Modeling and testing of reactive contaminant transport in drinking water pipes: Chlorine response and implications for online contaminant detection

Modeling and testing of reactive contaminant transport in drinking water pipes: Chlorine response and implications for online contaminant detection

ARTICLE IN PRESS WAT E R R E S E A R C H 42 (2008) 1397 – 1412 Available at www.sciencedirect.com journal homepage: www.elsevier.com/locate/watres ...

678KB Sizes 1 Downloads 14 Views

Recommend Documents

No documents
ARTICLE IN PRESS WAT E R R E S E A R C H

42 (2008) 1397 – 1412

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Modeling and testing of reactive contaminant transport in drinking water pipes: Chlorine response and implications for online contaminant detection Y. Jeffrey Yanga,, James A. Goodricha, Robert M. Clarkb, Sylvana Y. Lic a Water Supply and Water Resources Division, US EPA National Risk Management Research Laboratory, 26W Martin Luther King Drive, Cincinnati, OH 45268, USA b Environmental Engineering Consultant, 9627 Lansford Drive, Cincinnati, OH 45242, USA c Foreign Agricultural Service, US Department of Agriculture, 1400 Independence Avenue, S.W., Washington, DC 20250, USA

art i cle info

ab st rac t

Article history:

A modified one-dimensional Danckwerts convection–dispersion–reaction (CDR) model is

Received 28 November 2006

numerically simulated to explain the observed chlorine residual loss for a ‘‘slug’’ of reactive

Received in revised form

contaminants instantaneously introduced into a drinking water pipe of assumed no or

1 October 2007

negligible wall demand. In response to longitudinal dispersion, a contaminant propagates

Accepted 6 October 2007

into the bulk phase where it reacts with disinfectants in the water. This process generates a

Available online 13 October 2007

U-shaped pattern of chlorine residual loss in a time-series concentration plot. Numerical

Keywords: Adaptive early warning Contaminant detection Water distribution Convection–dispersion–reaction model Chlorine residual loss Chlorination Drinking water quality Aldicarb

modeling indicates that the residual loss curve geometry (i.e., slope, depth, and width) is a function of several variables such as axial Pe´clet number, reaction rate constants, molar fraction of the fast- and slow-reacting contaminants, and the quasi-steady-state chlorine decay inside the ‘‘slug’’ which serves as a boundary condition of the CDR model. Longitudinal dispersion becomes dominant for less reactive contaminants. Pilot-scale pipe flow experiments for a non-reactive sodium fluoride tracer and the fastreacting aldicarb, a pesticide, were conducted under turbulent flow conditions (Re ¼ 9020 and 25,000). Both the experimental results and the CDR modeling are in agreement showing a close relationship among the aldicarb contaminant ‘‘slug’’, chlorine residual loss and its variations, and a concentration increase of chloride as the final reaction product. Based on these findings, the residual loss curve and its geometry are useful tools to identify the presence of a contaminant ‘‘slug’’ and infer its reactive properties in adaptive contaminant detections. Published by Elsevier Ltd.

1.

Introduction

Small quantities of contaminant can enter a water distribution system due to back siphonage induced by transient pressures at pipe joints, leaks and breaks (LeChevallier et al., 2003; Kirmeyer et al., 2001), cross-connections and biological sloughing (US EPA, 2002) and through intentional sabotage Corresponding author. Tel.: +1 513 569 7655; fax: +1 513 569 7185.

E-mail address: [email protected] (Y. Jeffrey Yang). 0043-1354/$ - see front matter Published by Elsevier Ltd. doi:10.1016/j.watres.2007.10.009

and terrorist activities (US EPA, 2005). Such contamination is generally limited in time and space (Boyd et al., 2004), but can result in water quality deterioration and possibly unacceptable health risk (Burrows and Renner, 1999; GAO, 2005; WHO, 1996; Whelton et al., 2003). Detection and monitoring of the contamination events is a technical challenge, but is essential to the delivery of safe drinking water and to the security of

ARTICLE IN PRESS 1398

WAT E R R E S E A R C H

42 (2008) 1397– 1412

Nomenclature 

c¼C C* CR c^ Dc

D Deff E* fm kb kf

=CR chlorine concentration in water (dimensionless) chlorine concentration in water (M/V) chlorine concentration at reference point (M/V) transverse-averaged chlorine concentration (dimensionless) concentration difference between top and bottom of a parameter (e.g., chlorine residual, chloride, etc.) loss or increase curve (dimensionless) molecular diffusivity in water (L2/T) effective chlorine dispersion coefficient in water (L2/T) tracer or contaminant exit age, 1/T residual loss fraction from fast-reacting contaminants (dimensionless) bulk chlorine decay constant (dimensionless)

reaction rate constant for fast-reacting contaminants (dimensionless) k0s reaction rate constant for slow-reacting contaminants in the introduced contaminant slug (dimensionless) ks overall reaction rate constant of slow-reacting contaminants (dimensionless) L ¼ ns ro pipe length of interest in units of pipe radius or between two sensor stations along a straight pipe (L) ns scaling factor for pipe length of interests in units of pipe radius (dimensionless) PeD transverse Pe´clet number (dimensionless) PeL axial Pe´clet number (dimensionless)

Q flow rate (V/T) R1, R2 empirical model parameters in the competitive

r ¼ r =ro radial distance from center line of a straight pipe (dimensionless); variables r*and ro are dimensional radial distance and the pipe radius, respectively in a cylinder coordinate system RðcÞ ¼ RðC ; t Þ=CR dimensional reaction rate at a reference location, 1/T Re Reynolds number t ¼ t =tc dimensionless time normalized to tc; t* is absolute time (T) tc ¼ ns ro =u¯ x characteristic convection residence time (T)

tD ¼ ro 2 =Deff characteristic radial diffusion time (T) Dt half-width of the chlorine residual loss curve in a Dti u* u¯ x

c– t plot (dimensionless) sampling time interval in tracer testing (T)

water flow velocity (L/T) transverse-averaged water flow velocity (dimensionless) x ¼ x =ns ro axial distance along the pipe normalized to L* (dimensionless) y* dimensional azimuth angle in pipe cross-section in a cylinder coordinate system (Fig. 1a) ^ Wðc; tÞ chlorine decay rate including bulk demand and wall demand M, L, T, V unit symbol for mass, length, time and volume, respectively Subscript and superscript

b p inj * i

bulk phase pipe flow solution injection dimensional parameter ith measurement

second-order kinetics for chlorine decay (Clark and Sivaganesan, 2002) (dimensionless)

critical water infrastructure. Contaminant detection methodologies combined with model simulations have been suggested to detect and manage such incidents (Jain and McLean, 2006; Boulos et al., 1994; Harding and Walski, 2000; Caputo and Palagagge, 2002; Uber et al., 2004). Most research on predictive modeling in water supply has focused on chlorine transport in a distribution network (e.g., Lu et al., 1993; Rossman et al., 1994; Clark and Haught, 2005; Tzatchkov et al., 2002). The majority of these models describe the decay of chlorine and other disinfectants as a result of their reaction with water-born chemicals in the bulk phase and the reaction with stationary biofilm and pipe wall surfaces (Clark et al., 2001). Models such as EPANet (Rossman, 2002) and its commercial derivatives have been widely used in engineering design, network operation and management (Harding and Walski, 2000; US EPA, 2006). In describing chemical (chlorine) transport in water pipes, mass conservation principles apply. Biswas et al. (1993) and Rossman et al. (1994) used these principles and developed the widely used chlorine transport models that were evaluated and further

extended by Clark and Haught (2005). Wang and Postlethwaite (1997) developed mass transfer models for the flow of low Reynolds numbers in pipe corrosion studies. Heinz and Roekaerts (2001) explored mass conservation models for multi-scale mixing and reaction in turbulent pipe flows. Leonenko et al. (2005) applied the principles to pipe flow modeling of reactive droplets and polydispersed systems. Lu et al. (1993) derived simplified models for study of steadystate multi-species transport in a water distribution system, and Islam and Chaudhry (1998) numerically simulated transient pipe flows and contaminant transport. The same principles can be also found in studies of classic heat transfer in tubular conduits (Cussler, 1984) and of mass transfer in chemical reactors (Levenspiel, 1972; Chakraborty and Balakotaiah, 2002; Forney and Nafia, 2000; Levien and Levenspiel, 1999). Chlorine decay models help in understanding chlorine reactions with contaminants in the bulk phase (bulk demand) and with biofilm and pipe wall surfaces (wall demand) during water distribution. Few models have been developed to study

