0011-9164/03/$– See front matter © 2003 Elsevier Science B.V. All rights reserved

Desalination 159 (2003) 119–129

Nanofiltration of CIP waters from iodine X-ray contrast mediaproduction: process design and modelling

Anja Drewsa*, Thorsten Klahma, Berit Renka, Mahmut Saygilib, Goetz Baumgartenc,Matthias Kraumea

aInstitut für Verfahrenstechnik, TU Berlin, Sekr. ACK 7, Ackerstr. 71-76, D-13355 Berlin, GermanyTel. +49 (30) 314 72687; Fax +49 (30) 314 72756; email: anja.drews@tu-berlin.de

bSchering AG, Müllerstr. 178, D-13353 Berlin, GermanycAmafilter Deutschland GmbH, Am Pferdemarkt 11, D-30853 Langenhagen, Germany

Received 22 January 2003; accepted 12 February 2003

Abstract

CIP (cleaning in place) waters from X-ray contrast material production cause considerable disposal costs as fordischarge of iodine-containing effluents; a legal limit of only 1 ppm AOX has to be met. Therefore, CIP waters whichtypically contain more than 1000 ppm are currently incinerated, thus requiring a huge amount of energy. A two-stagenanofiltration process consisting of an enrichment and a purification step was designed for separating contrast agentsout of rinsing waters. Required areas for both stages were exemplarily calculated. Rejection was greater than 99% attransmembrane pressures of 20–40 bar and resulting fluxes of up to 200 L/(m²h). Besides designing an economicallyfeasible nanofiltration process which yields a permeate with concentrations below the legal limit and a retentate ofmuch less volume for incineration, the aim of this project was to further application of mathematical model descriptionsof NF transport phenomena and to find an appropriate design equation. It could be shown that in this case concentrationpolarisation or gel layer formation were of minor importance. As adsorption appeared to cause the principal resistanceto both solvent and solute flux, parameters of an extended solution–diffusion model were fitted in such a way that theyaccount for this physical effect.

Keywords: Nanofiltration; Iodine X-ray contrast medium; Process design; Modelling; Adsorption

1. Introduction

CIP waters from X-ray contrast agentsproduction give rise to considerable disposalcosts. The standards for wastewater discharge in

*Corresponding author.

Germany are set under the Federal Water Act [1]whereby any discharge of wastewater requires alicence. The German water laws are a collectiveterm for the Länder and federal regulations. Thecompetence as such is with the Länder. In thecase of iodine-containing effluents, the limitingvalue is not fixed but subject to individual assay.

A. Drews et al. / Desalination 159 (2003) 119–129120

(1)

(2)

(3)

For the manufacturing plant of concern in thisstudy it was set to 1 ppm AOX (i.e., 3.6 ppmAOI). CIP waters from X-ray contrast agentsproduction which typically contain more than1000 ppm are often incinerated, thus requiring ahuge logistic and expenditure of energy. Due totheir application, contrast media have to be bio-logically inert and are therefore practically non-biodegradable [2]. Alternative treatment pro-cesses include UV-promoted destruction of themolecules leading to unknown by-productswhich, in turn, may cause environmentalproblems. Evaporation, on the other hand, ishighly energy consuming. Therefore, a purelyphysical process with lower energy consumptionshall be employed.

By screening for suited membranes at varyingoperating parameters such as temperature,transmembrane pressure and retentate concen-tration, a membrane filtration process was to bedesigned to separate contrast agents out of rins-ing waters. The aim was to find an economicallyfeasible enrichment factor and to design ananofiltration (NF) process that yields a permeatewith concentrations below the legal limit and aretentate of much less volume for incineration. Inorder to further physical understanding ofmembrane transport phenomena and to find adesign equation, a suitable model description wassought. Besides molecular weight polarity andcharge of the molecules are important parametersinfluencing interaction with the membrane andtherefore retention [3]. The possibility ofdescribing NF by a universal model has beendiscussed by Straatsma et al. [4] but was found tobe unlikely. They set up a NF model based on theMaxwell–Stefan transport equations but con-cluded that fitted parameters lost their physicalmeaning as they contradicted literature values butinstead compensated for phenomena not includedin the model. It has to be noted that they did notconsider concentration polarisation.

