Keywords: fl fi fi Chemical dosing tuation of sewage ow, pH, sul de levels and also the perturbation from rainfall. Sul de production in e fl Corrosion the pipe is predicted using auto regressive models (AR) based on current ow measurements, which in Online control turn can be used to determine the dose of ferrous salt required for cost-effective sulfide control. Sewer systems Following comprehensive model-based assesment, the control was successfully validated and its effec- Sulfide control tiveness demonstrated in a 3-week field trial. The online control algorithm controlled sulfide below the target level (0.5 mg S/L) while reducing chemical dosing up to 30%. © 2018 Elsevier Ltd. All rights reserved.

1. Introduction preferential use of iron salt dosing in medium to large sewer systems, as its working principle makes it the best suited for sulfide Water utilities worldwide spend annually billions of dollars to control in such pipes (Ganigue et al., 2011). mitigate pipe corrosion and malodour nuisances caused by Various metal salts (e.g. ferrous chloride, ferric chloride and, in biogenic sulfide in sewers (Jiang et al., 2015; Pikaar et al., 2014; US some cases, ferrous sulfate) can react chemically with dissolved EPA, 1992). The dosing of chemicals (e.g. oxygen, nitrate, iron salt, sulfide to form relatively insoluble metal sulfides, thus decreasing magnesium hydroxide, among others) is commonly applied to the concentration of dissolved sulfide in sewage and subsequently 3þ prevent sulfide formation, remove it from solution or mitigate its H2S in sewer gas (WERF, 2007). Ferric ions (Fe ) oxidize sulfide to þ þ negative effects (WERF, 2007). Among these chemicals, the use of elemental sulfur while being reduced into ferrous ions (Fe2 ). Fe3 iron salt is widespread due to their cost-effectiveness (Barjenbruch, directly precipitate also with OH and phosphate, but will even- 2003; Bertran de Lis et al., 2007; Jameel, 1989; Padival et al., 1995). tually be available for sulfide removal (Zhang et al., 2009). Subse- þ A recent industry survey conducted in Australia highlighted the quently, Fe2 can combine with sulfide to form highly insoluble metallic precipitates (Dohnalek and Fitzpatrick, 1983). FeS is the most common iron sulfur precipitate in aqueous solutions, although other Fe-S precipitates such as Fe2S2 and Fe3S4 can also * Corresponding author. form (Rickard and Luther, 2007). Equations (1)e(4) illustrate the E-mail addresses: [email protected] (R. Ganigue), [email protected]. hydrogen sulfide equilibria, and the stoichiometries of sulfide au (G. Jiang), [email protected] (Y. Liu), [email protected] (K. Sharma), [email protected] (Y.-C. Wang), jose.gonzalez@ oxidation by ferric ions and FeS formation by ferrous ions. sydneywater.com.au (J. Gonzalez), [email protected] (T. Nguyen), [email protected] (Z. Yuan). https://doi.org/10.1016/j.watres.2018.02.022 0043-1354/© 2018 Elsevier Ltd. All rights reserved. R. Ganigue et al. / Water Research 135 (2018) 302e310 303

dynamics and sulfide production are yet to be achieved. In the 4 þ þ ¼ : H2S HS H pKa 7 05 (1) present work, these issues are addressed by the development of an online algorithm that is able to predict, in real time, sewage flows HS4S2 þ Hþ pKa ¼ 12:89 (2) several hours ahead. This enables the control system to adjust the iron salt dosing rates based on the predicted HRT and sulfide production. The flow-predicting algorithm and online controller were 3þ þ 2/ 2þ þ 0 Y 2Fe S 2Fe SðsÞ (3) tested by in-depth simulation analysis, and then validated in a full- scale trial. 2þ þ 2/ Y ¼ $ 18 Fe S FeSðsÞ Ksp 1 10 (4) 2. Materials and methods The dosage of iron for sulfide control can be estimated based on the theoretical stoichiometries (Equations (3) and (4)). However, 2.1. Field sewer site and trial real life Fe:S dosing ratios differ from those due to the fact that only þ þ the ionic sulfide species react directly with Fe2 /Fe3 . The forma- 2.1.1. The Bellambi pumping station and rising main tion of FeS precipitates requires the presence in solution of both Field studies were conducted at Bellambi sewage pumping þ Fe2 and S2 at concentrations sufficiently high so that the solu- station (SPS) from Sydney Water Corporation (Sydney, Australia). þ bility product, defined as [Fe2 ]$[S2 ], is higher than the solubility The rising main, which conveys wastewater from the SPS to Wol- constant (Ksp). Thus, higher ratios need to be applied to reach low longong Wastewater treatment plant (WWTP), has a total length of total dissolved sulfide (TDS) levels (WERF, 2007). This results also in 9930 m and a pipe diameter ranging from 603 to 750 mm. Sewage an increase of the concentration of residual iron in solution. As an is first collected in a 0.5 ML wet well and subsequently pumped into example, in the Los Angeles sanitation district, FeCl2 was added to the main through one of the three primary pumps. These are large diameter gravity sewers to control sulfide. An Fe:S molar ratio operated based on the water level. The sewer system has an average of 1.76:1 was necessary to reach target TDS levels above 0.4 mg S/L, dry weather flow of 21 ML/d and an average hydraulic retention while ratios of 3.77:1 and 25.4:1 were required to achieve sulfide time of 4 h. A flow meter is located at the inlet of the rising main, levels of 0.1e0.4 mg S/L and below 0.1 mg S/L, respectively (Padival and flow data is recorded every 15 min. The existing FeCl2 dosing et al., 1995). Besides, the availability of S2 is governed by pH. The unit includes a chemical storage tank, variable speed drive (VSD) lower the pH, the less S2 in equilibrium in the liquid phase, dosing pumps and a programmable logic controller (PLC). An iron 3þ 2þ requiring higher Fe /Fe concentrations to reach the FeS super- chloride solution (10e29% FeCl2; Orica chemicals, Australia) is