Development of Turbulence Intensity and Integral Length-Scale in a

1.5 Stage Axial Flow Turbine

Stephan Behre1 , Robert Kluxen1 , Peter Jeschke1 and Yavuz Guendogdu2

1 Institute of Jet Propulsion and Turbomachinery, RWTH Aachen University

Templergraben 55, 52062 Aachen, Germany
2 MTU Aero Engines GmbH
Dachauer Strasse 665, 80955 Munich, Germany

ABSTRACT
This paper describes experimental investigations conducted on a Subscripts
1.5 stage axial flow turbine in order to determine the turbulence in- 3D Turbulent quantity based on all spatial directions
tensity and integral length-scale development along the main flow E Approximated e-function
path. The turbulence level is evaluated using triple hot-wire probes EA Ensemble average method
which enable the 3D velocity vector to be measured by means of FFT Chopped FFT method
Constant-Temperature-Anemometry. Turbulence generating grids grid Turbulence grid
have been applied to introduce a definite level of velocity fluctua- i Number of samples per rotor revolution
tions. Three equidistant allocated measurement planes downstream j Number of rotor revolutions
of the grid enable the development of the turbulence quantities with- n Normalised
out airfoil influence to be measured. Furthermore, an investigation r Radial direction
is made of the turbulence decay between the blades’ axial gap by S1 First stator
measuring the turbulence level close to the trailing edge of the first sw Streamwise component
stator, in the center of the gap and close to the leading edge of the u Circumferential direction
subsequent rotor row. In all planes introduced, 2D-traverses have wf with blade passing frequencies
been performed. Two different methods for determining the tur- wo f without blade passing frequencies
bulence intensity are presented and compared, as well as a simple x Streamwise direction
approximation function for the autocorrelation.
Abbreviations
NOMENCLATURE ACF Autocorrelation function
BPF Blade passing frequency
a Coefficient for approximation function DNS Direct numerical simulation
b Coefficient for approximation function S1 First stator
c Absolute velocity S2 Second stator
c̄ Time averaged velocity R Rotor
c̃ Ensemble averaged velocity component
c′ Stochastic velocity component
INTRODUCTION
~c Velocity vector
f Blade passing frequency Since Direct Numerical Simulations (DNS) are still unsuitable
fs Sampling frequency for simulating multi stage compressors or turbines, in most aca-
fvK von Karman vortex street frequencies demic and industrial applications steady and unsteady Reynolds-
LTu Integral length-scale averaged Navier-Stokes equations are used to determine the flow
lg Distance between grid and S1 field within turbomachineries. The tremendous reduction in numer-
lgap Distance between S1 and R ical costs can only be achieved because small-scale structures in the
Rxx Normalised autocorrelation function flow, defined as turbulence, are - in contrast to DNS - not directly re-
rn Normalised radial coordinate solved, but modeled. Evidently, the coarse flow resolution leads in
T /τ Time turn to a loss in accuracy, in particular with respect to the turbulence
TTu Integral time-scale quantities . As a consequence, detailed experimental validations are
∆T Inverse of fs needed to evaluate the reliability and performance of the turbulence
t Pitch models used.
α Circumferential angle The majority of the models interpret turbulence as an elevated
γ Radial angle viscosity, the so-called ”eddy viscosity” . There are a number of op-
ω Vorticity tions available to determine eddy viscosity: up until now, the most
commonly used models are based on two partial differential equa-
tions. One is used to determine the turbulence kinetic energy k, and
the other the turbulence dissipation rate ω . The models introduced
by Menter in 1994 [1] and Wilcox in 1988 [2] are used in most
aerodynamic applications. In order to solve the Reynolds-averaged
 Grid MP02 MP2
Table 1: Turbine specifications
MP01 MP03 MP1 MP3

Stator 1 (S1) Rotor (R) Stator 2 (S2)

Blade count [-] 33 44 66 Inlet S1 R S2 Outlet
Reynolds number [-] 8.1 · 105 5.2 · 105 3.3 · 105 lg
Aspect ratio h/l 0.95 1.4 1.51

