You are on page 1of 17
Ultraviolet disinfection of secondary wastewater effluents: prediction of performance and design Frank J. Loge, Robert W. Emerick, Mark Heath, Joseph Jacangeio, George Tchobanogious, Jeannie L. Darby ABSTRACT: An empirical uluaviolet (UV) disinfection model was {developed to predict coliform inactivation in unfiltered secondary waste- ‘water effluent. The model was developed in the tailing region of the UV dose-response curve for log survival values less than ~3. Input parameters to the mode include the suspended solids concentration, the ‘unfiltered UV transmittance measured at a wavelength of 253.7 nm, the influent coliform concentration before exposure t0 UV light, and the applied UV dose. In the mode, UV dose is calculated asthe average UV intensity within the reactor (using the point source summation method) multiplied by the exposure time (based on the assumption of approxi- ‘ate plug flow conditions). Because the model is empirical. it should be calibrated to establish the statistical significance ofeach input parameter before use in a particular situation and t0 implicitly account for other factors influencing disinfection performance. The model was calibrated with data collected from a small-scale UV reactor operated at two north ‘em California wastewater treatment plans The values of suspended solids concentration, unfiltered UV transmittance, and UV dose were found tobe statistically significant with te water quality data set consi cred. The correlation coefficient (R’) ofthe calibrated model was 0:79. ‘The calibrated model was then used (oilustrate a design approach that imegrates model uncertainty, wastewater vanabiity, and variable permit requirements. In this approach, the number of UV lamps, modules ‘banks, and channels can be optimized to provide cost-effective designs Several design examples are used to illustrate both disinfection and hheadloss considerations in the recommended design approach. Water Environ. Res, 68, 990 (1996). KEYWORDS: ultraviolet disinfection, modeling, design, Monte Carlo analysis, uncertainty analysis. Ultraviolet (UV) radiation has grown in popularity over the last 10 years as a method of disinfecting wastewater. Increas- ingly, UV disinfection systems are being used to meet relatively stringent effluent coliform permit criteria (or example, effiuent coliform concentrations of 240, 23, and 2.2 most probable num- ber [MPNJ/100 mL). In meeting stringent permit criteria, UV uviamp T (17 mm, OD) 75mm K 150mm Figure 2—Small-scale UV disinfection system. (a) Side view and (b) cross sectional view of reactor. 10 Lis. Each reactor, preceded by an inlet box and followed by ‘weir box, contains four 26.7-W (rated UV output) low pressure ‘mercury vapor lamps. Each lamp is encased in a fused quart sleeve, The lamp centers are spaced 75 mm (3 in.) apart in the reactor and are oriented parallel to the flow. A side view of the UY system and a cross sectional view of one reactor are shown im Figure 2. Wastewater enters the unit and passes in series through reactors 1, 2, and 3, respectively. By sampling the effluent of each reactor, disinfection performance data at three UV doses were available for a particular wastewater quality. Operation of UV system at field sites. Identical UV disin- fection systems were operated at WWTPs at both the University of Califomia, Davis (UCD) and Central Contra Costa Sanitary District (CCSD) to collect performance data. Both plants use complete-mix activated sludge processes. To test wastewaters of varying quality, effluent from several points in the activated sludge process were blended as necessary. A wide range of 902 secondary effiuent quality was encountered during the field test- ing as summarized in Table 1 ‘During each experiment, samples were collected at the inlet to the UV system and from the effiuent of each reactor. The collection and handling of samples was controlled to minimize bacterial contamination and photoreactivation. The time be- ‘ween sampling and beginning laboratory analysis was less than 20 minutes. Samples were refrigerated at 7°C to reduce biologi- cal activity. Analysis was completed within 4 hours of sampling. Analytical procedures. The water quality of each sample ‘was characterized by measuring the suspended solids (SS) con- centration, filtered (FT) and unfiltered transmittance (UFT), tur- bidity, and estimating coliform concentrations. The SS. were measured according to Standard Methods, 17th edition (method 2540 D). Transmittance at 253.7 nm was measured in a spectro- photometer (Shimadzu UV160U) equipped with a I-cm rectan- gular cell, Filtered transmittance was determined after filtering Water Environment Research, Volume 68. Number 5 Table 1—Range of values observed for water quality parameters measured at UCD and CCCSD during collection of calibration data for UV disinfection model. Value Water quality parameter Unit Range Typical Suspended sotds mgt 39-528 187 Unitered trensmitance = % 35.1-82.0, 60.3 Fitered transmitance % 56.0-65.3, 723 Turbicity tu 27-180 62 Infuent total colferm — MPN/100 mL. 24 000-5 080.000 1 500 000 Panicle diameter yim 075-125 NA amici number umber. 