saturation point (ion product higher than the Ksp). According to dosed through the VSD pump into the wet well to control TDS at the Firer and co-workers, sewage pH below neutrality may require Fe:S discharge of the Bellambi rising main below 0.5 mg S/L. The dosing 2þ ratios much higher than the theoretical stoichiometry (Firer et al., rate (Kg Fe /d) at which the FeCl2 solution is added is adjusted on 2008). However, this should have limited influence on iron an hourly basis based on a pre-defined daily profile with 24 set- dosing for sulfide control in domestic sewers as sewage pH is points (Fig. A.1). Chemical dosing is automatically turned off mostly in the pH range 7.5e8.5 (Nielsen et al., 1998; Houhou et al., when sewage is not pumped into the main. 2009; Sharma et al., 2013). Sewers are very dynamic systems and sewage flow and char- 2.1.2. Field trials of on-line control for ferrous salt dosing acteristics show large variations throughout the day, week and year The field trial to validate the control algorithm consisted of three (Sharma et al., 2013). However, in current practice, iron salt dosing sequential periods, carried out in the same Bellambi pipeline: i) no rates are defined ad hoc, based on the typical sewage flow pattern of chemical dosing (14 days); ii) FeCl2 dosing according to the existing the system and sulfide levels at discharge (de Haas et al., 2008; pre-defined dosing profile (21 days); and iii) FeCl2 dosing based on Ganigue et al., 2011). As a result, generous chemical dosing safety the control algorithm developed here (21 days). Weather was fairly factors need to be applied, especially when targeting tight sulfide constant throughout the three test periods. Mean temperatures discharge limits. Overdosing can be more severe during wet were 23.1 C, 20.6 C and 21.7 C, respectively. Total accumulated weather conditions, when sulfide production declines due to the rainfall during each period was <20 mm, with no rainfall recorded shorter hydraulic retention time (HRT) of diluted sewage in the during the validation of the online control. pipe (as a consequence of the higher flow rates with the rainwater The control algorithm, which calculated the real-time FeCl2 inflow or infiltration) (Chen et al., 2014; Ganigue et al., 2016). On- dosing rate based on the readings from the sewage flow meter at line dosing control can adjust the chemical dosing rates to current Bellambi SPS, was implemented by coding the existing PLC. (and future) state of the sewer system, allowing high-precision, Considering that the ferrous salt is dosed into the wet well, and stable and effective sulfide control with minimal chemical dosing assuming that TDS in the wet well is negligible (confirmed with (Sharma and Yuan, 2009). Recently, Ganigue and co-workers re- data from a sampling campaign over a 24-h period), the FeCl2 ported the successful development of the first on-line chemical dosing rate is thus dependent only on the sulfide production during dosing controller for sulfide mitigation in sewers (Ganigue et al., the transport of the sewage in the rising main pipe section of 2016). The algorithm controlled in real-time the Mg(OH)2 dosing 9930 m length. The sulfide production rate was assumed uniform rate based on pH measurements and allowed improved sulfide along the Bellambi rising main. It was determined during a period control with reduced chemical consumption. without chemical dosing for different HRTs, by dividing the sulfide The sulfide control mechanism of iron salt relies on the chemical produced by the respective hydraulic retention time. The average þ oxidation of sulfide by Fe3 and/or the precipitation of dissolved value was used in both simulations and in the controller. The hy- þ sulfide with Fe2 into FeS (Dohnalek and Fitzpatrick, 1983), and is draulic retention time of each sewage slug in the rising main pipe fundamentally different from that of Mg(OH)2 (elevation of the can be predicted by the ARMA model using data from the flow- wastewater pH to minimise the fraction of sulfide as H2S, to prevent meter as an input, as described in section 3.1.2. After successful its transfer from the liquid to the gas phase) (Gutierrez et al., 2009). factory and site acceptance testings to identify and rectify any po- Therefore, it requires a completely different control approach. To tential errors or compatibility issues, the control algorithm was date there are no studies focusing on the online control of iron salt then brought into live for the control of the FeCl2 dosing pump. dosing in sewers. Besides, the real-time prediction of sewer flow During the trial, the chemical consumption was continuously 304 R. Ganigue et al. / Water Research 135 (2018) 302e310 monitored by recording the dosing rate and the levels of chemical 2.2.3. Long-term performance of the different TDS control strategies in the storage tanks. To estimate the savings, the chemical dosing of The long-term performance and benefits of the online control the real-time control was compared to the pre-defined dosing were investigated in a simulation study using the SeweX model. profile, whereas the effectiveness of sulfide control was assessed by Three different dosing control scenarios were tested, including: 1) online monitoring of TDS levels with a S::CAN unit (DCM Process no chemical dosing, 2) current profiled dosing used at Bellambi Control Ltd., New Zealand) at the discharge point of the Bellambi SPS; and 3) online control (taking into account SRB activity rising main. In addition, offline grab samples of sewage were taken reduction during wet weather). A concentration of 0.5 mg TDS/L at weekly for the determination of iron concentration in the wet well discharge was set as the control target for scenarios 2 and 3. of the SPS and the influent at the WWTP. For each scenario, the system was simulated for