1 s2.0 S0009250908004727 Main

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

Chemical Engineering Science 64 (2009) 126 -- 143

Contents lists available at ScienceDirect

Chemical Engineering Science


journal homepage: w w w . e l s e v i e r . c o m / l o c a t e / c e s

Use of angle resolved PIV to estimate local specific energy dissipation rates for up- and
down-pumping pitched blade agitators in a stirred tank
A. Gabriele, A.W. Nienow, M.J.H. Simmons ∗
School of Chemical Engineering, University of Birmingham, Edgbaston, Birmingham B15 2TT, UK

A R T I C L E I N F O A B S T R A C T

Article history: Up-pumping pitched blade turbines (and similar impellers) have recently been shown to be particularly
Received 3 April 2008 effective for achieving a variety of mixing duties. Here, their turbulent flow characteristics are analysed by
Received in revised form 17 July 2008 angle-resolved particle image velocimetry (PIV) for the first time and compared with their down-pumping
Accepted 3 September 2008
equivalent, the usual time-averaged parameters also being determined for each. The work was conducted
Available online 7 October 2008
in 0.15 m diameter vessel (T) with a 45◦ impeller of diameter D ( = 0.45T) in water. The angle-resolved PIV
Keywords: enables a number of novel features to be identified. Firstly, the two pumping directions are shown to give
Particle image velocimetry (PIV) very different vortex structures, even though the flow numbers, Fl, are the same ( = 0.79). In addition,
Stirred tank the `spottiness' of the normalized kinetic energy along a radius as the trailing vortex moved away from
Pitched blade turbine each impeller can be identified, which is not shown from time-averaged data. Often, the most important
Turbulence parameter for processing is the local normalized specific energy dissipation rate, ∗T and this is estimated
Local specific energy dissipation using three methodologies: by measurement of the components of the stress tensor directly, (∗T )DE ; by
Power
dimensional analysis, (∗T )DA , with measured integral length scales (ILS); and by the Smagorinsky closure
method, (∗T )SGS , to model unresolved scales (with a Smagorinsky constant used in the literature on stirred
vessels). Again, only the angle-resolved results show the spottiness of ∗Tmax and also higher values than the
time-averaged. Differences in the values obtained by the three methods are discussed and compared with
the existing literature. Most importantly, for the first time, the power input in the PIV-interrogated region
is calculated from the three methods and compared to the input based on the impeller torque. Both DA
and SGS methods are shown to overestimate the true power by a factor of 5 and 2, respectively, whilst the
DE method provided a significant underestimate (1/5th) due to the limitation of the resolved length scales.
The SGS method shows the greatest promise and by changing the value of the Smagorinsky constant in
accordance with recent recommendations, good agreement is obtained. Nevertheless, it is concluded that
there is still a need for improved methods for determining the important mixing parameter, ∗Tmax .
© 2008 Elsevier Ltd. All rights reserved.

1. Introduction The characteristics of turbulent flow in a stirred tank are highly


complex and substantial research effort has been undertaken to link
Batch and continuously fed, baffled stirred vessels are ubiquitous flow parameters to mixing characteristics. Zhou and Kresta (1998)
for mixing operations within the process industries, where they have stated that prediction of the local specific energy dissipation rate,
been used in applications involving liquid blending, chemical reac- T , is, in many cases, the key to successful process modelling. For ex-
tion, gas, immiscible liquid and solid dispersion in liquids as well ample, for fast competitive reactions in the liquid phase, it has been
as to enhance heat transfer. Whether the process involves single or shown that feeding reactants into regions of high T leads to higher
multiphase flow, the flow in the liquid phase is of critical impor- selectivity and yield with a consequent reduction in by-product
tance. In many cases, especially in larger scale vessels, the flow is formation and improved process economics. In multiphase systems,
turbulent and there have been a large number of studies that have mechanisms for the break-up and coalescence of drops and bub-
attempted to characterize both the fluid motion and the turbulent bles include this parameter and, in particular, its maximum value,
mixing for the purpose of performance optimization. Tmax . However, direct measurement of T is problematic since it
requires a technique that can resolve across all the turbulent length
scales in the flow, which is particularly important for calculation of
Tmax near the impeller blade. In addition, this region is also where

Corresponding author. Tel.: +44 121 414 5371; fax: +44 121 414 5324. the determination of T is most difficult because firstly, the value
E-mail address: m.j.simmons@bham.ac.uk (M.J.H. Simmons). depends very significantly on the precise location relative to the

0009-2509/$ - see front matter © 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2008.09.018
A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143 127

SGS (T )ta /N3 D2


impeller blade and secondly, of the complex and periodic nature of
the flow associated with the successive passages of the blades.
In recent years, laser-based flow visualization techniques have

1.52
1.38
0.51
been used extensively, with most studies being based on vessels

6
equipped with radial flow impellers such as the Rushton disc turbine

DE (T )ta /N3 D2


(RDT). Since time-averaged measurements do not take into account
the periodicity of the flow due to the blade passage, several studies
have performed measurements resolved by the azimuthal angle be-

0.091
0.086
0.25
tween the blades to obtain the turbulence intensities in the impeller

3.1

6
Comparison of experimental conditions, approaches to evaluating and estimates of maximum values of time-averaged local specific energy dissipation rate from different workers and this work.
region (e.g., Lee and Yianneskis, 1998). For example, laser Doppler

A = 1; u = u x  = x
anemometry (LDA) has been used to obtain angle-resolved instan-

 = (x +y +x y )1/2


A = 0.85; u = k1/2 ;
taneous velocities around RDTs by Yianneskis and Whitelaw (1993),

A = 1; u = k1/2
Ducci and Yianneskis (2006) and Wu and Patterson (1989), the latter

A = 1; u = uy ,
presenting an equation for calculation of specific energy dissipation

 = D/10

 = D/10
rate. In addition, Assirelli et al. (2005) demonstrated the importance
of differentiating between time-averaged and angle resolved values

8.14

3.33

3.31
2.26
22
of Tmax for fast chemical reactions. Although LDA offers distinct

A = 0.85; u = k1/2 ;
advantages in terms of temporal resolution of the order of several
thousand Hertz, particle image velocimetry (PIV), a whole field tech-

A = 1; u = k1/2

A = 1; u = ux ,

 = (3y 2 )1/2
nique, has gained in popularity due to its comparative ease of ap-

 = Vx ac *

 = D/10
plication. Using PIV, Sharp and Adrian (2001) calculated T around a

23.26
RDT based on angle-resolved measurements.

2.39

2.51
2.03

In the work of Wu and Patterson (1989) and Kresta and Wood (1993a), the value of ac was evaluated from autocorrelation and the Taylor hypothesis.
Some LDA studies have also been made for vessels equipped with
axial impellers such as down-pumping pitched blade turbines (PBT)

A = 0.85; u = k1/2  = (3x 2 )1/2


A = 0.85; u = k ;  = Uc *ac
by Schafer et al. (1998) and Kresta and Wood (1993a). More recent
studies have also used LDA to obtain local time-averaged T values

A = 1; u = k1/2 ;  = D/10
A = 1; u = ux ;  = D/10
(Zhou and Kresta, 1996) whilst Sheng et al. (2000) used time aver-

A = 0.85; u = k1/2 ;

A = 0.85; u = k1/2 ;
aged PIV measurements as a basis for evaluation of different meth-
1,2

ods for estimation of T for a 4-blade down-pumping PBT. Recently, A = 1; u = k1/2 ;


DA (T )ta /N3 D2

 = (3x 2 )1/
Khan et al. (2004, 2006) obtained both time-averaged and angle-


 = Uc ac *


resolved PIV velocity measurements on a 4-blade, down-pumping

 = y
PBT, the former work employing an ingenious `multi-block' method
6.98

2.47

4.22
1.96
0.42

3.20
23

25
to improve the resolution of the measurements. Khan et al. (2006)
considered a kinetic energy balance around the impeller to deter-
N3 D2 (m2 s−3 )

mine the fraction of total energy dissipated in the impeller region.


Many important features of these studies are summarized in Table 1.

