Separation and Purification Technology 108 (2013) 103–110

Contents lists available at SciVerse ScienceDirect

Separation and Purification Technology

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

Struvite recovery from municipal-wastewater sludge centrifuge supernatantusing seawater NF concentrate as a cheap Mg(II) source

Ori Lahav a,⇑, Marina Telzhensky a, Annette Zewuhn b, Youri Gendel a, Joachim Gerth b, Wolfgang Calmano b,Liat Birnhack a

a Faculty of Civil and Environmental Eng., Technion, Israel Institute of Technology, 32000 Technion City, Haifa, Israelb Environmental Technology and Energy Economics, Technical University of Hamburg, Eissendorfer Str. 40, Hamburg, Germany

a r t i c l e i n f o

Article history:Received 4 November 2012Received in revised form 31 January 2013Accepted 1 February 2013Available online 11 February 2013

Keywords:Struvite precipitationNanofiltrationMunicipal wastewaterPhosphate removalPHREEQC simulation

1383-5866/$ - see front matter � 2013 Elsevier B.V. Ahttp://dx.doi.org/10.1016/j.seppur.2013.02.002

⇑ Corresponding author. Tel.: +972 48292191; fax:E-mail address: agori@tx.technion.ac.il (O. Lahav).

a b s t r a c t

The paper describes the application of an alternative method for generating struvite (MgNH4PO4) solidsfrom wastewater, using a low-cost Mg(II) solution, previously separated from seawater by nanofiltration(NF). Mg2+ ions are present at high concentrations in the oceans (�1400 mg/L) and are well rejected bynanofiltration membranes, therefore, NF enables obtaining a concentrate that is rich in Mg(II) at a lowcost. Since the largest cost component in conventional methods for struvite precipitation from wastewa-ters stems from the cost of the Mg chemicals, the reduction in the overall struvite generation cost in thepresented method is considerable. However, the NF separation method includes inherent potential dis-advantages: along with Mg(II), other ions are also separated from the seawater (although with lowerrejection values) e.g. Cl(-I) and Na(I), which contribute to the salinity of the wastewater effluents, andCa(II) that may promote precipitation of unwanted calcium-phosphate solids, rendering the obtainedstruvite product less homogeneous thus less valuable. Consequently, the work focused on determiningthe best operational alternatives to attain the purest struvite product possible, using a fluidized bed(FB) reactor.

The results show the approach to be highly feasible. A theoretical simulation was first carried out todetermine the best operational conditions (pH, Mg(II) dosage through NF brine) to attain 90% P removalfrom the actual supernatant of a domestic-sludge dewatering facility. Total ammonia and orthophos-phate supernatant concentrations were adjusted to simulate high (�300 mg P/L and 600 mg N/L) andlow (100 mg P/L and 450 mg N/L) nutrient supernatant concentrations. The FB reactor was fed with amixture of supernatant and NF brine solutions, while pH was maintained constant via NaOH addition,and the hydraulic retention time was tested at 20, 30 and 60 min. In the six scenarios tested, P removalwas higher than 90% and the struvite purity obtained was very high (�95%), as demonstrated by both dis-solution experiments and XRD analyses. The struvite precipitation experiments were carried out in thepresence of an antiscalant agent, which was added to the NF brine to prevent the clogging of the NF mem-brane which was operated at 90% recovery to attain a concentrated Mg(II) brine.

� 2013 Elsevier B.V. All rights reserved.

1. Introduction seawater [10], bittern solution [13], MgCO3 – magnesite [14] and

Separation and recovery of struvite (MgNH4PO4) from superna-tant of sludge dewatering facilities has been extensively addressedin the last two decades (e.g. [1–10]) and is already implemented atfull-scale in more than a few wastewater treatment plants [11], allover the world. The required dosage of Mg2+ salts (MgCl2, MgSO4,Mg(OH)2, etc.) is undoubtedly the most significant operational ex-penses (OPEX) component associated with this approach, esti-mated to contribute up to 75% of the overall production costs [12].

To address this issue a few alternative magnesium sources havebeen suggested in the context of struvite precipitation, including

ll rights reserved.

+972 48228898.

MgO [15]. In a recent publication, Telzhensky et al. [16] suggestedto use the reject of seawater nanofiltration (NF) as a cheap Mg2+

source, showing the cost of Mg(II) to be much lower than corre-sponding magnesium chemicals costs. However, [16] showed re-sults related only to the NF separation step, while the struviteprecipitation step was only referred-to theoretically.

Potential implementation of NF brines as Mg(II) source for stru-vite precipitation, besides being advantageous cost-wise, raisessome interesting difficulties, which have to be addressed both the-oretically and empirically. As opposed to the dosage of a solutionmade up of a pure magnesium salt, the addition of an NF brinehas the potential to result in five unfavorable side effects: (1) theMg(II) concentration is much lower than that attained with pureMg salts, resulting in a certain blending of the wastewater stream

104 O. Lahav et al. / Separation and Purification Technology 108 (2013) 103–110

and consequent reduction in both the TAN (total ammonia nitro-gen, i.e. the sum of NHþ4

� �and [NH3]) and PT (total inorganic ortho-

phosphate) concentrations at the entrance to the precipitationreactor, thereby reducing the respective precipitation potential(PP) of struvite; (2) since nanofiltration-based separation of Mg(II)from seawater results in a solution which contains, on top ofMg(II), also relatively high concentrations of sulfates, chloridesand sodium ions (as compared with a pure chemical), as well asother ions present at lower concentrations, the precipitation po-tential of struvite may be significantly reduced by the formationof ions pairs (e.g. MgSO0

