Standard Article

You have free access to this content

Nanofiltration

Membrane Processes

  1. Bart Van der Bruggen

Published Online: 19 APR 2013

DOI: 10.1002/9781118522318.emst077

Encyclopedia of Membrane Science and Technology

Encyclopedia of Membrane Science and Technology

How to Cite

Van der Bruggen, B. 2013. Nanofiltration. Encyclopedia of Membrane Science and Technology. 1–23.

Author Information

  1. KU Leuven, Leuven, Belgium

Publication History

  1. Published Online: 19 APR 2013

1 Introduction

  1. Top of page
  2. Introduction
  3. Scope of Nanofiltration
  4. Membrane Materials and Synthesis Procedures
  5. Performance for Aqueous Separations and Modeling
  6. Solvent Filtration
  7. Fouling in Nanofiltration
  8. Applications
  9. Conclusions
  10. References
  11. Further Reading

Nanofiltration (NF) emerged as a separate pressure-driven membrane process in the second half of the 1980s, as a low energy purification process for the removal of color, organic matter, and hardness from drinking water sources (1). The low energy consumption for NF should be understood in comparison with reverse osmosis (RO), and refers to the proposed use as “low pressure” or “loose” RO membranes. This, however, is only part of the story. In reality, NF can be either related to RO or to ultrafiltration (UF), as will be demonstrated in this article. The classification as a separate technology emerged in 1988. Cadotte (in the journal Desalination) and Eriksson (in Environmental Progress, a few months earlier) described the new terminology and process scope in that year, using nearly the same words (2, 3). Even though the word “nanofiltration” appeared twice earlier in Desalination less than 3 months before Eriksson's publication, it is reasonable to consider Eriksson as the founding father of NF as a distinct technology. However, the development of the technology itself was rather a gradual process, which resulted from the optimization of RO membranes intended to have lower energy consumption. It was at that time, in 1984 during a meeting at the FilmTec company, that the name “nanofiltration” was chosen for “a RO process that selectively and purposely allows some ionic solutes in a feed water to permeate through” (4). Such more permeable membranes appeared to have features of particular interest for separations, combining partial salt removal with the reduction of organic matter. Nevertheless, the proposed membranes for these applications were not new, although they had been optimized specifically to operate in this range of applications. In fact, Lonsdale (5) gave an overview of membranes available in 1972, which had quite similar features as NF membranes that appeared many years later, even though they were rather denoted as loose RO membranes. At that time, these were still referred to as advanced RO membranes (6). It was because of the improvements in the stability, selectivity, and flux by FilmTec that NF was accepted as a separate unit operation (7). Since then, NF membranes are considered to be those that have pores in the nanometer range. It can be argued that those are really “pores,” but when this definition is related to the rejection of solutes in the same range of around a nanometer, it allows us to roughly define what NF is.

One may wonder whether nanofiltration really deserves indeed to be considered a different process than RO and UF. The reality is that the outcome would not be any different if NF is referred to as open RO or tight UF, in the sense that it would not influence the applications in water purification or water recycling. For solvent filtration, as discussed further in this article, the terminology becomes even more confusing because the size of membrane pores uses water as the reference. One nanometer, the reference size for NF, is approximately the size of a layer of three water molecules. Thus, effects of hydration are critical, and somewhat more or less hydration can alter the separation characteristics totally; when water is replaced by another solvent, it is evident that the notion of “nanopores” becomes somewhat vague, and disputable. For that reason, it is unclear whether NF should be considered a nanotechnology. It has been argued recently (8) that the term nanofiltration strongly suggests that it is a nanotechnological filtration method, but UF is not usually considered as a nanotechnology, although UF membranes contain intentionally engineered structures at the nanometer scale, so that they could logically be treated as nanomaterials as well (8). Thus, the terminology proposed 30 years ago by FilmTec has proven to have been a golden strategic choice, given the boom of nanotechnologies during the last decade. Nevertheless, it is clear that the success of the technology will not be determined by the terminology, but rather by the performance for advanced separation, by the filtration market, and by the decreasing cost of membrane systems.

2 Scope of Nanofiltration

  1. Top of page
  2. Introduction
  3. Scope of Nanofiltration
  4. Membrane Materials and Synthesis Procedures
  5. Performance for Aqueous Separations and Modeling
  6. Solvent Filtration
  7. Fouling in Nanofiltration
  8. Applications
  9. Conclusions
  10. References
  11. Further Reading

NF is, by definition, a multipurpose filtration process. The organic growth from “loose RO membranes” defines a first clear objective for NF membranes: partial rejection of salts. Starting from classical RO membranes, which have ion rejections in the range of 98–99%, the “loose” aspect affects particularly the smaller monovalent ions. These have a smaller hydrated radius, in general, because the charge/polarity interactions are less strong with a single charge. The decrease of sterical hindrance obtained in this way leads to more permeation. In addition, the charge effect also plays directly between ion and membrane. Multivalent ions are retained by an additional repulsion force, higher than for monovalent ions. Typical NF membranes have a negative surface charge at neutral pH, which explains higher rejections of multivalent, negatively charged ions. This, in turn, also influences the rejection of cations, which associate with the anions. Overall, no net charge is transferred through the membrane; permeation (or rejection) of anions also implies permeation (or rejection) of cations. Thus, the rejection of any ion depends not only on the membrane characteristics but also on the composition of the feed mixture. In general, a rejection sequence referred to as Donnan exclusion behavior is obtained. For a positively charged membrane, such sequence would be CaCl2 >  NaCl > Na2SO4; for a negatively charged membrane, the normal sequence is Na2SO4 >  NaCl > CaCl2 (9, 10). Nevertheless, many deviations are found when more complex mixtures are considered, so that this sequence is not always observed. In addition, it should be noted that steric effects and charge effects are strongly intertwined (11). This makes ion rejections in NF even more difficult to understand. A realistic expectation for ion rejections is in the range 90–99% for multivalent ions and 10–90% for monovalent ions.

