Chemosphere 92 (2013) 1375–1382

Contents lists available at SciVerse ScienceDirect

Chemosphere

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

Technical Note

A membrane-integrated advanced scheme for treatment of industrialwastewater: Dynamic modeling towards scale up

0045-6535/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.chemosphere.2013.05.006

⇑ Corresponding author. Tel.: +91 343 2755055; mobile: +91 9434788105; fax:+91 343 2754078.

E-mail address: parimalpaul1999@yahoo.co.in (P. Pal).

Ramesh Kumar, Parimal Pal ⇑Environment and Membrane Technology laboratory, Department of Chemical Engineering, National Institute of Technology, Durgapur 713 209, India

h i g h l i g h t s

�Mathematical model developed for an advanced integrated water treatment plant.� Highly integrated scheme is novel in ensuring by-product recovery and water reuse.� Model successfully predicted plant performance.� Study will facilitate industrial scale up of this new and advanced scheme.

a r t i c l e i n f o

Article history:Received 28 January 2013Received in revised form 30 April 2013Accepted 3 May 2013Available online 2 June 2013

Keywords:Aquatic environmentWastewater treatmentAdvanced processMathematical modeling

a b s t r a c t

Modeling and simulation was carried out for an advanced membrane-integrated hybrid treatment processthat ensures reuse of water with conversion and recovery of ammoniacal nitrogen as value-added struvitefertilizer from coke wastewater. While toxic cyanide was largely removed in a pre-chemical treatmentunit using Fenton’s reagents under optimized conditions, more than 95% of NHþ4 –N could be recoveredas a valuable by-product called struvite through addition of appropriate doses of magnesium and phos-phate salts. Water could be turned reusable through a polishing treatment by nanofiltration membranesin a largely fouling free membrane module following a biodegradation step. Mathematical modeling ofsuch an integrated process was done with Haldane–Andrew approach for the associated microbial degra-dation of phenol by Pseudomonas putida. Residual NHþ4 was degraded by nitrification and denitrificationfollowing the modified Monod kinetics. The model could successfully predict the plant performance asreflected in reasonably low relative error (0.03–0.18) and high Willmott d-index (>0.98).

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Since the early 1970s efforts have been directed towardsmodeling of wastewater treatment processes. Activated sludgeprocess (ASP) is one such wastewater treatment scheme for whichsome models were developed by the task group of InternationalWater Works Association. However, these models have been usedin carbon oxidation, nitrification, denitrification and biologicalphosphorus removal processes (Gujer and Henze, 1991) primarilyfor domestic wastewater treatment only and there is significantlimitation to their application to industrial wastewater treatment.With development of highly selective membranes in the area ofseparation and purification, membrane-integrated novel processesare now emerging with the promise of much better quality oftreated water. However, modeling studies on such membrane-integrated treatment processes are extremely scanty. In some

typical highly water-intensive industries like coke making andmetallurgical industries, problems of effective treatment of enor-mous quantities of complex wastewater persist in the absence ofscale up confidence. This is where modeling and simulation studiesof modern membrane-integrated treatment schemes are abso-lutely essential.

Coke wastewater generated in the coal carbonization plantscontains substantial amounts of toxic compounds such as thiocya-nate, cyanide along with NHþ4 and organic compounds like phenols,aromatic nitrogenated compounds and polycyclic aromatic com-pounds. Often wastewater discharges from coke making industriesenter into the river bodies causing severe pollution to surfacewaters. Thus for environmental preservation the enormous quanti-ties of coke plant wastewater should be treated at relatively lowcost. This could be done in biological treatment units for major pol-lutants like phenol, NHþ4 and hydrocarbons but presence of cyanidecompounds very often turns survival of the microbes difficult inthe environment of the treatment system and this often leads tofailure of such plants (Ghose, 2002). Individual studies on cyanide,phenol and nitrogen removal kinetics have been documented.