In contrast, a semi-empirical approach wasemployed here. For prediction of flux and

rejection this requires only little information suchas feed concentration and transmembrane pres-sure and thereby enables easy practical appli-cation. On the other hand, it only holds for theinvest igated membrane/solute/solventcombination. The influence of concentrationpolarisation is discussed.

2. Theory

Transport through RO and NF membranes isoften described by the solution–diffusion model(SDM). For dilute salt solutions, this consists oftwo equations for transport of solvent or per-meate P and solute j through the membrane [5,6].

(with since ).

Since each solvent/solute/membrane combi-nation is different, parameters A and B need to befitted to experimental data. As the basic SDMonly describes transport through the membraneand does not consider additional resistances, itgenerally overestimates flux. In addition tosolution and diffusion, however, other pheno-mena such as concentration polarisation, gel-layerformation or adsorption can occur, which makefurther modifications necessary.

2.1. Concentration polarisation

Concentration polarisation arises as a concen-tration gradient is established when solutemolecules are retained at the surface of themembrane. Assuming steady state and Fick’s law,a differential mass balance yields [5,6]:

A. Drews et al. / Desalination 159 (2003) 119–129 121

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

Applying the film theory [7], the unknownthickness * can be replaced by the mass transfercoefficient k which can be calculated fromcommon Sherwood correlations.

For spiral wound modules, Schock [8] suggestedthe following equation (see also [9,10] for acomprehensive review of Sh correlations):

Besides viscosity and diffusitivity, geometricalfeatures of the feed channel and spacer such asporosity g and hydraulic diameter dh must beknown to apply this equation. These, however,may not always be readily available.

As will be seen, permeate concentration cjP isnegligible in this case, thereby reducing Eq. (3) to

2.2. Adsorption

Adsorption onto surfaces is commonlydescribed by Langmuir or Freundlich equationswhich express the amount of adsorbed moleculesin terms of the surrounding concentration. Here,

a Freundlich isotherm is used:

3. Experimental

3.1. X-ray contrast media

X-ray contrast media are injected during X-rayexamination to provide visual contrast ofotherwise non-absorbing tissues and organs.Absorption of X-rays is facilitated by elements ofhigh atomic number such as iodine and barium.Derivatives of 2,4,6-triiodobenzoic acid are themost commonly used X-ray contrast agents. Theycan be of either ionic or non-ionic form withmolecular weights ranging from approximately600–1600 g/mol [11]. In 1996, 360 t of iodineX-ray agents were prescribed in Germany, 90%of which were non-ionic [2].

CIP waters from different contrast agents (ofboth ionic and non-ionic forms) production linesare typically collected in a common storage tank.In order to minimise the number of experiments,one representative substance each was selectedfor screening experiments. In addition, experi-ments were carried out with mixtures.

Concentration was measured using UV-spectroscopy (Specord 200, Analytik Jena). At245 nm, the relationship between absorbance andconcentration was found to be linear up to35 mg/L. Samples containing more were accord-ingly diluted with de-ionised water. All concen-trations are given in g iodine/L.

3.2. Membrane screening

Seven NF and six reverse osmosis membranesas listed in Table 1 were tested. It should be notedthat this classification was made according tomanufacturers terminology. Of those, three

A. Drews et al. / Desalination 159 (2003) 119–129122

Table 1List of tested membranes

Membrane type Manufacturer

NF:Desal 5 DK OsmonicsDesal 5 DL OsmonicsCK OsmonicsNitto LES 90 NittoNTR 7450 NittoN 30 NadirNF PES 10 NadirRO:Nitto LF 10 NittoTrisep X20 TrisepUTC 80 TorayDesal SG OsmonicsDesal SE OsmonicsDesal SC Osmonics

were selected for further investigations in whichtransmembrane pressure, temperature and reten-tate concentration were varied between 20 and40 bar, 20 and 35°C, and 0 and 25 g/L, respect-ively. Membrane screening was conducted in atest cell (effective membrane area 44 cm²,Amafilter GmbH) at cross flow velocities of2.3 m/s. For each run a new sheet was used andsome experiments were repeated to ensurereproducibility of results. Retentate and permeatewere recycled into the surge tank to avoid time-varying feed concentrations. As feed flow rateswere much higher than permeate flow rates, theconcentration was assumed to be constant overthe length of the membrane, too (cjF . cjR). Fluxwas measured on-line by weighing on a digitalscale and recording the signal over time.