0.0119
1.72

1.34

52.2

1.17
0.32

0.57
0.01

470
A few studies have used LDA or PIV to address the differences
between the flow patterns generated by up- and down-pumping im-
¯ T (m2 /s3 )

pellers (Jaworski et al., 2001; Aubin et al., 2001) and Aubin et al.
0.0025
0.086

(2004) also used PIV to study gas-liquid flows at low aeration rates.
0.09

0.1
Up-pumping impellers have been less commonly used within in-
dustry but have been shown to have great potential for gas disper-
0.34T

0.35T

0.45T
T/2

T/3

T/3

T/3

T/3

T/3

sion, solid suspension in 3-phase systems, ingestion of floating solids


D

and for fast reactions with surface feeding (Nienow and Bujalski,
0.46T

2004).
T/3

T/4

T/3

T/3

T/2

T/3

T/3

T/4
C

None of the above studies has attempted a direct comparison of


T (m)

up- and down-pumping PBTs in the same vessel with respect to T


0.152

0.292
0.27

0.24

0.15
0.1

0.2

0.1

0.3

or Tmax whether time-averaged or angle-resolved. Such measure-


ments would provide further insight into the applicability of both
Spatial res. (mm)

configurations to the successful achievement of many important pro-


cessing aims. In addition, computational fluid dynamics (CFD) is in-
creasingly being used to analyse such processes by calculating both
T and the angle resolved Tmax , so experimental data are important
1.38
1.08
0.5

0.1
5

for its validation.


4PBT-D

4PBT-D

4PBT-D

6PBT-D

6PBT-U

6PBT-D

Therefore, this paper is concerned with the characterization of


LDV
LDV

LDV

LDV

the hydrodynamics of a 4◦ , 6-blade pitch blade turbine (6-PBT) oper-


PIV

PIV

PIV
PIV

PIV

PIV
RT

RT

RT

RT

ated in both an up-pumping (U) and in a down-pumping mode (D) in


Baldi and Yianneskis (2004)

a stirred baffled vessel using time-averaged and angle-resolved PIV.


Lee and Yianneskis (1998)
Kresta and Wood (1993a)
Wu and Patterson (1989)

Sharp and Adrian (2001)


Zhou and Kresta (1996)

The effect of the different modes on the recirculating flow patterns,


on the turbulent kinetic energy (TKE), on flow number (Fl) and on
Sheng et al.(2000)

the local specific energy dissipation rates, T , have been elucidated.


Values of T have been evaluated directly from the measured fluc-
Present work
Khan (2005)

tuating velocity gradients (Baldi and Yianneskis, 2004) and via di-
Table 1

mensionally based relationships (Kresta and Wood, 1993a; Wu and


Patterson, 1989). Furthermore, since PIV can only resolve to a certain
128 A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143

minimum length scale, LIA , defined by the size of the interrogation


area, a large-eddy PIV method was also employed in order to ap-
proximate the dissipation of energy at scales below the resolution
of the measurements (Sheng et al., 2000). The local values of T ob-
tained are compared with data from the literature also in Table 1.
The summation of the local values of T obtained for each methodol-
ogy for the interrogation area is then compared to the overall energy
dissipation rate (or power input) for the whole vessel determined
from measurement of torque on the impeller shaft.

2. Materials and methods

2.1. Vessel configuration

A schematic of the flat-bottomed vessel with a diameter,


T = 150 mm, is given in Fig. 1, which also shows the interrogation
area. The vessel was constructed from borosilicate glass and im-
mersed in a glass rectangular tank filled with water, in order to
minimize the errors due to the refraction of the water and curved
glass of the inner vessel. The PBT impeller used had a diameter,
D of 67.5 mm ( = 0.45T) and six blades, each angled at 45◦ to the
horizontal. The vertical projected width of the blades, W, was 7 mm
(0.047T). Four longitudinal fixed baffles, 3 mm thick and made
from polymethyl methacrylate (PMMA) and of width = 0.1T were
mounted at 90◦ intervals on the vessel wall.
The power number, Po = P/ N3 D5 , where P is the power (W), N
is the impeller speed (rev s−1 ) and  is the fluid density (kg m−3 ),
were obtained for both (U) and (D) configurations by measurement
of torque (M, N m) on the impeller shaft using a Coesfeld Viscomix
unit (Coesfeld, Dortmund, Germany) since P = 2MN. More de-
tails are given in Simmons et al. (2007). The impeller speed was
300 rpm (N = 5 s−1 ) for both configurations giving a Reynolds num-
ber, Re = ND2 / = 2.28×104 , where  is the fluid dynamic viscosity
(Pa s). The choice of fully turbulent conditions ensures that macroin-
stabilities in the flow field are minimized which simplifies the later
Fig. 1. Schematic of experimental rig (with interrogation region indicated from which
analysis (Roussinova et al., 2000). For both (U) and (D) configura- the power input PIM was estimate by integration using the trapezoidal elements in
tions, P = 0.265 W to give Po = 1.55 and a tank mean specific energy Fig. 10). The impeller was rotated clockwise from the top.
dissipation rate, ¯ T = P/ V, (where V is the volume of fluid in the
vessel) of 0.1 W kg−1 .

Two dimensional PIV data were obtained for each vessel configu-
2.2. PIV experiments and analysis ration and 500 image pairs were recorded for each experiment. The
statistical reliability of the selection of the number of image pairs
The vertical laser plane in all of the experiments was placed at was tested by comparing root mean square velocity values between

angle of 5 from the nearest baffle, passing through the vessel axis 400 and 500 image pairs; the difference was less than 1%. The images
as shown in Fig. 1. Although most workers using PIV perform ex- were processed with a recursive Nyquist grid. For the first pass, the
periments with the laser plane equidistant between the baffles, i.e., dimensions of the IA were 32×32 pixels with 16×16 pixels adopted

at 45 , due to imperfect matching of refractive index between the for the second pass.
working fluid (water) and the PMMA baffle, distortions in the bulk The analysis was, for reasons of symmetry, performed only on
of the image by the baffle were observed in this work. Hence, an an- half of the vessel. The angle resolved study was carried out on the

gle of 5 was used, which gave a clear and undistorted image of the bottom half of the vessel, covering 0 < z/H < 0.5, 0 < r/R < 1, as shown
impeller and main flow loops. in Fig. 1 to capture velocities at a higher spatial resolution. The angle
The PIV equipment used comprised of a dual head Nd:YAG laser resolved images were obtained by use of a triggering device, which
(New Wave Research, Fremont, CA) and a single frame-straddling synchronized the impeller position to the image capture. The de-
CCD camera (TSI PIVCam 10-30, TSI Inc.) capable of capturing im- vice was capable of resolving angles of less than half of a degree.
ages at 15 image pairs per second with a resolution of 1000×1016 Since the 6-blade PBT may be assumed as having six-fold symmetry,
pixels2 . The camera and laser were controlled by a TSI LASERPULSE measurements between the two impeller blades covering angles of
610030 synchronizer (TSI Inc., Shoreview, MN) and a DELL Preci- 0–60◦ can be taken to resolve the entire flow from 0◦ to 360◦ .
sion 620 workstation running TSI Insight䉸 5.1 software. Fast fourier The degree of separation of the acquisition between two subse-
transform (FFT) cross correlation is used to interrogate the two im- quent angles was chosen on the basis of the thickness of the laser
ages, which are divided into interrogation areas (IA) of length LIA . sheet. The chosen degree of separation was 3◦ in order to obtain a
The seeding particles used were hollow 10 m diameter silver coated complete overlap of the laser sheet between acquisitions until a dis-
particles (Dantec Inc., DK) with a sufficiently small relaxation time tance of 19 mm away from the impeller shaft. Hence measurements
that they may be considered to follow faithfully the motion of the were taken for 20 angles between 0◦ and 60◦ , with 500 image pairs
fluid. The camera was mounted on a computer-controlled traverse. being acquired at each angle. Time-averaged data were obtained by
A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143 129

averaging over all the angles measured. Separate time averaged data 3. Theory
were not obtained since the rotational speed of the impeller was an
exact multiple of the sampling frequency. Hence any time-resolved Turbulent flow is characterized by temporal fluctuations in fluid
measurements would have been at a fixed, unknown blade angle. No velocity at any point. For any co-ordinate system, for flows, which
further processing, for example to remove the periodic component are steady on average, it is possible to decompose the instantaneous
due to blade passage, was performed. velocity signal, U, into an ensemble-averaged mean component, ū,
The resolution of the PIV is dependent upon the scale of the and a fluctuating component (deviation from the mean), u , thus,
individual IA; within each IA, the velocity vector obtained is a spatial
average. This characteristic measurement length scale, LIA , acts as U = u + u (4)
a filter on the velocity data, since only length scales greater than
Since the fluctuating component, u , can take positive or negative
this scale are resolved. To resolve to smaller length scales requires
values, it is normal practice to characterize it via the root mean
higher magnification and for a CCD camera, this requires the size of
square or RMS, ũ, i.e.,
the viewing area to be reduced. Hence, there is a trade-off between
 