4, NaPO2�4 , etc.), resulting in a need to main-

tain higher pH values to attain a given P removal efficiency; (3) thedosage of Cl� and Na+ to the wastewater, along with Mg(II), invari-ably raises the EC and SAR (Sodium Adsorption Ratio) values of thewastewater, which may be problematic if the treated wastewateris used for irrigation; and (4) the significant Ca2+ concentrationpresent in seawater NF brines may enhance the precipitation ofCa–PO4 solids (e.g. Ca3(PO4)2, Ca5(PO4)3(OH), etc.) thereby reducingthe purity (and value) of the struvite solid product. Moreover, (5)[16] showed that it was possible to operate the NF system at 64%recovery for attaining a brine solution comprising of 3500 mg/Lof Mg(II), without the need to use antiscalants (apart from seawa-ter acidification by HCl to prevent CaCO3 scaling). If a higher Mg(II)concentration is desired, for example in order to reduce dilution ef-fects (see comment (1) above) or when the wastewater treatmentplant (WWTP) is located away from the shore and transport costsshould be reduced, an antiscalant should be added to the concen-trate during the NF operation, to prevent both CaCO3 and CaSO4

precipitation and chemical fouling of the membrane. The sameantiscalant is thereafter dosed to the struvite precipitation reactorwith the NF brine. This raises a question regarding the effect of theantiscalant presence on both the kinetics of struvite (and other sol-ids having a positive PP value, e.g. calcium phosphate) precipita-tion and the composition and morphology of the formedprecipitant. The present work was designed to address these ques-tions, by operating a struvite precipitation fluidized bed (FB) reac-tor, which was fed with a mixture of centrifuge supernatant from adomestic WWTP and with seawater NF brine obtained at 90%recovery operation and containing a commercial antiscalant(Genesys CAS, manufactured by Genesys International, http://www.genesysamericas.com) at a concentration of �30 mg/L.

Table 1 shows the concentration range of TAN, Mg(II) and PT, asrecorded in the context of various struvite recovery endeavors. Theresults in Table 1 were used in order to decide on the conditionswith which the suggested method was tested in the current work.The anaerobic digester centrifuge supernatant used in the workwas collected from an operative WWTP (Hadera, Israel), and theTAN and PT concentrations were adjusted to simulate two TANand PT concentration scenarios: (1) PT = 310 mg P/L andTAN = 600 mg N/L and (2) PT and NT concentrations of �100 and

Table 1A few anaerobic digestion (and sludge dewatering) liquor compositions reported incontext of struvite precipitation.

WW treatment plant PT (mg P/L) TAN (mg N/L) Mg(II) (mg/L) Reference

Slough, UK 167 615 44 [2]Nine Springs, USA 204 624 4.7 [29]Tobu, Japan 207 756 7 [30]Ballarat South, Australia 293 775 16.2 [3]Nansemond, USA 300–700 820 NRa [11]Cape flats, South Africa 177 �1000 65 [31]Not disclosed, Italy 139 914 24 [1]Penticton, Canada 37–71 197–436 11–35 [6]Not disclosed, Japan 198–290 441–602 5–19 [4]

a Not reported.

�450 mg/L as P and N, respectively. The two scenarios representthe composition range in the majority of municipal anaerobic reac-tors that can be considered appropriate candidates for struviterecovery, where Scenario #1 is often linked with the operation ofenhanced biological P removal (EBPR) in the WWTP or with plantswhich receive an industrial stream containing significant ortho-phosphate concentrations (e.g. the Nancemod treatment plant inVirginia US and the Ballard South EBPR plant in Australia). Obvi-ously, lower P concentrations in EBPR plants have also been re-ported, e.g. the Tobu WWTP in Japan in which the reportedconcentration was 207 mg P/L however with TAN concentrationof 775 mg N/L or the Slough EBPR plant in the UK in which thePT concentration was merely 80 mg P/L (TAN = 750 mg N/L). Never-theless, the two Scenarios tested in this work were chosen to rep-resent WWTP sludge dewatering supernatants from which struviterecovery can be potentially cost effective. It is noted, though, thatstruvite recovery has been applied successfully at the pilot scalealso with lower PT concentrations (PT = 30–70, TAN = 250–450 mg/L), e.g. in the advanced WWTP in the City of Penticton,British Columbia, Canada (Table 1).

In the two scenarios tested in this work the FB reactor was oper-ated with a high struvite solids concentration (TSS > 20 g/L) in or-der to promote heterogeneous struvite precipitation throughoutthe experiments’ duration. In order to decide upon the requiredMg:P ratio in the influent to the precipitation reactor (which dic-tated the ratio between the flow rates of the NF brine and thewastewater) and on the constant pH to be maintained during theoperation, a theoretical simulation was first performed using theaquatic chemistry equilibrium-based program PHREEQC, run withthe Pitzer activity computation approach. The reactor was thereaf-ter operated according to the results of the PHREQC simulation andempirical results were collected both from a mass balance on theaqueous phase (effluent vs. influent characterization) and fromchemical and crystallographic characterization of the solid phasescollected from the reactor at steady state operation.

2. Materials and methods

2.1. Nanofiltration experiments

A cross-flow module accommodating a 2.4’’ diameter, 40’’ longcommercial NF membrane element (DS-5 DL, manufactured byOsmonics; average pore radii between 0.45 and 0.47 nm effectivemembrane area 2.6 m2 [17,18] was used within a lab-scale recircu-lated test cell. The initial solution consisted 50 L of UF-pretreatedMediterranean Seawater. Three repetitive experiments were per-formed with recycled feed flow rate of 75 L/(h m2), 17-bar constanttrans-membrane pressure and water temperature at 23 ± 2 �C. Thereject stream was returned to the feed tank to resemble a systemcomprising a few membrane elements operated in series. The per-meate stream was removed. Solute rejections, r, were defined asr(%) = (1 � CP/CF) � 100%, where CP and CF are the solute concen-trations in permeate and in feed streams, respectively. The concen-trations of ions in the permeate were either directly measured orcalculated indirectly by substituting the measured recovery rateand the ion concentrations measured in both the accumulatedbrine and the initial feed water into mass balance equations. Therecovery rate, c, was defined as c = (VP/VF) � 100%, where VP andVF are permeate volume and initial feed volume, respectively.