1376 R. Kumar, P. Pal / Chemosphere 92 (2013) 1375–1382

Chemical treatment of cyanide by oxidation using hydrogen perox-ide in presence of copper catalyst has been reported (Sarla et al.,2004) to follow pseudo-first order model. Biodegradation of phenolhas been described by Haldane–Andrew growth kinetics model forthe entire concentration range (Kumar et al., 2005). Dincer andKargi (2000) found that nitrogen removal in sequential nitrificationand denitrification processes follow the Monod kinetics. Develop-ment of new techniques to recover nitrogen is essential in the cur-rent scenario. NHþ4 –N in high concentration may be separated andrecovered by adding magnesium and phosphate salt which onreaction with ammonium nitrogen precipitate out as magnesiumammonium phosphate hexahydrate (MAP), a useful by-product(Lee et al., 2003; Wang et al., 2006). However, very little has beenreported till date on the modeling of an integrated processingscheme leading to reuse of water while simultaneously addingeconomy to the process. Through recovery of valuable struviteby-products from wastewater as proposed in the present study,economic viability is likely to improve substantially towards sus-tainable management. The new scheme also saves on consumptionof fresh water from external sources as it turns treated water reus-able within the same industry. Though, literature abounds inexperimental studies of activated sludge process, Fenton’s treat-ment and similar chemical treatments, studies on schemes inte-grating biological, chemical and membrane based treatmentsusing microporous and nanofiltration membranes have hardlybeen reported. Modeling and simulation studies of such plantsare absolutely essential for industrial scale-up. The present studyintends to develop a realistic mathematical model for such an inte-grated process. Such a model if succeeds in predicting the perfor-mance of the membrane-integrated bio-chemical treatment plantwill pave the way for industrial scale up. To our knowledge, nosuch studies have been reported in the literature.

2. Theory and model development

2.1. Chemical and biological treatment unit

Toxic cyanide compounds as present in coke oven wastewatercan be degraded into less harmful compounds using Fenton’s re-agent. Cyanide removal in such chemical treatment follows thefirst-order kinetics. Similarly, NHþ4 present in high concentrationmay also be precipitated out in the form of magnesium ammoniumphosphate or MAP (also known as Struvite, MgNH4PO4�6H2O) byadding magnesium and phosphate salts to such wastewater wherefirst order kinetics (Nelson et al., 2003) are followed.

Modeling of phenol, NHþ4 and nitrate degradation process in-volves mass balance of these pollutants, which in the present case,requires the knowledge of the flow pattern of the liquid phase andthe process kinetics. The following assumptions were made whileformulating the model: (a) micro-organisms are pre-acclimatedto the substrate and can metabolize substrate in the aqueous phaseonly; (b) during biodegradation, the pH of the system remains con-stant; (c) no degradation of substrate takes place in the settlingtanks; (d) dissolved oxygen (DO) content is sufficient enough tosustain the aerobic reactions in the nitrification process; (e) foodto microorganism ratio (F/M) was maintained in the ranges of0.06–0.1 kg COD kg�1 VSS d�1 in the nitrification unit; (f) dissolvedoxygen concentration (DO) was maintained at less than 0.5 mg L�1

in the denitrification unit and (g) substrate-specific treatment oc-curs in each reactor and other substrates remain inert if presentsimultaneously.

To simulate the continuous operation of the phenol, NHþ4 andnitrate degradation in the reactor, the treatment scheme as pre-sented in Fig. SM-1 of Supplementary material (SM) is modeled.The mass balance for the reactors has been done with variable

parameters like substrate, biomass and specific growth rate of mi-cro-organism. The reflux ratio, for all the three recycles is taken tobe 10%. The intentional wasting coefficient (b) is a parameter usedto maintain sludge retention time (SRT). When effluent substrateconcentration (Se) reaches below detection limit; b can be opti-mized to stabilize the system. Material balance and model devel-opment for the combined reactor and clarifier system may beseen online in SM Section 2.1.