Solutes with a charge opposite of that of the membrane (counter-ions) are attracted to the membrane, while solutes with a similar charge (co-ions) are repelled. At the membrane surface a distribution of co- and counterions occurs, which causes an additional separation. This is called Donnan exclusion (12). The rejection of ions decreases with increasing salt concentration. This is a typical phenomenon when electrostatic interactions are involved, due to compression of the electrical double layer at the membrane surface.

On the other hand, the rejection of uncharged (organic) solutes is inspired by the typical performance observed for UF. The ability of an UF membrane to retain uncharged solutes is indicated by the molecular weight cutoff (MWCO), which is the molecular mass of a solute that is retained for 90% (13). Making abstraction of the shape of the rejection curve (rejection as a function of molecular size, e.g., molecular mass), which can be sharp or diffuse, solutes larger than the MWCO are thought to be nearly completely rejected, whereas smaller solutes are thought to permeate. This concept is taken over in NF, even though it is even less reliable as a method to characterize the membrane performance than in UF. The separation in NF takes place in the (sub)nanometer range, with permeation rates determined not only by steric hindrance but also by effects of, for example, polarity (14); in general, molecular mass is only one of several descriptors, and is insufficient to characterize a membrane or to allow translation from one solute to another. Nevertheless, MWCO is still typically used to describe a membrane's rejection properties for uncharged solutes. The MWCO values for NF membranes range from circa 150 to 3000; most membranes are at the lower end of this range. This, however, creates the expectation of efficient removal of solutes with molecular mass down to 150. This is not always observed. Since the removal of substances in the molecular mass range between 150 and 300 is extremely important in water purification, mainly for the removal of pesticides and other micropollutants (including endocrine disruptors, pharmaceuticals, personal care products, and chemicals), the low MWCO values have attracted much interest (15-19). In spite of the potential of NF for such applications, it was found that high rejections, corresponding to nearly complete removal, were not always observed. This has led to increased knowledge on the determinants of transport of uncharged solutes in NF (20, 21). In general, problems may arise with small, hydrophobic solutes, which may have a lower rejection; this is particularly important for micropollutants such as N-nitrosodimethylamine (NDMA) and methyl tertiary butyl ether (MTBE), and will be further described specifically for this application. For most micropollutants, however, rejections in the order of 90% can be expected.

3 Membrane Materials and Synthesis Procedures

  1. Top of page
  2. Introduction
  3. Scope of Nanofiltration
  4. Membrane Materials and Synthesis Procedures
  5. Performance for Aqueous Separations and Modeling
  6. Solvent Filtration
  7. Fouling in Nanofiltration
  8. Applications
  9. Conclusions
  10. References
  11. Further Reading

Most NF membranes are polymers; for solvent filtrations, ceramic NF membranes and combined polymeric ceramic membranes are increasingly used as well (22-24). In general, they can be divided into two categories, depending on the material and synthesis method. These two categories can be denoted as “tight” and “loose” NF membranes.

Tight NF membranes are directly related to (loose) RO, and are synthesized in a similar way: they have a composite substructure, typically consisting of, for example, a polysulfone sublayer on a nonwoven support, and a top layer that is typically made of polyamide (PA). The top layer is made by interfacial polymerization, using a procedure in which the sublayer is immersed in an aqueous amine solution, followed by immersion in an organic acyl chloride solution. The reaction of the amine with the acyl chloride yields a thin polymeric layer with excellent separation capacity. The top layer is optimized to be more permeable compared to RO membranes; the concentration of monomers, reaction time, and the choice of monomers determine the eventual membrane performance (25). For RO membranes, this procedure is optimized toward (mainly) high salt rejections; the objective for NF membranes is rather to obtain more permeable membranes (for the solvent, but also for monovalent salts). The result is typically a membrane with performance roughly as shown in Figure 1a: rejections of multivalent ions (e.g., Ca2+ and SO42−) are 99 + %, rejections of monovalent salts (e.g., Na+ and Cl) are between 60% and 90%, and MWCO values are in the range of 200. Apart from PA, also polyimide (PI) membranes, often used in solvent NF, have characteristics in this range.

thumbnail image

Figure 1. Tight (a) versus loose (b) nanofiltration membranes.

In contrast, loose NF membranes do not make use of a top layer made by interfacial polymerization and are more similar to UF membranes. Such membranes can even serve as sublayers for NF membranes with an additional top layer, with higher rejections. Loose NF membranes are made by phase inversion, which involves a controlled transformation of a cast polymeric solution from a liquid into a solid state (24). This can be achieved in four ways: by immersion precipitation (immersion in a nonsolvent bath); by controlled evaporation; by thermal precipitation (when the temperature is lowered); and by precipitation from the vapor phase. Immersion precipitation involves a three-component system consisting of the polymer, a solvent for the polymer, and a nonsolvent. The nonsolvent diffuses into the polymer-rich phase; at the same time, the solvent diffuses to the polymer-lean phase. This process continues until insufficient solvent is left in the polymer-rich phase, leading to the formation of nuclei, which will grow and eventually coalesce, yielding the structure of a NF membrane.

This process can be significantly fine-tuned by considering, for example, the moment of demixing (instantaneous versus delayed demixing), the composition of the casting solution (such as polymer concentration, the choice of solvent and nonsolvent, the use of additives), and the postcasting treatment (the so-called curing of the membrane). This can lead to substantial differences in their structure (and therefore, also in differences in performance). For example, the formation of macrovoids can be favored or suppressed depending on the solvent/nonsolvent exchange and the nuclei that are formed during phase separation.