a period of 200 days using real flow data from the SPS as input. For the sake of simplicity wastewater composition was assumed constant 2.2. Desktop simulation studies throughout the day, except during rain events, when sewage composition was diluted by rainwater infiltration. Run-off was 2.2.1. SeweX model quantified by comparing the actual SPS flow with the typical dry Simulations studies were used to investigate the impact of rain weather flow profile, and sewage characteristics were corrected fi in ltration on sulfate reducing bacteria (SRB) activity (see sub- accordingly, assuming the same run-off composition as in Ganigue section 2.2.2), and to assess the effectiveness of the feedforward et al. (2016) (Table A.1). The performance of the different controls controller (see sub-section 2.2.3). For that, the Bellambi SPS and was evaluated based on the average sulfide discharge levels and the rising main were modelled using SeweX, a dynamic mathematical amount of FeCl2 dosed. model for the simulation of the physical, chemical and biological processes in sewer systems (Sharma et al., 2008). The model pre- 3. Results and discussion dicts the time and spatial variations of the main wastewater quality parameters in the liquid and gas phases by taking into account the 3.1. Developed strategy for ferrous salt dosing control biochemical processes (carbon, sulfur and nitrogen conversions under aerobic, anaerobic and anoxic conditions) occurring in sewer 3.1.1. Control architecture fi bio lms and in the bulk liquid, chemical processes and equilibria Iron salts oxidize and/or precipitate dissolved sulfide present in fi (e.g. sul de oxidation, precipitation reactions, and acid-base sys- sewage. The time span of these reactions is in the range of seconds tems) and physical processes such as liquid-gas mass transfer. The to minutes depending on sewage pH, buffering capacity or organic pH prediction is achieved based on charge balances (Sharma et al., matter content (Wei and Osseo-Asare, 1995; Kiilerich et al., 2017). 2013). As its inputs, the model requires the sewer system charac- Thus, sulfide control in sewers could be achieved by dosing iron salt … teristics (e.g. network layout, pipe sizes, lengths and slopes, ), at any locations before the discharge point. For practical reasons wastewater composition and hydraulic data (i.e. measured waste- dosing of iron salt is usually conducted in sewage pumping stations, fl fl water ows or details on wet well dimensions, pump ows and allowing to control sulfide along the entire pipe. While in the their operation). The Bellambi model used in this desktop simula- absence of sulfide iron salts may initially react with some other tion analysis had been calibrated in a previous study (Nguyen et al., anions (e.g. phosphate and/or hydroxide), iron ions will be made 2009). available for sulfide precipitation when the latter is produced due to the lower solubility of FeS in comparison to iron phosphate and 2.2.2. Impact of run-off on SRB activity iron hydroxide precipitates (Zhang et al., 2009). A recent lab-scale 3þ fi During heavy rain, the dilution of sewage due to inflow and/or study also demonstrated that the addition of Fe signi cantly fi infiltration (I&I) result in a dramatic decrease in the concentration inhibited SRB activity of anaerobic sewer bio lms (Zhang et al., fi of sulfate and readily biodegradable organic matter, both of which 2009). An additional bene t of upstream dosing could be the fi fi are substrates of SRB, to concentrations limiting SRB activity. Thus, reduction of sul de production rate by sewer bio lms, further I&I flow should be taken into consideration when determining the decreasing the iron dosing requirement. However, controlling iron iron dosing requirements. It is virtually impossible to accurately dosing in upstream locations becomes complex and challenging fl quantify the impact of increasing levels of run-off on SRB activity in due to the plug- ow nature of a sewer line. As the control objective real sewer systems. However, mathematical models allow the is typically to keep the TDS below a certain discharge limit, pre- systematic assessment of such effects. This was investigated using dictive control is needed so that the amount of iron added is fi the SeweX model. Taking as a starting point a default scenario with adequate for precipitating sul de, while avoiding chemical constant flow (8.75 ML/d), long HRT (8 h) and constant sewage overdosing. composition (Table A.1), multiple scenarios with increasing levels Assuming an ideal Fe:S ratio (Eqs. (3) and (4)), iron dosing re- of run-off were simulated, decreasing HRT and diluting sewage quirements (in mg Fe/L sewage) will depend mainly on the amount fi composition accordingly. Run-off characteristics were assumed as of sul de to be removed. This can be calculated based on the TDS in Ganigue et al. (2016) (Table A.1). In parallel, the same model was target set-point and the concentration of TDS at the end of the pipe, fi used to run simulations at the same HRTs by increasing the sewage which encompasses the sul de already present in sewage at the fi flow, but considering the default sewage composition. The reduc- dosing location and the sul de that will be produced during the tion in SRB activity as a function of the dilution was calculated by transport of the sewage from the dosing point to end of the pipe. comparing the sulfide produced in the run-off scenario with that in The former can be determined using an online UV-vis spectrometer dry weather conditions and same flow, according to Equation (5): probe and a pH sensor, located just before the dosing point (Sutherland-Stacey et al., 2008). However, the increase in concen- TDS tration due to the future production of sulfide during transport v ¼ runoff SRB acti ity (5) cannot be measured due to the plug-flow