Environ. Sci. Technol.

2010, 44, 89898995

Computational Fluid Dynamics

Investigation of Turbulence Models
for Non-Newtonian Fluid Flow in
Anaerobic Digesters
BINXIN WU
Philadelphia Mixing Solutions Ltd., Palmyra,
Pennsylvania 17078, United States

Received March 31, 2010. Revised manuscript received

October 17, 2010. Accepted October 18, 2010.

In this paper, 12 turbulence models for single-phase nonNewtonian fluid flow in a pipe are evaluated by comparing
the frictional pressure drops obtained from computational fluid
dynamics (CFD) with those from three friction factor correlations.
The turbulence models studied are (1) three high-Reynolds-number
k- models, (2) six low-Reynolds-number k- models, (3)
two k- models, and (4) the Reynolds stress model. The simulation
results indicate that the Chang-Hsieh-Chen version of the
low-Reynolds-number k- model performs better than the other
models in predicting the frictional pressure drops while the
standard k- model has an acceptable accuracy and a low
computing cost. In the model applications, CFD simulation
of mixing in a full-scale anaerobic digester with pumped circulation
is performed to propose an improvement in the effective
mixing standards recommended by the U.S. EPA based on the
effect of rheology on the flow fields. Characterization of the
velocity gradient is conducted to quantify the growth or breakage
of an assumed floc size. Placement of two discharge nozzles
in the digester is analyzed to show that spacing two nozzles 180
apart with each one discharging at an angle of 45 off the
wall is the most efficient. Moreover, the similarity rules of geometry
and mixing energy are checked for scaling up the digester.

1. Introduction
Anaerobic digestion is a biological process that treats
biodegradable materials such as agricultural manure and
slurry, food waste, and sewage sludge to produce a methanerich biogas. It has long been recognized that effective mixing
is a critical physical operation in achieving optimum process
performance for anaerobic digestion. Under most circumstances the mixing flow in anaerobic digesters is turbulent,
and the accuracy of a flow prediction is dependent on the
turbulence model being used because it has a pronounced
effect on the flow fields. Also, a non-Newtonian fluid such
as municipal wastewater sludge or liquid manure may exhibit
a flow behavior completely different from that of a Newtonian
fluid. Although numerous turbulence models are available
in the literature, there are no quantitative guidelines for
choosing appropriate turbulence models that characterize
non-Newtonian fluid flow in anaerobic digesters.
The performance of a turbulence model can be evaluated
by comparing the frictional pressure drop (or friction factor)
obtained from computational fluid dynamics (CFD) with that
* Author to whom correspondence should be sent: E-mail:
bwu@philamixers.com; phone: 717-832-8857; fax: 717-832-1740.
10.1021/es1010016

2010 American Chemical Society

Published on Web 11/03/2010

from a correlation analysis or an experimental measurement.

