A novel control system approach to enhance the efficiency of solar photo-Fenton microcontaminant removal in continuous flow raceway pond reactors

Chemical Engineering Journal 455 (2023) 140760 Available online 5 December 2022 1385-8947/© 2022 Published by Elsevier B.V. A novel control system approach to enhance the efficiency of solar photo-Fenton microcontaminant removal in continuous flow raceway pond reactors D. Rodríguez-García a , c , P. Soriano-Molina a , c , J.L. Guzm´ an S´ anchez b , c , J.L. García S´ anchez a , c , J.L. Casas L´ opez a , c , * , J.A. S´ anchez P´ erez a , c a Chemical Engineering Department, University of Almería, Ctra. Sacramento s/n, Almería, Spain b Department of Informatics, Ctra. Sacramento s/n, University of Almería, Almería, Spain c Solar Energy Research Centre (CIESOL), Joint Centre University of Almería – CIEMAT, Ctra. Sacramento s/n, Almería, Spain ARTICLE INFO Keywords: Advanced oxidation Photo-Fenton Wastewater decontamination Cost efficiency Control system ABSTRACT This work presents a control approach for the continuous flow operation of the solar photo-Fenton process in raceway pond reactors designed for micropollutant (MP) removal from urban wastewater treatment plant secondary effluents. The control system was designed using the mechanistic and semiempirical kinetic model of the photo-Fenton process at acidic pH developed and validated in previous work. Afterwards, a simulation study to demonstrate the viability of the control system was conducted under different operating conditions (hydraulic residence time and liquid depth) for solar irradiance and water temperature variation over the year. Two liquid depths (10 and 20 cm) and two hydraulic residence times (15 and 30 min) were selected as operating conditions to study the system performance for different MP removal control setpoints (70 % and 90 %). For 90 % MP removal setpoint, cost efficiency reductions of up to 21 % were reached, which denotes that the process efficiency is significantly influenced by the control setpoint. This is because highly demanding MP removal setpoints are achieved through high non-linear reagent consumption, resulting in low operating cost efficiencies of the process. On the other hand, cost efficiency improvements from 45 to 50 % were achieved when an 70 % MP removal setpoint was adopted, when comparing the manual operation of the process with the automatic mode. The presented results demonstrate not only the feasibility of implementing an automatic approach for the solar photo-Fenton process, but also the need for an optimised, efficient and controlled operation to upgrade its competitiveness versus conventional technologies. 1. Introduction Water protection and conservation have become particularly important in recent decades, as environmental studies predict a continuing and alarming increase in hydric stress as a result of population growth and caloric food demand [1], among other factors. The agricultural industry is the major water consumer worldwide, as it is estimated that 60 % of available fresh water in developed countries is used for crop irrigation, while this percentage can rise up to 90 % in developing countries [2]. As a result, there is growing awareness about the environmental sustainability of water in order to deal with these increasing demands. Moreover, there is currently great concern about the analytical detection of so-called micropollutants or contaminants of emerging concern (CECs) [3]. Adverse effects on the aquatic environment and public health due to the presence of micropollutants have been reported, such as the generation of antibiotic-resistant bacteria [4] and environmental imbalance brought about by endocrine disruptors [5]. Currently, conventional Wastewater Treatment Plants (WWTPs) are not designed for the removal of these compounds. It is therefore of great importance to improve WWTPs by enhancing tertiary treatments that achieve high quality effluents, allowing their discharge or reuse with all assurances of health and sustainability. These tertiary treatments rely on Advanced Oxidation Processes (AOPs) [6], which are based on in situ generation of highly oxidative transient species, meaning micropollutants are degraded and inactivated by oxidation. Thus, the main challenge of tertiary treatment based * Corresponding author at: Chemical Engineering Department, University of Almería, Ctra. Sacramento s/n, Almería, Spain. E-mail address: [email protected] (J.L. Casas L´ opez). Contents lists available at ScienceDirect Chemical Engineering Journal journal homepage: www.elsevier.com/locate/cej https://doi.org/10.1016/j.cej.2022.140760 Received 2 August 2022; Received in revised form 12 October 2022; Accepted 30 November 2022 Chemical Engineering Journal 455 (2023) 140760 2 on AOPs is to develop robust, efficient and cost-effective technologies, facilitating their implementation in WWTPs. Solar photo-Fenton operated in raceway pond reactors is well suited for this purpose due to high micropollutant removal