You are on page 1of 25

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

net/publication/274572831

On the presence of spectral shortcut in the energy budget of an asymmetric


jet─wake flow in a forward-curved centrifugal turbomachine as deduced from
SPIV measurements

Article  in  Journal of Turbulence · May 2015


DOI: 10.1080/14685248.2015.1013629

CITATIONS READS

6 762

3 authors:

Mohammad Reza Najjari Nader Montazerin


CoolIT Amirkabir University of Technology
19 PUBLICATIONS   129 CITATIONS    56 PUBLICATIONS   496 CITATIONS   

SEE PROFILE SEE PROFILE

Ghasem Akbari
Qazvin Islamic Azad University
22 PUBLICATIONS   134 CITATIONS   

SEE PROFILE

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

On the Role of Low Energy Modes of the Flow on Sub-Grid Scale Parameter Prediction View project

Pulsatile flow in a curved pipe View project

All content following this page was uploaded by Mohammad Reza Najjari on 22 December 2015.

The user has requested enhancement of the downloaded file.


This article was downloaded by: [George Washington University], [Mohammad Reza
Najjari]
On: 06 March 2015, At: 09:43
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Journal of Turbulence
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/tjot20

On the presence of spectral shortcut


in the energy budget of an asymmetric
jet–wake flow in a forward-curved
centrifugal turbomachine as deduced
from SPIV measurements
a a b
Mohammad Reza Najjari , Nader Montazerin & Ghasem Akbari
a
Department of Mechanical Engineering, Amirkabir University of
Technology, Tehran, Iran
b
Department of Mechanical Engineering, Faculty of Industrial and
Click for updates Mechanical Engineering, Qazvin Branch, Islamic Azad University,
Qazvin, Iran
Published online: 03 Mar 2015.

To cite this article: Mohammad Reza Najjari, Nader Montazerin & Ghasem Akbari (2015) On the
presence of spectral shortcut in the energy budget of an asymmetric jet–wake flow in a forward-
curved centrifugal turbomachine as deduced from SPIV measurements, Journal of Turbulence, 16:6,
503-524, DOI: 10.1080/14685248.2015.1013629

To link to this article: http://dx.doi.org/10.1080/14685248.2015.1013629

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015
Journal of Turbulence, 2015
Vol. 16, No. 6, 503–524, http://dx.doi.org/10.1080/14685248.2015.1013629

On the presence of spectral shortcut in the energy budget of an


asymmetric jet–wake flow in a forward-curved centrifugal
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

turbomachine as deduced from SPIV measurements


Mohammad Reza Najjaria , Nader Montazerina∗ and Ghasem Akbarib
a
Department of Mechanical Engineering, Amirkabir University of Technology, Tehran, Iran;
b
Department of Mechanical Engineering, Faculty of Industrial and Mechanical Engineering,
Qazvin Branch, Islamic Azad University, Qazvin, Iran
(Received 11 August 2014; accepted 26 January 2015)

Viscous dissipation and its contribution to turbulent kinetic energy (TKE) budget are
investigated in the asymmetric jet–wake flow of a forward-curved centrifugal turboma-
chine. Single-plane three-dimensional turbulent data are obtained using stereoscopic
particle image velocimetry (SPIV). Viscous dissipation is indirectly estimated from
subgrid-scale (SGS) dissipation (SGS energy flux) by filtering velocity field using a
top-hat filter. The filter scale should be within the inertial sub-range and this is ensured
by spectral analysis of the measured field. Reduction of turbulent energy flux for smaller
filter scales plus underestimation of viscous dissipation as compared with other TKE
terms both suggest the presence of spectral shortcut. This bypass energy transfer (from
intermediate scales towards dissipative scales) works in parallel with direct SGS en-
ergy transfer and affects the classical energy cascade. Analysis of TKE budget in the
rotor exit region shows significant radial/circumferential variations in the contributing
terms. These variations are mainly due to jet–wake–volute interactions, circumferential
asymmetry of volute area and expansion of flow toward the fan outlet.
Keywords: stereoscopic particle image velocimetry (SPIV); centrifugal turbomachine;
spectral shortcut; viscous dissipation; turbulent kinetic energy (TKE); SGS energy flux

1. Introduction
The study of the turbulent kinetic energy (TKE) equation unfolds the physics of a flow
and provides a basis for improvements in turbulence modelling.[1] This is particularly
important for complicated geometries/flows such as turbomachines that have intensive in-
teractions among jet/wake and flow structures.[2,3] The blade-induced three-dimensional
structures increase energy loss and deteriorate machine performance.[4] These non-linear
flow interactions significantly affect the validity of classical turbulence models in turboma-
chinery applications and reduce the accuracy of flow predictions.[5,6] The challenges are
even more intensive in the case of multi-blade centrifugal turbomachines due to the small
blade-to-blade distance and wide volute area.
Most TKE terms are calculable by approximations that are general consequences of the
limitations in measuring a well-resolved velocity field within a three-dimensional grid.[7–9]
Among various TKE terms, viscous dissipation is the most practical term whose estimation
is frequently discussed in the literature.[10–15] Direct numerical simulation (DNS) and


Corresponding author: Email: mntzrn@aut.ac.ir


C 2015 Taylor & Francis
504 M.R. Najjari et al.

experimental techniques are two reliable approaches for examination of viscous dissipation.
DNS of fine-scale turbomachinery flow structures accompanies enormous computational
costs which are not supported by current computational resources. Although experimental
techniques hardly provide a fine resolution data in the order of the Kolmogorov length scale,
they fulfil part of the limitations of numerical methods and are practical for turbomachinery
applications.[2,6,16–19]
Particle image velocimetry (PIV) provides instantaneous spatial velocity field in a
target plane and is appropriate for calculation of viscous dissipation.[7,8] Various methods
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

