Recovery of monosaccharides from dilute acid corncob hydrolysate by nanofiltration: modeling and optimization

In this work nanofiltration technology has been employed for removal of inhibitors and recovery of monosaccharides from dilute acid lignocellulose hydrolysates. The influences of feed solution pH, permeate flux, and Na2SO4 concentration on the rejection of monosaccharides and inhibitors were investigated. The results showed that the pH for the separation of carboxylic acids and furans from monosaccharides should be as low as possible. With increase of Na2SO4 concentration carboxylic acid and furan rejection decreased. Subsequently, the Donnan steric pore and dielectric exclusion model coupled with mass balance was used to predict the rejection of solutes at different permeate fluxes. In order to select a suitable permeate flux and operating time, multi-objective optimization was carried out to obtain the maximum total inhibitor removal efficiency, the maximum monosaccharide recovery rate, and the minimum water consumption. The optimal operating conditions were then verified using the real hydrolysate as feed solutions. More specifically, for the treatment of 6 L of a hydrolysate solution, 13 L of water and a treatment time of 35 min were required. This process allowed the removal of 90% inhibitors, while 93.55% glucose, 90.75% xylose, and 90.53% arabinose were recovered. Finally, a batch column equipped with a strong acid cation exchange resin was employed to recover the monosaccharides from the hydrolysate. Using water as an eluent, 95.37% of the sulfuric acid and 94.87% of the monosaccharides were recovered. In all, we demonstrated that the combination of nanofiltration with electrolyte exclusion chromatography is a promising integrated process for the recovery of monosaccharides and inorganic acids from dilute acid corncob hydrolysates.