ARTICLE IN PRESS WAT E R R E S E A R C H

42 (2008) 1397 – 1412

chemical and kinetic heterogeneity in bulk phase. Research by Powell et al. (2000) on UK water supplies showed the dependence of bulk demand on temperature, pH, total organic carbon and other water quality parameters. Clark and Sivaganesan (2002) proposed empirical equations that relate chlorine decay kinetics to several water quality parameters. When a discrete ‘‘slug’’ of contaminant is introduced to bulk flow, it will change the parameters of the receiving water cited by Powell et al. (2000) and consequently will generate local heterogeneity in chlorine decay rates. The associated convection–dispersion–reaction (CDR) process has not been fully investigated for water quality in drinking water distribution, nor represented in the chlorine decay models. In contrast, the phenomenon has been extensively studied in chemical and pharmaceutical engineering (Baldyga et al., 1996; Levien and Levenspiel, 1999; Forney and Nafia, 2000; Chakraborty and Balakotaiah, 2002) where an understanding and control of micro-scale mixing and reaction are important to product quality and quantity. Hladik et al. (2005) showed that chlorine and other disinfectants in drinking water can oxidize a number of herbicide compounds. Similar reactions occur resulting in the decomposition and transformation of recalcitrant pharmaceutical compounds subject to chlorination (Stackelberg et al., 2004). Some investigators (Khordagui, 1995; Burrows and Renner, 1999; Hamilton et al., 2006) have suggested that maintaining a high level of chlorine residual can protect against health risks from a number of hazardous contaminants in drinking water. The level of risk reduction depends on the quantity and composition of the contaminant ‘‘slug’’, the contaminant-chlorine reaction rate, and the traveling time of contaminated water reaching a consumer. Because of the importance of these types of localized reactions, new models and additional research are needed to adequately predict fate and transport of a small volume of contaminants introduced into drinking water. For this purpose, a series of experiments coupled with computer modeling have been undertaken using pilot-scale pipe flow devices. The objective is to define contaminant-chlorine reactions taking place during the transport of water in a pipe, to characterize the hydraulic dispersion of non-reactive chemicals, to assist in detecting contamination events using water quality sensors, and to establish a working model for predicting the fate and movement of a reactive contaminant ‘‘slug’’. This paper is the first of a series that describe these studies. It begins with an introduction followed by description of experiment testing and its results. To further explore temporal and spatial changes in chlorine residual concentrations, a CDR model is proposed to examine the relationships between the concentration changes, reactive properties of the contaminants and flow hydrodynamic properties in a water pipe. Finally, these results will be discussed in light of realtime contaminant detection in drinking water distribution systems.

2.

Experiments and results

For the purpose of this paper, results from two sets of experiments are presented. The first was conducted to

1399

investigate the dispersion of non-reactive sodium fluoride in a pilot-scale pipe flow device. The second was conducted on reactive aldicarb in a fiberglass-lined water pipe to qualify the effect of both dispersion and the fast reactions between aldicarb and chlorine disinfectant. The difference in results between the two sets of experiments is used to identify the effects of CDR process.

2.1.

Sodium fluoride tracer experiments

Li (2000) studied the transport of sodium fluoride in a 14.60cm inside diameter, horizontally placed new polyvinyl chloride (PVC) pipe with three 901 bends (Fig. 1a). Water exited at the end of the pipe 112.2 m downstream from the tracer injection port, where samples were collected and analyzed for fluoride ion concentration using a Hach 2000 spectrophotometer. In the experiments at Qp ¼ 69.1–69.3 L/min, 450-mL sodium fluoride solution (7300–16,900 mg/L) was injected into the pipe to form an instant fluoride ‘‘slug’’ at a calculated concentration of 208–562 mg/L after mixing. Based on the fluoride concentration measurements at the end of pipe (Table 1), tracer exit age (Levenspiel, 1972) and the fluoride dispersion coefficient in water are calculated: Ci ,   i¼1 Ci Dti

E ¼ Pn

2

   Deff Deff 2    1  eðu L =Deff Þ   2   L u L u Pn 2  Pn  !2 ti Ci C Pni¼1 i  ¼ Pi¼1  1. n  i¼1 Ci i¼1 ti Ci

(1)

ð2Þ

The tracer exit age defined a nearly symmetric E-curve (Levenspiel, 1972) with the peak around the hydraulic retention time (Fig. 2). In a 112.2-m flow distance, the instantaneously injected ‘‘slug’’ of tracer was dispersed into a water column of 7–13 min pipe flow at the pipe exist (Table 1). Maximum concentration detected was 1.45–3.33 mg/L after background corrections, representing 497 times of concentration decrease from the initial concentration 208–562 mg/L at the injection port. Li (2000) reported the conservation of recovered tracer mass in the experiments and attributed the large concentration reduction as a result of dispersion in the pipe. From Eq. (2), the experimental data yielded effective dispersion coefficients of 59.8– 216.1 cm2/s for fluoride which can be used in chlorine transport modeling.

2.2.

Reactive aldicarb experiments

The experiment apparatus for aldicarb consisted of a 7.62-cm inside diameter, 426.8-m-long straight ductile iron pipe in a configuration similar to that of the tracer experiments (Fig. 1b). The ductile iron pipe, fiberglass-lined inside for negligible wall demand, was long enough that two sensor stations could be used to quantify water quality changes along its length. In the experiments (Qp ¼ 83.2 L/min and Re ¼ 25,000), a volume of 50-L aldicarb solution was injected into the pipe through a port 24.4 m upstream from the sensor station #1 (Fig. 1b). Calculated initial aldicarb concentration was 0.2, 1.1 and 2.2 mg/L in the ‘‘slug’’ after assumed

ARTICLE IN PRESS 1400

WAT E R R E S E A R C H

42 (2008) 1397– 1412

Cincinnati tap water

Dimensions: 112.2 m – Total pipe length 14.60 cm – Inside pipe diameter

V=2841 L

112.2 m – From location 0 to 1 Injection port 0

Qp=69.1 – 69.3 L/min

Qinj=1.27 – 2.38 L/min 1 Sampling port

Cincinnati tap water

Dimensions: 426.8 m – Total pipe length 7.62 cm – Inside pipe diameter

V=2841 L

24.4 m – From location 0 to 1 335.4 m – From location 0 to 2

Injection port 0 Qinj=2.5 L/min

1

Qp=83.2 L/min

2

To drain

To sensor station

Fig. 1 – Schematic diagram showing the two pilot-scale water distribution pipe devices used in the experiments: (A) nonreactive tracer sodium fluoride, and (B) fast-reacting aldicarb. Sensor stations are marked in (1) and (2). Injection port is labeled in (0).

complete mixing. At two sensor stations, a side water stream approximately 0.5 L/min was diverted from the pipe to measure free chlorine, total chlorine and chloride concentration using ATI model A15 sensors and Hach CL-17 total chlorine analyzer. Measured free chlorine and total chlorine concentrations are shown in the time-series concentration (c– t) plots in Fig. 3. Experimental data are given in Szabo and Hall (2006). Clearly the chlorine residual loss shows a U-shaped curve pattern with a slightly wider opening at the top than at the bottom. This curve geometry is little changed between the two sensor stations in a 311-m flow distance, and is consistent among all three initial aldicarb concentrations. The depth of residual loss curve [Dc], which is defined as the concentration difference between the background and the curve’s flat bottom (Fig. 4), is 8–10%, 43–45% and 87–88% for the initial aldicarb concentrations of 0.2, 1.1 and 2.2 mg/L, respectively (Table 2). Measured Dc values and the initial aldicarb concentrations are linearly correlated. The chlorine and chloride measurement results lead to the suggestion that the reactive contaminant propagated

from the ‘‘slug’’ into the bulk phase and simultaneously reacted with chlorine producing a residual loss at the contaminant–water interface. Consequently, chlorine residual concentration varied continuously at both upstream and downstream sides of the ‘‘slug’’, forming the characteristic U-shaped pattern (Fig. 4). Furthermore, average width of the residual loss curve remained similar within the measurement error among the experiments of different initial aldicarb concentrations and for the two sensor stations spaced 311 m apart (Table 2). In the free chlorine c– t plots, for example, the average opening and bottom width is 20.370.9 min (m ¯  1s, n ¼ 6) in units of water flow time. Another geometric measurement is the curve width 19.870.3 min (m ¯  1s, n ¼ 6) at the 50% residual loss (Table 2). Both measurements are statistically identical to the 20-min duration of contaminant injection, strongly suggesting that the contaminant ‘‘slug’’ experienced no significant dispersion during the 335.4-m flow distance. This is in contrast to the greater degrees of dispersion observed in the non-reactive tracer experiments (Fig. 2). Even in the same aldicarb experiments, chloride as the reaction product with a noticeable concentration increase

