- Email: [email protected]
Modeling and efficiency of ultraviolet disinfection systems
Water Res. VoI. 19. No. 8, pp. 103%1046, 1985 Printed in Great Britain. All rights reserved
0043-1354 85 53.00+0.00 Copyright ~ 1985 Pergamon Press Ltd
MODELING AND EFFICIENCY OF ULTRAVIOLET DISINFECTION SYSTEMS ROBERT G. QUALLS and J. DONALD JOHNSON* Department of Environmental Sciences and Engineering. The University of North Carolina at Chapel Hill. Chapel Hill, NC 27514, U.S.A.
(Received December 1984) Abstract--Modeling can play a particularly useful role in ultraviolet (u.v.) disinfection because of difficulties in measuring u.v. dose and the immediate results of disinfection. This model predicts bacterial survival in flow-through systems. The calculation takes into account the complex intensity patterns. non-ideal flow patterns, and non-linear curves of log survival vs u.v. dose. Based on the organismal dose-response, the number of survivors in each fraction of the residence time distribution is calculated separately and summed to calculate the average survival. The model uses as input data: the average u.v. intensity within the system, the residence time distribution, and an experimentally determined dose-survival curve in a simplified system where dose can be directly measured. The predictions of the model corresponded welt with measured survival in a u.v. pilot plant study. The model was used to show the effects of flow dispersion on average survival by varying residence time distribution. Measures of capacity and efficiency of u.v. systems were derived and illustrated experimentally in simple cylindrical batch units and in two multiple lamp units.
Key words--ultraviolet disinfection, disinfection, modeling, coliform bacteria, flow dispersion, disinfection efficiency, bioassay
INTRODUCTION
NOMENCLATURE t = dosej = f(dose) = I= lj=
time interval dose to a particle or volume element function of dose intensity intensity to which a particle or volume element is subjected averaged over the exposure time i = i n d e x of fractions of residence time distribution j = index of fractions of distribution of intensities averaged over exposure times m = number of fractions of i n = number of fractions o f j No = initial number of viable organisms N0i = initial organisms in ith fraction N% = initial organisms in the j t h fraction of the ith fraction Ns=average number of organisms surviving irradiation Ns,' = number of organisms surviving irradiation in fraction (i,j) Ns = number of organisms surviving irradiation RT = residence time RTD = residence time distribution SA or S B= standard deviation of RTD A or RTD B t = residence time, or exposure time 7" = average residence time ti = residence time of fraction i of RTD G, or ta, = residence time of fraction i of RTD A or RTD B V = volume of fluid V,4' or Va, = volume of fluid in fraction i of RTD A or RTD B V,.j = volume of fluid in fraction i, j V, = volume of fluid in fraction i.
*Author to whom all correspondence should be addressed.
Disinfection o f wastewater effluents with ultraviolet (u.v.) light has recently become a rapidly growing alternative to chlorination. Over 30 large-scale u.v. systems (over 1000m 3 day -~ o f effluent), have been built or planned in the last several years in N o r t h America. N o t only does u.v. avoid the environmental problems associated with chlorination (Ward and DeGrave, 1978; National Research Council, 1980), but it can be economically competitive with chlorination (Scheible and Bassel, 1981). Because o f several difficulties in measuring u.v. dose, little comparative research has been done on the efficiency o f u.v. disinfection systems. However, with a systematic a p p r o a c h to measuring efficiency, u.v. disinfection systems can be improved and become even more competitive. A n o t h e r problem created by difficulties in measuring u.v. dose is evaluating the adequacy o f disinfection. The lack o f a residual disinfectant comp o u n d s this problem o f determining the extent o f disinfection with u.v. While there have been a number o f pilot and full scale investigations o f u.v. disinfection o f wastewater (Scheible and Bassel, 1981; Petrasek et aL, 1980; Roeber and Hoot, 1975; J o h n s o n and Quails, 1981) most o f these studies focused on d e m o n s t r a t i n g the reliability o f a given system. There have been several problems in research on u.v. systems which have prevented c o m p a r i s o n s o f efficiency: (a) until recently there has been no direct m e t h o d o f measuring u.v. intensities in the interior o f a complex u.v. unit
1039
ROBERT G. QUALLSand J. DONALD JOHNSO),
1040
(Quails and Johnson. 1983); (b) calculations of intensity have often relied on inappropriate equations (Qualls and Johnson. 1983) and (c) lacking reliable methods for measurement or calculation of u.v. dose, reliable and systematic comparisons of different u.v. disinfection designs and the disinfection efficiency in a given system have been difficult. Most current u.v. disinfection systems employ tubular germicidal lamps surrounded by a quartz tube submerged in a chamber through which the fluid flows. Flow may either be parallel or perpendicular to the tamp axes. In one novel system, the fluid flows through Teflon* tubes interspersed among the lamps. Intensity patterns are complex in the chambers especially those containing multiple lamps, u.v. absorbing water and reflective and absorptive surfaces. So average intensity cannot be directly measured with a meter. Hydraulic flow patterns are also complex resulting in a range of residence times. The objectives of this study were: (1) to develop a model for calculating overall survival which accounts for the complex intensity patterns, non-ideal flow patterns and non-linear survival curves found in flow-through u.v. disinfection systems, (2) to compare simulations to pilot plant data, (3) to use the model to evaluate factors influencing efficiency and (4) to develop measures to compare u.v. units. THE MODEL Dose is defined as: Dose = (intensity)(exposure time)
(1)
or, in units: mW-s cm--'
=
(mW cm-Z)(s).
(2)
The survival fraction N ] N o is: N,/No = f (dose)
(3)
where No and N~ are the density of organisms before and after irradiation, respectively, a n d f ( d o s e ) represents a function of dose. Equations (1) and (3) imply that intensity and exposure time may be varied reciprocally to obtain the same survival. For a homogeneous population of microorganisms following "single-hit" kinetics, the log survival fraction can be a linear function of the dose received by the microorganism (Chick, 1908; Jagger, 1967). However, curves of log survival vs dose for coliform bacteria in wastewater are distinctly non-linear partly because of the heterogeneous nature of the population and the presence of bacteria protected by inclusion in particles (Quails et al., 1983). The complex intensity patterns within an absorbing solution in a multiple lamp u.v. chamber caff be calculated by the point-source summation method (Quails and Johnson, 1983). Alternatively, the average intensity within the chamber may be measured *Trade name E. I. DuPont Co.
indirectly by a bioassay method (Quails and Johnson. 1983). The second factor in determining u.v. dose is the exposure time. In a flow-through system, it is also complex. The distribution of exposure times can be measured and accounted for by considering separate exposure times for each fraction of the residence time distribution (RTD) over many small time increments. The flow characteristics of a given unit may affect both the exposure time and the intensity to which various flow fractions are subjected. Intensity gradients will exist in any chamber and, ideally, the microorganisms should be well mixed across these gradients. Cortelyou et al. (1954) found an increase in disinfection efficiency in one unit as baffles were added to increase mixing across intensity gradients. Baffles can also aid in improving plug flow. In current u.v. systems, manufacturers claim that lamp spacing is sufficiently close and radial mixing across intensity gradients is adequate. Mixing in the direction of flow often at the expense of longitudinal dispersion is undesirable as shown by the results of this and earlier work (Johnson and Quails, 1981). Neither the residence time (RT) calculated from flow rate and volume nor the average RT measured from a tracer curve is sufficient to predict the average survival of microorganisms. The model presented here considers separately each fraction of the RTD, over small time increments. Each of these time fractions are further subdivided into a distribution of organism fractions subject to various average intensities. In a flow-through system a particle or volume element may be subjected to varying levels of intensity during the exposure period (t) (Fig. 1). The dose to which the particle is subjected during a small time interval is I dt and the dose for the exposure period is: dosej =
I dr.
(4)
i-FoI t
DOSEj =£'I dl =Ij I
TIME
Fig. 1. Illustration of the dose to which an individual particle is subjected (dosej) when passing through the varying fields of intensity in a u.v. chamber. It demonstrates that the dose can be expressed as the time multiplied by the intensity averaged over the exposure time (lj).
Modeling u.v. disinfection If the average intensity to which the particle is subjected over the exposure period is defined.
