You are on page 1of 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/280742702

Assessment of hydraulic flocculation processes using CFD

Article  in  Journal - American Water Works Association · September 2011


DOI: 10.1002/j.1551-8833.2011.tb11567.x

CITATIONS READS
18 966

4 authors, including:

Yamuna Vadasarukkai Graham Gagnon


Dalhousie University Dalhousie University
7 PUBLICATIONS   27 CITATIONS    215 PUBLICATIONS   2,418 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Phosphorus Removal from Wastewater using Alum Sludge View project

Canadian Water Network project View project

All content following this page was uploaded by Graham Gagnon on 17 August 2015.

The user has requested enhancement of the downloaded file.


The performance of flocculation tanks either in full- or pilot-scale systems cannot be fully

quantified using average velocity gradient values (G values) alone. The computational fluid

dynamics (CFD) process was used to investigate the turbulent flow characteristics of a

three-stage hydraulic flocculation facility at the J.D. Kline Water Supply Plant in Halifax,

Nova Scotia. Calculations of the local velocity gradient were achieved using predicted

energy dissipation rates. CFD analysis showed unbalanced mixing conditions in the

hydraulic flocculation tanks resulting from short-circuiting and from recirculation zones.

Inconsistent mixing energy, interrupted with severe spikes in the localized G values at the

weir columns, led to little or no mixing energy in most of the regions in the flocculation

tanks. Practical outcomes of operating at such varying mixing gradients can be detrimental

to the flocculation process and overall organic removal in the treatment plant.

A case study of the three-stage tapered


hydraulic flocculation tank at the
J.D. Kline Water Supply Plant in Halifax,
Nova Scotia was initiated using
computational fluid dynamics to evaluate
the mixing conditions in the tanks.

Assessment of hydraulic
flocculation processes
using CFD

C
oagulation and flocculation are key processes in optimizing natu-
ral organic matter removal and subsequent mitigation of disin-
fection by-products (DBPs) formation (Jarvis et al, 2008). Proper
use of coagulant type and dosage, optimal pH and alkalinity con-
ditions, and mixing arrangements are imperative for effective
destabilization of particles through the coagulation process. Subsequent
YAMUNA S. VADASARUKKAI, enhancement of contact between the coagulant, destabilized particles, and
primary particles in water occurs predominantly in the flocculation process
GRAHAM A. GAGNON,
through gentle mixing for 20 to 45 min (Crittenden & MWH, 2005). The
D. REID CAMPBELL, flocculation process promotes the agglomeration of particles—called flocs—to
form irregularly shaped aggregates that can be filtered or settled. These aggre-
AND SARAH C. CLARK
gates are then removed in downstream separation processes (e.g., sedimenta-
tion, filtration, dissolved air flotation).
Floc characteristics (i.e., size and structure) formed during coagulation–floc-
culation processes govern the performance of downstream processes for
improved particle and microbial removal (Jarvis et al, 2008). The flow condi-
tions in flocculation tanks are driven by localized fluid turbulence, and the
resulting flocs are subjected to varying shear rates. The resulting nonuniform
energy distribution (i.e., in the x, y, and z directions) within the tank relates

66 NOVEMBER 2011 | JOURNAL AWWA • 103:11 | PEER-REVIEWED | VADASARUKKAI ET AL

2011 © American Water Works Association


The J.D. Kline Water
Supply Plant (bottom) at
Pockwock Lake (top),
Halifax, Nova Scotia, is
located in a protected
watershed. The lake
water has a low pH, a low
alkalinity of calcium
carbonate, and a low
turbidity. Source water
is pumped into the
treatment facility at a
constant flow rate;
thereafter, it flows into
subsequent treatment
processes by gravity.

to the rate of par- of local G-value distribution (Bridgeman et al, 2010;


ticle collision Haarhoff & Van der Walt, 2001). Using computational
(Bridgeman et al, fluid dynamics (CFD)—a numerical modeling ap -
2008; Crittenden proach—Haarhoff and Van der Walt (2001) quantified
& MWH, 2005; high-turbulence zones in terms of estimated local G
Haarhoff & Van value as approximately 5% of the total flocculator vol-
der Walt, 2001). Preformed flocs may adhere or break ume for an around-the-end hydraulic flocculator. Review
up during the flocculation process depending on the of previous studies, however, reveals that the application
prevailing local hydrodynamics and physicochemical of CFD is more confined to hydrodynamics related to
conditions (Bouyer et al, 2005). Previous studies bench-scale mixers (impellers), with fewer examples
(Bridgeman et al, 2010, 2008; Bouyer et al, 2005; Greg- (Bridgeman et al, 2010; Samaras et al, 2010; Essemiani
ory, 2004) have determined that for a given hydrody- et al, 2002; Haarhoff & Van der Walt, 2001) available
namic flow condition, flocs reach a certain threshold with regard to the pilot- and full-scale flocculation pro-
size in their growth phase. Beyond this limit, the relative cess. As a result of wide variation in flocculator designs
fluid velocity that caused smaller particles to aggregate used in water treatment plants (WTPs), a detailed under-
into larger flocs can also result in their breakup. Bridge- standing of the fluid mechanics in these flocculators is
man et al (2008) expressed floc breakage threshold critical for enhancing floc growth and stability at opti-
value in terms of local velocity gradients prevailing in mal retention time.
a jar test apparatus. Table 1 shows examples of WTPs in Canada and the
The hydrodynamic flow condition depends on both the type of flocculator designs used in such conventional and
geometry of the flocculation tank and the impeller speed direct filtration plants. As shown, several direct filtration
and type. These hydrodynamic phenomena are often utilities are equipped with hydraulic flocculation designs
characterized by the average energy dissipation in the for treating high-quality surface water. In this study, a
tank (i.e., the global root mean square velocity gradient CFD modeling strategy was implemented in a direct
[G value]; Bouyer et al, 2005). Analogous to the high- filtration plant to investigate the performance of a
turbulence zones at the tip and edges of mechanical hydraulic flocculation process. The case study in Halifax,
impellers and paddles, hydraulic flocculation designs Nova Scotia, uses four sets of three-stage tapered hydrau-
suffer from zones of high turbulence at the edges of the lic flocculation, having an up-and-down flow arrange-
baffles. These zones could induce floc breakup. Having ment for mixing purposes. The main objective of this
invested time and energy in developing flocs however, it study was to evaluate the mixing condition for each of
is clear that these operating (hydraulic) conditions subse- the three-stage flocculation tanks in terms of the spatial
quently should not cause floc breakup and thus need to distribution of the local G value (Glocal) using CFD. The
be quantified (Bridgeman et al, 2010; 2008). effect of flow condition on the design parameters, i.e.,
The degree of variability in energy dissipation in the volume average G value and the residence time (t) of
hydraulic flocculators can be measured through analysis fluid flow, were investigated.