nature of sewers, and TDSdry weather needs to be predicted. Under non-limiting substrate conditions, An exponential decay model was fitted to experimental data there is a linear relationship between the HRT of the sewage in the using Sigmaplot 11 (Systat software, Germany), obtaining the pipe and the amount of sulfide produced (Sharma et al., 2008). model parameters and standard errors. Therefore, sulfide production along a pipe can be predicted if HRT is R. Ganigue et al. / Water Research 135 (2018) 302e310 305 known. Yet the time a wastewater slug will spend in the pipe is QFe_FFþFB ¼ QFe_FF þ kFB$(TDSsp-TDSave) (7) unknown, as it is governed by future wastewater flow that ‘pushes’ the slug through the pipe. The HRT of a slug can, however, be QFe_FFþFB is the iron salt solution dosing rate into the wet well estimated using predictive mathematical tools, for instance Auto (kg/d), kFB the feedback constant of the feedback loop (kg/d$pH Regressive Moving Average (ARMA) models (Chen et al., 2014). unit), TDSsp the TDS control set-point at the discharge point and The proposed control algorithm for the optimised dosing of TDSave the average TDS calculated based on Equation (8). þ þ Fe2 /Fe3 salts comprises two components (Fig. 1). A first feed- ð Þ¼ $ þð Þ$ ð Þ forward (FF) controller predicts the future HRT of a wastewater TDSave n a TDSn 1 a TDSave n 1 (8) slug entering the pipe based on current flow measurements using ARMA models, and calculates the concentration of TDS at the where TDSave (n) is the weighted average TDS at the present time, discharge point based on the TDS concentration at the dosing TDSave (n-1) is the weighted average TDS at the previous sampling point and the sulfide production during transport. The ferrous salt time, TDSn is the measured TDS at the present time, and a is a < < dosing rate is set based on these TDS concentrations and the weighting factor (0 a 1) to be tuned. The weighting factor a is sewage flow. important since it determines whether the feedback signal relies Given the plug-flow nature of sewer systems, the dosing station more on the recent TDS measurements or historical data. relies on the feedforward controller to determine the dosing rate based on the measurement of perturbations (in this case, sulfide 3.1.2. AR model development for HRT prediction, calibration and produced during the transport based on HRT in the pipe). This validation model assumes non-limiting substrate conditions (see Table A.2 for During rain events, rainwater can enter sewer systems due to further information on sulfate concentrations at the inlet and outlet I&I, increasing the sewage flow, but also diluting the sewage. An of the rising main). The controller equation used in the FF loop is: ARMA model for the prediction of future sewage flow into Bellambi SPS was developed following the methodology proposed by Chen et al. (2014). An iterative methodology similar to that described Q $l$b$g$HRT by Sharma et al. (2008) was applied to determine the HRT of a slug Q ¼ sew (6) Fe FF d entering the rising main. The algorithm predicts the flow incoming into the sewer at different time horizons, until the cumulative where QFe_FF is the iron salt solution dosing rate into the wet well volume pumped into the system is higher than the total volume of (kg/d), Qsew is the sewage flow into the rising main (L/d), l the “SRB the sewer pipe. At this time interval the slug has theoretically activity factor” (see section 3.1.3 for further details), b the sulfide reached the end of the pipe, and the total time required for that is production rate (mg S-TDS/L$h), g the FeS precipitation stoichi- defined as the future HRT of that slug in the pipe. A resolution of 2þ 3þ ometry based on Eq. (4) (mg Fe/mg S-TDS), and d the Fe (or Fe ) 15 min was deemed sufficiently accurate for control purposes, and content of the iron salt solution dosed into the wet well. yet not too computationally cumbersome. Although effective and efficient, feedforward controllers are An AR model was calibrated and validated based on more than a subject to a certain degree of uncertainty due to, for example, year of historical flow data from the Bellambi main flowmeter, modelling errors and/or unknown disturbances (Stephanopoulos, following the methodology described in Chen et al. (2014). For this 1984). To prevent this, an additional feedback (FB) loop can be case, a model order of three was found adequate, with the following included to correct control errors. In this case, a FB loop can adjust structure and parameter values: the dosing of the FF controller based on the overall long-term performance of the control, which can be calculated by a xðtÞ¼0:5961xðt 1Þ0:2791xðt 2Þ0:1126xðt 3Þ fi weighted weekly average of the total dissolved sul de concentra- þ vðtÞ (9) tion at the discharge point. This strategy has already been proven to allow the fine-tuning of chemical dosing in sewers (Ganigue et al., Results of 1 h, 3 h and 6 h-ahead predictions for dry weather and 2016). The combined FF and FB controller proposed here is: wet weather conditions are presented in Fig. 2. The Bellambi AR

Fig. 1. Ferrous salt online dosing control scheme. TDSsp: TDS set-point; FF: feedforward control; FB: feedback control. 306 R. Ganigue et al. / Water Research 135 (2018) 302e310 model was able to predict the daily flow dynamics during dry weather conditions, even when targeting the prediction of the future flow 6 h-ahead. The AR model also showed very good predictive performance in wet weather conditions, although a slight delay in the prediction can be observed, especially for longer prediction-time horizons. However, long-term predictions are rarely needed for chemical dosing control during wet weather periods due to the increased flow and reduced HRTs. These results demonstrate the good prediction capabilities of the AR model.