ARTICLE IN PRESS WAT E R R E S E A R C H

1401

42 (2008) 1397 – 1412

Table 1 – Measured fluoride concentrations, experimental conditions and the determined Deff values for four tracer tests at a pilot-scale 112.2-m-long straight water pipe flow device Time (min)

No. 18–1

No. 18–2

No. 18–3

No. 18–4

Measurement

0 18 19 20 22 23 24 25 25.25 25.5 25.75 26 26.25 26.5 26.75 27 27.25 27.5 27.75 28 28.5 29 29.5 30 31 32 33 36 39 40 50 70 120

0.93 0.97 0.97 – 0.95 – 1.09 1.93 – – – 2.12 – 2.4 – 1.75 – 1.30 – 1.18 1.08 1.06 1.01 1.00 0.99 0.97 0.96 0.96 0.96 – – 0.94 0.99

0.91 0.90 – 0.93 – 0.95 1.56 2.95 – – – 3.05 2.55 2.35 1.9 1.68 1.52 1.43 1.35 1.19 1.09 1.11 1.00 0.97 0.97 – 0.93 – – 0.93 0.91 0.93 –

0.84 0.86 – 0.85 – 0.91 1.11 2.88 – – – 3.7 3.25 2.92 2.64 2.28 2.02 1.94 1.64 1.57 1.43 1.25 1.12 1.04 1.00 – 0.94 – – 0.88 0.88 0.88 –

0.87 0.94 – 0.93 – 0.94 1.23 3.22 3.28 4.24 3.48 3.36 3.12 2.88 2.44 2.06 1.92 1.72 1.54 1.41 1.23 1.15 1.07 1.04 1.00 – 0.91 – – 0.93 0.94 0.88 –

Background Maximum (mg/L)

0–22

0.96 1.45

0.91 2.14

0.85 2.85

0.91 3.33

9020 69.1 1308 59.8

9020 69.3 481 162.7

9020 69.3 362 216.2

9020 69.3 704 111.2

28.1 7300 205.1 0.75

31.0 11,600 359.6 1.47

31.8 16,600 527.9 1.00

30.9 16,900 522.2 0.78

0.86 230

1.69 208

1.15 446

0.90 562

Parametersa Re Q (L/min) PeL Deff (cm2/s) Injectionb Volume (ml) Concentration (mg/L) Tracer mass (mg) Duration (s) Tracer slugc Volume (L) Concentration (mg/L)

Note: All fluoride concentrations analyzed using a Hach 2000 spectrophotometer for collected grab samples. PeL and Deff calculated from besting fit to the experiment results. See Li (2000). b Calculated from the testing conditions. c Calculated assuming instant complete mixing in the pipe at the injection port. a

exhibits a different c– t curve geometry than the reactant chlorine. An extended dispersion tail (Levenspiel, 1972) is noticeable for t*421.5 min at the 24.4-m sensor station and is more pronounced for t*442.2 min at the 335.4-m station, especially in the 2.2 mg/L experiments (Fig. 3).

3.

Modeling

In addition to the experimental work described above, the differences between non-reactive sodium fluoride and

ARTICLE IN PRESS 1402

WAT E R R E S E A R C H

42 (2008) 1397– 1412

0.4

0.4

No.18-2

No.18-1 0.3 E* (1/min)

E* (1/min)

0.3

0.2

0.2

0.1

0.1

0.0

0.0 0

500

1000

1500 2000 t = t*u*/(2r*)

2500

0

3000

1000

1500 2000 t = t*u*/(2r*)

2500

3000

2500

3000

0.4

0.4

No.18-4

No.18-3 0.3 E* (1/min)

0.3 E* (1/min)

500

0.2

0.2

0.1

0.1

0.0

0.0 0

500

1000

1500

2000

2500

3000

0

500

1000

t = t*u*/(2r*)

1500

2000

t = t*u*/(2r*)

Fig. 2 – Model-simulated E-curves (Levenspiel, 1972) and sodium fluoride experimental data for the four tracer tests at a 112.2-m-long PVC water pipe under turbulent flow conditions (Re ¼ 9020). Deff ¼ 33 cm2/s was used for the CDR model ^ tÞ ¼ 0 and for the analytical solution of Levenspiel (1972). Also shown for comparison are the calculation at k ¼ 0 s1 and Wðc; f

E-curves for the analytical model using experimental Deff values of Li (2000) in Table 1.

fast-reacting aldicarb were further examined using a one-dimensional CDR model modified from Danckwerts equations.

3.1. pipe

Fast-reacting contaminant ‘‘slug’’ transport in water

Chlorine concentration C ðx ; r ; y ; t Þ in a water pipe follows the mass conservation equation: 

















rt C þ u ðx ; r ; y Þ  rx C ¼ rr rx ðDeff C Þ  RðC ; t Þ,

(3)

where Deff is the effective chlorine dispersion coefficient in water; R(C*, t*) is the rate of chlorine decay including both bulk demand and wall demand. The velocity scalar u*(x*, r*, y*) represents the transverse velocity field as a function of axial distance (x*), radial distance (r*) and azimuth angle (y*) in a cylinder coordinate system (Fig. 5a). In the simplest term,

Eq. (3) describes the chlorine decay rate [qc/qt] as a function of convection XxC*, longitudinal and radial dispersion XrXxC*, chlorine bulk demand and wall demand R(C*, t*). Modified from Biswas et al. (1993) and Chakraborty and Balakotaiah (2002) and using dimensionless parameters, !   qc qc tc 1 q qc 1 q2 c  tc RðcÞ, þ uðx; r; yÞ ¼ r þ 2 (4) qt qx tD r qr qr ns PeD qx2 PeD ¼

tD ro 1 u¯  ¼ ¼ PeL . tc ns Deff x n2s

(5)

The parameters PeL and PeD are axial and transverse Pe´clet numbers, respectively. Dimensionless radial distance (r), axial distance (x), time (t) and concentration (c) are the dimensional equivalents normalized against pipe radius (ro ), pipe length of interest (ns ro ), characteristic convection time (tc) and chlorine concentration at a reference pipe location (CR*),

Fig. 3 – The concentration-time (c– t) plots showing free chlorine, total chlorine and chloride concentration variations measured at the 24.4 and 335.4-m sensor stations for initial aldicarb concentrations 0.2, 1.1 and 2.2 mg/L. Characteristic U-shape geometry in the chlorine residual loss curve is persistent for all three tested aldicarb concentrations and along the flow path. Chloride concentration increase is evident particularly at the high initial aldicarb concentrations. Concentrations for free chlorine (Clf), total chlorine (Clt), and chloride are normalized against arithmetic mean of first 16 measurements taken as the background.