are implemented to calculate viscous dissipation in different flow fields. These include
direct estimation on the basis of gradients (the gradient-based approach), spectral fitting in
equilibrium range, integral of the dissipation spectrum, energy flux across equilibrium range
and estimation on the basis of dimensional analysis (dimensional estimation).[10,11,14]
Comparison of spectral fitting and gradient-based methods in bottom boundary layer of
oceanic flow indicates that it is meaningful to assume the dissipative scales as isotropic, even
for a non-isotropic flow, and fit the measured data to a universal dissipation spectrum.[20]
Although the spectral fitting approach slightly overestimates viscous dissipation,[8,20,21]
it is superior to dimensional estimation and is useful for coarse-grid data.
If data resolution is close to the Kolmogorov length scale, the most accurate approach is
the gradient-based calculation.[13] This is a challenge for any experimental study, because
there is always a compromise between resolving fine-scale flow structures and capturing
sufficiently large flow regions. The former is required for the analysis of turbulent dissipative
scales and their contribution to the models, while the latter is required for the analysis of
larger coherent structures and the consequent turbulent flow dynamics. For instance, the
fine-grid PIV data of stirred tank flow (with a resolution that is three times larger than
the average Kolmogorov scale) can resolve 65%–80% of the viscous dissipation term.[10]
An order of magnitude decrease in PIV resolution results in two orders of magnitude
underestimation of viscous dissipation from the gradient-based method.
The large eddy (LE) PIV method approximates viscous dissipation from subgrid-scale
(SGS) dissipation in the inertial subrange (ISR) and is another candidate for estimation of
viscous dissipation from coarse-grid data.[22] The SGS dissipation is interpreted as the
SGS energy flux between resolved (large) and unresolved (small) flow scales and can be
calculated either from its definition or classical SGS models.
Comparison of model-based LE-PIV calculation of viscous dissipation with the dimen-
sional estimation in various regions of stirred tank flow shows that the former is more
accurate for inhomogeneous and anisotropic flows. The definition-based LE-PIV estima-
tion of viscous dissipation is also compared with gradient method and confirms their fairly
good agreement.[23]
Two necessary conditions are suggested for validity of LE-PIV approach [22]: (1)
the cut-off wave number should be in the ISR and (2) the sub-integral scales should be
in the dynamic equilibrium range. However, it is unknown whether these two conditions
are sufficient for a reliable estimation of viscous dissipation or other condition(s) is (are)
required.
The LE-PIV approach relies on direct cascading of energy between turbulent scales.
Consequently, presence of any bypass energy transfer from ISR or backward scattering of
energy is likely to reduce the accuracy of LE-PIV approach. Spectral shortcut is an energy
transfer mechanism of the former type (bypass transfer) that is local in physical space and
non-local in spectral space. It is direct extraction of TKE from large to fine turbulence scales
due to non-linear flow processes.[24] This occurs mainly due to existence of an object in
the flow (e.g. canopy elements or rotor blades) that causes vortex shedding. This non-linear
Journal of Turbulence 505
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

Figure 1. Schematic of: (a) the centrifugal fan and (b) the measurement area.[31]

energy transfer mechanism alters energy cascade and deviates the slope of energy spectrum
from Kolmogorov law in ISR.[24,25] In the presence of spectral shortcut, the slope of
energy spectra remains −5/3 up to injection scales and then becomes steeper at larger wave
numbers.[25–28] It can therefore affect performance of LE-PIV approach in the estimation
of viscous dissipation.
Examination of SGS energy flux with respect to filter size within and above a mature corn
canopy shows that the SGS energy flux significantly increases for smaller filter scales.[29]
The considered spectrum covers the ISR and dissipation range and this trend is interpreted
as the consequence of spectral shortcut as an additional energy source for fine scales of
turbulence.[24] Such behaviour is likely to be more significant in complicated flow fields
and affects the LE-PIV approach assumptions.
This paper aims to examine the estimated viscous dissipation via LE-PIV approach for
the complicated rotor exit flow of a centrifugal turbomachine. The existence of spectral
shortcut and backward scattering of energy is investigated using spectral analysis and TKE
budget. The prominent terms of TKE equation as well as effect of vortical structures on
energy backscattering are also discussed. To the best of our knowledge, there is only one
study on different terms of the TKE budget in the asymmetric turbomachinery wake flow
and it is performed downstream of a single row of stationary blades [30], while the flow of
this study is from a centrifugal turbomachine in its real working conditions.
This article is organised as follows. Experimental set-up and flow field characteristics as
well as the uncertainty analysis and post-processing of PIV data are introduced in Section 2.
Section 3 is devoted to calculation of viscous dissipation by LE-PIV approach. This follows
by calculation of two-dimensional (2D) energy spectra in Section 4. The status of LE-PIV
approach and contribution of spectral shortcut and backward-scattering mechanisms on it
are evaluated in Section 5.

2. Experimental set-up and flow field description


2.1. Fan facility and flow characterisation
A forward-curved centrifugal fan (Figure 1) is constructed and used in the experimental set-
up of the present and previous studies.[5,31,32] Table 1 summarises the main geometrical
506 M.R. Najjari et al.

Table 1. The main geometrical and operational characteristics of the centrifugal fan.[31].

Parameter Value Parameter Value

Inner diameter, D1 285 mm Blade entrance angle, β1 90◦


Outer diameter, D2 350 mm Blade exit angle, β2 155◦
Number of blades 43 Off-load angular velocity 750 RPM
Rotor width, b 165 mm Rotor tip velocity, Utip 13.74 m/s
Volute width, B 200 mm Reynolds number, ReD2 3.18 × 105
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

and operational characteristics of the fan.[31] The justifications for using the multi-blade
centrifugal fan for the study are:

(1) the wealth of previous experience with this turbomachine [33–36];


(2) dominant jet–wake interactions at the measurement plane, which are the main
contributors of flow complexities in any type of turbomachine;
(3) flat side walls with easy optical access and minimum light refraction during imaging
and calibration procedures for stereoscopic PIV in the rotor exit region.

Initial performance tests on the basis of ISO 5801 [37] are carried out and the perfor-
mance curves are obtained as shown in Figure 2. ϕ = 4Qv /π 2 nD23 and ψ = 2pF /ρπ 2 n2 D22
are the flow and pressure coefficients, respectively. ρ, Qv , n and pF are fluid density, vol-
umetric flow rate of the fan, rotational speed of the rotor and pressure rise by the fan.
The indices ‘s’ and ‘t’ in Figure 2 denote static and total pressures, respectively. The best
efficiency condition of the set-up corresponds to the flow coefficient of 0.48 (shown by
dashed line in Figure 2). Instantaneous velocity field in the rotor exit region is measured
for this condition using single-plane stereoscopic PIV.
The measurement plane is at 0.6B (B is the volute width) from the inlet (Figure 1(a)).
The measurement area should be large enough to include a sufficient number of jets and
wakes and yet small enough to resolve a major part of TKE. Such an area is selected from
previous studies at the top of the rotor where distinct jet/wake patterns are present due to

60
Static pressure coefficient
4 Total pressure coefficient
Efficiency 50

3 40
ψs , ψt

30
2
BEP

20
1
10

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7


ϕ

Figure 2. Performance curves of the fan. The best efficiency point (BEP) is considered as operating
condition of the fan in SPIV tests and is shown by dashed line.
Journal of Turbulence 507

proximity to the fan outlet. This measurement area is divided into four overlapping fields
of view (FOV) in order to capture a larger area without sacrificing resolution. Two of these
areas cover the upper section of the rotor, FOVs 1 and 2, and two others are located in the
volute region, FOVs 3 and 4 (Figure 1(b)). The dimension of each FOV is 100 × 74 mm2
and the overall area of these four FOVs is 190 × 128 mm2 , as shown in Figure 1(b).
−3/4
The Kolmogorov (η) and Taylor length scales (λ) are estimated from η∼ l Rel =
 3 3 1/4 √ −1/2
ν l/u and λ∼ 15 l Rel = (15νl/u) , respectively.[38] Although these relations
1/2
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

are valid for homogeneous isotropic turbulence, they are appropriate for rough estimation of
Kolmogorov and Taylor scales and frequently used in various anisotropic flows.[10,12,39]
For the present measurements, the root mean square of turbulent fluctuations in the rotor
exit region is selected as the characteristic velocity scale (u ≈ 2 m/s) and the blade-to-
blade distance is selected as the integral length scale (l ≈ 2.6 cm), respectively. The Taylor
and Kolmogorov length scales are therefore estimated as 1.71 mm and 57 μm, respectively
(λ ∼ 30η).