efficiencies being achieved with low investment and operating costs. It relies on the generation of hydroxyl radicals through the chemical reaction of hydrogen peroxide and iron (photocatalyst) in an acidic medium, under natural or artificial UV radiation [7]. A resulting feature of environmental sustainability emerges as the process can be run completely using solar radiation. The Photo-Fenton process has been extensively studied and some of the mechanism steps are represented by the reactions below [8]: Fe 2+ +H 2 O 2 → Fe 3+ +HO•+HO − (Reaction 1) Fe 3+ +H 2 O 2 → Fe 2+ +H + +HO 2 •(Reaction 2) Fe 2+ +HO•→ Fe 3+ +HO − (Reaction 3) H 2 O 2 +OH•→ H 2 O +HO 2 •(Reaction 4) Fe 3+ +HO 2 •→ Fe 2+ +O 2 +H + (Reaction 5) Fe 2+ +HO 2 •→ Fe 3+ +H 2 O 2 (Reaction 6) In the presence of ultraviolet visible radiation (UV–vis): Fe 3+ +H 2 O +hv → Fe 2+ +HO•+H + (Reaction 7) The first reaction known as the Fenton reaction, initiates a redox cycle in Fe 2+ /Fe 3+ through which hydroxyl radicals (HO•) are generated as the most active intermediate species, HO 2 •and O 2 • − being other species of this complex AOP (not shown [9]). In short, the process can be viewed as a source of oxidizing species able to react with micropollutants, leading to their degradation (effective reactant usage), or with hydrogen peroxide as Reaction (4) (ineffective reactant consumption). The photo-Fenton process applied for urban WWTP secondary effluent decontamination has been modelled by means of semi-empirical kinetic models [10], which incorporate some non-measurable intermediate species and chemical reactions in the mechanistic model in order to effectively model the system response. On the other hand, simple models for the control of the process have been developed using Experimental Designs [11]. Nevertheless, these models are static and their reliability and accuracy are much lower than the results obtained with semi-empirical kinetic models. Moreover, the simple models developed are focused on dissolved organic carbon (DOC) reduction in highly polluted industrial wastewater, so no micropollutant removal prediction can be made. Therefore, their use in the design and operation of a control system oriented towards micropollutant removal is not feasible. Solar photo-Fenton kinetics, micropollutant removal and operating costs have been studied in detail in previous works; both in batch and continuous flow operation mode [12,13]. It has been proven that the process is feasible from a physical, operational and economic perspective [10,14]. Hydrogen peroxide consumption plays an important role in determining the operating costs of the process due to both its high commercial price (compared to iron salts and acid prices) and its elevated consumption during operation. The measurement of dissolved oxygen concentration (DO) as a reaction stage indicator related to H 2 O 2 consumption during photo-Fenton industrial wastewater treatment has been widely studied [15,16,17]. These works present a functional, economic and valuable alternative for correlating H 2 O 2 concentration with DO measurement in the medium in highly polluted water, and mineralization was the primary goal. As for effluents with low H 2 O 2 concentration (such as WWTP secondary effluents regenerated by means of solar photo-Fenton), DO measurements cannot be correlated with slight variations in H 2 O 2 concentration. An automatic hydrogen peroxide dosage system would keep the concentration optimised throughout the process. Nevertheless, to achieve this aim, online measurements of hydrogen peroxide and micropollutants would be required. Hydrogen peroxide sensors available for industrial applications are not suitable for application in wastewater treatments mainly due to the interacting water matrix effect, as well as relatively low and rapid changes of hydrogen peroxide (as in the photo-Fenton process). As a result, the evaluation of new effective and economic alternatives for online sensing of hydrogen peroxide are needed in solar photo-Fenton processes. As far as the authors know, there is no previous work on the design of a functional and feasible control system for the solar photo-Fenton process operation in continuous flow mode, focused on the removal of micropollutants contained in WWTP secondary effluents. Most of the studies in this area are focused on the evaluation of the solar photoFenton process operated in batch mode [18]. In addition, studies on the continuous flow operation of the process are based on a manual operation [13]. The aim of this work is to design and evaluate the performance of a control system approach in a solar photo-Fenton raceway pond reactor (RPR) operating in continuous flow mode to ensure a minimum and constant micropollutant removal percentage as well as reducing reagent costs, which allows upgrading the process competitiveness versus conventional technologies. To this end, a simulation study to demonstrate the viability of an automatic control approach has been conducted under different operating