2.2. Nanofiltration membrane separation

Based on the present scheme of separation of ionic contami-nants specifically chloride and bicarbonate ions from chemicallyand biologically treated effluent using cross flow nanofiltrationmembrane module a mathematical model has been developed byusing extended Nernst–Planck approach (ENP). Model equationsare formulated with the assumptions that (a) the membrane con-sists of a bundle of identical straight cylindrical pores of radius rp

and length Dx (with Dx� rp); (b) the effective membrane volumecharge (Xd) is constant throughout the membrane and is mainlycontrolled by the feed concentration and (c) concentration of com-ponent i inside the membrane and electric potential are defined interms of radially averaged quantities.

In the modeling, the linearization approach of Bowen andWelfoot (2002) was adopted. In the present investigation, trans-port model has been developed for chloride and bicarbonate ionsthrough NF membrane. The details of nano-transport model devel-opment, determination of physico-chemical parameters and com-putational procedure may be seen online in SM Section 2.2. Thevalues of the parameters used in the model development havebeen summarized in Table SM-1.

3. Experimental

3.1. Materials

All the chemical reagents used during investigation were ofreagent grade and were used as such without further purification.NHþ4 and cyanide standard solutions were procured from Merck(Germany). The chemicals like methanol for calibrating phenylcolumn (Sigma–Aldrich, WI, USA), phenol and other chemicalswere purchased from Merck (India). Thin film composite polyam-ide nanofiltration (NF1, NF2, NF3 and NF20) as well as poly vinyl-idene fluoride (PVDF) microfiltration membranes were procuredfrom Sepromembranes (USA) and Membrane solution (USA)respectively.

3.2. Wastewater sampling and characterization

The industrial wastewater used in the reactor was actual coke-oven wastewater (ammoniacal liquor) from an operating plant(Durgapur Project, West Bengal, India) which is a coal based powerplant manufacturing coke as by product. Its physico-chemicalcharacterization is presented in Table 1.

3.3. Experimental set-up

The integrated wastewater treatment plant consists of a seriesof continuous stirred tank reactors (CSTR). All the CSTRs (madeof stainless steel, SS 316) are positioned in vertically lowering or-der allowing free flow of the liquid to the next stage reactors bygravity eliminating the need for use of pumps in between the reac-tors as shown in Fig. 1. Two membrane modules were connected inseries for micro and nanofiltration. While membrane moduleslike hollow fibre and spiral bound are prone to rapid fouling the

R. Kumar, P. Pal / Chemosphere 92 (2013) 1375–1382 1377

flat-sheet cross-flow membrane module used in the presentscheme is largely fouling-free. This permits the use of the samemembrane for long time without much decline in flux and reuseof the membrane with simple rinsing and cleaning.

3.4. Experimental procedure

The set up starts with a cyanide treatment reactor of volume 30L where real coke wastewater is first treated with Fenton’s reagent(FeSO4�7H2O and H2O2) for removal of cyanide. A second reactorwas used for conversion of NHþ4 –N into struvite through additionof magnesium and phosphate salts. The clear supernatant solutionpasses down to the series of three CSTRs via a settling tank androtameter. The sequentially arranged continuously stirred tankreactors of volume 15 L each had respectively the microbial cul-tures of Pseudomonas putida (NCIM 2152), nitrifying mixed culture(Nitrosomonas 5076 and Nitrobacter 5078) and Pseudomonasaeruginosa (NCIM 5032) in minimal salt medium (MSM) whereall the microbial strains were collected from national collectionof industrial microorganism NCIM (Pune, India). In the last step,biologically treated effluent was microfiltered under low pressure(around 0.2 MPa) and then subsequently passed to a second nano-filtration membrane cross flow membrane module (with effectivefiltration surface of 0.1 m2 each module) via a settling tank wherealmost all the charged as well as non-charged particles gotseparated out.

3.5. Analytical methods

Cyanide, NHþ4 , fluoride, sodium and chloride content weredetermined by Orion 4 Star pH ISE Benchtop ion meter withrespective electrodes. The phenol content was determined by HPLC(Agilent Technologies 1200 series, USA) with Zorbax SB-Phenyl col-umn (Germany). COD was measured by COD Vario Tube Test MR(0–1500 mg L�1) of COD analyzer of LoviBond (WTW) kits, Ger-many. Alkalinity (bicarbonate ion), phosphate, magnesium, nitrateand Oil and grease were determined following the standard proce-dures described in the standard methods (APHA, 1995). Total car-bon and total organic carbon were determined by TOC analyzer(Aurora 1030 USA). For measuring biomass wavelength was setat 600 nm using UV–vis double beam spectrophotometer (Cecil7000 series, England). Rejection of bicarbonate and chloride ionsfrom biologically treated effluent by the nanofiltration membraneswere computed using the initial value in the feed and the residualvalue in the treated effluent (permeate).