2.2. SPIV instrumentation and resolvable scales


The main elements of the stereoscopic particle image velocimetry (SPIV) set-up are illus-
trated in Figure 3(a) and more details are schematically shown in Figure 3(b). The laser
system is a double-cavity Quantel Brilliant Nd-YAG laser. The maximum energy of each
laser pulse is 150 mJ at 532 nm wave length. The laser frequency range is 4–10Hz and an
optical encoder triggers data acquisition at a particular rotor phase. The rotor rotates about
0.09◦ during the 20 μsbetween laser pulses. This is approximately 1.1% of the blade-to-
blade passage angle. The effective laser sheet thickness is adjusted to be about 1 mm after a
compromise between maximising the light intensity and minimising the loss of seed pairs.
Particle movements are captured with two FlowSense 1600 × 1186 pixel cross-
R

correlation charge-coupled device (CCD) cameras with AF Micro Nikkor lenses andR

interferential filters. A single-sided dot grid target is used for calibration.


About 2100 pairs of double-frame images are acquired for each FOV and then processed
using FlowManager software (ver. 4.6).[40] A third-order XYZ-polynomial scheme is
implemented in FlowManager software to build the required map for combining the 2D
R

vector maps. The seed particles are produced by a SAFEX F2010pluse fog generator. The
mean particle size is about 1 μm. The camera magnification factor is optimally adjusted
such that the diameter of particle image is about twice the pixel pitch.[41]
The Stokes number which is defined as the ratio of particle response time (τp ) to the
Kolmogorov time scale, must be significantly smaller than unity for acceptable tracking
accuracy. The particle response time scale is about 4.33 ns from τp = SG · dp2 /18νf [42]
where SG = 1.17 is the particle specific gravity, dp ∼ 1 μm is the particle diameter and
νf = 1.5 × 10−5 m2 /s is the kinematic viscosity of air. The Kolmogorov time scale is esti-
mated to be η2 /νf which is 216 μs in the present case. The Stokes number is consequently
2 × 10−5 , which confirms that the utilised particle tracers are appropriate.
The raw velocity vector maps are drawn after applying multi-stage adaptive cross-
correlation that started from 128 × 128 interrogation areas and ended at 32 × 32 interro-
gation areas (approximately 2 mm × 2 mm square areas). Setting the smallest resolvable
velocity scales in the 20η − 90η range covers a respective 95%–65% of TKE.[43] The
smallest resolvable scale of this study (2 mm) is approximately 35 times larger than the
estimated Kolmogorov length scale and therefore resolves about 85%–90%of TKE. Imple-
mentation of 50% of overlap between adjacent interrogation areas in the cross-correlation
508 M.R. Najjari et al.
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

Figure 3. (a) A view of SPIV setup and centrifugal turbomachine; (b) schematic representation of
experimental set-up elements.

procedure results in a final spatial vector spacing of about 1 mm, which is smaller than
the Taylor length scale and is approximately 18 times larger than the Kolmogorov length
scale.[5]
FlowManager uses a calibration map to calculate three components of instantaneous
velocity vector map (Ui , i = 1, 2, 3) along three Cartesian coordinates in the measurement
plane. The corresponding streamwise, lateral and spanwise coordinates of the volute are
x1 , x2 and x3 which are equivalent to x, y (in-plane coordinates) and z(normal coordinate),
Journal of Turbulence 509
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

Figure 4. (a) Convergence diagram of x-component of non-dimensional normal Reynolds stress


(−u u ) for point A before and after validation; (b) contour plot of −u u after validation in FOVs 1
and 2. Location of points A, B and C are also shown. Jet and wake trajectories are illustrated by solid
and dashed curves, respectively.

respectively (Figure 1(a)). Mean velocity components are indicated by U i , (i = 1, 2, 3) and


their respective velocity fluctuations are denoted by ui , (i = 1, 2, 3).

2.3. SPIV data post-process


2.3.1. Data validation
Data validation procedures of the FlowManager software include velocity-range, peak-
R

height and moving-average validations. Gaussian sub-pixel interpolation scheme is addi-


tionally applied in the software to avoid the pixel locking effect, enhance the accuracy of
correlation peak detection and resolve particle displacement to 1/64 of the pixel pitch.[40]
The overall uncertainty is estimated to be lower than 1.5% for the in-plane velocity com-
ponents and lower than 4% for the out-of-plane velocity component.[31]
Equation (1) gives ensemble average of any flow quantity ϕ. The over-bar denotes
ensemble averaging, N is the count of flow data in the averaging process for each point and
i is the snapshot index from 1 to Sequal to 2100.

1  (i)
S
ϕ (x, y) = ϕ (x, y). (1)
N i=1

Spatial derivative of any quantity ϕ along both in-plane directions is calculated using
second-order central finite difference scheme. The first-order (backward or forward) finite
difference method is used for points along FOV boundaries.
Convergence diagrams are used as a complementary validation procedure to detect
incorrect or uncertain data.[44] Figure 4(a) illustrates the convergence diagram for the
x-component of normal Reynolds stress (−u u ). This diagram corresponds to point A
in Figure 4(b), which illustrates distribution of the x-component of normal stress. In
Figure 4(a), vertical and horizontal axes denote ensemble average and the count of partici-
pating data in the averaging process, respectively. If one specific datum causes a significant
jump in the average of all previously counted data, it is tagged as uncertain.[44] These jumps
are either due to invalid velocity vectors that result from PIV cross-correlation procedure
(which may be itself due to loss-of-pair or low/high density of seeds in interrogation area)
510 M.R. Najjari et al.

0.03
3
Pk/(Utip/Rtip), Point B
0.08 Pk/(Utip3/Rtip), Point C
εSGS /(Utip3/Rtip), Point B 0.02
εSGS /(Utip/Rtip), Point C
3

0.06

εSGS /(Utip3/Rtip)
Pk/(Utip3/Rtip)
0.01
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

0.04
0

0.02 -0.01

0 -0.02
500 1000 1500
Number of participated vector maps

Figure 5. Convergence diagram of SGS dissipation and turbulent production. The location of points
B and C is shown in Figure 4(b). The vertical axis shows the averaged value of the corresponding
term.

or may also be due to some unusual behaviour of the flow (vortex shedding/ flip-flopping
of the structures). Both of these possible data groups are counted as uncertain data and they
are not used in calculations. Omission of 15 jumps from the convergence diagram of −u u
(shown by labels in Figure 4(a)) significantly enhances Reynolds stress convergence and
results in less noisy and more distinguishable patterns (Figure 4(b)). Validation procedure
rejects up to 14% of data near the rotor tip and up to 5% in other regions.