Introduction
Lignocellulosic biomass, such as woody materials and agricultural residues, is an abundant, readily available, and renewable feedstock for the production of biofuel. However, the utilization of such biomass generally requires pre-treatment processes, through which polymeric carbohydrates are decomposed to monosaccharides. 1 To date, a number of pretreatment methods have been proposed, including enzyme catalysis, hot water pretreatment, thermal pretreatment with mineral acids, or alkaline treatment, 2 with dilute acid pretreatment being the most commonly used method. 3 However, the dilute acid pretreatment method produces a number of by-products, such as furfurals, hydroxymethyl furfural (HMF), phenolic compounds, and acids (e.g., acetic and formic acid), 4,5 with the presence of such by-products during sugar fermentation being reported to seriously inhibit bacterial growth and the production of the desired bio-based products. For example, even low concentrations of phenolics are lethal to Clostridium, which is a widely used bacterium in the production of butanol and butyric acid. 5 The removal of these inhibitors from hydrolysates is therefore of particular importance.
To date, a number of techniques have been employed for hydrolysate detoxication, including evaporation, activated charcoal adsorption, overliming, neutralization, ion exchange, enzyme treatment, and electrodialysis. [6][7][8] However, as expected, these methods exhibit a number of different advantages and disadvantages. For example, overliming produces large quantities of gypsum during the neutralization and detoxication process, while evaporation increases the concentration of nonvolatile compounds despite removing volatile compounds. In addition, electrodialysis removes only the compounds that can be dissociated (e.g., p-coumaric acid, ferulic acid, syringaldehyde, and vanillin). 9 In the context of the various fermentation inhibitors mentioned above, furfural, HMF, and phenolic compounds can be removed by adsorption due to their hydrophobic properties, while the separation of acetic acid and formic acid from glucose and xylose is more problematic. Nanoltration (NF) is an efficient membrane separation technology that exhibits low energy consumption and unique separation properties. As such, Weng et al. 10 investigated the separation of furans and carboxylic acids from sugars in dilute acid rice straw hydrolysates using Desal-5 Dk nanoltration, which had a molecular weight cutoff of 150-300 Da. Using a pH of 2.9 and an applied pressure of 24.5-34.3 bar, they achieved maximum separation factors of acetic acid and HMF over xylose of 49 and 43, respectively. In addition, Brás and Guerra et al. 11 employed diananoltration mode to detoxify hemicellulosic hydrolysates from extracted olive pomace, and reported 99% removal of furans, acetic acid, and formic acid, but a monosaccharide loss of 40%. Furthermore, Maiti et al. 12 used the Donnan steric pore model (DSPM) to characterize the membrane and membrane transport. They concentrated a rice straw acid hydrolysate using a volume concentration ratio of 4, and increased the concentrations of xylose, glucose, arabinose, cellobiose, and inhibitors by 100, 104, 93, 151, and 3%, respectively. However, previous studies have ignored the existence of dilute sulfuric acid, which can have a signicant inuence on the separation performance of the nanoltration membrane. Optimization of the operating conditions would therefore be expected to minimize the monosaccharide removal rate and the consumption of water. Moreover, separation of the acid-sugar mixtures produced from the treatment of hydrolysates by nanoltration could be simplied if sulfuric acid could be recycled.
In this context, electrolyte exclusion chromatography, which is an efficient method for the separation of strong electrolytes from weak electrolytes and nonelectrolytes, 13 has recently been applied in the fractionation of acid-sugar mixtures. During this process, strong electrolytes are excluded from the strong ion exchange resins either completely or partially due to electrical repulsion caused by the xed ionic groups in the resin. 14 In addition, the strong electrolytes break through the resin bed at the interstitial volume due to complete exclusion at innite dilution. Weak electrolytes and nonelectrolytes are unaffected by the electrolyte exclusion and so propagate through the column slower than strong electrolytes. Thus, Sun et al. 15 used a Dowex 1X8 column to separate sulfuric acid and sugars in concentrated sulfuric acid hydrolysates of bamboo, and reported sulfuric acid, glucose, and xylose recoveries of 90.5-93.4, 94.9-99.7, and 82.8-88.3%, respectively. In addition, Heinonen and Sainio 16,17 investigated the recovery of monosaccharides and sulfuric acid from the concentrated acid hydrolysate of lignocellulosic biomass, while Xie et al. 18 employed the electrolyte exclusion chromatography technique for the separation of monosaccharides from dilute acid lignocellulosic hydrolysates. Furthermore, Springeld and Hester 19 investigated the fractionation of a solution containing sulfuric acid (10 wt%) and glucose (10 wt%) using a four-zone simulated moving bed for binary separations. These results suggest that the recovery of sulfuric acid can indeed be achieved in a number of systems.
Thus, we herein report the coupling of NF and electrolyte exclusion chromatography to remove inhibitors and recover monosaccharides from a dilute acid corncob hydrolysate. The effects of different operating conditions (i.e., ux, pH, and Na 2 SO 4 concentration) on the separation of acetic acid, formic acid, and furans from monosaccharides are examined, and coupling of the DSPM-DE with mass balance calculations will be employed to predict the rejection of solutes at different permeate uxes and to simulate the diananoltration process. To select a suitable permeate ux (j v ) and operating time (t) for the diananoltration process, multi-objective optimization was carried out to obtain the maximum total inhibitor removal efficiency (Pr inhibitor ), the maximum monosaccharide recovery rate (Y sugar ), and the minimum water consumption (EC). An authentic hydrolysate sample will also be employed to verify the optimized conditions. Following NF, recovery of the monosaccharides and sulfuric acid present in the retentate are attempted using a strong acid cation-exchange resin (PS-DVB) in a batch column.

Raw materials and pretreatment
Glucose, xylose and arabinose were purchased from Shanghai Sangon Biological Engineering Co. Ltd. Acetic acid was purchased from Shanghai Shen Bo Chemical Co. Ltd. Ferulic acids, vanillin, HMF, furfural, vanillic acids, formic acids and acetic acids were purchased from Aladdin Reagent Co., Ltd. Corncob was collected from Jingzhou, Hubei Province, China. The strong acid styrene-co-divinylbenzene cation-exchange resins Sa-2 was purchase from AnHui Sanxing Resin technology Co. Ltd. Synthetic solutions were prepared in de-ionized water. Solution pH was adjusted to 3, 5, 7 and 9 by addition of HCl/NaOH solutions. Hydrolysate sample was prepared by hydrolyzing corncob (20%, w/v) with 2% H 2 SO 4 for 150 min in an autoclave at 125 C. Aer pretreatment, the liquid fractions were separated via vacuum ltration and were stored at 4 C. Before nanoltration, hydrolysate was preltered with a lter of 0.45 mm.