Navier-Stokes equation, boundary conditions are necessary. In gen-

eral, the boundary conditions for the turbulence quantities are given Fig.1: Overview of the available measurement planes (MP)
either in form of turbulence intensity and length-scale, or turbulence
kinetic energy and dissipation rate.
According to Pope [3], the length-scale is linked to the turbulence MP10 MP11 MP12
dissipation rate which in turn has an influence on the decay of turbu-
lence within the flow. This means that inaccurate estimations of the
length-scales will lead to over- or underestimated dissipation rates
for the turbulence with the result - even though correct turbulence
intensities have been given as boundary conditions for numerical
investigations - that incorrect turbulence levels will be determined
in the flow of the turbomachinery. Since the free-stream turbulence,
particularly for small Reynolds numbers Re < 2 · 105 , has a con- S1 R S2
siderable effect on the behavior of the boundary layer, as well as
the acceptable positive pressure gradient at which separation occurs
[4], it is important to make a proper evaluation of experimental data lgap
in order to specify the correct quantities of turbulence. Detailed in-
vestigations of boundary layer behavior, when subjected to different Fig.2: Detailed measurement planes behind the first stator (S1)
free-stream turbulence levels, can be found in [5]. In addition to the
influence of the free-stream, the increased turbulence level within
the blade wakes passing through the subsequent row also has an Table 2: Grid specifications
impact on the boundary layer of the following blade. Bypass transi-
tion is the dominant transition process in most turbomachinery. For
Grid G1 Grid G2
that reason correct modeling of the turbulence dissipation within
the wake is crucial, meaning that there is a need to investigate the Rod diameter [mm] 1 3
turbulence decay between adjacent rows. Number of rods [-] 264 165
Pitch at rn = 50% [-] 6.5 10.4
Another important aspect making turbulence measurements and
Solidity [%] 16 30
modeling essential, is the impact of turbulence levels on heat trans-
fer. For example, Butler et al. [6] revealed that increased turbulence
levels lead to increased stagnation heat transfer on the leading edge.
Similar results have also been found by Carullo at al. [7]: they were
able to demonstrate increased heat transfer along the suction and Evaluated Measurement Planes
pressure side in a transonic cascade when free-stream turbulence Detailed flow measurements were conducted on the measuring
is increased. Evidently, turbulence measurements and proper post- plane MP01 - MP03, as illustrated in figure 1. In order to investi-
processing of experimental data is crucial in improving numerical gate the turbulence decay, including the influence of the blade, the
methods. In this paper therefore, detailed hot-wire measurements measurement planes MP10 to MP12 were used. All labeled loca-
have been taken in front of the first stator and within the axial gap tions allow circumferential, as well as radial traverses. The first
between the first stator and the rotor. The measurements allow the three measurement planes consist of a total of 1066 measurement
evaluation of turbulence intensities and integral length-scales so that points, allowing a high radial resolution of 1.25mm and 0.2◦ in
the decay rate of turbulence without blade influence and with the in- the circumferential direction. For the larger measuring field in the
fluence of a rotating blade can be determined . The post-processing planes within the axial gap, a total of 1763 measuring points were
methods applied are explained in detail and a simple approximation used. Keeping the radial resolution constant, the circumferential
method for the integral length-scale is also given. distance between each radial traverse is 0.33◦ . The measuring grids
described enable more than 3 rod pitches to be captured for G1 and
more than 2 pitches for G2. Within the turbine core measurement
EXPERIMENTAL SETUP of 1.3 S1 pitches are possible.
The 1.5 stage axial flow turbine rig under investigation, a cross- The notation of the different measurement planes, as well as their
section of which is given in figure 1, is operated and analyzed exper- position, is given in tables 3 and 4. The lengths between the rows
imentally. The configuration enables the investigation of different are normalized by the length of axial gap. The measurement plane
turbulence grids located at the beginning of the cylindrical annu- locations upstream of the first stator are given as a percentage of
lus. An overhung hub avoids the need for using struts, so allowing the total distance between the turbulence generator and the leading