2.2. Struvite recovery experiments

Fig. 1 depicts the laboratory fluidized-bed struvite precipitation(and settling) reactor used in the study. The reactor comprised twomajor units: (1) a fluidized bed component (transparent PVC pipe,

O. Lahav et al. / Separation and Purification Technology 108 (2013) 103–110 105

172 cm long, internal diameter 5.3 cm, overall volume 3.8 L), and(2) a settling zone (PVC pipe, 53 cm long, internal diameter18.9 cm, effective volume 8.8 L). To prevent accumulation of precip-itates in the settler, the two units were connected via a funnel madeof PVC (18.9 and 5.3 outer diameters, height 14 cm). Municipal-wastewater sludge centrifuge supernatant (from the HaderaWWTP, located in Northern Israel) was supplied into the reactorat desired flow rates via a peristaltic pump (1–100 rpm, Cole-Parm-er). Nanofiltration brine and KH2PO4 solution (required for adjust-ing the PT concentration in the influent) was dosed veryaccurately into the fluidized bed unit by two specifically pro-grammed automatic titrators (702 SM Titrino, Metrohm). The pHvalue in the fluidized bed unit was maintained constant atpH 7.57 and 8.13 (±0.05) in the two scenarios, respectively, throughautomatic addition of 1 N NaOH solution performed by an AlphapH 200 pH controller (Thermo Scientific) and a dosing pump(Gala100, Prominent). Recirculation of the solution through the flu-idized bed reactor (0.7 L/min, representing an upflow velocity of19 m/h) was done by a peristaltic pump (6–600 rpm, 36’’ neoprenetubing, Cole-Parmer). Since the recirculation flow rate was at least3.7 times higher than the influent flow rate (at HRT of 20 min) thereactor operated at completely mixed conditions (CSTR).

2.2.1. Struvite recovery experimental procedureSeven struvite recovery experiments were performed. The dura-

tion of the experiments varied between 18 and 36 h to ensure stea-dy state conditions in the reactor, particularly with respect to thesolid phase (aqueous phase steady-state was attained after �1 h).To ensure that the solids sampled in the reactor at the end of a gi-ven experiment represented the conditions tested, solids werewithdrawn from the reactor just before the beginning of eachexperiment (leaving only a minimum mass to promote heteroge-neous precipitation) to ensure that the solid phase present in thereactor was almost entirely (>95%) made-up of solids formed under

≈

Sludge dewatering supernatant

Effluent

Se�ler

Fluidized bed

NF brine

KH2PO4

solu�on

Sampling valve

Recircula�on pump

NaOH

pH controller

Flow meter

Fig. 1. Experimental fluidized bed reactor used for precipitating struvite fromsludge centrifuge supernatant, using nanofiltration brine as Mg(II) source.

the experimental conditions tested in that particular experiment.The first three experiments were aimed at assessing the effect ofthe hydraulic retention time (HRT) on the precipitation perfor-mance assuming conditions that resemble PT and TAN concentra-tions typically encountered in centrifuge supernatants in WWTPthat apply EBPR (i.e. PT � 300 mg P/L and TAN � 600 mg/L) andNF-brine Mg(II) addition of 10 mM (i.e. influent ratio Mg:P ofapproximately 1–1). The fourth experiment in this experimentalset examined the results obtained with the same wastewater char-acteristics however, with a higher Mg(II) dosage (15 mM). The lastthree experiments were carried out with similar operational condi-tions, however with lower phosphate and TAN concentrations (PT -� 100 mg P/L and TAN 450–470 mg/L), simulating the morerepresentative case, in which EBPR is not performed within theWWTP. In these experiments the Mg(II)-NF brine dosage targetwas 3.22 mM. Examined HRTs for both wastewater compositionswere 60, 30 and 20 min. In all experiments the TAN and inorganicphosphorous concentrations in the treated wastewater were ad-justed to the required concentrations, respectively, by an additionof NH4Cl into the supernatant solution (when required) and bycontrolled addition of KH2PO4 solution into the fluidized bed reac-tor unit, respectively. The concentration of Mg(II) in the NF brineinjected to the reactor was �7 g/L. Within each experiment, sam-ples were withdrawn from both the fluidized bed unit and the set-tling unit, filtered through a 0.22 lm syringe filter and analyzed forMg(II), Ca(II), TAN, PT, and alkalinity concentrations. Sampling fromthe aqueous phase was initiated only after three hydraulic reten-tions times have past to ensure steady state with respect to solublespecies. Additionally, precipitant samples were collected at the endof all the experiments, washed copiously with distilled water, dis-solved in concentrated HCl and analyzed for Mg(II), Ca(II), TAN andPT. Solid samples were also analyzed by XRD to determine crystalphases.

2.3. Analytical methods

TAN was determined by the method proposed by [19]. Ortho-phosphate was measured by the vanadomolybdophosphoric acidcolorimetric method [20]. Ca(II) and Mg(II) concentrations weredetermined by ICP (1CAP6300 Duo, Thermo Scientific). XRD wasperformed with a Siemens D500 diffractometer, using Cu Ka radi-ation at 30 mA and 40 kV with step-scanning in the range of 5–65� 2h at a rate of 1� 2h/min (step width 0.05� 2h for 3 s). XRD spec-tra were evaluated using the Bruker EVA software (version 2005).

3. Results and discussion

3.1. Separating Mg(II) from seawater using a nanofiltration membrane

The performance of the DS-DL5 NF membrane was quantified atthe high recovery range (c = 50–90%) by applying 17-bar externalpressure. The concentrations of the main seawater ions measuredin the brine (as a function of c) are shown in Table 2. It can be seenthat the Mg2+ concentration approached 8000 mg/L at c = 90%,indicating almost a 6-fold increase in the concentration of thision relative to Mediterranean seawater. The Mg(II) concentrationat c = 90% translated into a relatively low (�3%) dilution effect(of NT and PT) when the NF concentrate attained at this recoverywas used for dosing 10 mmol/L of Mg2+ into the reactor (Scenario#1). Table 3 shows the calculated rejection value of the main sea-water ions, as a function of c. As expected, as the recovery valueand the ion concentrations increased, the rejection values of allthe ions decreased. However, the rejection values of Cl(-I), K(I)and Na(I) decreased more steeply than the decrease in the rejectionof Ca(II) and Mg(II): e.g. the rejection of Na(I) (decreased from

106 O. Lahav et al. / Separation and Purification Technology 108 (2013) 103–110

12.4% at a = 50% to 8.3% at a = 90%, i.e. a reduction of �33%, whileboth Ca(II) and Mg(II) rejections decreased by only 9.2%. Similar re-sults were reported in [16,21]. As a result, the concentration ratiosof Cl(-I) to Mg(II) and Na(I) to Mg(II) in the brine became signifi-cantly lower at the higher recoveries. For example, the molar ratiobetween the concentrations of Na(I) and Mg(II) in the brinedropped from 5.26 (at c = 50%) to 1.65 (at c = 90%) mol Na/mol Mgand the molar ratio between the concentration of Cl(-I) and Mg(II)in the brine dropped from 6.46 (for c = 50%) to 2.37 (for c = 90%)mol Cl/mol Mg. Since the aim of the nanofiltration separationwas to produce a brine that was both rich in Mg(II) and also withas low as possible Na(I):Mg(II) and Cl(-I):Mg(II) concentration ra-tios (to minimize additional concentrations of these ions to thewastewater), a clear incentive existed for operating the processat the highest possible recovery rate, despite the need to add anantiscalant to the feed solution.