the area viewed and the resolution obtained. The resolution adopted
was 86.8 m pixel−1 so the measurements were resolved to 1.38 mm ũ = u2 = (U − ū)2 (5)
for a 16×16 pixel IA. Allowing for a 50% overlap of the IA, the vectors
For angle resolved data, a similar decomposition can be per-
are spaced 0.69 mm apart.
formed (Sharp and Adrian, 2001),
The delay between each frame in an image pair, td , was chosen
in relation to the maximum displacement that a particle could travel U = u + ū| (6)
in the interrogation window. td Can be calculated from:
where ū| is the angle resolved mean. The periodic fluctuations in
Mu Iw the flow may be determined as
td < (1)
4Utip
uP = ū| − ū (7)
where Mu is the magnification, Iw is the pixel dimensions of the in-
terrogation window and Utip is the tangential velocity at the tip of Hence
the blade (Utip = DN). The delay between two images was adopted
assuming that a particle present in the centre of the IA in the first im- u = u − uP (8)
age, moving at the maximum observed velocity of 0.5Utip , is present
Since the fluctuating component should only comprise of random
in the second image with a displacement of 0.25LIA (Adrian, 1986).
fluctuations due to the turbulence, u can be considered as the true
During post processing of the vector fields, using the TSI Insight䉸
turbulent velocity fluctuation since the periodic component included
software, a high pass filter was applied to remove velocities greater
in u is removed. In a stirred vessel, the impeller transfers kinetic
than the value of the impeller tip speed, Utip, together with vectors
energy to the fluid manifested by both the mean flow and the fluc-
that were three times bigger in magnitude than the standard devi-
tuating velocities within the flow. The TKE, k, is the portion of ki-
ation of the magnitude of surrounding vectors in a 9×9 grid around
netic energy due to the velocity fluctuations that provides a mixing
them. The velocity vectors removed were replaced by interpolated
mechanism due to turbulent dispersion. In Cartesian co-ordinates,
values from the same 9×9 grid. However, it should be noted that
TKE may be determined from 3-D velocity data as
these filtering procedures affected less than 1% of the processed
vectors. k = 12 (ũ2 + ṽ2 + w̃2 ) (9)

2.3. Flow number measurement For measuring techniques where only two components can be ob-
tained, as in 2-D PIV, the third fluctuating component of the velocity,
The flow number (Fl) is a dimensionless group, defined as the w̃, is determined from the isotropic approximation
ratio between the effective non-entrained pumped flow, and a
w̃2 = 12 (ũ2 + ṽ2 ) (10)
hypothetical impeller characteristic flow rate:
Q Hence,
Fl = (2)
ND3 k = 34 (ũ2 + ṽ2 ) (11)
For the volumetric flow rate, Q, to be calculated, a surface normal to
This assumption was verified experimentally by Khan (2005) for
the flow has to be created around the impeller. This surface is usually
a PBT using stereoscopic PIV data. In the later discussion,
 time-
taken as a circle close to the impeller in the direction of pumping and
with a diameter equal to the impeller swept diameter for an axial averaged TKE is based on time averaged RMS values (ũ= u2 ) 
whilst
flow impeller. For the PIV measurements, the surface was taken at a
angle-resolved TKE is based on angle resolved RMS values (ũ= u2 ).
distance of 1.38 mm away from the blade surface, which represents
The TKE is normalized using the square of the impeller tip speed,
the first row of IA, which is suitably far away from the blade not to
i.e., k∗ = k/Utip
2 .
suffer from interference in the data due to reflections off the impeller.
Assuming symmetry every 60◦ , or /3 rad, Q is calculated from the At high Reynolds numbers, it is possible to adopt energy cascade
angle resolved measurements as theory as a model for the fluid behaviour. Large primary anisotropic
eddies, with sizes comparable to the physical dimensions of the sys-
 /3  D/2 =
 /3 x=D/2
 tem, are generated by the impeller and decay to smaller isotropic
Q =6 xv(x, ) dx d = 6 xv(x)LIA  (3) eddies with higher frequencies until the Kolmogorov length scale
0 0
=0 x=0 is reached. At this scale, viscous and inertial forces are equivalent
(Rek = 1) and dimensionless analysis gives
where v is the axial component of the velocity, x is the radial distance
along the blade and  is the azimuthal angle between the two blades.  1/4
3
Discharge from the radial periphery of the impeller is not considered, K = (12)
and is shown later to be negligible. T
130 A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143

Fig. 2. Time-averaged flow fields for: (a) PBT-U impeller and (b) PBT-D impeller.

where K is the Kolmogorov length scale, T is the local specific en- Since the present PIV data only contains two velocity components,
ergy dissipation rate and v is the kinematic viscosity. At equilibrium, it is necessary to provide an estimate for the third component, in
the rate of generation of the TKE in the large, flow-dependent scales this case by using an isotropic approximation. Khan (2005) provides
is the same as the rate of dissipation, that manifests itself as heat and a detailed analysis of the various assumptions inherent in this pro-
occurs in the viscous sub-range, with a direct relationship between cedure. Sharp and Adrian (2001), used the isotropy assumption to
the Reynolds number and the difference between the largest and the simplify Eq. (13) into the following:
smallest scales /K = Re3/4 , where  is an ILS (Frisch, 1995). ⎛
Therefore, to measure the dissipation of the TKE, it is necessary j u 2 j v 2 jv 2
(T )DE = ⎝ +2 +3
to be able to resolve turbulent quantities down to the Kolmogorov jx jy jx
length scale. Saarenrinne et al. (2001) showed that in order to mea- ⎞
sure with sufficient accuracy to obtain 90% of the dissipation rate, j u 2 ju jv ⎠
+3 +2 (14)
the spatial resolution should be 2 K , and to reach 65%, it should jy jy jx
be 9 K . Clearly with the present image resolution, it is not possible
to resolve across the entire length scale from O() to O(K ) within Since the PIV measurements only resolve to a scale several times
one image acquisition. From Eq. (12), the average Kolmogorov length larger than K , this equation, termed the direct evaluation or DE
scale based on ¯ T = 0.1 W kg−1 is 56.2 m, hence the resolution of method, has only been applied for illustrative purposes to allow com-
the angle resolved data is 24.7 k . Hence, from the Helland spectrum parison with the approaches presented below, which attempt to
discussed by Saarenrinne et al. (2001), the average energy dissipated compensate for the limited resolution in other ways.
through the vessel that can be resolved is expected to be 15% of the Wu and Patterson (1989) proposed an equation to evaluate the
total. Various methods have therefore been adopted for a more ac- local energy dissipation from the TKE based on a relationship pro-
curate calculation of local specific energy dissipation rate from PIV posed by Batchelor (1953),
data.
k3/2
(T )DA = A (15)

3.1. Evaluation of the TKE dissipation rate
where A is a constant of proportionality (taken by them to be equal to
The energy dissipation can be defined as the rate of viscous dissi- 0.85 but other values have been used, see Table 1); and  is suggested
pation of mechanical energy in the turbulent motion (Hinze, 1975), to be a “resultant” macro ILS defined as:  =(2x + 2y + 2z )1/2 , which
thus:
given the 2-D PIV used here leads to  = (2x + 2y + y x )1/2 . x
⎧ ⎫
⎪ 2 2 2 2 ⎪ and y are evaluated using Eq. (17).