Metzner and Reed (1) introduced the concept of a generalized
Reynolds number for power-law fluids that relates to the
consistency coefficient and the power-law index. Subsequently, Dodge and Metzner (2) proposed an implicit
equation for predicting the friction factor of viscous fluids
in turbulent pipe flow. Later, several researchers (3-5)
developed explicit correlations for calculating the friction
factor. These correlations have been widely used in the study
of non-Newtonian fluid flow. Hartnett and Kostic (6) assessed
five friction factor correlations for the prediction of turbulent
pressure drop of purely viscous power-law fluids in rectangular channels and circular tubes and concluded that (1)
the Dodge-Metzner correlation gives the best agreement
with the experimental results over the entire range of available
power-law values (0.241 e n e 0.525), (2) the yoo correlation
and the Irvine correlation are within (10% of the experimental data for n g 0.3, and (3) the Tam-Tiu correlation
underpredicts experimental pipe flow data for n values
between 0.2 and 0.5. Pinho and Whitelaw (7) investigated
non-Newtonian fluid flow in a horizontal pipe and used a
laser Doppler velocimeter to measure the mean axial velocity.
The results quantified the delay in transition from laminar
to turbulent flow caused by shear-thinning, the suppression
of turbulent fluctuations, and the drag reduction at high
Reynolds numbers. Hemeida (8) presented an equation for
estimating the thickness of the laminar sublayer in turbulent
pipe flow of pseudoplastic fluids and validated the friction
factors against the reported correlations and field measurements. It was found that the hydraulically smooth pipes can
be used to determine the pressure loss in the tested pipeline
because the thickness of the laminar sublayer was greater
than the average roughness height. Escudier and Presti (9)
measured the mean velocity and velocity fluctuation levels
using a laser Doppler anemometer for fully developed pipe
flow of an aqueous solution of laponite and demonstrated
that under turbulent flow conditions the logarithmic law
was shifted upward by a small amount consistent with
progressively reduced levels of drag reduction as the Reynolds
number increased. Malin (10) performed numerical computations of fully developed turbulent flow of power-law
fluids in smooth tubes and tested a modified version of the
Lam-Bremhorst k- model. The predicted friction factor
and mean velocity profile closely agreed with the measured
data over a wide range of values for the power-law index and
the generalized Reynolds number. Chilton and Stainsby (11)
developed a numerical model for the simulation of laminar
and turbulent flow of Herschel-Bulkley fluids and used the
Launder-Sharma version of the low-Reynolds-number k-
model to characterize the turbulent flow. Generally, the
predicted pressure losses were lower than the experimental
values within 15% error. Rudman et al. (12) conducted a
direct numerical simulation (DNS) of the weakly turbulent
flow of power-law and Herschel-Bulkley fluids in a small
pipe and obtained very good agreement between the DNS
results and experimental measurements. The DNS is a useful
tool for performing fundamental research in turbulence
because it numerically solves the Navier-Stokes equations
without any turbulence model. However, the computational
resources required by a DNS would exceed the capacity of
the most powerful computers currently available (13). Wu
and Chen (14) applied the standard k- model to simulate
carboxymethyl cellulose (CMC) power-law fluid flow in a
horizontal pipe, verified the axial velocities with the experimental data from Pinho and Whitelaw (7), and claimed that the
overall prediction accuracies are acceptable. Wu (15) studied
VOL. 44, NO. 23, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

8989

gas and non-Newtonian fluid two-phase flow in a pipe at six

total solids (TS) concentration levels and reported that the shear
stress transport (SST) k- model could be used to simulate gas
mixing in anaerobic digesters when TS e 5.4%.
Even though both the CMC and manure slurry are nonNewtonian fluids, one question is whether a turbulence
model that is applicable to one non-Newtonian fluid will
apply to the others. The other question is whether the SST
k- model should be recommended when solving singlephase flow. Therefore, the primary goal of this research was
to evaluate 12 turbulence models on the basis of the criteria
of frictional pressure drops in a horizontal pipe and propose
the appropriate models that predict the single-phase nonNewtonian fluid flow at different TS levels. In addition,
simulations of mixing by pumped circulation in a full-scale
anaerobic digester were performed to investigate the effect
of non-Newtonian rheology on the mixing energy, velocity
gradient that impacts flocculation, optimum placement of
discharge nozzles, and scaleup mechanism.

2. Development of the Numerical Model

2.1. Assumptions. (a) The critical Reynolds number for the
laminar-turbulent flow transition in the pipe or the digester
is 2000. (b) Liquid manure behaves as a pseudoplastic fluid for
TS ranging from 2.5% to 12.1% at temperatures between 20
and 60 C (16). (c) The digestion temperature is constant at 35
C. (d) The model is single phase, in which the effects of any
suspended sediments entrained within the manure and phase
interaction due to the naturally produced biogas are negligible.
2.2. Reynolds-averaged Navier-Stokes (RANS) Turbulence Models. On the basis of the closure of the Reynolds
stress term in the momentum equations, the RANS turbulence models are briefly described as follows.
2.2.1. k- Models. The k- models consist of the highReynolds-number k- (high-Re k-) and low-Reynoldsnumber k- (low-Re k-) models. For the standard k-
model, the equations for k and are expressed as

(Fk) + (Fv
bk) ) [kk] + Gk - Yk + Sk
t

(1)

(F) + (Fv
b) ) [] + G - Y + S
t

(2)

where k and are the turbulence kinetic energy and its

dissipation rate, k and are effective diffusivity of k and
, Gk and G are the generation of k and , Yk and Y are the
dissipation of k and , and Sk and S are source terms.
In the low-Re k- models, the turbulent viscosity for the
standard k- model (t ) FC k2/) and the transport equation
for are modified as
k2
t ) FCf

where f, f1, and f2 are the dumping functions that account

for the effects of viscous damping and kinematic blocking.
2.2.2. k- Models. Similarly to the equation, the
transport equation for is expressed as