2.3.2. Convergence analysis


Convergence diagrams for different TKE terms of this study are investigated to show the
statistical convergence of the ensemble averages. Figure 5 shows the convergence of SGS
dissipation (εSGS from Equation (4)) and production term (Pk in Equation (6)) in the flow
field for points B and C as shown in Figure 4(b). According to this figure, there is no sudden
jump or significant variation in the calculated SGS dissipation and production, as the count
of snapshots exceeds 1000. This count is at most about 1500 for other TKE terms.
Table 2 presents more details about convergence of TKE terms and examines the ef-
fect of the last 100 snapshots on their ensemble averages for points B and C. The mean
values represent mean of the last 100 ensemble averages and the maximum deviation is
the maximum difference between ensemble averages and the mean. The standard devia-
tion is also calculated based on the last 100 ensemble averages and their mean. Maximum
deviation and standard deviation are normalised by mean values. The convergence anal-
ysis for velocity and vorticity has been already investigated in [31]. Table 2 shows that
the convergence of production and SGS dissipation terms is better than that of turbu-
lence diffusion and convection terms. The latter two terms are calculated after numerical
differentiation of velocity fluctuations and it is known that such process intensifies any
fluctuations.
Journal of Turbulence 511

Table 2. Statistics of TKE terms convergence at points B and C (Figure 5).

Mean of Normalised Normalised


ensemble maximum standard
Parameter Location average deviation (%) deviation (%)

SGS dissipation (εSGS ) B 61.91 3.01 1.79


C 89.10 2.43 0.79
Convection (Ck ) B −503.74 19.03 6.55
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

C 2897.92 4.15 1.44


Production (Pk ) B 130.92 1.00 0.55
C 789.81 0.77 0.38
Turbulence diffusion (Tk ) B 202.90 8.33 4.58
C −133.48 7.12 3.51

2.3.3. Velocity field


 √ 
The magnitude of the ensemble average velocity V = U1 2 +U2 2 +U3 2 is presented in
Figure 6. Jet and wake trajectories correspond to local maximums and minimums of
velocity magnitude in the rotor exit region and are plotted by solid and dashed lines, re-
spectively. Jet/wake structures are dominant near the rotor (FOVs 1 and 2), while in volute
region (FOVs 3 and 4) they are mixed with flow along the volute wall. The jet–wake–
volute interactions generate three-dimensional flow structures that strongly contribute to
geometrical/dynamical characteristics of flow structures.
Further explanation of experimental set-up, assumptions, data acquisition method and
uncertainty analysis of extracted velocity components is given in [5,31,32,44]. They include
explanation and discussion of rotor exit mean flow patterns, examination of geometrical
characteristics of flow structures and evaluation of common SGS models on such compli-
cated flow field.

Figure 6. The magnitude of ensemble average velocity. The jet and wake centrelines are labelled by
J and W, respectively.
512 M.R. Najjari et al.

3. Calculation of viscous dissipation from SPIV data


The PIV resolution of this study is considerably larger than the Kolmogorov scale. Conse-
quently, the gradient-based calculation of viscous dissipation significantly underestimates
this quantity and is not reliable. In contrast, the LE-PIV approach (calculation on the basis
of SGS energy flux) is not limited to fine-resolved data and is expected to suitably estimate
viscous dissipation.[22]
The SGS dissipation (εSGS ), that can be interpreted as the SGS energy flux, is
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

defined as
 
εSGS = − τij Sij , (2)

       
∂ U
i

where τij = Ui Uj − Ui
Uj is the SGS stress tensor, Sij = 12 ∂ xji + ∂ U ∂ xj
is the
filtered strain rate tensor and the ·
and over-bar symbols indicate spatial filtering and aver-
aging operations, respectively. Among three popular filter types, i.e. the top-hat, Gaussian
and cut-off filters, a top-hat filter is chosen since it is local in the physical space and non-
local in the Fourier space, while an opposite behaviour is true for the sharp cut-off filter. The
Gaussian filter is non-local, both in the physical and Fourier spaces. The calculations for
the SGS dissipation and other TKE terms are on the basis of physical space formulations.
Moreover, bounded fluid domains forbid the use of filters that are non-local in space. The
top-hat filter kernel is defined as

⎨1 x y
if |x| < and |y| <
G (x, y) = x · y 2 2 , (3)
⎩ otherwise.
0

where x and y denote the filter scale along x and y directions, respectively. Three filter
sizes, i.e. = 3δ, 5δ and 7δ, are implemented to calculate the filtered quantities in Equa-
tion (2) (δ is the PIV vector spacing).
The single-plane SPIV measurement gives all the in-plane velocity gradients. Ad-
ditionally, the normal out-of-plane component of strain tensor can be calculated from the
continuity equation for incompressible flow, i.e. S33
= − ( S11
+ S22
). Two approaches
are frequently used to estimate SGS dissipation from known components.

(1) Assuming isotropy and considering the same value for all cross terms in
 
τij Sij .[6,45,46]
(2) Estimating a 2D surrogate of the SGS dissipation based on the in-plane components
of SGS stress and strain tensors.[21,47–50]

This study develops the latter approach and estimates the SGS energy flux from the five
known components without assuming any limiting state of turbulence:

9 
εSGS = − τ11 S11
+ τ22 S22
+ τ33 S33
+ 2τ12 S12
. (4)
5

The multiplier 9/5 does not change the patterns of εSGS based on the five known
components and just scales SGS dissipation to account for the contribution of the unknown
components by an approximation.[22,27,28]
Journal of Turbulence 513

4. Spectral calculation
Spectral calculation shows the share of energy in different turbulence scales and the size
of important flow structures. Comparison of energy cascade for any anisotropic flow with
that of classical turbulent flows is helpful to discuss about possible non-linear/non-local
energy transfer mechanisms in such complicated flow. Quantification of the share of such
mechanisms in the overall energy transfer would be also possible if the entire range of flow
scales is resolved. Accordingly, this paper examines spectral characteristics of flow field
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

prior to approximating viscous dissipation by SGS energy flux. Two-hundred vector maps
with the largest number of valid data are sufficient to obtain convergent energy spectra.
Two-dimensional instantaneous energy spectra (Ejj 2D
, j = 1, 2, 3) are calculated by ap-
plying 2D discrete Fourier transform (DFT) on instantaneous vector maps of velocity
(r)
fluctuations.[45] Two-dimensional instantaneous radial energy spectrum (Ejj , j = 1, 2, 3)
is then calculated from averaging 2D energy spectra over an annular region with inner
radius Kr,i and outer radius Kr,o :

K12 +K22 <Kr,o


2

(r) Lr 
Ejj (κr ) = 2π 2D
Ejj (k1 , k2 )Kr,i (Kr,o − Kr,i ) , (5)
P
K12 +K22 ≥Kr,i
2

where Lr is the radial domain length and P is the number of points in each annular domain.
The instantaneous radial energy spectra are then averaged to obtain an ensemble average.
Two-dimensional DFT of each FOV is calculated for a 85 × 48 rectangular region (that is
shown in white in Figure 6) in order to remove the effect of near rotor region in FOVs 1
and 2 as well as the effect of non-accurate data in the boundary of each FOV.