Switching back to struvite recovery, it is noted that since thebrine attained in the 90% recovery NF experiment was used, a tar-get of 10 mM mg/L in the wastewater resulted in increasing theCa(II), Cl(-1) and Na(I) concentrations in the wastewater streamby approximately 45, 844 and 371 mg Ca/L, mg Cl/L and mg Na/L,respectively. These appear to be high additional concentrations(particularly with respect to Cl� and Na+) however the reader is re-minded that the supernatant line in WWTPs typically constitutes�1% of the overall flow-rate to the plant, thus the additional Cl�

and Na+ concentrations to the final WWTP effluents (when the PT

concentration is high, i.e. 310 mg P/L) are in fact merely �8.4 and�3.7 mg/L, respectively. Moreover, under such conditions, the dif-ference between the additional Cl� concentration, relative to thetypical case in which MgCl2 is dosed, was merely 2.1 mg/L. Theadditional Ca(II) concentration may, theoretically, become prob-lematic for attaining a high-purity struvite product (because ofpossible competition for PO3�

4 ions), and the Cl(-I), SO4(-II) andNa(I), besides adding to the concentrations of these ions in thewastewater effluents, also act to decrease the PP of struvite inthe precipitation reactor due to both increased ionic strength andion-pairing with either Mg(II), phosphate species or NH3. Effectsassociated with these phenomena were quantified in the theoreti-cal assessment presented in Section 3.2.

3.2. Theoretical assessment of struvite precipitation from centrifugesupernatants using NF brine as the Mg(II) source

Calculating the theoretical precipitation potential of a solid in acomplex solution requires solving simultaneously dozens of equa-tions, thus cannot be done manually. In this work the computerpackage PHREEQC was applied for this matter, using the Pitzer cal-culation method for accounting for activity and ion pairing effects[22]. The results obtained from the PHREEQC simulation were usedto determine the operational conditions (i.e. the constant pH andthe additional Mg(II) concentration) to be applied in the experi-mental examination of the process. A constant 25 �C temperaturewas assumed in the simulation, representing the approximate tem-perature of the supernatant shortly after flowing out of the meso-

Table 2Ion concentrations in the seawater feed (bottom row) and in the brine as a function of th

Recovery ratio (%) Cl� (g/l) Na+ (g/l) Ca2+

50 22.4 ± 0.4 11.8 ± 0.1 0.6260 23.3 ± 0.3 11.9 ± 0.1 0.7070 24.3 ± 0.5 12.1 ± 0.2 0.8675 24.8 ± 0.6 12.1 ± 0.1 0.9080 25.4 ± 0.4 12.4 ± 0.3 1.0485 26.2 ± 0.9 12.5 ± 0.12 1.2290 27.7 ± 0.24 12.5 ± 0.13 1.47Feed 22.0 ± 0.00 11.9 ± 0.13 0.42

philic anaerobic digesters. The program was used to provideinformation on the maximal (i.e. thermodynamic) mass of solidsthat could precipitate from the mixture (wastewater stream andNF brine) pumped into the FB reactor. Quite a few potential solidprecipitants may form as a result of blending centrifuge superna-tants with seawater NF brine obtained at c = 90%. Since a thermo-dynamic-based program cannot be used to provide informationregarding kinetic differences between solids which could poten-tially precipitate, kinetic patterns had to be assumed. Since struviteprecipitation kinetics has been reported to be much faster than thekinetics of any of the other solids that can possibly precipitate un-der the specified conditions (Ca/Mg – phosphate precipitants andCa/Mg – carbonate) (e.g. [23–25]), the presented theoretical exam-ination was first based on the assumption that at hydraulic reten-tion times of <60 min, the only precipitant would be struvite. Asshown further in Section 3.3, this assumption was strongly corrob-orated by the empirical results, under the conditions prevailing inboth tested scenarios. A second assumption was that the fast kinet-ics of struvite precipitation would lead to realization of the struvitePP, i.e. that in the effluent of the reactor the PP of struvite would bezero. Despite being incorrect (because true equilibrium cannot beattained at typical engineering HRTs, even in the presence of a highstruvite seed concentration) the latter assumption enabled attain-ing a fairly good assessment of the mass of struvite expected toprecipitate from a given solution. Additional assumptions were:(a) the operative pH was determined such that 90% of PT wouldbe removed from the WW stream (precipitation as struvite); (b)NT and PT concentrations in the WW stream were assumed to be600 and 450 mg N/L and 310 and 100 mg P/L, respectively; and(c) the composition of the NF brine dosed to the struvite reactoremanated from separating Mg(II) from seawater using a DS-DL5membrane (GE) at 90% recovery (Table 1). Note that the used brineunavoidably contains antiscalants, a fact that could not be ac-counted for in the theoretical simulation. Under these assump-tions, the combined effect of two operational conditionsdetermine the composition of the solution at equilibrium: the firstis the (constant) pH assumed to be maintained in the reactor (inthis case, for convenience, by NaOH addition; alternatively CO2

stripping can be used to minimize base costs), and the other isthe amount of Mg(II) added via the NF brine addition. At constanttemperature, all the other operational parameters depended uponthe choice of the these two values (e.g. the ionic strength of thesolution, ion pairing, the Ca(II) concentration, etc.). The constantpH maintained in the reactor was tested in the range pH 7.2–pH 8.4 while the simulated amount of Mg(II) added by the NF brinewas set at either 10 mM (or 15 mM) for the 1st scenario and3.22 mM in the 2nd scenario. For a given Mg(II) dosage, a higherpH value (within the simulated pH range) should result in an in-creased struvite PP (because the higher the pH the higher is thePO3�

4 fraction out of PT, whereas NH3 is negligible at the testedpH range). This is shown in Fig. 2, which presents the percentageof PT expected to precipitate from the wastewater stream, as afunction of pH, for the two examined Mg(II) additions (for the1st scenario, i.e. PT = 310 and NT = 600 mg/L). The combination of

e recovery ratio (membrane type: DS-DL5, applied pressure: 17 bar; n = 3).

