Performance evaluation and boron rejection in a SWRO system under variable operating conditions

It is well known that reverse osmosis (RO) is the leading desalination technology. As an energy intensive technology, the exploitation of renewable energy sources (RES) to power RO systems is a attractive option. A strategy to take advantage of all the available energy of an off-grid renewable system is to work with the RO system under variable operating conditions. This implies additional challenges in terms of water production and permeate quality, among others. Boron rejection is one of the main concerns in seawater RO (SWRO) systems. The aim of this work was to evaluate the performance and boron rejection of a single-stage SWRO system with 7 membrane elements per pressure vessel under variable operating conditions. The initial permeability coefficients of two SWRO membranes (TM820L-440 and TM820S-400) were calculated from experimental data of a full-scale SWRO desalination plant. These coefficients and the characteristics of the membranes were introduced in a simulation algorithm to estimate the behavior of the SWRO system. The results show that, compared with the TM820S-400 membrane, the TM820L-440 performed better in terms of boron rejection in the form of boric acid, but worse in terms of water production. When RES-powered SWRO systems are designed to work under variable operating conditions, consideration needs to be given to the safe operation window in terms of boron concentration in the permeate and to variation of the permeability coefficient of the membranes. © 2021 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ )


Introduction
Boron is an important nutrient, especially for plant growth . The margin between a deficient and a toxic concentration of boron is very small and the regulations in this respect are therefore usually quite restrictive ( Ruiz-García et al., 2019 ). Usually, boron concentration in seawater is around 5 ppm and is in the form of boric acid (H 3 BO 3 ) ( Hilal et al., 2011 ). This is an uncharged weak acid and the separation of species by reverse osmosis (RO) membranes depends mainly on their charge ( Qasim et al., 2019 ). As a result, boron rejection is a major concern in seawater reverse osmosis (SWRO) desalination plants ( Cengeloglu et al., 2008;Koseoglu et al., 2008b ). Considerable effort s have been made to make new RO membranes that increase boron rejection ( Li et al., 2020;Jung et al., 2020 ). Another well-developed research line is related to proposals for alternative processes to RO that separate boron from aqueous * Corresponding author.
The challenge to reduce boron concentration in the permeate produced by SWRO desalination plants has been studied extensively by the scientific community ( Farhat et al., 2013 ). This is an   ( Glueckstern and Priel, 2003 ). Boron rejection in SWRO systems depends, among other things, on feedwater pH, boron concentration, salinity and temperature ( T f ) and operating parameters such as pressure ( p), flux recovery ( R ) and membrane characteristics ( Hyung and Kim, 2006;Koseoglu et al., 2008b ). Membrane manufacturers provide information about boron rejection under test conditions which do not reflect the real operating conditions of full-scale desalination plants. Evaluation of the boron rejection of commercial SWRO elements under different conditions is of fundamental importance. Koseoglu et al. (2008a) studied the boron re-jection of two commercial high rejection SWRO membranes using a lab-scale flat-sheet configuration. They studied the influence of pH, p and salinity on boron rejection. Increasing the pH can have a positive effect on boron rejection but can also cause scaling problems in RO systems ( Hasson et al., 2011 ). The difficulty that SWRO membranes have to reject boron has caused, in some cases, redesigns of RO systems such as the incorporation of a second pass, etc. Gao et al. (2011) . Another additional factor that may affect boron rejection in SWRO systems is membrane fouling. This can be appreciated in the study published by Ruiz-García et al. (2019) in which two commercial SWRO membranes were evaluated in a fullscale desalination plant. Park et al. (2012) evaluated the decrease of boron removal and the reduction in the water production rate by membrane fouling. The study was based on simulations and a predictive model that estimates boron removal in SWRO desalination processes was used.