The typical performance of a membrane synthesized in this way is shown in Figure 1b. The rejection of multivalent ions such as Ca2+ and SO42− is typically between 90% and 99%, whereas monovalent salts such as Na+ and Cl have rather low rejections, typically between 60% and 90%. MWCO values are often in the range of 500–1000 (and higher, in the transition from NF to UF).

Ceramic NF membranes also have an asymmetric structure, consisting of at least three layers with different pore sizes: a macroporous support layer; a mesoporous intermediate layer; and a thin top layer with pores in the (sub)nanometer range. The macroporous support layer provides the mechanical strength of the membrane, and is usually made by sintering of a powder. The most common material for support layers is α-Al2O3, with a pore size of 1 µm or smaller (down to 0.1 µm).

The quality of the support layer determines the eventual quality of the membrane, because defects and irregularities in the support layer often also cause defects in the top layer. To avoid this, the surface roughness should be reduced. This is done by adding a mesoporous intermediate layer, formed by the sol–gel process (26). The sol is a colloidal suspension made by either the colloidal route, in which a precursor is mixed with excess water so that colloids of a few nanometers are formed, or the polymeric route, in which a precursor is mixed with a much smaller amount of water in an organic solvent; this yields branched polymeric molecules.

For the intermediate layer, the colloidal route is usually followed. Oxides of Al, Si, Ti, or Zr are used; Ti and Zr oxides are more stable than Al or Si oxides. Intermediate γ-Al2O3 layers are often found, but they are stable only at a pH between 3 and 11, while TiO2 and ZrO2 hardly corrode, and are also stable at lower and higher pH values.

Owing to the excess water, oxides are hydrolyzed to form colloids. Hydrolysis occurs as follows:

  • mathml alt image

The hydroxides further undergo a condensation reaction:

  • mathml alt image

Stabilization of the colloids is achieved by the addition of HNO3 or HCl to avoid aggregation. Addition of the intermediate layer is carried out by dip coating, that is, dipping into the dispersion. Drying below 100 °C yields a gel, which forms a network structure of particles. Organic binders such as polyvinylalcohol (PVA) are used to prevent cracks resulting from asymmetric drying (parallel to the support layer).

Finally, the membrane is calcinated; the calcination temperature is the main parameter that determines the membrane's pore size (27). In case of TiO2 layers, the desired phase is anatase, which is obtained during thermal treatment between 250 and 500 °C. At temperatures below 250 °C, an amorphous structure is formed, with little corrosion resistance. Above 500 °C, a phase transformation from anatase to rutile occurs. This yields larger pores, and is unfit for addition of a top layer. ZrO2 is quite similar: at room temperature an amorphous phase is obtained, whereas at higher temperatures a transition occurs to a tetragonal monoclinic or cubic phase. The desired tetragonal phase is obtained for calcinations temperatures between 360 and 550 °C. At 550 °C, there is a transition to the monoclinic phase, which has a more open structure, similar to Ti.

The formation of a thin top layer is carried out in the same way as the intermediate layers, but polymeric sols are used. A three-dimensional network is formed by the interpenetration of polymeric structures. This eventually leads to pore formation. For polymeric sols, the calcination temperature is even more crucial because of the separation requirements of the top layer. A moderate calcination temperature (e.g., 300 °C) usually yields the best results.

4 Performance for Aqueous Separations and Modeling

  1. Top of page
  2. Introduction
  3. Scope of Nanofiltration
  4. Membrane Materials and Synthesis Procedures
  5. Performance for Aqueous Separations and Modeling
  6. Solvent Filtration
  7. Fouling in Nanofiltration
  8. Applications
  9. Conclusions
  10. References
  11. Further Reading

Characterization of the membrane pore structure, such as pore radius, pore density, pore shape, pore length, tortuosity, and so on, is very important in view of understanding the process; therefore, characterization methods are essential to support the interpretation of solute transport, membrane fouling, and so on. Several characterizing methods have been applied, both based on direct instrumental observation and experimental methods. Instrumental methods include contact angle measurement, atomic force microscopy (AFM), Fourier transform infrared (FT-IR) spectroscopy, and scanning electron microscopy (SEM); these will be explained in a different article, to which the reader is referred.

Experimental observation is often applied as a direct characterization method for the membrane performance. It is usually accepted that the rejection of uncharged (organic) molecules is determined by the size of the dissolved molecules compared to the size of the membrane pores (28, 29). As stated above, other physicochemical effects such as dipole interactions may also play a role. Nevertheless, all models to describe the rejection of organic molecules that have been proposed are based on the sieving mechanism and neglect other interactions. These models make use of a parameter representing the size of the molecule (or a related parameter such as the diffusion coefficient), and a method to account for pore size distribution or steric hindrance. Rejections can be predicted, but the accuracy can be low when components are used that interact strongly with the membrane or cause fouling.

The volume flux (JV) and solute flux (JS) can be described as

  • mathml alt image(1)
  • mathml alt image(2)

in which c is defined as

  • mathml alt image(3)

The solute transport through the membrane is thus determined by three parameters: the water (hydrodynamic) permeability LP, the solute permeability ω, and the reflection coefficient σ. Other symbols in the equations refer to the pressure difference (ΔP), the osmotic pressure difference (Δπ), and the concentration in the permeate (cp) and in the feed (cf).

When testing the pure water flux (Δπ = 0) at different pressures, the permeability can be obtained as the slope of the volume flux as a function of the operating pressure, as shown in Figure 2.

thumbnail image

Figure 2. Calculation of the permeability as the slope of the volume flux as a function of the operating pressure.

