Environ. Sci. Technol. 2003, 37, 4435-4441
Rejection Efficiency of Water Quality Parameters by Reverse Osmosis and Nanofiltration Membranes WEIHUA PENG AND ISABEL C. ESCOBAR* Department of Civil Engineering, University of Toledo, Toledo, Ohio 43606
The objective of this study was to evaluate the effectiveness of reserve osmosis (RO) and nanofiltration (NF) membranes, under various solution chemistries, on water quality. The effects of organic carbon, divalent and monovalent cations, bacteria, and permeate drag on the rejection efficiencies of three different membranes were investigated through a series of laboratory bench-scale experiments. Quantitative models were successfully developed to predict the rejection of turbidity, divalent and monovalent cations, ultraviolet absorbance at 253.7 nm (UV254), and dissolved organic carbon (DOC) by membrane filtration. It was found that mechanical sieving (measured as molecular weight cutoff, MWCO) and electrostatic interactions were the most significant parameters since they were found to be important in nearly all models developed. For negatively charged membranes, under high ionic strength solution environments that repress electrostatic interaction between charged compounds and membranes, passage of compounds was mainly a function of size exclusion (i.e. MWCO). Further, of the feedwater parameters tested, bacteria concentration was observed to be the most significant influence on UV254, divalent cation and monovalent cation rejections. The developed models revealed that interactions between feedwater composition and membrane properties impacted the rejection efficiency of membranes as significantly as water composition and membrane properties individually.
Introduction As water becomes scarce and treatment regulations more stringent, membrane processes have drawn more attention because of their strong separation capabilities. Reverse osmosis (RO) and nanofiltration (NF) membranes are able to effectively remove most organic and inorganic compounds and microorganisms from raw water (1, 2) and have been applied to drinking water treatment and wastewater reclamation. Thus, to meet target contaminant removal goals, the rejection capability of membranes is a very important parameter to evaluate membrane processes. Membrane rejection is influenced by feedwater composition and membrane properties as well as the interactions between them (3, 4). Kwak et al. (5) determined that smoother surfaces with irregular ambiguous nodules led to higher water fluxes and lower rejections, whereas rough surfaces with uniform distinct nodule structures contributed to higher * Corresponding author phone: (419)530-8267; fax: (419)530-8086; e-mail:
[email protected]. Current address: Chemical and Environmental Engineering Department, University of Toledo, 3055 Nitschke Hall - Mail Stop # 305, Toledo, OH 43606-3390. 10.1021/es034202h CCC: $25.00 Published on Web 09/06/2003
2003 American Chemical Society
rejections. Yeom et al. (6) found that the electrostatic interactions and molecular sieving were important rejection mechanisms for membranes. The highly charged electrostatic double layer deposited on the surface of the membrane and natural organic matter (NOM) adsorbed to the membrane has been found to influence the hydrophobicity and surface charge of the membrane and, thus, impact rejection (7-14). In addition, Zander et al. (15) reported that the presence of divalent cations could decrease the rejection of both conductivity and organic matter, and low ionic strength solutions resulted in high organic carbon and conductivity rejections along with the high rates of flux decline. The membrane skin shrinkage in concentrated salt solutions was also observed to affect the rejection behavior of membranes (16). Microorganisms in feedwater are another important issue in membrane processes. They can attach and grow on the surface of the membrane to reduce the membrane flux and to decrease solute rejection because of the enhanced concentration polarization within the membrane biofilm (17). Finally, operation conditions, such as permeate flux (i.e. permeate drag), have also been found to influence the rejection capability of membranes (18, 19). In raw waters, various microorganisms and organic and inorganic compounds exist simultaneously. Thus, there must be complex relationships involved in the rejection efficiency of membrane processes. The objective of this study was to employ a statistical design simulating RO and NF membrane treatments of different feedwaters to develop quantitative models for predicting membrane rejection.
Experimental Section Model Development. For the evaluation of membrane rejection, the percent reduction of each target material concentration was calculated using eq 1.
(
R) 1-
)
Cp × 100 Cf
(1)
where R is the percent rejection of the target material and Cp and Cf are permeate and feed concentrations, respectively. The rejections of turbidity, divalent cation, monovalent cation, UV254, and dissolved organic carbon (DOC) were chosen as the variables of interest for indicating membrane rejection because they were are essential measurements of water quality. A total of 27 experiments were performed for model development according to a fractional factorial design. After all experiments were done, analysis of variance (ANOVA) was performed to evaluate the significance of the main factors and their two-way interactions with the level of significance at 0.05. Quantitative linear models were, then, developed using the experimental data based on the significant factors identified in the ANOVA and included the following quantitative parameters: (1) molecular weight cutoff (MWCO), roughness, and zeta potential were used to characterize membranes; (2) DOC, total organic carbon (TOC), turbidity, UV254, specific UV (SUV, equal to the ratio of UV254 and DOC), and humic acid component were employed to describe different organic compositions; (3) hardness as CaCO3; (4) total solved solids (TDS); and (5) bacteria concentration. In addition, the initial permeate flux of each experiment was included in the predictor pool to reflect the effect of operational conditions. All the potential predictor variables and their possible two-way interactions were considered in the model development process. The stepwise model deVOL. 37, NO. 19, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4435
TABLE 1. Membrane Characteristics membrane type chemical structure function groups molecular weight cutoff hydrophobic/hydrophilic zeta potential (pH ) 7.0) average surface roughness
TFC-S
ESPA1
NTR 7450
