Chapter 3 MEMBRANE FILTRATION REVIEW In this chapter, membranefiltration in water treatment is reviewed. The aim is to assess the current status and revealgaps in knozdedge from the wealth of literature. The backgrozmd on models and principles is summa;rised for the relevant processes; micrOJiltration WF)) ultrajltratioa (UF)) and nanoJltration @F). Reverse osmosis is bri&) considered to put NF, which is often desclibed as aprocess (%I between" UF and RO) in perspective. After a brief description of membrane materidls, membrane rejection and fouling wilI be addressed. Both rejection of and fouling b_y natural organics and inorganic colloids, wd be a major focus of this work. A further issue is the characteriration of clean and fouled membranes as well as fouling control. The last sections describe membrane application isszles in water treatment. The processes have been compared in terns of their volume of application and recent growth. This is obviou~b linked to treatment cost, an issue which will also be addressed bn.64. -Problems which have amkn in previous pilotplant or full scale studies will be pad $thefouli~g stzldies in this thesis, where efects can be imestigated on a smaller scale. Issues of concentrate disposal or treatment and membrane integm$ are not discussed in this review. The concluding remarks address research needs andplansfor this project. Copyright © 2001 by Andrea I. Schafer 40 MEMBRANE FILTRATION REVIEW 3.1 INTRODUCTION AND OVERVIEW There are many processes available for water treatment. Process selection depends on the required water quality, and therefore whch solutes or particles are to be retained. Of course the treatment cost also plays a major role in process selection. Unfortunately, environmental criteria - such as reduction of chemical addition or alternative operation modes, whlch allow the use of alternative energes - are, at best, only indirectly considered in cost evaluations which precede process selection. Conventional physico-chemical treatment involving addition of coagulants and sand filtration, competes with membrane separation processes, but often fails in the treatment of waters containing large amounts of natural organic matter. In Table 3.1, an overview of common processes as well as the sizes of solutes and particles of interest is presented. Table 3.1 Overview oftrtatmentprocesses and sol~te/partde dimensions (Cheryan (1986), Agbekodo (1994)). Copyright © 2001 by Andrea I. Schafer Introduction and Overview 41 As can be seen, membrane separation processes cover the entire size range, from suspended solids to mineral salts and small organics. Membrane processes also compete with some other processes such as activated carbon, ion exchange and to some extent coagulation and fdtration. Of the process options considered, microfiltration @F) is the membrane process with the largest pores. It is generally used for waters of htgh turbidity, and low colour or organics content. MF can remove bacteria and "turbidity". MF is also a common pretreatment process for NF and RO. The fact that MF pores are relatively large allows cleaning methods, such as air backflush or permeate backwash, whtch remove deposits from pores and surface. Ultrafiltration (UF) has only recently been recognised in water treatment and is becoming increasingly popular due to its ability to remove turbidity, microorganisms, and viruses, especially when issues such as Giardia and Cvptospon'ditlm are of concern (Jacangelo et al. (1995a)). The removal of lssolved organics is limited with UF. Nanofiltration (NF) is a relatively new process, though whle the number of applications is growing rapidly, the transport mechanisms are still poorly understood (Raman et al. (1994)). NF shows a high selectivity between mono- and multivalent ions. Its popularity in water treatment stems from its softening abilities and high organics (and micropollutant) rejection. Reverse Osmosis (RO) is used primarily in desalination, or for waters where rnicropollutants are difficult to remove with other processes. R0 removes both mono- and multivalent ions. However, for surface waters no full demineralisation is usually required and NF is more economic at a similar organics removal. Pressure driven membrane processes do not retain lssolved gases such as CO a (Rohe et al. (1990)) and some taste and odour compounds. Copyright © 2001 by Andrea I. Schafer 42 MEMBRANE FILTRATION REVIEW In ths section, the main models for membrane processes are summarised. This allows a basic understandmg of rejection and deposition principles and underlines the importance of certain parameters in the different processes. The four membrane types, MF, UF, NF, and RO, are considered in separate sections. Table 3.1 illustrates that the separation between the different processes is not precise, as the processes overlap. Therefore, filtration and separation models are generally applicable to more than one process. Often several phenomena are operative simultaneously and which one dominates depends on the membrane and the solute or particle in question. Concepts such as the resistance-in-series model, the osmotic pressure model or concentration polarisation are principles whch are applicable to any membrane