the undisturbed measurement of the turbulence decay in front of the edge of the first stator. To get a clearer overview of the measurement
first stator. A short description of the turbine’s most important fea- planes, the turbine’s core is illustrated in figure 2.
tures is given in table 1. For further details, the reader is referred to
[8] or [9]. Table 2 gives a short overview of the most important grid Measurement Technique
parameters for the turbulence generating grids G1 and G2 under in- Turbulence quantities - defined in the next section - can only
vestigation. Only radially oriented, cylindrical rods have been used. be evaluated using unsteady measurement techniques. The results
presented throughout this paper have been attained using hot-wire
 Table 3: Position of measurement planes MP01 - MP03 proposed by [12] was applied. For each hot-wire probe square wave
tests were conducted in order to estimate the cut-off frequency, de-
Normalized by lg fined in [13]. For each wire of the given probes, the cut-off fre-
Notation MP01 MP02 MP03 quency fc was measured to be in the range of 19 to 20kHz, which
Position [%] 24 45 66 is sufficiently high to capture all the necessary effects. For data ac-
quisition, a 16bit A/D converter with a maximum sample rate of
250kHz was used which enabled a maximum of 1, 000, 000 mea-
Table 4: Position of measurement planes MP10 - MP12 surement points to be recorded within 4 seconds. During data ac-
quisition, a trigger signal - crucial for ensemble averaging - gave a
voltage impulse per revolution and was recorded, likewise to deter-
Normalized by lgap
mine the position of each rotor blade.
Notation MP10 MP11 MP12
Position [%] 24 53 82 DEFINITION OF TURBULENT QUANTITIES
In the following section, a detailed description of the turbulence
quantities used and their calculation processes is given. Two of the
most important quantities, defining the turbulence character of a
anemometry, so enabling the investigator to calculate the random flow, are the turbulence intensity Tu and the integral length-scale
velocity field. In this paper, experimental results were attained us- LTu . The determination and calculation of both will be discussed
ing a triple hot-wire probe with a head diameter of 2.2mm, illus- here.
trated in figure 3. In addition to the only 4mm long prongs, figure 3
enables the reader to clearly identify the three 9µ m thick platinized Turbulence Intensity Tu
tungsten wires. All wires are welded to the prongs in such a way In the turbine rig illustrated, the probe’s position is stationary. As
that they overlap at the center of the probe (see figure 3 (c)). The a consequence, the flow field measured by the probe is exposed to
three wires are heated to a constant temperature of 284◦C. Due to periodic changes, due to the upstream effect of the rotor. Because of
the non-uniform temperature distribution along the wire - as a re- the periodic flow character, the periodic component has to be sub-
sult of the heat loss in the prongs - the maximum temperature is tracted from the velocity signal in order to calculate the turbulence
at the center of the wire [10]. Depending on the cold wire length intensity correctly. The periodic velocity change can be determined
and the wire diameter, Brunn [10] specifies a simple formula to es- using the ensemble average method devised by Gostelow [14]. The
timate the maximum temperature at the center of the wire. For the definition of the unsteady velocity signal, used throughout the pa-
given probe, the maximum temperature was calculated at 322◦C, per, is given in Eq. (1).
still leaving a sufficient distance to the material’s oxidation tem-
c(T ) = c̃(T ) + c′ (T ) (1)
perature. The triple hot-wire probe is operated using 3 Dantec CTA
modules installed in Dantec’s StreamlinePro Frame. Using the Dan- The velocity signal c(T ) is broken down into the ensemble-
tec Streamline Software, the required settings were applied to each averaged part c̃(T ) and the stochastic part c′ (T ). For the experi-
CTA module. mental setup indicated, the sampling rate of the unsteady voltage
was set to 250 kHz. With the given sampling frequency and the
maximum number of samples introduced in the previous section,
1
233 rotor revolutions were recorded and used to determine the pe-
riodically changing velocity component c̃(T ). This means that one
2 rotor revolution can be resolved with approximately 4285 samples.
3
1 For each of the 4285 samples (denoted by the index i), the ensemble
2 average will be achieved using 233 samples (denoted by the index
j).
3 1 233
c̃i (T ) = ∑ ci j (T ) i = 1...4285
233 j=1
(2)
(a) (b) (c)
Fig.3: (a) Triple hot-wire probe - (b) Detail of probe’s head - (c) 1 233
Wire location (c′i )2 = · ∑ (ci j (T ) − c̃i (T ))2 i = 1...4285 (3)
233 j=1