A higher LP indicates that the membrane has a more loose structure; this may also compromise the rejections of charged and uncharged solutes.

We can rewrite the expression for the solute flux as

  • mathml alt image(4)

where Δc is the concentration difference between the feed and the permeate and c is the mean logarithmic concentration.

From Equation 4, a graphical relationship to estimate ω and σ can be obtained (Fig. 3).

thumbnail image

Figure 3. Graphical relationship to estimate ω and σ for a given nanofiltration membrane.

Equation 2 suggests that transport of uncharged molecules is a combination of diffusion and convection. It can be rewritten as

  • mathml alt image(5)

When this equation is used as a macroscopic approximation, c should be the averaged value as given by Equation 3. When used in a differential form, c is a variable that can be solved when taken together with Equation 1. These are the transport equations of Spiegler and Kedem (30) for water flux and for the flux of a dissolved component. Diffusion is represented by the first term in Equation 5; the second term represents the contribution of convection to the transport of uncharged molecules.

The retention of a given molecule can be calculated from Equations 1 and 5 as

  • mathml alt image(6)
  • mathml alt image(7)

where R is the rejection of the considered solute; Jv is water flux [l/(h m2)]; P is the solute permeability [l/(h m2)], and σ is the reflection coefficient. In this way, the solute rejection R can be calculated as a function of the water flux Jv and the solute permeability P. The permeability P is a measure of the transport of a molecule by diffusion through the membrane.

From Equations 6 and 7, it can be seen that the rejection increases with increasing water flux and reaches a limiting value σ at an infinitely high water flux. The reflection coefficient σ of a given component is the maximum possible rejection for that component, at a theoretical infinite pressure (or infinite water flux). The reflection coefficient can be measured experimentally or derived mathematically (31). The resulting curve for the reflection coefficient as a function of the molecular diameter (rejection curve) can be used to estimate the maximum rejection that can be obtained by a given membrane.

Experimental observation of rejections of known solutes as a function of increased pressure allows us to determine the parameters in the Spiegler–Kedem equations. This in turn allows us to estimate indirectly the radius of the membrane pores, assuming ideal pores (identical, cylindrical, and perpendicular to the membrane surface), by applying idealized models such as the steric-hindrance pore (SHP) model (32). The pore radius is calculated using the following equation:

  • mathml alt image(8)

with

  • mathml alt image(9)
  • mathml alt image(10)
  • mathml alt image(11)

HF is a “wall-correction parameter” that represents the effect of the pore wall, and SF is a parameter that represents steric hindrance during transport through the pore. The solute radius and the pore radius are symbolized by rs and rp, respectively.

During transport, these molecules encounter a certain amount of steric hindrance and interactions with the pore wall. A molecule that is smaller than the diameter of the membrane is partially retained through these effects. A molecule with the same size as the pore diameter is completely retained.

Combining Equations 8-11, the following expression is obtained:

  • mathml alt image(12)

In the SHP model, the reflection coefficient is calculated from the pore size of the membrane and the diameter of the molecule. It is assumed that all the pores have the same size. Therefore, the uniform pore size should not be interpreted as a physical pore diameter. The calculated pore size corresponds to the pore size of an imaginary membrane with uniform pores, for which the rejection of uncharged molecules is equal to the rejection obtained for the real membrane. In reality, not every pore has the same cylindrical diameter; thus, the model is an idealization of the membrane's structure.

Other models than the SHP model can be applied, which allow the incorporation of the effects of charges as well. The space charge (SC) model (33), the Teorell–Meyer–Sievers (TMS) model (34), and the electrostatic and steric-hindrance (ES) model (35) derived from the SHP model and the TMS model can all be used to predict the transport performance of charged solutes through NF membranes based on their charged pore structure. All these are idealized models, and allow the interpretation of ion rejections to some extent.

Other models use a log-normal distribution of the pore diameters for the calculation of the reflection coefficient as a function of the solute size. In the log-normal (LN) model (31), no steric hindrance in the pores or hydrodynamic lag is taken into account, and the value of σ (reflection coefficient) reflects the fraction of membrane pores that are smaller than the molecules in solution. The equation that calculates the reflection coefficient with a molar radius r* is

  • mathml alt image(13)

This equation involves two variables, Sp and r, of which Sp is the standard deviation for the pore size distribution. This standard deviation is a measure of the distribution of pore sizes. As the retention curve corresponds to an integrated LN distribution, a small “Sp” represents a large slope of the retention and a large “Sp” represents a small slope. r is the size of molecule that has 50% rejection and is therefore a measure for the “average” pore size.

Although the molecular mass is not a direct measure of the dimensions of a molecule, it still reflects the molecular size, and it is a readily accessible parameter, whereas complicated calculations are necessary to obtain a reliable measure for the solute size (represented by effective molecular diameter, molecular width, or similar parameters). However, the LN model can be adapted by taking the correlation between molecular mass (or molecular weight) and the diameter of the molecule into account (Fig. 4). This relation was derived for the Stokes diameter and can be used for the effective diameter in the LN model. The proposed correlation is dS = A(MW)B, where A and B are empirical constants. For NF range solutes, A = 0.065 and B = 0.438.

thumbnail image

Figure 4. Correlation between molecular weight and the effective diameter used as solute size parameter in the LN model (36). Source: Reprinted from Reference 31, Copyright (2002), with permission from Elsevier.

The equations of the LN model can be written as

  • mathml alt image(14)

In this equation, σ (MW) is the reflection coefficient of a molecule for a given membrane, SMW is the standard deviation, and inline image is the average molecular weight where the retention is 50%.