Table 1Characteristics of wastewater after final treatment (membrane separation).

Parameter Influent concentration (mg L�1) After chemical & bi

Cyanide 120 NDa

Phenol 159 NDNHþ4 –N 4195 NDChloride ion 5930 18100COD 2470 1570Total carbon 1150 192Total organic carbon 246 15.5TDS 11600 37800Fluoride 112 110Sodium 1850 3540Oil and grease 51.5 3.5Conductivity 12.02 38.56(mS cm�1)Alkalinity-CaCO3 4250 29100

Salinity 8.4 25.9pH 8.5–9.5 8.5–9.0

a ND = non-detectable.

4. Results and discussion

4.1. Conversion in chemical treatment unit and precipitation ofstruvite

Fenton’s reagent (FeSO4�7H2O and H2O2) was used for the deg-radation of the cyanide. The ratio of H2O2:COD and H2O2:FeSO4�7H2O were 5:1 w/w and 1:50 w/v respectively which were foundto be appropriate for the complete removal of cyanide in minimumtime. The time taken for complete degradation of cyanide was10 min at pH 10.0 (Fig. 2a). The plot �ln(C/Co) vs. t is a straight line(R2 = 0.99) confirms the first order reaction kinetics of cyanide deg-radation, whereas Co and C are the initial concentration and reac-tant concentration at time t of cyanide respectively. Kineticconstant was obtained as 0.012 s�1 where H2O2/Fe2+ is a criticalparameter for improving the efficiency of the Fenton’s process.Cyanide can be mineralized to bicarbonate and NHþ4 by using Fen-ton’s reagent, as follows:

CN� þ 2H2O !H2O2 NHþ4 þHCOO� ð1Þ

Fenton’s oxidation for cyanide degradation is known as a highlypH-dependent process. It was observed that alkaline pH was betterfor degradation of cyanide. At high pH, cyanide is present as CN�

ions so it reacts easily with H2O2 and Fe2+ ions. But, it was observedthat at high pH degradation of aromatic compounds like phenolswas very low by Fenton’s reagent. In acidic pH, cyanide is presentmainly as HCN gas which is very difficult to oxidize. Ferrous doseslead to the generation of more �OH and acts as a catalyst. Hydroxylradicals are devoid of any charge and have a high affinity for elec-trons and can quickly strip any chemical of electrons including cya-nide causing their oxidation.

The effluent from Fenton’s treatment unit contained an en-hanced concentration of NHþ4 (4300 mg L�1) due to conversionof cyanide to NHþ4 and bicarbonate ion. NHþ4 –N was precipitatedout as MAP from the Fenton’s treated effluent in the struvite by-product section. In the first step of MAP precipitation, differentcombinations of the chemicals MgCl2�6H2O, Na2HPO4�12H2O,MgO, H3PO4, Ca(H2PO4)2�H2O and MgSO4�7H2O at different molarratios of Mg2+ and PO3�

4 with respect to NHþ4 concentration wereinvestigated to find out the best combinations for maximumprecipitation of NHþ4 –N as struvite. The experimental results asshown in Fig. 2b, indicate that the combination of MgCl2�6H2O,Na2HPO4�12H2O and NHþ4 was the best when used in the molarratio of 1:1:1 at pH 9.0. Struvite precipitation follows(Eq. (2)):

ological treatment (mg L�1) Final effluent Permissible limit (mg L�1)

ND <0.1ND <0.5ND <30181 <1000102 <2504.21.11360 –0.23 <1.535 <60ND <100.92 –

582 –

0.85 –7.5–8.5 –

Fig. 1. Schematic diagram for the coke-oven wastewater treatment plant.