In the last step the random velocity fluctuations (c′i )2 are time aver-
The calibration of the hot-wire probes was carried out in the in- aged over 4285 samples.
stitute’s free-jet wind tunnel. Due to the fact that the heat transfer
1 4285 ′ 2
along the wire - in addition to the fluid velocity and temperature (c′ )2 = · ∑ (ci ) . (4)
- depends strongly on the fluid density, calibration was carried out 4285 i=1
using the mass flow density ρ c instead of Mach number, as sug-
gested by [11]. This method takes into account the key fact that Normalised by the local time-averaged velocity component, Eq. (5)
the heat transfer along the wire is defined by the effective cooling shows the definition of the turbulence intensity Tu.
density. The procedure accounts for the obvious density differences q
between free-jet calibration and the test rig presented. For 8 differ- (c′ )2
ent mass flow densities, the circumferential angle - denoted as α - Tu = (5)
c̄
and the radial angle - denoted as γ - are varied respectively between
−20◦ and +20◦ , with an interval of 5◦ , leading in total to 648 cali- The aforementioned formulas were shown for the absolute velocity
bration points. At each point, a mean value for each voltage signal c. The experimentally investigated turbulence intensity is defined
of more than 500, 000 data points is determined and used for post- using the velocity fluctuations in each direction. c′x describes the
processing. Triple hot-wire measurements in the free-jet calibration stochastic velocity fluctuation respectively in a streamwise direc-
channel revealed a turbulence intensity of 0.5%. In order to account tion, c′u in a circumferential direction and c′r in a radial direction.
for temperature changes within the flow, the 1D-correction method The corresponding equation is given below. With respect to the
computing time, instead of a complete rotor revolution, the average the fvK were detected using a moving average filter. A detailed de-
is only attained using 487 samples, which corresponds to 5 rotor scription of the method is given in [16]. The result of the filtering
blade passings instead of 44. For the ensemble averaged turbulence process is shown in figure 4 by means of a single measurement point
intensity Tu3D a total of 113471 samples has been used. in measurement plane MP12. The filter used clearly detects all har-
s monic frequencies and no other frequencies will be deleted. This
1 (c′x )2 + (c′u )2 + (c′r )2 ensures that no information within the time signal - except the pe-
Tu3D = (6) riodic one - gets lost. The third picture shows the ACF calculated
c̄ 3 with (w f ) and without (wo f ) the blade passing frequency and its
Integral Length-scale LTu harmonics. After removing particular frequencies, the ACF shows
The second key turbulence quantity defining the character of a a typical behavior and the integral time-scale can be calculated. All
turbulence flow is the turbulence length-scale LTu . The size of the integral length-scales presented here are calculated using the fluctu-
energy-containing eddies can be estimated using the Taylor frozen ations of the absolute velocity c′ (T ) and a total number of 200, 000
turbulence hypothesis. At this point, it should be stated that Taylor’s samples. The fluctuations c′ (T ) are computed using Eq. (10).
hypothesis is generally only valid when the fluctuating velocity is
much smaller than the free-stream velocity. Only in that specific