This can be further extended by replacing inline image by the MWCO (inline image):

  • mathml alt image(15)

5 Solvent Filtration

  1. Top of page
  2. Introduction
  3. Scope of Nanofiltration
  4. Membrane Materials and Synthesis Procedures
  5. Performance for Aqueous Separations and Modeling
  6. Solvent Filtration
  7. Fouling in Nanofiltration
  8. Applications
  9. Conclusions
  10. References
  11. Further Reading

Solvent filtration, commonly known as organic solvent nanofiltration (OSN) or SRNF, is a relatively new application area for NF.

Although separation of solvents by means of a membrane was already proposed in the 1960s by Sourirajan (37) and Loeb (38), SRNF is a new technology that emerged in the early years of the twenty-first century. The main reason for its later development compared to other membrane processes is the lack of stability of membrane materials in contact with the solvent. However, the development of SRNF was catalyzed by its potential uses in the industry. Demands from industries to replace units such as distillation and crystallization by more (energy) efficient units were an incentive to improve the performance of NF membranes and their resistance toward harsh organic solvents (39).

The understanding of solvent and solute interactions with membrane materials is crucial for membrane design, performance evaluation, and prediction of SRNF.

Polymer–solvent interactions can be estimated, for example, by the Flory–Huggins theory for polymer solutions. Some of the parameters of solute, solvent, and membrane that determine fluxes and rejections in SRNF include (i) solvent properties, such as the polarity and the viscosity of the solvent, and the molar volume (size) of the solvent; (ii) solute properties, such as the molar volume (size) of the solute and its solubility in the solvent (measure of solute–solvent affinity) and in the membrane; and (iii) membrane properties, such as the surface energy of the membrane, hydrophilic/hydrophobic interactions with the solvent, and the degree of cross-linking of the membrane (resistance to swelling) (40).

Although transport processes in aqueous NF systems have been studied for several years and much knowledge has been gained, SRNF systems are not yet well understood. While some studies support the use of pore-flow models, others suggest using a solution-diffusion approach. Bhanushali et al. (41) suggested that a pore-flow model including an interaction parameter between the membrane and the permeating species could qualitatively describe SRNF rejection data. Machado et al. (42, 43) developed a resistance-in-series model and suggested that solvent transport consists of three main steps: (i) transfer of the solvent into the top active layer, which is characterized by a surface resistance; (ii) viscous flow through pores; and (iii) viscous flow through support layer pores, all expressed by viscous resistances, that is,

  • mathml alt image(16)

where f1 and f2 are solvent-independent parameters characterizing the sublayers, φ is a solvent parameter, γm is the critical surface tension of the membrane material, and γl the surface tension of the solvent. The surface resistance is proportional to surface tension difference between the solvent and the top layer, and viscous resistances are proportional to the solvent viscosity.

Darvishmanesh et al. (44) extended this approach by proposing a coupled series–parallel resistance model for transport of solvent through inorganic NF membranes.

In the solution-diffusion model, it is assumed that each permeating molecule dissolves in and diffuses through the membrane phase in response to a concentration gradient. There is no pressure gradient inside the membrane, and, based on the Stefan–Maxwell equation, the following equation can be derived (36) for the molar flux of each species through the membrane: the solvent flux Jv and the solute flux Js according to solute diffusion transport mechanism are given by

  • mathml alt image(17)
  • mathml alt image(18)

where k1 and k2 are the solvent and the solute permeability coefficients. In real membranes with imperfections and a microporous structure, the solution-diffusion mechanism might not be valid. The solution-diffusion model has often been used to describe the permeation of organic solvents through polymeric membranes (45-50). The transport and separation mechanism through dense membrane films (specifically poly(dimethyl siloxane)—PDMS) is generally described by the solution-diffusion model, originally developed by Lonsdale et al. (51) and recently updated by Paul (52) with parameters taken from both the pore-flow and the solution-diffusion model. Bhanushali et al. (41) suggested that solvent viscosity and surface tension are the dominant factors controlling solvent transport through NF membranes, and a solution-diffusion approach was proposed to predict pure solvent permeation. Stafie et al. (45) used the solution-diffusion model to describe sunflower oil/hexane and polyisobutylene/hexane permeation through a composite PDMS membrane with poly(acrylonitrile) (PAN) support. White (50) investigated the transport of normal and branched alkanes and aromatic compounds through a series of asymmetric PI membranes. These examples show that a solution-diffusion model may indeed be reliable, although results are not necessarily better than for the pore-flow approach.

In different industries, SRNF membranes have a very high potential to achieve difficult separations at a low (energy) cost. SRNF has the potential to replace or to complement (in a hybrid approach) traditional separation units such as distillation, evaporation, crystallization, adsorption, or extraction. Bioproducts made by fermentation fit in well with this hybrid approach. NF can be also very useful in those processes where low thermal stress is allowed. SRNF is investigated in different fields in the pharmaceutical industry (53, 54), in the food industry, and in the (petro)chemical industry. Applications include the recovery of solvent in lube oil dewaxing processes (55), the reuse of extraction solvent in the food industry (56-58), the recovery of homogeneous catalyst in chemical synthesis (59), decolorization of waste streams, and purification of pharmaceutically active ingredients (60).

6 Fouling in Nanofiltration

  1. Top of page
  2. Introduction
  3. Scope of Nanofiltration
  4. Membrane Materials and Synthesis Procedures
  5. Performance for Aqueous Separations and Modeling
  6. Solvent Filtration
  7. Fouling in Nanofiltration
  8. Applications
  9. Conclusions
  10. References
  11. Further Reading

One of the most often applied models to describe flux decline is the resistance-in-series model (61). For NF, the water flux is written as

  • mathml alt image(19)

where ΔP is the pressure difference (bar), η is the viscosity (Pa s), and Rtot is the total resistance against mass transfer through the membrane.

