Professional Documents
Culture Documents
Engineering Applications of Computational Fluid Mechanics
Engineering Applications of Computational Fluid Mechanics
Engineering Applications of Computational Fluid Mechanics
To cite this article: J. Bridgeman, B. Jefferson & S. A. Parsons (2009) Computational Fluid Dynamics Modelling of
Flocculation in Water Treatment: A Review, Engineering Applications of Computational Fluid Mechanics, 3:2, 220-241, DOI:
10.1080/19942060.2009.11015267
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of
the Content. Any opinions and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied
upon and should be independently verified with primary sources of information. Taylor and Francis shall
not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other
liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2, pp. 220–241 (2009)
ABSTRACT: The principal focus of this paper is to present a critical review of current approaches to modelling
the inter-related hydrodynamic, physical and chemical processes involved in the flocculation of water using
Computational Fluid Dynamics (CFD). The flows inside both laboratory and full scale mechanically-mixed
flocculators are complex and pose significant challenges to modellers. There exists a body of published work which
considers the bulk flow patterns, primarily at laboratory scale. However, there is little reported multiphase modelling
at either scale. Two-equation turbulence modelling has been found to produce variable results in comparison with
Downloaded by [New York University] at 18:48 13 June 2015
experimental data, due to the anisotropic nature of the swirling flow. However, the computational expense of
combining the sliding mesh treatment for a rotating mesh with the Reynolds Stress Model (RSM) in a full scale unit
is great, even when using a high performance computing facility. Future work should focus more on the multiphase
modelling aspects. Whilst opportunities exist for particle tracking using a Lagrangian model, few workers have
attempted this. The fractal nature of flocs poses limitations on the accuracy of the results generated and, in particular,
the impacts of density and porosity on drag force and settlement characteristics require additional work. There is
significant scope for the use of coupled population balance models and CFD to develop water treatment flocculation
models. Results from related work in the wastewater flocculation field are encouraging.
Keywords: computational fluid dynamics, turbulence, flocculation, mixing, multiphase modelling, fractal
dimension
220
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
flow processes in structures, there remain certain or inactivation of impurities, ranging in size from
areas which involve greater degrees of millimetres (grit, leaves) to microns (colloids,
computational complexity and where further viruses, protozoa). The principal processes
improvements are still required. One such involved in water treatment are identified in
example is flocculation. Table 1. Successful removal of these impurities
requires a range of different flow regimes
2. FLOW REGIMES AND SCENARIOS throughout a water treatment works (WTW). The
FOUND IN WATER TREATMENT accurate modelling of each process and flow
regime requires careful consideration, as each
2.1 Background presents its own subtleties and issues. Examples
Water treatment is based on a series of unit of the range of flow regimes and possible CFD
processes, each effecting some degree of removal modelling approaches are given in Table 2.
Raw Water Bulk storage of water > 1 day. Backup supply to WTW in event of source pollution.
Storage Some solids removal via sedimentation.
Coagulation Chemical (trivalent inorganic coagulant) Destabilisation of water via neutralization of colloidal
dose and short ( <30 s) rapid mix. material charge and precipitation of soluble compounds.
Flocculation Slow, extended (15–45 mins) mix. Encourage agglomeration of particles to form mass
fractal aggregates (“flocs”) up to 1000 μm.
Filtration Flow through porous granular media. Removal of smaller flocs and particles ( <100 μm).
Disinfection Chemical (chlorine, UV) dose and Killing or inactivation of potentially harmfully micro-
storage (chlorine only). organisms.
Table 2 Flow regimes and CFD modelling approaches adopted in water treatment.
221
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
Two-equation turbulence models are based on the surface charge and rendering them unstable,
eddy viscosity concept of Boussinesq. This eddy- whilst flocculation encourages particle
viscosity model states that the Reynolds stresses agglomeration via gentle mixing and the
are related to the local shear via eddy viscosity, νt , formation of irregularly-shaped, loosely
in the following manner (Rodi, 1993): connected mass fractal aggregates, known as flocs.
The size and structure of flocs are fundamental to
⎛ ∂U ∂U j ⎞ 2
− ui u j = ν t ⎜ i + ⎟ − k δ ij = 2ν t s ij − 2 k δ ij the efficient operation of WTWs. Ineffective
⎜ ∂x ∂xi ⎟ 3 3 coagulation and flocculation result in poorer
⎝ j ⎠
(1) quality feed water to downstream treatment
processes, potentially jeopardising treated water
The turbulent eddy viscosity, ν t , is defined in quality and increasing operational costs. Several
terms of the turbulent kinetic energy, k, and the interrelated criteria govern the efficiency of the
rate of dissipation of turbulent energy per unit coagulation and flocculation stages; viz.
mass, ε, according to: coagulant type and dosage, pH and mixing
k2 arrangements.
k-ε model: ν t = Cμ (2) Flocculation is the transformation of smaller
ε
destabilised particles into larger aggregates or
k-ω model: k flocs which are subsequently removed via
ν t = Cμ (3)
ω sedimentation or flotation processes. For all flocs
in their initial growth phase, the floc formation
ui u j ∂U i ⎛ ∂U i ∂U j ⎞ process is understood to be a balance between the
k= and ε = ν ⎜ + ⎟ (4)
2 ∂x j ⎜ ∂x ∂xi ⎟⎠ rate of collision-induced aggregation and the rate
⎝ j
of breakage for given shear conditions (Bouyer
δ ij represents the Kronecker delta and is defined et al., 2005; Mikkelsen, 2001; Biggs and Lant,
as: 2000).