1.04

1.10 1.05

1.10

1.04

1.05

1.03

1.00

1.02

0.95

1.01

0.90

1.00

0.90 0.2 mg/L 24.4 m

0.80 -40

0.85 0.96

-20

0

20

40

60

0.80

80

Clf Clt Chloride

0.2 mg/L 335.4 m

0.98 -40

-20

0

t* (min)

20

40

60

80

t* (min)

c (Clf,Clt)

0.60

1.00 0.8 0.99 0.6 0.98

1.1 mg/L 24.4 m

0.20 -40

0.4 0.94

-20

0

20

40

60

1.1 mg/L 335.4 m

0.2 -40

80

Clf Clt Chloride

0.96 -20

0

t* (min)

20

40

60

80

t* (min)

1.2

1.04

1.2

1.04

1.0

1.02

1.0

1.02

0.6 0.98 0.4 0.2

Clf Clt Chloride

2.2 mg/L 24.4 m

0.96

0.0

0.94 -20

0

20 t* (min)

40

60

80

1.00 0.6 0.98 0.4 0.2

Clf Clt Chloride

2.2 mg/L 335.4 m

0.0

0.96 0.94

-40

-20

0

20 t* (min)

40

60

80

1403

-40

0.8 c (Clf,Clt)

1.00

c (Chloride)

0.8 c (Clf,Clt)

0.97

c (Chloride)

0.40

0.96

Clf Clt Chloride

42 (2008) 1397 – 1412

0.98

c (Chloride)

0.80

1.01

1.0

1.00

ARTICLE IN PRESS

1.00

1.02

1.2

WAT E R R E S E A R C H

1.02

1.20

c (Clf,Clt)

0.99

c (Chloride)

0.85

0.98

Clf Clt Chloride

c (Chloride)

1.00

0.95

c (Clf,Clt)

1.00

c (Chloride)

c (Clf,Clt)

1.02

1.2

1.0

1.0 0.92± 0.005

0.6 0.55 ± 0.01

0.4 0.2

24.4 m -10

0.2 mg/L 1.1 mg/L 2.2 mg/L

0.12 ± 0.01 0

10

0.6 0.56 ± 0.01 0.4

Clf

0.0

0.90± 0.004

0.8 c (Clf)

c (Clf)

0.8

1404

1.2

0.2

335.4 m

0.13 ± 0.01

0.0 20

30

40

10

20

30

50

60

t* (min) 1.2

1.0

1.0 0.92± 0.005

0.92± 0.003

c (Clt)

0.8

0.6

0.6

0.57± 0.008

0.58± 0.005

0.4

0.4

0.2

Clt

0.2 mg/L 1.1 mg/L 2.2 mg/L

0.18± 0.007

24.4 m 0.0

0.2

Clt

0.2 mg/L 1.1 mg/L 2.2 mg/L

0.20± 0.012

335.4 m 0.0

-10

0

10

20 t* (min)

30

40

10

20

30

40

50

60

t* (min)

Fig. 4 – Progressive chlorine residual loss at initial aldicarb concentrations 0.2, 1.1 and 2.2 mg/L. Average chlorine concentration and one standard derivationðm ¯  1rÞ are marked for the chlorine (Clf) and total chlorine (Clt) measurements. Note that the U-shape chlorine residue loss is consistent at all initial aldicarb concentrations and persistent in the flow between two sensor stations of 311-m distance.

ARTICLE IN PRESS

1.2

42 (2008) 1397– 1412

c (Clt)

40

WAT E R R E S E A R C H

t* (min)

0.8

0.2 mg/L 1.1 mg/L 2.2 mg/L

Clf

ARTICLE IN PRESS WAT E R R E S E A R C H

1405

42 (2008) 1397 – 1412

Table 2 – Observed geometry of the free chlorine residual loss curve in the aldicarb pipe flow experiments Chlorine loss curve parameter

Residual loss curve geometry x* ¼ 24.4 m

x* ¼ 335.4 m

0.2 mg/L

1.1 mg/L

2.2 mg/L

0.2 mg/L

1.1 mg/L

2.2 mg/L

23 16 19.5 19.2 0.08 20

24 16 20.0 19.6 0.45 20

29 15 22.0 19.9 0.88 20

25 15 20.0 20.0 0.10 20

25 16 20.5 19.9 0.43 20

28 11 19.5 20.0 0.87 20

Opening width Dt* (min) Bottom width Dt1* (min) Average width (Dt*+Dt1*)/2 (min) 50% residual loss curve width (min) Residual loss curve depth (Dc) Slug width at injection (t* ¼ 0)

Note: All widths can be converted to equivalent pipe length by multiplying average flow velocity 30.5 cm/s.

Sensor #2

Sensor #1 L*

X1*

X2* water flow r*

ux

θ∗

Contaminant slug x* Reactive contaminant slug

Residual loss in contaminant slug co

Chlorine 2ˆ Deff ∂ c2 ∂x

ux

2ˆ D ∂ c2 ∂x

∂cˆ ∂x Reaction zone

x Chlorine

x

h

Reaction zone co

Non-reactive contaminant slug

Fig. 5 – Schematic diagrams for a contaminant ‘‘slug’’ introduced into a drinking water pipe: (A) a contaminant ‘‘slug’’ in transport between the two paired sensor stations, (B) the CDR process of a reactive contaminant ‘‘slug’’ reacting with chlorine in the bulk phase and (C) non-reactive contaminant or tracer dispersion in pipe flow. Dashed line—initial chlorine level Co. Other symbols are defined in the nomenclature.

respectively. Scaling factor (ns) allows flow modeling for a pipe length of interest in units of pipe radius; ns ¼ 2 for the characteristic pipe length. The first term on the right-hand side of Eq. (4) measures radial diffusion in transverse. Its constant is the ratio of characteristic time between convec-

tion and radial diffusion (Chakraborty and Balakotaiah, 2002): tc ns Deff ¼   . u¯ x ro tD

(6)

ARTICLE IN PRESS 1406

WAT E R R E S E A R C H

42 (2008) 1397– 1412

This constant can be sufficiently small to neglect the radial diffusion term in Eq. (4) for turbulent pipe flows in a paired configuration of online sensor stations (Fig. 5a) that Yang et al. (2007) described in adaptive contaminant monitoring. Using the dimensionless transverse-averaged flow velocity u¯ x for u(x, r, y), a dimensionless bulk decay constant (kb), and a ^ decay rate term Wðc; tÞ for wall demand, Eq. (4) becomes a modified one-dimensional Danckwerts equation for chlorine in water pipes: ! qc qc 1 q2 c ^ þ u¯ x ¼ 2 tÞ, (7)  kb c  Wðc; qt qx ns PeD qx2 kb ¼

ðns ro Þ  kb . u¯ x

The governing equation (7) is valid to describe mass transport in long and narrow pipes for which longitudinal dispersion is fully developed (Taylor, 1953, 1954). This condition is satisfied in the aldicarb experiments with the setup of two sensor monitoring stations (Fig. 5a). When the axial Pe´clet number ðPeL ¼ n2s PeD Þ is sufficiently large, longitudinal dispersion becomes negligible and subsequent simplifications yield the governing equations of Rossman et al. (1994) for the EPANet model (Rossman, 2002). The condition prevails in a distribution network except for laminar and stagnant flow regions such as distribution network dead ends. Furthermore, Clark and Haught (2005) evaluated the wall demand in Eq. (7) and compared the wall reaction limited model of Rossman et al. (1994) with the mass transfer-limited wall reaction kinetics; however, because of the condition under which these experiments were conducted, wall reaction was not considered for the fiberglass-lined pipe interior.

3.1.1.

Physical model

Fig. 5b shows a reactive contaminant ‘‘slug’’ introduced instantaneously into a flowing straight water pipe. Chlorine transport in this situation can be described using the modified one-dimensional Danckwerts equation. The instantaneous contaminant introduction would simulate a line break, back siphonage and even intentional sabotage. Contaminant back siphoning from pipe breaks, for example, usually lasts for o10 s before the negative transient pressure dissipates (AWWA, 2004). Following the introduction, a contaminant ‘‘slug’’ moves along the bulk phase, propagates by molecular diffusion and longitudinal dispersion, and reacts with chlorine disinfectants in the water. Simultaneously chlorine-contaminant fast reactions can rapidly reduce and even deplete chlorine residuals inside of the ‘‘slug’’, producing a chlorine concentration gradient from bulk phase toward the contamination. The resultant chlorine mass flux is in opposite direction to that of reactive contaminants at the contaminant–water interface (Fig. 5b).

3.1.2.