Membrane and nanoltration module
A commercial membrane, DK1812-34D (GE Company, USA), was used in this work which has been proven to have high rejection for monosaccharide. 11 From the information given by the manufacturer, the MWCO of the membrane are 150-300 Da. The effective ltration area of the membrane is 0.32 m 2 . The experimental module is purchased from Sundar Membrane Technology Co. Ltd. The nanoltration experimental setup used in this work is shown in Fig. 1, which has a feed tank, diaphragm pump, pressure gauge, membrane module, and pressure control valve.

Filtration experiments
Before the experiments were conducted, the membrane was washed with deionized water for several times. Pure water ux of the membrane was measured while the operating pressure varied from 6.0 bars to 24.0 bars. The permeability was then calculated as the slope of the pure water ux versus the operating pressure. All ltration experiments were performed in batch mode with the retentate and permeate fully recycled to the feed tank. The temperature was controlled to 25 C by circulating water into the jacket of the feed tank using a constant-temperature device. Feed and permeate samples were collected for each experimental conditions. The permeate ux J v was measured at each operating pressure and calculated using eqn (1).
where V p is the volume of permeation, t is the time, and A m is the effective membrane area. Rejection of glucose and xylose were performed at different pH to estimate the pore size. Rejection of Na 2 SO 4 was performed at varied concentrations to estimate the effective charge density.

Concentration-diananoltration experiment
The membrane separation process was operated at a concentration-diananoltration mode. The optimal J v and t were determined by Parallel Multi-objective Optimization. In the concentration process, 6 L of model solution and hydrolysate were concentrated to 3 L. In the diananoltration procedure, the permeate ow was continuously removed and equivalent volume of deionized water was added into the feed tank to keep the feed volume constant along the experiment. The samples were collected every 3 minutes.

Column experiments
The monosaccharides separation from sulphuric acid was performed in a batch column. The strong acid PS-DVB cationexchange resins (gel type) in H + form were used. The resin bed volume is 425 cm 3 and the bed height was 55 cm. The hydrolysate treated aer the nanoltration was fed in the column. The injection volume was 10 vol% of the resin bed volume. Water was pumped with a constant ow rate of 1 mL min À1 through the column. Samples were collected by an automatic collector.

Sample analyses
The concentrations of the monosaccharides, acetic acid and formic acid were measured by an on-line HP Agilent 1100 HPLC system equipped with a RID detector and a Bio-Rad Aminex HPX-87H column. The HPLC analyses were conducted at 55 C with injection volume of 10 mL. The 0.005 M H 2 SO 4 was used as an eluent. The concentration of 5-HMF, furfural and phenolic compounds were determined by an Agilent 1100 HPLC with a diode array detector working at 280 nm. The separation was carried out through a Zorbax XDB-C18 column at the temperature of 55 C. The mobile phase were 0.3% acetic acid (70%) and methanol (30%) mixture at a ow rate of 1.0 mL min À1 . Sulfuric acid concentrations were calculated by LeiCi DDBJ-350 conductivity meter with a DJS-1CF probe.

Modeling and calculations
The real (R i,real ) and the observed (R i,obs ) rejection of solutes represent the separation performance of the nanoltration membrane, which are dened as eqn (2) and (3) where C i,p is the concentration of solutes in the permeate, C i,m is the concentration near the membrane surface, which is difficult to be measured, and C i,b is the bulk concentration of solutes. Due to the concentration polarization, the bulk concentration is lower than the concentration near the membrane surface. Thus the following correlation of C i,b and C i,m is used to obtain C i,m .
Substitute eqn (2) and (3) for eqn (4), where K is the mass transfer coefficient. It can be calculated from eqn (6) in which d c is the hydrodynamic diameter, D represent diffusivity coefficient, Re is Reynolds number and Sc is Schmidt number.