Experimental flux decline values correspond to an increase in the total resistance against mass transport. Different mechanisms of flux decline can be distinguished (62, 63). Adsorption inside the pores or at the membrane surface narrows the pores. When the molecules have a similar size as the pores, permeation can lead to pore blocking, a phenomenon that can be enhanced or caused by adsorption. Pore blocking has been observed for UF, where macromolecules are filtered, and may occur in a similar way for NF.

The total resistance Rtot is the sum of different individual resistances, that is, Rtot = Rp + Ra + Rm + Rg + Rcp + Ri + Rd (Rp, resistance due to pore blocking; Ra, resistance due to adsorption inside the pores; Rm, intrinsic membrane resistance; Rg, resistance caused by the formation of a gel layer; Rcp, concentration polarization resistance; Ri, resistance caused by specific interactions; Rd, resistance from deposits on the membrane). This is shown schematically in Figure 5.

thumbnail image

Figure 5. Mechanisms contributing to the total resistance toward mass transport.

In the ideal case, for example, filtration of pure water, the membrane resistance (Rm) is the only resistance involved. This is an intrinsic membrane characteristic that corresponds to the resistance calculated from, for example, the Hagen–Poiseuille equation and does not change during filtration or by changing the feed solution. It reflects the minimal resistance of the system against mass transport and thus determines the maximum water flux at a given pressure. The other phenomena can only make pores narrower (or the membrane thicker), resulting in an increase of the total resistance or the addition of an extra resistance term to the intrinsic membrane resistance.

The gel layer resistance, the adsorption resistance, the pore blocking resistance, the deposition resistance, and the concentration polarization resistance depend strongly on the type of feed solution that is used. The formation of a gel layer resistance is not likely, as this is typically related to macromolecules, which are not often present in NF applications. After rinsing and mechanical cleaning of the membrane, the water flux is largely restored (assuming that the gel layer formation does not involve adsorption as well).

Concentration polarization, the change in concentration in the film layer at the membrane surface, is usually not considered a fouling phenomenon, although it is closely related. It is an effect that is inherent to any membrane separation, but it is always reversible and disappears when the flow conditions are adjusted. Concentration polarization is often minimized by increasing the cross-flow velocity above 1 m/s; particularly for laboratory-scale tests this is common practice. For large-scale applications, however, it should be taken into account by estimating the concentration increase at the membrane surface (64). This concentration increase is given by

  • mathml alt image(20)

where J is the water flux that is obtained for pure water; it is a constant for a given membrane and is not subject to further change. Rint is the internal rejection of the membrane (i.e., with the membrane concentration as reference, not the bulk concentration), not to be confused with the transport resistances. The mass transfer coefficient k can be related to the Sherwood number (Sh) or to the Reynolds number (Re) and Schmidt number (Sc):

  • mathml alt image(21)
  • mathml alt image(22)
  • mathml alt image(23)

The Reynolds number can be influenced by changing the feed velocity. Increasing the Reynolds number by increasing the feed velocity corresponds to a decrease of the boundary layer δ and thus to a change from laminar to turbulent flow pattern. When a turbulent flow pattern is obtained, concentration polarization will have only a minimal effect, due to the constant velocity profile. For undisturbed flow through a straight pipe, the change from laminar to turbulent flow occurs at a Reynolds number of about 2000.

For NF, pore blocking and adsorption have the most significant impact on the membrane's permeability, and should therefore be considered the main causes of fouling in NF (in the absence of biofouling, which may have significant consequences; it is not considered here since it is a more general phenomenon, not only relevant in NF). Molecules can attach to the membrane pores or to the membrane surface by adsorption or chemisorption. Inside the pores, they narrow the free pathway for the water flow, decreasing the net pore opening. From the Hagen–Poiseuille equation, it is clear that this should lead to a flux decline. When adsorption has a strong effect, it could even lead to pore blocking when the entire cross-section of the pore is filled by a molecule.

When fouling is caused by irreversible adsorption or chemisorption, a stable (although lower) flux may be expected once all adsorption sites in the membrane pores or at the surface are occupied and an adsorption equilibrium is established. This effect is usually irreversible, because desorption of the organic material is difficult.

Pore blocking, on the other hand, involves no physicochemical interaction and is thus in principle reversible. Nevertheless, this is generally not the case in practice since solutes can be stuck deep inside the membrane structure, from where they can never be removed. Furthermore, there usually is a certain degree of adsorption as well.

Finally, osmotic pressure also influences the water flux. This phenomenon is fully reversible and is not a fouling effect. However, for NF it can be problematic, particularly for highly concentrated product streams such as solutions with high salt concentrations, which are in part retained by the NF membranes.

7 Applications

  1. Top of page
  2. Introduction
  3. Scope of Nanofiltration
  4. Membrane Materials and Synthesis Procedures
  5. Performance for Aqueous Separations and Modeling
  6. Solvent Filtration
  7. Fouling in Nanofiltration
  8. Applications
  9. Conclusions
  10. References
  11. Further Reading

7.1 Application in Drinking Water Production

Historically, drinking water production was the first objective of NF, and today it is still the most significant application. Although the difference with “loose RO” might not be always so clear, the first reports on the use of NF appeared in the late 1980s and the early 1990s, in which it was emphasized that such membranes can remove color, natural organic matter, and hardness in one step (1, 65). The potential for using NF membranes as a barrier against micropollutants (66) and as a protection method against occurrence of disinfection by-products (DBPs) (67) was also recognized in an early stage, although this aspect still received less attention. The production plant in Méry-sur-Oise near Paris was probably the first large-scale NF plant that was successfully operated (68). Understanding and controlling membrane (bio)fouling rapidly emerged as the number 1 challenge for NF (69, 70), although the scientific community also paid much attention to understanding and modeling the membrane performance (71-73). Since then, many other applications emerged, such as production of potable water from sources contaminated with arsenic (74, 75), partial nitrate removal (76), removal of dissolved uranium (77), and, particularly, removal of pesticides (78-80). Van der Bruggen and Vandecasteele (81) reviewed explored and unexplored applications of NF for drinking water production in 2003, and this overview is still relevant and fairly complete, with the exception of the removal of emerging pollutants (17, 82-84). Since then, NF further matured as a technology, which resulted mainly in a broader range of commercially available membranes, increased technical expertise with plant operation, and better insights into solute–membrane interactions. The dominant applications, however, have remained the same.