conditions (HRT and liquid depth) for solar irradiance and water temperature variation over the year. 2. Methods 2.1. Simulation framework All the simulations were performed for a solar photo-Fenton RPR operating in continuous flow mode. The raceway (Fig. 1) has a total surface area of 0.52 m 2 and consists of two 1 m channels, each 0.2 m wide and connected by U-shaped bends. Liquid depths (D) of 10 and 20 cm were chosen and each one was simulated with both 15 and 30 min hydraulic residence time (HRT) based on previous experimental works [10]. The operating conditions (D and HRT) were simulated for different solar radiation and average water temperature values for three consecutive days over different seasons of the year 2020 (Table 1). These values were measured with a solar radiometer located at the University of Almería, Spain (36.83, −2.40). Maximum irradiance was 22 and 38 W⋅m −2 for low (February and November) and high (May and July) radiation months, respectively. Temperature data were collected from historical records at EL BOBAR WWTP (Almería, Spain). 2.2. Volumetric rate of photon absorption (VRPA) The average volumetric rate of photon absorption (VRPA) is defined as the number of photons that are absorbed per unit of time and unit of reaction volume. It depends on geometrical features of the reactor, the concentration of absorbing species in the medium, the optical properties Fig. 1. Scheme of the RPR used for simulations. D. Rodríguez-García et al. Chemical Engineering Journal 455 (2023) 140760 3 of the system and the photon source. For VRPA calculation the volumetric absorption coefficient of the photon absorbing species (Fe 3+ ) is considered, which is experimentally determined for every wavelength at the real water matrix. Therefore, the water turbidity is a parameter that is included in the computation of photon absorption as the model is calibrated in the real WWTP secondary effluent. Solar UV radiation is divided into two fractions: direct radiation and diffuse radiation. Direct radiation corresponds to the fraction of solar UV radiation that comes directly from the sun, while the radiation that is scattered in the atmosphere and reaches the earth’s surface corresponds to the diffuse radiation. Consequently, total VRPA is calculated as the sum of the VRPA values for direct (Eq. (1)) and diffuse (Eq. (2)) radiation [19]. VRPAdirect =1 D⋅∑ λ qwdirect,λ⋅[1−exp(−kλ⋅D cosθdirect)] (1) VRPAdiffuse =1 D⋅∫D 0∑ λ 2⋅qwdiffuse,λ⋅kλ⋅∫Φc 0 e−kλ⋅x cosΦdiffuse ⋅sinΦdiffuse⋅dΦdiffuse⋅dx (2) where q wdirect,λ and q wdiffuse,λ are the spectral distributions of direct and diffuse radiation at a specific wavelength, respectively (Einstein- ⋅m −2 ⋅nm −1 ), k λ is the spectral volumetric absorption coefficient of the photon absorbing species (m −1 ), D is the liquid depth (m) and ×the spatial coordinate (m). θ direct and Φ diffuse represent the refracted angle of the solar zenith angle (θ s ) inside the water and the angular coordinate of diffuse radiation, respectively. Φc is the critical angle of the angular coordinate (bounded in the range 0-90◦). The volumetric Napierian absorption coefficient of Fe 3+ at a specific wavelength is determined by the following equation: kFe3+,λ=2.3⋅ ε Fe3+,λ⋅CFe3+(3) where ε Fe3+,λ is the molar absorptivity of Fe 3+ at a specific wavelength (m 3 ⋅mol −1 ⋅m −1 ) and CFe3+is the molar concentration of Fe 3+ (mol•m −3 ). 2.3. Photo-Fenton dynamic model The plant simulation was performed using the semi-empirical kinetic model of acidic photo-Fenton process, developed and validated in a previous work [10]. To put it briefly, the model states for iron are Fe 2+ , Fe 3+ and Fe 3+* (activated). The reaction between H 2 O 2 and Fe 2+ (Fenton reaction) involves the generation of hydroxyl radicals, giving rise to secondary reactions with target micropollutant (GBP, O-DSMT and ODSMV), organic matter (OM) and hydrogen peroxide (H 2 O 2 ). The kinetic model includes target MP degradation and existing inefficient reactions between the hydroxyl radicals and hydrogen peroxide and organic matter, thus providing conservative, realistic and accurate predictions of the real process. The inorganic matter also affects the kinetics of the photo-Fenton process; however, as the process is operated at acidic pH, it is not necessary to consider it in the model because the HCO 3 − /CO 3 2+ are removed by desorption when acidifying the water at pH 2.8. The model was successfully validated for the operation of the solar photoFenton process in continuous flow mode under real operating conditions: real WWTP secondary effluent and solar radiation. In this way, the accuracy, robustness and applicability of the model provides a highly valuable tool for its application in control purposes. The reported kinetic model did not consider the effect of liquid temperature in the process, as all kinetic constants were estimated for an average liquid temperature of 25 ◦C. In this work, the kinetic model was enhanced by including the impact of liquid temperature on the plant efficiency. For