(6)

where F ui uj, Pi j, i j, Di j, and i j represent the Reynolds

stress, the stress-production, the pressure-strain, the diffusion, and the dissipation terms, respectively.
2.3. Non-Newtonian Rheology. The viscosity of manure
slurry () and density (F) are expressed as (14)
) K n-1eT0/T

(7)

F ) 0.0367(TS)3 - 2.38(TS)2 + 14.6(TS) + 1000

(8)

where K is the consistency coefficient, is the shear rate, n

is the power-law index, T0 is the reference temperature, T is
the digestion temperature, and TS is the weight percentage
of total solids in liquid manure.
2.4. Calculation of Frictional Pressure Drop. The frictional pressure drop per unit length in a pipe with a CFD
simulation can be calculated by
Pf_CFD )

( )

w
/L
A

(9)

where w is the viscous force exerted by the wall on the fluid,

A is the pipe cross-sectional area, and L is the pipe length.
2.5. CFD Simulation. The commercial CFD software
packages Gambit 2.4.6 and Fluent 12.0 were used for meshing
and solving the governing equations (17).

3. Theory of Correlation Analysis

The Reynolds number for water flow in a pipe is defined as
Re )

FuD

(10)

where u is the average velocity, D is the pipe diameter, and

is the dynamic viscosity.
The Fanning friction factor for turbulent flow is expressed as
f )

0.079
(Re)0.25

(11)

The generalized Reynolds number for non-Newtonian

fluid flow in a pipe is defined as (1)
Reg )

Fu2-nDn
k(0.75 + 0.25/n)n8n-1

(12)

Under turbulent flow conditions, several researchers have

proposed the following correlations for predicting the friction
factor.
(a) Dodge and Metzner (2):
1
0.4
4
) 0.75 log10(Regf (2-n)/2) - 1.2
n
n
f

(13)

(b) Yoo (3):

f ) 0.079n0.675Reg-0.25

(14)

2(2n/77n)1/(3n+1)
(0.75 + 0.25/n)nReg1/(3n+1)

(15)

(5)

where is the specific dissipation rate, is the effective

diffusivity of , G is the generation of , Y is the dissipation
of , and S is the source term.
8990

D
(Fuiuj) ) Pij + ij + Dij + ij
Dt

(3)

(F) + (Fv
b) ) [(t/)] + (f1C1Gk - f2C2F)
t
k
(4)

(F) + (Fv
b) ) [] + G - Y + S
t

2.2.3. Reynolds Stress Model. In the Reynolds stress model

(RSM), the Reynolds stresses are directly computed. The
second-order closure of the RSM can be expressed in a simple
form as

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 23, 2010

(c) Irvine (4):

f )

TABLE 1. Frictional Pressure Drop from Correlation Analysisa

K (Pa s )
n
min (Pa s)
max (Pa s)
F (kg/m3)
u (m/s)
Reg
f_DM
Pf_Corr (Pa/m)
f_Yoo
Pf_Corr (Pa/m)
f_Irv
Pf_Corr (Pa/m)
n

TS ) 2.5%

TS ) 5.4%

TS ) 7.5%

TS ) 9.1%

TS ) 12.1%

0.042
0.71
0.006
0.008
1000.36
2
6178
7.13 10-3
2854
7.07 10-3
2830
6.11 10-3
2446

0.192
0.562
0.01
0.03
1000.78
3
6316
6.1 10-3
5494
6.01 10-3
5410
4.65 10-3
4193

0.525
0.533
0.03
0.17
1001
4
4304
6.69 10-3
10715
6.38 10-3
10216
5.08 10-3
8131

1.052
0.467
0.07
0.29
1001.31
5
4865
5.9 10-3
14769
5.66 10-3
14163
4.19 10-3
10493

5.885
0.367
0.25
2.93
1001.73
6
2470
6.6 10-3
23801
5.7 10-3
20542
4.45 10-3
16046

The subscripts DM, Yoo, and Irv represent Dodge-Metzner, Irvine, and Yoo correlations, respectively.