(g/l) Mg2+ (g/l) K+ (g/l) SO2�4 (g/l)

± 0.01 2.37 ± 0.01 0.51 ± 0.01 6.8 ± 0.1± 0.02 2.80 ± 0.05 0.51 ± 0.05 8.1 ± 0.2± 0.05 3.64 ± 0.31 0.53 ± 0.02 10.9 ± 1.0± 0.02 3.91 ± 0.07 0.52 ± 0.07 11.7 ± 0.4± 0.02 4.72 ± 0.11 0.55 ± 0.02 14.2 ± 0.4± 0.02 5.86 ± 0.03 0.55 ± 0.05 17.5 ± 0.1± 0.02 7.97 ± 0.39 0.56 ± 0.03 23.0 ± 0.4± 0.01 1.46 ± 0.01 0.51 ± 0.010 3.8 ± 0.05

Table 3Rejection values for main seawater ions as a function of the recovery ratio (membrane type: DS DL5, applied pressure: 17 bar; n = 3).

Recovery (%) Cl� (%) Na+ (%) Ca2+ (%) Mg2+ (%) K+ (%) SO2�4 (%)

50 15.2 ± 0.7 12.4 ± 1.9 53.8 ± 3.1 78.1 ± 0.5 14.9 ± 2.9 97.2 ± 0.360 15.8 ± 0.6 12.0 ± 0.9 49.6 ± 4.2 77.2 ± 0.3 15.6 ± 1.0 97.1 ± 0.470 15.5 ± 2.4 11.0 ± 0.8 48.8 ± 1.9 75.0 ± 0.5 11.7 ± 2.4 96.3 ± 1.075 11.7 ± 2.2 11.3 ± 0.2 49.3 ± 0.2 74.3 ± 0.1 15.8 ± 1.7 96.7 ± 0.580 12.7 ± 1.1 9.9 ± 1.8 45.3 ± 4.3 72.6 ± 0.3 13.0 ± 3.7 96.5 ± 0.285 10.7 ± 0.9 9.8 ± 0.9 46.0 ± 2.4 71.6 ± 0.3 13.2 ± 2.7 96.1 ± 0.290 12.7 ± 3.9 8.3 ± 2.2 44.6 ± 3.3 69.0 ± 2.9 12.3 ± 3.3 94.8 ± 1.8

O. Lahav et al. / Separation and Purification Technology 108 (2013) 103–110 107

pH and Mg(II) addition unequivocally determined both the mass ofstruvite expected to precipitate and the NaOH dosage required tomaintain the chosen pH value constant. The percentages of P re-moval were calculated based on the PP of struvite in the mixtureof the wastewater stream and NF brine (denoted PPstruvite and ex-pressed in mmoles of struvite expected to precipitate from one li-ter of mixed solution), which was calculated by the program foreach scenario (i.e. for each pH and Mg addition combination), usingthe following Eq.:

%Premoval ¼ PPstruvite=ð%WW � ½PT�wwÞ ð1Þ

where %WW is the fraction of WW in the mixture (liter of WW perliter of mixture), and [PT]WW is the concentration of P in the WWprior to mixing with the NF brine (i.e. 10 mM).

In addition, for each scenario, the required moles of NaOH dos-age (denoted NaOHdose) per each kg of P removed were calculated.NaOHdose was calculated based on the consumption of NaOH (de-noted NaOHPHREEQC, and expressed in mmoles of NaOH per literof mixed solution) as calculated by the program, the PPstruvite andthe atomic weight of P:

NaOHdose ¼ NaOHPHREEQC=ðPPstruvite �MwðPÞÞ ð2Þ

Figs. 2 and 3 show the results of the simulation for the conditions ofScenario #1. From Fig. 2 it can be concluded that for attaining 90%PT removal, pH should be maintained at pH 7.27 and pH 7.55, forMg(II) dosages (via addition of NF brine; values refer to concentra-tions in the solution entering the struvite reactor) of 15 mM and10 mM, respectively. Based on Fig. 3 the required dosages of NaOHfor maintaining 90% PT removal are very close: 23.1 and 26.5 mmo-les per kg P precipitated, for 15 mM and 10 mM of Mg(II) addition,respectively, i.e. increasing the brine dosage by 50% results in 13%reduction in the required NaOH dosage, the reason being that thehigher Mg(II) concentration enabled operating at a lower pH value

70

75

80

85

90

95

100

105

7 7.2 7.4 7.6 7.8 8 8.2 8.4 8.6

PT

rem

oval

(%

)

pH

Mg addition = 10mMMg addition = 15 mM

Fig. 2. Simulated percentage of PT removed (out of the mass of P in the WW stream)as a function of the constant pH maintained in the reactor, for two given Mg(II)additions (1st scenario).

to attain a similar removal efficiency. The two examined cases (10and 15 mM Mg(II) addition) resulted in different calculated opera-tional costs (stemming from the cost of producing the NF brineand that associated with the NaOH requirement). Moreover, anunavoidable outcome of the elevated NF brine (i.e. the 15 mM case)dosage was an increased TDS (total dissolved solids) concentrationaddition (mainly Cl� and Na+) into the WWTP effluents. Finally, in-creased Mg(II) addition is necessarily accompanied by an increasedCa(II) addition. High Ca(II) concentrations may induce unwantedprecipitation of Ca-phosphate solids which have the potential to re-duce the purity of the struvite product, also requiring additionalNaOH dosage. As mentioned before, the theoretical examinationof the process relied on the assumption that the sole precipitantwas struvite. However, in order to evaluate the potential for Ca-phosphate precipitation in the two case studies, the PP of Ca3(-PO4)2(a) was also examined, following the complete precipitationof struvite: at this stage the PP of Ca3(PO4)2(a) was found to be0.42 and 0.46 mM (for Mg(II) addition of 15 and 10 mM, respec-tively), and the required modified NaOH dosages in this caseamounted to 26.8 and 30.3 mol per kg P removed, respectively, un-der the assumption that the PP of both struvite and Ca3(PO4)2(a)were fully realized, i.e. should Ca3(PO4)2(a) precipitate an increaseof 11% in the NaOH dosage should be expected (in the 10 mM Mg(II)addition case).