5. Results and discussion


Spectral characteristics are first examined to specify the relative location of filter wave
number with respect to energy cascade and particularly the ISR. The estimated viscous
dissipation is then investigated and compared with other terms of TKE equation.

5.1. Spectral characteristics


Figure 7 shows the 2D radial energy spectra based on fluctuations of velocity in near- and
far-rotor regions (FOVs 2 and 4, respectively). Spectra for other FOVs are also similar
(not shown here) and confirm the following conclusions. The spectrum starts proportional
to κ 1 in the energy containing range (low wave numbers).[38,51] The slope then changes
to steeper values and eventually for larger wave numbers to −5/3. The spectra show that
the ISR (−5/3 slope) starts at a wave number about κ = 1500 rad/m. The spectra slope
experiences noise at wave numbers greater than 2800 rad/m, which may be due to spectral
noise or aliasing (spectral tail up).
The maximum of spectra for FOV2 occurs at κ ∼ 280 rad/m, which corresponds to a
spatial scale of 22.4 mm. The blade-to-blade distance, which is approximately 26 mm, is
therefore an important contributor to the size of the most energetic coherent structures in
(r)
the rotor exit region. Maximum of Eii (i = u, v, w) in the volute region (FOVs 3 and 4)
is also comparable with volute dimension and occurs at larger scales as compared with that
of FOVs 1 and 2.
514 M.R. Najjari et al.

102 102

1 1
101 κ 101 κ
Eii (m /s )

Eii (m /s )
2

2
3

3
(r)

(r)
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

100 100
(r)
E
uu
Euu(r)
−5/3
κ
(r)
E
vv Evv(r)
−5/3
E
ww
(r)
Eww(r) κ
-1 -1
10 10
0

00

00

25 0
3000
00

00

00

25 0
3000
00
50

50

0
10

15
20

10

15
20
κ (rad/m) κ (rad/m)
(a) (b)

Figure 7. Two-dimensional radial energy spectra for three velocity components in (a) FOV 2 and
(b) FOV 4.

The smallest resolved scale from the SPIV measurements is about 35η (wave num-
ber 3140 rad/m) and the considered filter length scales for energy flux calculations are
3δ, 5δ and 7δ (2094 rad/m, 1257 rad/m, 898 rad/m). As we recognised from instanta-
neous energy spectra as well as ensemble averaged energy spectra (Figure 7), the three
implemented filters often locate around −5/3 slope that is expected to be the ISR. It will
be discussed in the following section that a part of ISR erases due to non-local energy
transfer and the slope of energy spectra deviates from −5/3 to steeper slopes. The validity
of estimation of viscous dissipation by energy flux through ISR (Equation (4)) in such
situation is evaluated in the next section.

5.2. Viscous dissipation and spectral shortcut


Figure 8 shows the ensemble average of estimated viscous dissipation as calculated from
SGS dissipation. According to the classical energy cascade, viscous dissipation should be
positive everywhere. However, this figure shows that there are a few regions with negative
viscous dissipation (shown in white). This is due to the comparable share of backward
energy transfer (energy scattering from smaller scales to larger ones) in every instantaneous
snapshot. Figure 9 further clarifies the point and shows the probability  density
 function
(PDF) of instantaneous SGS energy flux (dissipation), εSGS inst
= −τij Sij , for the entire
measurement area and snapshots. This figure indicates that about 40% of instantaneous
samples have backward scattering of energy. This energy transfer mechanism opposes the
direct energy cascade and its possible consequence can be underestimation of viscous
dissipation from LE-PIV. It will be shown in Section 5.3 that this is a likely scenario for
this study.
Further investigation of this issue reveals that the backscattering is highly correlated with
the vortices in the flow. An algorithm on the basis of 2D λ2 criterion [52] is implemented to
detect the vortical regions in instantaneous velocity vector maps. The PDF of SGS energy
flux is then extracted for either vortical or non-vortical regions for 800 instantaneous vector
maps and the results are presented in Figure 10 for a near-rotor FOV, i.e. FOV2 (a similar
pattern is observed for the other FOVs and not shown here). Table 3 summarises the share
of forward and backward scatters in vortical and non-vortical regions for various FOVs.
Journal of Turbulence 515
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

Figure 8. Estimation of viscous dissipation by the ensemble average of SGS energy flux for = 3δ.
Negative values indicate backward scattering of energy and are shown in white.

0.1

0.08

0.06
PDF

Backward Scatter Forward Scatter


40.2 % 59.8%
0.04

0.02

-0.03 -0.02 -0.01 0 0.01 0.02 0.03


εSGS
inst r / U 3
tip tip

Figure 9. Probability density function of instantaneous SGS energy flux for the entire FOVs and
snapshots.

Table 3. The share of forward and backward scattering in vortical and non-vortical regions.

FOV no. FOV 1 FOV 2 FOV 3 FOV 4


Energy
scattering Forward Backward Forward Backward Forward Backward Forward Backward

Non-vortical 0.73 0.27 0.81 0.19 0.80 0.20 0.80 0.20


region
Vortical 0.48 0.52 0.43 0.57 0.52 0.48 0.50 0.50
region
516 M.R. Najjari et al.
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

Figure 10. Probability density function of SGS energy flux in the vortical and non-vortical regions
of FOV2.

Figure 11. Schematic of the energy spectra in the presence of spectral shortcut mechanism. Size
of spirals indicates the largeness of flow scales and the arrow sizes are proportional to transferred
energy between different scales (modified from Poggi and Katul [25]).

These statistical distributions show that the contribution of backward scattering in energy
transfer mechanism intensifies in vortical regions. In other words, its share increases from
19%–27% in non-vortical regions to 48%–57% in vortical regions. Consequently, presence
of vortices and their interactions significantly affects the energy transfer mechanism and
modifies the TKE cascade.
Spectral shortcut is another energy transfer mechanism that, if it exists, affects the
energy cascade and hence the conclusions of LE-PIV approach. Figure 11 presents a
Journal of Turbulence 517