3.3. Module experiments

To represent more technically realisticcircumstances, 2.5" spiral-wound modules (widespacer, 1.77 m²) were utilised for obtainingdesign data. In screening experiments, NF

membranes Desal 5DK and 5DL (both byOsmonics) showed the best performance(compare 4.1.) and were therefore chosen formodule experiments. Both are made of polyamideand have a negative surface charge.

Here, too, retentate and permeate were re-cycled into the surge tank. By stepwise increasesin contrast agent content in the feed, themaximum possible retentate concentration whichstill allows an economically feasible permeateflux was determined. Operating parameters suchas temperature (20–35°C), transmembrane pres-sure (20–30 bar), and retentate concentration (0–70 g/L) were varied. Mathematical modelling wasbased on module results.

4. Results4.1. Screening

The non-ionic agent was identified as themore critical substance for membrane filtration asits rejection was lower than that of ionicsubstances. Therefore, it was chosen as a modelsubstance for further investigations. Fig. 1 showsmembrane performance at 40 bar and 20°C.Almost total rejection of contrast media could beachieved by use of the NF membranes Desal5DK, 5DL and Nitto LES 90. Desal 5DK exhi-bited the highest rejection rates of more than99.9%. Desal 5DL showed slightly lower rejec-tion but permeability, i.e., flux at equivalent pres-sures, was generally higher. For pure substances,similar results were achieved using Nitto LES 90,but here rejection decreased for mixtures ofdifferent contrast media. Unlike expected, thesepreliminary investigations proved that all testedRO membranes were less suited for the specifictask as they not only showed lower permeate flowrates but also lower retention than the threeselected NF membranes. Satisfactory rejectionwas achieved by the best RO membrane (NittoLF 10), but flux at the same pressure was lowerthus resulting in a higher energy consumption toreach the same goal.

A. Drews et al. / Desalination 159 (2003) 119–129 123

Fig. 1. Exemplary screening results.

Fig. 2. Two-stage NF-process.

As temperature increased rejection dropped inall cases due to elevated solute diffusion rates(data not shown). A 20-fold increase in feedconcentration did not impair membrane perfor-mance. It was observed that rejection increasedwith feed concentration.

4.2. Process design

On the basis of module experiments a two-stage process was designed where contrast agentsare concentrated in the first stage and the legallimit is met in the second (see Fig. 2). Assuminga feed concentration of 1000 ppm iodine and arejection of 99.9% (compare Fig. 1, Desal 5DK),the outlet requirements are safely met by this set-up. Even drops in rejection to 99.5%, e.g., due toa decreased flux in technical modules, should beeasily dealt with.

Required membrane areas were determined onthe basis of the derived model equations for 20°C

on the one hand and based on measuredseparation characteristics (Fig. 3) for 35°C on theother. Fig. 3 reveals a common relationshipbetween flux and feed concentration. As concen-tration was increased, flux decreased due toelevated osmotic pressure, concentration polari-sation or other resistances to permeation. Athigher temperatures these resistances declined asmolecular movement was increased. Fig. 3 alsoshows that to ensure an economically feasibleflux, cjR should not much exceed 60 g/L.

Both separation characteristics and modelequations (see 4.3) resulted from short-termmeasurements. Long-term experiments revealedthat during operation flux decreased down toapproximately 3/4 of the original value due tofouling (data not shown). Therefore, 70% of fluxat given conditions was used to estimate therequired areas. Considering also that these resultswere gained using 2.5" modules, flux was halvedto represent technical circumstances. In larger

A. Drews et al. / Desalination 159 (2003) 119–129124

Fig. 3. Separation characteristic of Desal 5 DK atdifferent sets of operating parameters.

modules flux is lower due to longer permeateflow channels and a resulting rise in pressuredrop which reduces the driving force. Fig. 4 givesa summary of calculated areas for an assumedfeed flow rate of 500 L/h for different sets ofoperating parameters. Results are given as per-centages of the required area at 20 bar and 20°C.