I~=tI f',,Idt
(5)
Residence time distribution data may be put in a form to use in these equations. The area under the RTD curve of relative concentration vs time is set equal to V. Then.
V, =
then from equations (I). (4) and (5) the dose received by the particle is dose, = (/j)(t).
(6)
Thus, the dose to which an individual particle or volume element is subject can be expressed as the average of the varying intensity during the exposure time, multiplied by the time that particle spends in the unit. Note that equation (6) is a simple consequence of the well established reciprocal effects of intensity and exposure time on survival (Jagger, 1967; Oliver and Cosgrove, 1975) and does not depend on the shape of the survival curve or the variety of times other volume elements spend in the unit. This considerably simplifies the model for application to complex intensity fields. The following equations show how the density of survivors (N,) may be predicted from the following data: (1) coliform influent density (No), (2) average intensity (I). either measured or calculated, (3) residence time distribution and (4) a dose-survival curve (determined in a simplified system where dose can be accurately measured). For an aliquot exiting the u.v. chamber at a certain time, the total volume (V) is composed of m fractions of volumes (V~) each of which have spent times ti in the u.v. chamber. The relative volumes of these fractions is described by the residence time distribution (RTD). Each of the fractions Vt can be further divided into fractions Vi.j which have been exposed to an intensity/j, averaged over the exposure time. Each fraction (V,j) contains an initial number N% of organisms. The dose for the volume element Is,..j is: doseij = ti~. The number of survivors in
V~.i(N,,J)
N,,j = N%f (61j).
(7) is: (8)
where the function f(tff~) is described by the experimental dose survival curve measured under conditions where t~ and 6 are single valued and easily determined. The total number of survivors in V(N~) is:
iV,= ~. i No,jf (tilj).
(9)
The average concentration of survivors is ~ , divided by V. If mixing across the intensity gradients is sufficient, then each particle or volume element is subject to the same intensity, 6, averaged over the exposure time. In that case the average concentration of survivors is: lq, 1 " ~: = ~ N0,f(rt~). (I0) v - ~ , °o,
fOal
(t) (relative concentration) V
(I t)
and,
V v,
No, =~ 0 ~ .
(12)
The simulation presented here has assumed good mixing across intensity gradients. From mapping intensity patterns in several types of designs in current use, we have found stratification across intensity gradients is minimized by the very short distances across these gradients relative to the chamber length, by baffles added to promote mixing, or by flow perpendicular to lamp axes. However, the bioassay method (Quails and Johnson, 1983) provides a means of detecting poor mixing across intensity gradients. In this method standardized bacterial spores were injected into a u.v. system and collected as a function of time from injection. Any systematic deviation from a linear relationship between assayed dose and time from injection for successive flow fractions would indicate poor mixing across intensity gradients (Quails and Johnson, 1983, Fig. 6). However, if the experiment indicates insufficient mixing across intensity gradients, then the bioassay results cannot be interpreted as the average intensity. Because the Aquafine unit used in this study was shown in experiments reported elsewhere (Quails et al., 1983) to have sufficient mixing across intensity gradients, we were able to use the average intensity in equation (10) but not able to verify the more general form of the model [equation (9)]. METHODS
Calculation of intensity The intensity at each point in the disinfection chamber was calculated by the point-source summation method (Quails and Johnson, 1983; Jacob and Dranoff, 1970). While this calculation is detailed elsewhere, we made several modifications to apply it to a multiple lamp reactor. Intensity at any point in the irradiated fluid was treated as the sum of contributions from each lamp, represented in turn by a line divided into point sources. We found that germicidal lamps transmit only about 250/0 of the 254nm u.v. light passing around them from a neighboring lamp (Johnson and Quails, 1981). So shadowing by neighboring lamps was included. Intensity was mapped by a computer program at each point on a grid in a cross-section of the chamber and averaged over the cross-section and along the length of the chamber (Johnson and Quails. 1981). Our calculations used these simplifications: (1) reflection from the reactor walls was negligible under operating conditions, (2) reflection and refraction by quartz tubes was ignored and (3) end effects on the lamp output measurement were negligible since the lamp length was 77 cm.
Determination of dose-surrical curces The survival of bacteria as a function of u.v. dose was determined by irradiating samples for various times in a
1042
ROBERT G. QUALLS and J. DONALD JOHNSON
stirred petri dish. The dish was located at the bottom of a collimating tube over which four germicidal lamps were suspended. Intensity at the liquid surface was measured with an International Light IL-500 radiometer (Qualls and Johnson, 1983). The average intensity within the suspension was calculated by an integration of Beer's Law over the fluid depth (Morowitz, 1950).
ultraviolet lamp quartz tube ~-
...., .-"
plexiqlass ~ cylinder ~ : l
Cylinder experiments To demonstrate the relationship between average intensity and fluid depth in a cylindrical geometry, we used the apparatus shown in Fig. 2. Average intensity within the cylinder was calculated by the point-source summation method. For this calculation lamp output was measured by the method o f Barrows (1951). To veri~ calculated average intensity in the cylinder, a bioassay measurement also was made (Quails and Johnson, 1983).
I
Pilot plant experiments The pilot u.v. units were located at the Sandy Creek contact stabilization wastewater treatment plant in Durham, NC. Effluent was pumped through either an Aquafine CSL-6 or a Pure Water Systems (PWS) IL-75 u.v, treatment system (Table I). While some treatments involved filtered effluents (Quails et al., 1983) only unfiltered runs are considered here. Thirty-six experiments were performed with: (1) flow rates either 2.27 or 4.92 1s -~ and (2) applied voltages o f either 60 or 128 V for each unit. Total coliforms were enumerated by the MPN procedure (APHA, 1975). The simulation involved only the data for the Aquafine unit. Residence time distributions were measured by injecting dye into the entrance o f the irradiation chamber and collecting samples o f the outflow in a rotating sampling tray as a function of time from injection.
stirring device
sliding black
I I
paper tube ] I between lamp [ I q
Model calculations The following data were needed for the simulation: (1) the coliform density in the influent (No), (2) the average intensity (I) calculated by the point-source summation method, (3) the RTD measured by dye injection and (4) the dose-survival curve, determined in the petri dish apparatus. Smooth curves were drawn through the RTD data and the log survival vs dose data. From these curves, data pairs for small increments of the independent variables (&time = 0.05 s and Adose = 2 mW-s cm--') were entered into arrays in a computer program. Intermediate values required in the calculations were generated by linear interpolation. For this simulation it was assumed that mixing was good across the intensity gradients and equation (10) was used to calculate the average survival.
Fig. 2. Cylindrical irradiation apparatus (from Quails and Johnson, 1983).
-I -J
-2 rv
-3
RESULTS AND DISCUSSION 0..J
As an example of the model of survival in a flow-through u.v. system we simulated runs with the Aquafine unit, These simulations were then compared to the observed survival in the pilot plant experiments with this unit. At the time of the pilot plant runs, methods to determine an accurate
-4
o
I 10
DOSE ( m W - s
I 20
I 30
cm -z)
Fig. 3. D o s e - s u r v i v a l curve f o r total coliforms in Sandy
Creek 2 = effluent (unfiltered) in a collimated u.v. beam,
Table 1. Comparison of Aquafine and pure water system units Characteristic Aquafine PWS u.v. Output, total (W) : :'. 62.3 77.8 Volume (1.) II .0 5.8 Calculated ave. intensity (mWcm--')* 9.7 18.8 Intensity × volume (roW cm -z) (I.) 107 107 Intensity-volume elficiency 0.45 0.31 (mWcrn-") (I.):'input wattage *At absorbance = 0.17.