3
Table 4. Spatial average of normalised SGS energy flux (εSGS Rtip /Utip ) for three
different filter scales.

FOV/filter size 3δ 5δ 7δ

1 0.0045 0.0085 0.0117


2 0.0030 0.0053 0.0069
3 0.0009 0.0018 0.0024
4 0.0007 0.0015 0.0021
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

schematic of energy cascade in the presence of spectral shortcut. Arrow thickness represents
transferred energy between different scales and spiral size shows the size of flow scales.
A demonstration of SGS energy flux in the presence of spectral shortcut depends on the
filter size as well as the location of filter within ISR. Larger eddies at the initial section
of ISR (lower wave numbers) transfer part of their energy to smaller scales through direct
cascading and the rest through spectral shortcut. The SGS energy flux only accounts for
direct cascading. If the spectral shortcut exists in the present flow, the SGS energy flux
should reduce for smaller filter size at the initial region of ISR.
Table 4 presents the effect of filter size on SGS energy flux for four FOVs. The trend
shows that at the initial section of ISR the energy flux decreases with reducing filter size.
Therefore, there is a removal of energy within that range of scales and it is likely that part
of their energy does not go through the direct cascade and instead transfers via spectral
shortcut. A similar conclusion was previously drawn from the high-wave number end of
ISR, close to dissipative scales.[29] It was shown there that the energy flux increases with
reducing filter size thus, smaller eddies at the end section of ISR and start of dissipation
range gain energy from larger eddies through either direct energy cascade or spectral
shortcut.
Further evaluation of estimating viscous dissipation by LE-PIV (εSGS ) approach is in the
next section within the TKE equation. The non-dissipative TKE terms are introduced first
and then a discussion on viscous dissipation and contribution of spectral shortcut follows.

5.3. Balance of the turbulent kinetic energy equation


The TKE equation is written as [38]

Ck Tk P

   k
     
k

∂ q2 ∂ q2 ∂  p ∂  q2 ∂U j
= −Ui − ui − u i −u i u j ,
∂t 2 ∂xi 2 ∂xi ρ ∂xi 2 ∂xi
Dk εk
    
  
   
 
∂  ∂u i ∂u j ∂u i ∂u j ∂u j
+υ uj + −υ + (6)
∂xi ∂xj ∂xi ∂xj ∂xi ∂xi

where q 2 = ui ui , p is the pressure fluctuation, υ is the kinematic viscosity and ∂ϕ/∂xi =
∂ϕ/∂xi for any quantity ϕ. The last right-hand side term is viscous dissipation (εk ) and
the first five terms are non-dissipative terms including convection (Ck ), pressure diffusion
(k ), turbulence diffusion (Tk ), production (Pk ) and viscous diffusion (Dk ), respectively.
518 M.R. Najjari et al.
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

Figure 12. The ensemble average of non-dimensional turbulent convection term. White regions
correspond to absolute values smaller than 0.01.

Among the nine components in each of the convection, turbulence diffusion and pro-
duction terms, six components include in-plane gradients and are directly calculable from
the present data. One component in each of these terms includes out-of-plane gradient of
normal velocity component (e.g. U3 ∂u∂z3 u3 in the convection term) and may be calculated
from continuity equation for incompressible flow. Consequently, only two components for
each of convection, turbulence diffusion and production terms are unavailable and should
be estimated. According to the discussion in Section 3 about calculation of viscous dissi-
pation from SPIV data, the two unavailable components of Ck , Tk and Pk are compensated
by multiplying the results by a 9/7 scalar. Viscous diffusion is much smaller than other
terms and is ignored in many previous studies on TKE budget.[1,7,30,53] Direct calcu-
lation of pressure term from PIV data is not possible and some previous studies neglect
the pressure term.[54] This study obtains the pressure term plus viscous dissipation from a
balance of other terms in TKE equation similar to previous studies.[1,7,30] We are aware
of the involved approximation due to these assumptions, but more accurate measurements
are rare and expensive and currently many researchers suffice to what may be gained
from such data.
Figure 12 presents the ensemble average of non-dimensional turbulent convection in
four FOVs. A positive TKE term indicates energy gain, while a negative term means energy
loss. Regions with absolute values less than 0.01 are shown in white. The absolute value of
convection, which is the portion of transferred TKE via mean flow, decreases away from
the rotor. Local maximum and minimum convections occur to the left of wakes and to the
right of jets, respectively. Non-symmetric trends in convection in the rotor exit flow field
could be due to an asymmetric pressure term and the complicated non-linear interactions
among rotating jet and wake structures and volute flow.[8]
Figure 13 shows turbulence diffusion in all four FOVs, which is generally positive in
jet and wake centrelines and negative in jet–wake interaction regions. This means that it is
a source term in jet/wake centrelines that diffuses energy through velocity fluctuations.
The ensemble average of non-dimensional production term is presented in Figure 14
for all four FOVs. Production is proportional to the work of mean velocity gradient against
Journal of Turbulence 519
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

Figure 13. The ensemble average of non-dimensional turbulence diffusion term. White regions
correspond to absolute values smaller than 0.001.

Figure 14. The ensemble average of non-dimensional production term. White regions correspond
to absolute values smaller than 0.0015.

Reynolds stress.[55] Velocity gradients are small on jet and wake centrelines and conse-
quently, minimum production occurs along these trajectories. The production is positive in
regions that have reached equilibrium between turbulent structures and strain rate.[56] The
shear layers between jet and wake structures in Figure 14 are the examples of such regions
that correspond to maximum production. However, in highly transient and intermittent re-
gions of flow, the strain rate experiences severe variations and this reduces or cancels out the
local equilibrium between flow structures and strain rate, which in turn results in near-zero
or even negative values of production. As Figure 14 shows, this particularly occurs in some
local regions in the volute that are due to jet–wake interaction with the cyclic volute flow.
520 M.R. Najjari et al.
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

Figure 15. Variation of convection and turbulence diffusion terms on two circular arcs: one in FOV 2
and another in FOV 4. R1 = 1.025 Rtip and R2 = 1.5 Rtip .

Figure 16. Variation of turbulent kinetic energy terms on the circular arc shown in Figure 5 from
jet #1 to jet #2. ‘J’ and ‘W’ indicate jet and wake, respectively.

Figure 15 shows convection and turbulence diffusion terms on circular arcs at R1 =


1.025 Rtip and R2 = 1.5 Rtip . Generally diffusion is stronger in flow fields with low mean
momentum and intensive fluctuations, while convection is more important for a dominant
mean flow momentum. Flow expansion results in smaller mean flow momentum in the
volute than that of the rotor exit region. Hence convection decreases more considerably
than diffusion away from the rotor (Figure 15).
Figure 16 shows the variation of all TKE terms on the circular arc at R = 1.025 Rtip
(Figure 6) in FOV1. Although the turbomachinery flow field is expected to be considerably
dissipative, viscous dissipation based on SGS energy flux is negligible as compared with
other TKE terms. This should not be surprising when combined with two further obser-
vations in this paper: (1) the presence of backward scattering and its opposition to the
Journal of Turbulence 521

classical energy transfer direction and (2) reduction of SGS energy flux for smaller filter
scales around the initial section of ISR (Section 5.2).
The possible scenario that sums up all these three observations is that non-linear inter-
actions between jet–wake–volute structures transfer a considerable portion of dissipative
energy away from the direct cascade via spectral shortcut and/or backward-scattering mech-
anisms. This has reduced the transferred energy as calculated from LE-PIV method and
has left a larger portion of energy unaccounted.
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

6. Conclusions
The three-dimensional instantaneous velocity field in the rotor exit region of a forward-
carved centrifugal turbomachine is obtained using SPIV. This provides the basis for under-
standing the dominant components of energy budget equation and analysing the possible
energy transfer mechanisms.
The TKE terms are calculated by relying on the available components of velocity gra-
dient from SPIV data. The smallest resolvable scale of this study is approximately 35 times
larger than the Kolmogorov length scale and the viscous dissipation is therefore indirectly
estimated via SGS energy flux (dissipation). Sum of pressure and viscous dissipation terms
is calculated from TKE equation balance and other TKE terms are directly calculated
from SPIV data without considering the simplifying assumptions such as isotropy of flow
structures. The main concluding remarks of the present research are as follows.

