Membrane Technologies For Boron Removal

 

Reverse osmosis (RO)
Reverse osmosis (RO) is widely recognized as one of the most im- portant technologies for fresh water production from saline water. Driven by an applied hydraulic pressure, the RO process preferentially allows water molecules to permeate through the membrane but retains the salt ions [24]. Under alkaline conditions (pH > 9) where boron exists mostly in the form of borate ions, B(OH)4−, RO membranes are completely capable of producing boron-free water due to size exclusion and Donnan repulsion of hydrated borate ions. However, RO is largely incapacitated under neutral or acidic conditions where the un-dis- sociated boron molecules are dominant [7]. This is ascribed to the smaller size of the poorly hydrated boron molecules and lack of elec- trostatic repulsion as aforementioned. Therefore, most of the state-of- the-art RO membranes fail to attain a high boron rejection [12]. In recent years, many studies are focusing on improving the boron re- moval ability of RO processes through membrane development, reg- ulation of operational parameters and system design.

RO membrane development
The membrane is the key element of a RO system. Conventional ROmembranes use cellulose-based polymers, while the polyamide thin film composite (TFC) membranes dominates the RO market in these days [24]. As shown in Fig. 4 [25], TFC RO membranes typically consist of three layers; namely, an ultrathin polyamide selective layer, a mi- croporous support layer and a polyester fabric substrate. In early stu- dies, commercial brackish water reverse osmosis (BWRO) membranes were reported to be ineffective to remove neutral boron with rejection ratios less than 65%, while seawater reverse osmosis (SWRO) opera- tions can achieve a fair rejection rate of 90% [7,11,12]. In recent years, new RO membranes were manufactured with special design for the treatment of boron-rich water. They have a tighter selective layer with reduced affinity to boron and increased affinity to water [26]. Table 2 presents examples of commercial RO membranes, with their specifica- tions provided by the manufacturers [27–30]. For example, Toray TM800K SWRO membranes could reach an impressive boron rejection rate as high as 96% at pH 8 under high pressure operations (800 psi)

 

In addition, novel RO membranes with higher boron rejections were tailored and evaluated in various laboratories. A novel polyamide

bilayer TFC membrane was fabricated by forming an additional hexa- fluoroalcohol-containing aromatic polyamide layer (HFAPA) on the surface of the conventional polyamide layer [31]. This membrane ex-hibited  a  50%  reduction  of  salt  passage  and  higher  boron rejection
without adding much resistance to water transport. More recently, a novel “charge-aggregate”  induced RO membrane was successfully devised by replacing the traditional m-phenylenediamine (MPD) with a new dipolar sulfonated diamine monomer 4,4′-(1,2-ethanediyldiimino) bis(benzenesulfonic acid) (EDBSA) during the interfacial polymeriza- tion [32]. Its boron rejection capability was demonstrated as high as 90.6% when desalinating a model seawater sample which comprised an overall ion strength of 37,000 ppm and 29 ppm boron under neutral conditions. The high rejection was due to the existence of several “charge-aggregate” induced cavities on the membrane surface.

 

 

Moreover, low pressure RO membranes were modified by the con- centration polarization-enhanced radical graft polymerization in a di- luted glycidyl methacrylate (GMA) solution [33]. With “defects” sealed in their selective layers, modified Dow-Filmtec LE membranes exhibited lower affinity to boron and hence, reduced the boron passage rate from
~40% to ~20% at pH 7. A moderate loss of permeability was also observed due to the denser membrane structure. Researchers also car- ried out an in-situ membrane modification of a commercial spiral wound element by a GMA solution at existing RO plants [34]. The modified element exhibited 2–4 times lower passage of salt (< 0.5%) and boron (~20%) but a similar permeability at a pH range of 6.5–7, which was superior to that of most BWRO membranes [34]. In addition, a controlled hypochlorite treatment was reported to improve the RO membrane performance in terms of water permeability, salt and boron rejection, by making the membrane surface more hydrophilic and ne- gatively charged [35]. After being treated in a 4000 ppm chlorine so- lution for 0.25 h, the BW30LE membranes demonstrated an improved boron rejection rate from ~75% to ~95% at pH 11.

RO operational parameters
Many operating parameters affect the effectiveness of boron rejec- tion via RO processes, such as temperature, pH, pressure, ionic strength, feed boron concentration and water recovery, fouling, etc. [10,36]. Among them, pH is a dominant factor. It was reported that the boron rejection of low pressure RO membranes dramatically increased from 30% to 83% at neutral pH to 90–99% at pH 10.5 [37,38]. The BWRO membrane (Dow FilmTecTM BW30) exhibited a boron rejection increase from ~50% at pH 7 to ~93% at pH 11 [39], and the membrane CPA2 from Hydranautics showed a nearly linear increase in boron rejection as a function of pH. [37]. Improving membrane rejections toward boron by pH adjustment has been widely investigated on other RO or NF membranes [37,39–45], such as BW30LE, ESPAB, ESPA2  [40], Woongjin Chemical RE4040-BE [46], Dow FilmTech™  BW30-2540  [43], etc. The dependence of boron rejection on pH can be attributed to boric acid dissociation and membrane surface charge at high pH [7]. As aforementioned, a higher pH may promote the formation of borate ions which have larger diameters and negative charges [10]. Meanwhile, most polyamide RO membranes are negatively charged and their charge densities increase at a high pH value, which favor the rejection of borate ions due to the electrostatic repulsion [10,38,47]. However, the elevated pH values also promote the scaling propensity and salt precipitation. Hasson, et al. defined the scaling boundary conditions of Pacific seawater aiming to control the boron content in desalinated seawater [48]. In the case of higher pH or water recovery, the potential of CaCO3 precipitation became higher with reduced induction periods and higher saturation indices for calcite and aragonite. Therefore, it is recommended that anti-scalant addition may be combined with optimal pH adjustment so that the single-pass RO process can produce boron- free water [48].