Modeling u.v. disinfection Table 2. Observed vs simulated survival IS) of total colit'orm~in a u.v. disinfected Sandy Creek 2 secondar.~ effluent Lamp Av. intensity Simulated Pilot plant voltage (roW cm--' ) log S log S 60 6.2 - 3.26 - 329 t "--O.13) 128 9.7 - 3.61 - 3.69 ( __ O. 16)
dose-survival curve were not developed. So dose-response data were determined some time later for three samples from the same site (Fig. 3). The curve levels out substantially between - 3 . 0 and - 4 . 0 log survival units in marked contrast to ideal Chick's Law log-linear behavior. The R T D for the Aquafine unit (Fig. 4) was skewed and shows considerable dispersion. The average log survival predicted by the simulation corresponded extremely well with that observed in the pilot plant runs (Table 2). Some deviation might have been expected since the dose-survival curve was based on samples taken at later dates. To illustrate the effect of flow dispersion, we used the model to simulate the average Survival for a set of hypothetical R T D curves with a constant mean R T D , but with varying degrees of dispersion. The R T D in Fig. 4 was transformed b y the following equation so as to vary the standard deviation of the distribution while maintaining the same average residence time T, the same relative shape and area. For a given value on the y-axis (VA,) of distribution A and a corresponding value on the time axis (tA,), a transformed value for distribution B (tB,) was calculated: ts, = T -I-(I
T -t4, ISs/S4)
(13)
where T is the mean residence time, SA and SB are the standard deviations of distributions A and B respectively. The areas under the curves were normalized by calculating a new y-axis value (Vn,) for each value of distribution B:
liB,
=
V~,lV~,
(14)
where VA, and Va, are the areas under curves of VA, and Va, vs tA, and ta,, respectively. The set of curves shown in Fig. 5 was generated from the Aquafine R T D (Fig. 4). The simulations of average survival all used the same dose survival curve shown in Fig. 3. The upper plot in Fig. 5 is the _~
8
1043
0 L
~
15 F
I
t
i -A
3
a
T i M E (S)
Fig. 5. Effect of varying amounts of dispersion on average survival in RT distributions with the same average RT. Curve C was that measured for the Aquafine system. The other curves (A, B and D) were generated by transforming the time axis symetrically about the average RT. Above the RTD's, the survival curve used for the simulation (Fig. 2) is shown as a function of RT for the indicated intensities. Note what portions of the RTD's correspond to the "steep" and "flat" portions of the survival curves.
survival curve from Fig. 3 shown as a functibn of time for the two levels of intensity used in the simulations. It is shown above the residence time distributions so that the portion of the survival curve which corresponds with each part of the R T D below can easily be compared using the same time axis. The most significant comparison to be made is whether a substantial portion of the R T D corresponds to the steeply sloping initial portion of the log survival curve. The effects of flow dispersion on disinfection efficiency are dramatic when a significant portion of the flow distribution is in a sensitive part of the dose-survival curve, in this case a dose less than about 10 mW-s cm -2 (Fig. 4) or 1.5-2 s for intensity of 9.7 and 6.2 mW-s cm-'- respectively (Fig. 5, Table 3). Thus the effect of dispersion in the Aquafine unit was greater for the lower intensity level. Beyond the 10 mW-s c m - " dose, the dose-survival curve (Fig. 4) levels out to such an extent tllllt dispersion of flow had relatively little effect on the average survival. In this region increasing dose by increasing either t or I
¢. w Z'O
4
_
2
L=J ¢:
0
I
2 TIME (S)
3
4
Fig. 4. Residence time distribution for Aquafine unit at a flow rate of 4.92 I s- L The curve is normalized so that the area = 1.0.
Table 3. Simulated average log survival for the RTD's (curves A, B. C arid D) in Fig. 5. The standard deviation of the RTD's are indicated by S Av. log S S [ = 6.2 / = 9.7 A 0 -3.50 -3.68 B 0.33 -3.44 -3.68 C 0.66 -3.26 -3.61 D 1.00 -2.33 -2.76
10-M.
ROBERT G. QUALLSand J. DONALD JOHNSON
has little effect on survival. Unfortunately, the most dispersed RTD, curve D shown in Fig. 5, was more typical of the actual RTD for most of the small one to six lamp u.v. disinfection systems tested by the National Sanitation Foundation (Bellen et al., 1981). The variation in average survival caused by flow dispersion demonstrates that caution must be used in interpreting curves of average log survival vs either the mean residence time, or especially the residence time calculated from volume/flow rate. in practical flow-through systems (e.g. Scheible and Bassel, 1981). Although average log survival vs mean residence time may be a valuable guide to operating a certain system, such a dose-response curve may not be comparable for the same organism in another unit with different RTD characteristics. The distinction between the "'true" dose-survival curve (e.g. batch exposure with defined time and intensity) and the ambiguities of the averaged relationship should be appreciated. The strongly curved relationship of survival to dose in actual dose-survival curves also makes the first fractions of the RTD with a large dispersion disproportionately important. Haas and Sakellaropoulos (1979) proposed a model of u.v. disinfection in a flow-through system. They assumed that log survival was linearly related to dose. By assuming that the u.v. inactivation was a first order kinetic process, they were able to use the classical equations for a first order chemical reaction in a series of continuously stirred tanks (CSTR's) (Levenspiel, 1972). However, the simple equation these authors used to calculate intensity was not valid for an absorbing fluid and u.v. inactivation is seldom first order with dose (Quails and Johnson, 1983; Jacob and Dranoff, 1970). Scheible et al. (1983) have used a modification of the same approach again assuming first order survival kinetics, but using instead the point-source summation method (Quails and Johnson, 1983) to calculate the intensity and the dispersed PlUg flow model for open vessels (Levenspiel, 1972, Chap. 9) instead of the CSTR model. The analytical solution of both of these models depends on using ideal first order kinetics for survival. The dose-survival data from wastewater samples and the analysis of the effects of residence time distribution in this study demonstrate that order of magnitude errors in survival can result from the assumption of first order kinetics. Severin et al. (1984) derived an e q u a - ~ tion similar to that of Haas and Sakellaropoulos (1979) for a one-lamp, single, cylindrical CSTR by a somewhat different solution. They pointed out the inadequacy of using the simple first order reaction expression to describe data of pure cultures which exhibited an initial lag in slope (or "'shoulder"). Severin et al. (1983) also derived a solution .using a multi-target model (and a series event model). He used these to describe a survival curve with an initial "shoulder" for the same single lamp cylindrical CSTR. Both the multi-target and multi-hit types of survival curve, however, are linear beyond an initial
lag in slope (Taylor and Johnson, 1974) unlike the typical dose-survival relationship we found (Fig. 3). The advantage of the model described in this paper is that it uses any of the types of survival curves actually found in the field. This approach is also amenable to describing the complex flow regimes found in practice. Easily obtained experimental dose survival and RTD data are used directly. While the CSTR series model or a dispersion model could be used to describe the RTD, such models still require experimental verification. They only describe certain families of curves and require additional measurements and calculations to determine the parameters of the equations (Levenspiel, 1972). Effects of lamp spacing
There are a number of divergent views on the efficient design of u.v. systems. It has been suggested that closely spaced lamps are needed in u.v. disinfection of wastewater because of the high u.v. absorbance of wastewater solutions. In order to scientifically approach the question of efficient use of u.v. intensity, we contrasted the survival for a given volume of flow with different schemes of lamp spacing. Any surface which absorbs u.v. light inside the irradiation chamber (e.g. walls, baffles, other lamps), besides the unavoidable absorbance of the water itself, reduces the efficient use of the u.v. energy. The product of average intensity times the volume of the chamber is a factor which is directly proportional to the capacity of the unit under ideal flow conditions. The volume serves as a weighting factor to express the larger capacity of units with larger volume. This measure isolates the tradeoff between intensity and volume from the effects of non-ideal flow or how well the volume is used. To investigate the intensity-volume relationship in a simple cylindrical geometry we showed how the distance the light was allowed to penetrate, before encountering a nonreflective wall, affected the efficiency of light use. The intensity-volume product is shown for a series of cylinders of varying radius r around a u.v. lamp (Fig. 6) for the experimental apparatus shown in Fig. 2. The solid lines represent the point-source summation calculations of average intensity times the volume. The data points were derived from actual bioassay measurements of average intensity. The point at which the lines level out is the radius beyond which little u.v. light penetrates at a given absorbance in the cylindrical geometry. Obviously the effectiveness, as expressed by the intensity-volume product, was greater for the larger cylinders. For an absorbance representative of a good secondary wastewater effluent (0.16), it can be seen that walls or other obstructions within 5 or 6 cm of the source can absorb a significant amount of the u.v. light. The most efficient design depends predictably on absorbance at 254 nm with closely spaced lamps
Modeling u.v. disinfection Ld
:~
I
zo L--
>-7~ L ~~ z- ~
0
8F
~ 0 0
2
4
6
RADIUS (cm}
Fig. 6. Intensity-volume product in cylinders of varying radii around a u.v. lamp. Lines represent calculations of average intensity times the volume in the cylinders in solutions of absorbances 0, 0.16 and 0.32. Data points represent the average intensity, measured by the bioassay, times the volume.
t045
responses. Both the intensity-volume product and RTD need to be reported for any system. The only single parameter which might be useful as an ultimate comparison of disinfection capacity would be a simulation using a standard dose-survival curve (e.g. Fig. 3) at a given flow rate. The parameter indicating effectiveness would be the maximum flow rate yielding--3.5 log survival units (for example) using the standard dose-survival curve. Such a parameter would provide an easily interpreted and realistic comparison of units operated under different conditions.