this purpose, the kinetic constants corresponding to Fenton and Fenton-like reactions (Reaction (1) and (5), respectively as numbered in [10]) were estimated for each month based on an experimentally obtained Arrhenius correlation (Table 2). It was found that the effect of temperature variation on the rest of reactions included in the kinetic model was not significant for the existing liquid temperature range in a photo-reactor operating annually under natural environmental conditions in Almería (14–30 ◦C, Table 1). The dynamic model for acidic photo-Fenton operation (Table 3) was obtained by imposing mass balances for each of the model states, assuming the hypothesis of continuous flow operation and perfect mixing. Q w (disturbance) is the water feed flow rate, while QFe and QH2O2 are the ferrous iron and hydrogen peroxide reactant feed flows (manipulated variables), respectively. The reactor volume (V) is set by overflow, and the outgoing flow rate is defined by Eq. (4). Qt=Qw+QFe +QH2O2(4) In all cases, the simulation starts with the steady state reached overnight through batch Fenton decontamination, assuming that initially the reactor is loaded with raw wastewater using 0.1 and 0.85 mM concentrations of Fe 2+ and H 2 O 2 , respectively. The corresponding values are shown in Table 3 as “Initial value”. Note that for practical purposes, microcontaminants and radical species are completely depleted inside the reactor before pumping starts. 2.4. Control strategy The main process variable in a photo-Fenton reactor used for wastewater tertiary treatment (WTT) is the micropollutant outlet concentration. Both H 2 O 2 and Fe affect micropollutant removal. Nevertheless, the impact of iron concentration is much greater because the process is predominantly limited by iron photoreduction, as H 2 O 2 is generally supplied to the reactor in stoichiometric excess. For this reason, the use of iron feed flow (QFe) as a manipulated variable is the obvious alternative to control micropollutant concentration ([MP]) inside the photo-reactor. Hydrogen peroxide concentration ([H 2 O 2 ]) must be controlled due to its great impact on the process operating cost. The direct relation between H 2 O 2 feed flow (QH2O2) and hydrogen peroxide concentration ([H 2 O 2 ]) determines the second pairing (Fig. 2). Variable pairing and coupling were also determined by using the relative gain array (RGA) method [20] in order to mathematically verify if the selected variable pairing is the best choice from an overall process control perspective. From this study, it was concluded that the best variable pairing is to control H 2 O 2 and MP concentration inside the reactor by modifying QH2O2 and QFe, respectively, as initially proposed. Moreover, by analysing the RGA it could be argued that coupling is negligible, meaning it is possible to tune the controllers independently. This fact has been verified according to the resulting RGA value of λ 11 = Table 1 Radiation measurement data (measured at the University of Almería, Spain) used for the plant simulations over different seasons (2020 year). Month Days Average Water Temperature (◦C) Night Day February 2, 3 and 4 14 18 May 11, 12 and 13 22 26 July 2, 3 and 4 26 30 November 5, 6 and 7 16 22 Table 2 Kinetic constants estimated for the photo-Fenton kinetic model at acidic pH operation. Kinetic constant Arrhenius eq. Unit k 1 (Fenton) 5.93⋅106⋅e −(34200 R⋅T )mM −1 •min −1 k 5 (Fenton-like) 6.28⋅105⋅e −(42000 R⋅T )mM −1 •min −1 D. Rodríguez-García et al. Chemical Engineering Journal 455 (2023) 140760 4 1.00011 (See Table S1, Supplementary Data). Water temperature (T), water feed flow (Q w ), and solar radiation measurements (Rad) are considered to estimate [MP] and [H 2 O 2 ] by means of model simulations, so this data is continuously fed back to the controller (Fig. 2). Total micropollutant concentration was calculated as the sum of the tracking pollutant concentrations considered in the dynamic model: Gabapentin, O-desmethyltramadol and O-desmethylvenlafaxine (GBP, O-DSMT and O-DSMV in Table 3). 2.5. Control system design The design of the control system is based on a linear black-box model obtained by using simulated data from the photo-Fenton dynamic model laid out in Section 2.3. The structure and the resulting parameters of the linear black-box model is shown in Subsection 2.5.1. The obtained linear models were exclusively used for controller tuning as presented in Subsection 2.5.2. Once the controllers were designed, the simulation of the plant was performed using the mechanistic photo-Fenton model (dynamic grey-box model, Section 2.3). 