⎧1 i= j R floc = α ij Rcol − Rbr (6)
δ ij = ⎨ (5) ij
⎩0 i≠ j
where Rfloc is the overall rate of floc growth, α is
A detailed discussion of the terms in the k-ε the collision efficiency factor (0 < α < 1), R col is
models can be found in Launder and Spalding the rate of particle collision and Rbr represents the
(1974). The principal turbulence models which rate of floc breakage.
are generally packaged with commercially
available CFD packages are the standard k-ε
model, the low Reynolds number k-ε model (LRN
k-ε), the renormalized group k-ε model (RNG k-ε),
the realizable k-ε model, the standard k-ω model,
the shear-stress transport (SST) k-ω model, and
222
Downloaded by [New York University] at 18:48 13 June 2015
Table 3 Comparison of available turbulence models (from Versteeg and Malalasekera, 1995, Marshall and Bakker, 2004, and Menter, 2003).
Standard k-ε • Semi-empirical modelling of k and ε. • Simplest model. • Poor performance in some scenarios (i.e.
• Valid for full turbulence only (molecular • Excellent performance for many flows. rotating flows, flow separation, adverse
diffusion ignored). • Well established. pressure gradients).
• Assumes isotropy in turbulence.
• Poor prediction of lateral expansion in 3-d
wall jets.
Renormalized • Based on statistical methods, not observed • Improved performance for swirling flow • Less stable than the standard k-ε model.
Group (RNG) k-ε fluid behaviour. compared to the standard k-ε model.
• Mathematics is highly abstruse. Texts only
quote model equations which result from it.
223
• Effects of small-scale turbulence
represented by means of a random forcing
function in N-S equations.
• Procedure systematically removes small
scales of motion by expressing their effects
in terms of larger scale motions and a
modified viscosity.
• Similar in form to sk-ε, but modified
dissipation equation to describe high-strain
flows better.
• Differential equation solved for μ t (changes
Cμ from 0.09 to 0.0845 at high Re), so good
for transitional flows.
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
Downloaded by [New York University] at 18:48 13 June 2015
Table 3 (Continued)
Realizable k-ε • Recent development. In highly strained • Suited to round jets, as well as swirling • Not recommended for use with multiple
flows, the normal Re stresses, u i2 , become flows and flows where separation occurs. reference frames.
negative (unrealizable condition), so μ t
uses variable Cμ .
• Cμ is function of local strain rate and fluid
rotation.
• Different source and sink terms in transport
equations for eddy dissipation.
• Good for spreading rate of round jets.
Standard k-ω • ω = ε /k = specific dissipation rate. • Valid throughout the boundary layer, • Separation is typically predicted to be
224
• sk-ε solves for dissipation of turbulent subject to sufficiently fine grid resolution. excessive and early.
kinetic energy, k-ω solves for rate at which
dissipation occurs.
• Resolves near wall without wall functions,
so can be applied through boundary layer.
Shear Stress • As k-ω except for gradual change from k-ω • Suitable for adverse pressure gradients and • Less suitable for free shear flows.
Transport (SST) k-ω in inner region of boundary layer to high pressure-induced separation.
Re version of sk-ε in outer part. • Accounts for the transport of the principal
• Modified μ t formulation to account for turbulent shear stress.
transport effects of principal turbulent shear
stresses.
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
Reynolds Stress • The most general of all models. • Accurate calculation of mean flow • Computationally expensive.
Model properties and all Reynolds stresses. • Not always more accurate than two-equation
• Yields superior results to k-ε models for models.
flows with stagnation points. • Harder to obtain converged result.
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
More accurately, the α ij R col ij term should be specific and interesting challenges for the
modeller.
broken down further into its constituent parts
Flocculation at WTWs is effected either via
where α ij R col ij is a function of
mechanical or hydraulic means. For mechanical
flocculation, the energy input is via an agitator
BM
α ij BM Rcolij , Shα ij Sh Rcolij , and DS α ij DS Rcolij (7)
which generates the necessary shear stress,
whereas in hydraulic flocculation, energy is
where i and j refer to discrete particles, and BM, imparted via the headloss across a baffled,
Sh and DS refer to the collision mechanisms of serpentine channel. The two processes are shown
Brownian Motion, shear and differential schematically at Fig. 1.
settlement respectively. (Derivations of these Coagulation and flocculation optimisation are
mechanisms are to be found in the literature, e.g. generally considered at the laboratory scale, using
Kusters, Wijers and Thoenes, 1997). As a result, a jar test apparatus and procedure. This well-
flocs do not continue to grow throughout the established process optimisation technique allows
flocculation stage, but rather, they attain a a rapid assessment of key variables (e.g.
limiting size beyond which breakage prevents coagulant dose, pH, mixing speed and
further overall growth. The flocculation process, flocculation time). The test apparatus typically
and hence this limiting size, is governed by comprises four or six glass vessels, each with a
physico-chemical conditions and by the shear
Downloaded by [New York University] at 18:48 13 June 2015
Fig. 1a Schematic diagram of sectional view of Fig. 1b Schematic diagram of plan view of an hydraulic
a mechanical flocculator. flocculator, from Haarhoff and Van der Walt (2001).