Applications of simulation of u.v. disinfection Simulation takes into account the factors of the intensity use, of flow dispersion or time characteristics and sensitivity of the target organisms. Model calculations can be a useful tool for research and development of reactor design. For example, it can be used to find optimum lamp configurations and tradeoffs with flow dispersion, water quality, etc. It can be used to predict the design parameters needed for a specific situation so that costly overdesign is not necessary. The predicted survival of a standard coliform sample at a given flow rate may be used to compare a number of different reactors. The difficulties in control of the process of u.v. disinfection has been regarded as a disadvantage of u.v. disinfection. However, model simulations may also be used to prepare empirical curves of predicted survival vs flow rate, lamp voltage, water quality, etc. as a guide to continuous operation for a particular installation (Dorfman, 1984).
decreasing efficiency except for highly absorbing solutions. Two units used in.the pilot plant experiments were compared on the basis of their intensity-volume products (Table 1). The unit with lamps placed close to one another and the walls (PWS unit) had an average intensity almost twice as high as the other unit (Aquafine). However, the PWS unit had a much smaller volume (and short residence time) so the intensity-volume products were almost equal. However, the PWS reactor required a larger total lamp wattage to produce this intensity-volumeproduct. The intensity-volume "efficiency" (intensity × volume/ input wattage) compares the efficiency of the use of the lamp wattage. The PWS system was less efficient in this respect. The proximity of the lamps to the Acknowledgements--We thank Kent H. Aldrich, Michael P. walls and each other causes loss of u.v. due to Flynn, Donald E. Francisco and Thomas S. Wolfe who contributed to the pilot plant study. absorption of the light. Since neighboring u.v. lamps The initial research in this paper was supported partially transmit little of the 254 nm light coming from adja- by a grant R80470010 from the U.S. Environmental Proteccent lamps, excessively close lamp spacing can result tion Agency, Albert D. Venosa, project officer. Modeling in reduced efficiency of u.v. light use. This reduction work was supported by a grant CEE82-05274 from the U.S. National Science Foundation, Edward H. Bryan, program in efficiency has been calculated for a large rectandirector. gular array of lamps using various lamp spacings (Johnson and Quails, 1981). REFERENCES It is important to note that the intensity-volume product does not account for the effects of nonideal APHA (1975) Standard Methods for the Examination of Water and Wastewater, 14th edition. American Public flow. Its value is simply to isolate the effectiveness of Health Association, Washington, D.C. the intensity use from hydraulic problems or the Barrows W. E. (1951) Light, Photometry, and Illuminating effectiveness of volume use so that these two probEngineering, 3rd edition. McGraw-Hill, New York. lems in design may be addressed separately. Although Bellen G. E., GoRier R. A. and Dormand-Herrera R. (1981) Water Disinfection Technology. National Sanitation the intensity-volume product of the two units used in Foundation, Ann Arbor, Mich. the pilot plant study were nearly equal (Table 1), the Chick H. (1908) An investigation of the laws of disinfection. PWS unit allowed 0.6-2.1 log units greater survival J. Hyg. 8, 92-158. than the Aquafine unit with the same Cortelyou J. R., McWhinnie M. A., Riddiford M. S. and intensity-volume product because of severe s h o r t Semrad J. E. (1954) Effects of ultraviolet irradiation on large populations of certain water-borne bacteria in circuiting and lack of plug flow in the PWS unit. Thus motion--I. The development of adequate agitation to the effects of the intensity-volume factor ancl-the provide an effective exposure period--II. Some physical effects of flow dispersion must be considered sepafactors affecting the effectivenessof germicidal ultraviolet rately. There is no simple single parameter to provide radiation. Appl. Microbiol. 2, 262-269. a balanced measure of the ultimate effectiveness of a Dorfman M. H. (1984) Applications of a model of ultraviolet disinfection. Masters thesis, Department of Enviu.v. disinfection system with different organismal
1046
ROBERT G. QUaLLS and J, DOYALOJo~yso~
ronmental Sciences and Engineering, University of North Carolina, Chapel Hill, NC. Haas C. N. and Sakellaropoulos G. G. (1979) Rational analysis of ultraviolet disinfection reactors. In Proceedings o f the National Conference on Environmental Engineering, pp. 540--547. American Society of Civil Engineers, Washington, DC. Jacob S. M. and Dranoff J. S. (1970) Light intensity profiles in a perfectly mixed photoreactor. Am. Inst. chem. Engng J. 16, 35%363. Jagger J. H. (1967) Introduction to Research in UV Photobiology. Prentice-Hall, Englewood Cliffs, NJ. Johnson J. D. and Qualls R. G. (1981) Ultraviolet disinfection of secondary effluent: measurement of dose and effects of filtration. Report of EPA project R 804770010, Municipal Environmental Research Laboratory, Cincinnati, OH. Levenspiet O. (1972) Chemical Reaction Engineering. Wiley, New York. Morowitz H. J. (1950) Absorption effects in volume irradiation of microorganisms. Science ! 11, 229-230. National Research Council (1980) Drinking Water and Health. National Academy of Sciences Press, Washington, DC. Oliver B. G. and Cosgrove (1975) The disinfection &sewage treatment plant effluents using ultraviolet light. Can. J. chem. Engng 53, 170-174. Petrasek A. C,, Wolf H. W., Edmond S. E. and Andrews D. C. (1980) UItraviolet disinfection of municipal wastewater effluents. EPA No. 600/2-80/102. U.S. Environmental Protection Agency, Washington, DC. Quails R. G. and Johnson J. D. (1983) Bioassay and dose
measurement in UV disinfection. Appl. envir. Microbiol. 45, 872-877. Quails R. G., Flynn M. P. and Johnson J. D. (1983) The role of suspended particles in ultraviolet disinfection. J. Wat. Pollut. Control Fed. 55, 1280-1285. Roeber J, A. and Hoot F. M. (1975) Ultraviolet disinfection of activated sludge effluent discharging to shellfish waters. EPA No. 600/2-75-060. U.S. Environmental Protection Agency, Washington, DC. Scheible O. K. and Bassel C. D. (1981) Ultraviolet disinfection of a secondary wastewater treatment plant effluent. EPA No. 600/$2-81-52. U.S. Environmental Protection Agency, Washington, DC. Scheible O. K., Forndran A. and Leo W. M. (1983) Pilot investigation of ultraviolet wastewater disinfection at the New York City Port Richmond Plant. In Municipal Wastewater Disinfection (Edited by Venosa A. D.), pp. 202-225. EPA. No. 600/9-83-009, U.S, Environmental Protection Agency, Washington, DC. Severin B. F., Suidan M. T. and Engelbrecht R. S. (1983) Kinetic modeling of u.v. disinfection of water. Water Res. 17, 1669-1689. Severin B. F., Suidan M. T., Rittmann B. E. and Engelbrecht R. S. (1984) Inactivation kinetics in a flow-through UV reactor. J. Wat. Pollut. Control Fed. 56, 164-169. Taylor D, G. and Johnson, J. D. (1974) Kinetics of viral inactivation by bromine. Chemistry of Water Supply, Treatment and Distribution (Edited by Rubin A. J.), pp. 369-408. Ann Arbor Science, Ann Arbor, MI. Ward R. W. and DeGrave G. M. (1978) Residual toxicity of several disinfectants in domestic wastewater. J. Wat. Pollut. Control Fed. 50, 46-60.