3.1.3. SRB activity correction due to sewage dilution The decrease of SRB activity due to substrate limitation as a result of sewage dilution by stormwater runoff was investigated in a simulation study. Fig. 3 depicts the reduction of SRB activity as a function of the dilution factor (ratio between actual flow and dry weather flow). SRB activity was only significantly reduced at dilution factors higher than 3 and the decrease in activity was non- Fig. 3. Simulated decrease in SRB activity due to sewage dilution during heavy rain linear. This behaviours was modelled using an exponential decay and empirical model fit. model. This empiric expression allowed to describe the reduction of SRB activity as a function of the sewage diluton factor in the pipe. To gain a deeper insights on why the online control out- fi 3.2. Desktop assessment of the on-line control algorithm performed the pro led dosing, the TDS and dosing rates for the performance scenarios under dry weather and wet weather conditions are presented in Fig. 5. In order to test the control algorithm and assess its long-term Fig. 5.C shows that under dry weather conditions both controls fi performance under different weather conditions, simulations behaved in a comparable manner by controlling sul de below the were conducted using the calibrated model of Bellambi SPS. The target concentration of 0.5 mg S/L. Dosing levels calculated during e 2þ 2þ online control was implemented in the model and simulated for the period 10 20 days were 175.6 Kg Fe /d and 152.5 Kg Fe /d, 2þ 200 days. For comparison, simulations were also run for two other respectively. The online control adjusted the Fe dosing rate to the fi scenarios, one with no addition of ferrous salt, and a second one in sul de production and allowed to reduce chemical dosing due to a lower dosing rate during early morning (see Fig. A.2 for further which FeCl2 was dosed according to a pre-defined profile (current dosing strategy at Bellambi). Fig. 4 depicts the TDS concentration at details). fi discharge, and Table 1 summarizes the average TDS at discharge During wet weather conditions the in ltration of rainwater þ fi fi and the average Fe2 daily dosage for each scenario. signi cantly reduced the amount of sul de produced within the e When no iron was dosed to the system, TDS levels at discharge pipe (e.g. days 141 145). The superior performance of the feed- ranged from 2 to 4.5 mg S/L depending on the HRT of the sewage. forward online control was accentuated under those conditions, fi Exception to that were the wet weather periods, when sulfide because it was capable of predicting the decrease in sul de concentration decreased to about 0.5e1 mg S/L due to the shorter production and adjusting the iron dose, reducing the overall fi fi HRTs and run-off infiltration. During the whole 200-day period, chemical dosing (Fig. 5F) without signi cantly affecting sul de both profiled dosing and online control were able to bring the TDS control (Fig. 5D). The average dosing rate during the period e 2þ concentration at discharge very close to the control goal (0.5 mg S/ 138 148 days was, respectively, 206.4 vs 158.0 Kg Fe /d. It is fl L). The average TDS at discharge of the profiled and online control important to note that the change in sewage ow during tran- were 0.11 mg S/L and 0.24 mg S/L, respectively. The online control, sition from dry to wet weather and vice versa (Fig. 5B) led to a however, allowed to reduce the chemical dosage by about 31.4% transient overdosage or underdosage, respectively. This can be þ þ (203.6 Kg Fe2 /d vs 139.7 Kg Fe2 /d), while still meeting the TDS clearly seen for the latter case in Fig. 5D (day 145). The TDS 2þ discharge limits. concentration spiked up to 1.44 mg S/L because the Fe dosing

Fig. 2. Measured and predicted inﬂow to Bellambi SPS wet well during dry weather (a) and wet weather periods (b). R. Ganigue et al. / Water Research 135 (2018) 302e310 307

Fig. 4. Simulation results of the scenarios without ferrous salt dosing, with proﬁled dosing and online control of dosing, at discharge of Bellambi SPS.

Table 1 3.4. Implications and future work Total dissolved sulfide at discharge and average daily dosage for the three scenarios. This work presents the development and validation in a field No dosing Profile dosing Online dosing test of an online algorithm for the control of ferrous salt dosing for Average TDS discharge 1.99 ± 1.08 0.11 ± 0.13 0.24 ± 0.27 sulfide mitigation in sewers. The control relies on online mea- (mg S/L) þ surements from affordable and robust online sensors and can be Average daily dosing (Kg Fe2 /d) e 203.6 139.7 applied to already existing or new dosing station with minimum investment and potentially significant reduction in chemical fl rate was not adjusted until this change in the ow regime was dosing. This is of special importance for iron salt, as these are detected by the AR model. commonly used in large systems with relatively high daily flows (Ganigue et al., 2011). In such systems the fine control of the dosing 3.3. Validation of the control in field trials rates based on the real-time state of the system can lead to significant savings in chemical usage and improved sulfide control. The online control algorithm was tested at Bellambi SPS for a This control was validated for the dosing of ferrous