TABLE 2. Frictional Pressure Drop from CFDa

TS ) 0%

SKE
RNG
RKE
SKO
SST
RSM
Abid
L-B
L-S
Y-S
A-K-N
C-H-C

TS ) 2.5%

TS ) 5.4%

Pf_CFD (Pa/m)

(%)

Pf_CFD (Pa/m)

_DM (%)

_Yoo (%)

_Irv (%)

Pf_CFD (Pa/m)

_DM (%)

_Yoo (%)

_Irv (%)

2651
2583
2657
2857
2280
2872
2705
2821
2482
2148
2678
2706

19
16
19
28
2.3
29
21
27
11
4
20
21

4799
4801
4799
3675
3981
3344
3252
3901
6073
3479
3409
2615

68
68
68
29
39
17
14
37
113
22
20
8

70
70
70
30
41
18
15
38
115
23
20
7

96
96
96
50
63
37
33
60
148
42
39
7

10908
10924
10908
8519
8997
7301
7525
8631
13992
8025
7904
5965

99
99
99
55
64
33
37
57
155
46
44
9

102
102
102
58
66
35
39
60
159
48
46
10

160
160
160
103
115
74
80
106
234
91
89
42

a
The subscripts DM, Yoo, and Irv represent Dodge-Metzner, Irvine, and Yoo correlations, respectively. SKE ) standard
k-, RNG ) RNG k-, RKE ) realizable k-, SKO ) standard k-, SST ) SST k-, RSM ) Reynolds stress model, L-B )
Lam-Bremhorst, L-S ) Launder-Sharma, Y-S ) Yang-Shih, A-K-N ) Abe-Kondoh-Nagano, and C-H-C )
Chang-Hsieh-Chen.

The frictional pressure drop with a correlation analysis

can be calculated by
Pf_Corr )

2fFu2
D

(16)

Accordingly, an error indicator based on a correlation analysis

can be calculated by
)

Pf_CFD - Pf_Corr
100
Pf_Corr

(17)

4. Results and Discussion

4.1. Examination of Turbulence Models. A horizontal pipe
(0.02 m in diameter and 0.4 m in length) was used to
examine 12 turbulence models that include (1) three
high-Re k- models (standard k-, RNG (Renormalization
Group) k-, and realizable k-), (2) six low-Re k-
models
(Abid,
Lam-Bremhorst,
Launder-Sharma,
Yang-Shih, Abe-Kondoh-Nagano, and Chang-Hsieh-Chen),
(3) two k- models (standard k- with shear flow
corrections and SST k- with low-Re corrections), and
(4) the RSM. These models have been incorporated into
Fluent CFD software. To achieve mesh resolution near the
pipe wall, a fine mesh having 52 836 cells was used (15).
The enhanced wall treatments were set for the high-Re
k- models and the RSM because this mesh was originally
designed for the low-Re k- models. The inlet velocities
of 2, 2, 3, 4, 5, and 6 m/s were set for TS ) 0%, 2.5%, 5.4%,
7.5%, 9.1%, and 12.1%, respectively, to ensure Re (or Reg)

> 2000. Using eqs 10 and 11, it can be determined that Re

) 40 000, f ) 5.59 10-3, and Pf_Corr ) 2230 Pa/m at TS
) 0%. Table 1 shows Pf_Corr from three different correlations for TS > 0, in which the rheological properties and
densities were taken from Wu and Chen (14). If the criterion
for selecting a turbulence model was set to be e 30%
on the basis of any one correlation analysis (15), all the
turbulence models overpredicted the frictional pressure
drops at TS g 7.5%. Table 2 shows Pf_CFD and at three
TS levels, indicating that (1) all the turbulence models are
applicable to TS ) 0% and (2) at TS ) 2.5% and 5.4% three
priority turbulence models rank as the Chang-Hsieh-Chen
version of low-Re k-, the Abid version of low-Re k-, and
the RSM. If computing time is not an issue, these three
models are recommended.
The prediction discrepancy mainly results from the
inherent limitations of using a model in an inappropriate
application. The high-Re k- models are developed on the
basis of the Boussinesq hypothesis that assumes locally
isotropic turbulence, which might not be true for flows with
heterogeneous turbulence. As an alternative approach to the
Boussinesq hypothesis, the RSM solves transport equations
for each Reynolds stress term along with an additional scaledetermining equation, which is usually superior for situations
where anisotropy of turbulence has a dominant effect on the
mean flows. The low-Re k- models are modified on the
basis of the k- model to enable them to be used at low
Reynolds numbers, which is appropriate for non-Newtonian
fluid flow in this study where the Reynolds number is found
VOL. 44, NO. 23, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

