Environ. Sci. Technol. 2007, 41, 6547-6553
Three-Component Adsorption Modeling to Evaluate and Improve Integrated Sorption-Membrane Processes LANCE C. SCHIDEMAN,† BENITO J. MARIN ˜ AS,† V E R N O N L . S N O E Y I N K , * ,† SHAOYING QI,† AND CARLOS CAMPOS‡ Department of Civil & Environmental Engineering and Center of Advanced Materials for the Purification of Water with Systems, University of Illinois, 205 N. Mathews Ave., Urbana, Illinois 61801, and R & I Alliance, Suez Environment, 75009 Paris, France
Integrated sorption-membrane (ISM) processes combining low-pressure membranes with adsorbents are increasingly popular because they cost-effectively expand low-pressure membrane treatment to include dissolved contaminant removal. However, contemporary ISM processes often exhibit antagonistic tradeoffs between adsorption and membrane performance that were investigated using stateof-the-art adsorption models that include both of the predominant competitive effects of natural organic matter: direct site competition and pore blockage. Two currently used ISM process configurations, powdered activated carbon-ultrafiltration (PAC-UF) and adsorptive floc blanket reactor-ultrafiltration (FBR-UF), were compared with a novel configuration, upflow adsorption-ultrafiltration (UAUF), which consists of a moving-bed of granular activated carbon upstream of a membrane. Model simulations quantitatively compared performance and evaluated potential improvements for each configuration. For instance, using contemporary PAC-UF practices and 90% atrazine removal as a baseline, alternative membrane backwashing procedures can lower carbon usage rates (CURs) by 75% but may also reduce membrane hydraulic performance. Using the same baseline, FBR-UF can reduce CURs by 92% while simultaneously improving membrane performance via pretreatment; however, process size increases 10-fold. The novel UA-UF configuration only increases process size modestly, but can still yield CURs 96% lower than the PAC-UF baseline while simultaneously providing beneficial membrane pretreatment and improving sustainability features by reducing residuals.
Introduction The use of low-pressure microfiltration (MF) and ultrafiltration (UF) membrane processes for large-scale water treatment applications is increasing dramatically due to efficient pathogen removal, smaller space requirements than conventional filtration processes, and lower energy demand * Corresponding author phone: +1 217 333 4700; fax: +1 217 333 6968; e-mail:
[email protected]. † University of Illinois. ‡ Suez Environment. 10.1021/es070410o CCC: $37.00 Published on Web 08/11/2007
2007 American Chemical Society
than high-pressure nanofiltration and reverse osmosis processes. However, MF/UF systems do not directly reject most dissolved contaminants, and thus, several integrated sorption-membrane (ISM) processes have been developed that combine adsorbents like activated carbon with lowpressure membranes to augment dissolved contaminant removal (1-5). ISM processes have been successfully used in several full-scale applications (6), but system optimization can still yield substantial process improvements. The original ISM configuration is powdered activated carbon (PAC)-UF (PAC-UF), which involves direct addition of PAC to the membrane influent (1, 7), as shown schematically in Figure 1a. With PAC-UF, dissolved contaminants adsorb onto PAC particles that are removed by the membrane and discharged during membrane backwashing. PAC-UF successfully expands MF/UF treatment capabilities to include removal of dissolved organics and has advantages including simplicity, low capital cost, and small space requirements (hydraulic retention time (HRT) of 1-10 min). Disadvantages include an inability to recover the adsorbent from other residual solids, inefficient usage of adsorbent capacity, and potentially negative effects of PAC particles on membrane hydraulic performance (7-9). The adsorption inefficiency of PAC-UF results from antagonism between adsorption and membrane components because typical membrane backwash intervals (MBI, 15-90 min) result in carbon retention times (CRTs) that are too short to achieve PAC equilibrium capacity. A second contemporary ISM configuration consisting of a floc blanket reactor (FBR) upstream of the membrane system (FBR-UF) (2, 10) improves adsorption efficiency and process synergy. FBR-UF is shown schematically in Figure 1b, which indicates that fresh PAC can be added either upstream of the FBR (option A) or just upstream of the membrane system (option B, dashed