31252-92390 50.000 ‘concentration? “Using a Coulter Muitsizer Darby et al 1995). the sample through 2 membrane filter with a nominal pore size of 1 ym. Turbidity was measured with a Hach Model 21004 turbidimeter. “The multiple-tube fermentation technique was used to deter- mine the MPN of total and fecal coliform per 100 mL as per Standard Methods, I7th Edition (9221 B and C, respectively), ‘The Poisson equation was used co combine all replicates and dilutions for each sample to generate a single MPN value. The Poisson equation was also used to establish confidence imervals for each MPN value. For the majority of influent samples, a ‘minimum of five dilutions with five tubes per dilution were used. Generally, either 50 tubes of a single dilution or three dilutions with five tubes per dilution were used to enumerate coliform density after exposure to UV light. The large nomber ‘of tubes per single dilution were used to decrease the 959% confidence intervals associated with the MPN test. Minimum detection limits were less than 0.1 per 100 ml. for the 50 tube test and 2.2 per 100 ml. for the standard test. Determination of UV dose. The UV dose was determined using the following equatic D=Ixt @) where IV dose (inW- sem?) verage intensity of the germicidal UV energy (mW! em’) 1 = exposure time (s) The average UV intensity within the reactors at 100% lamp ‘output was calculated using the Point Source Summation (PSS) method (Jacob and Dranoff, 1970; Qualls and Johnson, 1983; U.S. EPA, 1986). Output intensity of UV lamps typically de- creases exponentially during the first 100 hours of operation. Because lamp output during this break-in period is somewhat unstable, 100% lamp output is generally referred to as the lamp ‘output at 100 hours. Results from the PSS method for the partic: ular reactor geometry and lamp configuration used in this re- search are shown in Figure 3. At the UCD WWTP, where the majority of data was collected, the UV system was operated for ‘a minimum of 30 minutes before sampling and was tuned off after sampling. Lamp aging and fouling were neglected because the UV lamps were new at the start of the project and were usy/August 1996 Logo at a UV intensity, mWviem® TTT 8 3 Fikored transmitance, % Figure 3—Point source summation (PSS) curve relating the average intensity within the UV reactor to the filtered transmittance of the wastewater (PSS curve based on a 2 by 2 array with a 75-mm centerline lamp spacing). cleaned before each sampling event. At the CCCSD WWTP. the UV system was operated continuously and the UV lamps were cleaned frequently throughout the duration of the project. ‘Lamp aging was accounted for using the UV lamp manufacturer recommended guidelines. Based on the results of a tracer study on the UV system, ‘minimal short-circuiting and longitudinal mixing were found to ‘occur within each reactor (Darby et af., 1993). Because each reactor behaved hydraulically as a plug flow reactor, exposure time was calculated by dividing the net reactor volume by the low rate through the system, The net reactor volume is equal to the volume of wastewater contained over the lamp arc length, Model Development Results from the experimental study, shown as a plot of log survival versus UV dose, ate presented in Figure 4. Although higher UV doses generally resulted in lower log survival, there {is no apparent simple relationship between UV dose (calculated using the PSS method) and log survival. Multiple water quality Log survival, og (NIN,) yoo 200-500-400 ~«S00=«600«700 ‘Average UV Dose, mWes/om* Figure 4—UV inactivation of total coliform in activated sludge effluent of varying quality (average UV dose de- termined using PSS method and assuming plug flow in the UV reactor). Loge et al ‘Table 2—Summary of coefficients for the calibrated UV disinfection model. Parameter Value n ~187 7 re a 0.976 > 4083 Coetcients used in model develooment™ RMSE 04 1 (observations) 149 z 1 y 18 zZ 20 * Coefficients used in Equations 4 anc 6 * Copticents used in Equation 8. variables and UV dose were integrated into an empirical model as discussed below to predict coliform densities after exposure to UV light. Funetional form of model. Based on an analyses of the experimental data, the following empirical model was devel: ‘oped (Emerick and Darby, 1993), integrating both UV dose and ‘multiple water quality parameters. to describe coliform densities in the tailing region afier exposure to UV light. N= fy “ where N = effluent coliform density after exposure to UV light (MPN/100 mL) £ = empirical water quality factor D = UV dose (mW: siem*) 1h = empirical coefficient related to UV dose ‘The water quality factor was hypothesized to have the following general functional form: f= A(SSYUFTI"N,)* (3) where $8 = suspended solids concentration (mg/L) UFT = unfiltered UV transmittance at 253.7 nm (%) No = influent coliform density prior to exposure 1© UV light (MPN/100 mL) A. a, b,c = empirical coefficients ‘The specific functional form of the water quality factor and the corresponding values of the model coefficients are determined through a multiple linear regression of water quality data col lected at a particular WWTP. ‘Variables of significance in model. The following functional form was found to best describe the water quality factor for the UCD and CCCSD WWTPs: f= A(SS(UFT)® 6) ‘The empirical coefficient values developed in this study are summarized in Table 2. ‘A high statistical significance was found for SS (P = 0.0001), ‘UFT P = 0.002), and UV dose (P = 0.0001) in Equations 4 904 and 5. P reflects the probability that the observed relationship is due to chance alone. P < 0.05 indicates a statistically significant correlation, and P < 0.01 indicates a very significant correlation, The influent coliform density (N,) was found to be statistically insignificant and was subsequently dropped from Equation 5. ‘The roles of SS, UFT. and No in predicting coliform inactivation are discussed below. Role of particulate matter. Both UFT and SS were found to be sismificant variables in describing UV disinfection perfor- ‘mance for the wastewater characteristics considered in this study, The significance of these variables is likely the result of using the PSS method to quantity UV dose. In theory, if the ‘ue UY dose reaching targeted organisms were known. « model could be developed to predict the response ofthe targeted organ isms based on UV dose alone. However. existing methods of determining UV dose provide only a rough approximation of the actual dose reaching targeted organisms. In the PSS method, the average intensity within @ particular reactor geometry is calculated as a function of FT. Both absorption and scattering of UV light by particulate material also reduce the actual UV dose reaching targeted organisms but are not accounted for in the PSS method. Therefore, the inclusion of both UFT