0043-1354 85 53.00+0.00 Copyright ~ 1985 Pergamon Press Ltd
MODELING AND EFFICIENCY OF ULTRAVIOLET DISINFECTION SYSTEMS ROBERT G. QUALLS and J. DONALD JOHNSON* Department of Environmental Sciences and Engineering. The University of North Carolina at Chapel Hill. Chapel Hill, NC 27514, U.S.A.
(Received December 1984) Abstract--Modeling can play a particularly useful role in ultraviolet (u.v.) disinfection because of difficulties in measuring u.v. dose and the immediate results of disinfection. This model predicts bacterial survival in flow-through systems. The calculation takes into account the complex intensity patterns. non-ideal flow patterns, and non-linear curves of log survival vs u.v. dose. Based on the organismal dose-response, the number of survivors in each fraction of the residence time distribution is calculated separately and summed to calculate the average survival. The model uses as input data: the average u.v. intensity within the system, the residence time distribution, and an experimentally determined dose-survival curve in a simplified system where dose can be directly measured. The predictions of the model corresponded welt with measured survival in a u.v. pilot plant study. The model was used to show the effects of flow dispersion on average survival by varying residence time distribution. Measures of capacity and efficiency of u.v. systems were derived and illustrated experimentally in simple cylindrical batch units and in two multiple lamp units.
Key words--ultraviolet disinfection, disinfection, modeling, coliform bacteria, flow dispersion, disinfection efficiency, bioassay
INTRODUCTION
NOMENCLATURE t = dosej = f(dose) = I= lj=
time interval dose to a particle or volume element function of dose intensity intensity to which a particle or volume element is subjected averaged over the exposure time i = i n d e x of fractions of residence time distribution j = index of fractions of distribution of intensities averaged over exposure times m = number of fractions of i n = number of fractions o f j No = initial number of viable organisms N0i = initial organisms in ith fraction N% = initial organisms in the j t h fraction of the ith fraction Ns=average number of organisms surviving irradiation Ns,' = number of organisms surviving irradiation in fraction (i,j) Ns = number of organisms surviving irradiation RT = residence time RTD = residence time distribution SA or S B= standard deviation of RTD A or RTD B t = residence time, or exposure time 7" = average residence time ti = residence time of fraction i of RTD G, or ta, = residence time of fraction i of RTD A or RTD B V = volume of fluid V,4' or Va, = volume of fluid in fraction i of RTD A or RTD B V,.j = volume of fluid in fraction i, j V, = volume of fluid in fraction i.
*Author to whom all correspondence should be addressed.
Disinfection o f wastewater effluents with ultraviolet (u.v.) light has recently become a rapidly growing alternative to chlorination. Over 30 large-scale u.v. systems (over 1000m 3 day -~ o f effluent), have been built or planned in the last several years in N o r t h America. N o t only does u.v. avoid the environmental problems associated with chlorination (Ward and DeGrave, 1978; National Research Council, 1980), but it can be economically competitive with chlorination (Scheible and Bassel, 1981). Because o f several difficulties in measuring u.v. dose, little comparative research has been done on the efficiency o f u.v. disinfection systems. However, with a systematic a p p r o a c h to measuring efficiency, u.v. disinfection systems can be improved and become even more competitive. A n o t h e r problem created by difficulties in measuring u.v. dose is evaluating the adequacy o f disinfection. The lack o f a residual disinfectant comp o u n d s this problem o f determining the extent o f disinfection with u.v. While there have been a number o f pilot and full scale investigations o f u.v. disinfection o f wastewater (Scheible and Bassel, 1981; Petrasek et aL, 1980; Roeber and Hoot, 1975; J o h n s o n and Quails, 1981) most o f these studies focused on d e m o n s t r a t i n g the reliability o f a given system. There have been several problems in research on u.v. systems which have prevented c o m p a r i s o n s o f efficiency: (a) until recently there has been no direct m e t h o d o f measuring u.v. intensities in the interior o f a complex u.v. unit
1039
ROBERT G. QUALLSand J. DONALD JOHNSO),
1040
(Quails and Johnson. 1983); (b) calculations of intensity have often relied on inappropriate equations (Qualls and Johnson. 1983) and (c) lacking reliable methods for measurement or calculation of u.v. dose, reliable and systematic comparisons of different u.v. disinfection designs and the disinfection efficiency in a given system have been difficult. Most current u.v. disinfection systems employ tubular germicidal lamps surrounded by a quartz tube submerged in a chamber through which the fluid flows. Flow may either be parallel or perpendicular to the tamp axes. In one novel system, the fluid flows through Teflon* tubes interspersed among the lamps. Intensity patterns are complex in the chambers especially those containing multiple lamps, u.v. absorbing water and reflective and absorptive surfaces. So average intensity cannot be directly measured with a meter. Hydraulic flow patterns are also complex resulting in a range of residence times. The objectives of this study were: (1) to develop a model for calculating overall survival which accounts for the complex intensity patterns, non-ideal flow patterns and non-linear survival curves found in flow-through u.v. disinfection systems, (2) to compare simulations to pilot plant data, (3) to use the model to evaluate factors influencing efficiency and (4) to develop measures to compare u.v. units. THE MODEL Dose is defined as: Dose = (intensity)(exposure time)
(1)
or, in units: mW-s cm--'
=
(mW cm-Z)(s).
(2)
The survival fraction N ] N o is: N,/No = f (dose)
(3)
where No and N~ are the density of organisms before and after irradiation, respectively, a n d f ( d o s e ) represents a function of dose. Equations (1) and (3) imply that intensity and exposure time may be varied reciprocally to obtain the same survival. For a homogeneous population of microorganisms following "single-hit" kinetics, the log survival fraction can be a linear function of the dose received by the microorganism (Chick, 1908; Jagger, 1967). However, curves of log survival vs dose for coliform bacteria in wastewater are distinctly non-linear partly because of the heterogeneous nature of the population and the presence of bacteria protected by inclusion in particles (Quails et al., 1983). The complex intensity patterns within an absorbing solution in a multiple lamp u.v. chamber caff be calculated by the point-source summation method (Quails and Johnson, 1983). Alternatively, the average intensity within the chamber may be measured *Trade name E. I. DuPont Co.
indirectly by a bioassay method (Quails and Johnson. 1983). The second factor in determining u.v. dose is the exposure time. In a flow-through system, it is also complex. The distribution of exposure times can be measured and accounted for by considering separate exposure times for each fraction of the residence time distribution (RTD) over many small time increments. The flow characteristics of a given unit may affect both the exposure time and the intensity to which various flow fractions are subjected. Intensity gradients will exist in any chamber and, ideally, the microorganisms should be well mixed across these gradients. Cortelyou et al. (1954) found an increase in disinfection efficiency in one unit as baffles were added to increase mixing across intensity gradients. Baffles can also aid in improving plug flow. In current u.v. systems, manufacturers claim that lamp spacing is sufficiently close and radial mixing across intensity gradients is adequate. Mixing in the direction of flow often at the expense of longitudinal dispersion is undesirable as shown by the results of this and earlier work (Johnson and Quails, 1981). Neither the residence time (RT) calculated from flow rate and volume nor the average RT measured from a tracer curve is sufficient to predict the average survival of microorganisms. The model presented here considers separately each fraction of the RTD, over small time increments. Each of these time fractions are further subdivided into a distribution of organism fractions subject to various average intensities. In a flow-through system a particle or volume element may be subjected to varying levels of intensity during the exposure period (t) (Fig. 1). The dose to which the particle is subjected during a small time interval is I dt and the dose for the exposure period is: dosej =
I dr.
(4)
i-FoI t
DOSEj =£'I dl =Ij I
TIME
Fig. 1. Illustration of the dose to which an individual particle is subjected (dosej) when passing through the varying fields of intensity in a u.v. chamber. It demonstrates that the dose can be expressed as the time multiplied by the intensity averaged over the exposure time (lj).
Modeling u.v. disinfection If the average intensity to which the particle is subjected over the exposure period is defined.
I~=tI f',,Idt
(5)
Residence time distribution data may be put in a form to use in these equations. The area under the RTD curve of relative concentration vs time is set equal to V. Then.