chloride period of 3 weeks. Fig. 6 depicts the concentration of total dissolved through simulations and a field trial, but can be easily implemented sulfide and sewage pH at the discharge of the Bellambi rising main to control the dosing of ferric salts by adjusting the stoichiometry. for three different dosing regimen (no chemical dosing, pre-defined The methodology developed in this study is, however, not only profiled dosing and online dosing control) for a period of 2 days. constrained to iron salt. Oxygen injection and the addition of ni- The average sulfide concentration and ferrous chloride dosing trate salts are also widely used methods for sulfide mitigation in during the whole trial are summarized in Table 2, and compared to sewers (Gutierrez et al., 2008; Jiang et al., 2009). Both oxygen and the other two dosing regimen. Results show that when no ferrous nitrate are oxidant of sulfide (through biological and/or chemical chloride was dosed, sulfide levels at discharge of the rising main processes), although these reactions proceed at a much slower rate would be around 1.65 mg S/L, ranging between 1 and 5 mg S/L than the removal of sulfide by precipitation with iron salt, e.g. depending of the time of the day due to variations on sewage HRT. 30e60 min vs. few seconds (Gutierrez et al., 2008; Auguet et al., Average TDS levels at discharge were decreased to 0.13 ± 0.20 mg S/ 2015; Wei and Osseo-Asare, 1995). Although the working princi- L and 0.23 ± 0.19 mg S/L upon dosing of ferrous chloride based on a ple of these chemicals is different from ferrous salts, the present pre-defined dosing profile or online control, respectively. The TDS algorithm could be also applied with some modifications (i.e. 90% percentile of both control strategies were also very similar, 0.46 mainly the reaction stoichiometry) to control in real-time the and 0.54 mg S/L, respectively. However, ferrous chloride dosage dosing of oxygen or nitrate to sewers for improved control effi- was on average 30.5% lower for the same level of sulfide control. It ciency and chemical savings. It is important to highlight that both is thus clear that the online control achieved comparable control of oxygen and nitrate can be used as electron acceptors by facultative sulfide as the existing profiled dosing, but with reduced chemical anaerobes. In this respect, dosing the chemicals close to the control dosing. This equals to an annual saving of about 100,000 L of FeCl2. point should be preferred to minimise their consumption due to By comparing the dosing rate throughout the day for the pro- side-reactions (making controller tuning easier). Dosing locations filed and online dosing strategies (Fig. 7), the dosing rate of the downstream of the pipe may allow also to measure the concen- profiled control was consistently higher than that of the online tration of sulfide to be oxidized rather than predicting it. control, especially during morning periods (in accordance with the Also, the use of real-time HRT prediction can be of importance in fi ndings of the long-term simulation study). This extra dose of FeCl2 Mg(OH)2 online control. The reduction in chemical dosing achieved resulted in an increase of the chemical use of 30.5%, but did not in this study is of a much higher magnitude than that achieved in fi fi translate in a signi cantly better sul de control. Overall, the Ganigue et al. (2016) for Mg(OH)2 (only 10e15% chemical saving in improved dosing profile using online AR model achieved sulfide respect to the existing profiled control). This may be due to dif- control with lower iron dosage. It is important to highlight that no ferences in the performance of the default controls with which the rain was registered during the trial, and hence the superior per- online controls were compared against, but partly also to the formance of the online control can only be ensured in dry weather improvement derived from the real time HRT prediction by the conditions. 308 R. Ganigue et al. / Water Research 135 (2018) 302e310

þ Fig. 5. Simulated sewage ﬂow, TDS concentration at discharge and Fe2 dosing rates for the three different scenarios in dry (A,C,E) and wet (B,D,F) weather conditions.

ARMA model. Further work in this area should aim at refining flow data or data from online weather stations) into the flow prediction. prediction during the early stage of a wet weather event or when This approach has already been successfully applied to the pre- rain stops, to minimise over and under dosage. Delays or biases in diction of traffic flow based on vehicle detection (Salehinia et al., the prediction under those conditions could be minimised by using 2016) or air pollution based on solar irradiation, wind speed and autoregressive-moving average with exogenous terms (ARMAX) direction (Raimondi et al., 1997), among others. models (Ljung, 1999). This would allow taking into consideration In the present work, the influence of sewage pH on iron salt external information related to weather changes (e.g. rainfall radar stoichiometry has been assumed to be negligible because sewage pH R. Ganigue et al. / Water Research 135 (2018) 302e310 309

Fig. 6. TDS concentration and sewage pH at discharge of Bellambi rising main under different dosing regimens: no dosing, proﬁled dosing and feed-forward online dosing control.