The intensive use of energy in SWRO desalination plants has promoted the use of renewable energy sources (RES) to provide the power for this technology ( Nassrullah et al., 2020 ). The application of RES to power SWRO is not simple, and its viability depends on many factors such as water and energy accessibility ( Nassrullah et al., 2020 ), costs ( Elmaadawy et al., 2020;Rezk et al., 2020 ), regulations ( Sen and Ganguly, 2017 ), etc. There are two main factors in the operation of an SWRO desalination plant, permeate production and permeate quality in terms of total dissolved solids (TDS). Both factors are affected when an off-grid RES is powering the SWRO system under variable conditions. de la Nuez Pestana et al. (2004) studied the variable operation of an SWRO system directly connected to a wind turbine without any energy storage system. The membrane installed in this SWRO system was the TFC 2822-SS from Koch Fluid Systems TM , and actual seawater (as opposed to synthesized) was used. The feed pressures ( p f ) applied were 39, 49 and 60 bar, with flux recoveries of 19.74, 31.37 and 40% respectively. Permeate conductivity was between 429 and 292 μS cm −1 , with lower values at pressures of 49 and 60 bar. Ntavou et al. (2016) carried out a performance analysis of a multi-skid SWRO unit under variable power input. They used a FILMTEC TM SW30-4040 membrane and a synthesized feed solution by adding salt to tap water to reach 37,500 ppm. It was observed that the lower the TDS in the permeate ( T DS p ) the higher the power input that was required, with T DS p ranging between approximately 150 and 250 ppm. Dimitriou et al. (2017) validated a theoretical model for predicting SWRO systems under variable operating conditions. As in the previous study, the membrane used was the FILMTEC TM SW30-4040 in a small-scale SWRO unit with one pressure vessel (PV) and a Clark pump unit. p f ranged between 35 and 45 bar approximately and T DS p between 200 and 600 ppm. They also obtained lower T DS p with higher p f . The same SWRO system was used by C.-S. Karavas et al. (2018) , incorporating a short-term energy storage system. In this case, they operated the SWRO system in a p f range of between 39 and 51 bar, with a permeate conductivity of between 200 and 10 0 0 μS cm −1 . As is usual in these desalination systems, the higher the power input was, the higher the pressure and R and the lower the T DS p were. Calise et al. (2019) carried out an economic assessment study of SWRO desalination powered by photovoltaic panels. The performance analysis of the SWRO system was based on a simulation using the Water Application Value Engine (WAVE) software from Dupont® and a model proposed by the authors which showed similar results. The performance was assessed in terms of R and salt rejection. An off-grid solar energy system to power an SWRO desalination plant with integrated photovoltaic thermal cooling was proposed by Monjezi et al. (2020) . The modeling and operation of the SWRO system was simulated with the Reverse Osmosis System Analysis (ROSA) software. The authors studied the Filmtec TM SW30-2540 membrane, working with a single stage with an R of 40%. An inorganic feedwater was used as input, but boron concentration was non-existent and the concentration of each ion in the permeate was estimated by the software. Delgado-Torres et al. (2020) undertook a preliminary study of an SWRO process powered by a hybrid system (photovoltaic -tidal range). The performance of the SWRO system was simulated with ROSA, considering two SWRO membranes: FILMTEC TM SW30HRLE -440i and FILMTEC TM SW30XLE -440i. The inorganic composition of the feedwater was not detailed and only the T DS p of one operating point for each membrane was shown. Unfortunately, the issue of boron rejection in SWRO systems working under variable operating conditions has not been extensively studied. The permitted permeate boron concentrations according to different regulations are low. While the World Health Organization (WHO) established a maximum boron concentration of 2.4 ppm for drinking water in 2011, the EU limit is just 1.0 ppm. This limiting factor can reduce the safe operation windows (SOWs) of SWRO systems working under variable input power as supplied by RES.
The aim and novel contribution of this paper is the evaluation of the performance and the boron rejection under variable operating conditions of two commercial SWRO membranes (TM820L-440 and TM820S-400 from Toray) in full-scale PVs. To carry out this work, a simulation algorithm previously validated and published by the authors Ruiz-García and de la Nuez-Pestana (2018) was adapted to estimate the behavior of a single-stage SWRO system with the aforementioned membranes installed. The permeability coefficients of RO membranes are characteristic to each membrane and show membrane efficiency in terms of water production and solute rejection. The initial average water, salt and boron permeability coefficients that were used were taken from a previous published work ( Ruiz-García et al., 2019 ).