lines). With option B, PAC accumulates in the membrane reactor, and is recycled with the rest of the membrane backwash to the upstream FBR. For a dissolved organic carbon (DOC) removal application, Campos et al. (11) showed that FBR-UF reduced CURs by approximately 90% in comparison to PAC-UF, due primarily to much longer CRTs in the FBR (10-120 h). FBRUF also provides coagulation and clarification pretreatment that synergistically benefits membrane hydraulic performance. The major drawback of FBR-UF is its significantly larger space requirements because of the FBR’s much larger HRT (30-60 min). FBR-UF also suffers from the inability to separate the adsorbent from other residuals. Recognizing the major advantages and disadvantages of current ISM processes, a new configuration called upflow adsorption-ultrafiltration (UA-UF) was recently proposed that consists of an upflow bed of granular adsorbent, e.g., granular activated carbon (GAC), upstream of a low-pressure membrane system, as shown in Figure 1c (12). The upflow adsorbent bed can be operated in a counter-current, movingbed mode with spent adsorbent removal from the inlet as fresh/regenerated adsorbent is added at the outlet. UA-UF has several key advantages that result from synergistic interactions between the process components. For instance, upflow, moving-bed adsorbers are known to increase adsorption efficiency (13) but also liberate some fines (14), which has greatly limited practical applications. UA-UF resolves this dilemma because the downstream membrane easily removes any GAC fines. The adsorbent bed also improves membrane performance (15) by providing pretreatment that is space-efficient (typical HRTs of 2-8 min). Finally, the ability to easily separate GAC from other residuals with UA-UF VOL. 41, NO. 18, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
6547
and two fictive fractions of natural organic matter (NOM) that cause distinct competitive effects: (1) direct competition for adsorption sites caused by a small, strongly competing fraction of NOM that reduces equilibrium capacity; and (2) pore blockage caused by the accumulation of large NOM in activated carbon pores that reduces adsorption kinetics. The COMPSORB models use the Freundlich isotherm relationship and the ideal adsorbed solution theory (IAST) to describe adsorption equilibrium, and the homogeneous surface diffusion model (HSDM) to describe adsorption kinetics. These models then use equilibrium and kinetic parameters that decline over time according to the amount of strongly competing and pore-blocking NOM adsorbed. Previous models for flow-through ISM processes were compromised by not including pore blockage effects. For this study, a new adsorption model, COMPSORBFBR, was developed following the same approach to simulate the FBR-UF configuration when PAC is added upstream of the FBR only (option A on Figure 1b). Option A defines a limiting condition of minimum performance with FBR-UF, and one previous study showed that the performance with option A often closely approximates option B performance (11). Past research has also shown that PAC adsorption kinetics in an FBR can be adequately described by assuming continuously stirred tank reactor (CSTR) conditions and using the HSDM with the initial and boundary conditions provided below (2, 10, 11):
FIGURE 1. Integrated sorption-membrane process schematics.
(
)
∂qi(r,t) Dsi(t) ∂ 2∂qi(r,t) ) 2 r ∂t ∂r r ∂r
(i ) 1, 2, 3)
(1)
qi(0 e r e R,t ) 0) ) qi,0
(2)
∂qi(r ) 0,t) )0 ∂t
(3)
qi(r ) R,t) ) qi,R
(4)
improves process sustainability by facilitating adsorbent regeneration/reuse. This study presents the first attempt to directly compare several different ISM process configurations using the recently developed COMPSORB suite of multicomponent competitive adsorption models to efficiently evaluate a broad range of conditions. COMPSORB models have advanced the state-of-the-art by accounting for natural organic matter (NOM) competitive effects on both equilibrium and kinetic parameters as a function of NOM surface loading, which is tracked separately and allows model parameters to be determined in short-term, independent tests (16, 17). This study introduces a new model version, COMPSORB-FBR, created to evaluate FBR-UF and continuous feed-and-bleed operations with PAC-UF. This study also presents the first COMPSORB model application to optimize PAC-UF operations using natural water derived parameters. Finally, this study highlights several advantages of the novel UA-UF process that make it an attractive ISM process alternative.