While most applications remain the same, much progress has been achieved in improving membrane materials to allow better performance, particularly to increase fluxes, membrane stability, and fouling resistance, while keeping rejections at the same level. Ceramic NF (85) is an attractive option, but also polymeric membranes have improved significantly (86). Furthermore, NF is increasingly applied in hybrid configurations in view of synergetic effects. For example, the combination of NF and oxidation (using, e.g., ozone) has attracted the attention (87-89). Other examples include a hybrid coagulation- NF (90), the combination with catalytic reaction, and an anion exchange resin (91), a membrane bioreactor (92), and adsorption (93).

7.2 Wastewater Treatment and Water Recycling

The use of NF for wastewater treatment is more challenging because the risk of membrane fouling is much larger and because the separation is often more difficult since the concentrations are higher. Furthermore, the membrane resistance can be problematic when, for example, wastewater has extreme pH values. Wastewater purification in the pulp and paper industry was among the first nonpotable applications for NF (94, 95). Other industries came much later: the first reports appeared only in the second half of the 1990s. These, however, were very diverse and considered a very broad range of industries: cutting oil wastewater (96), textile processing effluents (97), tannery wastewater (98) landfill leachates (99), electroplating rinse water (100), and diary industry effluents (101) are only a few examples of trials with NF on laboratory scale or pilot scale.

Gradually, the scope of NF changed from wastewater treatment to water recycling. Food processing was considered by Mavrov and Belieres (102), and was found to allow very large savings in water consumption by reusing NF permeates. Other than the food industry, the textile industry (103), the plating industry (104, 105), and the pickling and tanning industry (106) are promising candidates for water recycling using NF. Furthermore, secondary effluents are also attractive for reusing after NF (107, 108). This is a clear tendency in the use of NF. Remarkably, the cited wastewaters are among the most demanding applications, typically with high concentrations of organics combined with high salinity. NF results in a permeate of sufficient quality for reuse, even in extreme conditions of feed quality. This gives NF the reputation of a process that works when all other alternatives fail. Nevertheless, solutions for the concentrate fraction have to be developed as well, and options for concentrates rather than the performance of the NF unit itself may be decisive for a successful application of using NF for wastewater recycling.

7.3 Nanofiltration for Fractionation

Fractionation of wastewater or process water can be used for recovery of valuable products such as phosphates. In a first approach, phosphates can be removed from wastewater to avoid excessive algae growth and deoxygenation of the water (109). However, phosphorus-containing compounds have the potential that they can be used as a fertilizer in agriculture or in aquaculture. Phosphorous-containing products can also be used as a substitute for phosphate rock (110). Given the depletion of phosphate resources, which is expected within the next 100–400 years, phosphate recovery is probably necessary. In worst case scenarios, about 40–60% of the current resource base would be extracted by 2100 (111). Demands for phosphorus have steadily increased during the last decades, because of an increase per capita income, to population growth, and global trends leaning toward more meat- and dairy-based diets. Phosphorus cannot be manufactured or synthesized and has no substitute in food production. Selective fractionation of phosphates is therefore of interest, and NF is one of the few processes that allow this (112).

Another well-known fractionation application for NF (and UF) is to concentrate and to purify cheese whey proteins (113). Similarly, a mixture containing the homologous series of cyclodextrins CD(6) to CD(60) obtained by enzymatic conversion of starch was considered fractionation using NF (114). Saccharides have been studied as well: galacto-oligosaccharides (115), glucose and raffinose (116), commercial oligosaccharide mixtures (117), and xylose and glucose (118). Vanneste et al. (119) proposed the concept of integrated countercurrent NF cascades, based on the work of Caus et al. (120), for three different sugar separations: raffinose–sucrose, fructose–glucose, and xylose–glucose. The proposed concept goes much beyond diafiltration, which is the standard configuration for fractionation of saccharides, and can in principle be applied for any other fractionation purpose as well.

In addition to saccharides, peptides also offer a large potential. NF can be used to fractionate peptides in P-lactoglobulin tryptic hydrolysates (121), for whey-derived peptides (122), and to fractionate the small peptides found in a rapeseed protein enzymatic hydrolysate (123). Thus, such applications for fractionation appear to have a huge potential, with numerous separations to be carried out, each one different but feasible.