225
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
6
6
25
150
127
25
76
53
30
76
100 115
6 6
A A B B
115
Plan Plan
Fig. 2 Typical jar test vessel configurations (not to scale, all dimensions in millimetres).
226
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
227
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
demonstrating the importance of turbulence if the volume fraction of water in any given cell
model selection. is represented by α w , then,
All turbulence models offer advantages and
α w = 0 implies that the cell is empty of water,
disadvantages over others in certain applications,
and no single turbulence model can be judged to
α w = 1 implies that the cell is full of water, and
be the optimum model in all circumstances. 0 < α w < 1 implies that the cell contains the
Consequently, it is necessary to evaluate each air/water interface.
circumstance individually and form a view as to Haque, Mahmud and Roberts (2006) simulated
which model would provide the best fit for the the flow field in unbaffled vessels mixed with a
particular flow under examination. Factors to paddle impeller and a Rushton turbine and
consider include, inter alia, the physics adopted the VOF approach to represent the
encompassed in the flow, the established practice vortex formed at the surface. Using the MRF
for a specific class of problem, the level of approach to rotating flow and the RSM to model
accuracy required, the available computational turbulence, Haque, Mahmud and Roberts (2006)
resources, and the amount of time available for reported a typical run time of 70 hours to achieve
the simulation. It is clear from the above that the convergence (Sun UltraSPARC III processor,
selection of turbulence model is not a 4Gb RAM, 900 MHz clock speed). Thus, it can
straightforward one. Menter (2003) correctly be seen that the computational expense of
considering the liquid surface shape is significant.
Downloaded by [New York University] at 18:48 13 June 2015
228
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
⎧⎪⎛ ∂u ∂v ⎞ 2 ⎛ ∂u ∂w ⎞ 2 ⎛ ∂v ∂w ⎞ 2 ⎫⎪
0.5
a single number (Cleasby, 1984; Clark, 1985;
G = ⎨⎜⎜ + ⎟⎟ + ⎜ + ⎟ + ⎜⎜ + ⎟⎟ ⎬ (8) Graber, 1994; Luo, 1997; Jones, 1999).
⎪⎩⎝ ∂y ∂x ⎠ ⎝ ∂z ∂x ⎠ ⎝ ∂z ∂y ⎠ ⎪⎭ Furthermore, the distribution of velocity
gradients within a stirred tank is clearly not
where u, v and w are the velocity components in
uniform, and local power consumption at a point
the x, y and z directions of a Cartesian coordinate
of high turbulence within a vessel (e.g. adjacent
system.
to the impeller) can be several orders of
Camp and Stein termed this the absolute velocity
magnitude in excess of the rest of the vessel (Luo,
gradient and related this to work done per unit
1997). When considering flocculation, this is
volume per unit time via:
unfortunate as it is precisely the magnitude and
Φ=μG
2
(9) fluctuations in local shear to which a floc is
subjected, and not the average value, which
From which: determine the success of flocculation.
However, using CFD it is possible to quantify
P /V ε (10) and understand the local impact of mean flow
G= =
μ ν and turbulence on floc formation and break-up
using the local velocity gradient, GL , where
where Φ work of shear per unit volume per
unit time at a point ε
Downloaded by [New York University] at 18:48 13 June 2015
GL = (13)
P is power dissipated ν
V is tank volume
Korpijarvi, Laine and Ahlstedt (2000)
μ is the dynamic viscosity of the water,
demonstrated the very large variability of the
ε is the energy dissipation rate per unit
local velocity gradient value (GL ) within a jar
mass
test vessel, thus questioning the validity of the
ν is the kinematic viscosity of the
water use of the G value to characterise the vessel.
Unsurprisingly, the largest GL values were to be
In theory, the absolute velocity gradient can be found near the paddles and particularly in the
calculated at any point within a mixing vessel, turbulent area behind the panel, stretching in a
provided that the power dissipated is known at radial direction towards the wall.