and SS in the model compensates for the effects of particulate material ‘on UV dose. effects not quantified in the PSS method. Role of influent coliform densirs. The number of influent coliform was found to be statistically insignificant in the model, ‘The influent coliform density is composed of both “ree swim- ming" and “particle associated” coliform. In the range of rels- tively high UV doses used to develop the model. most likely all of the free swimming coliform were inactivated. Therefore. changes in disinfection performance were due t0 changes in the number of particle associated coliform inactivated. In the particular water quality used 10 develop the model, the number of panicle associated coliform were more directly relaicd to the concentration of particulate material than to the total number of influent coliform in the wastewater. Therefore, the inclusion of the SS term in the model not only accounts for the reduction in UV dose reaching targeted organisms but also represents changes in the number of particle associated coliform. Because the above conclusion may not necessarily be tne for WWTP, the statistical significance ofthe influent coliform con- centration in Equation 5 should be verified by calibrating the model with the wastewater characteristics at a particular WWTP. Fit of mode! to experimental data. One hundred fifty-one samples were used 10 calibrate the model. The R® value of Equation 4 is 0.79, indicating 79% of the variance in N ean be explained by the variability in SS, UFT. and UV dose. The predicted versus actual coliform densities for the data used t0 develop the model are presented in Figure 5. Ninety five percent of the points lie within one order of magnitude of precision ‘The lack of a complete fit of the UV disinfection model to the experimental data (that is. R® < 1) is due to both the empiti- cal nature ofthe model and the uncertainty associated with the estimation of the input parameters. As with any empirical ‘model, the UV disinfection model is based on a specified func ‘ional form with input parameters that are related 10, but not necessarily a direct measure of, the factors affecting the dose reaching targeted organisms. The input parameters of the water quality factor have been hypothesized to be a qualitative mea- sure of the impact of particulate material on the UW dose reach Water Environment Research, volume 68. Number S Predicted log survival Observed log survival Figure 5—Log survival plot illustrating fit of UV disinfec- tion model to experimental data. ing target organisms, but the input parameters do not specifically quantify all of the absorption and light scattering effects of particulate material. ‘The uncertainty associated with the estimation of the values of the water quality variables used in Equation and the average intensity and exposure time used in Equation 3 affect the ® value of the calibrated model (Equation 4). Othe various te niques used in measuring the water quality variables. the Pois- son equation, used for the estimation of coliform concentration ‘most likely has the highest level of uncertainty. Inereasing the ‘number of tubes per dilution generally reduces the uncertainty ‘associated in the estimate of the coliform concentration. The ‘uncertainty associuted with the estimation of the suspended sol ids concentration can generally be reduced by increasing the umber of replicate samples. The uncertainty in the estimation of the unfiltered and filtered transmittance is strictly dependent ‘on the quality of the spectrophotometer. The exposure time used in the calculation of UV dose is based on the assumption of approximate plug flow conditions, Because deviations from ap- proximate plug flow conditions wil affect the accuracy of the UV disinfection model, tracer tests should be conducted before calibrating the UV disinfection model to verify this assumption. Calculation of the average intensity within the UV reactor using the PSS method is based on the assumption that the wastewater is ideally mixed in the radial direction along the overall arc length in a UV channel. The degree of radial mixing within 2 UV reactor has not been thoroughly investigated at this time ‘and is consequently a furure area of research. Because the UV disinfection model is empirical and the un- certainties associated with the estimation of the input parameters to Equation 4 are likely to change at different WW'TPs, the model should be calibrated before use ata particular WWTP. An. acceptable R? value is atthe discretion of the design engineer. A recommended minimum R? value is 0.7. Graphical illustration of model. Sithough the UV disinfee JuiyiAugust 1896 Loge at a. tion model is described by Equations 4 and 6. a graphical illus- tration of the model is insightful. Equations 6 and 4 are illus- trated graphically in Figure 6, a and b, respectively. As shown in Figure 6a, as water quality worsens (that is, as either SS increases or UFT decreases), the water quality factor increases. ‘The relationship between the water quality factor, final coliform levels in the effluent, and UV dose is illustrated in Figure 6b. As the water quality factor increases, a larger UV dose is required to maintain a specified effiuent coliform density. (One relationship apparent from Figure 6b is that coliform inactivation becomes more sensitive to water quality with more stringent effluent coliform standards. As water quality worsens, ‘greater increase in UV dose is required to maintain a more Stringent coliform standard than a less stringent standard. At less stringent permit standards, most likely all of the “free ‘swimming"* coliform and only a small percentage of the ““parti- cle associated" coliform need to be inactivated to maintain a specified permit limit, Free swimming coliform are easier to inactivate than particle associated coliform, However, as more stringent standards are adopted (lower values of N required), a ‘greater percentage of coliform bacteria associated with particles. and eventually those embedded deep within particles, will also need to be inactivated. Any decrease in water quality will sig- nificantly inhibit inactivation. ‘Testing of UV disinfection model with field results. A small-scale (operational flow rate of 8.0 12.5 L/s) UV disinfec~ tion system was operated at a WWTP in New Jersey to obtain ‘Suspended solids, mg/. ° 1x10 axt0 3x10 axe Water quality factor, ator quality factor, cr ne ee ee Dose, mWeslem= Figure 6—Graphical illustration of UV disinfection model. (a) Determination of water quality factor based ‘on SS (mg/L) and UFT (%) and (b) determination of UV dose (mW: s/cm?) based on water quality factor and dé sired effluent coliform concentration (MPN/100 mL). 