8991

FIGURE 1. Velocity contours and profiles at the pipe exit: (C) coarse mesh of 18 156 cells for the standard k- model and (F) fine
mesh of 52 836 cells for the Chang-Hsieh-Chen version of the low-Re k- model. Note that, in the velocity contours, blue
represents areas that have low velocity while red represents areas of high velocity as indicated in the velocity contour bar, and the
velocity magnitudes in the red area are greater than (or equal to) the maximum value specified in the bar. These representations
apply to all the contour figures in this paper.
center. Two nozzles were located in the same horizontal plane
with 180 spacing ( ) 180) and each one discharging at an
angle of 45 off the wall ( ) 45). The effluent pipe consisted
of two suction ports below the liquid surface, assuming that
the area of each port is equal to that of one nozzle outlet.
Table 4 shows five digester sizes, in which size 1 was used
in all simulations unless noted.
The mixing energy level can be calculated by (15)

FIGURE 2. Schematic diagram of the digester and velocity

contours for MEL ) 5 W/m3.
to be reduced significantly with an increase in TS. The k-
models having low-Reynolds-number effects can be used
with either a fine or a coarse near-wall mesh. Although two
low-Re k- models and the RSM are superior to the other
models when TS > 0%, they are computationally expensive
since a low-Re k- model requires a very fine mesh near the
wall while the RSM solves seven additional transport equations for three-dimensional flow. In comparison with the
SST k- and three high-Re k- models, the standard k-
model has a relatively low error when TS > 0%. To achieve
a compromise between accuracy and computing time, the
flow fields produced by the Chang-Hsieh-Chen version of
the low-Re k- model were set as the reference and a coarse
mesh having 18 156 cells without near-wall mesh resolution
was generated in the pipe to examine the standard k-
model. Figure 1 shows the comparison of flow patterns at
the pipe exit using two turbulence models under two sets of
meshes, from which it can be seen that the core flows are
similar along the pipe center while the flows in the near-wall
region are different because of the mesh density. From an
engineering standpoint, the standard k- model can be used
to simulate single-phase non-Newtonian fluid flow because
of its acceptable accuracy and low computing cost. Hence,
simulations hereafter were conducted using this model.
4.2. Effect of Rheology on the Mixing Energy. As shown
in Figure 2a, a full-scale anaerobic digester constructed from
a cylindrical tank with a conical bottom was used to
investigate mixing by pumped circulation. The digester was
equipped with a jet mixing loop comprised of two identical
discharge nozzles at a low level and one effluent pipe at the
8992

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 23, 2010

MEL ) E/V

(18)

E ) FgHQ

(19)

where E is the power input, V is the working volume of the

digester, H is the hydrostatic head of the liquid, and Q is
the discharging rate.
Given MEL ) 5 W/m3 at TS ) 0% and d ) 0.079 m for the
diameter of each nozzle, vin ) 6.46 m/s for the discharging
velocity of each nozzle can be determined from eqs 18 and
19. Because of insignificant density differences within TS e
5.4%, the discharging velocity calculated from TS ) 0% was
applied to the other two TS levels under the same mixing
energy. To perform quantitative analysis, the mixing intensity
was defined as the average velocity throughout the whole
j and the turbulence intensity was defined as
digester tank (v)
(ui2)1/2/vref, where ui are the turbulent velocity fluctuations
and ref ) 1 m/s. Parts b-d of Figure 2 show the velocity
contours along the central plane from the front view at three
TS levels, from which it can be observed that all the flow
patterns are symmetrical about the effluent pipe and that
the mixing intensity decreases with an increase in TS. Taking
TS ) 0% as an example, Figure 3 shows plots of the velocity
magnitude and turbulence intensity across the tank diameter
at two depths. At z ) 1.2 m the velocity and turbulence
intensity increase from the side wall to the nozzles, reach
their peaks at the nozzle exits, and then decline toward the
tank center, while at z ) 7.2 m relatively strong turbulence
can be observed from both the near-wall and effluent zones.
These predictions were generally consistent with the results
from a similar CFD study in which both inlets and outlets
were set at the tank wall (18).
To characterize the mixing intensity with respect to the
mixing energy, extensive simulation work was done to obtain
j
the v-MEL
function as

c ) 0.0415 TS ) 0%
vj ) c(MEL) c ) 0.0388 TS ) 2.5%
c ) 0.0365 TS ) 5.4%

(20)