where, qi is the solid-phase concentration of component i at radial distance r within the PAC particle, t is the time of adsorption, Dsi is the surface diffusion coefficient of component i, R is the radius of the PAC particles, qi,0 is the initial solid-phase concentration of component i, and qi,R is the surface concentration at the external particle surface (r ) R). By defining a new time variable, (21)
Mathematical Modeling Description and Approach
qi(r,θi(t)) )
COMPSORB Adsorption Models for ISM Processes. Each ISM process in Figure 1 incorporates flow-through adsorption, where adsorbent is held in place while water passes through. While this approach increases the amount of time carbon is contacted with water (CRT) without increasing reactor size (HRT), it also exposes the adsorbent to a continuously replenishing source of NOM that magnifies its competitive adsorption effects. Past studies showed that NOM reduces both capacity and rate of trace compound adsorption in proportion to the amount of NOM adsorbed (16, 18). Adsorption models that accurately describe these transient competitive effects of NOM have recently been developed and experimentally verified for PAC-UF (17, 19) and GAC adsorbers with a fixed or moving-bed (16, 20). These models are called COMPSORB and COMPSORB-GAC, respectively, and both separately track the adsorption of a trace compound 6548
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 18, 2007
θi(t) )
∫ D (λ)dλ t
si
0
(i ) 1, 2, 3)
(5)
an analytical solution to eqs 1 through 4 can be obtained for the surface concentration profile within a PAC particle that has been in a steady-state CSTR for time t (22)
(
qi,R + (qi,R - qi,0)
∞
2R
∑ πr
(-1)j
j)1
sin
j
( ) jπr R
])
[
j2 π 2 θi(t) R2 (i ) 1, 2, 3) (6)
exp -
Eq 6 can be integrated over the spherical adsorbent particle and over the exponential residence time distribution of a CSTR, to yield the following result:
qi,CSTR(τ) ) qi,R - (qi,R - qi,0)
6 π2 τ
∞
1
∑j ∫ j)1
2
∞
0
[
exp -
]
j2π2 θi(t) - t/τ dt R2 (i ) 1, 2, 3) (7)
The Supporting Information for this study provides additional
TABLE 1. Parameters Used in Various Models to Simulate Atrazine Removal from NCEL Water Using F-400 Activated Carbona
parameter description model name trace (TR) compound (atrazine) Freundlich isotherm coefficient, K (µg/mg)(L/µg)(1/n) Freundlich isotherm exponent, 1/n initial concentration, C0 (µg/L) initial film mass transfer coefficient, kf,0 (cm/sec) initial surface diffusion coefficient, Ds,0 (cm2/sec) strongly competing (SC) NOM fraction Freundlich isotherm coefficient, K (µg/mg)(L/µg)(1/n) Freundlich isotherm exponent, 1/n initial concentration, C0 (µg/L) initial film mass transfer coefficient, kf,0 (cm/sec) initial surface diffusion coefficient, Ds,0 (cm2/sec) pore blocking (PB) NOM fraction Freundlich isotherm coefficient, K (µg/mg)(L/µg)(1/n) Freundlich isotherm exponent, 1/n initial concentration, C0 (µg/L) film mass transfer coefficient, kf (cm/sec) surface diffusion coefficient, Ds (cm2/sec) external surface pore blockage parameters minimum film mass transfer coef. faction, kxf,min film mass transfer coef. decay exponent, R (g/mg) intraparticle pore blockage parameters critical PB surface loading, qcr (mg/g) surface diffusion coef. decay exponent, β (g/mg) carbon physical characteristics particle diameter, dp (µm) particle density, Fp(g/mL) bulk density, Fb(g/mL) adsorption reactor parameters hydraulic retention time, HRT (min) membrane backwash interval, MBI (min) carbon retention time, CRT (h) empty bed contact time, EBCT (min)
floc blanket reactor-ultrafiltration (FBR-UF)b
upflow adsorption-ultrafiltration (UA-UF)
COMPSORB
COMPSORB-FBR
COMPSORB-GAC
18.6 0.393 9.9 n/a 2.16 × 10-13
18.6 0.393 9.9 n/a 2.16 × 10-13
18.6 0.393 9.9 1.66 × 10-3 6.93 × 10-13
18.6 0.393 175 (87.5 as C) n/a 2.16 × 10-13
18.6 0.393 175 (87.5 as C) n/a 2.16 × 10-13
18.6 0.393 175 (87.5 as C) 1.66 × 10-3 6.93 × 10-13
1.653 × 10-4 1.77 2213 (as C) n/a 2.00 × 10-13
1.653 × 10-4 1.77 2213 (as C) n/a 2.00 × 10-13
1.653 × 10-4 1.77 2213 (as C) 1.58 × 10-3 5.20 × 10-13
n/a n/a
n/a n/a
0.250 0.210
1.01 0.0244
1.01 0.0244
1.01 0.0244
10 0.76 n/a
10 0.76 n/a
515 0.76 0.50
5 22.5-180 n/a n/a
30 n/a 15-120 n/a
See EBCT n/a n/a 2-16
powdered activated carbon- ultrafiltration (PAC-UF)b
a Experimentally determined parameters and the procedures for obtaining them were reported in a previous study and used to verify the COMPSORB-GAC model16. n/a ) not applicable. b Lower Ds values were used for processes with PAC that were determined by batch kinetic test measurements with a pulverized version of F-400 GAC.
