Nondeterministic Computational Fluid Dynamics Modeling of

May 4, 2015 - Log In Register · Cart · ACS · ACS Publications · C&EN ..... CURRENT ISSUELATEST NEWS. COVER STORY ... BUSINESS CONCENTRATES ...
0 downloads 0 Views 3MB Size
Article pubs.acs.org/est

Nondeterministic Computational Fluid Dynamics Modeling of Escherichia coli Inactivation by Peracetic Acid in Municipal Wastewater Contact Tanks Domenico Santoro,*,† Ferdinando Crapulli,† Mehrdad Raisee,‡ Giuseppe Raspa,∥ and Charles N. Haas§ †

Department of Chemical and Biochemical Engineering, Western University, London, Ontario, Canada N6A 3K7 Center of Excellence in Design and Optimization of Energy Systems (CEDOES), School of Mechanical Engineering, College of Engineering, University of Tehran, Tehran, Iran § Department of Civil, Architectural and Environmental Engineering, Drexel University, Philadelphia, Pennsylvania 19104, United States ∥ Department of Chemical Engineering, Material and Environment, La Sapienza University, 00185 Rome, Italy ‡

S Supporting Information *

ABSTRACT: Wastewater disinfection processes are typically designed according to heuristics derived from batch experiments in which the interaction among wastewater quality, reactor hydraulics, and inactivation kinetics is often neglected. In this paper, a computational fluid dynamics (CFD) study was conducted in a nondeterministic (ND) modeling framework to predict the Escherichia coli inactivation by peracetic acid (PAA) in municipal contact tanks fed by secondary settled wastewater effluent. The extent and variability associated with the observed inactivation kinetics were both satisfactorily predicted by the stochastic inactivation model at a 95% confidence level. Moreover, it was found that (a) the process variability induced by reactor hydraulics is negligible when compared to the one caused by inactivation kinetics, (b) the PAA dose required for meeting regulations is dictated equally by the fixed limit of the microbial concentration as well as its probability of occurrence, and (c) neglecting the probability of occurrence during process sizing could lead to an underestimation of the PAA dose required by as much as 100%. Finally, the ND-CFD model was used to generate sizing information in the form of probabilistic disinfection curves relating E. coli inactivation and probability of occurrence with the average PAA dose and PAA residual concentration at the outlet of the contact tank.

1. INTRODUCTION It is well-known that chlorine-based disinfection processes have played a very important role in the last century in improving public health protection and sanitation worldwide;1 however, the documented formation of chlorinated byproducts2,3 has precipitated environmental and public health concerns that have resulted in a more favorable consideration of alternative disinfectants such as ultraviolet light (UV) and peracetic acid (PAA). PAA is an organic peracid with the chemical formula CH3CO3H. At room temperature, it is a clear, colorless liquid with an oxidation potential of 1.81 eV and a pKa of 8.2. It can be manufactured either by reaction of acetic acid, acetyl chloride, or acetic anhydride with hydrogen peroxide (H2O2) (in the presence of a sulfuric acid catalyst) or by direct oxidation of © 2015 American Chemical Society

acetaldehyde. PAA is commercially available as an equilibrium mixture in water also containing hydrogen peroxide and acetic acid. As the PAA equilibrium tends to quickly revert toward acetic acid and hydrogen peroxide, stabilizers such as 1-hydroxyethylidene1,1-diphosphonic acid or 2,6-pyridinedicarboxylic acid are often employed.4 The use of PAA has been documented for a variety of applications, including color removal in textile effluents,5 surface sterilization in food processing,6 and Legionella control Received: Revised: Accepted: Published: 7265

December 9, 2014 March 26, 2015 May 1, 2015 May 4, 2015 DOI: 10.1021/es5059742 Environ. Sci. Technol. 2015, 49, 7265−7275

