THE CHEMICAL ENGINEERING JOURNAL ELSEVIER ‘The Chemical Engineering Journal 57 (1995) 163-176 Modelling electrodialysis using the Maxwell-Stefan description Gerrit Kraaijeveld, Vera Sumberova, Siemon Kuindersma, Hans Wesselingh * Deparment of Chemical Engineering, Univesity of Groningen, Niexborh 4, 747 AG Groningen, Netherlands Abstract, ‘The Maxwell-Stefan mass transfer description is applied to two electrodialysis systems. Firstly, a model system involving NaCl and HCI is investigated. For this system, all the model parameters are obtained either from separate experiments or existing correlations, Although it is not possible to determine all the parameters accurately, the values found are all within the ranges expected Secondly, a two amino acid system is investigated. The separation of two amino acids using electrodialysis is found to be feasible. Unfortunately, it is too difficult to obtain sufficient data for a proper quantitative description of the process. In both cases, the Maxwell-Stefan description appears to provide a thorough model of the electrodialysis process. Keywords: Electrolysis, Mass transfer; Multicomponent; Membrane; Diffusion 1, Introduction membrane. Similarly, the anions pass through the anion exchange membrane and not through the cation ex- change membrane. Thus if the membranes are atranged in an alternating fashion between the electrodes, this will result in concentrated (concentrate) and diluted (diluate) compartments being formed between the mem- branes as is shown in Fig. 2. In other words, electro- dialysis is a unit operation which can, starting from a Electrodialysis is a unit operation which uses ion exchange membranes to remove electrolytes from a solution under the application of an electrical field. A schematic representation of an electrodialysis cell or “stack” is shown in Fig. 1. The cell usually has a “plate and frame” design. On the outside there are two electrodes, over which the electrical field is applied. Inside, the cation and anion exchange membranes are Dituate A a placed in an alternating fashion between the electrodes. Spacers are used to separate the adjoining membranes, create compartments, promote turbulence and connect the appropriate fluid channels with the appropriate compartments in the stack. There are usually four fluid channels which traverse the entire stack: the concentrate and diluate feed and product streams (the electrode feed and product streams usually do not traverse the stack). The compartments formed between the mem- branes are connected in an alternating fashion to either the concentrate or the diluate streams. The principle used in the electrodialysis process is j tH the fact that the ion exchange membranes exclude co- ions, and hence only the counter-ions can diffuse through the membranes. Thus cations pass through the cation exchange membrane and not through the anion exchange Feed Fig. 1. Schematic representation of an electrodialysis cell. The elec- * Corresponding author. ‘odes and cation and anion exchange membranes are shown,164 G. Kraaljeveld et al. | The Chemical Engineering Joumal $7 (1995) 163-176 © a coa oe Fig. 2. Schematic concentration profile in an electrodialysis cll, The ‘exclusion of co-ions is illustrated. Cations (++) travel towards the ‘anode (—), In doing so they can pass through the cation exchange ‘membrane (c), but not through the anion exchange membrane (a). For the anions (—) a similar process takes place. Thus compartments will be created from which all the ions are removed (diluate com- partments) and to which ions are added and not removed (concentrate compartments). given electrolyte feed stream, produce a concentrated and a diluted product stream. Due to unwanted redox products being produced at the electrodes, its common practice to use a third process stream, the electrode fiuid, to prevent these unwanted products from entering the product streams. Thus the complete process usually involves three process streams. For a more elaborate description of the electrodialysis process, the reader is referred to standard texts such as Perry and Green [1] and the references cited therei Ever since the invention of the first electrodialysis apparatus by the German count Botho Scherwin in 1900 [2], the modelling of the mass transfer which occurs in an electrodialysis cell has been problematical. There are three major causes for these modelling problems. Firstly, the complete determination of the required properties of the ion