2.5.1. System modelling Typically in process control, the controllers are designed based on linear models obtained around the operating point of the process. Thus, to tune the system controllers, it is necessary to obtain the main linear dynamical models to relate process variables ([MP] and [H 2 O 2 ]) with the manipulated variables (QFe and QH2O2, respectively). As the photoFenton dynamic model (Table 3) is highly nonlinear, an empirical evaluation of the system is conducted using the reaction curve method and the first-order system. First-order models with time delay are linear models used in many different applications when the main dynamics are reasonably damped. Most industrial processes exhibit a dynamic response characteristic for first-order systems hence their application and study in industry is widespread [21]. These models define a simple mathematical input–output relationship, meaning no effort is made to determine what is actually inside the so-called black box. A first-order model with time delay, expressed in the Laplace domain is given by the following equation: G(s) = Y(s) U(s)=K τ s+1e−θs(5) where Y and U are the output and input variables of the system, respectively, K represents the static gain, τ stands for the time constant, and θ is the time delay of the system. The most widely-used method for parametric estimation is to apply a step-like change in the input variable of the system following the reaction curve method. As such, based on the system response, it is straightforward to determine the gain, the time constant and the time delay, as shown in [22]. Once the system is modelled, it is possible to tune the controller using appropriate tuning methods for first-order systems, depending on the process parameters and the control specifications. Along these lines, in this work, the system behaviour was evaluated through simulations by applying step-like changes on the control signals around the operating points to the photo-Fenton dynamic model. Specifically, six step changes for every input variable were applied (three positive and three negatives): one ±12,5% step followed by a ±25 % step and one single ±50 % step, with respect to the operating point. This study was performed in two different radiation conditions (Table S1, Supplementary Data). As a result, the first-order dynamic system parameters were calculated by averaging the parameters obtained for different step sizes and signs, in order to capture the effect of the high non-linearity of the real dynamic system. 2.5.2. Feedback controller Proportional-Integral-Derivative (PID) control is the most common choice as a feedback controller in industry. The majority of the industrial processes are controlled by PID controllers, and mainly by PI control [20]. It is simply to understand and implement, and in the vast majority of cases it offers good results from a process control perspective. Thus, micropollutants and hydrogen peroxide controllers were designed based on a classical PI control (Eq. (6)). Table 3 Dynamic model equations for acidic pH solar photo-Fenton. State variable Initial value (mM) Mass balance (mM•min −1 ) Fe 2+ 3.995•10 -4 d[Fe2+] dt =(QFe⋅[Fe2+]e−Qt⋅[Fe2+]) V−r1+r4+ r5 Fe 3+ 0.0996 d[Fe3+] dt = − Qt⋅[Fe3+] V−r2+r1+r3−r5 Fe 3+ •0 d[Fe3+*] dt = − Qt⋅[Fe3+*] V+r2−r3−r4 OM 0.3097 d[OM] dt =(Qw⋅[OM]e−Qt⋅[OM]) V−r6 MX 0.2204 d[MX] dt = − Qt⋅[MX] V+r6+9⋅r8+15⋅r9+16⋅r10 H 2 O 2 0.0307 d[H2O2] dt =(QH2O2⋅[H2O2]e−Qt⋅[H2O2]) V−r1− r5−r7 R 6.328•10 -11 d[R] dt = − Qt⋅[R] V+r1+r4−r6−r7−r8−r9−r10 GBP 1.957•10 -8 d[GBP] dt =(Qw⋅[GBP]e−Qt⋅[GBP]) V−r8 O-DSMT 1.919•10 -11 d[O-DSMT] dt =(Qw⋅[O-DSMT]e−Qt⋅[O-DSMT]) V− r9 O-DSMV 7.2287•10 -9 d[O-DSMV] dt =(Qw⋅[O-DSMV]e−Qt⋅[O-DSMV]) V− r10) Fig. 2. Control Scheme. D. Rodríguez-García et al. Chemical Engineering Journal 455 (2023) 140760 5 The PI control expression is shown in the following equation: Uc(s) = (1 Tis+1)KpE(s)(6) where T i is the integral time and K p the proportional gain. These are the classical parameters of a PI controller to be estimated using proper tuning methods. This type of controller shows proportionality with both the error signal as well as with the integral of the same. As a result, the controller initially responds proportionally to an error signal and then eliminates the residual error in steady state [23]. To tune the controller parameters properly, the Lambda Method is one of the most widely known approaches, mainly because of its capacity to generate smooth and non-oscillatory control responses. For a first-order process with gain K, time constant τ and delay time θ (see previous section), the tuning rule is the following [24]: Kp= τ K(λ+θ)Ti= τ (7) where λ is a parameter related to the closed-loop time constant, or in other words, the speed of the control system to reach the desired setpoint. This parameter should be selected based on each control type problem, and typically values of λ= ατ (with α ∈ [0.5,1]) and λ =θ are used. Note that this parameter can be used as a trade-off between performance and control effort (typically related with running costs). High values of λ give less control effort and worse system performance, while low values result in better performance but higher control effort. 