Fig. 2. (a) Concentration profile of cyanide when treated with H2O2:COD (5:1) w/w and H2O2:FeSO4�7H2O (1:50) molar ratio; and (b) efficiency of NHþ4 –N removal profilefrom coke wastewater by selection of different combination of chemicals added at different pH.

1378 R. Kumar, P. Pal / Chemosphere 92 (2013) 1375–1382

MgCl2 � 6H2Oþ NHþ4 þ Na2HPO4 � 12H2O

!MgNH4PO4 � 6H2O # þ2NaClþ 12H2OþHþ ð2Þ

The combination of MgCl2�6H2O and Na2HPO4�12H2O was themost efficient in NHþ4 removal, but this also led to high salt concen-tration in the effluent in comparison to other two combinations ofchemicals as evident from (Eq. (2)). The salts produced at this stagewere subsequently removed by nanofiltration membrane (as dis-cussed in Section 4.3). A first-order kinetic model was applied tothe experimental data obtained during NHþ4 precipitation withtime. By integrating and ordering the terms, the following linearform of the first order rate equation was obtained (Eq. (3)):

� lnC � CE

Co � CE¼ kt ð3Þ

where Co is the initial concentration of the reactant (NHþ4 ). Assum-ing first-order reaction kinetics of NHþ4 precipitation, a plot of�ln[(C � CE)/(Co � CE)] against reaction time gave a straight line(R2 = 0.99) with slope (k) equal to 0.12 min�1 and intercept equalto zero. Equilibrium concentration of NHþ4 (CE) was obtained fromthe intercept with abscissa at �dc/dt equal to zero when thisparameter was plotted against the concentration of NHþ4 remainingin the filtrate. It was found that the equilibrium, CE was 201 mg L�1

NHþ4 –N.

R. Kumar, P. Pal / Chemosphere 92 (2013) 1375–1382 1379

The MAP precipitate formed rapidly and settled quickly at thebottom in absence of stirring. The content of struvite in precipi-tates has been confirmed through scanning electron microscopywith energy dispersive X-ray analysis, Fourier transform infra-redanalysis and X-ray Diffraction as shown in Fig. 3a–d. The FT-IRanalysis, SEM-EDS and XRD patterns exhibited that the infraredspectrum of the precipitate and elemental profile were close tothat of the MAP as reported elsewhere (He et al., 2007). The Ther-mo-gravimetric analysis and differential thermo-gravimetric anal-ysis curve for struvite at 10 �C min�1 is shown in Fig. 3e. These dataindicate that mass loss begins at a temperature around 55 �C and isessentially complete when temperature reached above 250 �C and�51% corresponds to the following decomposition for the struvite(Eq. (4)):

MgNH4PO4 � 6H2O!MgHPO4 þ NH3 " þ6H2O " ð4Þ

DTGA curve of the struvite for heating rate 10 �C min�1 showsingle peak which is attained at temperature 103 �C due to simul-taneous loss of both ammonia and water molecules which indi-cates that precipitate material is struvite.

4.2. Biodegradation

Batch experiments were conducted to investigate the effect ofphenol, NHþ4 and nitrate concentration on the growth kinetics ofP. putida, nitrifying bacteria and P. aeruginosa for different initialsubstrate concentrations to find out the kinetic constants (ks ob-tained 14.0, 10.9, 26.9 mg L�1 respectively). The specific growthrate (l) was determined in the exponential growth phase for eachrespective micro-organism by using different substrate concentra-tion of phenol, NHþ4 and carbon source (methanol during denitrifi-cation). The variation of experimental specific growth rate(P. putida) as a function of phenol concentration was well fittedin Haldane-Andrews model curve with specific growth rate0.45 h�1. The substrate inhibition constant and the decay

Fig. 3. Surface characterization analysis of struvite (MAP precipitate) obtained during thedispersive X-ray analysis of struvite; (c) Fourier transform infrared spectroscopy analysisstruvite for heating rate 10 �C min�1.