Chlorine decay inside of a contaminant ‘‘slug’’

In general, the instantaneous introduction of a reactive contaminant ‘‘slug’’ into chlorinated drinking water can lead to changes in both chlorine bulk demand and wall demand. The new bulk demand in the contaminant ‘‘slug’’ after assumed instant complete mixing is described by the

generalized reactions: kf

a½A þ q1 ½Cl ! p1 ½P1 , ks

b½B þ q2 ½Cl ! p2 ½P2  in which [A] and [B] are fast- and slow-reacting contaminants that react with chlorine [Cl] to produce the reaction products [P1] and [P2] at the rate kf and ks , respectively, which yields two second-order equations (Clark and Sivaganesan, 2002). ^ Assuming negligible wall demand or Wðc; tÞ  0 for fiberglasslined pipe, chlorine residual decay in the ‘‘slug’’ is given by     ð1  R1 Þ ð1  R2 Þ þ ð1  f Þ c ¼ fm 0 m 1  R1 eð1R1 Þkf t 1  R2 eð1R2 Þðks þkb Þt ^  Wðc; tÞ,     ð1  R1 Þ ð1  R2 Þ c ffi fm Þ þ ð1  f . ð8Þ m 1  R2 eð1R2 Þks t 1  R1 eð1R1 Þkf t In Eq. (8), parameters kb, k0s and kf are dimensionless kinetic constants for the residual pre-mixing bulk demand, slow- and fast-reacting contaminants, respectively; ks ¼ kb þ k0s is the overall slow chlorine reaction rate in the ‘‘slug’’; fm is molar fraction of the chlorine consumption by the fast-reacting contaminants, while (1fm) denotes all other chlorine consumptions by slow reactions. R1 and R2 are empirical dimensionless parameters as defined in Clark et al. (2001).

3.1.3.

Chlorine reaction at contaminant–water interfaces

The chlorine residual concentration profile at the contaminant–water interface determines the extent and geometry of chlorine residual loss recorded at the sensor stations in Fig. 5b. Taking u¯ x ¼ 1 for turbulent pipe flows (Lu et al., 1993) and using transverse-averaged concentration ðc^ Þ, Eq. (7) becomes: ! qc^ qc^ 1 q2 c^ ^ c^ ; tÞ, þ ¼ 2  kb c^  Wð (9) qt qx ns PeD qx2 with boundary and initial conditions: c^ ð1;tÞ ¼ 1 at tX0   qc^ ¼ 0 at x ¼ 0, qx t   ð1  R1 Þ ð1  R2 Þ þ ð1  f Þ c^ ffi f m m 1  R1 eð1R2 Þks t 1  R1 eð1R1 Þkf t     h h pxp u¯ x t þ at u¯ x t  2 2

ð10Þ

and c^ o ¼



Gðc^ x¼x1 Þ 1

at t ¼ 0.

(11)

Gðc^ x¼x1 Þ is a chlorine concentration input function at sensor station #1 (Fig. 5a); Gðc^ x¼x1 Þ ¼ 1 for the contaminant introduced at this location. Boundary conditions in the contaminant ‘‘slug’’ ðu¯ x t  h=2Þpxpðu¯ x t þ h=2Þ are given in Eqs. (10) and (11); the dimensionless ‘‘slug’’ length ðh ¼ h =ro Þ changes with time in response to the contaminant front propagation or contraction. In this study, the model was solved using an explicit numerical scheme of forward time and center space with moving boundary conditions. Similar numerical computation techniques can be found in program BIO-1D for contaminant transport simulation in groundwater.

ARTICLE IN PRESS WAT E R R E S E A R C H

1407

42 (2008) 1397 – 1412

Table 3 – Modeling parameters for non-reactive sodium fluoride tracer and reactive aldicarb in pipe flow experiments Tracera

Parameters

Aldicarb 0.2 mg/L

1.1 mg/L

2.2 mg/L

112.2 15.24 0 69.1 6.98 0.0423 9020

335.4 7.61 0 83.2 30.5 0.0276 25,000

335.4 7.61 0 83.2 30.5 0.0276 25,000

335.4 7.61 0 83.2 30.5 0.0276 25,000

Dispersion and reactive properties Molecular diffusivity, D (cm2/s) Effective chlorine dispersion coefficient, Deff (cm2/s) Fast-reaction decay constantc, kf* (1/s) Slow-reaction decay constant, ks* (1/s) Fast- and slow-reaction molar fraction, fm Dimensionless fast-reaction parameter, R1 Dimensionless slow-reaction parameter, R2

1.44 105 33 0 0 0 – –

1.44 105 69 0.05 5 108 0.08 0.99 0.2

1.44 105 69 0.05 5 108 0.48 0.99 0.2

1.44 105 69 0.05 5 108 0.92 0.95 0.2

Parameters for high decay rate comparisons Fast-reaction decay constant, kf* (1/s) Slow-reaction decay constant, ks* (1/s) Effective chlorine dispersion coefficient, Deff (cm2/s)

– – –

– – –

– – –

0.3 3 106 1400

Physical and hydraulic properties Pipe length, L* (m) Pipe diameter, 2r* (cm) Pipe roughnessb Flow rate, Qp*(L/min) Velocity, u% x*(cm/s) Darcy–Weisbach friction factor Reynolds number, Re

a b c

Same parameters used in numerical modeling for all four tracer tests. Pipe roughness ¼ 0 for smooth PVC pipe and fiberglass-lined pipe used in the experiments. kf* ¼ 0.05 1/s for aldicarb at initial free chlorine concentration 0.90 mg/L. ks* is assumed to be 106 smaller than kf*.

3.2. pipes

Non-reactive contaminant ‘‘slug’’ transport in water

In contrast to the previous analysis for fast-reacting contaminants, conservative contaminants have no reactivity with chlorine. Longitudinal dispersion is the controlling factor over their mass transport. Between these two extremes are contaminants exhibiting slow-reacting kinetics, for which modifications to the model in Fig. 5b and the governing equations are required. A familiar example of non-reactive solutes is the conservative tracer sodium fluoride (Fig. 5c) often used in pipe flow testing (e.g., Taylor, 1953, 1954; Aries, ^ tÞ ¼ 0, 1956; Ekambara and Joshi, 2004). When kb ¼ 0 and Wðc; the CDR governing equation in Eq. (9) is simplified to qc^ qc^ 1 þ ¼ qt qx n2s PeD

! q2 c^ , qx2

(12)

which is identical to that of Levenspiel (1972) for tracer transport in pipe flows. When the contaminant or a tracer moves in the form of ‘‘slugs’’ (Fig. 5c), the exit age (E*) can be solved explicitly from Eqs. (13) and (14) of Levenspiel (1972). Li (2000) found these equations conformed to tracer experiment results under a wide range of flow conditions in a long water pipe. ( ) 1 ð1  t Þ2 ; E ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffi exp  4Pe1 L 2 pPe1 L

PeL 4100,

(13)

( ) 1 ð1  t Þ2 E ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi exp   1 ; 4t PeL 2 pt Pe1 L

PeL p100.

(14)

The simplified CDR model was calibrated in comparison with the Levenspiel’s (1972) analytical solution and against the tracer experimental results in Table 1. Using the model parameters in Table 3, both the numerical modeling (Eq. (12)) and the analytical solutions (Eq. (13)) generate E-curves consistent with the tracer experimental results (Fig. 2). The two models are consistent with each other except for the area at the E-curve peak where the numerical model slightly overestimated.

4.

Discussions

4.1.

Chlorine decay in a contaminant ‘‘slug’’

Mason et al. (1990) outlined the reactive pathways and proposed the reaction kinetics for aldicarb oxidation in chlorinated drinking water. In chlorine oxidation, the aldicarb carbamate structure and its thio function group are transformed to yield the intermittent products aldicarb-sulphone and aldicarb-sulphoxide. The initial fast oxidation is followed by slow reaction of more recalcitrant intermittent compounds at a rate several orders of magnitude smaller. Extrapolated from the experimental results of Mason et al. (1990), the aldicarb-chlorine fast reaction should have followed a kinetic

ARTICLE IN PRESS 1408

WAT E R R E S E A R C H

42 (2008) 1397– 1412

order of 1.78 at pH ¼ 7.0 for the initial chlorine 0.90 mg/L in the aldicarb experiments of this study.

c (clf)

Quasi-steady-state chlorine decay

As shown in Figs. 3 and 4, chlorine residual loss within the contaminant ‘‘slug’’ displayed no significant change between the two sensor stations in a 311-m flow distance. It follows that residual loss in the ‘‘slug’’ must have reached a quasisteady state before the water reached the 22.4-m sensor station at a 1.4-min hydraulic retention time. The quasi-steady-state condition is further evident in comparing the aldicarb pipe flow experiments with the aldicarb bench-scale studies of Haught et al. (2005). In the bench-scale study, concentrated aldicarb solution was added to a glass beaker reactor and mixed well with tap water. Initial aldicarb concentration was 0.2–2.2 mg/L after mixing (Haught et al., 2005). Measured chlorine residual concentrations normalized against the initial concentration at t* ¼ 0 were used to calculate the residual loss produced by the aldicarbinduced bulk demand; wall demand was negligible for the pre-cleaned glass beaker. The results showed a strong linear relationship between chlorine residual loss and initial aldicarb concentration (Fig. 6). In the pipe flow experiments, the chlorine residual loss within the aldicarb ‘‘slug’’ exhibits the same linear relationship as in the bench-scale testing (Fig. 6). This agreement strongly suggests that in the pipe flow experiments chlorine residual loss within the ‘‘slug’’ is primarily the result of bulk demand in aldicarb oxidation, and also reinforces the notion that complete mixing and quasi-steady-state chlorine decay in the ‘‘slug’’ must have occurred during its transport from the injection port to the first sensor station.