Operational pressure also influences the boron rejection. It has been demonstrated by many experimental studies that the boron rejection usually increases with increasing the transmembrane pressure [40,41,44,49–52]. It was found that SWRO membranes, i.e., Toray UTC-80AB and Dow FilmTec SW30HR, exhibited a boron rejection in- crease from 74% to 84% at 15.5 bar to 92–97% at 48.3 bar [41]. For the BWRO membrane (Dow FilmTec BW30), boron rejection increased from 53% to 75% with increasing pressure from 6.9 to 20.7 bar, while a further increase to 31 bar reduced the boron rejection, probably caused by the high degree of concentration polarization [41]. According to the solution-diffusion model, water flux increases as a proportional func- tion of the difference between the applied pressure and the osmotic pressure, while the boron solute flux does not [7,47]. The increase in applied pressure results in an increase in concentration polarization, and the increased boron concentration at the membrane surface leads to the increase of boron diffusion. In addition, it has been pointed out that boron transport via convective water flow should not be ignored [7,50], especially at low pH and high pressure conditions. Overall, in most cases, the increase of water flux is greater than that of boron flux at elevated applied pressures, leading to a higher boron rejection. On the other hand, the membrane property change under high pressure as well as the convective flow of boron should also be taken into consideration. An interesting observation is that pressure influences on boron flux significantly if the feed pH is low but becomes insignificant if the feed pH is high [50].

In addition, both high feed salinity [37,39,45,53] and elevated temperature [42,44,45,49] enhance boron passage through RO mem- branes, despite of the increased borate ions. For example, the boron rejection of the CPA2 RO membrane declined from ~80% to 45% when

 

 

the feed salinity increased from 0 to 15,000 mg L−1 at pH 9 because of charge neutralization of membrane surface potential at a high salinity [53]. However, in another study, an increase in boron rejection was observed by increasing the feed salinity from 0 to 5000 µs/cm, which was explained by the facilitated ionization of boron at the same pH due to the increased salinity [45]. Unlike the seawater (40,000 µs/cm), the relatively low salinity increase did not cause serious concentration polarization or significant negative impacts, allowing the increase in boron rejection [45]. Similarly, a higher feed temperature usually leads to a lower boron rejection. Experimentally, the boron rejection of the TFC 4040HR membrane exhibited  a  considerable  drop  from  ~40%  to < 10% with an increase in temperature from 10 to 35 °C [42]. The influence of temperature on boron rejection is a trade-off between the increase of boron permeability through the membrane and the decrease of pKa. Since the former is the dominant factor, boron rejection would generally decrease at elevated temperatures [7].
Lastly, membrane fouling and scaling have been reported to influ- ence the boron removal  efficiency  in  RO  processes  [54,55].  In  Tu,  et al.’s work, membrane fouling caused by sodium alginate, colloidal silica or CaSO4 led to a decrease in boron rejection [54]. The major mechanism was the cake-enhanced concentration polarization, which neutralized the membrane surface charge and inhibited the impact of pH change on surface charge properties. On the contrary, humic acid fouling benefited the boron rejection, mainly due to the enhanced electrostatic repulsion by the surface adsorption of highly negatively charged humic acid [54]. The effect of membrane fouling on boron removal was also simulated in full-scale seawater RO processes [55]. It was found that boron rejection could be improved by controlling the fouling, replacing the SWRO element by boron-specific adsorbents or increasing the pH in the second pass [55].