DSPM-DE model
The Donnan-steric-pore-dielectric-exclusion (DSPM-DE) model, 20,21 which was derived from the extended Nernst-Planck equation, was used in this work to simulate the NF process. The equation can be expressed as: For the uncharged solutes (like the xylose), the electrical potential gradient can be ignored. So the rejection of the solutes can be expressed as: in which F i , K i,c and Pe are the model parameters. Pe can be obtained from eqn (9) and where r i,s is Stokes radius of solute, r p is average membrane pore radius, J w is the pure water ux and DP is the transmembrane pressure. All the model parameters were determined in Section 4.1 in details.
For the charged solutes, the concentration gradient and potential gradient can be expressed as eqn (15) and (16) , respectively.
The electroneutrality conditions should be fullled: where X d is effective membrane volume charge density, Z i is valence of ion i. In order to solve the above-mentioned ordinary differential equations, the boundary conditions should be included which can be obtained from the Donnan steric equilibrium partition coupled dielectric exclusion effect.
f i is steric partitioning coefficient, Dj is Donnan potential difference and DW i is Born solvation energy barrier. The DW i can be calculated based on the method proposed by Bowen and Welfoot: 22 3 p and 3 b are the pore and bulk dielectric constant, respectively. The variation of the average pore dielectric constant was estimated as proposed by Bowen and Welfoot 22 as follows:

Mass balance equations in diananoltration
The diananoltration procedure is a batch-continuous process. The solution ux J v and volume of feed solution keep constant. The mass balance of the diananoltration process can be expressed as eqn (22) which has been adopted by Brás and Guerra. 11 Solve the eqn (22) the concentration of solutes at any time (C i,f,t ) can be expressed as eqn (23) In order to evaluate the diananoltration process, we dene the remove rate of solutes (G i ) as: Based on the denition of G i , the following parameters were dened as well. They were used in the model optimization process. The total inhibitor remove rate is dened as eqn (25).
The water consumption efficiency in the diananoltration process is described as eqn (26). The inhibitors remove efficiency is described as eqn (27) The recovery rate of monosaccharides is described as eqn (28) among which, W 1,i is the mass fraction of inhibitors. W 2,i is the mass fraction of monosaccharides.
In order to evaluate the tness of the model predictions to the experimental data, the average relative deviation (ARD%) between experimental and predicted data was calculated by the following equation: where R exp and R pred is the experimental and predicted rejection, respectively. N is the number of experiment data points.

Calculation of the model parameters
The structural parameters of the membrane, i.e., the membrane pore radius (r p ) and the membrane thickness (DX/A k ), have a great inuence on prediction of the membrane performance, and these parameters can be obtained from physical methods, such as atomic force microscopy or scanning electron microscopy. 23 In this context, Liu et al. 24 proposed a correlation between the molecular weight cut-offs (MWCO) and the r p , and reported an r p of approximately 0.39 nm for the DK1812 membrane. In our case, the rejection data were t to the Spiegler-Kedem and the steric hindrance pore models to nd the pore radius, as these methods were previously employed by Fang et al. 25 Thus, the r p of the DK membrane calculated from the model was 0.395 nm at pH 3. With the value of r p in hand, the value of DX/A k could be calculated from the Hagen-Poiseuille equation (eqn (18)), in which the pure water permeability J w /DP was approximately 2.269 Â 10 À11 m Pa À1 s À1 at pH 3, as determined by a pure water permeate experiment. Indeed, this value of J w /DP was similar to that reported by Almazán et al. 26 (i.e., 2.79 Â 10 À11 m Pa À1 s À1 ). Thus, the r p and DX/A k values calculated from the model were 0.395 nm and 1.661 mm, respectively, and so these values were employed in the following DSPM-DE model. An additional membrane parameter, namely the membrane volume charge density (X d ), which is essential for calculating the rejection of an ionic compound, was obtained by tting the rejection data of a Na 2 SO 4 solution obtained at a range of concentrations (C b ). The relationship between X d and C b is dened as a form of the Freundlich isotherm 27 as follows: For the system of interest herein, the correlation between X d and C b is shown in Fig. 2, which gives a and n values of 57.94 and 0.5379, respectively. Consequently, when the concentration of Na 2 SO 4 is known, the volume charge density X d can be easily determined.