⎪2
ju jv jw ju ⎪




⎪ jx + jx + jx + jy ⎪


This method, termed the dimensional analysis or DA method, is

⎪ ⎪
⎪ subject to some key assumptions in addition to the value of the
⎨ 2 2 2 2 ⎬
(T )DE =
jv jw ju jv (13)
+2 + + + constant A. Energy enters the turbulent flow at the largest scales of

⎪ j j j j ⎪
⎪
y z z z

⎪  ⎪
⎪ motion () and is contained in the larger eddies. This energy is

⎪ 2 ⎪

⎪ +2 jw +2


ju jv ju j w jv jw ⎪


⎩ jz jy jx + jz jx + jy jx ⎭ transferred to smaller scales without loss by dissipation and in
equilibrium. Various constant values of ILS have been adopted
A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143 131

Fig. 3. Angle resolved average flow fields for different angles relative to the blade: (a) PBT-U and (b) PBT-D.

throughout the flow field, from  = D/6 (Costes and Couderc, 1988), Normally, the normalized autocorrelation, for increasing , con-
to D/12.5 (Wu and Patterson, 1989). Also, the assumption that the tinuously decreases to an asymptotic value. If the chosen is too
value of  is spatially constant is clearly not the case in reality large, the function correlates to a portion of the field not influenced
since in the bulk region of the flow the typical dimension of the by the same vortex structure leading to a discontinuity. The opti-
turbulence is greater than in the proximity of the impeller. mized value of max = x = y was found to be 7 mm.
Local values of ILS, e.g., ux , within the flow field can be eval-
uated from the 2-D spatial autocorrelation, Rux ( x , y ), of the fluc-
3.2. Large eddy PIV method with Smagorinsky closure model
tuating velocity components (Rao and Brodkey, 1972; Escudié and
Line, 2003):
Assuming dynamic equilibrium within turbulent flows, the
u (x)u (x + x , y + y ) largest scales where TKE is generated are strongly flow dependent; at
R ux ( x , y ) = (16) the smallest scales where the energy is dissipated, the structures are
urms (x)2
more universal. Large eddy simulation (LES) exploits this approach.
to give u , It applies a low pass filter to the Navier Stokes equations, which are
 +∞  +∞ solved for the larger scales, whilst the sub-grid scale (SGS) stress,
 ux = Ru ( x , y ) d x d y (17) ij, , introduced by the filtering process, is found from a model for the
−∞ −∞
smaller scales. This reduces computational expense when solving
In practice, the limits of x and y are not infinite but bounded, since these equations via finite volume methods such as those employed
the flow field has finite dimensions. Values of x and y were chosen in CFD packages. Clearly, there is an analogy between these SGS
by performing a normalized autocorrelation at different points in the filtering methods in LES-CFD and data obtained from PIV, as pointed
flow for increasing sizes of the integration area (2 x , 2 y ), where out by Sheng et al. (2000), amongst others. They adopted the SGS
square areas were chosen i.e., x = y . models to provide the contribution to the total energy dissipation
132 A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143

Fig. 3. (continued).

rate below the resolution of their time-averaged PIV measurements. the resolution of the PIV measurements does not approach the Kol-
The eddy viscosity model proposed by Smagorinsky (1963) was mogorov scale, as expected, but would be expected to lie within
found to be the most reliable closure model, although very little the inertial sub-range. These magnitudes are comparable to previous
difference was found between the various SGS closure models ex- works applying LES-PIV (e.g., Sheng et al. 2000; Khan, 2005).
amined. Similar conclusions were found by Hall et al. (2005) for In Cartesian co-ordinates, (T )SGS can be calculated from the fil-
experiments performed on small high throughput stirred reactors. tered gradients obtained from 2D- PIV using (Khan, 2005):
The SGS stress is obtained from the Smagorinsky model as ⎛ ⎞3/2
j u 2 jv 2 j u 2 j v 2
2
ij = −(Cs ) |S̄|S̄ij (18) (T )SGS = (Cs ) 2⎝
4 +4 +2 +2 ⎠ (19)
jx jy jy jx
where Sij = 12 (jui / jxj + juj / jxi ), is the filter width or cut-off scale,
In Eq. (19), isotropic assumptions have been used to compensate
and (Cs2 2 |S̄|) is the eddy viscosity where |S̄| is the filtered rate of for the unknown z component. The constant Cs is the Smagorinsky
strain |S̄| = (2S̄ij S̄ij )1/2 . The local specific energy dissipation rate can constant, which has usually been assumed in the earlier work in
then be estimated since T ≈ (T )SGS  = −2ij S̄ij . Table 1 to vary from 0.17 (Sheng et al., 2000) to 0.21 (Sharp and
Since the length of the interrogation area determines the spa- Adrian, 2001). In this work, Cs = 0.17 has been generally adopted,
tial resolution of the system, this implies for the angle resolved and was also first used by Lilly (1963). Hence, here it is called the
PIV measurements that = 1.38 mm. Values of constant ILS may be Lilly–Smagorinsky constant, Cs,∞ . However, since the precise value
taken from 5.4 mm ( = D/12.5) to 11.25 mm ( = D/6) whilst the of this constant is not generally agreed and is dependent on certain
Kolmogorov length scale was calculated to be K = 56.2 m. Hence basic assumptions, the use of other values will be discussed later.
A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143 133

The values of T calculated using the three methods above can


be normalized by the tank averaged specific energy dissipation rate
as ∗T = T / ¯ T , where ¯ T = 0.1 W kg−1 as determined from measure-
ment of power via the torque on the impeller shaft and the results
are generally presented in that format. Angle resolved T was cal-
culated in all cases from angle resolved data, i.e., in Eqs. (14)–(16)
and (19), the fluctuating velocity values used were calculated from
Eq. (6). Time averaged values of T were obtained by using fluctuat-
ing velocities calculated from Eq. (4).

4. Results and discussion

4.1. Flow fields and flow numbers, Fl

Time-averaged flow fields for the up-pumping (PBT-U) and down-


pumping (PBT-D) configuration are shown on Fig. 2(a) and (b),1
respectively. The velocity vectors shown are normalized by the im-
peller tip speed, Utip. Fig. 2 (a) illustrates the flow pattern for the
PBT-U. The impeller discharges at a high velocity of ∼0.5 Utip at an
angle upwards of ∼45◦ and extends to a height of z/H = 0.5 at the
wall. Here, it splits to give a two loop structure, one a relatively
strong downward flow at the wall, which then recirculates below
the impeller back into the hub with velocities from 0.2–0.5 Utip . The
other loop, which contains a smaller proportion of the flow, forms
a counter-rotating flow loop in the upper part of the vessel for z/H
> 0.5. Velocity magnitudes in the upper flow loop are less intense,
typically ∼0.1 Utip .
Fig. 2(b) shows the typical flow pattern observed for the PBT-D;
the fluid is discharged from beneath the impeller with a maximum
velocity magnitude of ∼0.5 Utip and is swept around the vessel bot-
tom and sides to form a lower flow loop. This loop extends to a
height of z/H = 0.7, in agreement with Bittorf and Kresta, (2001) and
the velocities in the main circulation loop do not drop below ∼0.3
Utip . Close to the impeller shaft, in the upper reaches of the vessel
at z/H = 0.8 and above, the flow is less affected by the flow loop and
relatively quiescent, with velocities of 0.1 Utip . A similar effect is
noted beneath the impeller hub where the flow is reversed, giving a
region where solid suspension is difficult (Nienow, 1997). A notable
observation is that velocity magnitudes for the upflow configura-
tion are, in general, substantially lower in the upper reaches of the
vessel, i.e., ∼0.05–0.15 Utip for z/H > 0.6, 0.2 < r/R < 0.8, than for the Fig. 4. Radial variation of time-averaged axial velocity from impeller discharge from:
(a) PBT-U and (b) PBT-D.
downflow configuration (∼0.1––0.3 Utip ). The results for both PBT-U
and PBT-D with this size D/T ratio and clearance are similar to those
observed in previous work using ensemble-averaged data (Aubin
et al., 2004; Jaworski et al., 2001). is constricted by the base and baffle geometry, causing a local veloc-
Angle resolved flow fields for both configurations are given in Fig. ity increase as the fluid accelerates from a downward to an upward
3. At an angle of 0◦ , the middle of the impeller blade is in the plane of direction. A similar situation is observed in the up-pumping config-
the laser sheet. Plots are given for every 15◦ of blade rotation and, as- uration, but with this clearance and since the flow is split into an
suming perfect symmetry of the blades, the results have a periodicity upper and lower flow loop, the incremental increase in velocity at
every 60◦ (1/6 of a revolution). Hence results at angles of 0, 15, 30 and the wall is less evident.
45◦ are shown. Considering the PBT-U (Fig. 3a), the highest values of The flow numbers, Fl, for both upflow and downflow configura-
velocity, of 0.55 Utip , are in the impeller discharge region when the tions were 0.87, as shown in Fig. 4. Kresta and Wood (1993b) per-
blade is in plane (0◦ ). The velocities in this region are sensitive to the formed measurements on a 45◦ 4-blade PBT-D and for 0.59 > C/D
angle relative to the blade and the region moves slightly upwards > 0.33 and D = 0.5 T with N = 450 rpm, they found a negligible influ-
and away from the impeller as the angle increases. However, the ence of clearance on the value of Fl of 0.82. Since they used a 4-blade
spatial distribution of velocity in the lower flow loop and throughout PBT, these results are in reasonable agreement. The values obtained
the bulk of the vessel otherwise remains relatively constant. also compare well with values from other studies, e.g., 0.88 for a 6-
Similar observations may be made for the PBT-D impeller in blade 60◦ PBT-D (Hockey and Nouri, 1996) but slightly higher than
Fig. 3(b). There is an additional area behind the baffle with higher the values of 0.73 for 6-blade 45◦ PBT-D (Nienow, 1998) and 0.68
velocity. This velocity can be explained by considering that the flow and 0.75 reported by Aubin et al. (2001) for a PBT-U and PBT-D, re-
pumped downwards from the impeller (at C = H/4) towards the wall spectively. As shown on the figure, the values of Fl were evaluated
at a distance of 1.38 mm from the blade. It can be noted that the
shape of the velocity profile in the up and the down mode differ
slightly, even though they give essentially the same value for Fl. As
mentioned in earlier in Section 2.3, radial discharge from the vertical
1
For clear interpretation of Figs. 2, 3, 5–9, the reader is advised to view colour edge of the blade is not considered in these calculations; inspection
versions available online at www.sciencedirect.com. of Figs. 2 and 3 confirm this flow to be negligible.
134 A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143