As indicated by the U.S. EPA, MEL ) 5-8 W/m3 is recommended for effective mixing in anaerobic digesters by
pumped circulation (19). However, this guideline does not
address the type of anaerobic digester being used and the
fluid being mixed. In this study, it was presumed that MEL
) 5-8 W/m3 could be applied to mix water in a cylindrical
digester, and then vj ) 0.208-0.332 m/s at TS ) 0% could be

predicted on the basis of this assumption. Through inversely

solving eq 20, MEL ) 5.36-8.55 and 5.7-9.1 W/m3 were
proposed to be the ranges of mixing energy levels for acquiring
vj ) 0.208-0.332 m/s at TS ) 2.5% and 5.4%, respectively.
Thus, the rheological properties of a fluid being mixed should
be taken into consideration when using the effective mixing
standards recommended by the U.S. EPA.
4.3. Characterization of the Velocity Gradient. The
average velocity gradient (G) is defined as (20)
G)

E
V

(21)

The local velocity gradient is defined as (20)

GL )

(22)

where is the kinematic viscosity.

The breakup number is defined as (20)
B)

F
J

(23)

where F is the rupture force and J is the aggregate strength.

The floc breaks up when B > 1, while its size is maintained
or increased when B < 1.
The rupture force can be approximated by
F df2

(24)

where is the hydrodynamic stress ( ) GL in the viscous

subrange, while ) F C2( df)2/3 in the inertial subrange)
and df is the floc diameter.
Substituting ) /(*k) in the k- model developed by
Wilcox (21) and ) /F for non-Newtonian fluids in eqs 22
and 23 yields

F*k

(25)

C1F*k/df2
J

(26)

FC2(F*k/)2/3df8/3
J

(27)

GL )

Bvsr )
Bisr )

where * ) 0.09, C1 ) 5/8, C2 ) 0.7, Bvsr is the breakup

number in the viscous subrange, and Bisr is the breakup
number in the inertial subrange.
Equations 25-27 can be solved using three custom field
functions in the Fluent 12.0 software. Figure 4 shows the
contours of the velocity gradient for MEL ) 5 W/m3 at three
TS levels, from which it can be shown that the maximum GL
occurs at the nozzle exits while the minimum GL occurs at
the tank bottom. Theoretically, eq 21 does not apply to a
non-Newtonian fluid in which the viscosity () varies with
the shear rate ( ) unless an equivalent for this fluid is
available. Given MEL ) 5 W/m3 and ) 0.001 Pa s at TS )
0%, G ) 70.7 s-1 can be acquired by solving eq 21. However,
checking the GL distributions shows that the percentages of
volume with respect to GL are 84.5%, 7.9%, and 7.6% for GL
< 10 s-1, 10 s-1 e GL < 70 s-1, and GL g 70 s-1, respectively.
The average GL can be predicted as 3.44 s-1 via the integration
of GL over the whole computational domain, demonstrating
that only a small percentage of volume has a GL over 70 s-1.
Clearly, the velocity gradients in the digester are nonuniform.
These results were in qualitative agreement with the findings
from related research (22). Therefore, the concept for the
local velocity gradient is more reasonable than the traditional
definition of the G value that characterizes a complex flow
field with a single number.
To investigate the effect of turbulence on flocculation,
two points in the digestion tank were specified as P1 (5, 0,
1) and P2 (5.5, -0.5, 1.2). Note that the Cartesian coordinate
system was fixed at the bottom center with metric units.
Given df ) 100 m and J ) 5 10 -9 N for a floc (20), Bvsr

FIGURE 3. Mean velocity and turbulence intensity for MEL ) 5 W/m3 at TS ) 0%.