Based on this theoretical examination it was decided to operatethe struvite reactor in the 1st scenario first with 10 mM Mg(II)addition at varying hydraulic retention times (HRT) of 60, 30 and20 min. A Mg(II) dosage of 15 mM was tested only with HRT of20 min.

The results of the simulation of the 2nd scenario (PT = 100 andTAN = 450 mg/L) are not presented in detail, however using thesame set of assumptions and 3.22 mM Mg(II) dosage (throughthe NF brine), the pH calculated for 90% PT removal was found tobe 8.13.

18

21

24

27

30

33

36

39

42

7 7.2 7.4 7.6 7.8 8 8.2 8.4 8.6

Req

uire

d N

aOH

dos

e (m

oles

per

kg

PT

rem

oved

)

pH

Mg addition = 10mMMg addition = 15 mM

Fig. 3. Simulated required NaOH dosage (normalized per g of PT removed) from theWW as a function of the pH maintained in the reactor, for two given Mg(II) dosages(1st scenario).

Table 5Characterization of the solids by complete dissolution (n = 3).

Exp# HRT(min)

Mg(II) addedvia NF brine(mM)

Molar ratio between precipitatedcomponents

Ca/Mg PT/NT mg/NT

1 20 10.32 0.059 ± 0.054 1.100 ± 0.035 1.020 ± 0.0072 30 10.01 0.006 ± 0.003 1.065 ± 0.016 1.036 ± 0.0283 60 10.52 0.030 ± 0.009 0.990 ± 0.05 0.928 ± 0.0234 20 14.91 0.048 ± 0.026 1.206 ± 0.07 1.070 ± 0.0285 20 3.02 0.045 ± 0.004 1.017 ± 0.035 0.945 ± 0.0246 30 3.16 0.042 ± 0.007 1.043 ± 0.038 0.972 ± 0.0267 60 3.16 0.052 ± 0.0045 1.096 ± 0.012 0.947 ± 0.011

108 O. Lahav et al. / Separation and Purification Technology 108 (2013) 103–110

3.3. Results of struvite precipitation experiments

Table 4 shows the P removal efficiency, as well as the ratio be-tween the molar concentrations of Mg, PT and TAN which precipi-tated from the aqueous phase within every experiment (i.e. thedifference between influent and effluent concentrations). Table 5shows the ratio between the fractions of PT, TAN, Mg(II) and Ca(II)in the solids collected in each experiment, the ratios emanatingfrom complete dissolution of the solids.

The results of the 1st scenario (high PT concentration) can besummarized as follows: First, the results were very closely pre-dicted by the PHREEQC simulation (as also reported by [26]). Sec-ond, by dosing NF brine to add 10 mM of Mg(II) to the supernatantit is feasible to attain >90% PT removal at a pH value of 7.57, whileat the same time obtain an almost pure solid struvite product, at aretention time as short as 20 min. From the characterization of thesolids it is clear that the presence of a relatively high Ca(II) concen-tration in the precipitation reactor, induced by the NF brine addi-tion, hardly affected the obtained struvite purity, presumablybecause the kinetics of struvite precipitation are much faster ascompared to both hydroxyapatite and amorphous calcium phos-phate. The molar Ca(II) component in the solids obtained in allthe experiments was not higher than �5%, indicating struvite pur-ity >95%. In the 4th experiment, in which 15 mM of Mg(II) weredosed to the water, the higher ratio obtained between P and Mgin the solids seemed to indicate a deviation from this pattern, i.e.a more significant formation of (most probably, according to theXRD results in which these solids are not represented) Ca3(PO4)2

and Mg3(PO4)2 amorphous precipitants. Nevertheless, increasingthe NF dosage to result in addition of 15 mM of Mg(II), on top ofbeing more costly, did not improve the overall PT removal consid-erably and also resulted in 50% higher addition of unwanted ions tothe water. The conclusion was thus that adding Mg(II) at a ratio of1:1 with the measured PT was sufficient to attain the required P re-moval efficiency and higher Mg(II) dosages were not further at-tempted. Since the wastewater contained an intial Mg(II)concentration in the range 1.1–2.1 mM, the initial Mg(II) to PT ratioin the experiments was in fact around 1.15–1 in Scenario #1 and�1.5–1 in Scenario #2.

As mentioned, the solids obtained in the precipitation experi-ments were subjected to XRD analysis, in which all the sampleswere identified as struvite, although the spectra differed in the rel-ative intensities of some lines (a common phenomenon in engi-neered solids). Dittmarite, newberyite or other potential crystalswere not identified. The results, shown in Fig. 4 for the solids ob-tained in Experiments #4 and #7, indicate the presence of onlyone crystalline form, i.e. struvite. Clearly, XRD analysis cannot re-veal the presence of amourphus solids, such as Ca3(PO4)2 or Mg3(-PO4)2, which were likely present to some extent in the solidsproduct, as indicated by the complete dissolution results (Table 5),with emphasis on the composition of the solids obtained in Exper-iment #4. Having said this, the XRD results proved yet again the

Table 4Results of precipitation experiments based on influent and effluent aqueous phase measu

Exp# HRT (min) Actual Mg(II) concentrationdosed with NF brine (mmole/L)

P removal ef(aqueous pha

1 60 10.52 95.10 ± 2.302 30 10.01 92.60 ± 0.923 20 10.32 97.43 ± 1.584 20 14.91 97.76 ± 1.555 60 3.16 91.84 ± 0.826 30 3.16 91.30 ± 0.847 20 3.02 90.74 ± 0.52

feasibility and robustness of the suggested struvite precipitationmethod.