Inuence of solution pH on the membrane separation performance
The molecular formula, dissociation constant, diffusion coefficient, and Stokes diameter of typical sugars, furans, phenolic compounds, and carboxylic acids present in the dilute acid corncob hydrolysate are shown in Table 1. As previously reported, the sieving mechanism and the Donnan exclusion are the two main mechanisms of molecular separation in the NF process. 28 In the case of uncharged solutes present in the hydrolysate, such as glucose, xylose, arabinose, HMF, and furfural, their separation performances depend mainly on the sieving mechanism. Thus, the rejection percentages of these components at a range of pH values are shown in Fig. 3. More specically, at pH 3.14, glucose exhibited the largest rejection, followed by xylose, arabinose, HMF, and furfural. This trend is in accordance with the particle sizes of the ve molecules, as indicated in Table 1. Upon increasing the pH to 9.05, the rejection of glucose, xylose, and arabinose decreased from 97.84, 94.38, and 95.25%, to 93.36, 82.47, and 81.28%. Indeed, similar results were previously reported, 29,30 it was assumed that the increase in solution pH may facilitate membrane swelling. In contrast, the rejection of carboxylic acid and phenolic compounds increased signicantly upon increasing the solution pH (Fig. 3). In this case, an increase in the pH from 3.14 to 6.93, resulted in the rejections of acetic acid, formic acid, vanillic acid, and ferulic acid increasing from 24.58, 11.94, 37.9, and 45.14% to 62.1, 85.4, 94.6, and 95.26%, respectively. A further increase in pH to 9.05 resulted in rejections of almost 100% for the four acids, thereby indicating that these compounds essentially did not pass through the membrane, a phenomenon also observed by Li et al. 31 This large variation in the rejection of carboxylic acid and phenolic compounds at a given pH value was therefore expected to be correlated to their respective pK a values. As shown in Table 2, the pK a values for acetic acid, formic acid, vanillic acid, and ferulic acid are within the range of 3.5-5. Thus, upon variation in the solution pH within this region, the dissociation degrees of the acids changed dramatically. Indeed, at pH 3.14, only 1.9% of acetic acid, 16.1% of formic acid, 4.32% of vanillic acid, and 5.63% of ferulic acid are dissociated, and so the sieving mechanism dominated during the NF process. However, when the pH was increased to levels greater than the pK a values, these compounds were essentially fully dissociated in solution. Moreover, the isoelectric point of the DK membrane (i.e., a value of approximately 4) 32 should also be considered, as it resulted in similar changes in the membrane surface charge upon varying the solution pH. At pH values higher than the membrane isoelectric point, the surface of the membrane was negatively charged. As such, the increased rejection of carboxylic acid and phenolic compounds at pH 6.93 and 9.05 was attributed to the enhanced electrostatic repulsion between the membrane and the negatively charged solute. 31