4.1.2.

Bench-scale test data

(15)

Fig. 7 shows a comparison between the second-order and first-order kinetics models. Also plotted are chlorine loss (Dc) observed at the 22.4- and 335.4-m sensor stations in the aldicarb pipe flow experiments. Both chlorine decay models approximated the observed residual loss using modeling parameters in Table 3, and also predicted the same quasisteady-state chlorine decay in the extended time. They differ in the period immediately following introduction of the aldicarb ‘‘slug’’ (Fig. 7). Furthermore, the second-order kinetic model in Eq. (10) provides additional information on the nature of reactive contaminants in the ‘‘slug’’. The secondorder model parameters (fm, R1 and R2) are related to water quality parameters such as total organic carbon, UVA, pH, initial chlorine concentration and temperature (Clark et al., 2001). Therefore, the second-order kinetic model was selected to represent the CDR model boundary condition in Eq. (10). It should be noted, however, that the second-order competitive

0.6

y=0.413x-0.014, R2=0.999

0.4

0.2

0.0 0.0

0.5

1.0

1.5

2.0

2.5

Co* (mg/L)

Fig. 6 – Experimentally determined chlorine residual loss within the aldicarb ‘‘slug’’ is consistent with the linear relationship defined in the bench-scale bulk decay experiments reported by Haught et al. (2005). Error bars indicate one standard deviation associated with the geometric mean shown in cross.

1.2 Clark & Sivaganesan (2002) Parallel first order

1.0 0.2 mg/L aldicarb

0.8 0.6

1.1 mg/L aldicarb

0.4

Chlorine decay kinetics and boundary conditions

The CDR model boundary condition in Eq. (10) relies on the competitive second-order chlorine decay kinetics of Clark and Sivaganesan (2002) under assumed negligible wall demand. Gang et al. (2003) and Warton et al. (2006) described a firstorder kinetic model for drinking water, which may be applicable to approximate the residual loss in the multi-step successive oxidation of aldicarb: c^ ffi f m ekf t þ ð1  f m Þeðks þkb Þt .

244.4-m station 335.4-m station

0.8

c (Clf)

4.1.1.

1.0

0.2

2.2 mg/L aldicarb

0.0 0

5 15

20

t* (min)

Fig. 7 – Modeling results for free chlorine residual loss within the aldicarb ‘‘slug’’ using the competitive secondorder chlorine decay kinetics in Eq. (8) and the parallel firstorder kinetics in Eq. (15). Measured free chlorine residual loss (Dc) at the 24.4-m (t* ¼ 1.8 min) and 335.4-m (t* ¼ 18.3 min) sensor stations is consistent with the modeling results at 0.2, 1.1 and 2.2 mg/L initial aldicarb concentrations. Modeling parameters are given in Table 3.

model examined in this study for aldicarb-chlorine reaction may not work as well for some other contaminants. This variance should be considered in model definition.

4.2.

Residual loss curve geometry and control variables

Fig. 8 shows experimental data for the 24.4-m sensor station (x ¼ 320) and the CDR modeling results at initial aldicarb concentrations of 0.2, 1.1 and 2.2 mg/L. Numerical simulation is based on Eqs. (9)–(11) and the model parameters in Table 3.

ARTICLE IN PRESS WAT E R R E S E A R C H

1.02

42 (2008) 1397 – 1412

*

fm=0.08, kf =0.05

Δt=1420

1.00

0.96

Δc=0.09

c (Clf)

0.98 x=640 x=320

x=0

0.94 0.92 0.90 0

1.1

500

1000

1500

2000

2500

2000

2500

2000

2500

*

fm=0.48, kf =0.05

1.0

Δt=1420 Δc=0.47

c (Clf)

0.9 x=640

0.8

x=320

x=0

0.7 0.6 0.5 0.4 0

1.2

500

1000

*

fm=0.92, kf =0.05

Δt=1420

1.0

x=0

0.6

x=320

Δc=0.92

Deff=1400 cm2/sec

0.8 c (Clf)

1500

x=640

0.4 0.2

*

fm=0.92, kf =0.3

0.0 0

500

1000

1500

t = t*u*/(2r*)

Fig. 8 – The CRD-model simulated and the experimentally determined residual loss of free chlorine at initial aldicarb concentration: (A) 0.2, (B) 1.1 and (C) 2.2 mg/L. The model simulation shows the response of residual loss curve geometry to three variables: fm ¼ 0.09, 0.48 and 0.92, kf ¼ 0:05and 0.3 s1, and Deff ¼ 33 and 1400 cm2/s. The residual loss curve width (Dt) and depth (Dc) are marked for the 24.4-m sensor station at x ¼ 320. When Deff increases from 33 to 1400 cm2/s at the 2.2 mg/L aldicarb concentration, the geometry of chlorine residual loss curve is changed slightly.

relations between the residual loss curve geometry (width, depth and slope), contaminant reaction rate constant and chlorine longitudinal dispersivity. The influence of contaminant reactivity on the residual loss curve geometry is assessed by using kf ¼ 0:05 and 0:3 s1 in the CDR model (Table 3). The experimental results by Mason et al. (1990) revealed a logarithmic relationship between observed reaction rate of aldicarb oxidation and the initial chlorine concentration. Extrapolation from their results yielded a pseudo-first-order constant of 0.02–0.18 s1 for the 0.2–2.2 mg/L initial aldicarb concentrations of this study. In another approach, model fitting using Eq. (8) to the experimental residual loss data yielded kf ¼ 0:05 1 that falls into the range 0.02–0.18 s1 based on Mason et al. (1990). Thus kf ¼ 0:05 1 was selected for the CDR model simulations producing the residual loss curves consistent with experimental results for all initial aldicarb concentrations (Fig. 8a–c). When the constant increases to 0.3 s1, model-calculated residual loss at a quasi-steady state is rapidly established within to50 or t* ¼ 0.5 min and the loss curve slope is steepened (Fig. 8c). In this comparison, the larger value kf ¼ 0:3 1 is the first-order decay constant that Brosillon et al. (2006) reported for glyphosate, a fast-reacting herbicide of the same chemical class. For the fast-reacting contaminants— aldicarb and glyphosate, their chlorine reactivity controls the degree of chlorine decay within the ‘‘slug’’ which serves as CDR model boundary condition in Eq. (10) and therefore can significantly affect the geometry of a residual loss curve. This effect remains significant, and the U-shaped chlorine residual loss curve can persist in a straight pipe when mechanical mixing is absent and the rate of chlorine-contaminant reaction inside of the ‘‘slug’’ exceeds longitudinal dispersion flux of the solute (chlorine or aldicarb), namely kf c^ bð1=n2s PeD Þ ð@2 c^ [email protected] Þ in Eq. (9). Compared to reactivity, the longitudinal dispersion has relatively small effect on the chlorine residual loss curve geometry. The modeled curve geometry has small change when effective dispersivity increases from 33 to 1400 cm2/s (Fig. 8e). However, the effect of longitudinal dispersion becomes important for less reactive contaminants; the most extreme is the non-reactive contaminants or tracers. For the latter, the dispersion curve geometry is directly related to the effective dispersion coefficient of the solute and to the pipe flow Pe´clet number (Aries, 1956; Levenspiel, 1972; Ekambara and Joshi, 2004). Evidently in this study, substantial longitudinal dispersion was observed in the sodium fluoride tracer experiments (Fig. 2). Even in the aldicarb experiments, chloride as the reaction product also displayed an asymmetric dispersion profile spreading beyond the zone of chlorine residual loss (Fig. 3).