VADASARUKKAI ET AL | PEER-REVIEWED | 103:11 • JOURNAL AWWA | NOVEMBER 2011 67

2011 © American Water Works Association


BACKGROUND DESIGN CONCEPTS Precise hydraulic measurements are required to account
G value and its calculation. The G value in Eq 1 is used for the actual energy loss at the tank entrance and during
to calculate the intensity of mixing energy input required the passage of water from one tank to another. This addi-
in a flocculation process (Crittenden & MWH, 2005; tional energy loss could result in an increase of about 25%
AWWA/ASCE, 1997). in the estimated G value (Hargrave & Loucks, 1990).
Previous studies (e.g., Bridgeman et al, 2010, 2008;
Ducoste & Clark, 1999; Clark, 1994; Cleasby, 1984) have
G 冪莦
  冪莦

µ
m

P
µV (1) evaluated the validity of the G value to represent the aver-
age velocity gradient in a complex turbulent field. The
in which m is the mean value of the work input per unit authors argued that the distribution of turbulent energy is
of time per unit of volume (J/s), µ is the dynamic viscos- more dependent on the tank geometry and impeller con-
ity of water (kg/m s), P is the average power of mixing figuration than the average G value (i.e., calculated using
input to the entire mixing vessel (J/s), and V is the volume Eqs 1 and 3). Consequently, these studies suggest the need
in the tank. for a detailed investigation of the hydrodynamic flow
The G value in Eq 1 can also be interpreted in terms of characteristics. The approach to evaluating the turbulent
the root mean energy dissipation per unit of mass (Haar- flow field presented in this study would provide adequate
hoff et al, 2001; Cleasby, 1984), as given in Eq 2, information to describe the distribution of local energy
dissipation rates at various locations in the tank.
– With the use of CFD it is possible to compute the local
G 冪莦v (2)
G value (Glocal) at any location in the tank as given in Eq
4 (Bridgeman et al, 2010). The local turbulent dissipation
in which – is the average energy dissipated per unit of mass rate () is calculated using an appropriate turbulence
(m2/s3) and v is the kinematic viscosity of water (m2/s). closure model.
The G value for the hydraulic flocculation process is
largely a function of flow and is measured in terms of the Glocal  冪莦v (4)
total head loss estimated across the tank (Bridgeman et
al, 2010; Haarhoff & Van der Walt, 2001; Haarhoff, Equation 5 is then used to determine the overall power
1998; AWWA/ASCE, 1997) as shown by Eq 3. consumption by numerically integrating the local power
consumption over the entire vessel contents (Bridgeman
gH et al, 2008).
G 冪莦

µ (3)

P   dV (5)
in which  is the density of water, g is acceleration from
gravity, H is total headloss (m), and  is the hydraulic in which dV is the numerical integration over the entire
retention time of the fluid (s). volume.

TABLE 1 Examples of flocculator designs used in conventional and direct-filtration WTPs in Canada

Type of Flocculator
Treatment Capacity
WTP Type ML/d Mechanical Hydraulic Reference

Conventional
Mannheim WTP (Kitchener, Ontario) 72.6 Yes No Arnold, 2008
R.L. Clark WTP (Toronto, Ontario) 415 Yes No City of Toronto, 2010b
R.C. Harris WTP (Toronto, Ontario) 453 No Yes Hargrave et al, 1990
Direct filtration
Island WTP (Toronto, Ontario) 410 No Yes City of Toronto, 2010a
Prescott WTP (East of Brockville, Ontario) 8.2 Yes No Doyle et al, 2002
Point Pleasant WTP (Kingston, Ontario) NA Yes No Kingston Utilities, 2009
Sydenham WTP (Kingston, Ontario) NA No Yes Kingston Utilities, 2009
F.J. Horgan WTP (Toronto, Ontario) 570 Yes Yes Doyle et al, 2002
J.D. Kline WTP (Halifax, Nova Scotia) 95 No Yes

NA—not available, WTP—water treatment plant

68 NOVEMBER 2011 | JOURNAL AWWA • 103:11 | PEER-REVIEWED | VADASARUKKAI ET AL

2011 © American Water Works Association


Residence time distribution and its measurement. Resi- in the TIS model are statistically independent of one
dence time distribution (RTD) curves are used to identify another because of a perfect mixing that exists at any
the imperfect hydraulic behavior of the treatment units. boundary between two tanks (Teefy & Singer, 1990).
These curves are usually expressed in terms of stagnant Mean residence time and variance calculations. The
(dead) spaces and bypassing flow. Applying material bal- mean residence time (tm) of a fluid element in the floc-
ance principles and Laplace transform solution tech- culation tank is calculated from the first moment of the
niques, Martin-Dominguez and colleagues (2005) devel- E(Θ) curve, as given in Eq 8, whereas the magnitude of
2
oped an RTD function for reactors in series with the spread of the distribution ( Θ , in which 2 indicates
stagnation spaces and bypassing over the whole series of variance, the second moment of the centroid of the RTD
reactors. The resultant three-parameter tank-in-series curve about the mean of the distribution [s2]) is deter-
(TIS) model is given in Eq 6a. mined from the second moment about the mean, as indi-
cated in Eq 9.
NN
冕 E( )d
nN
1 n N–1
E( )   
M (N – 1)! 冢 冣

M
tm  t苶N
0

E( )d
(8)

(6a)
2

Nn Nn

冢 冣
exp – 
(1 – n) 
M M 冢 冣 2  
t2m

0
(1 – )2 E( )d (9)

2
These two important parameters, tm and Θ , are used to
in which E( ) is a normalized exit age distribution curve; characterize RTD curves (Teefy & Singer, 1990). Com-
n is a fraction of flow rate effectively used in the treat- parison of the calculated tm with the theoretical mean

ment process (with no global bypassing); N is the number (HRT t N) yields an insight into the importance of prefer-
of reactors in the series; M is the fraction of a unit’s ential flow paths and stagnation zones in the tank (Ken-
volume effectively used in the treatment process (with no nedy et al, 2006).
stagnation of space); is the normalized residence time, The dimensionless exit-age distribution function E(Θ)
– –
⬅ t/t N (dimensionless); t N is the total mean residence can be statistically related to the cumulative exit-age
– – –
time in N tanks (s), t N = N t s = Vs/Q; t s is the mean distribution, also known as F(Θ) curve, given in Eq 10.
hydraulic residence time (HRT) of a single tank, calcu-
lated as the ratio of volume of the flocculation tank (Vs)
冕 n
in m3 to the bulk flow rate (Q) in ML/d; and is the F( )  E( )d  冱 E  (10)
0  1
Dirac delta function, = 0.0069 (assumed).
The TIS model represents the flow through a series of
equal volume of N continuously mixed flow reactors The function F(Θ) has an S-curve when plotted with
(CMFRs), allowing for dead-spaces (1 – M) and global respect to time t bounded between a minimum of 0 and
bypassing (1 – n) over the whole series of reactors. For a maximum of 1 (Templeton et al, 2006). The F(Θ) curve
n = 1 (no bypassing) and M = 1 (no dead-space), it can therefore represents the cumulative probability of x frac-
be seen that Eq 6a simplifies to a well-known single- tion of fluid element leaving the tank at a residence time
parameter RTD function for a series of completely of Θx, which is less than or equal to the normalized resi-
stirred reactors (Levenspiel, 1999), as given in Eq 6b. dence time (Θ). Certain flow-through curve characteristics
(e.g., tm, Θ
2, Θ , Θ
x max [normalized time at which E(Θ) is
equal to maximum {dimensionless}]) of both the E(Θ) and
N
E( )   (N )N–1 exp (– N ) (6b) F(Θ) are used as “indicators of flow.” These parameters
(N – 1)!
were used in characterizing the flow behavior in the floc-
culation tanks, which is consistent with other studies
The parameter Θ in the E(Θ) curve represents the num- (Stamou, 2008; Kennedy et al, 2006). The F(Θ) curves
ber of reactor volumes of fluid based on entrance condi- generated from the TIS model were used to compare the
tions that have flowed through the reactor in time t results of the CFD particle-tracking simulations.
(Fogler, 2006), given as For each residence time (Θ), the difference (or error)
t between the F(Θ) predicted by the TIS model and the CFD
⬅ 

tN (7) analysis was calculated (Martin-Dominguez et al, 2005),
given as
– –
in which the total reactor volume is t N = Nt s = Vs /Q.
The function E(Θ) in Eqs 6a and 6b is a normalized
RTD. It characterizes the probability distribution function
of the residence time (Θ) of a fluid element in the total tank
冑冱 冤
imax

i=1

F( i)CFDpredicted – F( i)TISpredicted
2

erf (M,n,N)   (11)


(Stamou, 2008). The RTD functions of the individual tanks (imax – 1)