exchange membranes is often a difficult and laborious task. Secondly, there is very little information available on the required mem- brane diffusion coefficients. Thirdly, the mass transfer in an electrodialysis cell is often not only of a diffusive nature, Other transport mechanisms may occur si- multaneously with the diffusive transport, and this usu- ally results in a loss of efficiency. In this paper, we describe our attempts to overcome the problems associated with the modelling of the electrodialysis process. Two model systems are inves- tigated: the NaCI-HCl mixed salt system combined with Tonics membranes, and a system containing two amino acids combined with Selemion membranes. 2. Process model ‘The mass transfer model of the electrodialysis process consists of four elements, which are described in this section. The elements are: the bulk solution (the well- mixed part of the solutions); the polarization films (the unstirred part of the solutions adjacent to the mem- branes); the ion exchange membranes; the electrodes. In most laboratory applications, the electrodialysis process is operated in a batch mode. In industrial applications, the process can also be operated in a feed-and-bleed or single-pass mode. As is shown in Fig. 3, the electrodialysis process in batch mode can be described as three circulating batch systems. For the three circulating batch systems, the diluate, concentrate and electrode fluid, component mass bal- ances are given by de, ad AN,+Vrj—Acju @) where r; is the rate at which component i is generated by either electrode reactions or acid-base or dissociation, equilibria. If only strong electrolyte solutions are used, the second type of reaction need not be considered However, the electrode reactions cannot be ignored if, the number of cell pairs is less than 100, as is the case in most laboratory scale electrodialysis cells, The last term in the mass balance is the viscous convection contribution, which is a possible transport mechanism that may occur in the electrodialysis cell. It should also be noted that, for most laboratory applications, the amount of salt removed per pass is small, so that a model using average fluxes instead of local fluxes is sufficiently accurate. In order to proceed with the calculations of the fluxes through the membranes, further assumptions concerning Electrode fluid Bae “Mp Diluate a Concentrate qo Fig. 3. Schematic flowsheet of electrodialysis in batch mode.G Kraaljeveld et al. | The Chemical Engineering Joumal 57 (1995) 163-176 16s the mass transfer situation near the membrane are required, Fig. 4 illustrates the model used for the mass transfer calculations. ‘The three mass transfer resistances (per membrane) in the model are the two diffusion films on either side of the membrane and the membrane itself. For each of these resistances, a system of partial differential equations needs to be solved. The system of partial differential equations for the polarization film consists of the following. For an N component system, there are N—1 independent Maxwell-Stefan equations [3] 7 Rr aad 2s ead, ” o) Here the average species velocities are related to the fluxes by 0s 8) Furthermore, for each component there is a com- ponent mass balance. Since under normal laboratory operation, the amount of mass transfer per pass is low, it is assumed that the mass balances may be solved under the pseudo-steady state assumption -VN+n=0 @ In the case of strong electrolytes, the reaction term reduces to zero, and the above equation indicates that the flux is constant, ie. independent of location. When there are charged species in the system, an additional relation is provided by the electroneutrality condition 0 6) ‘A further relation between the fluxes is provided by the fact that the sum of the fluxes of the charged species is equal to the electrical current density Fig, 4. Mass transfer resistances near the membrane, ‘San © Finally, a reference frame needs to be chosen. For the diffusion film, this is the volume-fixed reference frame Devil @” Foran componentsystem, there are 2N +1. unknown variables in the polarization films (N fluxes, N’ con- centrations and the electrical potential). These can be calculated from the system of 2N++1 differential equa tions, consisting of N- 1 Maxwell-Stefan equatior component mass balances, the electroneutrality con- dition and the reference frame. Eqs. (2)-(6) can also be used to describe the mass transfer in the ion exchange membranes. There