205 Predicted log survivat 6 “4 3 2 Observed log survival Figure 7—Model testing with UV disinfection perfor- mance data from field site in New Jersey. performance data to test the UV model. The WWTP used a conventional activated sludge process, The UV system in New Jersey was different than the system used to develop the model but was thought to provide essentially plug flow conditions. A plot of predicted (using the water quality factor expressed in Equation 6 and the model coefficients summarized in Table 2) versus observed log survival at the New Jersey site is shown in Figure 7. Within the region of log survival that the model ‘was calibrated (log survival less than —3), the observed results followed a similar trend to mode! predictions (illustrated by a one-order-of- magnitude band of precision encompassing 95% of the data points.) The band of precision extends to 1.5 orders of magnitude for values of log survival greater than ~3. At values of log survival greater than —3, inactivation is likely dominated by first-order kinetics rather than tailing phenomenz. ‘Although moderately successful in this case, use of the UV disinfection mode! with the model coefficients reported in Table 2 and the functional form of the water quality factor expressed in Equation 6 may not always provide an accurate estimate of Eee EE eee (X.- 8\" (x23 Z-2 0.05) (Shapiro and Wilk, 1965 ). The error term accounts for the exror Luly/August 1996 associated with the estimation of the empirical coefficients in the model (A, a, 6, and n), and the error associated with the model not being the exact proper functional form. Use of the Model to Develop Allowable UV Loading To design a UV system, the allowable UV loading must be established. The UV loading, defined as liters per second per watt (L-s"!-W"}, represents the maximam flow rate ia liters per second (L/s) that can be disinfected adequately per watt of UV energy. To use the UV model to determine the UV loading at a WWTP, three factors must be considered: the uncertainty in empirical model predictions, wastewater variability, and vari- able permit criteria. Permits vary in both the stated MPN dis- ‘charge limit (for example, 2.2, 23, 240, 1 000 per 100 ml.) and in the method used to calculate the limit (for example, 7- oF 30-day running median or value not to be exceeded more than once in a 30-day period). Determination of UV loading. A detailed step-by-step pro- cedure, taking into account the above three factors, for determin= ing the allowable UV loading ata particular WWTP is presented in Table 3. Data obtained from the UCD WWTP (see Figure 9) are used to illustrate the procedure. A summary of the proce- dure is provided below. ‘The Monte Carlo analysis proceeds by obtaining approxi- rately 1 year of daily water quality data (SS, UFT, and FT values) from the WWTP effluent. Each day of water quality data is placed into Equation 9 along with an error term, €, ‘obtained from a random number generator to predict an effivent coliform concentration, The above procedure is repeated for various flow rates, which are subsequently converted into UV Toading values. The tabulated set of predicted MPN values is manipulated to represent various methods of calculating permit requirements (for example, I-day maximum or 7- of 30-day running median). Cumulative probability distribution (CPD) plots of the revised MPN values are then developed. Representa~ tive plots, based on a 7-day running median. forthe predicted MPN values at the UCD WWTP are shown in Figure 10. The CPD plots are used to determine the percent of time a permitted MPN discharge limit is exceeded for each UV loading analyzed. AA plot is then developed of the probability a specified MPN discharge limit will not be exceeded as a function of UV loading ‘on log probability paper. Representative plots for discharge lim- its of 23 and 240 MPN are shown in Figure 11 for the UCD WWTP. The allowable UV loading is defined as the UV loading at which there is an acceptable likelihood (for example. 99, 99.9, or 99.99%) the specified permit requirement will not be exceeded (an acceptable likelihood of not exceeding a specified ‘permit requirement is left to the discretion of the design engi- heer). Effect of model uncertainty on UV loading. A similar anal- ysis to the one outlined in Table 3 was repeated neglecting the variability in model predictions. To accomplish this analysis, the error term, ¢, was set equal to zero (Step 3 of Table 3). ‘A companson of the required UV loading predicted by both neglecting and accounting for model uncertainty is summarized in Table 4 for various permit conditions. These qumbers are specific to the UCD WWTP effluent quality. As shown in Table 44, neglecting model uncertainty increases the allowable UY loading, For example, for a permit of 23 MPN based on a 7- day running median, the allowable UV loading increases from 0.0591 to 0.0650 L=s"'+W" when model uncertainty is ne- sor Loge et a ‘Table 3Procedure using UV model to determine allowable UV loading based on model uncertainty, water quality variability, and permit criteri: stop Deseription 1 (Catan or simulate 1 year of daly $8, UFT, and FT valves ‘rom WWIP effuent(nerein data set 1). 2 sng a 2 * 2 lamp array and a specifed centerine lamp spacing (see Figute 2), develop | versus FT curve with PSS method (PSS curve. 