System design
Owing to the difficulties to meet the regulations of boron content from various water sources by means of a single stage RO process, several multi-stage systems have been designed [56,57], such as two- stage RO systems [45,58,59], integrated NF-RO systems [60], RO-ad- sorption systems [12,61] as well as polyol-assisted RO systems [62–64], etc. As shown in Fig. 5 [10,65], two-stage RO systems were developed by combining the SWRO and BWRO stages. To prevent scale formation, a higher pH was usually employed in the second BWRO stage [65]. Bick, et al. utilized an analytic hierarchy process (AHP) model to evaluate the optimal boron removal methods [66]. According to the AHP model and Hasse diagram analysis, two-stage RO were preferred and could produce four permeate streams with different boron con- centrations by using permeate split, while in series ion-exchange treatment of low quality permeates can improve the performance of a single stage RO. Similarly, a model based mixed integer nonlinear programming (MINLP) optimization framework was developed for the evaluation of boron rejection in a two-stage RO system [59]. To achieve a target level of boron rejection (i.e., 90%), the authors investigated the effects of pH and temperature on system design. They found that the optimal RO system for different seawater pH and temperature condi- tions varies from one-stage to two-stage SWRO depending on the BWRO addition. In Farhatet, et al.’s work, new generation commercial RO membranes were employed in a two-stage RO system with the SWRO- BWRO configuration [58]. Without pH adjustment, the two-stage RO system could achieve a significant boron rejection as high as 96% using real seawater. Alternatively, novel boron selective resins (BSRs) were also developed to adsorb boron from RO permeate. By functionalizing monodisperse porous poly(VBC-co-DVB) beads with N-methyl-D-glu- camine (NMDG) groups, those BSRs were able to remove 97% of boron from the RO permeate of geothermal water with the aid of functiona- lized beads at a dosage of 4 g L−1 [61].

Moreover, various polyol compounds have been integrated in var- ious NF and RO processes [60,62–64]. Polyol compounds carries mul- tiple hydroxyl groups, which can chelate with the boric acid or borate ions as aforementioned, resulting in large anionic complexes. It has been reported that the addition of polyols including glycerol, mannitol and sorbitol to the feed could significantly improve the boron rejection of NF and RO membranes [60]. The complexation capability of polyols with boron followed the order of sorbitol > mannitol > glycerol, and the complexation reaction was significantly affected by the solution pH. With a boron: polyol molar ratio of 1:1, NF90 membranes could achieve a boron rejection of more than 90% at pH 10 [60]. In Liu, et al.’s work, sorbitol was chosen as the chelating additive in a dual-stage NF desa- lination process [64]. By adjusting chelation conditions and pH values, the boron rejection of dual-stage NF was improved and the boron concentration  in the NF  permeate  was reduced  to  0.1 mg L−1. NMDG and sodium D-gluconate were also applied to enhance the boron re- jection of RO membranes [62], which helped to reduce boron content to less than 1 mg L−1 at a high water recovery rate of 90%. In addition, the effects of polyol addition were investigated in seawater condition. It was found that the polyol-boron chelation reaction occurred at low boron and polyol concentrations without significant interference by other ions [63]. The reaction was so fast that it may be applicable to full-scale SWRO processes. However, the recycle or disposal of polyols in the retentate was rarely addressed by the literature, so additional treatment of the RO or NF retentate may be necessary in order to re- generate the polyols.

Forward osmosis (FO)
As an emerging membrane technology, forward osmosis (FO) has gained increasing interest in recent years. Compared with pressure- driven membrane processes, FO is driven by the osmotic pressure dif- ference across the semi-permeable membrane (Fig. 6), with the net movement of water from a diluted feed solution to a concentrated draw solution [67,68]. FO processes could be highly attractive in water treatment because of their low fouling tendency and high rejections to a wide range of contaminants if the regeneration of draw solutions is not

required or can be achieved by cost-effective methods [69]. FO mem- branes made from both cellulose-based and thin film composite (TFC) materials have been investigated for boron removal  [70–76]  and  Table 3 gives a performance comparison. At neutral pH, the permeation of un-dissociated boric acid molecules through the selective layers of FO membranes could be explained by the solution-diffusion mechanism while the convection-diffusion mechanism was employed to explain the boron transport in the membrane substrates. Thus, it was reported that boron rejection was susceptible to the membrane orientation, water flux, boron permeability of the selective layer, structure parameter [70–72] as well as the draw solute type [72,73].
According to the modeling of boric acid permeation through FO membranes, boron rejection is dependent on FO water flux [70]. As water flux increased from 1 to 8 µm s−1 (or 3.6–28.8 LMH), boron re- jection under the AL-FS mode (i.e., the active layer faces the feed so- lution or forward osmosis (FO) mode) improved from ~20% to ~60%, while boron rejection under the AL-DS mode (i.e., the active layer faces the draw solution or pressure retarded osmosis (PRO) mode) increased slightly to ~15% and then declined to below 10% due to the con- centration polarization of boron within the membrane substrate [70]. Moreover, the effects of draw solution pH, draw solute type, cross-flow velocity, and feed solution pH were examined in the FO process [72]. Key mechanisms governing boron transports are reverse salt diffusion and internal concentration polarization within the FO membrane. Ac- cording to the experimental results, the draw solutes with a small hy- drated radius had higher reverse diffusion, which hindered the boron passage and facilitated the boron rejection [73]. However, a boron rejection similar to BWRO (≤ 65%) was achieved in most laboratory FO operations at neutral pH. If the feed boron solution was adjusted to pH > 10, the boron solute flux would decrease substantially, due to the relatively larger hydrated radius of charged borate anions and the electrostatic repulsion by the membrane. Thus, boron rejection higher than 90% could be achieved with boron feed solutions at pH 11[73,75]. To enhance the boron rejection, a double-skinned FO mem- brane was specially developed by fabricating two polyamide TFC layers on top and bottom surfaces of the sulfonated PPSU substrate. It dis- played superior boron separation with a rejection as high as 84% at neutral pH [76]. Overall, the reported FO membranes showed pro- mising potential for boron separation from aqueous systems. However, additional treatment steps and energy input are needed in order to concentrate draw solutions and generate pure water. Also, the boron permeation through FO membranes may become significant if the feed water recovery rate is greatly increased. Nevertheless, the novel dual- layer design and modification of FO membranes for boron separation are helpful to hinder the boron passage and improve the separation performance.