Article

Environmental Science & Technology in hospitals.7 In the past decade, PAA has gained popularity for municipal wastewater disinfection, as reported by a number of investigators who have conclusively demonstrated good antimicrobial properties against a wide range of microorganisms, including bacteria,8,9 viruses,10 and protozoa.11 However, its mode of action is still debated and not yet thoroughly understood. Baldry and Fraser12 suggested that the PAA disrupts the chemiosmotic integrity of the lipoprotein cytoplasmic membrane and allows ion transport between the cell interior and its environment through a shift or rupture of the cell wall. It has also been hypothesized that bacterial inactivation could be augmented by hydroxyl radicals formed during transition metal-catalyzed PAA decomposition.13 Studies have also been conducted to compare the disinfection efficiency of PAA to those of other disinfectants. Stampi et al.14 evaluated the performance of PAA and chlorine dioxide for the inactivation of microbial indicators such as total coliforms, Escherichia coli, Enterococcus, and Salmonella, concluding that PAA was more efficient than chlorine dioxide, and that the extent of coliform inactivation generally increased with temperature and decreased with high 5-day biochemical oxygen demand (BOD5). Veschetti et al.15 reported that the microbial activity of PAA against fecal Streptococci and coliphages was lower than that of hypochlorite. Lubello et al.16 evaluated the possibility of combining PAA with UV and H2O2 to disinfect secondary wastewaters for municipal effluent reuse in agriculture. Their results showed that the combination of PAA and UV was more effective than the combination of PAA and H2O2. Koivunen and Heinonen-Tanski10 studied PAA, H2O2, and sodium hypochlorite (NaClO), alone and in combination, against several microbial targets, namely, E. coli, Enterococcus faecalis, Salmonella enteritidis, and MS2 coliphage. They concluded that the combined use of PAA and UV achieved significant synergistic disinfection against enteric viruses, while H2O2 and UV improved microbial inactivation only slightly when compared to that with UV alone. PAA has also been reported to be a viable chlorine-alternative disinfection process for treating municipal wastewater for agricultural reuse,17 particularly when it is applied to tertiary filtered effluents where good disinfection efficiency was reported even at very low PAA concentrations (0.5) was displayed by the following pairs: Np and kPAA, Np and DPAA, Np and kp, and kp and kPAA. The observed correlation between Np and kPAA (0.501) and between Np and DPAA (0.599) suggests that an increase in the level of particle-associated coliform

3. RESULTS AND DISCUSSION 3.1. Kinetic Parameters. Figure 2 shows a typical subset of batch inactivation experiments after their reassessment using the model-based analysis presented in section 2.2. As expected, 7269

DOI: 10.1021/es5059742 Environ. Sci. Technol. 2015, 49, 7265−7275

Article

Environmental Science & Technology

Figure 3. Observed and stochastically simulated cumulative distributions function (CDFs) of Np, Nd, kp, kd, kPAA, and DPAA.

could be attributed to the fact that the decomposition of matrix Σ returned a small negative eigenvalue (−0.074). To make the latter matrix positive definite, a minor modification was necessary (i.e., the negative value was replaced with zero). Thus, the multivariate Gaussian model used for the simulation of the sextuples should be regarded as an approximation of the joint distribution of the kinetic parameters in a Gaussian framework. As a further verification step of the stochastic inactivation model proposed in this study, the observed and simulated E. coli log removals were compared in Figure 4, confirming that the model was able to faithfully reproduce the magnitude and the extent of variability observed for both the particle-free and tailing regions of the batch inactivation curves. Additionally, the observed and simulated E. coli inactivation data were quantitatively compared to determine whether the stochastic model was able to accurately represent the PAA dose-dependent variability observed in Figure 4. To do so, the first step was to remove the log-linear trend from each set of data by subtracting the value returned by the fitted inactivation model (at each PAA dose) from the corresponding discrete value (Figure S3 of the Supporting Information). Then, the detrendized data were subjected to a normality test, which indeed confirmed that residuals belonged to normally distributed populations with means equal to zero (Figures S5 and S6 of the Supporting Information). This allowed the hypothesis of equality of variances between the two detrendized sets of data using a statistical F test to be tested. By stating as the null hypothesis H0: σobserved = σstochastic, we were able to conclude that the variances associated with the two sets of data were equal at the 95% confidence level (see the Supporting Information).

bacteria, as would occur when the quality of the secondary settled wastewater deteriorates, is accompanied by a greater disinfectant demand and disinfectant decay rate. Such a trend is also confirmed by the inverse correlation existing for the kd−DPAA pair (−0.305), suggesting an effluent with a lower disinfectant demand, DPAA (typical of a wastewater that has been biologically oxidized well and settled well) displaying a greater inactivation rate of dispersed microorganisms. There is also a significant correlation (0.601) for the Np−kp pair indicating that a change in the concentration of particleassociated E. coli (Np) could be accompanied by a positively correlated change in their inactivation rate constant (kp). Such behavior could be explained by considering that an effluent characterized by a low concentration of particle-associated E. coli may have microorganisms embedded in larger particles or flocculated biomass. These larger particles offer a greater barrier to disinfectant penetration and hence a lower inactivation rate, leading to the positive correlation between Np and kp. Finally, a positive correlation (0.734) between kPAA and kp is also to be expected, because of the mathematical relationship existing via Np between the positively correlated pairs of Np and kPAA (0.501) and Np and kp (0.601). 3.2. Stochastic Inactivation Kinetics. To verify the accuracy of the simulated stochastic kinetics, the observed and predicted correlation matrices were compared, as well as the distributions of the six kinetic parameters involved in the model. As shown by Table 1 and Figure 3, the distribution and statistical dependency of the simulated sextuples were in good agreement with the observed data. In the case for the Np−DPAA, DPAA−kPAA, and Np−kPAA pairs, the discrepancies between observed and simulated correlations 7270