7.4 Nanofiltration Challenges

It is interesting to extrapolate the state of the art in NF to further progresses to be made in the coming decades. These are challenges for researchers and process developers, and should stimulate broader industrial application when successful solutions are found. Van der Bruggen et al. (124) outlined several of these challenges, of which some are specific for NF and others are more general for many, if not all, membrane processes. With some creativity, they can be logically derived from the current state of the art. Fouling has already been discussed in this article as a priority challenge in NF. In spite of many studies on this topic, it remains difficult not only to control but also to predict. Among the various types of fouling, the most difficult ones in NF appear to be organic fouling due to adsorption and biofouling. Adsorption is closely related with the membrane performance itself. When solutes are retained as they should be, their concentration at the membrane surface is increased. This favors adsorption on the membrane, or even in the inner structure of the membrane, due to diffusive mechanisms. It has been shown (125) that this can even lead to breakthrough of solutes. This often comes as a surprise since it takes time for the membrane to become saturated, especially when concentrations are low (as is the case for micropollutants). Adsorption makes the membrane more hydrophobic: the effect is clearly cumulative, and caused by (changing) hydrophobicity. It was shown that the rejection of triazines, for example, is influenced by membrane fouling (126); membranes turned more hydrophobic by adsorption of humic substances. One approach is to use more hydrophilic membranes through modifications, which could suppress this effect (127). However, fouling cannot be totally avoided in this way; some modifications are also questionable on longer time scales. A more sustainable approach could be the development of in situ engineered solutions.

Work on membrane synthesis is particularly needed to increase the membranes' lifetime and chemical resistance. In aqueous applications, chlorine resistance is a specific challenge. Chlorine is used in most water filtration applications, and could be a solution against biofouling. This is not compatible with the chemical resistance of the membrane. Even though the ultimate solution would be in nonoxidative in situ disinfection methods, the search for more resistant membranes is important as well. This, of course, also applies to solvent filtration, where the membrane's chemical resistance determines the success of the application.

The performance of NF membranes to reject solutes is generally good. Nevertheless, some compounds are still problematic, such as small chemicals and pharmaceuticals. Classical examples are MTBE, NDMA, ibuprofen, and some other pharmaceuticals. Such compounds can easily slip through the structure of a NF membrane, and should be closely followed.

This relates to the need for modeling and simulation tools to control fluxes and rejections in NF. Expertise in NF is to a large extent empirical, and, although it is possible to develop quantitative tools (as shown previously), this is not generally applied yet. Engineering would be obviously easier with uniform calculation tools that can be used for all membranes and all manufacturers. This would also make new applications possible in which NF membranes are used as separation tools rather than as purification tools. The chemical engineering approach of integrating processes in cascades with recycle streams and a tuned separation performance is only possible when a single membrane separation can be reliably calculated, based on a limited set of membrane parameters, which should be available for any NF membrane.

A final challenge, shared with RO, is the further treatment of the concentrate fraction. Evidently, this is a general membrane challenge, because the concentrate fraction is intrinsic to any membrane separation, but the volume, composition, and discharge possibilities differ significantly from process to process. NF, in particular, has a concentrate challenge, because volumes are large, and the composition (often containing high concentrations of salts and recalcitrant organics) prevents easy solutions. Boosting the process recovery is a possible way to achieve this; this entails the integration with other processes such as oxidation or adsorption (128, 129), which is in fact an integrated concentrate treatment.

8 Conclusions

  1. Top of page
  2. Introduction
  3. Scope of Nanofiltration
  4. Membrane Materials and Synthesis Procedures
  5. Performance for Aqueous Separations and Modeling
  6. Solvent Filtration
  7. Fouling in Nanofiltration
  8. Applications
  9. Conclusions
  10. References
  11. Further Reading

During the three decades since its conception, NF has matured from a drinking water production process with a somewhat vague profile, closely related to both UF and loose RO, to an all-round technology with applications ranging from the original drinking water production to demanding wastewater purification processes as well as fractionation of waste and process streams. The latter is possibly the largest growth market for NF, also in view of the added value of the streams resulting from fractionation. This is an application quite different from the classical purification, in which the ambition is to decrease the concentration of a wide range of components below the applicable standard. For fractionation, two fractions have to be considered (permeate and concentrate), with a double requirement for each: product recovery and product yield. Membranes with narrower pore size distributions could do this, but engineered solutions using integrated membrane cascades can also be applied. Current NF membranes make use of a limited number of materials (at least on commercial scale), and much progress is expected here to develop tailored NF membranes for specific applications. This includes SRNF, which is another area with much potential for growth in purification of high added value products. Furthermore, fouling-resistant materials could complement the need for pretreatment and fouling control. Apart from improved polymers, much is expected of hydrophilic ceramic NF membranes for aqueous applications, and hydrophobized ceramic NF membranes for filtration of organic solvents.

Models for NF have been developed for aqueous filtration, and even for solvent filtration. Translation of these models into generally applicable simulation tools is the progress to be expected in the coming decade; this would allow a reasonable estimation of the membrane performance, without the need of extensive preparatory studies.

References

  1. Top of page
  2. Introduction
  3. Scope of Nanofiltration
  4. Membrane Materials and Synthesis Procedures
  5. Performance for Aqueous Separations and Modeling
  6. Solvent Filtration
  7. Fouling in Nanofiltration
  8. Applications
  9. Conclusions
  10. References
  11. Further Reading

Further Reading

  1. Top of page
  2. Introduction
  3. Scope of Nanofiltration
  4. Membrane Materials and Synthesis Procedures
  5. Performance for Aqueous Separations and Modeling
  6. Solvent Filtration
  7. Fouling in Nanofiltration
  8. Applications
  9. Conclusions
  10. References
  11. Further Reading
  • Schäfer AI, Fane AG, Waite TD, editors. Nanofiltration: Principles and Applications. Oxford: Elsevier; 2005.
  • Van der Bruggen B, Geens J. Nanofiltration. In: Li NN, Fane AG, Winston Ho WS, Matsuura T, editors. Advanced Membrane Technology and Applications. Hoboken (NJ): John Wiley & Sons; 2008. p 271296.
  • Van der Bruggen B, Mänttäri M, Nyström M. Drawbacks of applying nanofiltration and how to avoid them: a review. Sep Purif Technol 2008;63:251263.
  • Van der Bruggen B, Vandecasteele C.. Environ Pollut 2003;122(3):435445.