Permeability coefficients
Determination of the initial water permeability coefficient ( A ), salt permeability coefficient ( B s ) and B B was carried out using the experimental data of the initial operating point of a fullscale SWRO desalination plant with TM820L-440 and TM820S-400 membranes installed. A more detailed description of this desalination plant can be found in a previous work published by one of the authors Ruiz-García et al. (2019) . More specifically, the initial data of trains 2 (TM820L-440) and 9 (TM820S-400) were used. These trains were selected as their initial operating time was closer to 0 than the other trains and so their membranes were the least used. The number of PVs was 90 and 79 for trains 2 and 9, respectively. Each PV had 7 SWRO membrane elements in series ( Fig. 1 ). As the operating data were usually collected at the input and output of the PVs, this allowed calculation of the average permeability coefficients. The detailed procedure for calculating the permeability coefficient can be found in a previous study by Ruiz-García et al. (2019) . Conductivity was measured using a Hanna® Instruments EC 215 conductivity meter, and boron concentration in the permeate was determined using the carmine method. The pH was close to 7, so B B was calculated as boric acid. The permeability coefficients per SWRO membrane module were estimated from the initial operating data. Table 1 show the dimensional characteristics and performance under manufacturer test conditions ( p f = 5.52 MPa, feed-brine temperature ( T fb ) = 25 • C, C f = 32,0 0 0 ppm NaCl, R = 8%, pH p = 8 and 5 ppm boron added to feedwater) of the SWRO membranes and Table 2 shows the initial operating values per PV and the permeability coefficients for each membrane. For these calculation the Equations of Table 3 were used.

Process modeling
The simulation algorithm ( Ruiz-García and de la Nuez-Pestana, 2018 ) used the solution-diffusion ( Qasim et al., 2019;Al-Obaidi et al., 2017 ) transport model. This is the most commonly used model for simulating ( Joseph and Damodaran, 2019 ) and predicting RO system performances ( Alsarayreh et al., 2020a;Al-Obaidi et al., 2019 ) as usually provides results close to the real behavior of these systems, despite its limitations ( Alsarayreh et al., 2020b ). The transport equations are applied for each membrane element considering averages. The temperature T and pressure drop in the permeate along the RO system were disregarded. More details about the simulation algorithm can be found in a previous work ( Ruiz-García and de la Nuez-Pestana, 2018 ). A fouling factor ( F F ) of 1 was considered (new membrane) along with a T fb of 25 • C, and so the temperature correction factor ( T CF ) was 1. S m is the active membrane surface, NDP the net driving pressure, L the membrane length, v fb the feed-brine velocity, ε the porosity in the feed-brine channel (considered 0.89 for both membranes), h the feed-brine spacer height (28 milli-inches = 7 . 11 × 10 −4 m), P F the polarization factor, k s the solute mass transfer coefficient and η the dynamic viscosity of water (0.0 0 0891 kg m −1 s −1 ). To determine all the above variables, the aforementioned algorithm was imple- 3 . 15 × 10 −7 5 . 65 × 10 −7 Table 3 Transport equations.
Permeate flow Seawater density Osmotic pressure Shearwood number Boron mass transfer coefficient ( Taniguchi et al., 2001 ) k Rejection mented in MATLAB®. The results are presented per PV of 7 membrane elements each (as in the actual SWRO system). The simulations were carried out with the following operating ranges per PV: Q f from 5 to 16 m 3 h −1 since for Q f between 3 and 5 m 3 h −1 the operating window was small and with low R, C f from 32 to 45 g L −1 and p f from 4 to 7 MPa. Boron concentration in the feedwater was considered as 5 ppm in all cases in accordance with the analysis of the feedwater carried out in the desalination plant ( Ruiz-García et al., 2019 ). Only NaCl was increased. The specific energy consumption ( SEC) was determined considering the ideal performance of the high pressure pump (100% efficiency for the electrical engine and pump). No energy recovery devices were considered in the simulations.

Results and discussion
The results shown in Figs. 2-9 were obtained considering C f = 35,0 0 0 ppm. Figs. 2 and 3 show the R of the TM820L-440 and TM820S-400 membranes, respectively. It can be observed that higher R was achieved with the TM820S-440 membrane for the same operating points. The influence of coefficient A was higher than the active area of the membranes in terms of R . High R values can bring some membrane elements close to thresholds imposed by manufacturers including, for example, R higher than 15% per element or brine flow rates ( Q b ) very close to 3 m h −3 . Such operating conditions can decrease notably the system performance over long periods of operation and mean that more frequent chemical cleanings are required, etc. The highest R values obtained were 57.56 and 58.45% for the TM820L-440 (7 MPa, 7.25 m 3 h −1 ) and TM820S-400 (7 MPa, 8.5 m 3 h −1 ) membranes, respectively.