4.3. The model predictions agree closely with the experimental results at all tested aldicarb concentration levels (Fig. 8a–c). Calculated half-width of the residual loss curve is approximately Dt ¼ 1420 or 11.8 min; the full width is 23.6 min slightly larger than the experimentally determined width of 19.5–22.0 min (Table 2). As the modeling replicates experiment data for both reactive aldicarb and non-reactive tracer (see Fig. 2), use of the CDR model is extended to investigate the

1409

Implications for real-time contaminant detection

The non-reactive sodium fluoride in the tracer experiments showed a large concentration decrease over 97 times in a 112.2-m turbulent flow distance. In general, the longitudinal dispersion can lead to a large reduction in concentration making it difficult to measure the target compounds in realtime field applications. This often results in false negative detections (US EPA, 2005; Klise and McKenna, 2006). However,

ARTICLE IN PRESS 1410

WAT E R R E S E A R C H

42 (2008) 1397– 1412

1.2 1.0

c (Clf)

0.8 0.6 0.4 0.2 0.0 0.0005 0.0000

Δc/Δt

-0.0005 -0.0010 Model, fm=0.08 Model, fm=0.48

-0.0015

Model, fm=0.92 -0.0020

Test data, 0.2 mg/L aldicarb Test data, 1.1 mg/L aldicarb Test data, 2.2 mg/L aldicarb

-0.0025 -0.0030 4e-5

Δ2c/Δt2

2e-5

0

-2e-5

-4e-5

-6e-5 0

500

1000

1500

2000

t = t*u*/(2r*)

Fig. 9 – Experimental and CDR modeling results showing the use of first- and second-order concentration time derivatives to mark the front edge of an incoming contaminant ‘‘slug’’. These two variables can better detect the edge than the absolute concentration measurements, even at the low aldicarb concentration 0.2 mg/L. CDR model parameters are given in Table 3.

many studies (e.g., Donohue and Lipscomb, 2002) have pointed to the fact that health risk from contaminant exposure depends on both the concentration and mass intake. In the sodium fluoride experiments, Li (2000) reported that the entire tracer mass was conserved in less than 7–13 min of the pipe flow or 484–900 L of water column (Table 1). If the chemical was highly toxic, risk from the small volume of contaminated water could not be underestimated.

The chlorine residual loss curve and its geometry potentially offer a more reliable detection marker for a reactive contaminant ‘‘slug’’ in drinking water. This potential relies on a close relationship between the residual loss, chloride concentration increase, and the contaminants reactivity as examined in the aldicarb experiments and corresponding CDR modeling. The concentration changes can be detected using real-time chlorine and chloride sensors, and the

ARTICLE IN PRESS WAT E R R E S E A R C H

42 (2008) 1397 – 1412

relationships can be quantified in an adaptive monitoring scheme of Yang et al. (2007) that requires an accurate characterization of the chlorine loss curve geometry, particularly at the leading edge of a contaminant ‘‘slug’’. For this purpose, first- and second-order concentration derivatives ðDc=Dt; D2 c=Dt2 Þ are better monitoring parameters than the absolute chlorine concentrations in detecting the aldicarb ‘‘slug’’, even at the low contaminant concentration (Fig. 9).

5.

Conclusions

Experimental testing and numerical simulation using a CDR model have quantified the transport of non-reactive sodium fluoride and reactive pesticide aldicarb in the pilot-scale drinking water pipes under turbulent flow conditions. Major conclusions are:

When instantaneously introduced into turbulent flow of a







straight pipe, reactive contaminants can move in a discrete ‘‘slug’’ bounded by a reactive interface where the contaminants are oxidized by chlorine in the bulk phase. Longitudinal dispersion forces the ‘‘slug’’ propagating into the uncontaminated bulk phase, while the propagation is counteracted and inhibited by aldicarb-chlorine fast reactions. This CDR process is considered to have produced the characteristic U-shaped curves in the chlorine residual concentration as consistently observed in the aldicarb pipe flow experiments. In examination of the CDR process, a modified onedimensional Danckwerts model reproduced the observed chlorine profiles across the reactive aldicarb ‘‘slug’’. Quantitative analysis shows that the geometry (width, depth and slope) of a residual loss curve is a function of the fast-reaction kinetic constant (kf ), molar fraction of fast and slow reactions (fm) and the axial Pe´clet number (PeL). In the aldicarb experiment using a fiberglass-lined water pipe, the quasi-steady-state condition in the residual loss was established within the ‘‘slug’’ in less than 1.3 min. The chlorine decay in the ‘‘slug’’, which serves as the boundary condition in the CDR model, can be described by the competitive second-order kinetic model of Clark and Sivaganesan (2002) at no or negligible chlorine wall demand. For the fastreacting aldicarb, rapidly developed chlorine decay controlled the geometry of observed residual loss curve. This generic relationship can be potentially explored to identifying the chemical and kinetic properties of a reactive contaminant. In the sodium fluoride tracer experiments, determined exit ages are consistent with E-curves generated by both the Levenspiel’s (1972) analytical model and the numerical CDR model in Eq. (13) when wall demand and bulk demand are set to zero. Both the experimental and modeling results show that a significant degree of dispersion is responsible for 497 times of tracer concentration decrease in the 112.2-m flow distance, yielding

1411

potential difficulties in direct measurement of the target concentrations.

Acknowledgment The authors wish to acknowledge contributions of the following individuals: Dr. Jeff Szebo of US EPA and Mr. Greg Meiners of Shaw Environmental for aldicarb experiments and manuscript review; Dr. Tomas Speth, Dr. Michael Elovitz, Mr. John Hall, Dr. Lewis Rossman, and Mr. Donald Brown, all of US EPA, for their reviews and constructive comments. The authors acknowledge the review and comments from two anonymous reviewers. This work is financially supported in part by US EPA National Homeland Security Research Center. Conclusions and opinions are those of the authors, and do not necessarily represent the position of the US EPA and USDA. Lastly this paper is in memory of late Prof. Sun Dianqing, a senior academician at the Chinese Academy of Science to whom the leading author is indebted for his early academic guidance. R E F E R E N C E S

Aries, R., 1956. On the dispersion of a solute in a fluid flowing through a tube. Proc. R. Soc Lond. 235 (1200), 67–77. AWWA, 2004. Verification and Control of Pressure Transients and Intrusion in Distribution Systems. AWWA Res Foundation, Denver, USA, 126pp. Baldyga, J., Bourne, J.R., Hern, S.J., 1996. Interaction between chemical reactions and mixing on various scales. Chem. Eng. Sci. 52 (4), 457–466. Biswas, P., Lu, C., Clark, R.M., 1993. A model for chlorine concentration decay in drinking water distribution pipes. Water Res. 27 (12), 1715–1724. Boulos, P.F., Altman, T., Jarrige, P-A., Collevati, F., 1994. An eventdriven method for modeling contaminant propagation in water networks. Appl. Math. Model. 18, 84–92. Boyd, G.R., Wang, H., Brittonm, M.D., Howie, D.C., Wood, D.J., Funk, J.E., Friedman, M.J., 2004. Intrusion within a simulated water distribution system due to hydraulic transients. I: Description of test rig and chemical tracer method. J. Environ. Eng. 130 (7), 774–777. Brosillon, S., Wolbert, D., Lemasle, M., Roche, P., Mehrsheikh, A., 2006. Chlorination kinetics of glyphosate and its by-products: modeling approach. Water Res. 40, 2113–2124. Burrows, W.D., Renner, S.E., 1999. Biological warfare agents as threats to potable water. Environ. Health Perspect. 107 (12), 975–984. Caputo, A.C., Palagagge, P.M., 2002. An inverse approach for piping networks monitoring. In: Fourth International Conference on Inverse Problems in Engineering, Brazil. Chakraborty, S., Balakotaiah, V., 2002. Low-dimensional models for describing mixing effects in laminar flow tubular reactors. Chem. Eng. Sci. 57, 2545–2564. Clark, R.M., Haught, R.C., 2005. Characterizing pipe wall demand: implications fro water quality modeling. J. Water Res. Plann. Manage. 1, 208–217. Clark, R.M., Sivaganesan, M., 2002. Predicting chlorine residuals in drinking water: second order model. J. Water Res. Plann. Manage. 128 (2), 152–161. Clark, R.M., Rossman, L.A., Sivaganesan, M., Scheneck, K., 2001. Modeling chlorine decay and the formation of disinfection