Apart from temperature and transmembranepressure, retentate concentration cjR can be varied.A rise in cjR results in decreased flux and hencelarger area requirements, but permeate yield isincreased (Y = 96.7% at cjR = 30 g/L and 98.3% atcjR = 60 g/L). On the other hand, as flux declinesretention drops (compare Fig. 6a), in the worstcase to such an extent that permeate con-centration exceeds the legal limit. Therefore, anoptimisation of operating conditions should becarried out on the basis of costs for the following:C transport of concentrate to incinerationC incineration/re-use/reclamation of iodineC membrane areaC energy

while always ensuring that the second-stagepermeate concentration falls short of the legallimit. To be on the safe side, process designshould be performed for a transmembrane

Fig. 4. Calculated minimum areas for different operatingconditions as percentages of required area at 20 bar and20°C, cjF = 1 g/L; = 500 L/h.

pressure well below the module’s maximumoperating pressure. This way a pressure increasecan still be carried out to counteract a severe fluxdecline due to changing specifications (e.g.,rising retentate concentration).

4.3. Modelling

In the following, modifications to thesolution–diffusion model are derived which holdfor transport of a non-ionic contrast mediumthrough Desal 5DK at 20°C. Initially, focus islaid on flux, and later on rejection to round up thederived model.

4.3.1. FluxFig. 5a shows measured flux over trans-

membrane pressure at different feed concen-trations. In contrast to the commonly assumedconstant hydraulic permeability, a non-linearrelationship can be observed even when pure (de-ionised) water is used. This might be caused bymembrane compression or by an increasedpressure drop on the permeate side at higher fluxand a resulting decrease in driving force. In any

A. Drews et al. / Desalination 159 (2003) 119–129 125

(12)

(13)(14)

(15)

Fig. 5a. Flux over transmembrane pressure.

case A is not constant but depends on )p, whichleads to a first modification of Eq. (1):

where )B = 0 for pure water. The correlation forA()p) takes the form typical for cake filtrationwith compressible cake or, as in this case,compressible filtration medium:

In Fig. 5a, Eqs. (12) and (13) are plotted withfitted parameters A0 = 6.7968 kg/(m²hbar0.8727)and m = !0.1273. As expected, higher feedconcentrations gave rise to an increasedresistance. To identify this concentrationinfluence a dimensionless flux wasdefined in such a way that permeate flux at agiven pressure is divided by the pure water flux atthe same pressure. Plotting this dimensionlessflux over transmembrane pressure reveals analmost constant relationship for each feedconcentration for pressures greater than approxi-

Fig. 5b. Dimensionless flux over transmembrane pres-sure; curves calculated from Eq. (14).

mately 5 bar (see Fig. 5b), i.e., can now bedescribed by multiplying pure water flux by afactor C (cjF) only depending on feed concen-tration. In other words, at )p o )B the permea-bility decrease with pressure is merely caused bythe decreasing hydraulic permeability.

For low pressures, however, the otherwiseconstant dimensionless flux decreases rapidly,highlighting the influence of osmotic pressurewhich can reach values of up to a few bars forhigher concentrations and thus reduces thedriving force considerably in this region.

Eq. (14) is plotted in Fig. 5b with C(cjF) fittedfor each concentration and )B(cjF) calculated bythe van’t Hoff equation, which proved to beapplicable.

A. Drews et al. / Desalination 159 (2003) 119–129126

(16)

4.3.2. RejectionTo achieve a satisfactory fit of the complete

model Eq. (2) for solute flux must be modified,too. In Fig. 6a full lines represent cjP as calculatedfrom Eq. (2) with B = 0.013 L/(m²h), where agood match for values at 20 g/L is found.Permeate concentrations at higher feed concen-trations, however, are overestimated when thisvalue is used (smaller cjF vice versa). This meansthat B, too, depends on cjF.

Apart from the over- and underestimation ofpermeate concentration B = const erroneouslypredicts falling rejection with increasing feedconcentration (see Fig. 6b). For cjF o cjP, cjF canbe cancelled from the equation for rejection:

Thus cjF only influences , which leads to adecrease of rejection with rising feed concen-tration. This confirms that B, too, must depend on

(a)

feed concentration. This dependency must beexpressed in such a way that resistance againstsolute transport rises with increasing feedconcentration.