VADASARUKKAI ET AL | PEER-REVIEWED | 103:11 • JOURNAL AWWA | NOVEMBER 2011 69

2011 © American Water Works Association


in which erf is error function, imax is the number of mea- by gravity. The treatment system consists of three rapid-
sured concentration, and i is the index representing cur- mixing tanks in series, four parallel units of a three-stage
rent concentration, tank number or time value. tapered hydraulic flocculation tank, eight direct dual-
media filtration units (filled with a 0.6-m-deep layer of
METHODOLOGY anthracite over a 0.3-m layer of sand), followed by dis-
Description of the full-scale system. The direct filtration infection with free chlorine. A four-flat-bladed propeller
system of J.D. Kline Water Supply Plant (JDKWSP) treats is installed in the rapid-mixing tanks for mixing the
water from Pockwock Lake, which is located in a pro- chemicals added during preoxidation and coagulation
tected watershed. The lake water is characterized by a processes (i.e., lime, potassium permanganate, carbon
low pH (5.0–5.2), a low alkalinity of ~ 0 mg/L (i.e., below dioxide, aluminum sulfate).
detection limit) calcium carbonate, and a low turbidity Flow from the third rapid-mixing tank spills over a
of 0.3–0.5 ntu. The design capacity of JDKWSP is about weir into the premixed shaft and then into a 1.2-m (48-
220 ML/d, with an average daily intake of 95 ML/d. The in.) distribution header system (Figure 1). This header
source water is pumped into the treatment facility at a delivers the flow to four identical hydraulic flocculation
constant flow rate through a 1.2-m (48-in.) inlet pipe; tanks through four 0.6-m (24-in.) tees. Each 0.6-m tee is
thereafter, it flows into subsequent treatment processes placed 10.7 m equidistant. Isolation butterfly valves are

FIGURE 1 Flow distribution arrangements of the three-stage hydraulic flocculation process of JDKWSP’s direct filtration*

4.0 m

Raw water supply Rapid mix Tank 2 Tank 3


1.2 m- (48 in.-) diameter pipe tank 1

4.0 m
Outflow of coagulated water from the
third rapid mix tank through a weir
1.2 m- (48 in.-) diameter conduit of length 19.5 m Rectangular 1.2 m- (48 in.-) diameter conduit
tank of length 12.8 m
0.6 m- (24 in.-)
diameter pipe

5.0 m

10.7 m
5.0 m

Pipe size and distance between the pipe dimensions Tank 1


are same for other pipes exiting from 1.2 m (48 in.)
m
2
1.

Tapered hydraulic
flocculation tank
Tank 2
m
5
1.

Tank 3

Floc water conduit of 45.4 m length

Dual media
filtration unit

JDKWSP—J.D. Kline Water Supply Plant

*Not to scale

70 NOVEMBER 2011 | JOURNAL AWWA • 103:11 | PEER-REVIEWED | VADASARUKKAI ET AL

2011 © American Water Works Association


provided on each of the flocculation tank inlets but are and 638,575, respectively for the flow distribution and
not used to control the flow split. Flow-splitting is based flocculation tank analysis, were selected.
on hydraulics associated with piping and conduits. The material properties (e.g., density and viscosity) of
As shown in Figure 1, each flocculation tank consists of water were changed to simulate summer and winter tri-
three rows of two parallel sets of cells (i.e., a total of six als. The temperature, density, and viscosity of water
cells). The length, width, and depth of each flocculation during the summer trial were set to 15°C, 998.23 kg/m3,
cell are 5.0 m, 5.0 m, and 8.3 m, respectively. Two tapered and 0.001002 kg/m s, respectively; whereas the fluid
vertical shafts, each with corresponding volumes of 11.95 properties of the winter trial were changed to 2.5°C,
m3 and 18.68 m3, are provided for flow transfer between 999.9 kg/m3, and 0.001645 kg/m s, respectively. The
the three rows of flocculation chambers. Water flows over total raw water flow into the plant was varied to four
a weir, through a vertical shaft, and enters the next cell at discharges of 69, 75, 90, and 120 ML/d for the flow
the bottom. The design of such an up-and-down flow distribution analysis. The Reynolds number for those
arrangement provided in the subsequent second and third discharges was estimated as 90,000–162,000 during the
flocculation chamber facilitates in achieving the G value flow-distribution investigation, whereas it ranged from
for mixing purposes. After flocculation, the flow exits the 95,000 to 169,000 in each flocculation tank. A fully
tanks through a port near the top of the last row of the turbulent flow regime developed in the tanks was mod-
chambers and is distributed to eight filters through a floc eled using the standard k- model, which is consistent
water conduit. Flow split to the filters is controlled by the with other studies (Bridgeman et al, 2010; Stamou,
effluent control values at the discharge of each filter. 2008; Templeton et al, 2006; Essemiani & De Traversay,
CFD modeling procedure. All numerical simulations of 2002; Haarhoff & Van der Walt, 2001), and in which
the present study were carried out using a finite volume– k = turbulent kinetic energy (m2/s2).
based software program.1 Using the program’s built-in Under steady-state conditions, the distribution of k and 
features, the Navier-Stokes equations coupled with turbu- at various locations in the tank was predicted by solving the
lence flow transport equations were approximated to a corresponding transport equations for k and  of the stan-
system of algebraic equations using discretization tech- dard k- model. Equation 4 was then used to calculate the
niques (Essemiani & De Traversay, 2002; Versteeg & Glocal values for each of the control volumes (cells) using the
Malalasekera, 1995). These equations governing the fluid corresponding , whereas Eqs 1 and 5 were used to estimate
flow were solved iteratively on a computer. A three-dimen- the average velocity gradient in the flocculation tanks.
sional, single-phase (i.e., water being the only fluid Boundary conditions. Mass-flow inlet boundary condi-
medium) simulation was performed to evaluate the flow tions (BCs) were assigned to the 1.2-m (48-in.) inlet pipe
distribution from the rapid mixing to the flocculation tanks for the flow distribution. Simulations were performed for
(Baek et al, 2005). The resultant flow distributed to each the flow distribution analysis by varying the mass-flow
flocculation tank inlet was then used to investigate the rates from 796.6 to 1,386.4 kg/s at the 1.2-m (48-in.) raw
flow-field characteristics in each flocculation tank. Two water inlet. The total mass flux exiting the 0.6-m (24-in.)
separate computational domains were defined using a tee in the flow distribution analysis was used to set the
preprocessing tool2 for flow distribution and flocculation mass-flow inlet BC for the flocculation tank analysis, sat-
tank simulations. The full-scale drawing of JDKWSP was isfying continuity principles. The mass-flow rate in accor-
used to create an exact geometry of the two flow problems. dance with the results obtained from the flow distribution
Symmetrical geometries of the first, second, and third floc- analysis varied from 76.2 to 194.84 kg/s. Pressure-outlet
culation chambers of a hydraulic flocculation tank were BCs assigned for the outlets were specified with a 0-Pa
modeled, as shown in Figure 1. gauge pressure. A zero gradient was assumed for the flow
In this study, a mesh-density analysis was conducted to variables (except pressure) at the outlet. The k and  at the
obtain a mesh-independent flow solution. The effect of inlet and outlet BCs were estimated based on the turbulence
mesh density on the average G-value calculation (Eqs 1, 5) intensity (I) that typically ranged between 5 and 10%.
was investigated for the flow distribution and the floccula- The normal velocity component and the normal gradi-
tion tank analysis separately. For instance, a mesh-density ent of all other flow variables were set to zero using rigid-
analysis was conducted at three mesh conditions (i.e., lid approximation at the symmetry plane of the floccula-
coarse, medium, and finer grids) for the flocculation tank tion tank. The free surface was also modeled using the
analysis. Medium and finer mesh densities of 524,584 and symmetric BCs because of a zero flux condition across
638,575, respectively, were obtained by refining the high- the symmetry, which was consistent with other studies
gradient regions in the coarse mesh (i.e., 100,130 tetrahe- (Goula et al, 2008; Stamou, 2008; Wang et al, 2008; Baek
dral elements). A maximum percentage difference in the et al, 2005; Zhou & McCorquodale, 1992). No-slip BC
predicted average G values between the coarse and medium was specified for the wall boundary, signifying a zero-
meshes was determined to be 12.86% for the flow distribu- fluid velocity occurred at the solid–fluid interface relative
tion and 21.92% for the flocculation tank analysis. To to the wall boundary, whereas the standard wall function
ensure the quality of mesh, finer mesh densities of 518,425 was used to model the viscosity-affected region between