FIGURE 4. Contours of the velocity gradient for MEL ) 5 W/m3.

VOL. 44, NO. 23, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

8993

TABLE 3. Mixing Intensity versus Orientation of the Nozzles

v (m/s) versus at ) 45
TS (%)
0
2.5
5.4

45

90

135

180

v (m/s) versus at ) 180

15

30

60

TABLE 4. Design Parameters and Mixing Intensity for Five

Digestersa

75

0.200 0.204 0.210 0.208 0.209 0.209 0.197 0.150

0.187 0.189 0.199 0.195 0.201 0.194 0.188 0.145
0.175 0.177 0.180 0.181 0.179 0.180 0.180 0.138

and Bisr at P1 can be predicted as 0.13 and 0.003, respectively,

indicating that mixing at P1, which is near the conical wall,
favors the floc growth. Similary, Bvsr ) 13.86 and Bisr ) 1.71
can be achieved at P2, indicating that mixing at P2, which is
near the right discharge nozzle, causes the floc breakage. In
physics either Bvsr or Bisr at a point is true depending on into
which subrange this point falls. It should be pointed out that
rheological and turbulent effects on solids flocculation and
breakup are among the most challenging topics in anaerobic
digestion research, which requires experimental work as an
aid in further investigation.
4.4. Optimum Placement of Discharge Nozzles. To test
the effect of the spacing between two nozzles () on the
mixing, simulations of flow fields were carried out at three
TS levels for ) 45, 90, and 135, in which the discharging
angle for each nozzle () was kept at ) 45. Similarly,
sensitivity analysis of the discharging angle ranging from 15
to 75 while keeping ) 180 was conducted. Table 3 shows
that (1) the mixing intensity changes insignificantly with an
increase of from 45 to 180 and (2) the mixing intensity
remains almost unchanged over the range of ) 15-30
while it decreases with an increase of from 60 to 75. Taking
TS ) 5.4% as an example, Figure 5 shows that the flow patterns
for ) 180 are more uniform than those for < 180 and
that an increase in results in moving the intensive mixing
from the near-wall region to the central region. In terms of
the mixing intensity and uniformity, mixing design by )
180 and ) 45 for two discharge nozzles is recommended.

V (m )
DT (m)
H (m)
h (m)
C (m)
de (m)
d (m)
v at TS ) 0% (m/s)
v at TS ) 2.5% (m/s)
v at TS ) 5.4% (m/s)

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 23, 2010

size 2

size 3

size 4

size 5

792
12
6.7
0.9
1.2
0.158
0.079
0.208
0.195
0.181

1188
13.74
7.672
1.031
1.374
0.182
0.09
0.208
0.196
0.182

1584
15.12
8.442
1.134
1.512
0.199
0.099
0.208
0.196
0.183

1980
16.29
9.095
1.222
1.629
0.214
0.107
0.208
0.197
0.184

2376
17.31
9.665
1.298
1.731
0.228
0.114
0.208
0.198
0.185

a
V ) working volume of the digester, DT ) tank
diameter, H ) cylindrical height, h ) conical height, C )
clearance of each nozzle, de ) diameter of the effluent
pipe, and d ) diameter of each nozzle.

4.5. Scaleup Mechanism. The purpose of scaleup is to

design a large mixing system that will achieve the same mixing
characteristics as in a small one. In this study, the scaleup
with geometric and mixing energy similarities was examined.
The working volumes for four digesters were assumed to be
V2 ) 1.5V1, V3 ) 2V1, V4 ) 2.5V1, and V5 ) 3V1, in which the
subscripts 1, 2, 3, 4, and 5 represent size 1, size 2, size 3, size
4, and size 5, respectively. If the tank diameter (DT) was chosen
as the dimension basis, then H/DT, h/DT, C/DT, de/DT, and
d/DT in each digester were kept constant. Under the same
mixing energy level, the discharging velocity of each nozzle
at any digester size can be predicted as vin ) 6.46 m/s by
solving eqs 18 and 19. As shown in the last three rows of
Table 4, the mixing intensity stays unchanged at TS ) 0%
while it increases slightly with an increase in the digester
size at TS ) 2.5% and 5.4%. If the numerical errors are taken
into account, the flow fields at any digester size remain much

FIGURE 5. Velocity contours versus placement of discharge nozzles at TS ) 5.4%.