ARTICLE IN PRESS 1412

WAT E R R E S E A R C H

42 (2008) 1397– 1412

by-products (DBPs) in drinking water. In: Clark, R.M., Boutin, B.K. (Eds.), Controlling Disinfection By-Products and Microbial Contaminants in Drinking Water, EPA/600/R-01/110, US EPA, Washington, DC. Cussler, E.L., 1984. Diffusion, Mass Transfer in Fluid Systems. Cambridge University Press, New York, 525pp. Donohue, J.M., Lipscomb, J.C., 2002. Health advisory values for drinking water contaminants and the methodology for determining acute exposure values. Sci. Tot. Environ. 288, 43–49. Ekambara, K., Joshi, J.B., 2004. Axial mixing in laminar pipe flows. Chem. Eng. Sci. 59, 3929–3944. Forney, L.J., Nafia, N., 2000. Eddy contact model: CDF simulations of liquid reactions in nearly homogeneous turbulence. Chem. Eng. Sci. 55, 6049–6058. Gang, D.C., Clevenger, T.E., Banerji, S.K., 2003. Modeling chlorine decay in surface water. J. Envoron. Inform. 1 (1), 21–27. Government Accounting Office, 2005. Protection of chemical and water infrastructure: Federal requirements, actions for selected facilities, and remaining challenges. GAO-05-327, Washington, DC Hamilton, P.D., Gale, P., Pollard, S.J.T., 2006. A commentary on recent water safety initiatives in the context of water utility risk management. Environ. Int. 32 (8), 958–966. Harding, B.L., Walski, T.M., 2000. Long-time-series simulation of water quality in distribution system. J. Water Res. Plann. Manage. 126 (4), 199–209. Haught, R., Hall, J., Rahman, M., Richardson-Coy, R., Piao, H., 2005. The Bench-Scale Minimum Threshold Experiment. WQTC, Toronto, Canada. Heinz, S., Roekaerts, D., 2001. Reynolds number effects on mixing and reaction in a turbulent pipe flow. Chem. Eng. Sci. 56, 3197–3210. Hladik, M.L., Roberts, A.L., Bouwer, E.J., 2005. Removal of neutral chloroacetamide herbicide degradates during simulated unit processes for drinking water treatment. Water Res. 39, 5033–5044. Islam, M.R., Chaudhry, M.H., 1998. Modeling of constituent transport in unsteady flows in pipe networks. J. Hydraulic Eng. 124 (11), 1115–1124. Jain, S., McLean, C.R., 2006. An integrating framework for modeling and simulation for incident management. J. Homeland Security Emerg. Manage. 3 (1), 1–24. Khordagui, H.K., 1995. Fate and control of nerve chemical warfare agents in the desalination industry of the Arabian–Persian gulf. Envoron. Int. 21 (4), 363–379. Kirmeyer, G.J., Friedman, M., Martel, K.D., Noran, P.F., Smith, D., 2001. Maintaining distribution system water quality. J. AWWA 93 (7), 62–73. Klise, K.A., McKenna, S.A., 2006. Water quality change detection: multivariate algorithms. In: Saito, T.T., Lehrfeld, D.(Eds.), Proc. SPIE, vol. 6203, pp. J1–J9. LeChevallier, M.W., Gullick, R.W., Karim, M.R., Friedman, M., Kunk, J.E., 2003. The potential for health risks from intrusion of contaminants into the distribution system from pressure transients. J. Water Health 1 (1), 3–14. Leonenko, Y., Vonogradov, O., Braverman, L., 2005. Using discrete simulation of microprocesses in dispersed phases in analyzing multiphase flows with reactions. Chem. Eng. Sci. 60, 1565–1574. Levenspiel, O., 1972. Chemical Reaction Engineering. Wiley, New York, 578pp. Levien, K., Levenspiel, O., 1999. Optimal product distribution from laminar flow reactors: Newtonian and other power-law fluids. Chem. Eng. Sci. 54, 2453–2458.

Li, S.Y., 2000. Chlorine transport and decay studies in a pilot scale drinking water distribution dead end pipe system. Ph.D. Dissertation, University of Cincinnati, Cincinnati, OH. Lu, C., Biswas, P., Clark, R.M., 1993. Modeling of breakpoint reaction in drinking water distribution pipes. Environ. Int. 19, 543–560. Mason, Y., Coshen, E., Rav-Acha, C., 1990. Carbamate insecticides: removal from water by chlorination and ozonation. Water Res. 24 (1), 11–21. Powell, J.C., Hallam, N.B., West, J.R., Forster, C.F., Simms, J., 2000. Factors which control bulk chlorine decay rates. Water Res. 34 (1), 117–126. Rossman, L.A., 2002. EPANET Version 2 User’s Manual. US EPA, Cincinnati. Rossman, L.A., Clark, R.M., Grayman, W.M., 1994. Modeling chlorine residuals in drinking-water distribution systems. J. Environ. Eng. 120 (4), 803–820. Stackelberg, P.E., Furlong, E.T., Meyer, M.T., Zaugg, S.D., Henderson, A.K., Reissman, D.B., 2004. Persistence of pharmaceutical compounds and other organic wastewater contaminants in a conventional drinking-water-treatment plant. Sci. Tot. Environ. 329, 99–113. Szabo, J.G., Hall, J.S., 2006. Water Quality Sensor Responses to Contamination in a Single Pass Water Distribution System Simulator. Water Information Sharing and Analysis Center (WaterISAC), Washington, DC. Taylor, G., 1953. Dispersion of soluble matter in solvent flowing slowly through a tube. Proc. R. Soc. Lond. 219 (1137), 186–203. Taylor, G., 1954. Conditions under which dispersion of a solute in a stream of solvent can be used to measure molecular diffusion. Proc. R. Soc. Lond. 225 (1163), 473–477. Tzatchkov, V.G., Adama, A.A., Arregiun, F.I., 2002. Advection–dispersion–reaction modeling in water distribution networks. J. Water Res. Plann. Manage. 128 (5), 334–342. Uber, J., Murray, R., Janke, R., 2004. Use of system analysis to assess and minimize water security risks. J. Contemp. Water Res. Ed. 129, 34–40. US EPA, 2002. Potential contamination due to cross-connections and backflow and the associated health risks: an issue paper. NTIS, Springfield, VA, 46pp. US EPA, 2005. Technologies and Techniques for Early Warning Systems to Monitor and Evaluate Drinking Water Quality: State-of-the-Art Review. ORD, NHSRC, Washington, DC. US EPA, 2006. Water Distribution System Analysis: Field Studies, Modeling and Management. US EPA, National Risk Management Research Laboratory, Cincinnati, OH. Wang, Y., Postlethwaite, J., 1997. The application of low Reynolds number k–e turbulence model to corrosion modeling in the mass transfer entrance region. Corrosion Sci. 39 (7), 1265–1283. Warton, B., Heitz, A., Joll, C., Kagi, R., 2006. A new method for calculation of the chlorine demand of natural and treated waters. Water Res. 40, 2877–2884. Whelton, A.J., Jensen, J.L., Richards, T.E., Valdivia, R.M., 2003. The cyanic threat to potable water. Proceedings of AWWA ACE 2004, Anaheim, CA. World Health Organization, 1996. Guidelines for Drinking-Water Quality, vol. 2, second ed., Health Criteria and Other Supporting Information. WHO, Mastercom/Wiener Verlag. Yang, Y.J., Haught, R.C., Hall, J., Szabo, J., Clark, R.M., Meiners, G., 2007. Adaptive water sensor signal processing: experimental results and implications for online contaminant warning systems. EWRI World Environmental and Water Resources Congress, Tampa, FL.