2.6. Cost estimation 2.6.1. Reagent cost calculation Reagent cost was determined based on the characteristics and prices of commercially available reagents. Hydrogen peroxide stock solution cost was estimated based on a commercial price of 0.67 € •kg −1 (50 % w/ v, Brenntag, Essen, Germany), and ferrous sulphate heptahydrate (99.9 % w/w, Brenntag, Essen, Germany) was used as ferrous source, with a price of 0.47 € •kg −1 . All simulations were performed using 168.37 mM concentration for H 2 O 2 stock solution and 8.55 mM for Fe 2+ . As a result, the stock solutions prices considered in the simulations were 9.17 and 1.18 € •m −3 , for hydrogen peroxide and ferrous iron stock solution consumption, respectively. Reagent cost is calculated as the sum of Fe 2+ and H 2 O 2 operating cost based on reactant feed flows (QFe and QH2O2). Therefore, by integrating the instantaneous reactant feed flows the stock solution volume consumed during the simulation in terms of m 3 (VFe and VH2O2, for Fe 2+ and H 2 O 2 consumption, respectively) is obtained. Total reagent cost is calculated (ReagentCost)considering commercial prices of the reagent stock solutions, as shown in Eq. (8). ReagentCost =VFe⋅1.18 € m3+VH2O2⋅9.17 € m3 N⋅tday 24 (8) where VFe and VH2O2 are the Fe 2+ and H 2 O 2 stock solution volumes consumed in the operating time (m 3 ), respectively, t day is the daily operating time (h•day −1 ), which depends on daylight hours and N is the number of consecutive operating days. ReagentCost is expressed in terms of € •day −1 . 2.6.2. Cost efficiency calculation Cost efficiency (mg• € -1 •m −2 ) is calculated for open-loop (OL) and closed-loop (CL) operation based on the numerical integration (A MP , Eq. (10)) of MP instantaneous removal rate (Removalinst) obtained in the simulations (mmole•min −1 ), which is calculated as shown in Eq. (9). Removalinst(t) = Qw(t)⋅[MPinlet −MPoutlet(t) ] (9) AMP =∫N⋅ 1440min day 0 Removalinst(t)dt (10) where MP inlet is the micropollutant concentration of the inlet water feed flow (7.09 •10 -6 mM) and MP outlet is the micropollutant concentration of the treated effluent, corresponding to [MP] as perfect mixing is assumed in the photo-reactor. A MP is used to estimate the MP removal capacity of the plant (RemovalMP) for each operating scenario in terms of mmole of microcontaminant removed per operating day (Eq. (11)), as shown in Fig. 3.a (open-loop) and Fig. 3.b (closed-loop). Note that Removal inst is variable throughout the operation for open-loop and constant for closed-loop operation, related to the control setpoint. These integrals are calculated only for the continuous flow operation of the process in order to estimate the cost efficiency of the plant just considering continous flow mode decontamination. RemovalMP =AMP N⋅tday 24 (11) Finally, cost efficiency (CE, mg• € -1 •m −2 ) is calculated in each case as shown in Eq. (12), using ReagentCost and RemovalMP, previously obtained in Eq. (8) and Eq. (11), respectively. CE =RemovalMP⋅Mmav Reagentcost⋅S (12) where S is the reactor surface area (m 2 ) and Mmav is the molar mass averaged for the three tracking MPs in the outlet effluent (mg•mmol −1 ). As MP degradation kinetics are very similar, the tracking MP ratio is considered to remain unchanged throughout the process. 3. Results and discussion 3.1. Linear model and control tuning Based on the modelling study performed (Subsection 2.5.1), the characteristic parameters for the first-order systems (averaged) are shown in Table 4. The linear models obtained present dynamic response speeds very similar to each other, as both time constants ( τ ) are very close to 4.5 min. It means that 4.5 min elapse until 63 % of the steady state value of the output variable ([H 2 O 2 ] and [MP]) is reached when a step-like change in the input variable of the system (QH2O2 and QFe, respectively) is carried out. Results are in concordance with reactant feed flows used in simulations, which are also very similar. Considering that 15 and 30 min of HRT were simulated, it can be concluded that the system response is affordable from a control point of view. Time delays (θ) are linked to characteristic operational factors of a real plant operation (pipeline transport and sample analysis devices). As there is no physical real plant operation in this study, θ values equal to zero were considered for both systems. Proportional gain (K) is positive for the first transfer function as QH2O2 and [H 2 O 2 ] are directly related to the dilution rate. The negative gain obtained for the second transfer function is related to the embedded kinetic mechanism since lower MP concentrations in the photo-reactor are achieved by higher Fe 2+ supply. The PI controllers were tuned using the Lambda method for firstorder systems (Subsection 2.5.2), considering a closed-loop time constant value (λ) equal to 0,9 times the open-loop time constant ( τ ), λ = 4.22, for the hydrogen peroxide controller, and 0,5 times for the micropollutant controller, with λ =2.27 (See Table 5). Notice that these values for the closed-loop tuning parameter (λ) were selected looking for speeding up the micropollutant removal and keeping a conservative hydrogen peroxide usage, based on typical values used in process control [20]. D. Rodríguez-García et al. Chemical Engineering Journal 455 (2023) 140760 6 3.2. Plant simulation example The present section discusses the results obtained by simulating a photo-Fenton RPR used for WTT in continuous flow mode under different operating and environmental conditions. All simulations were obtained for open-loop (OL) operation or manual mode (classical solution), as well as closed-loop (CL) operation or automatic mode, in order to compare the effect of the implementation of the control system in the decontamination performance and reagent consumption. Simulation time was fixed in all cases for three consecutive operating days and control set points were set to 0.35 mM of H 2 O 2 concentration in the photo-reactor and 2.1•10 -6 and 7.9•10 -7 mM of MP, which involves 70 % and 90 % removal, respectively. Fig. 4 shows a plant simulation example for a 70 % MP removal setpoint in July and the following operating conditions: 10 cm & 30 min (D & HRT). Note that the whole simulation study was performed for all the days shown in Table 1. Simulation starts at midnight (00:00). Initially, reagents (QH2O2 and QFe) and water feed (Q w ) flows remain off (Figs. 4.a and 4.e), since the plant is designed to operate during daylight. Thus, overnight, the photoreactor remains loaded with wastewater and reagents, producing batch decontamination due to the Fenton reaction. At night, MP concentration in the reactor (Fig. 4.f) decreases to zero. At sunrise, feed pumps are switched on (Figs. 4.a and 4.e) starting continuous flow Fig. 3. Numerical integrals (A MP ) used for cost efficiency calculation for the following simulation example for three consecutive days in July (N =3): 10 cm depth, 30 min HRT and 70 % MP removal. Open-loop (a) and closed-loop (b). Table 4 Characteristic first-order parameters for linear models obtained by computer simulation. Open loop system First-order system parameters (averaged) Gain, K (mM•m −3 •min −1 ) Time constant, τ (min) Time delay, θ (min) QH2O2 vs [H 2 O 2 ] 2.99 •10 4 4,69 0 QFe vs [MP] −3.66 •10 -2 4,54 0 Table 5 Tuning of micropollutant and hydrogen peroxide controllers. Controller Proportional gain, K p (m 3 •min −1 •mM −1 ) Integral time constant, T i (min) [H 2 O 2 ] 3.72 •10 -5 4,69 [MP] −3.66 •10 -2 4,54 D. Rodríguez-García et al. Chemical Engineering Journal 455 (2023) 140760 7 operation. MP concentration in the reactor increases to a maximum value due to the wastewater injection effect and hydrogen peroxide absence in the reactor, since it has been completely consumed during the night batch reaction. Fig. 4.f is a typical illustrative example that shows how the disturbance variability (Fig. 4.c) is transferred to the manipulated variable in order to achieve the control setpoint previously set up. For open-loop operation MP removal is variable over time as the radiation (process disturbance) is also variable throughout operation. Nevertheless, it remains unchanged for closed-loop operation based on the control setpoint (assuming perfect setpoint tracking). By implementing the control system (closed-loop), MP concentration in the reactor remains constant, according to decontamination control setpoint (70 % inlet MP removal). To reach the setpoint, the controller modifies iron feed flow rate (Fig. 4. e) depending on radiation conditions. Therefore, when solar radiation is low, the controller increases iron consumption to achieve target decontamination through the Fenton reaction. Throughout the day, the controller decreases iron feed flow, as the photo-Fenton effect on MP removal increases, achieving iron feed flow reductions of up to 83 % during maximum radiation hours with respect to early operation hours of the day (Fig. 4.e). As a result, iron concentration in the photo-reactor is also reduced by up to 81 % (Fig. 4.d). Regarding hydrogen peroxide concentration in the reactor (Fig. 4.b), it decreases in OL operation as solar radiation increases, reaching its minimum concentration in the middle of the day, when the effect of the photo-Fenton cycle (Fe 3+ photo-reduction to Fe 2+ ) is accelerated due to maximum radiation. Fig. 5 shows a one-day plant simulation (1 June 2021) for a cloudy operating day to show the disturbance rejection capabilities of the proposed control approach. For OL operation, QH2O2 and QFe are fixed at 0.871•10 -5 and 2.018•10 -5 m 3 •min −1 , respectively, so that total dissolved iron in the photo-reactor remains constant at a value of 0.1 mM, Fig. 4. Plant simulation for three consecutive July days (N =3) corresponding to 10 cm and 30 min (D and HRT, respectively) as operating conditions for open-loop and closed-loop operation (OL and CL, respectively). H 2 O 2 and Fe 2+ feed rates (a and e), H 2 O 2 and Fe concentration profiles in the reactor (b and d), MP profile (f) and global and diffuse UV solar radiation (c). D. Rodríguez-García et al. Chemical Engineering Journal 455 (2023) 140760 8 as performed above (see Fig. 4). At midday (12:00–15:00) the regular operation of the photo-reactor is strongly disturbed by an abrupt and intense cloud (Fig. 5.a). Consequently, the incident solar UV radiation is sharply reduced by 89 %, giving rise to an increase in MP concentration of up to 172 % for OL operation (Fig. 5.b). However, for CL operation the controller rapidly increases iron concentration in the reactor (Fig. 5.d) by modifying QFe (Fig. 5.c) in order to cancel the system disturbance by compensating the decontamination through the Fenton reaction. As a result, MP concentration remains almost unchanged based on the control setpoint (Fig. 5.b), proving the robustness and effectiveness of the control system for disturbance rejection. Note that this is a very important advantage to ensure autonomous operation for these types of processes. 