V, =
then from equations (I). (4) and (5) the dose received by the particle is dose, = (/j)(t).
(6)
Thus, the dose to which an individual particle or volume element is subject can be expressed as the average of the varying intensity during the exposure time, multiplied by the time that particle spends in the unit. Note that equation (6) is a simple consequence of the well established reciprocal effects of intensity and exposure time on survival (Jagger, 1967; Oliver and Cosgrove, 1975) and does not depend on the shape of the survival curve or the variety of times other volume elements spend in the unit. This considerably simplifies the model for application to complex intensity fields. The following equations show how the density of survivors (N,) may be predicted from the following data: (1) coliform influent density (No), (2) average intensity (I). either measured or calculated, (3) residence time distribution and (4) a dose-survival curve (determined in a simplified system where dose can be accurately measured). For an aliquot exiting the u.v. chamber at a certain time, the total volume (V) is composed of m fractions of volumes (V~) each of which have spent times ti in the u.v. chamber. The relative volumes of these fractions is described by the residence time distribution (RTD). Each of the fractions Vt can be further divided into fractions Vi.j which have been exposed to an intensity/j, averaged over the exposure time. Each fraction (V,j) contains an initial number N% of organisms. The dose for the volume element Is,..j is: doseij = ti~. The number of survivors in
V~.i(N,,J)
N,,j = N%f (61j).
(7) is: (8)
where the function f(tff~) is described by the experimental dose survival curve measured under conditions where t~ and 6 are single valued and easily determined. The total number of survivors in V(N~) is:
iV,= ~. i No,jf (tilj).
(9)
The average concentration of survivors is ~ , divided by V. If mixing across the intensity gradients is sufficient, then each particle or volume element is subject to the same intensity, 6, averaged over the exposure time. In that case the average concentration of survivors is: lq, 1 " ~: = ~ N0,f(rt~). (I0) v - ~ , °o,
fOal
(t) (relative concentration) V
(I t)
and,
V v,
No, =~ 0 ~ .
(12)
The simulation presented here has assumed good mixing across intensity gradients. From mapping intensity patterns in several types of designs in current use, we have found stratification across intensity gradients is minimized by the very short distances across these gradients relative to the chamber length, by baffles added to promote mixing, or by flow perpendicular to lamp axes. However, the bioassay method (Quails and Johnson, 1983) provides a means of detecting poor mixing across intensity gradients. In this method standardized bacterial spores were injected into a u.v. system and collected as a function of time from injection. Any systematic deviation from a linear relationship between assayed dose and time from injection for successive flow fractions would indicate poor mixing across intensity gradients (Quails and Johnson, 1983, Fig. 6). However, if the experiment indicates insufficient mixing across intensity gradients, then the bioassay results cannot be interpreted as the average intensity. Because the Aquafine unit used in this study was shown in experiments reported elsewhere (Quails et al., 1983) to have sufficient mixing across intensity gradients, we were able to use the average intensity in equation (10) but not able to verify the more general form of the model [equation (9)]. METHODS
Calculation of intensity The intensity at each point in the disinfection chamber was calculated by the point-source summation method (Quails and Johnson, 1983; Jacob and Dranoff, 1970). While this calculation is detailed elsewhere, we made several modifications to apply it to a multiple lamp reactor. Intensity at any point in the irradiated fluid was treated as the sum of contributions from each lamp, represented in turn by a line divided into point sources. We found that germicidal lamps transmit only about 250/0 of the 254nm u.v. light passing around them from a neighboring lamp (Johnson and Quails, 1981). So shadowing by neighboring lamps was included. Intensity was mapped by a computer program at each point on a grid in a cross-section of the chamber and averaged over the cross-section and along the length of the chamber (Johnson and Quails. 1981). Our calculations used these simplifications: (1) reflection from the reactor walls was negligible under operating conditions, (2) reflection and refraction by quartz tubes was ignored and (3) end effects on the lamp output measurement were negligible since the lamp length was 77 cm.
Determination of dose-surrical curces The survival of bacteria as a function of u.v. dose was determined by irradiating samples for various times in a
1042
ROBERT G. QUALLS and J. DONALD JOHNSON
stirred petri dish. The dish was located at the bottom of a collimating tube over which four germicidal lamps were suspended. Intensity at the liquid surface was measured with an International Light IL-500 radiometer (Qualls and Johnson, 1983). The average intensity within the suspension was calculated by an integration of Beer's Law over the fluid depth (Morowitz, 1950).
ultraviolet lamp quartz tube ~-
...., .-"
plexiqlass ~ cylinder ~ : l
Cylinder experiments To demonstrate the relationship between average intensity and fluid depth in a cylindrical geometry, we used the apparatus shown in Fig. 2. Average intensity within the cylinder was calculated by the point-source summation method. For this calculation lamp output was measured by the method o f Barrows (1951). To veri~ calculated average intensity in the cylinder, a bioassay measurement also was made (Quails and Johnson, 1983).
I
Pilot plant experiments The pilot u.v. units were located at the Sandy Creek contact stabilization wastewater treatment plant in Durham, NC. Effluent was pumped through either an Aquafine CSL-6 or a Pure Water Systems (PWS) IL-75 u.v, treatment system (Table I). While some treatments involved filtered effluents (Quails et al., 1983) only unfiltered runs are considered here. Thirty-six experiments were performed with: (1) flow rates either 2.27 or 4.92 1s -~ and (2) applied voltages o f either 60 or 128 V for each unit. Total coliforms were enumerated by the MPN procedure (APHA, 1975). The simulation involved only the data for the Aquafine unit. Residence time distributions were measured by injecting dye into the entrance o f the irradiation chamber and collecting samples o f the outflow in a rotating sampling tray as a function of time from injection.
stirring device
sliding black
I I
paper tube ] I between lamp [ I q
Model calculations The following data were needed for the simulation: (1) the coliform density in the influent (No), (2) the average intensity (I) calculated by the point-source summation method, (3) the RTD measured by dye injection and (4) the dose-survival curve, determined in the petri dish apparatus. Smooth curves were drawn through the RTD data and the log survival vs dose data. From these curves, data pairs for small increments of the independent variables (&time = 0.05 s and Adose = 2 mW-s cm--') were entered into arrays in a computer program. Intermediate values required in the calculations were generated by linear interpolation. For this simulation it was assumed that mixing was good across the intensity gradients and equation (10) was used to calculate the average survival.
Fig. 2. Cylindrical irradiation apparatus (from Quails and Johnson, 1983).
-I -J
-2 rv
-3
RESULTS AND DISCUSSION 0..J
As an example of the model of survival in a flow-through u.v. system we simulated runs with the Aquafine unit, These simulations were then compared to the observed survival in the pilot plant experiments with this unit. At the time of the pilot plant runs, methods to determine an accurate
-4
o
I 10
DOSE ( m W - s
I 20
I 30
cm -z)
Fig. 3. D o s e - s u r v i v a l curve f o r total coliforms in Sandy
Creek 2 = effluent (unfiltered) in a collimated u.v. beam,
Table 1. Comparison of Aquafine and pure water system units Characteristic Aquafine PWS u.v. Output, total (W) : :'. 62.3 77.8 Volume (1.) II .0 5.8 Calculated ave. intensity (mWcm--')* 9.7 18.8 Intensity × volume (roW cm -z) (I.) 107 107 Intensity-volume elficiency 0.45 0.31 (mWcrn-") (I.):'input wattage *At absorbance = 0.17.