Fig. 5. Normalized angle-resolved TKE for different angles relative to the blade compared to time-averaged values: (a) PBT-U and (b) PBT-D.

4.2. Normalized TKE sometimes overlap. At 0◦ , the region furthest away from the im-
peller is due to the vortex generated by the previous blade and with
The normalized TKE, calculated from the angle-resolved data, increasing angle, they both move away from the blade and at the
(k* )ar , is shown in Fig. 5. In Fig. 5(a) for the PBT-U, the region same time, these regions become larger and more diffuse, with val-
of highest TKE is in the impeller discharge with maximum values ues of (k* )ar > 0.03 in the bulk of the discharge flow. In Fig. 5 (b) for
of (k* )ar = 0.071. Two areas of maximum TKE are observed which the PBT-D, there are also two areas of high TKE ((k* )ar > 0.045) at
A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143 135

Fig. 5. (continued).

each angle, but the regions are more distinct and localized. For ex- Also shown are the time-averaged normalized TKE, (k* )ta , for each
ample at  = 30◦ , there is an area at the impeller tip, and another mode. In each case, the time-averaging smears out the volume in the
at coordinates r/R = 0.55, z/H = 0.07. Again, the latter area is due to trailing vortex over which the TKE is a maximum and the effect is
the vortices from the previous blade. Comparing the results for the greatest in the up-pumping case. In addition, the detailed movement
two configurations, it is clear that the regions of high TKE in the of the vortex is lost. On closer inspection of the maximum angle-
PBT-D configuration are more localized, whilst in the discharge of resolved normalized TKE, it is found to be very peaked and for the
the PBT-U, they cover a greater area of the vessel. However, typical down-and up-pumping mode, respectively, it is 0.060 at an angle of
maximum values and bulk values in the discharge loops are similar 39◦ and 0.071 and 57◦ compared to 0.038 and 0.045 for the time-
in each configuration. averaged values (see Table 2).
136 A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143

4.3. Normalized local specific energy dissipation rates with the SGS method ((∗Tmax )ar > ∼ 10). The distribution in Fig. 6
of (∗T )ar in the SGS method follows more closely the distribution
In Fig. 6, the three methods for calculating the normalized angle- obtained from the DE method, but the relative values are higher,
resolved local specific energy dissipation rate, (∗T )ar , are compared typically (∗T )ar = 6 up to 10 for the SGS compared with 0.25–0.5 for
for the PBT-U impeller at an angle of  = 0◦ . The DE method the DE.
(Eq. (14)) is shown in Fig. 6(a); maximum values are observed in Comparing the value of ((∗Tmax )ar between DE and SGS methods,
two locations corresponding to the location of vortices from the the maximum values vary by a factor of 20, whereas for the DE and
current blade passage and the previous blades. It is important to DA methods, they vary by a factor of 40. Assuming the Helland spec-
note that different scales are present for (∗T )ar for each method trum for turbulence holds, at a resolution of 24.7 K , only 15% of the
and values for the DE method are considerably lower than for the energy dissipation would be expected to be resolved (Saarenrinne
other two methods with (∗Tmax )ar less than 1, even close to the et al., 2001). Clearly a factor of 20 between DE and SGS methods
impeller. would suggest that ∼5% was resolved. The situation is even worse if
Results using the DA method (Eq. (15)) with measured local val- one compares DE and DA methods, which brings into question the
ues of  and the SGS method (Eq. (19)) are shown in Fig. 6(b) and (c), validity of this comparison. Values of (∗T max )ar between the DA and
respectively. The area of maximum (∗Tmax )ar is larger using the DA SGS methods vary by a factor of 2.
method and the values are much higher ((∗Tmax )ar > ∼ 20) compared Since the values from the DE method were so low, Fig. 7 only
compares the DA and SGS methods for estimating (∗T )ar for the PBT-
U impeller at other angles. As expected, the SGS method (Fig. 7b)
shows that (∗Tmax )ar reduces as the discharge jet moves away from
Table 2 the impeller blade as r/R, z/H and  increase, though, in general, a
Maximum values of observed angle-resolved normalized specific energy dissipation peak from successive blades is seen. Oddly, the DA method shows
rate, (∗Tmax )ar , for the DA and SGS method (CS = 0.17) and of the normalized TKE, that the area of (∗Tmax )ar grows with increasing angle with little
(k*)ar , for the two pumping modes and the angle relative to the blade at which each reduction in magnitude. Similar results are observed for the PBT-D,
are found.
the area of (∗Tmax )ar in the impeller discharge moves downstream
(∗Tmax )ar (DA) (∗Tmax )ar (SGS)
Pumping mode (k*)ar
with increasing  (Fig. 8). However, with PBT-D, (∗T )ar in the vor-
PBT-U 48 19 0.071 tex from the second of the two succeeding blades is much lower.
Angle 57◦ Angle 54◦ Angle 57◦ (∗Tmax )ar > ∼ 20 for the DA method compared with (∗Tmax )ar > ∼ 10
PBT-D 34 15 0.060
for the SGS method for the PBT-D, as it did for the PBT-U. Again, the
angle 36◦ Angle 30◦ Angle 39◦
actual peak (∗Tmax )ar is very sharp for both the PBT-U and PBT-D for

Fig. 6. Angle-resolved specific energy dissipation rates,(∗T )ar , in the vicinity of the PBT-U impeller at  = 0 calculated using: (a) direct evaluation from Eq. (14); (b)

dimensional analysis from Eq. (17); and (c) the SGS method from Eq. (19).
A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143 137

Fig. 7. Comparison of angle-resolved specific energy dissipation rates,(∗T )ar , from: (a) dimensional analysis and (b) SGS method for the PBT-U impeller at other azimuthal
angles.
138 A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143

Fig. 8. Angle-resolved specific energy dissipation rates,(∗T )ar , in the vicinity of the PBT-D impeller calculated using (a) dimensional analysis and (b) SGS method.
A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143 139

the angle-resolved measurements and the DA measurements and


the absolute value and the angle at which it occurred depended
on the pumping direction. These values are also summarized in
Table 2. It is also interesting to compare the maximum angle-
resolved and the time-averaged SGS values for the PBT-D (15 at 30◦
and 7.9, respectively), which are in the ratio, (∗Tmax )ar /(∗Tmax )ta , of
∼2, which is of the same order as that found by Ducci and Yianneskis
(2006) for radial flow Rushton turbines (∼3).
The values obtained by the DA method are significantly higher in
the proximity of the ◦ impeller and in the areas not directly influ-
enced by the discharge jet compared to the SGS. This is despite the
use of local values of  in Eq. (17) where, for example, at 15◦ with
the PBT-U, the values (normalized by D/10) vary from ∼0.2 close
to the impeller to ∼0.8 near the wall. The values are of the order
found in the earlier literature (Lee and Yianneskis, 1998; Wu and
Patterson, 1989). Well away from the impeller in the bulk, the SGS
and DA methods give comparable values. The DA method is based on
simple dimensional reasoning requiring assumptions about A and ,
whilst the SGS method is developed from the Navier–Stokes equa-
tions governing the fluid motion. It is unclear whether the differ-
ences observed are due to the assumptions inherent in each method
or the resolution of the experiments.