that point. In practice, however, the flow
characteristics vary within the mixing vessel
from point to point, and so too does the energy
dissipation. Consequently the velocity gradient is
a function of both time and position. Given the
difficulties associated with calculation of G,
workers have traditionally replaced the absolute
velocity gradient with an approximation of the
exact value; that is its average value throughout
the vessel, G :
Pave
G = (11)
Vμ
229
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
values where, subject to a de minimus, one absolute terms and normalized by dividing by the
would expect coalescence (i.e. floc growth) to G value. They then introduced the concept of a
occur. Similar to Korpijarvi et al. (1999) and 95th percentile of the normalized G value
Korpijarvi, Laine and Ahlstedt (2000), the distribution in the flocculator as an arbitrary (but,
authors found that the distribution of local in their opinion, realistic) performance indicator
velocity gradients indicated high GL intensity for flocculator optimisation (i.e. a G value which
adjacent to the impeller, decreasing with distance is exceeded only in 5% of the total flocculator
from it (Fig. 5). volume). The authors argued that use of this G95
Essemiani and de Traversay (2002a) found that figure enables the designer to compare absolute
for the system considered, G and the average GL and normalized G values on a more
values exhibited reasonably close correlation, representative basis than by simply using the
although there was a great range of GL values; a average. Using this approach, Haarhoff and van
conclusion corroborated by Bridgeman, Jefferson der Walt were able to conclude that the slot
and Parsons (2008). Essemiani and de Traversay width ratio is the single most important
concluded that G does not account for variables geometric ratio for design purposes. The effect of
such as impeller position, geometry, operating the overlap ratio was found to have a less
mode and residence time. However, these pronounced impact on the G values, whilst the
parameters do affect the velocity gradient effect of the depth ratio was inconsequential if
Downloaded by [New York University] at 18:48 13 June 2015
distribution, and therefore, spatial floc size the channel velocity was maintained.
distribution. As a result, Essemiani and de
Traversay suggested that for the same G value, 3.5 Floc breakage
changing mixing rates and impeller type will Having expended time and energy in developing
affect performance. flocs, it is important that operating conditions do
Craig et al. (2002) developed a model using not subsequently cause their break-up. As a
Fluent to simulate the operation and performance result, floc strength, growth and breakage have
of a simple flocculation tank equipped with an been the subject of detailed research in recent
axial impeller generating turbulent flow years.
conditions. No details of modelling technique or Floc size may be considered to be a balance
turbulence model used were provided. Craig et al. between the hydrodynamic forces exerted on a
used the model to demonstrate variations in the floc and the strength of the floc. Where the floc
GL value, and to identify maximum and strength is resistant to the hydrodynamic forces,
minimum GL values associated with different one would expect floc size either to remain
operational scenarios. The authors identified that constant or for growth to occur. Where the
use of GL values in this manner enabled the hydrodynamic forces exceed floc strength,
identification within the flocculation tank of breakage will occur. Consequently, the
those areas where both coalescence and floc conceptual growth/breakage mechanism may be
break-up would occur. Similar to Essemiani and expressed as follows:
de Traversay (2002a), this led the authors to
Hydrodynamic Forces F
question the validity of the use of the G value as B= = (14)
Floc Strength J
a design parameter.
Haarhoff and van der Walt (2001) used the where F represents the hydrodynamic forces
lesser-used Flo++ code to study flow in an exerted by the flow, and J represents the strength
around-the-end hydraulic flocculator. Building of the floc (Coufort, Bouyer and Liné, 2005). It
on previous work (van der Walt, 1998), Haarhoff is clear from Eq. (14) that breakage will occur
and van der Walt used CFD to develop when B > 1, and floc size will be maintained or
qualitative guidelines for three fundamental increased when B < 1. Floc strength, J, is a
geometric ratios (slot width ratio, overlap ratio function of the physico-chemical conditions (raw
and depth ratio) (see Fig. 1b). No details of water type, coagulant type and dose) and the floc
modelling strategy were provided in the paper. structure.
Haarhoff and van der Walt also studied the Yeung, Gibbs and Pelton (1997) suggested that
variation of the G value in hydraulic flocculators. the hydrodynamic force required to pull apart a
It is known that hydraulic flocculators incur floc in tensile mode may be expressed as:
zones of high turbulence at the baffle edges π
which can cause floc break-up. The authors F≈ .σ .d 2 (15)
4
showed variations in GL in the flocculator in both
230
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
where F represents the floc rupture force, σ performance of existing flocculators or to design
represents the floc strength, and d is the floc new installations, thus permitting optimisation on
diameter. In the viscous sub-range, σ = μ . ε ν . the basis of implied floc strength. Indeed,
Substituting into Eq. (14) shows that the Bridgeman, Jefferson and Parsons (2007)
breakage mechanism in the viscous sub-range, modelled two full scale flocculators (one
BVSR , may be expressed as: mechanically mixed, the other an hydraulic
installation) and assessed their performance in
C1 μ ε ν .d 2 terms of the breakage threshold concept.
BVSR = (16)
J However, the computational expense of using
sliding mesh to model a three dimensional full
where C1 is a constant. scale flocculator using the realizable k-ε model
Thomas (1964) suggested that in the inertial sub- was significant. (Using a high performance
range computing facility, the full scale mechanical
2 2 2 flocculator sliding mesh simulations required up
σ = ρ .u ′ and u ′ = C2 (ε .d ) 3
(17) & (18) to 12 days’ CPU usage).