VADASARUKKAI ET AL | PEER-REVIEWED | 103:11 • JOURNAL AWWA | NOVEMBER 2011 71

2011 © American Water Works Association


FIGURE 2 Flow distribution pattern of the four flocculation tank inlets at a minimum discharge condition
of 69 ML/d (summer simulation at 15°C)

1.2194
1.1585
1.0976
1.0367
0.9758
0.9149
Velocity Magnitude—m/s

0.8540
0.7930
0.6712
0.6103 Inflow (Q total) = 69 ML/d
0.5494
0.4885
Q4 (Qmax) =
0.4276 19.3 ML /d
0.3666
0.3057
Q 3 (Qmin) = 13.6 ML /d
0.2448
0.1839
0.1230 Q 2 = 17.5 ML /d

0.0621
0.0011 Y
Z Q 1 = 18.5 ML/d
X
Qmax—maximum discharge, Qmin—minimum discharge, Qtotal—total inflow into the treatment plant, Qx—discharge in flocculation
tanks, X—depth of tank, Y—width of tank, Z—length of tank

Red circle indicates the inlet location of the third flocculation tank, where minimum discharge occurs.

the wall and the fully turbulent region (Stamou, 2008; initial sets of values and BCs were solved iteratively under
Wang et al, 2008; Baek et al, 2005). steady-state conditions until convergence was attained.
A pressure-based solver using a segregated algorithm Convergence of solution was distinct when the monitor-
was used for all the simulations because of an incompress- ing residuals of the continuity equation were below the
ible flow condition. The numerical algorithm (interpola- tolerance limit of 1 × 10–5 and the momentum equation
tion methods) used to solve the governing equations for along the x, y, and z directions was < 0.001; the total
each individual control volume was “simple” for pres- mass flow rate imbalance between inlet and outlet was
sure-velocity coupling, “standard” for pressure, and less than 0.2%; and the velocity profile of the outlet sur-
“second-order upwind” for momentum, turbulent kinetic face remained constant with the increment in the number
energy, and turbulent dissipation energy. The flow-field of iterations (Vadasarukkai & Gagnon, 2010). All the
problems that were thus generated using appropriate computations were converged within ~ 2,000 iterations.

TABLE 2 Effect of total discharge on flow distribution to flocculation tank inlets (summer simulation at 15oC)

Flow Distribution to the Four Flocculation Tanks Standard


Average Deviation
Total Plant ML/d % Flow From the Average
Flow Rate (Qavg) Flow Rate ( )
ML/d Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 ML/d ML/d

69 18.5 17.5 13.6 19.3 26.9 25.4 19.7 28.0 17.2 2.54
75 20.1 19.0 14.8 21.0 26.9 25.4 19.7 28.0 18.8 2.76
90 24.2 22.8 17.8 25.2 26.9 25.4 19.7 28.0 22.5 3.31
120 32.2 30.5 23.7 33.6 26.9 25.4 19.7 28.0 30 4.41

Qavg—average discharge across the four flocculation tanks, Qx—discharge to flocculation tanks

72 NOVEMBER 2011 | JOURNAL AWWA • 103:11 | PEER-REVIEWED | VADASARUKKAI ET AL

2011 © American Water Works Association


RESULTS AND DISCUSSION average of turbulent viscosity (µt) was 1.5 kg/m s during
Analysis of flow distribution. Figure 2 shows the charac- summer and winter conditions. Changing the absolute
terization of the flow distribution to each flocculation tank value of viscosity (µ) and density of water for summer
inlet (i.e., Q1, Q2, Q3, and Q4) at a total inflow of 69 and winter trials did not alter the flow-field characteris-
ML/d. From the velocity profile, it was evident that there tics; however, the distribution of temperature was
was an inequity in flow distribution among the floccula- assumed to be spatially uniform throughout the tank.
tion tanks. The inlet pipe located at Q4, which delivered Future research, based on the work of Goula and col-
water to the outermost flocculation tank, had a maximum leagues (2008), will require investigating the influence of
discharge of 19.3 ML/d. The inlet pipe connected to floc- temperature gradient in the influent and bulk water on
culation tank Q3 received a minimum discharge of 13.6 the mixing characteristics in the flocculation tank.
ML/d. The percentage of flow distribution across Q3 was Through the flow-distribution analysis, the design of the
79% of the average flow across the four flocculation tanks flow distribution header system was shown to impart
(Qavg = 17.2 ML/d). unequal hydraulic losses and thus an unequal flow split to
The effect of variation in the actual plant’s intake rate the flocculation tanks. In particular, the Q3 inlet was situ-
on the flow distribution pattern was investigated. Three ated at a hydrostatic depth of 7.8 m immediately below
simulations were performed by increasing the total inflow the weir outflow. The pipe diameter of Q3 was half the size
to 75, 90, and 120 ML/d, respectively. As shown in Table of other conduits (i.e., 0.6 m or 24 in. each) connected to
2, the discharge in the third flocculation tank inlet was the distribution header. A turbulence region was created
significantly less compared with the Q1, Q2, and Q4 inlets. at the entrance of the third inlet (Q3) resulting in an
Increasing the total inflow significantly increased the devi- increased energy loss of nearly 0.22 m3/s3, which caused a
ation from the average flow. For instance, the standard minimum discharge condition at the Q3 inlet. However,
deviation ( ) was calculated to be 2.54 and 4.41 ML/d at the Q4 inlet, which experienced the maximum flow, had a
minimum and maximum inflow conditions of 69 and 120 relatively low average  of about 0.0026 m3/s3.
ML/d, respectively. Analysis of flow-field pattern in the flocculation tank.
A comparison of summer and winter simulations Inequity in the flow distribution resulted in a range of
revealed no significant variation in the percentage of flow minimum and maximum flow conditions at the floc-
distribution. The viscous effect of water was negligible in culation tank inlets. Initial simulation was carried out
the flow field because of dominant inertial effects in the third flocculation tank experiencing a minimum
(Cleasby, 1984); however, there was a marginal increase inflow of 13.6 ML/d at Q3, as predicted from the flow
of about 0.01% of flow into Q3. The volume-weighted distribution analysis. Equal distribution of the jet flow

FIGURE 3 Spatial variation of the G local values for the three-stage tapered hydraulic flocculation tank*

50.0000 Qmin1 = 6.8 ML/d (assuming equal flow into the


47.5111 symmetric tanks)
45.0222
42.5332 Large recirculation zone
40.0443
Local Velocity Gradient—s–1

37.5554
35.0665
32.5776
30.0886
27.5997 G floc1 = 9.5 s–1
G floc2 = 3.0 s–1
25.1108 G floc3 = 2.0 s–1
22.6219
20.1329
17.6440
15.1551 G local = 40–50 s–1
12.6662
10.1773
7.6883
5.1994 Y
2.7105 Z
0.2216 X

Glocal — local velocity gradient, Gt—shear work, a dimensionless quality, Gflocx — average velocity gradient
in each flocculation chamber, Qmin1—minimum discharge condition at the third flocculation tank, X—depth of tank,
Y—width of tank, Z—length of tank