operation. These will be described in the MF section. Rejection (R,) is defined by equation (3.1). This definition is the apparent rejection calculated from the bulk concentration CB and the permeate concentration cl], for sample i. The true membrane rejection is higher due to concentration changes in the boundary layer. However, the values of concentration in the boundary layer are not accessible. The most critical parameter in the characterisation of membranes is their flux. For the characterisation of clean membranes flux is measured with MilliQ water as 'pure water flux'. The definition of the instantaneous flux is given in equation (3.2), where V is the filtrate volume, t the filtration time, and A the membrane surface area. Alternatively the hydrodynamic permeability (Lv) can be used to describe water throughput. This parameter is very useful when different processes or transmembrane pressures are to be compared, as it is normalised bp the transmembrane pressure AP. Both, flux and rejection tend to vary with time. The underlying mechanisms are described below by a summary of models for each process. Some models apply to several processes and others only to a particular process under certain conhtions. The application of models requires caution as membrane- solute interactions will depend on many factors. These include solute size, charge and morphology; membrane pore size, charge, surface roughness and chemical characteristics; solution chemistry; and, hydrodynamics, whch influence permeation drag, shear forces, and cake compaction. A surface water system is complex and cannot easily be explained by simplified models, especially when solute-solute interactions are poorly understood. Nevertheless, the awareness of existing models is essential to recognising trends and to develop model extensions and improvements. Copyright © 2001 by Andrea I. Schafer Fundamental Principles and Mechanisms 3.2.1 Microfiltration (MF) Rejection Mechanisms Physical sieving is believed to be the major rejection mechanism for MF with water convecting through the membrane due to an applied transmembrane pressure. The deposit or cake on the membrane can act as a self-rejecting layer, and retain even smaller particles or solutes than would be expected to be removed given the pore size of the membrane ("dynamic membrane"). Thus a fouled MF membrane may have UF rejection characteristics and flux may decline significantly due to the build-up of this deposit. Electrostatic interactions, dispersion forces, and hydrophobic bonding may pla~7 some role in rejection. Little is known about effects such as particle adhesion, deposit compressibility, particle shape, and particle mixtures. Filtration Models Pure water flux under lanlinar condtions through a tortuous porous barrier may be described, according to Carman (1 938) and Bowen and Jenner (1 995), by equation (3.4). AP is the transmembrane pressure difference, q rhe dynamic solvent viscosity, and RAI the clean membrane resistance (i.e. the porous barrier). Units of the symbols are explained in the symbols section at the end of this thesis. The Resistance in Series Model describes the flux of a fouled membrane. Ths is given in equation (3.4). The resistances Rm, RP and RC denote the addtional resistances which result from the exposure of the membrane to a solution containing particles or solute. RCp is the resistance due to concentration polarisation, RP the internal pore fouling resistance, and RC the resistance due to external deposition or cake formation. These resistances are usually negligible in RO, where the osmotic pressure effects become more important (Fane (1997)). However, the osmotic pressure can also be incorporated into RCP. The Osmotic Pressure Model, as shown in (3.6), is an equivalent description for macromolecules according to LVijmans et a/. (1985). All is the osmotic pressure difference across the membrane. The osmotic pressure difference can usuallj~ be neglected in MF and UF, since the rejected solutes are large and their osmotic pressure small. However, even polymeric solutes can develop a significant osmotic pressure at boundary layer concentrations (Ho and Sirkar (1 992)). Thls naturally implies that the resistance in series model (equation (3.4)) would be more appropriate in MF, whlle the osmotic pressure model (equation (3.6)) may be more useful in NF and RO. Both models have been applied to UF. Copyright © 2001 by Andrea I. Schafer 44 MEMBRANE FILTRATION REVIEW Reversible flux decline can be reversed by a change in operation conditions, and is referred to as concentration polarisation. Irreversible fouling can only be removed by cleaning, or not at all. Irreversible fouling is caused by chemical or physical adsorption, pore plugging, or solute gelation on the membrane. Concentration Polarisation is the accumulation of solute due to solvent convection through the membrane and was first documented by Shenvood (1965). It appears in every pressure driven membrane process, but dependng on the rejected species, to a very dfferent extent. It reduces permeate flux, either via an increased osmotic pressure on the feed side, or the formation of a cake or gel layer on the membrane surface. Concentration