3 Use Equation 9 to calculate the predicted value of N (etfuent MPN) for each of the 365 days. Because the predicted value of Vig a function of exposure time {fi Steps 3 end 4 are repeated for vanous flow rates. Flow rates are converted 10 equivaiant UV loadings by ‘assuming each UV lamp in the 2 % 2 lamp array emits 26.7 W (rated UV output, a) A, 2, b, are emprncal parameters, valves ven Tele 2 ) SS, UFT from data set t (Step 1) 9} Irom PSS curve (Step 2) ©} Tis calodated from fow rate assuming plug flow) @) «is @ normally disvibuted random error term with mean 0 and standard devation equal to RMSE’ of the ‘model (RMSE" from Equation 8) 4 Reuse predicted MPN valves in accordance with pert ‘calculation procedure (tor exarnple, 2.2, 23 or 260 MPN/100 ml. based on a 1-day max. ora 7- oF 80 cay running median). Generate a cummuatve probabilty detrbuton (CPD) plot for each UV loading ‘The ordinate represents the percent of ime each value (of MPN on the abscissa is precicted to be exceeded 5 Using the CPD plots, record the percent of te the ‘permited MPN vale 1S exceeded (Pan) f° each UY Toading anaiyzed. Plot the probably tne permited valve will not be exceeded (100 ~ Py) as a function (of UV toading (Ls-W) on log probability paper. The lowable UV loading is defined as the UY toading at which there és an acceptable fasinood (tor exemple, ‘98, 99.9, or 99.90%) the spectied permit requrement ‘wll not be exceeded. For example, using an efiuent tischarge lm ct 240 MPN/100 mi and 2 99.9% lovel ff confience, the allowable UV loading is 0.0886. 0.201, end 0.813 Lis-W based on a f-day max. and a 7 and 30-day running medan, respectvey. Mustration Objective of results To characterize SS, UFT, and FT variation Figure 9 To daterrine average UV intensity values, Figure 8 for each water qualty condition Prediction of efuent MPN based on water ‘quality varabilty and uncertainty of ‘exnpircal model predictions. Determine predicted offuent MPN in ight Faure 10 ‘of the diferent mettods of calcuiating ‘the permit requirements. Determine probabil that permit wil not ‘be exceeded as function of UV loading. For ilustratwe purposes, the results are shown for permit requrements of 23. ‘and 240 MPN/100 mL based on (3) 7 day running median, (2) 30-day running median, and (2) 1-day maximum. When the procedure is used at @ particular WWTP, plots would be developed for the WWTP's spectic permit citer Frgure 11 glected, Using an empirical UV disinfection model without ac- counting for model uncertainty may overestimate the allowable ‘UV loading at « WWTP. An overestimate of the allowable UV ‘oading would result in an underestimate of the number of lamps required to adequately disinfect the wastewater. Effect of permit basis on UV loading. For 2 specified effiu- ent MPN value (that is, 23, 240, 1000), the allowable UV loading is greater for a permit based on a running median (that is, 7- or 30-day) than for a permit based on a 1-day maximum. For example, using a discharge limit of 23 MPN in Table 4, the allowable UV loading increases from 0.0207 L+s~!-W"! for a 1-day maximum to 0.0591 and 0.118 Ls“! W~' for a 7- ‘and 30-day running median, respectively, when model uncer- tainty is considered. However, typically WWTPs have variable effluent MPN values based on the method of calculating the permit. For example, a permit might include @ 1-day maximum 208 of 240 MPN and « 7-day running median of 23 MPN, In these cases, typically the running median permit limit results in a Tower allowable UV loading than the 1-day maximum permit limit. In the case of the UCD WWTP, for I-day maximum permit requirement of 240 MPN and a 7-day running median Of 23 MPN, the allowable UV loedings are 0.0886 and 0.0591 Lis"! W"!, respectively. Effect of water quality on UV loading. The SS and transmit- tance values in the dataset collected at UCD were modified to produce three addtional data seis to evaluate the influence of ‘arious water quality parameters on the allowable UV loading ‘Table 5 contains a statistical summary ofthe SS, UFT, and FT values for each of the three data sets. Data set 1 represents 2 high quality effient with low SS (mean of 6:9 mg/L) and high UFT and FT (786 and 80.5%, respectively). Data set 2 repre- sents a poorer quality effluent that might result from increased Water Environment Research, Voure 68, Number 5 Loge ot a Suspended sols (mg/L) Unfitered transmittance (%) Fillored transmittance (%) 20 | Uvioading, User 0.0148 meen 0.0085 —- 0.0443 = 0.0581 = = - 0.0738 0.0886 N value ‘specifi Percont probably of exceed ied He i 0 5 1 1% 2 2 30 35 Effluent coliform, MPN/4100 mL. Figure 10—Cumulative probability distribution plots for predicted effluent total coliform MPN values calculated on a 7-day running median basis (values specific to the UCD WWTP effluent). values (0.130 L371» W"' for data set 3) than for effluents with poorer transmittance values (0.171 L-s"!= WT"! for data set 2). ‘Thus, in the above example, an increase in SS was more im= portant in controlling UV performance than the decrease in transmittance when less stringent discharge requirements pre- vail, Although the model coefficients and the statistical signifi- cance of the water quality Variables is likely to change at differ- 80 100 169 200 280 300 360.400 ‘Sample day Figure 9—Water quality data set, collected and partially simulated from the UCD WWTP, consisting of (a) sus- pended solids, (b) unfiltered transmittance, and (c) fil- tered transmittance. ‘amounts of soluble UV absorbing compounds. The SS concen- tration is identical to that of data set 1, but the transmittance is much lower ($4.8 and 60.9% for UFT and FT. respectively). Data set 3 represents another poorer quality effluent that might result from increased amounts of particulate matter; the mean $8 conceatration is 30.1 mg/L, whereas the transmittance values ‘are identical to those of data set 1 ‘The allowable UV loading calculated for each of the three data sets is summarized in Table 5 for discharge limits of 23, 240 and 