*Flocculation design criteria: typical Gt values for flocculation process of 40,000 to 75,000 at 20°C
(AWWA/ASCE, 1997)

VADASARUKKAI ET AL | PEER-REVIEWED | 103:11 • JOURNAL AWWA | NOVEMBER 2011 73

2011 © American Water Works Association


FIGURE 4 Frequency distribution function of the Glocal for the first flocculation chamber had a range of Glocal values
the three-stage tapered hydraulic flocculation  9.5 s–1, the average G value, whereas the maximum
tank at a minimum flow of 13.2 ML/d limit increased to 104.1 s–1 when 95% of the tank vol-
ume was considered. The remaining 5% of the tank
volume had a range of Glocal values between 104.1 and
Tank 1
Tank 2 1,230.5 s–1. Only 1% of those control volumes (cells)
100 Tank 3 located near the influence of high turbulence zones (e.g.,
the tank inlet and exit) experienced a high set of Glocal
90
values (i.e., 254.8 s–1 < Glocal < 1,230. 5 s–1). Similarly,
80 the upper limit of the Glocal value for 95% of the tank
Cumulative Occurrence—%

70 volume was 24.87 and 14.4 s–1 in the subsequent second


60
and third chambers, respectively.
Table 3 outlines the 5th, 50th, and 95th percentile of
50
the Glocal values in the third tapered hydraulic floccula-
40 tion tank. The 50th percentile values in the first, second,
30 and third chambers represented a highly skewed Glocal
20
distribution; with 50% of the Glocal values concentrated
toward the left of the average G value, which was calcu-
10
lated using Eqs 1 and 5. The median value was 2.11 s–1
0 only as compared with the average G value of 9.5 s–1 in
0.1 1 10 100 1,000
the first flocculation tank.
Glocal—s–1 A short-circuiting path was formed near the inlet region
Glocal—local velocity gradient because of a recirculation zone in the interior of the first
flocculation chamber. In addition, a jet velocity of nearly
0.4–0.5 s–1 at the entrance caused some portion of the
incoming flow to rapidly exit into the second flocculation
into the symmetrical chambers was assumed so that chamber. The observed flow phenomenon was consistent
each chamber had a net inflow of 6.8 ML/d. Figure 3 with the interpretations made by other researchers (Goula
shows the spatial distribution of the Glocal values in the et al, 2008; Zhou & McCorquodale, 1992) for the sedi-
third tapered hydraulic flocculation tank. Similar to mentation tanks. Their studies showed an intense forma-
other studies (e.g., Bridgeman et al, 2010; Haarhoff & tion of recirculation zones and short-circuiting effects in
Van der Walt, 2001), this study also found a nonuniform those sedimentation tanks at a low turbulent kinetic
spatial variation in the Glocal values at different locations energy–dissipation rate. In contrast, the design of the inlet
within the tank. openings at the bottom of the second and third floccula-
A quantitative description of variation in the Glocal tion chambers directed the flow more toward the side-
values in the first, second, and third symmetrical floc- walls (i.e., in the y direction, as shown in Figure 3). Thus,
culation chambers was expressed in terms of a frequency the short-circuiting path was avoided. However, a rela-
distribution function. Figure 4 shows the cumulative tively flat hydraulic gradient line (HGL) of about 0.23 m
frequency of occurrence of the total number of control between the second and third flocculation chambers
volumes (cells) within a particular range of Glocal values. caused an overall reduction in the average G values to
For instance, about 76.4% of the total tank volume in 3.0 and 2.0 s–1, respectively.

TABLE 3 Percentiles of Glocal values for flocculation tanks 1, 2, and 3 at a minimum flow of 13.2 ML/d

Glocal Values—s–1

Flocculation Tank 5th Percentile 50th Percentile (Median) 95th Percentile Average G Value*—s–1

1—Total number of cells = 216,475 1.07 2.11 104.1 9.5


2—Total number of cells = 224,907 0.41 1.08 24.87 3.0
3—Total number of cells = 197,193 0.33 0.8 14.4 2.0

Glocal—local velocity gradient

*The average G value was calculated using Eqs 1 and 5 for each flocculation tank.

74 NOVEMBER 2011 | JOURNAL AWWA • 103:11 | PEER-REVIEWED | VADASARUKKAI ET AL

2011 © American Water Works Association


Particle-tracking and RTD curve analysis. To visualize
the recirculation and short-circuiting effects, the trajec- FIGURE 5 Path lines of neutrally buoyant particles
tories of simulated particles were investigated as a func- released at the inlet of the hydraulic
tion of space and time (Kukura et al, 2004). The parti- flocculation tank at 308 min
cles were intended to represent a conservative tracer that
was in equilibrium with the fluid motion (Kennedy et
309.3 69 of 77 particles exit at a theoretical
al, 2006). Using the flow modeling simulation software’s 293.9
t of 308 min
278.4
flow visualization feature, such neutrally buoyant par- 262.9
247.5
ticles were released at the inlet surface of the third 232

Path Lines—min.
216.5
hydraulic flocculation tank. The time required for those 201.1
185.6
170.1
particles to reach the outlet was tracked. Initially, par- 154.7
139.2
ticles were released from the inlet surface. This number 123.7
108.3
was generated automatically in the flow modeling sim- 92.8
77.33
ulation software depending on the two-dimensional 61.86
46.4
facets (i.e., the number of two-dimensional triangular 30.93
15.47
Y
Z Dead (stagnant/nonmixing) zones
0
faces) in the inlet surface. Sensitivity analysis was con- X 77 path lines of particles injected at t = 0
ducted to study the effect of increasing the total number
of particles released (Np = 77, 180, 383, 766, 1,665, and t—detention time, X—depth of tank, Y—width of tank, Z—length
of tank
3,363) on the particle residence time. (Np is the total

FIGURE 6 Dead (stagnant/nonmixing) zones represented


Optimization of hydraulic flocculator through a velocity contour plot along the vertical
plane at the mid-depth location in the second
designs are required to provide better flocculation tank (A) and the third flocculation
mixing for particle–chemical interaction tank (B)

and growth of flocs. A


0.100
0.095
0.090
Contours of Velocity Magnitude—m/s

0.085
0.080
0.075
number of neutrally buoyant particles released using 0.070
0.065
particle tracking analysis.) 0.060
0.055
Figure 5 shows the trajectories of 77 particles, which 0.050
were injected at the inlet surface of the first flocculation 0.045
0.040
chamber and tracked for 308 min. At a total inflow of 0.035
0.025
6.6 ML/d, the theoretical HRT was calculated to be 154 0.020
min (i.e., HRT = volume/bulk flow rate). At that theo- 0.015 Y Z
0.010
retical HRT, the path lines of less than half the number 0.005
X
of particles (i.e., 41 of 77 total particles) were traced at 0.000

the outlet of the third flocculation tank. With an


increase in the time interval to twice the theoretical HRT B
0.100
of 308 min, the number of path lines that reached the 0.095
0.090
outlet surface increased from 41 to 69 particles. How-
Contours of Velocity Magnitude—m/s

0.085
ever, about 10.4% (i.e., 8 of 77) particles were still 0.080
0.075
residing in the tank. This phenomenon can be correlated 0.070
0.065
to the recirculation flow, which was identified in the 0.060
first flocculation chamber during the flow-field analysis. 0.055
0.050
Most of the particles were entrained in the recirculation 0.045
region, whereas the short-circuiting path carried the 0.040
0.035
remaining particles to the second flocculation chamber. 0.025
0.020
As shown in Figures 5 and 6, the path lines of particles 0.015 Z
Y
never visited some portions of the second and third floc- 0.010
0.005
culation chambers. 0.000 X

It was evident from the analysis of particle-tracking


that each individual particle had a unique flow path. The
X—depth of tank, Y—width of tank, Z—length of tank
path lines were analyzed to develop a probability distri-