are, however, two important differences. Firstly, there is an extra component present (the membrane itself), and secondly, the diffusion coefficients in the membrane differ from those in the aqueous solution. In this case, Eg. (7) is usually replaced by the membrane-fixed reference frame N, ® For an N component system, the system of differential equations thus consists of N Maxwell-Stefan equations, N+1 mass balances, the electroneutrality condition and the membrane-fixed reference frame. These contain 2N+3 unknown variables, namely the N+1 fluxes and the N+1 concentrations. The systems of differential equations for the polar- ization film and the membrane need to be linked to each other by means of the interfacial equilibrium. For every ionic species present, an equilibrium relation needs to be established, For a “1-1 salt”, the Donnan equilibrium coefficient may be used aay K @) where the quantities inside the membrane are indicated by an overline. For the amino acids, it was found to be more convenient to use the following relation & Kae (10) ‘The water content of an ion exchange membrane can be determined experimentally, and the water con- centration in the membrane is then given by - _ %0H0XPm = “T00xM,, (uy) Furthermore, the experimental determination of the ion exchange capacity in combination with the elec-166 G. Kraaijeveld et al. | The Chemical Engineering Joumal 57 (1995) 163-176 troneutrality condition provides another useful equation Zn (22) Using a combination of Eqs. (9)-(12), the equilibrium can be completely characterized, and all the concen- trations in the membrane may be calculated if the external concentrations are known and if equilibrium is assumed to exist on the interface, Thus the previously mentioned link between the two sets of differential equations has been established. What remains is to establish the appropriate boundary conditions so that the differential equations can be solved. The boundary conditions required are the bulk concentrations on both sides of the membrane and the electrical current passing through the membrane (or the electrical potential ap- plied across the membrane). Finally, to make the model complete, the phenomena at the electrodes need to be described, In the systems considered here, three reactions need to be taken into account. Firstly, the hydrogen evolution reaction occurs at the cathode 2H*+2e7 =H, Ag. =0.000V (a3) This reaction has been studied extensively, and the Kinetic parameters can be found in standard texts on electrochemistry [4]. For a platinum electrode, the Butler-Volmer equation reduces to —0.8x10-3 ooo 32" (a4) ‘The second electrode reaction is the oxygen evolution reaction. This reaction occurs at the anode 2H,0 = 0,+4H* +4e7 AG2=1228V (15) Kinetic data on this reaction can again be found in standard texts on electrochemistry [5]. Contrary to the parameters for the hydrogen evolution reaction, the Kinetic parameters reported for the oxygen evolution reaction exhibit much variation. For a platinum elec- trode, the Butler-Volmer equation becomes 10-° on 7] (16) Finally, at higher salt concentrations under acidic conditions (at neutral or basic pH, hypochlorite is formed), the chlorine evolution reaction may occur at the anode 2CI" 2 Ch, +2e7 AG,°=1.395 V ay) In this case, the Butler—Volmer equation in the following form may be used [6] RT 0.0601cq)- onl Ee | RT as) —0.005968(Pq,)"* ox a | In this equation, the concentration should be expressed in moles per litre and the partial pressure in bar. The kinetics for the three reactions discussed above are only valid on platinum electrodes, Since the working electrodes in an electrodialysis stack are at best pla- tinized (usually they are made of a cheaper material, eg, lead or stainless steel), the above kinetics cannot be used for these electrodes. However, for mass balance purposes, only the mounts of the ionic species which are removed from the solution by electrode reactions need to be known, At the cathode, where only reaction (13) occurs, this can be calculated directly from the electrical current. At the anode, reactions (15) and (17) are competing. In this case, an electrode current efficiency (which specifies the fraction of the current used for one of the two competing reactions at the electrode) can adequately describe the situation, Having established the entire model, the next step is to obtain the required model parameters. These can be subdivided into two categories: the membrane prop- erties and the diffusion coefficients. 