1.000 MPN/100 mL based on a 7-day running median, [AS expected, the allowable UV loading decreases for either a decrease in UV transmittance or an increase in SS as indicated bby a comparison of data set 1 with data sets 2 and 3, respec tively, A less expected result is related to the relative importance of SS and UFT in the operation of 2 UV system. For a UV system to meet celatively strict discharge limits (for example, 23 MPN) the allowable UV loading (0.0177 Ls!» W~) was determined 10 be the same for an effluent with either poorer transmittance values (data set 2) or higher SS values (data set 3). However, for a UV system to meet less stringent discharge limits (for example, 1000 MPN), a lower allowable UV loading was determined to be required for effluents with higher SS LlyAugust 1866 UV loading, LiseW wees geses 8 3 Propabilty of not exceeding specties parmit requirement Figure 11—Probability of not violating specified permit criteria a8 a function of UV loading (values specific to the UCD WWTP effluent). Loge et a Table 4—Comparison of allowable UV loadings accounting for and neglecting UV model uncertainty for various permit critert Predicted UV loading, L-s"'-W Basis for permit Neglecting Accounting for cciterta model uncertainty model uncertainty* 23MPN ‘day maximum 0.0295 0.0207 “7-day running mecian 0.0850 0.0891 30-day running median 0.138, ote 240 MPN 5-day maximum ont 0.0886 ‘cay nureing median 0.236 0.201 80-day running median 0.458 org * Biased on # 99.9% tkelnood the spectied permit requrements wil ot be violated (Figure 11) cent WWTP, the above discussion illustrates the potential relative importance of SS over UV transmittance in the performance of a DV disinfection system. Design of UV Disinfection Facilities ‘The principal components of a UV disinfection system and the typical terminology used in design ate illustrated in Figure 12a, ‘Typically, the design flow rate is divided equally among a number ‘of channels, Each channel typically contains two or more banks ‘of UV lamps in series, and each bank contains a specified number ‘of modules (or racks of UV lamps). Modules are used to increase accessibility to the lamps within each bank. Each module contains 4 specified number of lamps: typically manufacturers of UV sys- tems use 2, 4, 8 12 or 16 lamps per module, A spacing of 75 ‘mm between the centers of UV lamps is currently the most fre- quently used lamp configuration by UV manufacturers (Figure 12b). However, asthe use of UV disinfection increases, manufac turers will likely provide alternate numbers of lamps per module as well as alternate lamp centerline spacings with the goal of providing cost-effective designs. An alternative lamp centerline spacing is ilustrated in Figure 12 (One of the primary goals in the design of a UV disinfection system is to minimize the total number of Jamps in the system ‘while simultaneously meeting both disinfection and headloss requirements. The lowest allowable UV loading, corresponding to the most stringent permit requirement, can be used to deter- ‘mine the number of lamps ina UV system required for disinfec- tion, However, the headloss along the length of a UV channel must also be constrained to acceptable values as discussed in detail below The specific steps and calculations involved in generating a design of a UV disinfection system that simultaneously meets both disinfection and headloss requirements are summarized in ‘Table 6 (Example |) in reference to a hypothetical WWTP with the following properties: (1) effiuent water quality identical to ‘Table S—Summary of the statistics of various water quality parameters and the allowable UV loading for data sets 1, 2, and 3. UV loading sccounting for summary Suspended Unitered Fitered model uncertainty, statistics sols, may. transmittance, % transmittance, % tes wo Data set 1 Mean 68 788 208 Median 66 798 a5 Maxum 185 ase a7 Penn basis* 2 0:08 240 0.366 +000 0865 am set 2 Mean 39 sae cog Median 85 594 e186 Maximum 185 598 650 Perit basis? 23 oor? 240 0.0886 +000 oar Data set 3 Moan 201 786 a5 Median 2 796 ars Maxum sos 636 887 Pernt base 28 oir 240 0.0850 +1000 0.190 * Biased on a 99.9% Ikeihood the specified pert requirement wil not be violated Pernt basis (MPN/100 mL) based on a 7-day running medien. oro Water Environment Research, Volume 68, Number 5 [ Loge et a UV modules (UV lamps shown as dotted lines below housng covering tt channel) Wt Hii j UV banks fifiye rfitita (Typical) oe Hhifela fifife a: Pyefaa ilife ififile iafifa TO 0 fifa fife 75mm 150 mm AL UREA ppipiy ta too 150 mm Channels —1 ©) Plan view of UV disinfection system (not to scale) a Pog 4 r. oo00 200 = 0000 0000 UV lamps -O Oo oo oo 00 ooo 100 mm 0000] looove tto o | Cross section A-A 200 mm (not to scale) (a) (ce) Figure 12—Components in a UV disinfection system. (a) Typical configuration of modules, banks, channels and lamps (shown is a system with four modules per bank, two banks per channel, and two channels with a total of 96 lamps}, (0) 75-mm centeriine lamp spacing (currently the industry standard), the UCD WWTP (see Figure 9). (2) effluent otal coliform permit criteria of 1 000 MPN/100 mt. based on a 7-day running median, and (3) average flow rate of 4.38 m/s (peak weekly flow rate of 7.67 m’/). The assumptions used in the design ‘example are summarized in Table 7. Although the assumptions used are likely o vary considerably among designers to be most appropriate for a particular WWTP. the step-by-step procedure ‘outlined in Table 6 can be used at any WWTP in the design of 4 UV disinfection system. In most cases. adapting the steps t0 a spreadsheet program is advisable due to the relatively large siumber of system configurations that must be evaluated Discussion of Design Examples A similar analysis to the one outlined in Table 6 was performed for two additional WWTP scenarios. Results trom key steps in eiyiAugust 1996 ind (¢) 100-mm centerline lamp spacing. the design process for Examples 1, 2, and 3 are summarized in Table 8. Design Examples 1 and 2 