polarisation creates a high solute concentration at the membrane surface compared to the bulk solution. This creates a back diffusion of solute from the membrane which is assumed to be in equilibrium with the convective transport. At the membrane, a laminar boundary layer exists (Nernst type layer), with mass conservation through this layer described by the Film Theory Model in equation (3.7) (Staude (1992)). c[: is the feed concentration, Ds the solute diffusivity, CBI, the solute concentration in the boundary layer and x the &stance from the membrane. A schematic of the concentration profiles and the mass balance leadmg to equation (3.7) is shown in Figure 3.1, where 6 is the boundary layer thickness. After integrating with the boundary conditions c = cw for X = 0 and c = CI, for X = 6. for similar solute and solvent densities, constant diffusion coefficient, and constant concentration along the membrane, equation (3.7) can be derived. The wall concentration which determines adsorption is cw, gel formation or precipitation, and ks the solute mass transfer coefficient as defined in equation (3.59, Copyright © 2001 by Andrea I. Schafer Fundamental Principles and Mechanisms D.,. where k. r- S Concentration polarisation can be minimised with turbulence promoters on the feed side of the membrane, such as spacers or introduction of crossflow. The Gel Polarisation Model is based on the fact that at steady state flux reaches a limiting value, where increases in pressure no longer increase the flux. According to the Gel Polarisation model, at this limiting value, the solubility limit of the solute in the boundary layer is reached and a gel formed. For 100% rejection, the expression for this limiting flux (Jrm) is described by equation (3.10). cc is the gel concentration, beyond which the concentration in the boundary layer cannot increase. The model does not include membrane characteristics, and tends An improvement can be achieved in using Ds for the gel layer to predict a lower flux than observed. rather the bulk solution (Bowen and Jenner (1995)). McDonogh et al. (1984, 1989) molfied this model and included charge effects. Bacchin et al. (1995) included effects of pH and ionic strength on surface interactions. Belfort et al. (1994) proposed five stages of fouling. These are, (1) fast internal sorption of macromolecules, (2) build-up of a first sublayer, (3) build-up of multisublayers, (4) densification of sublayers, and (5) increase in bulk viscosity. The fifth stage can be neglected for dilute suspensions like surface water. The dependence on particle size can be described as dparticlc ? dporc: deposit on pore walls, restricting pore size dpartic~e - dporr: pore plugging or blockage dprriciclc ) dporc: cake deposition, compaction over time. For particles much smaller than the membrane pores, internal deposition eventually leads to the loss of pores. Particles of a similar size to the membrane pore will cause pore blockage. Particles larger than the pores will deposit as a cake, with the porosity depending on a variety of factors including particle size dmribution, aggregate structure and compaction effects. The process of small particles adsorbing in the pores may be a slow process compared to pore plugging, where a single particle can completely block a pore and therefore flux decline should be more severe for the latter case. Hermia (1982) introduced the Filtration Laws, which aim to describe fouling mechanisms. The models are valid for unstirred, dead-end filtration (deposition without cake dmurbance due to shear and no gravity settling) and complete rejection of solute by the membrane (but obviously allowing pore penetration). Under conditions where permeate drag dominates, the effect of stirring may be negligible. The The and constant pressure filtration law is shown in equation (3.1 1). d't - = k [S) nv2 basic equation leads to four filtration models have been derived by Hermia (1 982). By plotting t/V Exp(t) over filtration time t and volume V, it is possible to determine which filtration mechanism is Copyright © 2001 by Andrea I. Schafer 46 MEMBRANE FILTRATION REVIEW dominant. According to Bowen et a/. (1995), all mechanisms occur in a complete filtration experiment either successively or superimposed due to pore and particle size size dtstributions. The Complete Blocking Model @ore blocking) is valid for particles which have a very similar size to the pores. The particles seal the pores filtration law can be written as d't - d~' and do not accumulate on each other. The constant pressure k l which, on integration, gives V = Jo (l - e- ) (3.1 3) where Jo is the initial flux. The Standard Blocking Model (Fore Constriction Model) describes pore blocking for particles that are much smaller than the pores. Particles pass through the pores and deposit on the surface of the pores. The pore volume will decrease proportionally with the filtrate volume. The Intermediate Blocking Model describes long term adsorption. Every particle reaching a pore will contribute to blockage and particles accumulate on each other. Again, the modified constant pressure filtration law is d't and the integration kV= ln(l+kt Jo) (3.1 7) The Cake Filtration Model describes the