• Flow structures are directly affected by the rotor-induced asymmetry of pressure


field. The consequences of this asymmetry are generation of non-uniform jet/wake
interactions as well as creation of an asymmetric field for the convection term as a
major contributor of TKE.
• Jet–wake interacting regions (shear layers) are the main sources of turbulence en-
ergy generation that result in maximum turbulence production in those regions. The
production term has a larger maximum along the rotor pressure side as compared
with that of the suction side. This is expected to be due to the presence of more ener-
getic flow structures along the rotor pressure sides and consequently more intensive
jet–wake interactions in these regions.
• The calculated averaged viscous dissipation on the basis of SGS energy flux is
negative in a few regions of the rotor exit regions. Instantaneous calculations show
that the backward scattering of energy in an opposite direction to classical dissipative
cascade contributes to this behaviour and results in reduction of averaged SGS energy
flux. Detection of vortical regions in instantaneous velocity vector maps on the basis
of 2D λ2 criterion shows that the backscattering is highly correlated with the vortices
in the flow.
• Negligibility of estimated viscous dissipation from SGS energy flux as compared
with the calculated TKE terms plus reduction of SGS energy flux with decreasing
filter scale (when the filter scale is around the initial section of ISR) also implies the
presence of spectral shortcut mechanism. The larger eddies in such situation transfer a
significant part of turbulent energy to fine scale eddies via an indirect energy transfer
mechanism.

The final conclusion of the paper is that correct estimation of viscous dissipation by
SGS energy flux requires one extra condition in addition to those introduced in [22], which
are the accepted norms. There should be neither bypass nor backward mechanisms in TKE
522 M.R. Najjari et al.

transfer, since they affect the direct energy cascade. Dominancy of indirect energy transfer
mechanisms such as spectral shortcut and deviation from the classical energy cascade are
important features that should also be predicted by turbulence models.

Disclosure statement
No potential conflict of interest was reported by the authors.
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

References
[1] Liu X, Thomas FO. Measurement of the turbulent kinetic energy budget of a planar wake flow
in pressure gradients. Exp Fluids. 2004;37:469–482.
[2] Chow YC, Uzol O, Katz J. Flow nonuniformities and turbulent “hot spots” due to wake-blade
and wake-wake interactions in a multi-stage turbomachine. J Turbomachinery. 2002;124:553–
563.
[3] Ubaldi M, Zunino P, Ghiglione A. Detailed flow measurements within the impeller and the
vaneless diffuser of a centrifugal turbomachine. Exp Thermal Fluid Sci. 1998;17:147–155.
[4] Sarraf C, Nouri H, Ravelet F, Bakir F. Experimental study of blade thickness effects on the
overall and local performances of a controlled vortex designed axial-flow fan. Exp Thermal
Fluid Sci. 2011;35:684–693.
[5] Akbari G, Montazerin N. A-priori study of subgrid-scale models for the flow field in the rotor
exit region of a centrifugal turbomachine. Int J Heat Mass Transfer 2013;66:423–439.
[6] Sinha M, Katz J, Meneveau C. Quantitative visualization of the flow in a centrifugal pump with
diffuser vanes–II: addressing passage-averaged and large-eddy simulation modeling issues in
turbomachinery flows. J Fluids Eng. 2000;122:108–116.
[7] Piirto M, Saarenrinne P, Eloranta H, Karvinen R. Measuring turbulence energy with PIV in a
backward-facing step flow. Exp Fluids 2003;35:219–236.
[8] Yue W, Meneveau C, Parlange MB, Zhu W, Kang HS, Katz J. Turbulent kinetic energy budgets
in a model canopy: comparisons between LES and wind-tunnel experiments. Environ Fluid
Mech. 2008;8:73–95.
[9] Hsiao SC, Hsu TW, Lin JF, Chang KA. Mean and turbulence properties of a neutrally buoyant
round jet in a wave environment. J Waterway Port Coast Ocean Eng. 2011;137:109–122.
[10] Baldi S, Yianneskis M. On the direct measurement of turbulence energy dissipation in stirred
vessels with PIV. Ind Eng Chem Res. 2003;42:7006–7016.
[11] Doron P, Bertuccioli L, Katz J, Osborn TR. Turbulence characteristics and dissipation es-
timates in the coastal ocean bottom boundary layer from PIV data. J Phys Oceanography.
2001;31:2108–2134.
[12] Kresta S, Wood PE. The flow field produced by a pitched blade turbine: characterisation of the
turbulence and estimation of the dissipation rate. Chem Eng Sci. 1993;48:1761–1774.
[13] Xu D, Chen J. Accurate estimate of turbulent dissipation rate using PIV data. Exp Thermal
Fluid Sci. 2013;44:662–672.
[14] Liberzon A, Gurka R, Sarathi P, Kopp GA. Estimate of turbulent dissipation in a decaying grid
turbulent flow. Exp Thermal Fluid Sci. 2012;39:71–78.
[15] Antonia RA. On estimating mean and instantaneous turbulent energy dissipation rates with hot
wires. Exp Thermal Fluid Sci. 2003;27:151–157.
[16] Chow YC, Uzol O, Katz J, Meneveau C. Experimental study of the structure of a rotor wake in
a complex turbomachinery flow. 4th ASME-JSME Joint Fluids Engineering Conference; 2003
July 6–11; Honolulu (HI: ASME).
[17] Pedersen N, Larsen PS, Jacobsen CB. Flow in a centrifugal pump impeller at design and
off-design conditions–part I: particle image velocimetry (PIV) and laser Doppler velocimetry
(LDV) measurements. J Fluids Eng. 2003;125:61–72.
[18] Sinha M, Katz J. Quantitative visualization of the flow in a centrifugal pump with diffuser
vanes–I: on flow structures and turbulence. J Fluids Eng. 2000;122:97–107.
[19] Uzol O, Chow YC, Katz J, Meneveau C. Experimental investigation of unsteady flow field
within a two-stage axial turbomachine using particle image velocimetry. J Turbomachinery.
2002;124:542–552.
Journal of Turbulence 523