derivation details and discusses how to combine eq 7 with reactor mass balances and equilibrium isotherm equations to resolve the steady-state effluent concentrations of the three components tracked by COMPSORB-FBR. Model Input Parameters. A set of COMPSORB-GAC model parameters was developed in a previous study using atrazine as the trace compound, F-400 GAC, and a natural groundwater with a DOC of 2.4 mg/L from the Newmark Civil Engineering Laboratory (NCEL) in Urbana, Illinois (16). These parameters were used to successfully predict experimental breakthrough data for atrazine in a moving-bed GAC reactor at four different column depths, which makes them directly applicable to modeling the UA-UF process performance in this study. Table 1 lists this set of verified model parameters under the column heading for UA-UF. The same parameters were used for PAC-UF and FBR-UF model simulations except that a smaller particle diameter, dp, and lower initial surface diffusion coefficients, Ds,0, were used to reflect that these configurations use PAC. The dependence of Ds on particle size has been noted in past studies, (13, 23) and was confirmed for this case using batch kinetic tests with pulverized F-400 GAC (PGAC) and NCEL water that were reported elsewhere (16) for atrazine. Thus, simulations for PAC-UF and FBR-UF used the lower Ds,0 values measured for PGAC as listed in Table 1. The size of each ISM process was defined in terms of the hydraulic retention time (HRT) and is also listed in Table 1. The specific conditions investigated in this study were a normalized average atrazine effluent concentration (C/C0) range of 0.01-0.80 for all three ISM configurations (PAC-UF, FBR-UF, and UA-UF), MBI
of 45-180 min for PAC-UF, CRT of 15-120 h for FBR-UF, and EBCT of 2-16 min for UA-UF.
Results and Discussion Evaluation of PAC-UF Process Performance. PAC-UF is the simplest ISM process configuration and provides a baseline for evaluation of the other configurations. Figure 2a shows an example of the COMPSORB model output with effluent atrazine concentration plotted as a function of membrane filtration cycle time for two different PAC dosing methods- pulse and step. For a step (continuous) feed of PAC, the effluent concentration drops gradually over the filtration cycle as PAC accumulates in the reactor. With pulse dosing, the same overall carbon mass is added at the beginning of the filtration cycle, which leads to a rapid initial concentration decline, followed by a gradual increase as water continues to pass through the reactor. For both dosing methods, the accumulated PAC is typically wasted with the membrane backwash, and the process is then repeated with new carbon. Since both concentration profiles in Figure 2a are based on a 5 mg/L CUR and a 90 min MBI, they can be directly compared, and it can be calculated that pulse dosing lowers the average effluent concentration by about 40%. However, pulse dosing also requires individualized PAC feed piping to each membrane unit and intermittent PAC feed is more prone to clogging. Quantitative modeling with COMPSORB provides the information needed to determine when the additional cost and difficulty of pulse dosing are justified. To facilitate comparison of different ISM process configurations and summarize many model simulations for a VOL. 41, NO. 18, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
6549
FIGURE 2. Normalized atrazine effluent concentrations versus operating time for various ISM process configurations, operating conditions and a carbon usage rate of 5 mg/L: (a) PAC-UF; (b) FBR-UF and UA-UF. variety of treatment goals, it is helpful to convert the atrazine effluent data into a form with CUR plotted versus the normalized average atrazine effluent concentration (C/C0) as shown in Figure 3a. In this figure, reactor efficiency is quantified in terms of the CUR needed to meet a certain treatment goal, and a lower CUR indicates a more efficient process. Adsorption performance is presented for two PAC dosing methods (pulse and step) and a range of MBIs from a low of 45 min, reflecting the midpoint of current membrane operations, to a high value of 180 min, representing a “foreseeable future” value that may become attainable