filtration of particles whch are much larger than the pores and will be retained, without entering the pores. The particles deposit on the membrane surface contributing to the boundary layer resistance. Included in this model is deposition due to concentration polarisation. d't dv2 - Another model, known as the Solids Flux Model, was developed bp Belfort et al. (1994). Ths was proposed for sticky particles, whch do not backmigrate from the membrane to the bulk solution and cause irreversible fouling. The constant b describes the characteristics of the sublayer and 0s the solids volume fraction in the feed. Copyright © 2001 by Andrea I. Schafer Fundamental Principles and Mechanisms 47 In the filtration of aqueous solutions, all of these models may be combined and their importance in the overall filtration behaviour map change over time. The particle size has a strong influence and only very little is known about the filtration of mixtures where a variety of particle sizes and shapes are present in solution. In most publications, single filtration laws are considered, while very little work has been done on the coupling of different processes. 3.2.2 Ultrafiltration (UF) UF can be used to remove colloids and macromolecules. UF can be used as a pretreatment to NF or RO, whch may lengthen the filtration cycle of these processes compared to a MF pretreatment. Rejection Mechanisms As in MF, physical sieving is an important rejection mechanism in UF and convection dictates solvent passage. The deposit can also act as a self-rejecting layer and charge interactions, as well as adsorption, may play an important role. Rejection is usually evaluated with macromolecules of dfferent molecular weights, such as dextrans or proteins, which leads to the determination of a molecular weight cut-off (MYVCO). Filtration Models The Mechanical Sieving Model (Ferry) suggests hindered transport of solute due to convection, limited by steric effects (Braghetta (1995)). Rejection is determined by the ratio of solute macromolecular dameter to pore diameter, h. R=[A(~-A)]~ for ~<l (3.21) R=l for A21 (3.22) The model does not account for solute velocity drag, diffusional limitations, or concentration effects at the membrane surface. The Modified Sieving Rejection Model (Munch et al: (1979)), accounts for the double layer thickness surroundmg a charged solute whtch leads to a modfied L. Th~s double layer thickness, or Debye length, K-' will affect the packing of colloids on a membrane (McDonogh (1984)). E is the dielectric constant, k~ the Boltzmann constant, T the absolute temperature, z the ion valence, e the fundamental electron charge, N,\ the Avogadro constant and CS the electrolyte concentration. The double layer thickness is strongly influenced by the solution ionic strength. Copyright © 2001 by Andrea I. Schafer 48 MEMBRANE FILTRATION REVIEW The Pore Flow M,odel uses the Hagen-Poiseuille Equation to describe solvent flow through cylindrical pores of the membrane. No membrane characteristics other than pore size or pore density are accounted for, and neither limitation of flux due to friction nor Qffusion is considered. Flux occurs due to convection under an applied pressure. The equation is derived from the balance between the driving force pressure and the fluid viscosity, which resists flow (Braghetta (1995), Staude (1992)). Solvent flux 0) is described bp equation (3.26) and solute flux us) by equation (3.27), where rp is the pore radius, n,, the number of pores, z the tortuosity factor, Ax the membrane thckness and o the reflection coefficient. The flow rate is predicted to be proportional to pressure and proporuonal to the fourth power of pore radius. Two mechanisms were proposed for solute transport, physical sieving and equilibrium partitioning between solute in pores and outside pores. Bhattacharjee and Datta (1996) predicted mathematically that the resistance due to solute backtransport was responsible for flux dedine, whereas osmotic pressure, as well as cake and gel formation were negligible. Rosa and dePinho (1994) used different sized organics to model mass transfer resistance as a function of pore size distribution. Transport for the relatively high concentrations was typical for pore flow (steric and hydrodynamic forces) and good agreement between model and experimental data was achieved. Huisman et al. (1997) studied the effect of temperature and ionic strength on UF membrane resistance. Temperature showed no effect, although the permeability increased with ionic strength. 'Ihs was attributed to lower zeta potentials and thnner double layers - thus electroviscous effects. In Chapter 6, additional models covering filtration through cakes will be described. 