DOI: 10.1021/es5059742 Environ. Sci. Technol. 2015, 49, 7265−7275

Article

Environmental Science & Technology

process occurring at the mesoscale. CFD disinfection simulations were conducted for a constant average PAA dose of 47.8 mg L−1 min and an initial PAA concentration of 4.5 mg/L. The inactivation kinetic parameters were also kept constant and equal to the observed 50th percentiles taken from their distributions. In Figure 6, contours of E. coli inactivation for the two contact tanks are presented for a horizontal midplane. In both tanks, recirculating zones corresponding to large PAA doses were clearly visible. Such regions, induced by the presence of longitudinal baffles, were more evident for the case of the four-pass reactor because of a greater cross-sectional velocity characterizing this contact tank resulting in a reattachment point of the recirculating region located further along the channel than in the case of the eight-pass contact tank having a lower crosssectional velocity. Such recirculating zones are generally undesirable as they correspond to the higher-retention time regions where the PAA dose is large and the disinfection byproduct formation potential could also increase.17 Because of the low Morrill dispersion index displayed by both contact tanks (Table 2), the E. coli inactivation predicted by the Eulerian model reached a similar extent for both contact tanks (2.56 and 2.54 logs for the four-pass and eight-pass reactors, respectively), confirming that both disinfection systems were characterized by a nearly ideal plug flow behavior. Additional Eulerian CFD simulations conducted for two sets of kinetic parameters corresponding to the 25th and 75th percentiles (not included in the manuscript for the sake of brevity) confirmed a similar disinfection efficiency of both contact tank designs. 3.4. NDMF Analysis. NDMF simulations were conducted for two different secondary settled effluent qualities, namely (a) an effluent of average quality (Avg-SSE) represented by the entire population of simulated sextuples and (b) an effluent of high quality (High-SSE) represented by sextuples where the number of particle-associated E. coli was minimal and the rate of inactivation maximal. Simulation results were qualitatively compared against previously published PAA pilot data gathered in different geographies with wastewaters of different qualities. As shown in Figure 7, the variability measured in the field compared well with that from the simulation using the ND-CFD model. Finally, the ND-CFD model was used to generate probabilistic disinfection tables relating the PAA dose, E. coli inactivation, and probability of occurrence for Avg-SSE and High-SSE (see Table S5 of the Supporting Information). It was found that the probability of occurrence plays an equally important role in determining the PAA dose when compared to the fixed E. coli concentration target. For example, for an Avg-SSE effluent, the PAA dose required for 2 log E. coli inactivation doubled from 60 to 120 mg L−1 min when the percentage of occurrence increased from 50 to 90%. This clearly highlights the importance of taking wastewater quality into account during process sizing when determining the PAA dose for meeting regulations. This is confirmed by the results shown in Table S5 of the Supporting Information, where a dose of 120 mg L−1 min was able to yield, on a 50th percentile basis, 2.8 log for Avg-SSE and up to 4.2 log for High-SSE. A similar table presenting effluent E. coli concentration data instead of log removal was also generated using the ND-CFD and is included as Table S6 of the Supporting Information. As another example highlighting the importance of taking into account disinfection variability during process sizing, we estimated the PAA dose required for meeting three hypothetical disinfection regulations having the same concentration

Figure 4. E. coli inactivation by PAA: observed and stochastically simulated batch inactivation data.

3.3. Deterministic CFD Simulations. Tracer experiments were conducted and compared with deterministic CFD simulations to assess whether the variability induced by uncertainty in flow distribution was negligible compared to that induced by microbial inactivation kinetics. The satisfactory agreement between simulated and empirical data, reported in Figure 5,

Figure 5. Comparison between simulated and experimental tracer tests in breakthrough and washout mode. Probes were installed in two different locations, p1 (end of the first pass) and p2 (end of the third pass).

indicated that the flow field distribution within the contact tank reactor could be treated and predicted deterministically. A quantitative comparison between the four-pass and eightpass contact tanks was conducted and is reported in Table 2. In general, CFD simulations were able to predict well the flow distribution in the two contact tanks showing a maximal error of