coefficient of P. putida during phenol biodegradation were foundto be 500.8 mg L�1 and 3 � 10�5 h�1 respectively. Experimentaldata indicated that nitrification by nitrifying bacteria followedMonod kinetics with l value of 0.24 h�1. Degradation of nitratewas found to be of zero order whereas growth pattern of P. aerugin-osa on carbon source methanol followed the Monod kinetics with lvalue of 0.29 h�1. Fig. 4 shows that experimental and model pre-dicted profiles of P. putida, nitrifying bacteria and P. aeruginosawith initial phenol, NHþ4 –N and nitrate concentration of 159, 200and 191 mg L�1 respectively was degraded in 13, 23 and 14 hrespectively. The kinetic parameters obtained from the batch studyhave been used here to model a continuous system in an integratedplant.

After precipitation of NHþ4 , the supernatant fluid (effluent ‘a’)was transferred for the biological treatment under controlled flowrate. Fig. 5 shows the biodegradation potential of different micro-organisms in degrading their respective substrates and their bio-mass yields during continuous process. As continuous flow ofwastewater was allowed containing phenol, NHþ4 and nitrate intothe respective reactor system, another exponential growth phaseof the microbe population was observed. The exponential growthphase ends as the microbial population becomes high enough toconsume influent substrate instantly. At this point, the systemwas allowed to bleed a part of the biomass to maintain its concen-tration within a desired level. The hydraulic residence time andSRT were optimized at 19 h and 20 d, 23 h and 16 d, 15 h and30 d respectively for the successful treatment of phenol, NHþ4 andnitrate respectively. Phenol was almost completely removedregardless of the loading variation. The removal efficiency was al-ways higher than 99% even at very low hydraulic retention time.The nitrification reactions were generally coupled and proceededrapidly to form nitrate from NHþ4 . Therefore, nitrite levels at any gi-ven time were usually low as detected in the relevant analysis.Nitrification required a long retention time of 25–30 h, a low foodto microorganism ratio (F:M) 0.08 g COD g�1 VSS d�1, a high mean

chemical treatment of NHþ4 . (a) Scanning microscopy analysis of struvite; (b) energyof struvite; (d) XRD pattern of the struvite crystel; and (e) TGA and DTGA curves for

Fig. 4. Experimental and model predicted profiles during batch studies (a)Pseudomonas putida during phenol degradation. (b) Mixed nitrifying bacteria(Nitrosomonas and Nitrobacter) during conversion of NHþ4 to nitrate. (c) Pseudomonasaeruginosa during nitrate and carbon source (methanol) degradation.

Fig. 5. Experimental and model predicted profiles during continuous treatment of(a) phenol; (b) NHþ4 and (c) nitrate.

1380 R. Kumar, P. Pal / Chemosphere 92 (2013) 1375–1382

cell residence time (sludge age) 15–25 d, adequate buffering(alkalinity) 3.0 g NaHCO3 L�1 and dissolved oxygen concentrationof 4.5 mg L�1. Concentration of nitrate in the reactor increasesslowly at the beginning of the continuous process. This was dueto the continuous removal of reactor contents to maintain flow.As the NHþ4 concentration depleted nitrate concentration in-creased. P. aeruginosa was used as denitrification heterotrophicbacteria and methanol was used as the carbon source for thegrowth of the bacteria. The optimum dose of methanol that wouldresult in maximum removal of nitrate was determined experimen-tally. The methanol:nitrate ratio was 2:1. The facultative bacteriadraw dissolved oxygen that results from nitrate molecules. Theprocess is performed under anoxic conditions, when the dissolved

oxygen concentration is less than 0.2 mg L�1. Methanol was se-lected as a suitable external carbon source because of its low costand very effective role in denitrification in an environment of rel-atively low carbon concentration. The rate of denitrification wasreported to be zeroth order for NO3 concentrations below400 mg L�1 (Foglar et al., 2005). The (Eq. (5)) describing the nitrifi-cation reaction is as follows:

6NO�3 þ 5CH3OH! 3N2 " þ5CO2 þ 7H2Oþ 6OH� ð5Þ

The recycle ratio was maintained at 0.1 in each pair of reactorand settling tank.