3.2.3 Nanofdtration (NF) NF is a process located between UF and RO. Some authors refer to NF as charged UF (Simpson et al. (1987)), softening, low pressure R0 (Rohe et al. (1990)), or do not distinguish at all between NF and RO. NF is generally expected to remove 60 to 80% of hardness, >90% of colour, and all turbidity. The process has the advantage of low operating pressures compared to RO, and a high rejection of organics compared to UF. Monovalent salt is not retained to a significant extent, however this is not normally required in water treatment of surface water. Rejection of membranes is usually evaluated by the manufacturer with NaCl or MgS04 solutions, as opposed to a MWCO specification as in UF. Rejection Mechanisms Both, charge and size are important in NF rejection. At a neutral pH most NF membranes are negatively charged, whle they might be positively charged at low pH (Zhu et al. (1 995), Peeters (1 997)). The principal transport mechanisms of NF are depicted in Figure 3.2. Copyright © 2001 by Andrea I. Schafer [...]... as described in Chapter 2 3. 4.2 Ultrafiltration (UF) Rejection of natural organics by UF membranes has been discussed briefly in the natural organics characterisation and size fractionation by UF section of Chapter 2 The MWCO ranges from 0.5 to 30 0 kDa in UF and t h s governs retention of natural organics Hagmeyer et al (1996) reported that DOC removal varied between 26 and 37 % for UF in long term operation... inhibition by natural organics was recognised as a possibility requiring further investigation Overall, there appears to be a definite lack of research in the area of natural organics effects on inorganic precipitation, CO-precipitationof inorganics and natural organics, as well as the precipitation of calcium-organic complexes The limited Copyright © 2001 by Andrea I Schafer Fouling by Natural Organics. .. Rejection of Natural Organics and Colloids 57 The rejection of both MF and UF can be increased by an appropriate pretreatment (see pretreatment section) This raises the question of whether substantial organics removal using either MF/UF with pretreatment or NF is more economic 3. 4 .3 Nanofdtration and Reverse Osmosis The MWCO of NF and R 0 is in the 100 to 1000 Da range with "pores" < 1 nm in diameter Organics. .. micropollutant rejection will not be investigated in this project, a brief review of rnicropollutant rejection abilities of NF membranes is included due to the importance of natural organics Two ultra-low pressure R 0 membranes were compared (TFC-S, TFC-ULP, Fluid Systems) Both membranes showed different salt rejections, but pesticides were removed to a very high l degree (>94O/) (Takigawa et a (1995))... water Multivalent ions are believed to enhance natural organics adsorption The effect depends on the organic type Clark and Jucker (19 93) determined that the effect of calcium on FA is lower than with HA Binovi (19 83) found that the gel layer formed on R 0 membranes was composed primarily of l organics and iron Nystrom et a (1994) filtered HA with 3 mgL-1 iron and whlle a gel layer was formed, the flux... acheved by producing asymmetric membranes or by mounting a thin layer on a more porous support (Noble and Stern (1995)) While MF membranes are symmetric, UF membranes are mostly asymmetric due to the smaller pore size 3. 3.2 Membrane Materials for N F and R 0 A comprehensive R 0 and NF membrane materials overview was published by Petersen (19 93) NF membranes may be porous or non-porous depending on the material... together with convection Fouling by macromolecules, especially pore fouling, also increased organic rejection over time (Fane et al (19 83) ) Copyright © 2001 by Andrea I Schafer Fouling by Natural Organics and Colloids 3. 5 FOULING NATURAL BY ORGANICS AND COLLOIDS Generally, membranes with larger pores exhibit a greater flux decline as filtration proceeds T h s is due to the significantly hgher intrinsic... Thorsen et a (19 93) l recommend the use of highly hydrophilic membranes with a pore size of 1-2 nm and low operating pressure to reduce fouling in the filtration of soft waters high in organics Fouling was worst for positively charged membranes which interact strongly with the negatively charged organics (Nystrom et al (1996)) In a later study Thorsen et a (1997) found hydrophllic membranes to be more... removal of Cyptospoadium depends on size, adsorption and cake layer built-up Jacangelo et al (1995a) observed that fouling of MF membranes increased rejection of various species Consequently, Icumar et al (1998) found a significant removal of trihalomethanes (THMs) bp MF in an extended pilot study The retention of natural organics has to date not been studied on a small scale, although fouling of natural. .. average solute concentration across the membrane Solvent flux J =L, (AP-oAn,) Solute flux J , =L, AII+(l-0) Jc,, (3. 31) , (3. 32) Solute flux increases with solvent flux (and pressure) and with increasing osmotic pressure The Solution Diffusion Model assumes that solute and solvent dissolve in the membrane, which is imagined as a dense, non-porous layer The membrane also has a layer of bound water at the . fragile colloid-organic matrices, as described in Chapter 2. 3. 4.2 Ultrafiltration (UF) Rejection of natural organics by UF membranes has been discussed briefly in the natural organics characterisation. solute concentration across the membrane. Solvent flux J =L, (AP-oAn,) (3. 31) Solute flux J, =L, AII+(l-0) Jc,, , (3. 32) Solute flux increases with solvent flux (and pressure) and with. section of Chapter 2. The MWCO ranges from 0.5 to 30 0 kDa in UF and ths governs retention of natural organics. Hagmeyer et al. (1996) reported that DOC removal varied between 26 and 37 % for