To discuss the influence of concentrationpolarisation, the following module geometry [8]and typical material properties are assumed:

dh = 1.39 mmdch = 1.51 mmg = 0.81l = 0.9 m< = 10!6 m2/sD = 1.5@10!9 m2/s

With the experimental flow rate of =1400 L/h, the mass transfer coefficient can nowbe calculated from Eq. (5), yielding k = 3.4@10!5

m/s. According to Eq. (10), e.g., at JP = 50 L/(m²h), a concentration increase of 50% at themembrane surface occurs for any given feedconcentration. The relevant concentration forsolute transport through the membrane now is cjMand Eq. (2) changes into:

(b)

Fig. 6. (a) Permeate concentration over flux for different feed concentrations with B = 0.013 L/(m²h). Full lines, withoutconcentration polarisation; dotted lines, with concentration polarisation. (b) Rejection over feed con-centration with B =0.013 L/(m²h).

A. Drews et al. / Desalination 159 (2003) 119–129 127

(17)

(18)

(19)

(20)

Eq. (17) implies a linear relationship between cjMand cjP as shown in the following equation:

(with since cjM o cjP).Therefore, concentration polarisation at JP =

50 L/(m²h) yields a 50% increase in permeateconcentration for each feed concentration, too.This resulting shift towards higher permeateconcentrations at increasing flux (see dotted linein Fig. 6a), however, means falling resistancewith rising flux and does not explain theconcentration dependency of B.

Another possible effect is adsorption whichwas observed in an experiment specificallycarried out to detect its presence. Feed concen-tration was suddenly increased from 0 to 0.5 g/L.At this low level osmotic pressure is negligible.This was done at a low level of )p (here 5 bar) toadditionally avoid the effect of concentrationpolarisation. A flux decline can now only beattributed to adsorption [12,13]. As can be seen inFig. 7, flux decreased from 25.8 to 25 L/(m²h).

This observation led to the idea of repre-senting C(cjF) by a term analogical to aFreundlich adsorption isotherm [Eq. (11)]. Intro-ducing this into Eq. (14) yields a commonequation for solvent flux with adsorptionresistance. The factor aN, however, is not identicalwith the Freundlich coefficient a as otherwisezero flux would result for maximum adsorption,which would defy any physical explanation.

As discussed earlier, it was observed thatsolute flux did not decrease linearly with feed

Fig. 7. Flux decline during adsorption experiment.

concentration as suggested by Eq. (2) but thatinstead the solute permeability B decreased withrising feed concentration. A physical explanationcould be that adsorbed molecules screen offfurther molecules due to their polarity. In thiscase, solute permeability would be inverselyproportional to the amount of adsorbed mole-cules. This led to the following implementation ofthe same Freundlich-type term:

In Fig. 8, Eqs. (19) and (20) are plotted withfitted parameters:

A0 = 6.7968 kg/(m²hbar0.8727)aN = 0.0014 (mg/L)!0.524

B0 = 0.00007 L/(m2h)m = !0.1273n = 0.524

Experimental data are well represented by thederived model equations. This means that astransport can be modelled with an implementedFreundlich-type equation, adsorption indeedseems to cause a significant mass transferresistance to both solvent and solute flux.

A. Drews et al. / Desalination 159 (2003) 119–129128

(a) (b)

Fig. 8. Results of the complete solution–diffusion model including adsorption, Eqs. (19) and (20). (a) Permeateconcentration over flux for different feed concentrations. Full lines: without concentration polarisation, model using cjF;dotted lines: with concentration polarisation, model using cjM from Eq. (10) and k = 3.4@10!5 m/s. (b) Rejection over feedconcentration.

Including cjM in the model (dotted lines) yieldsa better match of the curve shape but as discussedearlier requires knowledge of the mass transfercoefficient k.

These equations, however, only hold for T =20°C. For rising temperatures, the derived modelwould erroneously predict decreasing fluxbecause so far it contains temperature only in theosmotic pressure term. Osmotic pressure riseswith increasing temperature leading to a higherresistance especially at low transmembrane pres-sures. Temperature-dependency of both diffusionand adsorption has so far not been implemented.