Table 2 to 6 when receiving the discharge of brewery wastewater Comparison of sewage parameters (average ± standard deviation) at Bellambi rising (Sudarjanto et al., 2011). In such conditions, the Fe:S stoichiometry main discharge for three dosing scenarios: no dosing, profiled dosing, feed-forward will need to be adjusted based on sewage pH. In a similar way, in this dosing and. study we have controlled sulfide at discharge below 0.5 mg S/L, Parameters No dosing Profiled dosing Online control based on current practice of the water industry in Australia. How- (n ¼ 21 days) (n ¼ 14 days) (n ¼ 21 days) ever, water utilities may decide to target more stringent sulfide Sewage flow (ML/d) 21 20.9 20.9 levels, which could affect the Fe:S ratios required to attain such pH 7.39 ± 0.23 7.33 ± 0.22 7.38 ± 0.29 sulfide concentrations at discharge. Thus, this should be taken also ± ± ± Average TDS (mg S/L) 1.65 1.15 0.13 0.20 0.23 0.19 into consideration when calculating the dose of iron salt. Last but not 90% TDS (mg S/L) 3.08 0.46 0.54 þ Iron dosage (Kg Fe2 /day) 0 213.5 ± 30.4 148.28 ± 6.40 least, this study did not consider the fact that rain and storm events may have a direct impact on the sulfide producing capacity of sewers, by detaching part of the sewer biofilm or removing sewer sediments. To date there is little knowledge on these phenomena, is usually above neutrality (Nielsen et al., 1998; Houhou et al., 2009; but further research should elucidate whether or not theycan lead to Sharma et al., 2013). However, this may only be true for sewers a significant decrease in sewer sulfide production and its duration. conveying purely domestic sewage. Certain sewer systems may Finally, corrosion control in sewers can be also targeted at a receive the discharge of trade waste or industrial wastewater that network level, as recently shown by Liu and co-workers (Liu et al., can lead to more acidic pH conditions. For instance, Sudarjanto et al. 2013, 2016). Magnesium hydroxide and iron salt are “conserved” observed a decrease in the pH of lab-scale sewer systems from pH 7 chemicals, and thus a single dosing point may allow the control of

Fig. 7. Comparison of FeCl2 dosing profiles for the profiled dosing and feed-forward online control. 310 R. Ganigue et al. / Water Research 135 (2018) 302e310 sulfide corrosion at multiple locations of the network. This Dohnalek, D.A., Fitzpatrick, J.A., 1983. The chemistry of reduced sulfur species and approach will result in fewer dosing stations, decreasing the capital their removal from groundwater supplies. J. Am. Water Works Assoc. 75 (6), 298e308. expenditure for sulfide control. On the contrary, control complexity Firer, D., Friedler, E., Lahav, O., 2008. Control of sulfide in sewer systems by dosage will multiply as the size of the network grows. Network-wide of iron salts: comparison between theoretical and experimental results, and e corrosion control is not incompatible, but rather complementary, practical implications. Sci. Total Environ. 392 (1), 145 156. Ganigue, R., Gutierrez, O., Rootsey, R., Yuan, Z., 2011. Chemical dosing for sulfide to the online control of chemical dosing in single pipes. If the control in Australia: an industry survey. Water Res. 45 (19), 6564e6574. number of corrosion and odour hot-spots is limited, and/or these Ganigue, R., Jiang, G., Sharma, K., Chen, J., Vuong, L., Yuan, Z., 2016. On-line control are geographically isolated, dosing control should be preferably of magnesium hydroxide dosing for sulfide mitigation in sewers: algorithm fi fi development, simulation analysis and eld validation. J. Environ. Eng. 142 (12), carried out at a single-pipe level. On the contrary if sul de issues 04016069. are widely distributed within a network and the hot-spots are Gutierrez, O., Mohanakrishnan, J., Sharma, K.R., Meyer, R.L., Keller, J., Yuan, Z., 2008. concentrated in a small area, it would be more cost effective to Evaluation of oxygen injection as a means of controlling sulfide production in a fi sewer system. Water Res. 42 (17), 4549e4561. address sul de control on a network basis. In this light, the choice Gutierrez, O., Park, D., Sharma, K.R., Yuan, Z., 2009. Effects of long-term pH elevation of strategy should be based on a preliminary analysis of the on the sulfate-reducing and methanogenic activities of anaerobic sewer bio- network and the needs for sulfide control. films. Water Res. 43 (9), 2549e2557. Houhou, J., Lartiges, B.S., Hofmann, A., Frappier, G., Ghanbaja, J., Temgoua, A., 2009. Phosphate dynamics in an urban sewer: a case study of Nancy, France. Water 4. Conclusions Res. 43 (4), 1088e1100. Jameel, P., 1989. The use of ferrous chloride to control dissolved sulfides in inter- e - A novel and robust online dosing control strategy was estab- ceptor sewers. J. Water Pollut. Control Fed. 61 (2), 230 236. Jiang, G., Sharma, K.R., Guisasola, A., Keller, J., Yuan, Z., 2009. Sulfur transformation lished for ferrous salt, delivering improved sulfide control in in rising main sewers receiving nitrate dosage. Water Res. 43 (17), 4430e4440. sewers and a reduction of chemical dosing. The control con- Jiang, G., Sun, J., Sharma, K.R., Yuan, Z., 2015. Corrosion and odor management in e siders the fluctuation of sewage flow, pH, sulfide levels and also sewer systems. Curr. Opin. Biotechnol. 33, 192 197. Kiilerich, B., van de Ven, W., Haaning Nielsen, A., Vollertsen, J., 2017. Sulfide pre- the perturbation from rainfall. cipitation in wastewater at short timescales. Water 9, 670e682. - ARMA models can predict sulfide production in sewers based on Liu, Y., Ganigue, R., Sharma, K., Yuan, Z., 2013. Controlling chemical dosing for current flow measurements, which in turn can be used to sulfide mitigation in sewer networks using a hybrid automata control strategy. Water Sci. Technol. 