3. Membrane characterization For the electrodialysis of the NaCI-HCI solutions, two Tonics membranes were used: the cation exchange membrane 61 CZL 386 and the anion exchange mem- brane 204 UZL 386. The required properties are listed in Table 1 and are discussed briefly below. The ion exchange capacity of a membrane which contains strong ion exchange groups is a constant which can be determined as follows. In the case of a cation exchange membrane, the membrane is first converted completely to the hydrogen form, It is then washed with distilled water to rinse out any sorbed electrolytes. ‘The membrane is then placed in a concentrated sodium, chloride solution, which contains an exactly known small amount of sodium hydroxide, After the ion exchange process is finished, the ion exchange capacity can be determined by titration of the excess sodium hydroxide. In principle, a similar method can be used for the anion exchange membrane, but in practice these usually disintegrate under high pH conditions. Hence the pre- ferred method is to use the membrane in the chloride form, remove the chloride from the membrane using nitrate and then directly titrate the chloride using silver nitrate. ‘The membrane thickness was measured using a screw micrometer. For the cation exchange membrane, theG Kraaijeveld et al. | The Chemical Engineering Journal 57 (1995) 163-176 167 Table 1 Properties of Ionics membranes* Membrane property 61 CZL 386 204 UZL 386 1827 mol m™? wet membrane Capacity 1690 mol m=? wet membrane Thickness 0559-0566 mm 0551 mm Density in HCI solution 11625-27504 11130+800 Density in NaCl solution 11675-7356 11000415. +y in HI solution 0280806 0.25052 + in NaCl solution 0574028 0560296 26H,0 in HCI solution 31.66—054e 3294-0.16¢ £614,0 in NaCl solution 3017-083 3838-1420 * Where ¢ isthe total external salt cone. in mol =" first value in the table was measured for the hydrogen “rer or UL form of the membrane and the second for the sodium == . form. All measurements were made at infinite dilution, 3 “* and the effect of the external salt concentration on the “bos thicknesses was neglected. ee The wet densities of the membranes were measured oo by accurate determinations of both the masses and the cas Gn volumes involved, As can be seen in Table 1, the effect oe me of the external concentration on the density was taken “a Son Ua into account. Ton exchange equilibrium measurements were also é performed. For these measurements, the membrane os. ¥ ‘was equilibrated with a given solution. The membrane © was then removed from the solution, any liquid on the 00 ool surface of the membrane was removed and the mem- ou tm ee tno) brane was placed in distilled water. The electrolyte sorbed in the membrane enters the solution and its concentration can be determined by titration. From a knowledge of the masses and volumes involved in the experiment, the co-ion concentration in the membrane can be calculated, Eq. (9) was used to describe the equilibria. The liquid phase activity coefficient was calculated using the equation suggested by Bromley [7] ey. (19) where A is the Debye-Hiickel constant (0.51), Bis a parameter of the electrolyte under consideration and Tis the ionic strength. For HCI the value for B is 0.1433 and for NaCl the value is 0.0574 [8]. ‘The membrane phase activity coefficients were then fitted to the experimental data points which are shown in Fig. 5. Clearly the description used is adequate. Surprisingly, the experimental results indicate that the activity coefficients in the membrane can be described by a linear relation between the external concentration and the activity coefficient. On the other hand, this might not be so surprising, since the standard state Fig. 5. Sorption characteristics of two Tonies membranes in HCl and NaCl solutions used for the activity coefficients in the membrane is that of infinite dilution in an aqueous solution, ‘The water contents listed in Table 1 were determined by equilibrating the membrane with a given external solution, When the membrane is removed from the solution, any liquid on its surface is removed (this inevitably results in some experimental error). The wet weight of the membrane is then determined, the mem- brane is dried in an oven and finally the dry weight of the membrane is obtained. The water content is then expressed as the percentage by which the weight of the wet membrane has been reduced. Unfortunately, ‘most membranes cannot be reused once they have been dried. For the amino acid system, Selemion membranes were used. In this case, the concentration ranges ex- amined were relatively small, so that most of the membrane properties could be considered constant. These properties are listed in Table 2 (where CMV is the abbreviation for the cation exchange membrane and AMV is the abbreviation for the anion exchange membrane) and were obtained using the same methods as discussed above.168 G. Kraaijveld et al. | The Chemical Engineering Joumal 57 (1995) 163-176 ‘Table 2 Selected membrane properties of Selemion membranes operty cu AMV Capacity (meq g~* wet) 23 28 Density (kg m™?) 