Figs. 4 and 5 show the SEC values of the TM820L-440 and TM820S-400 membranes, respectively. As usual, the membrane with higher R had the lower SEC. The TM820L-440 membrane had a larger SOW, mainly due to the A coefficient. This membrane had lower production, and so there were more operating points where restrictions such as minimum Q b or maximum R per membrane element were not reached. For this reason, depending on the operating conditions, it is not advisable to have many high production membrane elements in series. Minimum values of SEC were 3.12       and 3.27 kWh m −3 for membranes TM820S-400 and TM820L-440 respectively (considering ideal performance). Usually these operating points are very close to the thresholds imposed by manufacturers and it is convenient to take safety margins in real operation. The lowest SEC values obtained were 3.2683 and 3.1180 kWh m −3 for the TM820L-440 (5.85 MPa, 6 m 3 h −1 ) and TM820S-400 (5.55 MPa, 6 m 3 h −1 ) membranes, respectively.
For both membranes, the higher Q f and p f were the lower C p was ( Figs. 6 and 7 ). The higher Q p and the lower CF were, the lower C p was. (Eq. (17)). High values of Q p required high values of Q f and p f . The mentioned two figures show the role of coefficient B in solute rejection. The difference is very small as the coefficients are quite close ( 1 . 55 × 10 −8 vs 1 . 75 × 10 −8 ). It should be considered that fouling could have a different im pact on both membranes making the coefficient B vary differently. However, the difference between the two membranes in terms of B B was more pronounced. This can be observed in Figs. 8 and 9 , which show  boron concentration in the permeate. The TM820L-440 membrane was able to maintain lower C pB in a wider operating range, making it more suitable than the TM820S-400 in terms of boron rejection as boric acid, although the energy requirements were higher. For this reason, two types of SWRO membranes (one with high boron rejection and another with high production) are installed in some SWRO desalination plants in order to balance the SEC and C pB ( Ruiz-García et al., 2019 ). If the SWRO system is working under variable operation this task becomes much more complex, this is why to have a SOW is crucial. Fig. 10 shows the SEC vs C pB for the two membranes. Considering a limiting factor of 1 ppm for C pB , it can be seen that the TM820L-440 membrane had a wider operating window that allowed an appropriate C pB . While there were various operating points that could allow the same C pB , the actual operating point of the SWRO system will depend on the high pressure pump installed. Fig. 11 shows the displacement of the SOW of the TM820L-440 membrane considering two different C f (35 and 42,0 0 0 ppm). It can be seen how the SOW for the higher C f was wider. This was due to higher osmotic pressure resulting in the SWRO membrane element produce less permeate under the same operating conditions and lower C f . This decrease in production meant the SWRO membrane did not exceed the safety operation margin for most operating points, although many operating points showed an unacceptable C p and high SEC. It should be noted that all the results were obtained considering constant permeability coefficients.
In actual operation, fouling can cause the permeability coefficients to change ( Ruiz-García and Nuez, 2020 ), in which case all the possible operating points would be displaced with respect to the calculated ones. A decrease in coefficient A can produce decreases in R and/or increases in SEC if a constant R is desired by increasing p f . A change in coefficient B B could be delicate as it may result in the need for changes in the operating conditions to ensure the quality criteria are met, as well as, in some cases, premature membrane replacement.

Conclusions
An evaluation was undertaken in this work of boron rejection (in the form of boric acid) of two commercial membranes (TM820L-440 and TM820S-400) under variable operating conditions. The results show that the TM820L-440 membrane is the more suitable option in SWRO systems where the required C pB is low ( < 1 ppm) and a priority. However, as the TM820L-440 requires more energy than the TM820S-400, the latter is more appropriate when C pB is not a priority or when the limiting concentration is not very low, such as for agricultural purposes (non boron-sensitive crops). It should be noted that this was a static study (constant permeability coefficient) and that in SWRO systems the permeability coefficient is usually not constant, mainly due to the effect of fouling on the system. This can produce important variations in performance and may require changes to the operating conditions. Works on the sizing and techno-economic assessment of RES-powered SWRO systems should take into account not only the SOW, considering new membranes or constant permeability coefficients, but also the change of these coefficients and its impact on the entire plant. The decrease in coefficient A could result in lower permeate production or the need to oversize the RES system to provide the energy required to ensure constant longterm permeate production. Future works should consider different permeability coefficients under variable operating conditions to enable the determination of the most suitable membranes in terms of performance in certain operating ranges.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.