4.3. Membrane separation

After the two steps of chemical treatment and then biologicaltreatment, the effluent b got enriched in terms of total dissolvedsolids (TDSs), salinity and conductivity due to introduction ofnew ionic species such as bicarbonate and chloride ions (Table 1).Nanofiltration played a significant role in removal of these ionic

Fig. 6. Comparative (a) flux behavior and (b) rejection of bicarbonate ions and (c)chloride ions performances of the membranes under varying pressure at cross-flowrate of 800 L h�1 and pH 9.5.

R. Kumar, P. Pal / Chemosphere 92 (2013) 1375–1382 1381

species and the salts in the final stage of treatment of cokewastewater. Four different polyamide composite nanofiltrationmembranes (NF1, NF2, NF3 and NF4) fitted to a largely foul-ing-free cross flow module were examined during nanofiltration.Fig. 6a shows the permeate flux data from the experimentalinvestigations as well as from the model. It shows that the per-meate flux increases with increase of transmembrane pressurefor all the NF membranes and varied linearly with applied pres-sure. NF2 membrane yielded the highest flux (at 1.5 MPa,294 L m�2 h�1 flux) due to its large pore radius (0.57 nm)whereas the NF1 membrane (0.53 nm) yielded the lowest flux(at 1.5 MPa, 116 L m�2 h�1 flux) as permitted by its smallest poreradius. However, at 1.5 MPa pressure, the flux of NF1 membraneof around 115–116 L m�2 h�1 was reasonably good and could beaccepted for scale up. The NF20 and NF3 membrane yielded aflux intermediate between the two former types as its pore ra-dius (0.54 and 0.55 nm) was in between. The developed modelpredicts such increase of flux following increase in operatingpressure and the experimental findings agreed well with suchpredictions.

Fig. 6b and c shows the model vs. experimental data of removalof bicarbonate and chloride ion by different nanofiltration mem-branes under varying pressures (from 0.5 to 1.6 MPa) at a pH ofaround 9.5. However, beyond the operating pressure of 1.5–1.6 MPa, no further improvement in rejection of bicarbonate andchloride ions were observed except a very marginal increase in fluxunder the investigated maximum pressure range. It was observedthat the rejection order followed NF1 > NF3 > NF20 > NF2 at anyparticular applied pressure. The data set as shown in Fig. 6b andc were fitted to the linearized model in order to obtain the effectivemembrane charge density Xd. Two transport mechanisms maywork in case of NF membranes. In solution diffusion mechanism,solute flux and solvent flux are uncoupled and as a result increaseof solvent flux occurs with increase of transmembrane pressurewithout increasing solute flux. So with increasing pressure, purewater flux will increase, while the solute flux (bicarbonate andchloride ion) is constant and due to the low concentration of thesolutes in the permeate side, the overall solute passage decreases.In the size exclusion mechanism, relative sizes of the membranepore and solute dimension assume important role in determiningdegree of separation. Higher pH values lead to an increase in osmo-tic pressure and ionic strength thus reducing the membrane per-meability and increasing rejection. High pH favours an increasedthickness of diffuse double layer of charged functional groupsdue to deprotonation over the surface of membrane and thus re-duces the apparent pore size and results in greater rejection ofcharged solutes (Braghetta et al., 1997). Donnan exclusion orcharge repulsion plays the dominant role as it remains as nega-tively charged ion. So after nanofiltration, the final effluent maybe recycled in the same plant.

5. Error analyzes

During the experimental investigations, random errors werekept at the minimum possible levels by averaging. The mean ofthree values was always considered as the final value. To validatethe model, the experimental data were compared with the mod-el-predicted values. The deviations of the experimental findingsfrom the model predictions were calculated in terms of the modelerrors by calculating relative error (RE) and Willmott index (d). Thevalues of d should take from 0 to 1.0, with an index of 1.0 indicat-ing perfect agreement (Willmott, 1982) and expressed as:

d ¼ 1�Pn

i¼1ðPi � EiÞ2Pni¼1 jðPi � EÞj þ jðEi � EÞj� �2

24

350 6 d 6 1 ð6Þ

where Ei is experimental value; Pi is predicted (model) value; n is noof points analyzed.