Polymer-enhanced ultrafiltration (PEUF)
The PEUF process, or named as liquid-phase polymer-based reten- tion (LPR) process, has been put forward for the removal of pollutant ions from aqueous solutions for decades [77]. Fig. 7 shows the sche- matic of the PEUF process which combines the polymer–pollutant ion complexation and membrane filtration. In a typical PEUF process, the target species can be solely concentrated and removed by using a spe- cific chelating polymer; namely, chelatogen, based on their complexa- tion chemistries. To date, this process has been extensively applied to separate many heavy metal ions including cadmium, copper, mercury, lead and zinc, etc [78–80].

Boron removal via the PEMF process has been assessed based on the strong interactions between boron and poly-hydroxylated organics as aforementioned. Many researchers focused on the development of highly efficient chelating polymers in the past 10 years. In order to form a stable complex, the chelating polymer should bear multi-hydroxyl groups which must orient properly to match the structural parameters of the tetrahedrally coordinated boron [10,56]. A good example is the vic diol which consists of two hydroxyl groups occupying vicinal cis positions. In addition, proton acceptor groups on the inert polymer backbone are able to enhance the chelation reaction by neutralizing the proton generated during the complexation process (e.g. tertiary amine group [56]).

Various water soluble polymers have been reported as poly- chelatogens in the literature. Commercially available polymers, e.g. polyvinyl alcohol [81] and poly(ethylene imine) [82], were evaluated and low boron rejections were achieved due to either the slow chelating kinetics or the lack of functional groups. Later on, polymers grafted with highly efficient functional groups were synthesized to attain a high boron-chelating capacity. N-methylglucamine (NMG) is a well-known functional group and has provided reactive sites in boron-adsorption resins [83,84]. Some researchers grafted NMG and its derivatives onto water soluble polymers for use in PEUF processes and observed im- provements in boron rejection [85–88]. Recently, Yilmaz and co- workers [89] synthesized two novel boron-chelating polymers; namely, hydroxyethylamino glycerol-grafted poly(glycidyl methacrylate) (PNS) and poly(vinyl-ethanediol-co-vinyl alcohol) (COP). The maximum boron rejection was about 55% when using PNS at a polymer loading of 1000 g g−1  and  a  pH  of  9.0.  The  performance  was  not  impressive

probably due to the negative effects of the carboxylic ester group[86,89] or the lack of proton acceptors [89]. To overcome it, the team further designed a group of new polymers; namely, poly(vinyl aminoN,N-bis-propane diol) [90] and its derivative copolymer [91], which showed high boron rejections of 92% and 96%, respectively. Most recently, Tang et al. devised two hyper-branched polymers, hyperbranched polyglycidol (HPG) and 2, 3-Dihydroxypropyl-hyperbranched polyethylenimine (Diol-HPEI) [92]. Due to the high density of diols, a high rejection of 66–91% was achieved with a Diol-HPEI polymer loading of 100 g per g boron in the pH range of 6.9–9.0. In addition, compared with linear polymers, hyperbranched polymers have lower solution viscosities [93] and hence, lower membrane fouling tendency in the PEUF process owing to their globular characteristics and low inter-molecular entanglements. The authors also verified the importance of the proton acceptor groups by comparing the protonation of the tertiary amine group and the ether group, which provides a good guide to design new chelating polymers.

 

An ultrafiltration (UF) membrane is used in the PEUF process due to the fact that UF is a well-established process with advantages of low energy consumption and high productivity. The most important parameters of the UF membrane are the pore size and anti-fouling properties. The pore size should be adapted to the size of the chelating polymer to minimize the loss of chelatogens and avert the pore clogging by chelatogens. Fouling is another important concern. The addition of polymers in the boron-containing solution may create serious concentration polarization and membrane fouling. Fouling can be affected by chelating polymer loading, membrane materials, surface charge, pore size, etc. Tang et al. employed a hydrophilic polyethersulfone(PES) UF membrane and observed that water flux showed a remarkable decreasing trend as a function of the chelating polymer loading, signifying an increasingly significant fouling tendency [92]. Water flux was also found to be affected by solution pH. The authors suggested that polymer fouling on membrane surface may be the main cause of these findings. The electrostatic interaction between chelating polymers and the membrane surface exerted a critical influence on water flux.

 

Similar observations on fouling could be found in other literatures[79,89,94]. In contrast, Yilmaz and co-workers reported that the permeate flux was almost independent of polymer concentration for
both cellulose and PES membranes, which was achieved by a proper control of operating conditions [86,90,91].