NF TFC polyamide carboxylate carb. acid ∼200 Da hydrophilic -9.3 mv 35.64 nm
RO aromatic polyamide carboxylic acid and amide 100∼300 Da hydrophobic -2.87 mv 78.93 nm
NF poly (ether sulfone) N/A 500 Da hydrophobic -6.26 mv 27.84 nm
velopment method was employed to seek the most appropriate models. Selected models were found to be significant based on global F-statistics and had the approximately highest adjusted multiple determination coefficients (R2), the relative lowest mean square of errors (MSE), and a reasonable number of independent variables in the model. After the appropriate models were built, paired-samples t-tests at the level of significance of 0.05 were conducted to evaluate the difference of predicted variables and actual variables, assuming the relative frequency distributions of each variable’s populations were approximately normal. It computed the differences between values of predicted and actual variables for each case and tested whether the average differed from 0. In addition, the Wilcoxon signed ranks test, a nonparametric test, was used at a 95% confidence level to eliminate constraints associated with paired t-test assumptions. Nonparametric tests do not depend on the distribution of the sampled populations, and they focused on the location of the probability distribution of the sample population. The Wilcoxon signed ranks test had the null hypothesis that two population probability distributions were identical against the alternative hypothesis that one was shifted to the right or left of the other; thus, different. Significant predictive models were developed for turbidity, divalent cation, monovalent cation, UV254, and DOC rejections. All other parameter rejections were not predictable by the independent variables chosen. Thus, other factors must be significant in their rejection mechanisms. The analysis was performed using SPSS 11.0 software (Chicago, IL). Synthetic Raw Water. A matrix of 27 synthetic feed solutions was prepared to reflect various feedwater conditions (20-22): (1) organic composition: (a) DOC concentration of 138.68 mg/L and 31% humic acid; (b) DOC concentration of 67.5 mg/L and 74% humic acid; and (c) DOC concentration of 287.75 mg/L and 0% humic acid; (2) hardness concentrations of 1000 mg/L, 3000 mg/L, and 5000 mg/L as CaCO3; and (3) monovalent cation concentrations of 3000 mg/L, 6000 mg/L, and 9000 mg/L. High concentrations of these components were employed to account for bulk feed concentrations and back diffusion, to reflect the full-scale increase in solute concentration at the membrane surface, and to account for membrane treatment processes that contain a second stage that uses first stage concentrate. In addition, pure bacteria strains, Pseudomonas fluorescens strain P17 and Spirillum volutans strain NOX, were chosen to be inoculated in raw water with a total concentration of 2000 CFU/mL (colony forming units per milliliter) because they are able to metabolize a wide variety of organic carbon compounds (23). They led to biofilm formation on the membrane and reflected the effect of biofilms on membrane rejection. Four liters of synthetic feedwater were prepared for each experiment. pH of the solution was adjusted to approximately 7 using 0.1 M NaOH solution. To avoid effects of unexpected microorganisms, the synthetic feedwater was pasteurized for 1 h at 70 °C before the test bacteria were seeded. The 4436
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 19, 2003
prepared raw water was incubated at 25 °C for 3-5 days before use to allow bacteria to grow. Membranes. Three membranes, TFC-S (Koch Membrane, San Diego, CA), ESPA1 (Hydranautics, San Diego, CA), and NTR7450 (Hydranautics, San Diego, CA), were chosen because they are commonly used in water industries and have distinctly different properties (Table 1). To assess changes in membrane surface charge at various solution chemistries, membrane surface zeta potentials were determined using a streaming potential analyzer (BI-EKA, Brookhaven Ins., Holtsville, NY), located at the University of Central Florida, Orlando, FL. The zeta potential of the membrane surface was calculated from the measured streaming potentials using the Helmholtz-Smoluchowski equation (13). The pH titration was first performed at a background electrolyte concentration of 10-2 M NaCl to determine the isoelectric point of the membrane. The matrix solution was distilled deionized water with 10-2 M NaCl as electrolyte background. For pH titrations, HCl and NaOH were added to vary the pH from 4 to 10. AFM was used to examine the surface structure and morphology of the membranes. The AFM investigation was conducted using a Nanoscope IIIa Scanning Probe Microscope (Digital Instruments, Santa Barbara, CA). It was equipped with a surface roughness determination and peak counting function that assisted in determining the impact of the irradiation on the membranes. Ra, the mean value of the surface relative to the center plane, was calculated using
Ra )
1 LxLy
∫∫ Ly
0
Lx
0
|f(x, y)|dx dy
where ×a6(x,y) is the surface relative to the center plane and Lx and Ly are the dimensions of the surface. Other characteristics of the membranes were provided by the manufacturers. Of specific interest was the measurement of MWCO. Manufacturers used the dextran procedure, ASTM E1343-90, to determine MWCO values ranging from 100 to 4500 Daltons. Flat Sheet Test Apparatus. All experiments were performed in a batch mode using a flat sheet test unit (Osmonics, Minneetonka, MN) having a cell with 155 cm2 active membrane area. The concentrate was recycled back to the reservoir, and the permeate was collected. New membranes were used in every experiment, and before an experiment was started, the membrane was rinsed with deionized (DI) water and then was soaked in DI water overnight (24). TFCS, ESPA1, and NTR7450 were run at their normal operating pressures of 280, 120, and 120 psi, respectively. The feedwater was maintained at 35 ( 5 °C by a cooling water bath. Before filtration of the synthetic feedwater, membranes loaded in the unit were precompacted by filtering 2 L of DI water. After precompaction, 3 L of the synthetic feedwater was run through the test unit. Measured Parameters. Approximately 150 mL of the raw water was collected before the feedwater was loaded, which was tested to determine feedwater quality. Then, the first 2
TABLE 2. Chosen Model’s Regression Statistics dependent variable turbidity rejection divalent cation rejection monovalent cation rejection UV254 rejection DOC rejection
adjusted R2 F-statistics P-value
MSE
0.86 0.67 0.62
43.36 16.19 14.92