The solids obtained in the struvite precipitation experimentswere characterized by a very low SVI value of �10 mL/g, indicatinggood settlability and dewatering properties. The typical size distri-bution of the solids following their drying at 55 �C is given in Ta-ble 6. In all the experiments the majority of the obtained solidswere in the size range 63–250 lm, i.e. much smaller than struvitecrystals reported in similar pilot operations (e.g. [27]). This reasonfor the small size particles may stem from the fact that the exper-iments were operated for relatively short times (dozens of hours),determined for obtaining steady state with respect to the newkygenerated solids. Other reasons may be either the high shear forcesenduced by the hydraulic conditions in the FB reactor or perhapsthe presence of the antiscalant at �30 mg/L. Further work is re-quired to determine the effect (if at all) of the antiscalant presenceof the morphology of the struvite solids. Having said this, the re-sults of this work show that, as expected, the antiscalant did nothave an apparent effect on the precipitation of struvite per se,nor (down to HRT of 20 min) on the precipitation kinetics.

3.4. Approximate cost analysis

To roughly estimate the cost effectiveness of the suggested pro-cess, the cost of producing one kg of Mg(II) by seawater NF separa-tion was compared with cost of magnesium chemicals potentiallyused for struvite precipitation. To assess the cost of the NF separa-tion approach, the cost of desalinating brackish water using ROmembranes was considered. The reason for this was that lowTDS brackish plants are operated in Israel with similar operationalconditions (i.e. 17 bar pressure, 90% recovery and a similar antisca-lant dosage). From personal knowledge, the cost of producing 1 m3

of conditioned desalinated water in a 10 million m3/y brackishwater (TDS = 4000 mg/L) RO application in Israel is �$0.25. Thisis roughly also the cost quoted in the literature for low-TDS BWROapplications (e.g. [28]). Considering this cost (which includes a posttreatment stage, not required in the suggested NF application andrelates to 1 m3 of permeate rather than to 1 m3 of feed water), it

rements (n = 3).

ficiencyse results) (%)

Molar ratio between precipitated components

Ca/Mg PT/NT mg/NT

0.21 ± 0.001 1.01 ± 0.001 1.03 ± 0.010.1 ± 0.01 1.06 ± 0.06 1.12 ± 0.05

0.26 ± 0.05 1.13 ± 0.07 1.06 ± 0.060.04 ± 0.04 1.06 ± 0.07 1.15 ± 0.10.11 ± 0.02 0.97 ± 0.08 1.02 ± 0.090.64 ± 0.02 1.05 ± 0.07 1.13 ± 0.040.45 ± 0.01 1.11 ± 0.08 0.91 ± 0.08

Fig. 4. XRD pattern of the precipitated solids (Experiments #4 and #7) as compared to standard struvite.

Table 6Particle size distribution of the solids attained in the struvite precipitationexperiments.

Exp# >250 lm (%) 150–250 lm (%) 63–150 lm (%) <63 lm (%)

1 6.62 48.08 33.94 11.362 1.62 31.18 60.86 6.353 3.68 5.83 64.02 26.464 2.66 10.63 59.21 27.495 3.52 17.99 53.64 24.856 5.26 21.07 56.55 17.127 8.26 19.78 51.04 20.92

O. Lahav et al. / Separation and Purification Technology 108 (2013) 103–110 109

can be deduced that the cost associated with obtaining 1 kg ofMg(II) in the brine using the NF method will be roughly similar,based on the following assumptions: 70% magnesium rejection(i.e. approximately 1 kg of Mg(II) separated from each m3 of seawa-ter passed through the system), cost of NF membranes similar tothat of brackish RO membranes, and a plant of a similar size. Incomparison, the cost of magnesium chemicals was estimated at275 and 140 $/ton for MgSO4�7H2O and MgCl2�6H2O respectively,i.e. $2.787 and $1.171 per kg Mg. Thus, even if the cost of Mg(II)separation via the NF unit was 0.5 $/kg Mg (i.e. 100% higher thanthe estimated cost) it would still be less than half of the cost ofthe currently used chemicals. Under the reasonable assumptionthat the cost associated with the addition of magnesium chemicalsis 50% of the overall production cost, applying the suggested ap-proach will result in (at least) 25% decrease in the struvite produc-tion cost, under the condition that the WWTP is located close to theshore.

4. Conclusions

The results presented in the paper indicate that the concept ofusing seawater NF brine as a magnesium source within struviteprecipitation reactors is highly feasible, both economic and engi-neering wise. Both PT removal efficiency and struvite purity werevery high with Mg(II) addition (through the brine) that equaledthe initial PT concentration in the wastewater. Ion additions tothe wastewater were estimated to be relatively low. When thetreated wastewater is reused for irrigation the addition of 3.7 (Sce-nario #1) and 1.2 (Scenario #2) mg Na/L should be considered.Additional Cl� concentration is negligible relative to the case thatMgCl2 is used for magnesium supply. The cost of magnesium pro-duction using the NF separation method was estimated to be lessthan 50% of the cost of magnesium chemicals, provided that theWWTP is located close to the sea. Further research is required on

the possible effect of antiscalants on the morphology of the ob-tained struvite particles and for identifying nanofiltration mem-branes with more selective Mg(II) separation properties.

Acknowledgements

This project was funded by the Joint German-Israeli WaterTechnology Research Program (BMBF/MOST) and the Israel ScienceFoundation.

References

[1] P. Battistoni, R. Boccadoro, P. Pavan, F. Cecchi, Struvite crystallization in sludgedewatering supernatant using air stripping: the new full scale plant at Treviso(Italy) sewage works. in: Proceedings of the 2nd International Conference onphosphorus Recovery for Recycling from Sewage and Animal Wastes,Noordwijkerhout, Holland, March 12–14, 2001.

[2] Y. Jaffer, T.A. Clark, P. Pearce, S.A. Parsons, Potential phosphorus recovery bystruvite formation, Water Res. 36 (2002) 1834–1842.

[3] E. Munch, K. Barr, Controlled struvite crystallization for removing phosphorusfrom anaerobic digester sidestreams, Water Res. 35 (2001) 151–159.

[4] M. Yoshino, M. Yao, H. Tsuno, I. Somiya, Removal and recovery of phosphateand ammonium as struvite from supernatant in anaerobic digestion, Water Sci.Technol. 48 (1) (2003) 171–178.

[5] Q.Z. Wu, P.L. Bishop, Enhancing struvite crystallization from anaerobicsupernatant, J. Environ. Eng. Sci. 3 (2004) 21–29.