 

In summary, the PEMF process is a capable approach for efficient boron removal. Table 4 benchmarks the performance of PEUF for boron removal. However, the industrial scale up of PEUF is highly in- capacitated by two factors. First of all, the regeneration of chelating polymers may not be easy because of the partial loss of chelatogens and process complexity, although it can theoretically be accomplished by a simple acidity adjustment. No report has unveiled the regeneration process so far. Secondly, flaws either in low kinetics or low boron re- jection can be found in most developed polymers. The design of new polymers with a high chelation capacity is one of the most important imperatives in order to make PEUF a workable approach for boron separation. Thirdly, the morphology, charge, and chemical properties of the membrane also affect the performance and operational cycles. Membranes with good chemical resistance, mechanical strength and anti-fouling properties are always preferred for the PEUF process.

Adsorption-membrane filtration (AMF)
Adsorption-membrane filtration (AMF) is a hybrid process which involves the use of both boron selective resins (BSRs) and microfiltration (MF) membranes in the following scheme: i) boron absorption by
BSRs suspended in a feed solution; ii) separation of the saturated BSRs from the treated feed solution by MF; iii) desorption of boron from the saturated BSRs using mineral acids; iv) separation of fresh BSRs by MF;v) regeneration of BSRs using a mineral base; vi) recovery and recirculation of the regenerated BSRs back into the feed solution [95–97].

 

Fig. 8 plots the flow chart of a typical AMF process. AMF is a cheaper alternative to the conventional fixed-bed ion-exchange column for the removal of boron from aqueous solutions [97]. This is because the boron adsorption rate of BSRs increase as their size decreases. However, it is unrealistic to consider the use of fine BSRs in a conventional fixed bed column because the resulting pressure drop can be significant [10]. According to the Ergun's equation [98], which is widely used to determine the pressure drop across a packed bed for all flow conditions, the pressure drop is inversely proportional to the particle size.

 

Where ΔP is the pressure drop, H is the height of the packed bed, µ is the dynamic viscosity of the fluid, U is the superficial velocity, x is the spherical equivalent diameter, ɛ is the bed voidage and ρf is the density of the fluid. In other words, reducing the BSR size in the conventional fixed bed column will significantly increase the pressure drop and the energy requirement of the system. On the other hand, for AMF, the pressure drop is mainly a function of the membrane resistance and is independent of the BSR size [97]. Hence, it is highly plausible to reduce the BSR size to improve the process performance without increasing the pressure drop [10]. As a result, the required quantity of BSRs can be reduced, which leads to lower both BSR and operational costs [10,97]. In addition to the needs for high operating capacity and fast kinetics, there are some other requirements for BSRs when they are employed in the AMF configuration. They should have a small diameter, uniform size and smooth edge and be resistant to abrasion and cracking[56]. It was revealed that the BSRs particle size largely affected the

economics of the AMF configuration [56]. Unfortunately, the current commercially available BSRs are designed for the conventional opera- tions in the packed bed columns. Hence, their particle size usually ranges from 300 to 500 µm which are too large for the AMF config- uration [11]. As such, the development of monodispersive BSRs with diameters ranging from several to 50 µm is a challenge and has received great attention [11].

To date, attention has been mainly given to the development of the resins, while other aspects of the AMF process may have been ne- glected. However, in order to optimize the entire AMF process, a de- tailed cost analysis should be conducted as well. Additional efforts should be given to the membrane separation unit in order to optimize the AMF process. Firstly, the membranes used in the AMF configuration should be as hydrophilic as possible because a significant drop in permeate flux would be otherwise observed due to the strong foulant- membrane adhesion [99]. The increased hydrophilicity would not only reduce the water transport resistance across the membranes and miti- gate any reduction in permeate flux but also reduce the fouling ten- dency of the membranes. Secondly, the MF membranes should have high acid and alkali resistance. Currently, most commercially available polymeric MF membranes cannot withstand harsh acidic or alkaline conditions for a prolonged period of time [100]. For example, Kabay   et al. used a Telfone Fluropore MF membrane [95], which usually has a pH operating range of 1–11 [101], for the separation of BSR from aqueous solutions. Güler et al. used polypropylene membranes which have a pH operating range of 2–14 [102]. In contrast, Darwish em- ployed polyvinylidene fluoride (PVDF) membranes which have a pH operating range of 2–10.5 for the MF separation of the resin suspension [103]. The relatively narrow pH operating range of MF membranes prevents the use of either concentrated acids during the desorption process or concentrated bases during the regeneration step, and this results in a greater volume of waste. Onderkova believed that the use of ceramic membranes might be more advantageous in this case as ceramic membranes have high chemical stability [104]. However, ceramic membranes are much more expensive than polymeric mem- branes. They are also very brittle and difficult to handle. Thirdly, par- ticular attention should be given to the submerged hollow fiber and tubular membranes [10] because they potentially requires less oper- ating energy compared to the cross flow UF/MF system (i.e., external loop system) [105].