68 (12), 2584e2590. determine the dose of ferrous salt required for cost-effective Liu, Y., Ganigue, R., Sharma, K., Yuan, Z., 2016. Event-driven model predictive control sulfide control. of sewage pumping stations for sulfide mitigation in sewer networks. Water - The control strategy was evaluated using both a state-of-the art Res. 98, 376e383. Ljung, L., 1999. System Identification - Theory for the User. Prentice-Hall. sewer model and in a real sewer site. It achieved high level of Nielsen, P.H., Raunkjaer, K., Hvitved-Jacobsen, T., 1998. Sulfide production and sulfide control at <0.5 mg S/L while reducing chemical dosing wastewater quality in pressure mains. Water Sci. Technol. 37 (1), 97e104. up to 30%. Nguyen, T., Nobi, N., Soliman, A., Sharma, K.R., 2009. A case study of using sulphide fi model to optimise chemical dosing for odour and corrosion control. In: - The control strategy proposed here can be adopted or modi ed OzWater’ 09 Australia's National Water Conference and Exhibition 16e18 for the dosing control of other commonly used chemicals in March 2009 Melbourne (Australia). sewers. Padival, N.A., Kimbell, W.A., Redner, J.A., 1995. Use of iron salts to control dissolved sulfide in trunk sewers. J. Environ. Eng. 121 (11), 824. Pikaar, I., Sharma, K.R., Hu, S., Gernjak, W., Keller, J., Yuan, Z., 2014. Reducing sewer Acknowledgements corrosion through integrated urban water management. Science 345 (6198), 812e814. The authors acknowledge the financial support provided by the Raimondi, F.M., Rando, F., Vitale, M.C., Calcar, A.M.V., 1997. A Short-term air pollution predictor for urban areas with complex orography. Application to the town Australian Research Council and many members of the Australian of Palermo. In measurements and modelling in environmental pollution. Trans. water industry through LP0882016 the Sewer Corrosion and Odour Ecol. Environ. 19, 261e271, 1743e3541. fi Research (SCORe) Project (www.score.org.au) and the ARC Rickard, D., Luther III, G.W., 2007. Chemistry of iron sul des. Chem. Rev. 107, 514e562. LP160101040. The authors would like to thank Syndey Water Cor- Salehinia, S., Ghaffari, A., Khodayari, A., Khajavi, M.N., Alimardani, F., 2016. poration for supporting the field trial, sampling campaign and data Modelling and controlling of car-following behavior in real traffic flow using collection. Ramon Ganigue gratefully acknowledges support from ARMAX identification and model predictive control. Int. J. Automot. Technol. 17 (3), 535e547. Ghent University BOF postdoctoral fellowship (BOF15/PDO/068). Dr Sharma, K.R., Yuan, Z., De Haas, D., Hamilton, G., Corrie, S., Keller, J., 2008. Dynamics Guangming Jiang is the recipient of a Queensland State Govern- and dynamic modelling of H2S production in sewer systems. Water Res. 42 ment's Early Career Accelerate Fellowship and an Australian (10e11), 2527e2538. Sharma, K.R., Yuan, Z., 2009. A model based investigation of chemical dosing control Research Council DECRA Fellowship (DE170100694). for sulfide management in sewers. In: Instrumentation, Control and Automa- tion, Cairns. Appendix A. Supplementary data Sharma, K., Ganigue, R., Yuan, Z., 2013. pH dynamics in sewers and its modeling. Water Res. 47, 8086e6096. Stephanopoulos, G., 1984. Chemical Process Control: an Introduction to Theory and Supplementary data related to this article can be found at Practice. Prentice-Hall, Englewood Cliffs, New Jersey, USA. https://doi.org/10.1016/j.watres.2018.02.022. Sudarjanto, G., Sharma, K.R., Gutierrez, O., Yuan, Z., 2011. A laboratory assessment of the impact of brewery wastewater discharge on sulfide and methane production in a sewer. Water Sci. Technol. 64 (8), 1614e1619. References Sutherland-Stacey, L., Corrie, S., Neethling, A., Johnson, I., Gutierrez, O., Dexter, R., Yuan, Z., Keller, J., Hamilton, G., 2008. Continuous measurement of dissolved Auguet, O., Pijuan, M., Guasch-Balcells, H., Borrego, C.M., Gutierrez, O., 2015. Im- sulfide in sewer systems. Water Sci. Technol. 57 (3), 375e381. plications of downstream nitrate dosage in anaerobic sewers to control sulfide USEPA, 1992. Detection, Control and Correction of Hydrogen Sulfide Corrosion in and methane emissions. Water Res. 68, 522e532. Existing Wastewater Systems. U.S. Environmental Protection Agency, Wash- Barjenbruch, M., 2003. Prevention of odour emergence in sewage networks. Water ington, D.C.. Technical Report, 832-R-92e9001. Sci. Technol. 47 (7e8), 357e363. Wei, D., Osseo-Asare, K., 1995. Formation of iron monosulfide: a spectrophotometric Bertran de Lis, F., Saracevic, E., Matsche, N., 2007. Control of sulphide problems in study of the reaction between ferrous and sulfide ions in aqueous solutions. pressure sewers. In: Vortrag (Ed.), 6th International Conference on Sustainable J. Colloid Interface Sci. 174 (2), 273e282. Techniques and Strategies in Urban Water Management, pp. 965e972. Lyon, WERF, 2007. Minimization of Odors and Corrosion in Collection Systems, Phase I. France. Water Environment Research Foundation WERF, IWA, Alexandria, VA, Chen, J., Ganigue, R., Liu, Y., Yuan, Z., 2014. Real-time multi-step prediction of sewer pp. 22314e31177. flow for online chemical dosing control. J. Environ. Eng. 140, 04014037. Zhang, L., Keller, J., Yuan, Z., 2009. Inhibition of sulfate-reducing and methanogenic de Haas, D., Sharma, K.R., Corrie, S., O'Halloran, K., Keller, J., Yuan, Z., 2008. Odour activities of anaerobic sewer biofilms by ferric iron dosing. Water Res. 43 (17), control by chemical dosing: a case study. Water 35 (2), 138e143. 4123e4132.