4.4. Vortex structure

It is possible to convert the PIV data in Cartesian co-ordinates,


resolved by the azimuthal angle between the blades, , to cylindrical
co-ordinates by simply mapping (x, y, ) to (r, z, ). The trailing
vortices can then be reconstructed in 3-D space (Schafer et al., 1998).
Iso-surfaces of the vorticity, defined as

juy jur
z= − (20)
jr jy
are shown in Fig. 9. It is evident that in the same z/H plane of the
vessel, two different zones are present, one where the levels of z
are higher than 60 s−1 (one vortex attached to the blade and the
passage of the vortex from the previous blade) and one where z
are lower than 60 s−1 where vortices are not present. In the up-
pumping configuration (Fig. 9a), the volume with z > 60 s−1 is more
extended than in the down-pumping configuration (Fig. 9b), and in
this case, it is only at the very end of the vortex from the previous
blade that values of z higher than 60 s−1 can be seen. Fig. 9. Location of the trailing vortex for: (a) PBT-U and (b) PBT-D configurations,
identified by the plot of the vorticity.

4.5. Calculated power input from local specific energy dissipation rates
compared to that from shaft torque
the impeller, it is to be expected the PIX ≈ P, the actual power input
Though not done previously in the literature, it was decided that into the vessel, P, based on the measured torque, i.e., 0.265 W.
a check on the quantitative values given by the different methods The results of this integration for both pumping directions are
of estimating the local T should be made. Firstly, the total energy shown in Fig. 11. On average, the DA method gives PIX /P ≈ 4, whilst
dissipation rate (power, PI ) over the region interrogated was deter- the SGS method gives a value of ∼2. The DE method, unsurprisingly
mined by integrating t (=∗T ¯ T ) over it. The region considered is in since much of the fine scale structure is missed, underestimates PIX /P
the bottom half of the vessel shown in Fig. 1, defined by 0 < z/H < 0.5 by a factor of ∼5. All of the methods suffer from the intrinsic limi-
and 0 < r/R < 1. Integration was done using trapezoidal elements of tation of the equipment used, such as the need to use the isotropic
dimensions defined in Fig. 10 with an azimuthal angle, , of 3◦ , and approximation to estimate the third component of the velocity fluc-
dy and dr equal to LIA /2. The width of the laser sheet is consid- tuation and also the assumptions made in treating the experimental
ered to be infinitesimal, thus every T value is associated with one data. Clearly, as a result, very different values of PIX /P are obtained
integration volume. The integral over this volume for the discrete from each of these methods and none of them are close to 1, the
system to obtain PI is value that should be found as most of the energy is dissipated in the
 region covered by the integration.
PIX =  (∗T ¯ T )Mij Vij (21)
i j 
4.6. Choice of Smagorinsky constant
where the subscript index X relates to the method used to estimate
T and Vij is the volume of trapezoidal element. Since as can be seen The SGS method provides the best estimate, yet the overestima-
here and in all earlier work, most of the energy is dissipated close to tion by a factor of about 2 is still very significant. One aspect of the
140 A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143

so that, with a Kolmogorov constant,  = 1.5,

1 2 3/4
Cs,∞ = = 0.173 (23)
 3

Thus, Cs,∞ = 0.173 comes from various rigid assumptions, such as


the implication of an infinite Reynolds number and imposing a very
simple model for the inertial sub-range energy spectrum.
Recent work has reconsidered these assumptions and as a result,
has taken into account an explicit dependence between the Reynolds
number and the filter width, , and . Thus, Meyers and Sagaut
(2006), as suggested by Pope (2000), used a more complex model for
the inertial sub-range energy spectrum and introduced the concept
of a variable Smagorinsky coefficient, (CS( /  ) ), which was strongly
K
dependent on the local ratio of the filter width, , to the Kolmogorov
scale, K. . Thus, (CS( /  ) ) was shown to increase monotonically from
K
zero at /K = 0 to the value used here, Cs,∞ = 0.17, for /K > 102 .
In the present work, / K has an average value of ∼25, implying that
a value of the Smagorinsky coefficient of ∼0.13 would be appropriate
(Meyers and Sagaut, 2006).
There is also an extensive literature on the application of the
Smagorinsky-SGS- method applied to LES CFD and the choice of
Fig. 10. Schematic of the trapezoidal integration volume.
the correct value for CS . The reason for the use of the SGS in
CFD is to economize on the calculations. Meyers et al. (2007) ap-
proached the problem by minimizing the error associated with
the relatively approximate LES technique using SGS compared to
a precise solution for the same energy spectrum obtained with a
direct numerical simulation (DNS) of the flow field. They predicted
a range of values of CS , which minimize the error of the spatial
resolution, . Introducing the as a cut-off length scale in the
graphs presented in their work, the value of CS which minimizes
the error ranges from 0.1 to 0.13 depending on the Runge–Kutta
numerical method adopted to resolve the convective and viscous
fluxes.
Both the above more recent studies suggest that a significantly
lower value of CS of the order of 0.1 should be used. In fact, if a
value of 0.11 is adopted, the power obtained by integrating the local
specific energy dissipation rates, PIX , equals that measured directly
via the shaft torque, P, i.e., a value of PIX /P = 1 is obtained.
Clearly, the absolute value of PIX relative to P cannot be estab-
lished, especially since measurements were not done for the whole
vessel. Indeed, Khan (2005), who used a value of CS = 0.17 in his
Fig. 11. A comparison of the integrated energy dissipation rate over the bottom half
work, rejected comparing PIX to P because his measurements did
of the vessel (the interrogation volume, see Fig. 1) by the three different methods
of evaluating the local specific energy dissipation rate (Eq. (21)) compared to the not encompass the entire fluid. Nevertheless, it seems that such a
power input into the vessel based on the power number. comparison between PIX and P is important to ensure that the data
analysis is not totally unrealistic.

4.7. Comparison of (∗Tmax )ta with that in earlier work


use of this approach is the value chosen for the Smagorinsky con-
stant, Cs . Various approaches have been taken in the literature to Since Tmax is extremely important for mixing processes, the
correctly choose the value of Cs and much of the work is devoted to its values obtained from this work using each method are compared
meaning and general theoretical aspects. A good resumé on the eval- with values from other published studies in Table 1, (expressed as
uation and use of the Smagorinsky closure model for the analysis of Tmax /N3 D2 , the most common mode of comparison in the literature,
turbulence can be found in Meyers and Sagaut (2006). The first study where Tmax has also been obtained by time-averaging). In this work,
by Lilly (1963) evaluated it theoretically (and as explained above, the value of  adopted for the DA method was (2x + 2y + y x )1/2 .
this value was used in this work and called the Lilly–Smagorinsky However, in Table 1 in addition to this definition of , in order to
constant, Cs,∞ = 0.173). His approach was to equate the energy dis- compare the present study with previous work, length scales eval-
sipated using the SGS method to that from a homogeneous isotropic uated by autocorrelation or as  = (32x )1/2 and as  = (32y )1/2
LES-filtered energy spectrum, Ē(k). He also assumed that the inertial are also reported for the same spatial location; and (Tmax )ta has
sub-range spectrum, which was not resolved by the filter, could be been used. In performing this comparison, it is important to note
2/3
exactly modelled by E(k) = T k−5/3 . This approach gives rise to that Tmax would be expected to vary significantly with impeller
type, the number of blades and D/T (Zhou and Kresta, 1996) and C/T
(Kresta and Wood, 1993a) ratios for the same type of impeller; and
 ∞ 3/2
  even between worker and worker because of the assumptions that
T = (Cs,∞ )2 2 k2 Ē k dk (22)
0 have to be made especially for A and . However, an order of mag-
A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143 141

Table 3
Time-averaged (∗Tmax )ta = ((T )max )ta /¯T for studies on PBT-D impellers in Table 1 and this work.