[6] A. Britton, F.A. Koch, D.S. Mavinic, A. Adnan, A. Oldham, B. Udala, Pilot scalestruvite recovery from anaerobic digester supernatant at an enhancedbiological phosphorus removal wastewater treatment plant, J. Environ. Eng.Sci. 4 (2005) 265–277.

[7] K.S. Le Corre, E. Valsami-Jones, P. Hobbs, S.A. Parsons, Kinetics of struviteprecipitation: effect of the magnesium dose on induction times andprecipitation rates, Environ. Technol. 28 (12) (2007) 1317–1324.

[8] M. Iqbal, H. Bhuiyan, D.S. Mavinic, Assessing struvite precipitation in a pilot-scale fluidized bed crystallizer, Environ. Technol. 29 (11) (2008) 1157–1167.

[9] N. Marti, A. Bouzas, A. Seco, J. Ferrer, Struvite precipitation assessment inanaerobic digestion processes, Chem. Eng. J. 141 (1–3) (2008) 67–74.

[10] D. Crutchik, J.M. Garrido, Struvite crystallization versus amorphousmagnesium and calcium phosphate, Water Sci. Technol. 64 (12) (2011)2460–2467.

[11] R. Prasad, A. Britton, B. Balzer, G. Schafran, Nutrient recovery by struvitecrystallization process: virginia experience. in: Proceedings of the WaterEnvironment Federation, WEFTEC, 2007, pp. 344–358.

[12] T. Dockhorn, About the economy of phosphorus recovery. in: InternationalConference on Nutrient Recovery from Wastewater Streams, Vancouver,Canada, IWA publishing, London, UK, 2009, pp. 145–158(ISBN:978843392323).

[13] Z.L. Ye, S.H. Chen, M. Lu, J.W. Shi, L.F. Lin, S.M. Wang, Recovering phosphorus asstruvite from the digested swine wastewater with bittern as a magnesiumsource, Water Sci. Technol. 64 (2) (2011) 334–340.

[14] A. Gunay, D. Karadag, I. Tosun, M. Ozturk, Use of magnesit as a magnesiumsource for ammonium removal from leachate, J. Hazard. Mater. 156 (2008)519–623.

[15] E.J.C. Borojovitch, M. Munster, Z. Rafailov, Precipitation of ammonium fromconcentrated industrial wastes as struvite: a search for the optimal reagents,Water Environ. Res. 82 (7) (2010) 586–591.

[16] M. Telzhensky, L. Birnhack, O. Lehmann, E. Windler, O. Lahav, Selectiveseparation of seawater Mg2+ ions for use in downstream water treatmentprocesses, Chem. Eng. J. 175 (2011) 136–143.

110 O. Lahav et al. / Separation and Purification Technology 108 (2013) 103–110

[17] G. Bargeman, J.M. Vollenbroek, J. Straatsma, C.G.P.H. Schroën, R.M. Boom,Nanofiltration of multi-component feeds. Interactions between neutral andcharged components and their effect on retention, J. Membr. Sci. 247 (2005) 11–20.

[18] B. Cuartas-Uribe, M.I. Alcaina-Miranda, E. Soriano-Costa, A. Bes-Pia,Comparison of the behavior of two nanofiltration membranes for sweetwhey demineralization, J. Dairy Sci. 90 (2007) 1094–1101.

[19] R.B. Willis, M.E. Montgomery, P.R. Allen, Improved method for manual,calorimetric determination of total Kjeldahl nitrogen using salicylate, J. Agric.Food Chem. 44 (1996) 1804–1807.

[20] Standard Methods for the Examination of Water and Wastewater, 20th ed.,APHA, AWWA, WPCF, Washington DC, USA, 1998.

[21] A.Z. Abdullatef, A.M. Farooque, G.F. Al-Otaibi, N.M. Kither, S.I. Al Khames,Optimum Nanofiltration Membrane Arrangements in Seawater Pretreatment:Part-1, IDA World Congress – Maspalomas Gran Canaria – Spain, October 21–26, 2007.

[22] L.N. Plummer, D.L. Parkhurst, G.W. Fleming, S.A. Dunkle, A computer programincorporating Pitzer’s equations for calculation of geochemical reactions inbrines, US Geological Survey, Water-Resources Investigations Report, 1988.88–4153.

[23] E.V. Musvoto, M.C. Wentzel, G.A. Ekama, Integrated chemical-physicalprocesses modeling – II. Simulating aeration treatment of anaerobic digestersupernatants, Water Res. 34 (2000) 1868–1880.

[24] V. Babic-Ivancic, J. Kontrec, L. Brecevic, D. Kralj, Kinetics of struvite tonewberyite transformation in the precipitation system MgCl2–NH4H2PO4–NaOH–H2O, Water Res. 40 (18) (2006) 3447–3455.

[25] B. Lew, S. Phalah, H. Sheindorf, M. Kummel, M. Rebhun, O. Lahav, Favorableoperational conditions for obtaining high value struvite-product from filtrateof sludge dewatering systems, Environ. Eng. Sci. 27 (9) (2010) 733–741.

[26] J.C.L. Warmadewanthi, Recovery of phosphate and ammonium as struvite fromsemiconductor wastewater, Sep. Purif. Technol. 64 (2009) 368–373.

[27] H. Huang, D.S. Mavinic, K.V. Lo, F.A. Koch, Production and basic morphology ofstruvite crystals from a pilot-scale crystallization process, Environ. Technol. 27(2006) 233–245.

[28] I.C. Karagiannis, P.G. Soldatos, Water desalination cost literature: review andassessment, Desalination 223 (2008) 448–456.

[29] P. Barak, A. Stafford, Proceedings of the Wisconsin Fertilizer, Aglime and PestManagement conference, 17–19 January, Madison, Virginia 45 (2006) 199–204.

[30] N. Fujimoto, T. Mizuochi, Y. Togami, Phosphorus fixation in the sludgetreatment system of a biological phosphorus removal process, Water Sci.Technol. 23 (1991) 635–640.

[31] P. van Rensburg, E.V. Musvoto, M.C. Wentzel, G.A. Ekama, Modelling multiplemineral precipitation in anaerobic digester liquor, Water Res. 37 (2003) 3087–3097.