Modeling u.v. disinfection Table 2. Observed vs simulated survival IS) of total colit'orm~in a u.v. disinfected Sandy Creek 2 secondar.~ effluent Lamp Av. intensity Simulated Pilot plant voltage (roW cm--' ) log S log S 60 6.2 - 3.26 - 329 t "--O.13) 128 9.7 - 3.61 - 3.69 ( __ O. 16)
dose-survival curve were not developed. So dose-response data were determined some time later for three samples from the same site (Fig. 3). The curve levels out substantially between - 3 . 0 and - 4 . 0 log survival units in marked contrast to ideal Chick's Law log-linear behavior. The R T D for the Aquafine unit (Fig. 4) was skewed and shows considerable dispersion. The average log survival predicted by the simulation corresponded extremely well with that observed in the pilot plant runs (Table 2). Some deviation might have been expected since the dose-survival curve was based on samples taken at later dates. To illustrate the effect of flow dispersion, we used the model to simulate the average Survival for a set of hypothetical R T D curves with a constant mean R T D , but with varying degrees of dispersion. The R T D in Fig. 4 was transformed b y the following equation so as to vary the standard deviation of the distribution while maintaining the same average residence time T, the same relative shape and area. For a given value on the y-axis (VA,) of distribution A and a corresponding value on the time axis (tA,), a transformed value for distribution B (tB,) was calculated: ts, = T -I-(I
T -t4, ISs/S4)
(13)
where T is the mean residence time, SA and SB are the standard deviations of distributions A and B respectively. The areas under the curves were normalized by calculating a new y-axis value (Vn,) for each value of distribution B:
liB,
=
V~,lV~,
(14)
where VA, and Va, are the areas under curves of VA, and Va, vs tA, and ta,, respectively. The set of curves shown in Fig. 5 was generated from the Aquafine R T D (Fig. 4). The simulations of average survival all used the same dose survival curve shown in Fig. 3. The upper plot in Fig. 5 is the _~
8
1043
0 L
~
15 F
I
t
i -A
3
a
T i M E (S)
Fig. 5. Effect of varying amounts of dispersion on average survival in RT distributions with the same average RT. Curve C was that measured for the Aquafine system. The other curves (A, B and D) were generated by transforming the time axis symetrically about the average RT. Above the RTD's, the survival curve used for the simulation (Fig. 2) is shown as a function of RT for the indicated intensities. Note what portions of the RTD's correspond to the "steep" and "flat" portions of the survival curves.
survival curve from Fig. 3 shown as a functibn of time for the two levels of intensity used in the simulations. It is shown above the residence time distributions so that the portion of the survival curve which corresponds with each part of the R T D below can easily be compared using the same time axis. The most significant comparison to be made is whether a substantial portion of the R T D corresponds to the steeply sloping initial portion of the log survival curve. The effects of flow dispersion on disinfection efficiency are dramatic when a significant portion of the flow distribution is in a sensitive part of the dose-survival curve, in this case a dose less than about 10 mW-s cm -2 (Fig. 4) or 1.5-2 s for intensity of 9.7 and 6.2 mW-s cm-'- respectively (Fig. 5, Table 3). Thus the effect of dispersion in the Aquafine unit was greater for the lower intensity level. Beyond the 10 mW-s c m - " dose, the dose-survival curve (Fig. 4) levels out to such an extent tllllt dispersion of flow had relatively little effect on the average survival. In this region increasing dose by increasing either t or I
¢. w Z'O
4
_
2
L=J ¢:
0
I
2 TIME (S)
3
4
Fig. 4. Residence time distribution for Aquafine unit at a flow rate of 4.92 I s- L The curve is normalized so that the area = 1.0.
Table 3. Simulated average log survival for the RTD's (curves A, B. C arid D) in Fig. 5. The standard deviation of the RTD's are indicated by S Av. log S S [ = 6.2 / = 9.7 A 0 -3.50 -3.68 B 0.33 -3.44 -3.68 C 0.66 -3.26 -3.61 D 1.00 -2.33 -2.76
10-M.
ROBERT G. QUALLSand J. DONALD JOHNSON
has little effect on survival. Unfortunately, the most dispersed RTD, curve D shown in Fig. 5, was more typical of the actual RTD for most of the small one to six lamp u.v. disinfection systems tested by the National Sanitation Foundation (Bellen et al., 1981). The variation in average survival caused by flow dispersion demonstrates that caution must be used in interpreting curves of average log survival vs either the mean residence time, or especially the residence time calculated from volume/flow rate. in practical flow-through systems (e.g. Scheible and Bassel, 1981). Although average log survival vs mean residence time may be a valuable guide to operating a certain system, such a dose-response curve may not be comparable for the same organism in another unit with different RTD characteristics. The distinction between the "'true" dose-survival curve (e.g. batch exposure with defined time and intensity) and the ambiguities of the averaged relationship should be appreciated. The strongly curved relationship of survival to dose in actual dose-survival curves also makes the first fractions of the RTD with a large dispersion disproportionately important. Haas and Sakellaropoulos (1979) proposed a model of u.v. disinfection in a flow-through system. They assumed that log survival was linearly related to dose. By assuming that the u.v. inactivation was a first order kinetic process, they were able to use the classical equations for a first order chemical reaction in a series of continuously stirred tanks (CSTR's) (Levenspiel, 1972). However, the simple equation these authors used to calculate intensity was not valid for an absorbing fluid and u.v. inactivation is seldom first order with dose (Quails and Johnson, 1983; Jacob and Dranoff, 1970). Scheible et al. (1983) have used a modification of the same approach again assuming first order survival kinetics, but using instead the point-source summation method (Quails and Johnson, 1983) to calculate the intensity and the dispersed PlUg flow model for open vessels (Levenspiel, 1972, Chap. 9) instead of the CSTR model. The analytical solution of both of these models depends on using ideal first order kinetics for survival. The dose-survival data from wastewater samples and the analysis of the effects of residence time distribution in this study demonstrate that order of magnitude errors in survival can result from the assumption of first order kinetics. Severin et al. (1984) derived an e q u a - ~ tion similar to that of Haas and Sakellaropoulos (1979) for a one-lamp, single, cylindrical CSTR by a somewhat different solution. They pointed out the inadequacy of using the simple first order reaction expression to describe data of pure cultures which exhibited an initial lag in slope (or "'shoulder"). Severin et al. (1983) also derived a solution .using a multi-target model (and a series event model). He used these to describe a survival curve with an initial "shoulder" for the same single lamp cylindrical CSTR. Both the multi-target and multi-hit types of survival curve, however, are linear beyond an initial
lag in slope (Taylor and Johnson, 1974) unlike the typical dose-survival relationship we found (Fig. 3). The advantage of the model described in this paper is that it uses any of the types of survival curves actually found in the field. This approach is also amenable to describing the complex flow regimes found in practice. Easily obtained experimental dose survival and RTD data are used directly. While the CSTR series model or a dispersion model could be used to describe the RTD, such models still require experimental verification. They only describe certain families of curves and require additional measurements and calculations to determine the parameters of the equations (Levenspiel, 1972). Effects of lamp spacing
There are a number of divergent views on the efficient design of u.v. systems. It has been suggested that closely spaced lamps are needed in u.v. disinfection of wastewater because of the high u.v. absorbance of wastewater solutions. In order to scientifically approach the question of efficient use of u.v. intensity, we contrasted the survival for a given volume of flow with different schemes of lamp spacing. Any surface which absorbs u.v. light inside the irradiation chamber (e.g. walls, baffles, other lamps), besides the unavoidable absorbance of the water itself, reduces the efficient use of the u.v. energy. The product of average intensity times the volume of the chamber is a factor which is directly proportional to the capacity of the unit under ideal flow conditions. The volume serves as a weighting factor to express the larger capacity of units with larger volume. This measure isolates the tradeoff between intensity and volume from the effects of non-ideal flow or how well the volume is used. To investigate the intensity-volume relationship in a simple cylindrical geometry we showed how the distance the light was allowed to penetrate, before encountering a nonreflective wall, affected the efficiency of light use. The intensity-volume product is shown for a series of cylinders of varying radius r around a u.v. lamp (Fig. 6) for the experimental apparatus shown in Fig. 2. The solid lines represent the point-source summation calculations of average intensity times the volume. The data points were derived from actual bioassay measurements of average intensity. The point at which the lines level out is the radius beyond which little u.v. light penetrates at a given absorbance in the cylindrical geometry. Obviously the effectiveness, as expressed by the intensity-volume product, was greater for the larger cylinders. For an absorbance representative of a good secondary wastewater effluent (0.16), it can be seen that walls or other obstructions within 5 or 6 cm of the source can absorb a significant amount of the u.v. light. The most efficient design depends predictably on absorbance at 254 nm with closely spaced lamps
Modeling u.v. disinfection Ld
:~
I
zo L--
>-7~ L ~~ z- ~
0
8F
~ 0 0
2
4
6
RADIUS (cm}
Fig. 6. Intensity-volume product in cylinders of varying radii around a u.v. lamp. Lines represent calculations of average intensity times the volume in the cylinders in solutions of absorbances 0, 0.16 and 0.32. Data points represent the average intensity, measured by the bioassay, times the volume.
t045
responses. Both the intensity-volume product and RTD need to be reported for any system. The only single parameter which might be useful as an ultimate comparison of disinfection capacity would be a simulation using a standard dose-survival curve (e.g. Fig. 3) at a given flow rate. The parameter indicating effectiveness would be the maximum flow rate yielding--3.5 log survival units (for example) using the standard dose-survival curve. Such a parameter would provide an easily interpreted and realistic comparison of units operated under different conditions.