Membrane adsorptive filtration (MAF)
To overcome some of the limitations of AMF and PEUF in removing boron, the use of UF/MF polymeric membranes grafted with chelating groups has been proposed. The resultant membranes are also known as adsorptive membranes. Similar to AMF and PEUF, The membrane ad- sorptive filtration (MAF) involves both the sorption and membrane filtration processes. However, it helps to simplify the AMF and PEMF processes by combing the sorption and filtration steps into a single operation [106]. The water soluble polymer-induced fouling in the PEMF process can be completely averted by the MAF process. In ad- dition, compared with the regeneration of chelating polymers in the PEUF process, the regeneration of adsorptive membranes is more vi- able. Thus, it may be a promising alternative to replace both AMF and PEUF for boron removal.

There are generally 4 approaches to prepare the adsorptive mem- branes. The first method is post-functionalization which involves (1) membrane fabrication and (2) functionalization of the membrane sur- face. Wei et al. adopted this technique by firstly grafting poly(glydicyl methacrylate) on the surface of a commercially available regenerated cellulose membrane via atom transfer radical polymerization (ATRP) [107]. Subsequently, N-methylglucamine was introduced onto the membrane surface via epoxy ring-opening reaction. The resultant membrane had a maximum adsorption capacity of 0.75 mmol g−1 after 7 days of immersion in a 500 ppm boron solution at a pH of 7 and 20 °C. This performance was comparable to those of commercial BSRs. In a similar manner, Meng et al. fabricated adsorptive membranes by suc- cessively grafting hyperbranched polyethyleneimine (HPEI) and gly- cidol onto the surfaces of PA200 polyacrylonitrile (PAN) membranes produced by Sepro Membrane Inc. [108]. The modified membrane adsorbed  3.2 mmol g−1  of  boron  within  4 mins  of  immersion  in  a 100 ppm boron solution at a pH of 8 and 30 °C, which is much better than other reported sorbents. The superior performance was attributed to the use of HPEI that optimized the glycidol distribution. This fabri- cation method is preferred because of (1) the ability to optimize the grafting yield and (2) the ease of forming a functional layer on the membrane surface [108].

It is preferably to have a functional layer on membrane surface because the diffusion path of boron to the ligand can be significantly reduced and thus drastically increase the adsorption rate. A high ad- sorption rate is of utmost importance since the feed passes through the membrane rapidly [109]. However, this technique also suffers from some drawbacks because harsh grafting conditions would often (1) limit the membrane materials and functional groups that can be used and (2) result in membrane structural change. For example, the polar aprotic solvents employed during the grafting reactions may deform or even dissolve most polymeric membranes. Hence, it may be necessary to use solvent resistant membranes as the base membranes. In addition, the high temperature used to speed up the grafting reactions may densify the membranes, leading to a reduced permeability.

The second approach is pre-functionalization which involves the functionalization of the polymer followed by the membrane fabrication. Shi et al. prepared their adsorptive membranes from the blends of a polysulfone (PSF) polymer and a hydrophilic PSF-based glycopolymer [110]. The glycopolymer was synthesized by (1) chloromethylation of polysulfone, (2) ATRP reaction of glycidyl methacrylate and (3) grafting of N-methyl-D-glucamine onto the polymer chains. Different weight ratios of PSF to glycopolymer were employed to fabricate the adsorptive membranes. Their best membrane had an adsorption capa- city of 0.193 mmol g−1 within 30 min of contact with a 300 ppm boron solution at a pH of 9 and 30 °C. This approach is better than the first method because it enables to use a variety of membrane materials and functional groups since the harsh grafting reaction is carried out on the polymers instead of the resultant membranes. Additionally, this method results in a bulk-functionalized membrane, which may lead to a higher adsorption capacity. However, as the amount of hydrophilic 

glycopolymer increases, both the porosity and average pore size of the membranes increase. This may cause membranes with reduced me- chanical strength and selectivity [110]. Thus, there is an amount limit of functionalization polymers to be employed during the membrane fabrication.

The third method is a combination of the first two methods. Pre- functionalization of the polymer is carried out to graft an intermediate link on the polymer chains and then post-functionalization is performed to decorate the membrane with boron-specific groups. Meng et al.
[106] and Du et al. [109] firstly carried out the chloromethylation of polysulfone and cast membranes from the resultant chloromethylated polysulfone (CMPSF). Meng et al. subsequently grafted poly(2-gluco- namidoethyl methacrylate) via ATRP for complete functionalization of the membrane surface [106]. The resultant adsorptive membrane could adsorb  more  than  2.0 mmol g−1  of  boron  after  soaking  in  a  300 ppm boron solution for 2 h at a pH of 9 and 30 °C. Du et al. on the other hand, grafted a variety of multi-hydric methacrylate monomers which included 2,3-dihyrdoxypropyl methacrylate, 3-(N-glucidol-N-methyl) amino-2-hydroxypropyl methacrylate and 2-bis(2,3-dihydroxypropyl) amino ethyl methacrylate on the CMPSF membrane via ATRP [109]. The resultant membranes had an adsorption capacity ranging from 0.20 to 0.46 mmol g−1 at a similar condition. This approach is advantageous because it is possible to split up the functionalization process into 2 separate operations and protect the fabricated membrane from harsh grafting conditions.