5. Conclusions

A two-stage NF process consisting of anenrichment and a purification step was designedfor cleaning of CIP-waters from iodine X-raycontrast agents production. Rejection of solutedecreases with decreasing flux and feed concen-tration but was always greater than 99%. Mass

transport through the membrane can be describedby an extended solution–diffusion model. Theeffect of concentration polarisation is discussed.To account for a significant resistance againstboth solvent and solute transfer an adsorptionisotherm was implemented and parameters fittedto represent experimental data. In order to findout if adsorption really prevails, the adsorptionisotherm should be measured and compared withresults gained here. Also, further experiments atdifferent temperatures are necessary to validatethis assumption and to account for temperature-sensitive processes such as adsorption anddiffusion within the model.

6. Symbols

a — Freundlich coefficient, (mg/L)-n

aN — Adsorption factor, (mg/L)-n

A — Solvent permeability, kg/(m2 hbar)A0 — Coefficient in Eq. (13), kg/(m²hbar1+m)B — Solute permeability, L/(m²h)B0 — Solute permeability, L/(m²h)

A. Drews et al. / Desalination 159 (2003) 119–129 129

c — Concentration, kg/m3

C — Factor in Eq. (14)d — Diameter, mD — Diffusion coefficient, m2/sJ — Volumetric flux, L/(m2h)k — Mass transfer coefficient, m/sl — Length of membrane cushion, mm — Exponent in Eq. (13)

— Mass flux, kg/(m2h)n — Freundlich exponent)p — Transmembrane pressure, barR — RejectionU — Gas constant, J/(mol K)Re — Reynolds numberSc — Schmidt numberSh — Sherwood numbert — Time, sT — Temperature, °Cw — Velocity, m/sY — Yield

Greek

" — Dimensionless membrane area* — Film thickness, mg — Spacer porosity< — Kinematic viscosity, m2/sB — Osmotic pressure, barD — Density, kg/m3

> — Mass fraction, kg/kg

Subscripts

ads — Adsorbedch — Channeleff — EffectiveF — Feedh — Hydraulicj — SoluteM — Membrane surfacemax — MaximumP — Permeate, solventR — Retentatew — Water

References

[1] Federal Water Act (Wasserhaushaltsgesetz, WHG),Federal Law Gazette I p. 632, 2000.

[2] S. Wischnack, A. Putschew and M. Jekel, IodierteRöntgenkontrastmittel in Abwässern, Wässern undTrinkwasser, Colloquium ProduktionsintegrierteWasser-/Abwassertechnik, GVC, Bremen, 2000.

[3] B. van der Brugge, J. Schaep, D. Wilms and C. Van-decasteele, Influence of molecular size, polarity andcharge on the retention of organic molecules bynano-filtration, J. Membr. Sci., 156 (1999) 29–41.

[4] J. Straatsma, G. Bargeman, H.C. van der Horst andJ.A. Wesselingh, Can nanofiltration be fully pre-dicted by a model? J. Membr. Sci., 198 (2002)273–284.

[5] R. Rautenbach, Membranverfahren, Springer-Verlag,Berlin, 1997.

[6] M. Mulder, Basic Principles of Membrane Tech-nology, 2nd ed., Kluwer Academic, Dordrecht, 1996.

[7] R.B. Bird, W.E. Stewart and E.N. Lightfoot, Trans-port Phenomena, Wiley, New York, 1960.

[8] G. Schock, A. Miquel, Mass transfer and pressureloss in spiral wound modules, Desalination, 64(1987) 339–352.

[9] V. Gekas and B. Hallström, Mass transfer in themembrane concentration polarisation layer underturbulent cross flow. Part 1. Critical literature reviewand adaption of existing Sherwood correlations tomembrane operations, J. Membr. Sci., 30 (1987)153– 170.

[10] V.S. Minnikanti, S. DasGupta and S. De, Predictionof mass transfer coefficient with suction for turbulentflow in cross flow filtration, J. Membr. Sci., 157(1999) 227–239.

[11] W. Krause, H. Miklautz, U. Kollenkirchen andG. Heimann, Physicochemical parameters of X-raycontrast media, Investigative Radiology, 29(1)(1994) 72–80.

[12] B. Goers, Spülwassermanagement undTensidrückge-winnung in Mehrproduktanlagen mitMembran-verfahren. PhD Thesis, TU Berlin, 2000.

[13] A. Maartens, P. Swart and E.P. Jacobs, Charac-terisation technique for organic foulants adsorbedonto flat sheet UF membranes used in abattoireffluent, J. Membr. Sci., 119 (1996) 1–8.