8994

size 1

the same with a constant mixing energy level and a specified

TS level. Therefore, the design of scaleup digesters with
pumped circulation should follow the similarity rules of
geometry and mixing energy.

Literature Cited
(1) Metzner, A. B.; Reed, J. C. Flow of non-Newtonian fluidssCorrelation of laminar, transition and turbulent-flow regions. AIChE
J. 1955, 1, 434440.
(2) Dodge, D. W.; Metzner, A. B. Turbulent flow of non-Newtonian
systems. AIChE J. 1959, 5, 189204.
(3) Yoo, S. S. Heat transfer and friction factors for non-Newtonian
fluids in circular tubes. Ph.D. Dissertation, University of Illinois,
Chicago, IL, 1974.
(4) Irvine, T. F., Jr. A generalized blasius equation for power law
liquids. Chem. Eng. Commun. 1988, 65, 3947.
(5) Tam, K. C.; Tiu, C. A general correlation of purely viscous nonNewtonian fluids flowing in ducts of arbitrary cross-section.
Can. J. Chem. Eng. 1988, 66 (4), 542549.
(6) Hartnett, J. P.; Kostic, M. Turbulent prediction factor correlations
for power law fluids in circular and non-circular channels. Int.
Commun. Heat Mass Transfer 1990, 17 (1), 5965.
(7) Pinho, F. T.; Whitelaw, J. H. Flow of non-Newtonian fluids in
a pipe. J. Non-Newtonian Fluid Mech. 1990, 34, 129144.
(8) Hemeida, A. M. Effect of wall roughness in turbulent pipe flow
of a pseudo plastic crude oil: An evaluation of pipeline field
data. J. Pet. Sci. Eng. 1993, 10, 163170.
(9) Escudier, M. P.; Presti, P. Pipe flow of a thixotropic liquid. J.
Non-Newtonian Fluid Mech. 1996, 62, 291306.
(10) Malin, M. R. Turbulent pipe flow of power-law fluids. Int.
Commun. Heat Mass Transfer 1997, 24 (7), 977988.

(11) Chilton, R. A.; Stainsby, R. Pressure loss equations for laminar

and turbulent non-Newtonian pipe flow. J. Hydraul. Eng. 1998,
124 (5), 522529.
(12) Rudman, M.; Blackburn, H. M.; Graham, L. J. W.; Pullum, L.
Turbulent pipe flow of shear-thinning fluids. J. Non-Newtonian
Fluid Mech. 2004, 118, 3348.
(13) Jiang, X.; Lai, C. Numerical Techniques for Direct and LargeEddy Simulations; CRC Press: London, U.K., 2009.
(14) Wu, B.; Chen, S. CFD simulation of non-Newtonian fluid flow in
anaerobic digesters. Biotechnol. Bioeng. 2008, 99 (3), 700711.
(15) Wu, B. CFD simulation of gas and non-Newtonian fluid twophase flow in anaerobic digesters. Water Res. 2010, 44 (13),
38613874.
(16) Achkari-Begdouri, A.; Goodrich, P. R. Rheological properties of
Moroccan dairy cattle manure. Bioresour. Technol. 1992, 40,
149156.
(17) ANSYS-Fluent Inc. Fluent 12.0; Lebanon, NH, 2008.
(18) Meroney, R. N. CFD simulation of mechanical draft tube mixing
in anaerobic digester tanks. Water Res. 2009, 43 (4), 10401050.
(19) U.S. Environmental Protection Agency. Process Design Manual
for Sludge Treatment and Disposal; EPA/625/1-79/011; EPA:
Cincinnati, OH, 1979.
(20) Coufort, C.; Bouyer, D.; Line, A. Flocculation related to local
hydrodynamics in a Taylor-Couette reactor and in a jar. Chem.
Eng. Sci. 2005, 60, 21792192.
(21) Wilcox, D. C. Reassessment of the scale-determining equations
for advanced turbulence models. AIAA J. 1988, 26 (11), 1299
1310.
(22) Bridgeman, J.; Jefferson, B.; Parsons, S. A. The development
and application of CFD models for water treatment flocculators.
Adv. Eng. Software 2010, 41, 99109.

ES1010016

VOL. 44, NO. 23, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

8995