VADASARUKKAI ET AL | PEER-REVIEWED | 103:11 • JOURNAL AWWA | NOVEMBER 2011 75

2011 © American Water Works Association


bution function of the particle residence time in the third which was significantly less than the theoretical HRT. As
tapered hydraulic flocculation tank (Figure 7). The prob- shown in Figure 8, the cumulative exit-age distribution
ability distribution function also represented the dimen- [i.e., the F(Θ) curve] was developed from the E(Θ) curve.
sionless exit-age distribution [E(Θ)] curve as shown in The Θ10, Θ50, and Θ90 values (i.e., the time taken by 10,
Figure 8. The dimensionless exit-age distribution was 50, and 90% of the fluid particles to exit the outlet) were
computed as the ratio of the number of particles (Ni) that determined from the F(Θ) curve as 44.7, 95.48, and 138.6
exited the flocculation tank in a discrete time increment min, respectively. Increasing the total number of particles
of dΘ to the total number of particles initially released from NP = 77 to 3,363 particles did not significantly
(NP). In the E(Θ) curve, the principal peak of the particles improve the predictive accuracy of the E(Θ) curve. The
exiting the outlet occurred at about Θ = 0.45 (i.e., t = 69.3 tm calculated for NP = 3,363 particles was 111.3 min with
min), before the theoretical HRT at Θ = 1 (i.e., t = 154 a standard deviation of 7.3 min. As a result, it was con-
min). This reflected the stagnant backwater (dead) regions cluded from a computational standpoint that 77 particles
in the tank (Figure 6), which reduced the effective tank should be selected for particle-tracking analysis.
volume. As a result, a larger portion of flow reached the Martin-Dominguez et al (2005) showed that the tall and
outlet at a shorter residence time. The residence time of asymmetric characteristics of the E(Θ) curve were preva-
those particles entrapped in the dead space and recircula- lent in a small number of reactors in series (N < 11), and
tion zone was represented by the tail elongated to the in reactors with a large amount of dead space (M < 1).
right of the theoretical HRT (i.e., Θ > 1). Accordingly, different combinations of parameters of the
Using Eq 8 and 9, the mean residence time (tm) and the TIS model (i.e., N, n, and M) were analyzed for the three-
standard deviation (Θ) were calculated from the E(Θ) stage tapered hydraulic flocculation tank. Using Eq 10,
curve as 107.9 and 6.82 min, respectively. The E(Θ) curve the F(t) curve was generated from the E(t) curve for the
indicated that about 68% of the fluid elements spent proposed TIS model. Figure 9 shows a comparison of the
between 101.1 and 114.7 min in the tank (i.e., tm ± Θ), F(t) curve predicted using the particle-tracking analysis in

FIGURE 7 Probability distribution function of the residence time of particles in the hydraulic flocculation tank

Fitted lognormal distribution


Density of observation

1.4

1.2

1.0

0.8
Ni/NP

0.6

0.4

0.2

0.0
0.00 0.45 0.90 1.35 1.80 2.25 2.70
t
Normalized Residence Time — = –
tN

Ni—number of particles that exited the flocculation tank in a discrete time increment of d , NP—number of neutrally buoyant

particles initially released, t—detention time, tN—total mean residence time in N tanks, —normalized residence time

The bar represents the density of observations falling within each bin (i.e., residence time); the height of each bar represents
the proportion of particles (ratio) having that range of . The area under the curve gives the probability distribution function of
the particle residence time.

76 NOVEMBER 2011 | JOURNAL AWWA • 103:11 | PEER-REVIEWED | VADASARUKKAI ET AL

2011 © American Water Works Association


CFD and the TIS model. Initially, bypassing and dead ated at a low range of G values for most other inflow
space in the tank were neglected (i.e., n = M = 1) to gener- conditions, reflecting inadequate mixing in the floccula-
ate an ideal or a single-parameter TIS model (Eq 6b). The tion tanks. In contrast to the observed insufficient G
predicted tx values from the CFD simulations were sig- value in the tanks, the nondimensional product Gavgt
nificantly lower than the ideal tx values. For instance, the was still maintained at an adequate flocculation range
predicted t50 value obtained from the CFD simulation was of 44,788 to 71,596. The prolonged residence time (
about 36 min below the ideal t50 value of ~ 132 min. With
a decrease in the n value to 0.95 (i.e., 5% bypassing) and
M value to 0.80 (i.e., 20% dead space), the F(t) curve Hydrodynamic flow condition depends on
shifted to the left of the ideal curve (i.e., the single-param-
eter TIS model). The three-parameter TIS model showed both the geometry of the flocculation tank
a reasonable agreement with the CFD-predicted tx values and the impeller speed and type.
for M = 0.62 and n = 0.98, with an error function of
0.00823 (Eq 11). Allowing a high fraction of stagnant
space (38%; i.e., M = 0.62) in the entire flocculation tank,
the maximum cumulative fraction of tracer predicted 30 min) in the flocculation tanks created an offset in the
using the TIS model was nearly 88%, with 12% of the product Gavgt. This situation may give rise to an inad-
tracer still residing in the tank. equate number of effective particle collisions and con-
Implications on future operations. Table 4 shows the sequently the formation of filterable flocs. The finding
expected design parameter calculated from the CFD of Bernhardt and Schell (1993) agrees with the interpre-
analyses and the typical range of G value and detention tation that an average G value of < 30 s–1 cannot be
time recommended by AWWA/ASCE (1997) for a floc- offset by a longer aggregation time ( 30 min) in order
culation process. The average G values of the existing to attain the optimum turbidity removal ( 80%).
hydraulic flocculation process did not achieve the high- The applicability of the Gt concept as an effective
energy input required for a direct filtration system. High design parameter in classifying the performance of a
G values of 83 s–1 in the distribution header and 40 s–1 particular flocculation system has been previously dis-
in the flocculation tank were achieved only at a maxi- cussed. Argaman (1971) showed that the performance of
mum inflow of 120 ML/d. Otherwise, the plant oper- a flocculation system is directly dependent on the product

FIGURE 8 E( ) and F( ) curves predicted using the results of particle tracking CFD analysis for the hydraulic
flocculation tank

E ( ) Exiting particles older than the theoretical hydraulic time


1.4 F ( ) 1.0

0.9
1.2
0.8 Particles Leaving Tank at
Cumulative Fraction of

1.0 0.7

0.6
0.8
E( )

0.5
0.6
0.4

0.4 0.3

0.2
0.2
0.1

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2


= t/ tN
Normalized Time—


CFD—computational fluid dynamics, E( )—exit age distribution, F( )—cumulative exit age distribution, t—detention time, tN—total mean
residence time in N tanks, —normalized residence time

The hatched area indicates the fraction of those particles exiting the outlet—because of dead space and recirculation zones—that are
older than the theoretical hydraulic retention time.