RE and d were found to be in the range of 0.03–0.18 and 0.98–0.99 respectively indicating capability of the model to successfullypredict the system performance.

6. Conclusion

The dynamic mathematical modeling effort of the present studyeffectively paves the way for industrial scale up of a very advancedwastewater treatment scheme. Statistical analysis indicates highcapability (with overall correlation coefficient being of the orderof 0.99) of the model to successfully predict the performance ofthe hybrid treatment plant. The model first of its kind deals witha novel, continuous treatment process for hazardous industrialwastewater that integrates conventional chemical and biologicaltreatments with membrane based treatment following a wellinvestigated, logical sequencing of operations and leading to pro-duction of valuable struvite by-product and reusable water. The

1382 R. Kumar, P. Pal / Chemosphere 92 (2013) 1375–1382

treatment scheme and the model depart from the traditional,conventional activated sludge-based processes and appear to besustainable by virtue of its capability of fast treatment of industrialwastewater in an operationally and economically attractive waywhere its applicability will not remain confined to the coke makingindustry alone.

Acknowledgment

Authors are thankful to the Department of Science and Technol-ogy, Government of India for financial support under DST-FISTProgram.

Appendix A. Supplementary material

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.chemosphere.2013.05.006.

References

APHA, 1995. Standard Methods for the Examination of Water and Wastewater, 19thed. American Public Health Association, Washington, DC.

Bowen, W.R., Welfoot, J.S., 2002. Modeling the performance of membranenanofiltration critical assessment and model development. Chem. Eng. Sci. 57,1121–1137.

Braghetta, A., DiGiano, F.A., Ball, W.P., 1997. Nanofiltration of natural organicmatter: pH and ionic strength effects. J. Environ. Eng. 123, 628–641.

Dincer, A.R., Kargi, F., 2000. Kinetics of sequential nitrification and denitrificationprocesses. Enzyme Microb. Technol. 27, 37–42.

Foglar, L., Briski, F., Sipos, L., Vukovic, M., 2005. High nitrate removal from syntheticwastewater with the mixed bacterial culture. Bioresour. Technol. 96, 879–888.

Ghose, M.K., 2002. Complete physico-chemical treatment for coke plant effluents.Water Res. 36, 1127–1134.

Gujer, W., Henze, M., 1991. Activated sludge modeling and simulation. Water Sci.Technol. 23 (4–6), 1011–1023.

He, S., Zhang, Y., Yang, M., Du, W., Harada, H., 2007. Repeated use of MAPdecomposition residues for the removal of high ammonium concentration fromlandfill leachate. Chemosphere 66, 2233–2238.

Kumar, A., Kumar, S., Kumar, S., 2005. Biodegradation kinetics of phenol andcatechol using Pseudomonas putida MTCC 1194. Biochem. Eng. J. 22, 151–159.

Lee, S.I., Weon, S.Y., Lee, C.W., Koopman, B., 2003. Removal of nitrogen andphosphate from wastewater by addition of bittern. Chemosphere 51, 265–271.

Nelson, N.O., Mikkelsen, R.L., Hesterberg, D.L., 2003. Struvite precipitation inanaerobic swine lagoon liquid: effect of pH and Mg:P ratio and determination ofrate constant. Bioresour. Technol. 89, 229–236.

Sarla, M., Pandit, M., Tyagi, D.K., Kapoor, J.C., 2004. Oxidation of cyanide in aqueoussolution by chemical and photochemical process. J. Hazard. Mater. B 116, 49–56.

Wang, J., Song, Y., Yuan, P., Peng, J., Fan, M., 2006. Modeling the crystallization ofmagnesium ammonium phosphate for phosphorus recovery. Chemosphere 65,1182–1187.

Willmott, C.J., 1982. Some comments on the evaluation of model performance. Bull.Am. Meteor. Soc. 63, 1309–1313.