1200 1200 Water content (mim%) 42 219 ‘Thickness (mm) 014 03 ay © Glycine -- 3 Aspartic Acid © Glutamic Acid AMY Capacity Cone (mol/l) pH (-) Fig. 6. Amino acid equilibria for the CMV. ‘yine (Asp) Glycine (Glu) Aspartic Acid Cone (mol/l) pH () Fig. 7. Amino acid equilibrie for the AMV. The only membrane properties of the Selemion mem- branes which were investigated extensively were the equilibria of the amino acids. Figs. 6 and 7 show the absorption on the CMV and AMV respectively, which occurs when these membranes are equilibrated with a 0.03 M solution of the amino acids at the relevant pH. ‘The results presented in these figures seem to indicate that the amino acids only enter the membranes if an amino acid species of the appropriate charge is present in the solution. In other words, only the cationic forms of the amino acids enter the CMV, and only the anionic forms of the amino acids enter the AMV (for the investigated pH range, this means that no glycine enters the AMV!). Thus the AMV only contains amino acids at high pH values, whereas the CMV only contains amino acids at low pH values. At very low pH values, there is an increasing amount of hydrogen present in the CMV, which is the reason why there is a maximum in all the curves shown in Fig. 6. From these figures, it may thus be concluded that, for modelling purposes, itwill be sufficient for the amino acid systems to consider only the counter-ion transport through the membranes. ‘The equilibria shown in Figs. 6 and 7 were described using Eq. (10). Unfortunately, the equilibrium coefficient K, isnot constant. Indeed, itis rather difficult to describe its behaviour using a simple formula. Eventually the best results were obtained by fitting minus the logarithm ‘of K according to pK/=aq+a,pH+azpHl? +a3ph?? (20) The values for the required coefficients are given in Table 3. The full curves shown in Figs. 6 and 7 were calculated using Eqs. (10) and (20) and Table 3. In this section, all the parameters required to describe the equilibrium completely have been obtained. The only model parameters which still need to be established are the diffusion coefficients and these will be deter- mined in the next section. 4, Diffusion coefficients For simple salts in aqueous solution, there is usually sufficient information available on the Maxwell-Stefan diffusion coefficients. Table 4 shows the values of the diffusion coefficients required for the NaCI-HCI system. However, for the amino acids, diffusion data in aqueous solution are less readily available. Indeed for both aspartic acid (Asp) and glutamic acid (Glu) no diffusion data were found. To obtain a reasonable estimate of the amino acid diffusion coefficients, the data published by Cohn and Edsall [11] on the diffusion coefficients of some amino acids and peptides were correlated. The following relation was found between the Fick diffusion coefficient and the molecular weight of the compound ‘Table 3 Coefficients for Eq. (20) System ae a a a GWICMY 0.58444 —0.86095 0.18878 0.023500 GyCMY — -096774 031155 - 0.078498 ASpICMY — ~031674 — =1.0087 0.27951 GIWCMV 0.23755 1.0160 0.32188, AspiAMV ~73716 5.6468. ~ 1.06805 0.15953, GIWAMV — -70357 3.2705 0.60505 0.036508 acid Giy, glycine; Asp, aspartic acid; Glu, glutamiG. Kraaijeveld et al. / The Chemical Engineering Jounal 57 (1995) 163-176 169 Table 4 Free solution diffusivities (based on data by Chapman [9] and Mills and Lobo (10})* Components Diftusviey (10~? mi s-4) HY, HO 9.308 Na*, HO 1.333 cr, HO 2.033 HY, Cr 0.0320 Na*, Cm o.o01sc%# He, Na* o.o0sace “Where C is the chloride concentration in mol I~ D (10°m‘/s) oo a 100 200 300 M, () Fig. 8. Amino acid diffusion coefficients as & function of the molar D= nit @ This correlation corresponds to the full line in Fig. 8, Which also shows the data points used for the correlation. It is interesting