Substituting into Eq. (14) shows that the
breakage mechanism in the inertial sub-range, 3.6 Residence time distribution
BISR , may be expressed as: One means by which flocculator performance
Downloaded by [New York University] at 18:48 13 June 2015
231
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
Term Definition
t10 , t50 , t90 Time at which 10, 50 and 90% of tracer has passed through reactor
ti / τ Index of short-circuiting. Ideal plug flow reactor = 1. Tends to 0 with increased short-
circuiting
232
Downloaded by [New York University] at 18:48 13 June 2015
Ducoste & Clark, 1999 FIDAP Code evaluation for Lab 3D sk-ε None LDV
modelling flocculator fluid
mechanics.
Bridgeman, Jefferson & Parsons, 2008 Fluent Evaluation of floc strength Lab 3D sk-ε, Rz k-ε, RNG DPM LDA
and trajectory in jar tester. k-ε, k-ω, SST k-ω,
RSM
Bridgeman, Jefferson & Parsons, 2007 Fluent Flow characteristics in two Lab 3D sk-ε, Rz k-ε, RNG DPM LDA, PIV
jar test configurations. k-ε, k-ω, SST k-ω,
RSM
Flow characteristics in
233
mechanical flocculator. Full sk-ε, Rz k-ε, RSM DPM, RTD, Power
RTD measurements
optimization. velocity
measurements
Korpijarvi, Laine & Ahlstedt, 2000 CFX GL variations in jar tester. Lab 3D sk-ε, RNG k-ε, None LDA, PIV
RSM
Downloaded by [New York University] at 18:48 13 June 2015
Table 5 (Continued)
Korpijarvi et al., 1999 CFX Turbulence modelling and Lab 3D sk-ε, RNG k-ε, None LDA, PIV
mesh density analysis jar LRN k-ε, RSM
tester.
Lainé et al., 1999 Fluent Flow characteristics to Full 2D RNG k-ε DPM None
determine cause of
underperformance.
Nopens, 2007 Fluent Coupling CFD and PBM in Full 2D ske PBM None
wastewater clarifier. (QMOM)
Prat & Ducoste, 2007 Phoenics Evaluation of Lagrangian Lab 3D ske Lagrangian, Experimental data
and Eulerian approaches Eulerian,
234
for simulating flocculation PBM
in stirred vessels.
N/A – not applicable, ND – not discussed, ADV – acoustic Doppler velocimetry, LDV – laser Doppler velocimetry, LDA – laser Doppler anemometry,
PIV – particle image velocimetry, RTD – residence time distribution, MFM – multi-fluid model.
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
Most commercially-available CFD software has mass fractal aggregates—i.e. they demonstrate
an in-built discrete phase model which can be self-similarity irrespective of scale and
used to model the trajectories of individual demonstrate a power law relationship between
particles within a flow via integration of the force mass (or volume) and length, such that:
balance on a particle in a Lagrangian reference df
frame. Equating particle inertia with the forces MαL (31)
acting on it, and applying a Cartesian co-ordinate where df is a non-integer mass fractal dimension.
system, yields: A low fractal dimension (< 2.0) indicates an open
du p g x (ρ p − ρ ) structure, whereas a higher fractal dimension
= FD (u − u p ) + + Fx (25) indicates a more compact structure. (For regular,
dt ρ
three-dimensional objects, df = 3.0).
where, u is the fluid velocity, up is the particle This fractal nature has some very important
velocity, μ is the dynamic viscosity of the fluid, consequences on floc settlement and transport and
and ρ and ρp are the fluid and particle densities, the modelling thereof. For example, as floc size
respectively, FD (u − u p ) is the drag force per unit increases, the density decreases (Tambo and
Watanabe, 1979), with the floc effective density
particle mass, and
represented by
18μ CD Re
FD = . (26) ρE = B . a − y (32)
Downloaded by [New York University] at 18:48 13 June 2015
ρ p d p 2 24
where a is floc size and B and y are constants.
where dp is the particle diameter, and Re is the The effective density is proportional to the
relative Reynolds number, defined as: volume fraction of solid in the floc, such that
ρd p u p − u ρ E = φs ( ρ p − ρ ) (33)
Re = (27)
μ
and hence y is related to df via
The drag coefficient, CD , is defined as:
df = 3− y (34)
α2 α3
C D = α1 + + (28)
Re Re 2 Furthermore, because flocs are not spherical, the
drag coefficient expression must be amended.
where α n are constants that apply to smooth Assuming the sphericity of flocs to be
spherical particles over several ranges of Re given approximately 0.8, at low Reynolds numbers,
by Morsi and Alexander (1972) ( α n values are Tambo and Watanabe (1979) suggested that
shown in Appendix 2). CD ≈ 45 / Re (35)
The term Fx in Eq. (25) relates to additional forces
in the particle force balance that are relevant only i.e. almost twice the value predicted via Stokes’
under special circumstances; for example, forces law (= 24/Re).
on particles that arise due to the rotation of the Whilst this deviation from sphericity may be
reference frame. Considering rotation about the z easily addressed via a simple adjustment to
axis, the forces on the particles in the x and y Eq. (28), the porosity of flocs (and hence the
directions may be expressed as: possibility of flow through the floc itself) poses
additional modelling challenges. The net effect of
⎛ ⎞ ⎛ ⎞
⎜1 − ρ ⎟Ω 2 x + 2Ω⎜ u y , p − ρ u y ⎟ (29) porosity is that a floc will experience reduced
⎜ ρ ⎟ ⎜ ρ p ⎟⎠ drag compared to an impermeable sphere of the
⎝ p ⎠ ⎝
same size and density. This effect becomes
where uy , p and uy are the particle and fluid significant for more open flocs with low fractal
velocities in the y direction, and dimensions (df < 2.0), settling faster than solids of
the same size and density.