Effect of SO 4 2À concentration
It has been widely conrmed that the inorganic salt concentration of a solution has a signicant inuence on the membrane separation performance during NF. 33 As the concentration of H 2 SO 4 is approximately 0.2 mol L À1 in hydrolysate solutions, the effect of SO 4 2À concentration on the rejection of monosaccharides, furans, and carboxylic acids should be examined. In this case, to avoid the presence of additional hydrogen ions inuencing the solution pH, we employed Na 2 SO 4 rather than H 2 SO 4 to vary the SO 4 2À concentration (see Fig. 4). In addition, to ensure a constant permeate ux, the operating pressure was adjusted according to the increased Na 2 SO 4 concentration. As    Fig. 4, upon increasing the Na 2 SO 4 concentration in the hydrolysate from 0.05 to 0.5 mol L À1 , the rejection of glucose, xylose, and arabinose decreased by 2, 3.5 and 4%, respectively. Interestingly, the rejection of HMF and furfural decreased signicantly from 7.13 and 6.25% to À2.804 and À3.834%, respectively. This trend corresponded with previous literature reports, 33,34 it was assumed that the decrease in rejection may be attributed to salt-induced pore swelling 35 or reduction of the hydration layer on the pore walls. 36 In the case of the organic acids, signicant changes in rejection were observed upon increasing the concentration of Na 2 SO 4 in the mixture from 0.05 to 0.5 mol L À1 . More specically, the rejection of formic acid, vanillic acid, ferulic acid, and acetic acid decreased from À4.66, 18.01, 21.25 and 5.24% to À54.96, À16.728, À10.011 and À22.83%,respectively. Weng et al. 10,37 also reported a similar negative rejection of acetic acid and HMF during the NF of dilute acid rice straw hydrolysate. They assumed that this decreased rejection was attributed to interactions between the concentration polarization layer of the sugars and the inhibitors, while other reports have suggested that this phenomenon may be attributed to a combination of electrostatic screening and a reduction in steric hindrance. 38 As Na + has a smaller ionic radius and moves more rapidly in solution than SO 4 2À , it passes more easily through the membrane. In addition, the electrostatic repulsion between SO 4 2À and the membrane is higher than those of the carboxylic acids, thereby leading to the increased rejection of SO 4 2À compared to the carboxylic acids. Thus, upon increasing the Na 2 SO 4 concentration in solution, increased quantities of organic acids pass through the membrane to maintain charge balance at the membrane outlet, thereby leading to negative retention of the carboxylic acids.