to note here that the power to which the molar mass is raised in the above equation is almost the same as the power of the molar volume of the solute in the Wilke-Chang correlation [12] 7.4 X1075(GMp)"8T naVa* Dass (2) Furthermore, for these amino acids, there is a virtually linear relationship between the molar mass and the molar volume. Hence Eq. (21) may be seen as a Wilke-Chang-like correlation for amino acids, The diffusion coefficients correlated by Eq. (21) are those at the isoelectric point. At this point, the larger part of the amino acid is present in the neutral form, which means that this diffusion coefficient is probably, the best estimate of the diffusion coefficient of the neutral amino acid species. In this section, the diffusion coefficients of the cationic, neutral and anionic amino acid species are assumed to be equal. Because of the low concentrations used, and the scarcity of data on amino acid diffusion, the ion-ion interactions in the Maxwell-Stefan equations were ne- glected. Thus, in the aqueous phase, only the Maxwell~ Stefan interactions between the solute and the solvent were taken into account, Another important parameter for the description of the mass transfer in the aqueous phase is the film thickness of the polarization layer. For an electrodialysis cell fitted with tortuous path spacers, the correlation presented by Sonin and Isaacson [13] may be used sn=19se%(4) Re? 300 exp(gn) 3 2 expen) (30) § cq +a exp(bn) Eq, (30) illustrates that the anodic current efficiency is a function of the chloride concentration (in moles per litre) and the applied overpotential. Since no ex- perimental data on the overpotentials were available, the potential applied across the stack was substituted in the equation instead (it must, however, be borne in mind that this can only result in a rather crude ap- proximation). The constant a in Eq. (30) represents the ratio of the exchange currents of the two competing electrode reactions. The constant b in Eq. (30) tep- resents the difference between the exponential factors of the two competing reactions (see Eqs. (16) and (18)). It was found that the experimental resuits could be described using Eq. (30) if the constants a and & had values of 30.72 and —2.136 respectively. Both of these values appear to be feasible when compared with “similar” data available in the literature. Fig. 10 il- lustrates the agreement between Eq. (30) and the experimental results, where the symbols represent the experimental data and the full line was calculated using Eq. (30). As can be seen from the figure, a reasonable agreement exists. Finally, the inefficiencies involved in the mass transfer through the ion exchange membranes should be ex- amined. The model used incorporates two possible mechanisms by which inefficiencies can occur: the fusion of co-ions through the membrane and viscous convection. Before discussing the results on viscous convection, another inefficiency is examined. Fig, 10. The anodic current efficiency as a function of the chloride concentration in a mixed NaCI-IICI system and the applied electrical potential. One of the inefficiencies which will occur is that of current leakage through the ports made in the mem- branes for distributing the fluid to the various com- partments. This inefficiency can be described according to [17,18] Re(N+2)(N-+1) ZAR, +3NRet RFD where N represents the number of cell pairs, Re is the resistance of a cell pair, R, is the resistance of the port though which the leakage occurs and R,, is the resistance of the pathway between the port and the compartment. It was found that, if Eq, (31) was used to explain the inefficiencies of the mass transfer, the required resistances were about ten times smaller than the actual, resistances. This indicates that, although the leakage through the ports does contribute to the inefficiencies, 1)G. Kraaijeveld et al. | The Chemical Engineering Joumal 57 (1995) 163-176 173 ‘Table 7 Ratio of viscous velocity and pressure gradient for the Tonics mem- branes ‘Membrane IAP (10-8 m s*! bar~) 204 UZL 386 59 61 CZL 386 2a it can only be used to account for a small fraction of the inefficiencies which occur. As can be seen in Table 6, the inefficiencies can be explained if viscous convection occurs, with an average viscous velocity of 1.0 10-7 ms~?. To ascertain whether these viscous velocities are possible, some additional experiments were performed. In these experiments, the water flux as a function of the