Applications of simulation of u.v. disinfection Simulation takes into account the factors of the intensity use, of flow dispersion or time characteristics and sensitivity of the target organisms. Model calculations can be a useful tool for research and development of reactor design. For example, it can be used to find optimum lamp configurations and tradeoffs with flow dispersion, water quality, etc. It can be used to predict the design parameters needed for a specific situation so that costly overdesign is not necessary. The predicted survival of a standard coliform sample at a given flow rate may be used to compare a number of different reactors. The difficulties in control of the process of u.v. disinfection has been regarded as a disadvantage of u.v. disinfection. However, model simulations may also be used to prepare empirical curves of predicted survival vs flow rate, lamp voltage, water quality, etc. as a guide to continuous operation for a particular installation (Dorfman, 1984).
decreasing efficiency except for highly absorbing solutions. Two units used in.the pilot plant experiments were compared on the basis of their intensity-volume products (Table 1). The unit with lamps placed close to one another and the walls (PWS unit) had an average intensity almost twice as high as the other unit (Aquafine). However, the PWS unit had a much smaller volume (and short residence time) so the intensity-volume products were almost equal. However, the PWS reactor required a larger total lamp wattage to produce this intensity-volumeproduct. The intensity-volume "efficiency" (intensity × volume/ input wattage) compares the efficiency of the use of the lamp wattage. The PWS system was less efficient in this respect. The proximity of the lamps to the Acknowledgements--We thank Kent H. Aldrich, Michael P. walls and each other causes loss of u.v. due to Flynn, Donald E. Francisco and Thomas S. Wolfe who contributed to the pilot plant study. absorption of the light. Since neighboring u.v. lamps The initial research in this paper was supported partially transmit little of the 254 nm light coming from adja- by a grant R80470010 from the U.S. Environmental Proteccent lamps, excessively close lamp spacing can result tion Agency, Albert D. Venosa, project officer. Modeling in reduced efficiency of u.v. light use. This reduction work was supported by a grant CEE82-05274 from the U.S. National Science Foundation, Edward H. Bryan, program in efficiency has been calculated for a large rectandirector. gular array of lamps using various lamp spacings (Johnson and Quails, 1981). REFERENCES It is important to note that the intensity-volume product does not account for the effects of nonideal APHA (1975) Standard Methods for the Examination of Water and Wastewater, 14th edition. American Public flow. Its value is simply to isolate the effectiveness of Health Association, Washington, D.C. the intensity use from hydraulic problems or the Barrows W. E. (1951) Light, Photometry, and Illuminating effectiveness of volume use so that these two probEngineering, 3rd edition. McGraw-Hill, New York. lems in design may be addressed separately. Although Bellen G. E., GoRier R. A. and Dormand-Herrera R. (1981) Water Disinfection Technology. National Sanitation the intensity-volume product of the two units used in Foundation, Ann Arbor, Mich. the pilot plant study were nearly equal (Table 1), the Chick H. (1908) An investigation of the laws of disinfection. PWS unit allowed 0.6-2.1 log units greater survival J. Hyg. 8, 92-158. than the Aquafine unit with the same Cortelyou J. R., McWhinnie M. A., Riddiford M. S. and intensity-volume product because of severe s h o r t Semrad J. E. (1954) Effects of ultraviolet irradiation on large populations of certain water-borne bacteria in circuiting and lack of plug flow in the PWS unit. Thus motion--I. The development of adequate agitation to the effects of the intensity-volume factor ancl-the provide an effective exposure period--II. Some physical effects of flow dispersion must be considered sepafactors affecting the effectivenessof germicidal ultraviolet rately. There is no simple single parameter to provide radiation. Appl. Microbiol. 2, 262-269. a balanced measure of the ultimate effectiveness of a Dorfman M. H. (1984) Applications of a model of ultraviolet disinfection. Masters thesis, Department of Enviu.v. disinfection system with different organismal
1046
ROBERT G. QUaLLS and J, DOYALOJo~yso~
ronmental Sciences and Engineering, University of North Carolina, Chapel Hill, NC. Haas C. N. and Sakellaropoulos G. G. (1979) Rational analysis of ultraviolet disinfection reactors. In Proceedings o f the National Conference on Environmental Engineering, pp. 540--547. American Society of Civil Engineers, Washington, DC. Jacob S. M. and Dranoff J. S. (1970) Light intensity profiles in a perfectly mixed photoreactor. Am. Inst. chem. Engng J. 16, 35%363. Jagger J. H. (1967) Introduction to Research in UV Photobiology. Prentice-Hall, Englewood Cliffs, NJ. Johnson J. D. and Qualls R. G. (1981) Ultraviolet disinfection of secondary effluent: measurement of dose and effects of filtration. Report of EPA project R 804770010, Municipal Environmental Research Laboratory, Cincinnati, OH. Levenspiet O. (1972) Chemical Reaction Engineering. Wiley, New York. Morowitz H. J. (1950) Absorption effects in volume irradiation of microorganisms. Science ! 11, 229-230. National Research Council (1980) Drinking Water and Health. National Academy of Sciences Press, Washington, DC. Oliver B. G. and Cosgrove (1975) The disinfection &sewage treatment plant effluents using ultraviolet light. Can. J. chem. Engng 53, 170-174. Petrasek A. C,, Wolf H. W., Edmond S. E. and Andrews D. C. (1980) UItraviolet disinfection of municipal wastewater effluents. EPA No. 600/2-80/102. U.S. Environmental Protection Agency, Washington, DC. Quails R. G. and Johnson J. D. (1983) Bioassay and dose
measurement in UV disinfection. Appl. envir. Microbiol. 45, 872-877. Quails R. G., Flynn M. P. and Johnson J. D. (1983) The role of suspended particles in ultraviolet disinfection. J. Wat. Pollut. Control Fed. 55, 1280-1285. Roeber J, A. and Hoot F. M. (1975) Ultraviolet disinfection of activated sludge effluent discharging to shellfish waters. EPA No. 600/2-75-060. U.S. Environmental Protection Agency, Washington, DC. Scheible O. K. and Bassel C. D. (1981) Ultraviolet disinfection of a secondary wastewater treatment plant effluent. EPA No. 600/$2-81-52. U.S. Environmental Protection Agency, Washington, DC. Scheible O. K., Forndran A. and Leo W. M. (1983) Pilot investigation of ultraviolet wastewater disinfection at the New York City Port Richmond Plant. In Municipal Wastewater Disinfection (Edited by Venosa A. D.), pp. 202-225. EPA. No. 600/9-83-009, U.S, Environmental Protection Agency, Washington, DC. Severin B. F., Suidan M. T. and Engelbrecht R. S. (1983) Kinetic modeling of u.v. disinfection of water. Water Res. 17, 1669-1689. Severin B. F., Suidan M. T., Rittmann B. E. and Engelbrecht R. S. (1984) Inactivation kinetics in a flow-through UV reactor. J. Wat. Pollut. Control Fed. 56, 164-169. Taylor D, G. and Johnson, J. D. (1974) Kinetics of viral inactivation by bromine. Chemistry of Water Supply, Treatment and Distribution (Edited by Rubin A. J.), pp. 369-408. Ann Arbor Science, Ann Arbor, MI. Ward R. W. and DeGrave G. M. (1978) Residual toxicity of several disinfectants in domestic wastewater. J. Wat. Pollut. Control Fed. 50, 46-60.