⎛ ⎞ ⎛ ⎞
⎜1 − ρ ⎟Ω 2 y + 2Ω⎜ u x , p − ρ u x ⎟ (30) However, the Lagrangian particle models treat all
⎜ ρ ⎟ ⎜ ρ p ⎟⎠
⎝ p ⎠ ⎝ particles as point masses, and therefore cannot
consider the effect of porosity directly. One
where ux , p and ux are the particle and fluid
means of addressing this is to assume that the
velocities in the x direction.
porosity has an effect on the drag forces and
However, the work of Morsi and Alexander (1972)
consequently amend the drag coefficient
was predicated on the assumption that particles
accordingly. However, the means by which the
are spherical, whereas flocs are irregularly-shaped,
information required to do this with any degree of
235
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
accuracy is unclear. This also involves further use of the QMOM. The QMOM offers reduced
amendment of a coefficient to which adjustment computational expense compared to the discrete
has already been made to accommodate deviation model, but it differs from the SMM by replacing
from sphericity. This clearly adds an additional the exact closure with an approximate closure,
degree of uncertainty into the modelling process. requiring only a small number of scalar equations
to track population moments (McGraw, 1997).
3.8 Population balance modelling (PBM) The reduced computational cost of the moment
Referring back to the modelling work undertaken methods means that they can be combined with
on water treatment flocculation, it can be seen that CFD calculations such that the effects of local
much work in the water treatment field has (rather than global average) flow field
addressed flow patterns. However, it is not just characteristics can be considered.
the hydrodynamic behaviour of flocculators Whilst the development of PBM techniques has
which is of interest, but the agglomeration, advanced, the coupling of PBMs with CFD
growth and breakage processes are of great remains in its infancy, particularly in water
significance also. Multiphase models which treatment applications with most (but not all)
incorporate a particle size distribution (PSD), work being undertaken using wastewater. Prat
such as those found in a flocculator, require a and Ducoste (2007) considered the evolution of
population balance model (PBM) to describe flocs using the QMOM and used CFD to solve the
Downloaded by [New York University] at 18:48 13 June 2015
particle population changes. Several solution turbulent flow field within a 28 litre stirred vessel
methods exist, including, inter alia, the discrete, with clay particles. Whilst the results provide
class size method (Hounslow, Ryall and Marshall, proof of concept for the decoupling of fluid
1998), the standard method of moments (SMM) flow and flocculation dynamics in the
(Randolph and Larson, 1971), and the quadrature Lagrangian/QMOM approach, they do not
method of moments (QMOM) (Marchisio, Vigil provide information regarding the growth and
and Fox, 2003). A full description of each is strength of flocs in specific water treatment
outside the scope of this paper and only brief applications. Other recent work includes that of
comments on each are made. Nopens (2007) (CFD and QMOM-based PBM in
The discrete method requires the particle wastewater clarifier) and Feng and Li (2008)
population to be discretised into a finite number (theoretical PBM approach, encapsulating internal
of size bins and is based on the representation of body forces and fluid shear stress). A detailed
the PSD in terms of those bins. This method has review of all recent PBM work is outside the
the advantage that the PSD is calculated directly; scope of this paper; the interested reader is
however, the bins must be defined from the outset referred to Hounslow, Ryall and Marshall (1988),
and a large number may be required. Further, the Nopens and Vanrolleghem (2006) and Coufort
coupling of CFD and PSD models requires the et al. (2007). It is the work of Nopens which
incorporation of transport equations in each bin, would appear to have advanced the application of
making the process computationally expensive. CFD and PBM modelling to the greatest extent
Unlike the discrete method, moment methods thus far. However, it is clear from the literature
simulate statistical information about the PSD, that there has been no work undertaken on raw
rather than deriving an accurate description of the waters abstracted from WTW at either lab or full
PSD. With the SMM, the population balance scale. Consequently, there remains the need to
equation is transformed into a set of transport simulate accurately realistic water treatment
equations for moments of the distribution which, flocculation processes at both laboratory scale
for lower-order moments, are relatively easy to using raw water samples taken from WTW, and
solve. No prior assumptions are made with regard also at full scale.
to the size distribution and the moment equations
4. CONCLUSIONS
are closed (involving only functions of the
moments themselves). However, this latter point
The use and application of CFD within water
prevents particle aggregation and breakage being
treatment has expanded significantly over the
written as functions of moments, thus meaning
period 1995–2008. Examples of the technique’s
that the method is only applicable in situations of
application can be found for most unit processes,
constant aggregation, size independent, and
from the storage of raw water, through detailed
growth. For the simulation of flocculation in
and technical water treatment processes, to
water treatment, these are significant limitations.
chemical disinfection. Model complexity varies
However, these limitations can be overcome via
from application of two-equation turbulence
236
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
237
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
14. Ducoste JJ, Clark MM (1999). Turbulence in and Wastewater Treatment VI, Proceedings
flocculation: comparison of measurements of the 9th Gothenburg Symposium. Ed. Hahn
and CFD simulations. AIChE J. 45(2):432– HH, Hoffmann E, Odegaard H, Springer, 89–
436. 99.