Modeling and optimization
4.4.1 Effect of permeate ux. Fig. 5 shows the effect of permeate ux on the rejection of the main hydrolysate components. As indicated, the rejection increased for all solutes upon increasing the permeate ux 0.84 to 2.84 L m À2 min À1 , and this effect was particularly pronounced for formic acid, acetic acid, and HMF, where their rejections increased from 0.95, 5.44 and 1.38% to 5.59, 9.65 and 15.6%, respectively. Moreover, the Na 2 SO 4 retention also increased slightly with an increase in the permeate ux. This phenomenon could be explained by the convection-diffusion mechanism. 33 More specically, at higher ux rates, water passes more easily through the membrane, leading to a lower solute concentration in the ltrate and higher solute retentions. However, upon further increasing the permeate ux, greater quantities of the solute accumulate at the membrane surface, thereby leading to severe concentration polarization. Thus, solute diffusion through the membrane would be enhanced, resulting in a decrease or plateau of the solute retention. This would be more likely to take place in the case of high-rejection solutes. In addition, Fig. 5 also shows the tting data for the DSPM-DE model, where it is apparent that the DSPM-DE model ts well with the experimental data (2.15% deviation for monosaccharides, 4.6% deviation for Na 2 SO 4 ). As indicated, the rejection of both monosaccharides and inhibitors increased upon increasing the permeate ux. Although an increased rejection of monosaccharides is benecial for their recovery from the hydrolysate, the rejection of inhibitor also increased, and so it is apparent that the selection of a suitable permeate ux plays an important role in the NF process. As such, we moved on to optimize the permeate ux, as described in the following subsection.
4.4.2 Multi-objective optimization of the permeate ux (j v ) and operating time (t) during the diananoltration process. To further improve the purities of the monosaccharides present in the retentate, a dialtration step was introduced for inhibitor removal. Multi-objective optimization on the basis of DSPM-DE model was then carried out to select a suitable permeate ux and operating time. Three objective functions were selected, namely maximized Pr inhibitor and Y sugar , and minimized EC. The permeate ux j v and the operating time t are the two decision variables, where the upper value of the j v was 2.4 L m À2 min À1 Fig. 4 Effect of Na 2 SO 4 concentration on the rejection of the various solutes present in the hydrolysate. A permeate flux of 1.35 L m À2 min À1 was employed along with a pH of 3 and a feed temperature of 25 C (TMP is transmembrane pressure). and the diananoltration time was limited to 40 min. A total inhibitor removal rate (RM inhibitor ) of $90% was set as the constraint for the purication of monosaccharides. To reduce the search region and prevent the generation of unrealistic results, a total monosaccharide removal rate (RM monosaccharides ) of #35% was set as an additional constraint.
A parallel optimization strategy was proposed for this multiobjective optimization study. The optimization ow chart is shown in Fig. 6. More specically, the decision variables were discrete rstly. By systematically scanning a dense grid of variable values (10 000), approximately 3320 values were then found to fulll the set constraints. Among these values the individual operating conditions were nally obtained corresponding to the maximum Y sugar , maximum Pr inhibitor , and minimum EC values (i.e., P3, P2, and P1 in Fig. 6). The optimization process was carried out using MATLAB Soware (MathWorks). Fig. 7 shows the resulting 3D plots in which all the points can fulll the design constraints, but no points exist which can meet all the optimization objectives simultaneously. For a better view, the projection planes are presented in Fig. 8. As can be seen in Fig. 8a, the Pr inhibitor increases with an increase of EC. This is because increasing the water consumption results in greater quantities of inhibitors passing through the membrane. The maximum Pr inhibitor value locates in point P2, while P1 and P3 have similar Pr inhibitor values. Similarly, Fig. 8c shows the correlation between Pr inhibitor and Y sugar . With the increase of Y sugar , the Pr inhibitor increases rst and then decreases. The maximum of Y sugar lies in the point P3. The Y sugar depends strongly on the water consumption EC (see Fig. 8b). Increasing the EC the Y sugar decreases. This is due to the fact that water enhances the permeation of the monosaccharide into the ltrate. Thus, to increase the recovery rate of monosaccharide, the EC should be low, and at the same time the operating time should be short, and the permeate ux should be small as well. The minimum EC lies in the point P1. However, the minimum EC (P1) and maximum Y sugar (P3) points are very close. To analyze the optimization results, we can conclude that the objective functions in terms of Pr inhibitor , Y sugar and EC contradict one another. The optimal operating conditions were actually non-existed. In such case, only suitable operating  conditions can be selected among the P1, P2, and P3 points. Table 2 compares the Pr inhibitor , Y sugar and EC values of the points P1, P2, and P3. It can be observed P1 and P3 give the similar performance. The aim of this work is to obtain the monosaccharide as much as possible. Hence, the operating conditions of P3 were selected as the optimal conditions which would be veried use of real hydrolysate solution as the feed.

Verication of the optimized result.
To verify the optimized result, the diananoltration mode was used to process the real hydrolysate solution which have a lower pH (pH < 1). The concentration proles are shown in Fig. 9. Due to the complexity of the hydrolysate solution, the rejection for each solute was higher than the model predictions. The operating time of 35 min and 13 L of water were required to reach the target values. However, the monosaccharide loss was lower than the predicted value, with only 6.45% glucose, 9.25% xylose, and 9.47% arabinose. The deviations of approximately 4% were calculated for arabinose, glucose, H 2 SO 4 , formic acid, HMF, and furfural, while the deviation of acetic acid was >7%. Aer ltration 0.78 g L À1 acetic acid was detected in the retentate, the furan concentration was also reduced to 0.08 g L À1 , and the total monosaccharide concentration in the retentate was in the range of 85-90 g L À1 . These results showed that the change of solution pH might have no signicant inuence on the separation of monosaccharides and inhibitors when pH < 3. We could therefore conclude that the optimized operating conditions were feasible to deal with the hydrolysate solution. The remaining sulfuric acid (around 72%) has to be separated from monosaccharide by the following electrolyte exclusion chromatography.