applied pressure gradient across the Ionics membranes was measured. For both membranes, the relation between the viscous velocity and the applied pressure gradient was linear, and the constants of proportionality are listed in Table 7. The values in the table are relatively high, which indicates that the membranes have a fairly open structure. ‘As can be seen in Table 7, @ viscous velocity of 1.01077 m s”? would require a pressure gradient of less than 0.05 bar. In practice, the cell is operated at an equal inlet pressure of 1 atm. However, these inlet pressures are not extremely accurately controlled; hence a difference of 0.05 bar is not at all unlikely, thus making this a feasible explanation for the observed efficiencies. Finally, the sodium-hydrogen diffusion coefficient in the cation exchange membrane was fitted. Unfortu- nately, this parameter did not have much effect on the quality of the fit and, as can be seen in Table 6, the values for this parameter differ considerably between the various experiments. The average value for this diffusion coefficient is 9.66 x 10-* ms~? with a standard deviation of 7.50X10-" m s~?, and this result must therefore be regarded as unreliable. 6. Amino acid electrodialysis ‘The electrodialysis of the amino acid solutions was performed in a cell which was manufactured by the engineering fim Kuipers B.V. The cell consisted of ‘wo platinized titanium electrodes, two AMVs and one CMV. The cell was used as shown in Fig. 3 and as discussed earlier. The electrode fluid consisted of 4 1 of 0.1. mol I HC. The initial concentrate and diluate solutions con- sisted of 4 | of solution containing 0.03 mol 1~? of each of the two amino acids present, and had the same initial pH. The concentration of amino acids used was fairly low due to the low solubility of Asp (5 g 17? at 25 °C [19]). The initial pH of the concentrate and diluate solutions was adjusted using either concentrated HCI or NaOH. During the experiment, the pH in the diluate compartment was controlled using a pH con- troller which added either HCl or NaOH. The pH values of the concentrate and electrode compartment were allowed to vary during the experiment. ‘The initial electrical current was 1 A, and the electrical potential applied across the stack was kept constant during the experiments (between 3 and 24 V depending on the pH of the system). The flow rates through the compartments were kept at approximately 10 ml s~?. All experiments were carried out at room temperature. During the experiments, the electrical current, applied electrical potential, compartment volumes, pH of the diluate compartment and various flow rates were mon- itored. Every hour, a 50 ml sample was taken from each holding tank for analysis purposes. For these samples, the pH was measured and the chloride ions were determined by the Mohr method [16]. The amino acids in the sample were determined using high per- formance liquid chromatography (HPLC) analysis, sim- ilar to that described by Tapuhi et al. [20]. The essence of the method is as follows. The pH of the solution containing the amino acids is first increased to pH 9.5. The amine group of the amino acids is then derivatized with dansyl chloride (1-dimethylaminonaphthalene-5- sulphony! chloride) (Dns-Cl) according to the reaction AA-NH+Dns-Cl— AA-NH-Dns+HCl (32) ‘The derivatized amino acids are then separated on an. HPLC column and detected by a UV detector at 254 nm. Unfortunately, both the derivatized products of Glu and Asp decompose, and hence the analysis was always performed over similar time spans for improved accuracy. ‘The amino acid electrodialysis experiments were per- formed atvarious diluate pH values for both the Gly-Asp and Gly-Glu systems. When the electrodialysis of the amino acids is per- formed at low pH values, the positively charged Gly ions will move through the cation exchange membrane to the concentrate compartment. The negatively charged Asp ions (or Glu ions) will move through the anion exchange membrane to the electrode compartment. Thus the diluate solution will be depleted of amino acids, and a Gly solution (the concentrate) and an Asp or Glu solution (the electrode fluid) will be created. ‘To describe this process, a total of six membrane diffusion coefficients need to be established for each system (four in the cation exchange membrane and two in the anion exchange membrane). ‘Unfortunately, the determination of the amino acid concentrations in the various solutions was not suffi