15. Essemiani K, de Traversay C (2002a). 28. Kusters K, Wijers J, Thoenes D (1997).
Optimisation of the flocculation process using Aggregation kinetics of small particles in
computational fluid dynamics. Chemical agitated vessels. Chem. Eng. Sci. 52(1)107–
Water and Wastewater Treatment VII, 121.
Proceedings of the 10th Gothenburg 29. Lainé S, Phan L, Pellarin P, Robert P (1999).
Symposium. Ed. Hahn HH, Hoffmann E, Operating diagnostics on a flocculator-
Odegaard H, IWA Publishing. settling tank using FLUENT CFD software.
16. Essemiani K, de Traversay C (2002b). Water Science and Technology 39(4):155–
Optimum design of coagulation / flocculation 162.
vessels. WQTC 2002 conference, Seattle, 30. Launder BE, Spalding DB (1974). Lectures in
USA. Mathematical Models of Turbulence.
17. Feng X, Li XY (2008). Modelling the kinetics Academic Press, London, UK.
of aggregate breakage using improved 31. Luo C (1997). Distribution of velocities and
breakage kernel. Wat. Sci. Tech. 57(1):151– velocity gradients in mixing and flocculation
Downloaded by [New York University] at 18:48 13 June 2015
238
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
integration of a population balance model and density function and aluminium floc. Water
a CFD model. PS-IWA 2007: Particle Research 13:409–419.
Separation, Toulouse. 47. Thomas, DG (1964). Turbulent disruption of
41. Nopens I, Vanrolleghem PA (2006). flocs in small particle size suspensions.
Comparison of discretization methods to J. AIChE 19(4):517–523.
solve a population balance model of activated 48. Van der Walt JJ (1998). The application of
sludge flocculation including aggregation and computational fluid dynamics in the
breakage. Mathematical and Computer calculation of local G values in hydraulic
Modelling of Dynamical Systems 12(5):441– flocculators. Proceedings of the Biennial
454. Conference of the Water Institute of Southern
42. Paul EL, Atiemo-Obeng VA, Kresta SM Africa, 1998, Cape Town, South Africa.
(2004). Handbook of Industrial Mixing: (Available at: http://www.ewisa.co.za/
Science and Practice. Wiley-Interscience. literature/files/1998%20-%2079.pdf, accessed
43. Prat OP, Ducoste JJ (2007). Simulation of February 2009.)
flocculation in stirred vessels—Lagrangian 49. Versteeg HK, Malalasekera W (1995). An
versus Eulerian. Chem Eng Res. and Des. Introduction to Computational Fluid
85(A2):207–219. Dynamics. The Finite Volume Method.
44. Randolf AD, Larson MA (1971). Theory of Prentice Hall.
Downloaded by [New York University] at 18:48 13 June 2015
Particulate Processes. Academic Press, New 50. Yeung AKC, Gibbs A, Pelton RP (1997).
York. Effect of Shear on the Strength of Polymer-
45. Rodi W (1993). Turbulence Models and Their Induced Flocs. Journal of Colloid and
Application in Hydraulics. 3rd Ed., IAHR. Interface Science 196:113–115.
46. Tambo N, Wanatabe Y (1979). Physical
aspects of flocculation—I The flocculation
APPENDIX 1
∂ ∂ ∂ ⎡⎛ μt ⎞ ∂k ⎤
( ρk ) + ( ρkui ) = ⎢⎜⎜ μ + ⎟ ⎥ + Gk − ρε (A1)
∂t ∂xi ∂x j ⎣⎢⎝ σ k ⎟⎠ ∂x j ⎦⎥
∂ ∂ ∂ ⎡⎛ μt ⎞ ∂ε ⎤ ε ε2
( ρε ) + ( ρεui ) = ⎢⎜⎜ μ + ⎟⎟ ⎥ + C1 Gk − C2 ρ (A2)
∂t ∂xi ∂x j ⎣⎢⎝ σ ε ⎠ ∂x j ⎦⎥ k k
where the turbulent viscosity, μt , is calculated according to
k2
μt = ρCμ (A3)
ε
and the generation of turbulence kinetic energy due to mean velocity gradients, Gk ,
⎛ ∂u ∂u ⎞ ∂u
Gk = μt ⎜ i + j ⎟ j (A4)
⎜ ∂x ⎟
⎝ j ∂xi ⎠ ∂xi
σ k and σ ε are the effective Prandtl numbers for k and ε, respectively.