¯ T (m2 s−3 ) (∗Tmax )ta = ((T )max )ta /¯T

Zhou and Kresta (1996) (D = T/3; C = T/3 ) 0.086 38.83 (DA)


Zhou and Kresta (1996) (D = T/2; C = T/4) 0.652 15.5 (DA)
Kresta and Wood (1993a,b) (D = T/2; C = T/4) 0.38a 36.5 DA)
Sheng et al.(2000) (D = 0.35T; C = 0.46T 6.27×10−4 a 8.5 (SGS, Cs = 0.17)/7(DA)
Khan (2005) (D = 1/3T; C = T/3) 0.09 78 (SGS, Cs = 0.17 )/43(DA)
Present work (D = 0.45T; C = T/4) 0.1 7.8(SGS, Cs = 0.17)/13(DA)b
3.27 (SGS, Cs = 0.11)
a
Value obtained using Po = 1.4.
b
Based on  = (x +y +x y )1/2 .

nitude agreement might be expected if that was the only difference. The angle-resolved PIV enabled a number of novel features to be
In fact, in other work, as here too, where more than one technique identified. Firstly, the two pumping directions were shown to give
has been used, there is a noticeable discrepancy between them. In very different vortex structures. Secondly, the `spottiness' (defined
addition, for the DE method, the values obtained in this study are here as being highly localized in time and space) of the normalized
an order of magnitude lower than in other work, perhaps due to the TKE, (k* )ar , along a radius as the trailing vortex moved away from
lower resolution used here. each impeller can be identified, with the regions of highest TKE with
In some ways, a more useful comparison is on the basis of ∗T = the PBT-D being more localized, whilst in the discharge of the PBT-U,
max
Tmax / ¯ T as it indicates immediately the maximum impact of an im- they cover a greater area. These features cannot by seen with time-
peller for causing drop break-up, micromixing, mechanical stress on averaging, which smears out the volume in the trailing vortex over
dispersed entities, etc., when a certain power is imparted to the ves- which the TKE is a maximum and reduces the value by about 50%.
sel (which is related to the running costs of the process). It is also The turbulent local specific energy dissipation rate can be con-
easier to see the differences between the results of different workers sidered to be proportional to the rate of dissipation of a scalar
and impeller types. However, it requires measurement of power or concentration, which is a direct measure of the micromixing effec-
an assumption regarding the power number. Some examples from tiveness of a system (Sharp and Adrian, 2001). Three methods were
very directly comparable studies are given in Table 3. used to estimate the local specific energy dissipation rate, T ; via
The results in Tables 1 and 3 are worth some more discussion. direct evaluation from the definition of T (DE method); via dimen-
Firstly, when ∗Tmax is determined by DA, it was shown by Zhou and sionless analysis (DA method using measured ILSs); and via use of
Kresta (1996) for four impeller types that the value of ∗Tmax is signif- an SGS closure model to account for unresolved length scales (SGS
icantly bigger when D/T is smaller. This difference is well illustrated method) using mainly CS = 0.173 (a value close to that used by other
in Table 3 by the results for a T/2 and T/3 PBT. When comparing the workers studying stirred vessels) but also CS = 0.11, as set out below.
results of different workers and different methods, it is clearly im- The results were expressed as the normalized local specific energy
portant to bear this in mind. Nevertheless, inspection of the Tables dissipation rate, ∗T = T / ¯ T . Both DE and SGS methods gave the same
shows that there is a huge variation in the results of different work- qualitative distribution of ∗T , although the values for the SGS method
ers even for the same geometry. The SGS method gives the maximum were considerably higher, typically two orders of magnitude. The DA
value of (∗Tmax )ta (Khan, 2005) and the two lowest values (Sheng method gave a different distribution and estimated the highest val-
et al., 2000 and this work), the latter two being in reasonable agree- ues of ∗T . Again, from the SGS method, the angle resolved ∗Tmax was
ment for either value of Cs , though the D/T values are different in the greater than the time- averaged and their ratio (∼2) was similar to
two cases. In each case where both are given, the value of (∗Tmax )ta that found for Rushton turbines (∼3) by Ducci and Yianneskis (2006).
from DA is less than from SGS except in this work; and the three For the first time, integration of the total measured specific energy
DA results for the T/2 PBT differ by a factor of about 2. Overall, it is dissipation rate to give the power input in the PIV interrogated region
clear that, in spite of the improvement in measuring techniques and of the tank has been undertaken and compared to the actual total
numerical sophistication now available, considerably more work is power input based on the measured torque. This analysis showed
required before the values of this very important parameter are es- that the DE method only gave 20% of the total whilst the DA and
tablished with confidence. SGS methods overestimated it by a factor of 5 and 2, respectively.
Clearly, the SGS method gives more realistic result than the DA (even
when measured values of the ILS are used) and DE methods, but the
5. Conclusions prediction depends strictly on the value of the CS chosen. If, on the
basis of some other recent theoretical and experimental studies, CS
For the first time, angle-resolved PIV has been used to study the is changed to 0.11, reasonable agreement is reached between the
flow generated by 45◦ up-pumping pitched blade turbines (PBT-U), two power values.
which have recently shown to be very effective for many mixing The use of a variable CS function to estimate local macro flow
duties (Nienow and Bujalski, 2004). The results have been com- properties, such as  is already available in software for LES-CFD
pared to the few other studies of PBT-U with PIV and the more and could be adopted with some simplifications to the filtered flow
commonly studied down pumping configuration (PBT-D). Both tech- field from PIV. Thus, it can be concluded that the SGS method shows
niques showed that the mean flow with the PBT-U generally devel- the greatest promise for estimation of the local and maximum spe-
oped lower values of velocity than the PBT-D. Also, the PBT-U led to cific energy dissipation rate from PIV experimental data. However, a
a more radial discharge profile with a strong lower flow loop and careful comparison of measurements in this work and the literature
weak upper flow loop present. The PBT-D yielded only one main flow by DA and SGS of ∗Tmax , a very important parameter for many mix-
loop, which encompassed most of the vessel. These results agree well ing processes, shows very wide discrepancies even for recent work
with time-averaged earlier work (Aubin et al., 2001; Jaworski et al., for the same geometry whichever method is used. More work and
2001). Values of flow number and power number for both configu- improved experimental, theoretical and computational methodolo-
rations were essentially the same. gies are clearly still required.
142 A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143