VADASARUKKAI ET AL | PEER-REVIEWED | 103:11 • JOURNAL AWWA | NOVEMBER 2011 77

2011 © American Water Works Association


107.9 min at an inflow of 69 ML/d) was used instead of
FIGURE 9 Comparison of cumulative exit age distribution
the theoretical HRT (i.e., 154 min).
curves predicted from the CFD simulation ideal
CFD analyses of the hydrodynamics of flow distribu-
or TIS models
tion and the flocculation process raised a number of
CFD simulation concerns that may have an effect on overall treatment
Ideal or single-parameter TIS model
(n = 1, M = 1, N = 3, erf = 0.02355) efficiency. Figure 10 illustrates the inference of the CFD
Three-parameter TIS model outcomes on Halifax Water in terms of meeting the
(n = 0.95, M = 0.80, N = 3, erf = 0.0250) water quality standards and regulatory amendments.
Three-parameter TIS model
(n = 0.98, M = 0.62, N = 3, erf = 0.00823) Previous work on floc formation and breakage (Bridge-
man et al, 2008; Bouyer et al, 2005; Gregory, 2004) have
1.0
shown significant evidence of floc breakage with an
0.9 increase in the level of shear and a limit in their regrowth
Cumulative Fraction of Particles

0.8 capacity when the shear rate was reduced. Although the
0.7 JDKWSP has a relatively flat HGL during the plant’s
Leaving Tank at t

average inflow of 95 ML/d, there is a major break in the


0.6
HGL at the weir located immediately near the down-
0.5 stream of the third premix tank. The surface water eleva-
0.4 tion drops to ~ 0.94 m at that weir location during an
0.3 average flow of 95 ML/d, imparting an increased level
of shear as the flocs travel through the distribution
0.2
header toward the flocculation stage.
0.1 An unbalanced mixing condition in the flocculation
0.0 tank offers a limited supply of active mixing for floc
0 40 80 120 160 200 240 280 320
aggregation, apart from high spikes in the G values at the
Residence Time—min tank inlet and outlet locations. Thus the flocs are sub-
CFD—computational fluid dynamics, erf—error function, jected to an uneven proportion of high and low G values,
M—fraction of unit’s volume effectively used in the treatment which can potentially cause damage to the flocs that have
process (with no stagnation space), n—fraction of flow
rate effectively used in treatment process (with no global only been partially developed. The next step will be to
bypassing), N—number of reactors in series, t—detention time, account for the interaction of the local physicochemical
TIS—tank-in-series model
flocculation processes in the complex flow field developed
through a passive scalar description of the particle phase
Gt only for low values of G or for very strong flocs. (Samaras et al, 2010).
However, the author also indicated that a fine structure The present flocculation process of a relatively quies-
of turbulence field is required, which has a significant cent condition, interrupted with severe spikes in the
effect on performance. There are certain limitations with mixing energy, can be remediated by promoting a more
the Gt concept because it fails to consider the floc- consistent G value in the design. Neutralizing spikes in
breakup phenomenon. This study shows that product Gt the mixing gradient through a uniform tapered floccula-
alone was not an adequate parameter for describing floc- tion will improve floc formation, thereby optimizing the
culation performance. A better estimation of the Gt value efficiency of the flocculation process. Haarhoff (1998)
was obtained when the mean residence time (tm; e.g., developed a design equation to determine the exact floor

TABLE 4 G, t, and Gavgt values of hydraulic flocculators and the recommended design criteria for the flocculation process

G—s–1 t—s* Gavgt†

Treatment Design Present Design Present Design Present


Process Criteria‡ Study Criteria‡ Study Criteria‡ Study

Distribution channels mixer to flocculator 100–150 36–83 Varies 173–300 NA NA


High-energy flocculation for direct filtration 20–75 2–40§ 900–1,500 3,600–9,240 40,000–75,000 44,788–71,596

G—average velocity gradient value, Gavgt—shear work, a dimensionless quality, t—detention time

*Detention time for the design is the overall mean residence time in N number of flocculation tank-in-series (AWWA/ASCE, 1997).
†Gavgt is approximately calculated as the product of average of velocity gradients in the compartment, assuming equal size tanks, i.e., Gavg = (Gfloc1 + Gfloc2 +

Gfloc3)/N and t N (Hargrave et al, 1990).
‡Flocculation design criteria: typical G values and detention time for flocculation process at 20oC (AWWA/ASCE, 1997).
§Average G value in each flocculation tank, calculated corresponding to the inflow condition (i.e., 69–120 ML/d).

78 NOVEMBER 2011 | JOURNAL AWWA • 103:11 | PEER-REVIEWED | VADASARUKKAI ET AL

2011 © American Water Works Association


slope and water depth that would satisfy a range of
flocculation requirements for different flow rates.
FIGURE 10 Flow diagram of implication of CFD analyses
Another alternative for providing a controlled mixing
on future operations
gradient is to install mechanical mixers in each of the
flocculation chambers.

CONCLUSIONS Research outcomes from CFD analysis

This study reports on a computational analysis for


• Uneven flow distribution
hydraulic flocculation tanks. The CFD results indicated • Inadequate and unbalanced mixing conditions in flocculators
that the present design offered inadequate mixing in • Short-circuiting and large dead volumes with lower G values
• Inflexibility in adjusting the flow rates and therefore G value
these tanks. The mixing intensity (i.e., the G value) cal-
culated for the flocculation tank was dependent on the
total inflow into the treatment plant and its distribution
across the tanks. High spikes in the G values (~ 40–50 Water quality Effect on Future plan
Halifax
s–1) were observed at the bottom of the tank; the high Water
spikes accounted for as much as 20% to as little as 5%
of the total mixing in the flocculation tank. Most floc-
• Poor (low-density) floc formation
culation tanks have little to no mixing. From a tapered • Reliance on polymer in the winter • Mechanical
mixing perspective, the average G value significantly required for minimizing floc breakups flocculators
• Load on downstream filtration unit • Premix modification
drops from the first to the second tank, with a minimum
• Imbalance in ripening performance • Coagulant change
variation existing between the G values of the second among filters
and third flocculations.
This work focused on a plant in Halifax; however,
Regulatory Surface water treatment Regulatory
several plants across Canada have similar infrastructure. standards (e.g., DBP
These designs are based on an older design approach; rules)

as a result, there is a potential need to evaluate the


design in terms of local hydrodynamic parameters,
CFD—computational fluid dynamics, DBP—disinfection by-product,
which can be predicted effectively by CFD applications. G value—average velocity gradient value
Optimization of hydraulic flocculator designs are
required to provide better mixing for particle–chemical
interaction and growth of flocs. Practical outcome of
operating at ineffective mixing conditions might lead to modeling of unit processes for drinking water treatment
potential negative consequences such as poor floc for- optimization. Graham A. Gagnon (to whom
mation, reliance on excess flocculant aid, decreased correspondence should be addressed) is a professor and
incorporation of dissolved organics in floc, and NSERC/Halifax Water Industrial Research Chair,
increased particle load to the filters. Dalhousie University, Sexton Campus, 1360 Barrington
St., Halifax, Nova Scotia, Canada B3J 2X4; graham.
ACKNOWLEDGMENT gagnon@dal.ca. D. Reid Campbell is director of water
The authors thank Halifax Water and Natural Sciences services at Halifax Regional Water Commission. Sarah
and Engineering Resource Council of Canada (NSERC) C. Clark is a senior project manager at HDR
for financial support to the NSERC/Halifax Water Indus- Engineering, Denver, Colo.
trial Research Chair. The authors also acknowledge
Atlantic Computational Excellence Network for provid- Date of submission: 12/16/2010
ing the high-performance computing facility. Date of acceptance: 09/12/2011

ABOUT THE AUTHORS FOOTNOTES


1ANSYS FLUENT flow modeling simulation software, version 6.3.26,
Yamuna S. Vadasarukkai is a
ANSYS, Lebanon, N.H.
graduate research assistant at the 2GAMBIT meshing software, ANSYS, Lebanon, N.H.