illustrate two methods of reduc- ing headloss through the use of alternative lamp spacings, Design Example 3 illustrates an optimal design based on the conventional lamp spacing of 75 mm, Comparison of the results from the various steps in the design process illustrate four general features in the design of UV disinfection systems: (1) considerations involved in meeting disinfection requirements, (2) methods of reducing head- loss, (3) the lamp savings associated with the use of wider center- line lamp spacings in configurations that are headloss constrained, and (4) the impact of permit requirements on determining whether 1 design is headloss constrained. Considerations in meeting disinfection requirements. The First step in design is to generate a process configuration based only on meeting disinfection requirements. as illustrated in on Loge ot a ‘Table 6—Example + illustrating the steps in the design of a UV isinfection system, Determine the design requirements using a 75-mm centerline lamp spacing (conventional lamp spacing) 1. Determine the number of lamps requred fer aisinfection based on the allowable UV loading (calculated as per Table 3) Peak weekly flow rate “Flow rate per lamp 7.87 m/s GAS 10 WY) x 15.08 Wamp) Number of lamps = Number of ams = 1324 2. Assuming various system configurations, select @ configuration that meets disinfection criteria with the minimum nurnber of lemps. ‘2. Assume number of banks per channel Ne = 2, initially b. Aspume a lamp array N, x Nie, where N, is the number of lamps per module and Ny is the number of modules per bank. Recall that Ny = Nye 1.75 N, ard Ni = 2, 4, 6, 12, oF 16. For example, assume N, = 12 and Nw = 14 «c, Calculate the number of channels, Ne, gen Mp. Nz, Nw and the number of lamps required for disinfection ‘Number of iamps required for disinfection UN. * Nu) lamps/bank) % Ne banks 41324 lamps (02% 19 tamps/bank) XP banks Number of channels = Number of channels = 3.94, use 4 d. Recompute the total number of lamps in the configuration. Total lamps = No X Ny X No ¥ Ne Total lamps = [(12 x 14) lamps/bank) x (2 banks/channel) x 4 channels = 1 344 . Caleulate the number of excess lamps. Excess lamps ~ Total lamps ~ Lamps require tor disinfection = 1 S44 ~ 1 S24 = 20 {. Repeat Steps ae forall possibie amp arrays that ic, all possibie values of Nz X Ny) ‘9, For arays that result in values of Nc > 20, increase Np in Step a by 1 (that is, Np = No + 1) and repeat Steps c-. fh. From the set of configurations developed in Stops -g, select the one with the fewest excess lamps that best fits the available ‘Space at the WWTP. In this design example, the optimal configuration, given the design assumptions, was 12 lamps per module, 14 ‘modules per bank, with 2 banks per channel in each of 4 channels, @ » 14 » 12 x 4[1934@75), 3, Check whether the headloss for the selected configuration is acceptable 8, Determine the channel cross-sectional area. Cross sectional area of channel = (12 x 0.075 m) x (14 x 0.075 m) = 0.948 m? ». Determine the net channel cross-sectional area by subtracting the cross sectional area of the quartz sleeves (@.18 x 10~* m'lamp). Net channet area = 0.945 mr? ~ [(12 X 14) lamps/bank) x 4.18 x 10~* map = 0.875 m* . Determine the velocity in the channel 1.87 mes Velocity = 287 is___ a9 mis ‘Tchannels x (0875 mvTenanned ~ * 4, Determine the headloss per UV channet. (2.19 ver - tn = 22 EE AT 3 * (2 bank/channel = 1076 mm Comment: Because the headioss per channel is greater than 50 mm, the design is unacceptable. A modified system configuration is ‘required 10 reduce headloss to an acceptable valve. 4, Repeat Step 3 forall configurations (using a 75-mm centerline lamp spacing) developed in Step 2. For all configurations that have unacceptable values of headioss, add additional channels in paralle! to split the flow, thereby reducing the velocity and headloss in each channel, The 12 x 14 lamp array, which results in 1076 mm of neadloss, wil be used to itustrate the steps. ‘2. Determine the velocity that will result in S0 mm or less of headloss per channel v 0472 re, if Ne = 2 ». Determine the number of channels required to reduce the velocity in each channel to less than that determined in Step a. Pune No. of channels = ————— PE _____ : Vie Net channel area rom step 8b) 1.87 mls No. of channels = <7 87 mits _ Srannels = EAT en) x BTS MF) = 1857, use 19 cc. Determine the revised total number of larnps in the configuration that meets headioss constrains, Total lamps = MN, % Nav X Na X Ne Total lamps = (12 x 74) lamps/bank] (2 panks/channel) x 19 channels = 6384 one Water Environment Research, Volume 68, Number § Loge et al Table 6—(Continued) 4. Calculate the number of excess lamps. Excess lamps = Total lamps ~ Lamps required for disinfection = 6384 ~ 1 924 = 5060 h. From the set of configurations developed in Step 4, select the one with the fewest excess lamps which best fits the available space at the WWTP. In this design example, the optimal configuration, given the design assumptions, was ¥6 lamps per module, 22 modules per bank, with 2 banks per channel in each of 9 channels, resulting in an excess of 5.012 lamps. Comment: For a conventional lamp spacing, the optimal configuration based on disinfection requirements alone was 2 x 14 x 12 x 4 (1344075). This process configuration contained only 20 excess lamps, but had an unacceptable value of headioss per channel (1076 Imm). Headloss could be reduced to an acceptabie value by adding 15 additional channels and increasing the lamp array size in the above system configuration. However, the resulting configuration, 2 x 22 x 16 x 9 (6 396@75), contains a significant number of total lamps (6.336), many of which are excass lamps not required for disinfection. Another method of reducing headloss that does rot Increase the total number of lamps a8 much is to widen the centertine lamp spacing, which is discussed nert Determine the design requirements using a 100-mm centerline lamp spacing 5, Determine the number of lamps requied for disinfection based on the allowable UV loading (Repeat step 1). 