as membrane systems improve. These parameters can be manipulated by the system designer and have a major impact on adsorption performance. For instance, a horizontal line at a constant CUR of 20 mg/L shows that the average atrazine effluent concentration decreases approximately 10 times as the carbon dosing switches from step to pulse and the MBI increases from 45 to 180 min. The effect of MBI and carbon dosing method are discussed in the subsequent section. Effect of Membrane Backwashing Interval (MBI) and PAC Dosing Method on PAC-UF Performance. Contemporary PAC-UF systems generally use step dosing and waste PAC with each backwash cycle. Thus, MBI determines the CRT, which is the average amount of time PAC is contacted with water. MBIs typically range from 15 to 90 min and are determined primarily to maximize membrane hydraulic performance with little attention given to the adsorption impacts. Unfortunately, most commercially available PACs need several hours to achieve equilibrium adsorption capacity. For the range of simulated conditions in Figure 3a, increasing CRT by extending the MBI or by implementing pulse PAC dosing always improved adsorption performance. However, the impact of switching from step to pulse dosing was generally greater than extending the MBI for step dosing. 6550
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 18, 2007
FIGURE 3. Effect of treatment goal and various design parameters on carbon usage rates for the ISM configurations: (a) PAC-UF; (b) FBR-UF; (c) UA-UF.
FIGURE 4. Effect of membrane backwash interval (MBI), backwashing method, and carbon dosing method on PAC-UF carbon usage rates for 90% atrazine removal (avg. C/C0 ) 0.1). Figure 4 highlights the effect of MBI and PAC dosing method on adsorption performance for a fixed treatment goal of 90% contaminant removal (average C/C0 of 0.10). This figure shows that adsorption performance with step dosing benefits more from increasing MBI than pulse dosing,
but pulse dosing always maintains a distinct advantage. For the same MBI, changing to pulse dosing reduced CUR by 55-75%. Figure 4 also shows that for a constant CUR of 30 mg/L, pulse dosing can have a 22 min MBI, whereas step dosing requires an MBI that is eight times larger. This is important because membrane hydraulic performance generally decreases as MBI increases. The effect of longer MBIs on hydraulic performance is a complex function of the raw water quality, membrane characteristics, and membrane operating conditions that precludes predictive modeling at this time. However, a few published studies have investigated the effect and found that membrane flux was reduced by 10-35% for each time that MBI was doubled (24-26). Since the membranes are a significant cost component for PACUF, it is unlikely that a system design based on extending MBI to increase adsorption efficiency would be advantageous. Pulse PAC dosing is a more promising alternative that significantly improves adsorption performance without compromising membrane performance, although the practical problems of intermittent PAC feed to multiple locations must be addressed. Effect Of Alternative Backwashing Procedures On PACUF Performance. Because step dosing is simpler and less expensive to implement, it also makes sense to pursue step dosing alternatives that do not require increasing MBI. The lower adsorption performance of step dosing results from inefficient use of PAC that arrives in the membrane shortly before backwashing, and from a high effluent concentration at the beginning of the filtration cycle (See Figure 2a). This occurs because membrane permeate water typically used for backwashing is usually wasted at the end of the backwash cycle and the reactor refilled with membrane influent water. However, it is possible to refill the membrane reactor with permeate water, which has a lower contaminant concentration. This approach was modeled with COMPSORB, and the benefit is shown on Figure 4, where the net CUR needed to achieve 90% contaminant removal is 20-60% lower for the permeate refilling approach. Net CURs were calculated to account for additional water that must be filtered for permeate refill, which also requires slightly more membrane area (typically