Monosaccharide and sulfuric acid recovery by electrolyte exclusion chromatography
The chromatographic recovery of both sulfuric acid and the monosaccharides from the hydrolysates pretreated by nano-ltration was then examined using batch column experiments, and an elution chromatogram of the hydrolysate obtained using a strong acid cation-exchange resin is shown in Fig. 10. As indicated, sulfuric acid was eluted rst, with the breakthrough point of the sulfuric acid peak being close to the void volume of the resin bed. Subsequently, all monosaccharides were eluted simultaneously due to their similar structures, and this was followed by the elution of acetic acid. Formic acid, HMF, furfural, and the phenolic compounds were not considered here due to their low concentrations in the diananoltrated hydrolysate.
Following pretreatment of the hydrolysates by nano-ltration, the sulfuric acid concentration was reduced to 0.3 mol L À1 , thereby indicating that the electrolyte exclusion is sufficiently strong to prevent the SO 4 2À ions from entering the resin pores. In addition, due to the electroneutrality of the solution, cations were also unable to enter the pores, thereby resulting in the poor adsorption of sulfuric acid onto the resin and consequently, its rapid elution. It should also be noted that some overlap was observed between elution of the sulfuric acid and the monosaccharides, in addition to between the monosaccharides and the acetic acid. Following the recovery of 90.37% sulfuric acid (98% pure), the overall monosaccharide yield and purity were 94.87% and 95.6-98.5%, respectively. We  therefore expect that to achieve a further increase in the yields and purities of the monosaccharides and the sulfuric acid, a continuous chromatography process would be required.

Conclusions
We herein described the successful application of a combined membrane-chromatography process for the removal of inhibitors and the recycling of monosaccharides and sulfuric acid from dilute acid corncob hydrolysates. Initially, we investigated the effect feed pH, permeate ux, and Na 2 SO 4 concentration on the retention/rejection of monosaccharides and inhibitors in a model solution to obtain the optimal conditions for the nanoltration process. More specically, we found that optimal separation of the carboxylic acids and furans from the monosaccharides was achieved at pH 3, while carboxylic acid and furan rejection decreased upon increasing the Na 2 SO 4 concentration. In addition, coupling of the Donnan steric pore and dielectric exclusion model with mass balance measurements was successful both in predicting solute rejection at different permeate uxes and in simulating the diananoltration process. Furthermore, to determine a suitable permeate ux and operating time for the process, multi-objective optimization was carried out to obtain the maximum total inhibitor removal efficiency, the maximum recovery rate of monosaccharides, and the minimum water consumption. Indeed, the operating conditions that maximized the monosaccharide recovery rate were optimal. Subsequently, a cheap strong acid cation-exchange resin (PS-DVB) was employed to recover both the monosaccharides and the sulfuric acid from the nano-ltered hydrolysate, with elution and column regeneration being facile using water as the eluent. The suitability of the optimized operating conditions was then conrmed using hydrolysate solutions, with nanoltration resulting in the removal of 90% of inhibitors, including HMF, furfural, phenolics, and carboxylic acids, in addition to the recovery of 93.55% glucose, 90.75% xylose, and 90.53% arabinose following treatment using a batch column packed with the strong acid cationexchange resin. Over the whole combined process, monosaccharide losses ranged from 10 to 15%, and the recovery of dilute sulfuric acid ranged from 65 to 70%. The recovered sulfuric acid was then added directly to the subsequent hydrolysis process, while the monosaccharides were continuously supplemented to the fermenter. As such, our results clearly demonstrated that the combination of nanoltration with electrolyte exclusion chromatography is an effective strategy for the removal of inhibitors and the recovery of monosaccharides from dilute acid corncob hydrolysates.

Conflicts of interest
The authors declare there is no conicts of interest regarding the publication of this paper. Hindrance factor for diffusion dimensionless m i Mass concentration of solute g L À1 m inhibitor Mass concentration of inhibitor g L À1 M monosaccharides Mass concentration of monosaccharides g L À1 Fig. 10 Elution chromatogram of the hydrolysates pretreated by nanofiltration. The feed temperature was maintained at 25 C and an elution flow rate of 1 mL min À1 was employed.

Abbreviation
This journal is © The Royal Society of Chemistry 2018