The fourth approach is the in-situ polymerization of a pre-polymer within another polymer matrix to attain an interpenetrating polymer network (IPN). The functional polymer is synthesized within a plain polymer matrix which helps to provide better processability and higher mechanical strength. Palencia et al. adopted this method by conducting the in situ polymerization of boron selective monomers (which are based on N-methyl-D-glucamine and vinylbenzyl chloride) inside the pores of cellulose ultrafiltration membranes [111,112]. Although this technique allowed the combination of chemical and physical properties of a number of polymers into the same material, it significantly reduced the permeability (> 50%) of the resultant membranes [111]. It is worth noting that unlike aforementioned works which focused on the ad- sorption capacity of the membranes, The researchers also evaluated the filtration performance of their membranes [112]. Feed solutions con- taining 20.0 mg L−1 of boron at different pH values (of 5.0, 7.0 and 9.0) were used. To ensure a low flow rate and a long contact time, gravity was the only driving force during their filtration experiments. Despite that, relatively low boron retention values (≤20%) were obtained and this highlighted the limitations of the adsorptive membranes under the actual pressure-driven filtration process. Under the actual pressure- driven filtration process, high pressures or flow rates are generally employed to increase throughput of the filtration process. However, this can often lead to incomplete adsorption of boron due to the short contact time. Thus, adsorptive membranes should not only have a high adsorption capacity but also a high adsorption rate.

In general, the use of adsorptive membranes is a relatively new approach in the removal of boron and it is still not massively studied. There are a few areas worthy of further investigations. Firstly, most of the current research focuses on improving the adsorption capacity and uptake rate. However, there are still limited studies evaluating the performance of the adsorptive membranes under the pressure-driven filtration process. In other words, it is necessary to establish a re- lationship between the adsorption rate/capacity and the membrane rejection. The effects of membrane adsorption capacity on regeneration frequency should be determined as well. Secondly, particular attention should be paid to develop hollow fiber adsorptive membranes. This is because among the various configurations, the hollow fiber configura- tion has a greater surface area to volume ratio which may translate to a greater boron adsorption capacity per unit volume. Thirdly, similar to AMF, high acid and alkali resistance of the adsorptive membranes are crucial for stable MAF operations. Hence, the exploration of novel

membrane materials is of great significance.

Electrodialysis (ED) and Donnan Dialysis (DD)
To date, ion-exchange membranes (IEMs) in electrodialysis (ED) and Donnan dialysis (DD) systems have been advanced with some competitive edges for boron removal. IEMs are made of synthetic ion exchangers immobilized on membrane materials, including cation-ex- change membranes (CEMs), anion-exchange membranes (AEMs), and bipolar membranes (BPMs) [56,113,114]. In an ED system as shown in Fig. 9 [115], CEMs and AEMs are positioned alternately and exposed to an external electric field, which drives cations and anions to migrate from the diluate compartment into the concentrate compartment. At neutral pH, the boron flux strongly depends on (1) membrane type, (2) initial boron concentration in the diluate, (3) ion types moving across membranes as well as (4) electric current density [116]. Dydo in- vestigated different IEMs in pairs: AMX–CMX (Neosepta, Japan), AMV–CMV (Selemion, Japan), PC–SA and PC–SK (PCA, Germany) and AM(H)–CM(H) (Ralex, Czech Republic). PC-SA AEMs had the highest boron  flux  of  ~720 µg m−2 s−1  with  an  initial  diluate  boron  con- centration of 0.1 M and an electric current density of 100 A m−2 [116]. Since un-dissociated boric acid molecules dominated in the boron so- lution under neutral conditions, the main mechanism of boron transport across AEMs and CEMs was diffusion driven by the boron concentration gradient, coupled with a convective drag from ion flux [117]. Experi- mentally, most boron molecules were found to transport through AEMs rather than CEMs, and were greatly influenced by the anion and cation types [117,118]. Goli, et al. examined the effects of salts and acids on boron diffusion through CEMs, they found that boron flux decreased with  the  cation  type  of  K+ > Na+ > Ca2+ > Mg2+  [118].  Kijański et al. studied the effectiveness of boron transport from the neutral dil- uate (pH 6) into the alkaline concentrate (pH > 11) in ED, with the optimizations of concentrate boron content, pH and voltage drop [119]. They estimated the cost of boron removal from 75 to 0.8 mg L−1 with simultaneous  boron  concentration  up  to  5000 mg L−1  to  be  about
$0.22 m−3.