Notation Acknowledgements

A constant in Eq. (15) The PIV equipment was purchased using funds from EPSRC Grants
Cs Smagorinsky constant GR/R15399 and GR/R12800. AG was funded by the School of Chem-
Cs,∞ Lilly–Smagorinsky constant ical Engineering under an exchange agreement with the University
D impeller diameter, m of Pisa, Italy. We wish to acknowledge useful discussions with and
(D) down-pumping configuration assistance with calculation of the integral length scale by Prof. Chris
Fl flow number, dimensionless Rielly, University of Loughborough and also Prof. Elisabetta Brunazzi
Iw interrogation pixels, m2 of the University of Pisa, Italy.
k turbulent kinetic energy, m2 s−2
k* normalized turbulent kinetic energy, m2 s−2 References
LIA side length of interrogation window, m
M impeller torque, N m Adrian, R.J., 1986. Image shifting technique to resolve directional ambiguity in
Mu Magnification double-pulsed velocimetry. Applied Optics 25 (21), 3855–3858.
N agitator speed, s−1 Assirelli, M., Bujalski, W., Eaglesham, A., Nienow, A.W., 2005. Intensifying
micromixing in a semi-batch reactor using a Rushton turbine. Chemical
P power input, W Engineering Science 60 (8–9), 2333–2339.
PI power input associated with the region of interro- Aubin, J., Mavros, P., Fletcher, D.F., Bertrand, J., Xuereb, C., 2001. Effect of axial
gation agitator configuration (up-pumping, down-pumping, reverse rotation) on flow
patterns generated in stirred vessels. Chemical Engineering Research and Design
Po power number, dimensionless 79 (A), 845–856.
Q impeller pumping rate Aubin, J., Le Sauze, N., Bertrand, J., Fletcher, D.F., Xuereb, C., 2004. PIV measurements
R ux ( x , y ) autocorrelation function, dimensionless of flow in an aerated tank stirred by a down- and an up-pumping axial flow
impeller. Experimental Thermal and Fluid Science 28 (5), 447–456.
Re impeller Reynolds number, dimensionless Baldi, S., Yianneskis, M., 2004. On the quantification of energy dissipation in
Rek eddy Reynolds number, dimensionless the impeller stream of a stirred vessel from fluctuating velocity gradient
|S̄| see Eq. (18) measurements. Chemical Engineering Science 59, 2659–2671.
Batchelor, G.K., 1953. The Theory of Homogeneous Turbulence. Cambridge University
S̄ij see Eq. (18)
Press, Cambridge.
u, v, w radial, axial and tangential components of the Bittorf, K.J., Kresta, S.M., 2001. Three-dimensional wall jets: axial flow in a stirred
velocity, m s−1 tank. A.I.Ch.E. Journal 47 (6), 1277–1284.
ū mean velocity, m s−1 Costes, J., Couderc, J.P., 1988. Study by laser Doppler anemometry of the turbulent
flow induced by a Rushton turbine in a stirred tank: influence of the size of
u fluctuating components of the velocity, m s−1 the units II. Spectral analysis and scales of turbulence. Chemical Engineering
ũ root mean square of the fluctuating velocity, m s−1 Science 43, 2765–2772.
ū| angle-resolved mean velocity, m s−1 Ducci, A., Yianneskis, M., 2006. Turbulence kinetic energy transport processes in
the impeller stream of stirred vessels. Chemical Engineering Science 61 (9),
uP periodic velocity component of the flow, m s−1 2780–2790.
(U) up-pumping configuration Escudié, R., Line, A., 2003. Experimental analysis of hydrodynamics in a radially
U instantaneous velocity, m s−1 agitated tank. A.I.Ch.E. Journal 49 (3), 585–603.
Frisch, U., 1995. Turbulence. Cambridge University Press, Cambridge. p. 107.
Utip impeller tip velocity ( = DN), m s−1 Hall, J.F., Barigou, M., Simmons, M.J.H., Stitt, E.H., 2005. Comparative study of different
x,y,z direction of Cartesian co-ordinates, m mixing strategies in small high throughput experimentation reactors. Chemical
Engineering Science 60 (8–9), 2355–2368.
Hinze, J.O., 1975. Turbulence, An Introduction to its Mechanism and Theory. McGraw-
Greek letters Hill, New York.
Hockey, R.M., Nouri, J.M., 1996. Turbulent flow in a baffled vessel stirred by a 60◦
pitched blade impeller. Chemical Engineering Science 51, 4405–4421.
 Kolmogorov constant, dimensionless
Jaworski, Z., Dyster, K.N., Nienow, A.W., 2001. The effect of size, location and pumping
grid spacing or filter width, m direction of pitched blade turbine impellers on flow patterns: LDA measurement
td delay between image pairs, s and CFD predictions. Chemical Engineering Research and Design 79 (A),
T local specific energy dissipation rate, m2 s−3 887–894.
Khan, F.R., 2005. Investigation of turbulent flows and instabilities in a stirred vessel
¯ T mean specific energy dissipation rate, m2 s−3 using particle image velocimetry. Ph.D. Thesis, Loughborough University.
∗T local normalized specific energy dissipation rate, Khan, F.R., Rielly, C.D., Hargrave, G.K., 2004. A multi-block approach to obtain angle-
dimensionless resolved PIV measurements of the mean flow and turbulence fields in a stirred
vessel. Chemical Engineering & Technology 27 (3), 264–269.
 azimuthal angle between blades, ◦ Khan, F.R., Rielly, C.D., Brown, D.A.R., 2006. Angle-resolved stereo-PIV measurements
K Kolmogorov length scale, m close to a down-pumping pitched-blade turbine. Chemical Engineering Science
 integral length scale, m 61, 2799–2806.
Kresta, S.M., Wood, P.E., 1993a. The flow field produced by a pitched blade
kinematic viscosity, m s−2 turbine: characterization of the turbulence and estimation of the dissipation
z vorticity, s−1 rate. Chemical Engineering Science 48 (10), 1761–1774.
Subscripts Kresta, S.M., Wood, P.E., 1993b. The mean flow field produced by a 45◦ pitched
blade turbine: changes in the circulation pattern due to off bottom clearance.
ar angle-resolved Canadian Journal of Chemical Engineering 71 (1), 42–53.
Lee, K.C., Yianneskis, M., 1998. Turbulence properties of the impeller stream of a
DA by dimensional analysis Rushton turbine. A.I.Ch.E. Journal 44 (1), 13–24.
DE by direct evaluation Lilly, D.K., 1963. On the application of the eddy-viscosity concept in the inertial sub-
max Maximum range of turbulence. National Center for Atmospheric Research, NCAR Manuscript
123.
X method of estimating T
Meyers, J., Sagaut, P., 2006. On the model coefficients for the standard and the
SGS using Smagorinsky sub-grid scale variational multi-scale Smagorinsky model. Journal of Fluid Mechanics 569 (1),
ta time-averaged 287–319.
Meyers, J., Geurts, B.J., Sagaut, P., 2007. A computational error-assessment of central
Superscripts finite-volume discretizations in large-eddy simulation using a Smagorinsky
model. Journal of Computational Physics 227 (1), 156–173.
 fluctuating quantity from time-averaged data Nienow, A.W., 1997. The suspension of solid particles. In: Harnby, N., Edwards, M.F.,
 fluctuating quantity from angle-resolved data Nienow, A.W. (Eds.), Mixing in the Process Industries. Butterworth Heinemann,
London, pp. 364–393.
A. Gabriele et al. / Chemical Engineering Science 64 (2009) 126 -- 143 143

Nienow, A.W., 1998. Hydrodynamics of stirred bioreactors. Applied Mechanics Sheng, J., Meng, H., Fox, R.O., 2000. A large eddy PIV method for turbulence
Reviews 51, 3–32. dissipation rate estimation. Chemical Engineering Science 55, 4423–4434.
Nienow, A.W., Bujalski, W., 2004. The versatility of up-pumping hydrofoil agitators. Simmons, M.J.H., Zhu, H., Bujalski, W., Hewitt, C.J., Nienow, A.W., 2007. Mixing in
Chemical Engineering Research and Design 82 (A9), 1073–1081. a model bioreactor using agitators with a high solidity ratio and deep blades.
Pope, S.B., 2000. Turbulent Flows. Cambridge University Press, Cambridge. Chemical Engineering Research & Design 85 (A5), 551–559.
Rao, M.A., Brodkey, S.R., 1972. Continuous flow stirred tank turbulence parameters Smagorinsky, J., 1963. General circulation experiments with the primitive equations.
in the impeller stream. Chemical Engineering Science 27, 137–156. Monthly Weather Review 91, 99–164.
Roussinova, V.T., Grgic, B., Kresta, S.M., 2000. Study of macro-instabilities in stirred Wu, H., Patterson, G.K., 1989. Laser-Doppler measurements of turbulent flow
tanks using a velocity decomposition technique. Chemical Engineering Research parameters in a stirred mixer. Chemical Engineering Science 44 (10), 2207–2221.
and Design 78 (A7), 1040–1052. Yianneskis, M., Whitelaw, J.H., 1993. On the structure of the trailing vortices around
Saarenrinne, P., Piirto, M., Eloranta, H., 2001. Experiences of turbulence measurement Rushton turbine blades. Chemical Engineering Research & Design 71 (A5),
with PIV. Measurement Science and Technology 12, 1904–1910. 543–550.
Schafer, M., Yianneskis, M., Wachter, P., Durst, F., 1998. Trailing vortices around a Zhou, G., Kresta, S.M., 1996. Impact of geometry on the maximum turbulence energy
45◦ pitched-blade impeller. Fluid Mechanics and Transport Phenomena 44 (6), dissipation rate for various impellers. A.I.Ch.E. Journal 42, 2476–2490.
1233–1246. Zhou, G., Kresta, S.M., 1998. Correlation of mean drop size and minimum drop
Sharp, K.V., Adrian, R.J., 2001. PIV Study of small-scale flow structure around a size with the turbulence energy dissipation and the flow in an agitated tank.
Rushton turbine. Fluid Mechanics and Transport Phenomena 47 (4), 766–778. Chemical Engineering Science 53 (11), 2063–2079.

You might also like