[20] Luznik L, Gurka R, Nimmo Smith WAM, Zhu W, Katz J, Osborn TR. Distribution of energy
spectra, Reynolds stresses, turbulence production and dissipation in a tidally driven bottom
boundary layer. J Phys Oceanography. 2007;37:1527–1550.
[21] Nimmo Smith WAM, Kats J, Osborn TR. The effect of waves on subgrid-scale stresses,
dissipation and model coefficients in the coastal ocean bottom boundary layer. J Fluid Mech.
2007;583:133–160.
[22] Sheng J, Meng H, Fox RO. A large eddy PIV method for turbulence dissipation rate estimation.
Chem Eng Sci. 2000;55:4423–4434.
[23] Kilander J, Rasmuson A. Energy dissipation and macro instabilities in a stirred square tank
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

investigated using an LE PIV approach and LDA measurements. Chem Eng Sci. 2005;60:6844–
6856.
[24] Finnigan J. Turbulence in plant canopies. Annu Rev Fluid Mech. 2000;32:519–571.
[25] Poggi D, Katul GG. Two-dimensional scalar spectra in the deeper layers of a dense and uniform
model canopy. Boundary-Layer Meteorol. 2006;121:267–281.
[26] Yue W, Parlange MB, Meneveau C, Zhu W, van Hout R, Katz J. Large-eddy simulation of plant
canopy flows using plant-scale representation. Boundary-Layer Meteorol. 2007;124:183–203.
[27] Zhu W, van Hout R, Luznik L, Kang HS, Katz J, Meneveau C. A comparison of PIV measure-
ments of canopy turbulence performed in the field and in a wind tunnel model. Exp Fluids.
2006;41:309–318.
[28] Zhu W, van Hout R, Katz J. On the flow structure and turbulence during sweep and ejection
events in a wind-tunnel model canopy. Boundary-Layer Meteorol. 2007;124:205–233.
[29] van Hout R, Zhu W, Luznik L, Katz J. PIV measurements in the atmospheric boundary layer
within and above a mature corn canopy. Part I: statistics and energy flux. J Atmospheric Sci.
2007;64:2805–2824.
[30] Sideridis A, Yakinthos K, Goulas A. Turbulent kinetic energy balance measurements in the
wake of a low-pressure turbine blade. Int J Heat Fluid Flow. 2011;32:212–225.
[31] Akbari G, Montazerin N, Akbarizadeh M. Stereoscopic particle image velocimetry of the flow
field in the rotor exit region of a forward-blade centrifugal turbomachine. Proc Inst Mech Eng,
Part A. 2011;226:163–181.
[32] Akbari G, Montazerin N. On the role of anisotropic turbomachinery flow structures in inter-
scale turbulence energy flux as deduced from SPIV measurements. J Turbulence 2013;14:44–
70.
[33] Montazerin N, Damangir A, Mirian S. A new concept for squirrel-cage fan inlet. Proc Inst
Mech Eng, Part A. 1998;212:343–349.
[34] Montazerin N, Damangir A, KazemiFard A. A study of slip factor and velocity components at
the rotor exit of forward-curved squirrel cage fans, using laser Doppler anemometry. Proc Inst
Mech Eng, Part A. 2001;215:453–463.
[35] Nikkhoo M, Montazerin N, Damangir A, Samian RS. An experimental study of leaning blades
on the half-cone rotor of a squirrel cage fan. Proc Inst Mech Eng, Part A. 2009;223:973–980.
[36] RezaeiNiya SM, Montazerin N, Damangir A, Dehkordi AH. Performance and laser Doppler
anemometry experimental investigation of squirrel cage fans with half-cone rotors. Proc Inst
Mech Eng, Part A. 2006;220:753–763.
[37] The International Organization for Standardization. International Standard ISO 5801, Industrial
fans, fan performance testing using standardized airways; 2007.
[38] Hinze JO. Turbulence. New York (NY): McGraw-Hill; 1975.
[39] Zhou G, Kresta S. Distribution of energy between convective and turbulent flow for three
frequently used impellers. Trans Inst Chem Eng. 1996;74:379–389.
[40] Dantec Dynamics A/S. FlowManager software and introduction to PIV instrumentation.
Skovlunde (DK): Dantec Dynamics A/S; 2002.
[41] Westerweel J. Fundamentals of digital particle image velocimetry. Meas Sci Technol.
1997;8:1379–1392.
[42] Schroeder A, Willert CE. Particle image velocimetry. Berlin: Springer; 2008.
[43] Saarenrinne P, Piirto M, Eloranta H. Experiences of turbulence measurement with PIV. Meas
Sci Technol. 2001;12:1904–1910.
[44] Najjari MR, Montazerin N, Akbari G. Statistical PIV data validity for enhancement of velocity
driven parameters in turbomachinery jet/wake flow. 20th Annual International Conference on
Mechanical Engineering-ISME2012; 2012 May 16–18; Shiraz; 2012.
524 M.R. Najjari et al.

[45] Liu S, Meneveau C, Katz J. On the properties of similarity subgrid-scale models as deduced
from measurements in a turbulent jet. J Fluid Mech. 1994;275:83–119.
[46] Liu S, Katz J, Meneveau C. Evolution and modeling of subgrid scales during rapid straining
of turbulence. J Fluid Mech. 1999;387:281–320.
[47] Chen J, Katz J, Meneveau C. Implication of mismatch between stress and strain-rate in tur-
bulence subjected to rapid straining and destraining on dynamic LES models. J Fluids Eng.
2005;127:840–850.
[48] Chen J, Meneveau C, Katz J. Scale interactions of turbulence subjected to a straining–
relaxation–destraining cycle. J Fluid Mech. 2006;562:123–150.
Downloaded by [George Washington University], [Mohammad Reza Najjari] at 09:43 06 March 2015

[49] Hong J, Katz J, Meneveau C, Schultz MP. Coherent structures and associated subgrid-scale
energy transfer in a rough-wall turbulent channel flow. J Fluid Mech. 2012;712:92–128.
[50] Wu H, Miorini RL, Katz J. Analysis of turbulence in the tip region of a waterjet pump rotor.
ASME third Joint US-European Fluids Engineering Meeting; 2010 Aug 1–5; Montreal (QC):
ASME; 2010.
[51] Garde RJ. Turbulent flow. New Delhi: New Age International (P) Ltd; 2000.
[52] Schram C, Rambaud P, Riethmuller ML. Wavelet based eddy structure education from a back-
ward facing step flow investigated using particle image velocimetry. Exp Fluids. 2004;36:233–
245.
[53] Hussein HJ, Capp SP, George WK. Velocity measurements in a high-Reynolds-number,
momentum-conserving, axisymmetric, turbulent jet. J Fluid Mech. 1994;258:31–75.
[54] Panchapakesan NR, Lumley JL. Turbulence measurements in axisymmetric jets of air and
helium, Part 1: air jet. J Fluid Mech. 1993;246:197–223.
[55] Pope SB. Turbulent flows. Cambridge: Cambridge University Press; 2000.
[56] Gatski TB, Hussaini MY, Lumley JL. Simulation and modeling of turbulent flows. New York
(NY): Oxford University Press; 1996.

View publication stats

You might also like