3.3. Water treatment capacity Water treatment capacity is determined by the operating conditions: liquid depth and hydraulic residence time. Nevertheless, the treatment capacity will also vary for the same operating conditions throughout the different seasons of the year (Fig. 6) since the daily operating time of the plant depends on daylight hours. As a result, the highest water treatment capacity is obtained in July, being 3.6 % higher than in May and 40.7 % higher than in November and February. Values obtained are similar to previous empirical studies of continuous flow operation mode of the solar photo-Fenton process applied to urban wastewater regeneration using raceway technology [10]. The authors reported a water treatment capacity of 2.250 m 3 •m −2 •day −1 at 15 cm and 30 min of D and HRT, respectively, which is very close to the treatment capacity shown in this simulation study for 10 cm & 30 min. 3.4. Cost efficiency of micropollutant removal Cost efficiency of the plant (Figs. 7 and 8) is defined as the mass of micropollutants (mg) removed per monetary unit ( € ) and square meter of reactor surface area (m 2 ), in order to compare micropollutant removal performance between OL and CL operations in terms of reagent cost. Fig. 5. Plant simulation for a cloudy operating day (1 June 2021). Global and diffuse UV solar radiation (a), MP concentration profile (b), H 2 O 2 and Fe 2+ feed rates (c) and reagent concentration (d). Fig. 6. Water treatment capacity (WTC) of the plant for the different operating conditions (D and HRT) and seasons of the year. D. Rodríguez-García et al. Chemical Engineering Journal 455 (2023) 140760 9 Cost efficiency of the process was lower in CL than for OL operation when a 90 % MP removal setpoint was imposed (Fig. 7). However, by reducing it to 70 % it can be observed that the control system increases the cost efficiency of the process in all cases with respect to OL operation (Fig. 8). It can therefore be concluded that MP removal setpoint affects the process efficiency in automatic mode (CL), as reactant consumption imposed by the controller to reach decontamination target does not exhibit a linear correlation with MP removal. In this way, demanding MP removal setpoints are achieved through high reagent consumption, resulting in low operating cost efficiencies of the process. Note that as mentioned in Subsection 2.5.2, a trade-off between performance and control effort can be reached by tuning the closed-loop parameter λ. As such, this provides a degree of freedom to the plant designer for situations when the operating costs are too high. For OL operation, in all cases reagent costs ( € •day −1 ) do not vary as reactant flows remain constant throughout the operation (Fig. 4.a and 4. e). Nevertheless, as solar radiation increases, a higher MP removal is obtained, meaning the cost efficiency of the system increases. For CL operation, MP removal remains unchanged (control setpoint, Fig. 4.f), so the improvement in the system efficiency is linked to reagent cost reduction. The controller decreases Fe 2+ consumption as radiation increases due to the effect of the solar photo-Fenton process on the decontamination. For the highest treatment capacity, at 20 cm & 15 min (Fig. 6), the system shows the lowest operating cost efficiency. In these operating conditions the effect of the photo-Fenton cycle is limited due to the increase in liquid depth. Moreover, as HRT is minimal, the extent of the reactions involved is reduced. As a result, the cost efficiency of the process decreases both in OL (decrease in MP removal) and CL operation (increase in reagent consumption in order to achieve decontamination control setpoint). There is optimum cost efficiency at 10 cm & 30 min, as this operating condition maximizes the VRPA (photo-Fenton cycle) in water decontamination. Nevertheless, the water treatment capacity of the plant is minimal for the given operating conditions (Fig. 6). As expected, there is a strong correlation between water treatment capacity and process efficiency. Fig. 7. Cost efficiency for open-loop (OL) and closed-loop (CL) operation for a 90% MP removal setpoint. Fig. 8. Cost efficiency for open-loop (OL) and closed-loop (CL) operation for a 70% MP removal setpoint. D. Rodríguez-García et al.