When the pH of boron-containing water increases to above 9, a significant increase in boron transport in ED could be observed in the range of 80–420 µg m−2 s−1 (up to 3500 µg m−2 s−1) [113,114]. As a negatively charged species, borate ions are preferentially transported across AEMs rather than CEMs. Nagasawa et al. employed both bipolar membrane electrodialysis (BMED) and BPMs to remove boron [115]. As shown in Fig. 9(b), within BPMs, water dissociates into H- and OH- ions under the external electric field. Then the solution close to the anion- exchange part of BPM becomes alkaline, leading to the formation of borate ions. Borate ions then migrate across AEMs to the concentrate solution side, and react with protons generated by the water-splitting reaction in BPMs to form boric acid. Consequently, boron is removed from the wastewater feed and concentrated in the concentrate com- partment. Promisingly, over 90% of boron could be removed from a 100 mg L−1 solution over a wide range of initial pH from 2.3 to 12.0, and the electric current efficiency approached 24.5% in the absence of NaCl [115]. However, the ED process for boron removal is relatively expensive because of limited electric current efficiency for boron transport. Improvements on energy efficiency and the use of cheaper membranes are suggested to decrease the ED capital cost [12].

Alternatively, Donnan dialysis (DD) processes are employed to se- parate boron by utilizing an AEM without an external electric field. When borate ions transport from the feed solution into the receiving solution, an equivalent amount of “driver” anions needs to be trans- ported from the receiving solution to the feed [113,114]. AEMs were evaluated on BSRs regeneration by DD, which revealed that the sodium chloride salt could drive borate transport more effectively than sodium sulfate [120]. Similar results were also reported by [121,122]. More- over, Kir et al. observed that the boron flux increased proportionally with the feed boric acid concentration in a range of 0.001–0.1 mol L−1. In addition, an electron cyclotron resonance (ECR) plasma treatment of the AFX AEM membrane could significantly improve the boron flux to
104.5 µg m−2 s−1  (under  0.01 M  H3BO3  feed)  in a  DD  process  due  to the change of pore morphologies [121]. However, the presence of Cl− or SO 2− ions in the feed dramatically reduced the boron flux by 57–78%. Bryjak and Duraj grafted polyethyleneimine onto porous Celgard membranes via dielectric barrier discharge (DBD) plasma to prepare well-permeable membranes for a DD process [122]. Compared with commercial FumaTech AEMs, the modified membranes were more efficient in boron removal by reducing 60–70% of boron over 6 h when using Na2SO4 as the receiving phase. However, the grafted membranes were not selective for anion transport due to incomplete pore coverage, while the salt passed through the membrane several orders of magni- tude faster than boron. In addition, DD processes are time-consuming and can only be used for boron removal from diluted solutions.

Overall, the economic and technical applicability of ED or DD for boron removal may require further discussion and demonstration. Although borate fluxes across IEMs at elevated pH are higher than those of boric acid, a higher pH indicates a higher risk of scaling with calcium or magnesium. In addition, the removal efficiency is still low since OH- or other anions under the alkaline conditions migrate faster than borate ions. As a result, borate converses to neutral boric acid molecules in the diluate [114].

Membrane distillation (MD)
MD is an emerging alternative method for boron removal. It is a thermally driven separation process which involves transport of water vapor across a hydrophobic membrane [123–127]. This technology possesses several unique advantages: (1) larger contact area and lower operation temperature compared to the conventional distillation; (2) lower operating pressure than pressure driven membrane processes such as RO; (3) 100% theoretical rejections to non-volatile solutes; (4) less sensitive to high feed concentrations.

The superior performance in boron removal via the direct contact membrane distillation (DCMD) process (Fig. 10) has been demonstrated in limited literatures [128–132]. By applying a polyvinylidene fluoride (PVDF) hollow fiber membrane in the DCMD process, Hou et al. re- vealed a high boron retention rate about 99.8% at a wide range of boron concentrations in the feeds. A high boron rejection was also achieved by desalinating a boron-containing real seawater solution [129]. Similarly results were reported elsewhere [130,131]. In another work, Wen et al. employed a commercial polypropylene (PP) mem- brane in DCMD to concentrate highly saline radioactive wastewater containing cobalt, strontium, caesium and boron [132]. A boron re- jection above 99.9% was maintained even at a high boron concentra- tion of 5000 ppm or feed salt concentration of 300 g L−1. However, the boron rejection was affected by membrane scaling due to the co-pre- cipitation of B and CaSO4 on the membrane surface. Owing to its high B rejection, MD was also employed in hybrid systems to produce boron- free water from seawater or wastewater. Macedonio and Drioli had proposed a RO-MD integrated system to remove boron as shown in Fig. 5 [133].

Despite of its advantages in boron removal, MD is still considered to be an energy-intensive process. To develop an economically feasible MD approach, one must use (1) high flux MD membranes and (2) solar energy or waste heat to power the processes in order to lower the en- ergy consumption. The scaling on MD membranes particular for boron removal is another important issue. Serious membrane scaling was observed in literatures [129,130,132]. Decreases in permeation flux, liquid entry pressure (LEP) and boron rejection were reported due to pore clogging, membrane wetting and co-precipitation of boron with other chemicals, respectively. Low-cost cleaning methods are of great importance in various MD processes to reduce scaling and maintain the solute retention capacity. For example, polyacrylic acid was dosed in the feed solution as an anti-scalant to delay the deposit formation [129]. On the other hand, scaling can be minimized by devising low- scaling membranes which is more viable to reduce operating costs. Efforts may also be given to either tuning the membrane morphology or developing new membrane materials.