Center for Water Resource Studies at


Dalhousie University, Halifax, Nova
Scotia, Canada. She is involved in the
research activity of Natural Sciences
and Engineering Research Council
JOURNAL AWWA welcomes
(NSERC)/Halifax Water Industrial
comments and feedback
Research Chair at the J.D. Kline Water Supply Plant.
at journal@awwa.org.
Vadasarukkai’s research work involves water quality

VADASARUKKAI ET AL | PEER-REVIEWED | 103:11 • JOURNAL AWWA | NOVEMBER 2011 79

2011 © American Water Works Association


REFERENCES Gregory, J., 2004. Monitoring Floc Formation and Breakage. Water
Sci. & Technol., 50:12:163.
Argaman, Y.A., 1971. Pilot-plant Studies of Flocculation. Jour.
AWWA, 63:12:775. Haarhoff, J. & Van der Walt, J.J., 2001. Towards Optimal Design
Parameters for Around-the-End Hydraulic Flocculators. Jour.
Bernhardt, H. & Schell, H., 1993. Effects of Energy Input During
Water Supply Res. & Technol.–Aqua, 50:3:149.
Orthokinetic Aggregation on the Filterability of Generated
Flocs. Water Sci. & Technol., 27:10:35. Haarhoff, J., 1998. Design of Around-the-End Hydraulic Floccula-
tors. Jour. Water Supply Res. & Technol.—Aqua, 47:3:142.
Arnold, A., 2008. Evaluation and Quantification of Engineered Flocs
and Drinking Water Treatability. Master’s thesis, University of Hargrave, W.J. & Loucks, A.L., 1990. Toronto-R.C. Harris Water
Waterloo, Canada. Treatment Plant. Queen’s Printer for Ontario, Toronto.
AWWA/ASCE (American Water Works Association/American Soci- Jarvis, P.; Jefferson, B.; Dixon, D.; & Parsons, S.A., 2008. Treatment
ety of Civil Engineers), 1997 (3rd ed.). Water Treatment Plant Options and Their Effect on NOM-Coagulant Floc Structures.
Design. McGraw-Hill, New York. Jour. AWWA, 100:1:64.
Baek, H.K.; Park, N.S.; Kim, J.H.; Lee, S.J.; & Shin, H.S., 2005. Exami- Kennedy, M.G.; Ahlfeld, D.P.; Schmidt, D.P.; & Tobiason, J.E., 2006.
nation of Three-dimensional Flow Characteristics in the Distri- Three-dimensional Modeling for Estimation of Hydraulic
bution Channel to the Flocculation Basin Using Computational Retention Time in a Reservoir. Jour. Envir. Engrg., 132:9:976.
Fluid Dynamics Simulation. Jour. Water Supply Res. & Tech-
Kingston Utilities, 2009. Kingston Annual Water-Quality Reports—
nol.–Aqua, 54:6:349.
Point Pleasant Water Treatment Plant. www.utilitieskingston.
Bouyer, D., Coufort, C.; Liné, A.; & Do-Quang, Z., 2005. Experimental com/water/waterreports.aspx?wopenpane=0 (accessed Sept.
Analysis of Floc Size Distributions in a 1-L Jar Under Different 15, 2011).
Hydrodynamics and Physicochemical Conditions. Jour. Colloid
Kukura. J.; Baxter, J.L.; & Muzzio, F.J., 2004. Shear Distribution and
& Interface Sci., 292:2:413.
Variability in the USP Apparatus 2 Under Turbulent Conditions.
Bridgeman, J.; Jefferson, B.; & Parson, S.A., 2010. The Development Intl. Jour. Pharmaceutics, 279:1-2:9.
and Application of CFD Models for Water Treatment Floccula-
Levenspiel, O., 1999 (3rd ed.). Chemical Reaction Engineering. John
tors. Adv. Engrg. Software, 41:1:99.
Wiley & Sons, New York.
Bridgeman, J.; Jefferson, B.; & Parson, S.A., 2008. Assessing Floc
Strength Using CFD to Improve Organics Removal. Chem. Martin-Dominguez, A.; Tzatchkov, V.G.; Martin-Dominguez, I.R.; &
Engrg. Res. Des., 86:8:941. Lawler, D.F., 2005. An Enhanced Tanks-in-Series Model for
Interpretation of Tracer Tests. Jour. Water Supply Res. & Tech-
City of Toronto, 2010a. Raw Water Intake—Island Water Treatment nol.—Aqua, 54:7:435.
Plant. www.toronto.ca/water/supply/supply_facilities/island/
raw.htm (accessed Feb. 15, 2010). Samaras, K.; Zouboulis, A.; Karapantsios, T.; & Kostoglou, M.,
2010. A CFD-based Simulation Study of a Large Scale Floccu-
City of Toronto, 2010b. Raw Water Intake—R.L. Clark Water Treat- lation Tank for Potable Water Treatment. Chem. Engrg. Jour.,
ment Plant. www.toronto.ca/water/supply/supply_facilities/ 162:1:208.
rlclark/filtration_process.htm (accessed Feb. 15, 2010).
Stamou, A.I., 2008. Improving the Hydraulic Efficiency of Water Pro-
Clark, M.M., 1994. Selection and Design of Mixing Processes for cess Tanks Using CFD Models. Chem. Engrg. & Processing,
Coagulation. AwwaRF, Denver. 47:8:1179.
Cleasby, J.L., 1984. Is Velocity Gradient a Valid Turbulent Floccula- Teefy, S.M. & Singer, P.C., 1990. Performance and Analysis of Tracer
tion Parameter? Jour. Envir. Engrg. ASCE, 110:5:875. Tests to Determine Compliance of a Disinfection Scheme With
Crittenden, J. & MWH, 2005 (2nd ed.). Water Treatment: Principles the SWTR. Jour. AWWA, 82:12:88.
and Design. John Wiley & Sons, Hoboken, N.J. Templeton, M.R.; Hofmann, R.; & Andrews, R.C., 2006. Case Study
Doyle, E., 2002. Production and Distribution of Drinking Water. Com- Comparisons of Computational Fluid Dynamics (CFD) Modeling
missioned Paper 8. The Walkerton Inquiry. Queen’s Printer for Versus Tracer Testing for Determining Clearwell Residence
Ontario, Toronto. Times in Drinking Water Treatment. Jour. Envir. Engrg. & Sci.,
5:6:529.
Ducoste, J.J. & Clark, M.M., 1999. Turbulence in Flocculators: Com-
parison of Measurements and CFD Simulations. AIChE Jour., Vadasarukkai, Y.S. & Gagnon, G.A., 2010. Determination of Conven-
45:2:432. tional Velocity Gradient (G) Using CFD Technique for a Pilot-
Essemiani, K. & De Traversay, C., 2002. Optimisation of the Floccu- Scale Flocculation System. Jour. Water Supply Res. & Tech-
lation Process Using Computational Fluid Dynamics. Proc. nol.–Aqua, 59:8:459.
Intl. Gothenburg Symposium on Chemical Water and Waste- Versteeg, H.K. & Malalasekera, W., 1995. An Introduction to Compu-
water Treatment VII (H.H. Hahn et al, editors). Gothenburg, tational Fluid Dynamics: The Finite Volume Method. Pearson
Sweden. Education Limited, Harlow, UK.
Fogler, H.S., 2006 (3rd ed.). Elements of Chemical Reaction Engi- Wang, X.L.; Zhou, S.; Li, T.; Zhang, Z.; Sun, Y.; & Cao, Y., 2008. Three-
neering. Prentice Hall International Series in the Physical and dimensional Simulation on the Water Flow Field and Sus-
Chemical Engineering Sciences, Upper Saddle River, N.J. pended Solids Concentration in the Rectangular Sedimenta-
tion Tank. Jour. Envir. Engrg., 134:11:902.
Goula, A.M.; Kostoglou, M.; Karapantsios, T.D.; & Zouboulis, A.I.,
2008. The Effect of Influent Temperature Variations in a Sedi- Zhou, S.P. & McCorquodale, J.A., 1992. Influence of Skirt Radius on
mentation Tank for Potable Water Treatment—A Computa- Performance of Circular Clarifier With Density Stratification.
tional Fluid Dynamics Study. Water Res., 42:13:3405. Intl. Jour. Numerical Methods Fluids, 14:8:919.

80 NOVEMBER 2011 | JOURNAL AWWA • 103:11 | PEER-REVIEWED | VADASARUKKAI ET AL

2011 © American Water Works Association


View publication stats

You might also like