2.87 mis Number of lamps = Sse ag mre Wo) x (13.08 Whamo) % 2256 ‘Comment: Increasing the centerine lamp spacing from 75 to 100 mm reduces the allowable UV loading from 4.43 x 10~* to 2.60 x 10~* mms" W", which results in an increase in the number of lamps required for disinfection from 1 374 to 2 256 lamps. However, as shown below, increasing the centerine lamp scacing also reduces the headloss. 8. Assuming various system configurations, select a configuration that moets disinfection crteria with the minimum number of lamps Pepeat Step 2). “The resulting process configuration based only on meeting disinfection requirements is 2 x 19 x 12 x § (@ 2808100} 7. Check whether the headloss for the selected configuration is acceptable (Repeat Step 3) 4. Cross sectional area of channel = 2.28 m’ b. Net channel area = 2.18 mé 6, Velocity = 0.704 mis 9. Headioss = 111 mm, (an unacceptable value) 8, Repeat step 4 for all configurations (using a 100-mm centering lamp spacing) developed in Step 6. ‘The resuiting process configuration that contains the fewest number of excess lamps and meets headioss constraints is 2 x 18 x 16 x 5 (3.86@100) Comment By making use of an alternative lamp spacing, the number of total lamas required to meet both disinfection and headloss requirements was reduced by 6 336 (Step 4) to 2456 (Step 2). The usefulness of an even wider centerine amo spacing will be evaluated next. Determine the design requirements using a 150-mm centertine lamp spacing 9. Determine the number of lamps required for disinfection based on the aliowable UV loading (Repeat Step 1). 187 m/s (Cae TO SW) x (18.08 Wamp) ‘Comment: Increesing the centerine lamp spacing from 100 to 150 mm reduced the allowable UV loading significantly resutting n 6 199 lamps required for disinfection. This is greater than the number of lamps required in the acceptable process configuration using 100-mm centerline lame spacing (2 456). Therefore, the process canfiguratin that meets both disinfection and heaaloss requirements and resuits In the fewest number of total lamps is 2 x 18 x 16 x 6 (@-456@100). All three methods of reducing headloss (increasing the centering lamp spacing, the lamp array size, and the number of channeis) were necessary to generate an optimal process configuration, given the design assumptions. A centering lamp spacing between 100 and 150 mm may result in @ process configuration with fewer than 3 456 lamps, but the above example itustrates the important points in the design process. Number of lamps = = 6199 Steps | and 2 in Table 6. A process configuration consists of the number of banks per channel, modules per bank. lamps per module, and total number of channels. The primary consider ation in this step is to generate a process configuration that ‘minimizes the difference between the number of lamps required for disinfection and the actual number of lamps in the system (Step 1). The slight increase in the tozal umber of lamps is a result of adding integer numbers of components together 10 develop an acceptable system configuration. For example, the umber of lamps required for disinfection in Examples | and 2 using a 75-mm lamp spacing are 1 324 and 334. respectively Combining integer numbers of system components (For exam: July/August 1996 ple, lamps per module, modules per bank, and channels) to ‘generate an acceptable design resulted in 2 total number of lamps for Examples 1 and 2 of 1344 and 336, respectively. Very infrequently will the system configuration contain exactly the number of lamps required for disinfection. as in Example 3 (128 lamps). ‘Methods of reducing headloss. The headloss associated with this process configuration is then calculated as in Step 3 of ‘Table 6. If the process configuration is acceptable (for example. produces an acceptable value of headloss), then the resulting design is based on meeting disinfection requirements but subse quently meets headloss constraints. However, if the process a8. Loge ot a ‘Table 7—Assumptions used in the design of UV disinfection systems. + Horizontal amp configuration with flow paralie to the lamps, 1 the number of UV disinfection channels was assumed to be at least 2 but no more than 20 (that is, 2 = Ne = 20). 1 intially, two UV banks per channel are used, but additional banks are added to each channel ifthe total number of channels exceeds 20. In most designs, the minimum number of banks per channel proves optimal Headioss per channel is constrained to 50 mm calculated as ree 1(¥2) x ttre where K = 2.2 (Based on Trojan Technolosies, inc) = velocity in the channel (m/s) 1g = aoceleration due to gravity = 9.81 mst ‘The number of lamps required for disinfection was based on the UV disinfection model; mode! uncertainty, water culty variability, and the probabilistic nature of the permit criteria were inciuded as described herein. lamp fouling, to produce an effective output of 13.08 Wilamp. ‘The new lamp output of 26.7 W (rated UY output) is reduced by 20% to account for lamp aging and an addtional 30% to account for ‘The extemal diameter of the quartz sleeve surrounding the UV lamp was assumed to be 23 mm. All lamp arrays wore assumed! to meet the following criteria: N. = Na = 1.75 N, where lamp array is defined as the number of amps per module (N,) by the number of modules per bank (Nj. Modules were assumed to have 2, 4, 8, 12, or 16 lamps per module as per the current industry standard (tral is, N, was constrained to values of 2, 4, 8, 12, oF 16) + Lamp centerine spacing is assumed initially to be 75 mm, but altemative values of 100 and 150 mm are considered if ital cesign is. headioss constrained configuration produces an unacceptable value of headloss. then steps are taken to reduce the headloss to an acceptable value. and the resulting design is said 10 be based on meeting headloss constraints but subsequently meets disinfection requirements. ‘The criteria for any acceptable design is that it must meet both

You might also like