The empirical model constant values are generally accepted as C μ = 0.09 , C1 = 1.44, C2 = 1.92, σk = 1.0 and
σε = 1.3.
∂ ∂ ∂ ⎡ ∂k ⎤
( ρk ) + ( ρkui ) = ⎢(α k μeff ) ⎥ + Gk − ρε (A5)
∂t ∂xi ∂x j ⎣⎢ ∂x j ⎦⎥
239
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
∂ ∂ ∂ ⎡ ∂ε ⎤ ε ε2
( ρε ) + ( ρεui ) = ⎢(α ε μeff ) ⎥ + C1ε Gk − C2 ρ − Rε (A6)
∂t ∂xi ∂x j ⎢⎣ ∂x j ⎥⎦ k k
μeff = μ + μt
α k and α ε are the inverse effective Prandtl numbers for k and ε, respectively.
∂ ∂ ∂ ⎡⎛ μt ⎞ ∂k ⎤
( ρk ) + ( ρku j ) = ⎢⎜⎜ μ + ⎟ ⎥ + Gk − ρε + S k (A7)
∂t ∂x j ∂x j ⎣⎢⎝ σ k ⎟⎠ ∂x j ⎦⎥
∂ ∂ ∂ ⎡⎛ μt ⎞ ∂ε ⎤ ε2 ε
( ρε ) + ( ρεu j ) = ⎢⎜⎜ μ + ⎟⎟ ⎥ + ρC1 Sε − ρC 2 + C1ε C3 Gb + Sε (A8)
∂t ∂x j ∂x j ⎢⎣⎝ σε ⎠ ∂x j ⎥⎦ k + νε k
C2 is a constant. σk and σε are the turbulent Prandtl numbers for k and ε, respectively. Sk and Sε are user-
defined source terms. The empirical model constant values are generally accepted as C1ε = 1.44,
C1 = max ⎡0.43, η ⎤ , C2 = 1.9, σk = 1.0 and σε = 1.2.
Downloaded by [New York University] at 18:48 13 June 2015
⎢ η + 5 ⎥⎦
⎣
k-ω Model
∂ ∂ ∂ ⎡ ∂k ⎤
( ρk ) + ( ρkui ) = ⎢Γk ⎥ + Gk − Yk + S k (A9)
∂t ∂xi ∂x j ⎣⎢ ∂x j ⎥⎦
∂ ∂ ∂ ⎡ ∂ω ⎤
( ρω ) + ( ρωui ) = ⎢Γϖ ⎥ + Gω − Yω + Sω (A10)
∂t ∂xi ∂x j ⎢⎣ ∂x j ⎥⎦
Gk represents the generation of turbulence kinetic energy due to mean velocity gradients. Gω represents the
generation of ω . Yk and Yω represent the dissipation of k and ω due to turbulence. Sk and Sε are user-defined
source terms.
∂ ∂ ∂ ⎡ ∂k ⎤
( ρk ) + ( ρkui ) = ⎢Γk ⎥ + Gk − Yk + S k (A13)
∂t ∂xi ∂x j ⎣⎢ ∂x j ⎥⎦
∂ ∂ ∂ ⎡ ∂ω ⎤
( ρω ) + ( ρωui ) = ⎢Γϖ ⎥ + Gω − Yω + Dω + Sω (A14)
∂t ∂xi ∂x j ⎢⎣ ∂x j ⎥⎦
240
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)
Considering an incompressible flow without body forces and ignoring transport due to rotation, the exact
equation for the transport of the Reynolds stresses (= ρ u i u j ) is
∂
∂t
( ρ ui′u′j ) +
∂
∂xk
( ρuk ui′u′j ) = −
∂
∂xk
[ ]
ρ ui′u′j u′k + p(δ kj ui′ + δ ik u′j ) +
∂
∂xk
⎡ ∂
⎢μ
⎤
(ui′u′j )⎥
⎣ ∂xk ⎦
local time convection turbulent diffusion molecular diffusion
derivative
⎛ ∂u ∂u ⎞ ⎛ ∂u′ ∂u′ ⎞ ⎛ ∂u′ ∂u′j ⎞
− ⎜⎜ ρ ui′u′k j + u′j uk′ i ⎟⎟ + p⎜ i + j ⎟ − 2μ ⎜⎜ i ⎟⎟
∂xk ∂xk ⎜ ∂x ∂x ⎟ ∂x ∂x
⎝ ⎠ ⎝ j i ⎠ ⎝ k k ⎠
stress production pressure strain dissipation
Downloaded by [New York University] at 18:48 13 June 2015
APPENDIX 2
α2 α3
C D = α1 + +
Re Re 2
Re α1 α2 α3
< 0.1 0 24 0
0.1–1.0 3.69 22.73 0.0903
1.0–10 1.222 29.1667 -3.8889
10–100 0.6167 46.5 -116.67
100–1,000 0.3644 98.33 -2778
1,000–5,000 0.357 148.62 -4.75 x 104
5,000–10,000 0.46 -490.546 57.87 x 104
10,000–50,000 0.5191 -1662.5 5.4167 x 106
241