Velocity Amplitudechopped [m/s]

case can the assumption be reasonably made that changes in the

Velocity Amplitude [m/s]

motion of the turbulence are due simply to the passage of a frozen
pattern of turbulent motion over the fixed point [15]. This allows the
investigator to measure the fluctuations at a stationary point in space
and relate those fluctuations to the ones some distance downstream
of that point. This assumption enables the autocorrelation function
(ACF) Rxx , defined in Eq. (7), to be used [16]. The denominator
corresponds to the value of highest correlation (τ = 0). 0 5000 10000 15000 20000 0 5000 10000 15000 20000
Frequency [Hz] Frequency [Hz]
Instead of taking measurements at two different locations at the
same time, the stochastic velocity signal c′ (T ) is correlated to itself. (a) (b)
As suggested by O’Neill et al. [17], the autocorrelation function Rxx 1

is then integrated up to the first zero crossing in order to determine 0.8

ACFwo f
ACFw f
the integral time-scale TTu (see Eq. (8)). The integral time-scale
0.6
can be interpreted as a time period, for which there is a reasonably
high degree of correlation. The result of multiplying the integral 0.4
Rxx[-]

time-scale TTu and the local time-averaged velocity c̄ is called the 0.2

integral length-scale LTu . Since the presented experimental data is 0

based on time discrete signals, the integrals are replaced by sums

and the infitesimal time interval d τ is replaced by ∆T .
-0.2

-0.4

R∞ ′ 0 0.4 0.8 1.2

c (T ) · c′ (T + τ )dT
T [ms]

0 (c)
Rxx (τ ) = (7)
R∞
c′ (T ) · c′ (T )dT Fig.4: Signal preprocessing. (a) Original FFT-signal, (b) Chopped
0
FFT-signal, (c) Autocorrelation function
τ =ZτRxx =0
TTu = Rxx (τ )d τ (8)
τ =0 Approximation of LTu . The measurement data was post-
processed using the above criteria. One observation is that in free-
LTu = TTu · c̄ (9) stream regions in MP11 - characterized by a very small turbulence
With respect to Camp et al [16], the definition of the integral length- intensity - it became difficult to calculate the integral time-scale,
scale is only valid for random, continuously measured signals with- since the ACF did not cross zero (solid line figure 5 (c)). Depend-
out periodic influences. The proposed ensemble average method vi- ing on the flow region - this results in large integral length-scales of
olates the requirement for a continuously measured signal and fails up to 10 times the channel height. In contrast to regions with very
to remove the periodic parts due to the von Karman vortices shed- low turbulence, regions of higher turbulence do not show this kind
ding from the cylindrical rods of the grid with distinct frequencies. of problem. In order to estimate the scale of the integral length-
Without proper preprocessing of the random measurement signal, scale, the ACF is approximated using the approach indicated in Eq.
the ACF will oscillate, leading to a sharp and rapid decline in the (11). A similar approach has been introduced by Roach [18].
ACF (see dashed line in figure 4). The preprocess, introduced in
[16], is applied to the velocity signals which enables the integral Rxx,approx. (τ ) = a · eb·τ (11)
time-scale TTu to be determined using a long and continuous signal.
Instead of Eq. (1) for the velocity fluctuation c′ (T ), the following The coefficients a and b are determined using the least squares
calculation of c′ (T ) provides the starting point for the upcoming method and the first 15 points of the autocorrelation function. The
preprocess. number of points required depends on the sampling frequency and
c′ (T ) = c(T ) − c̄ (10) should be chosen with some caution. Because of the approxima-
tion, coefficient a is always slightly higher than 1 which implies
Before evaluating the ACF, the periodic components of the random a violation of Rxx,approx. (τ = 0) = 1. For that reaosn, values for
velocity signal are filtered out by Fourier-transforming the signal Rxx,approx. (τ ) > 1 have been set equal to 1. The approximated cor-
into the frequency domain. The amplitudes, as well as their argu- relation function is then integrated until it falls below the thresh-
ments at the blade passing frequencies and the von Karman frequen- old of 0.001. The results of the actual autocorrelation function -
cies fvK and their harmonics, have been set to zero. Since the blade subsequently labeled as ACF - and the approximated one - labeled
passing frequency, as well as its harmonics, can be easily calculated, ACFE - are shown in figure 5. In order to demonstrate the effect
of the turbulence level, the integral length-scales are calculated for of the reverse transformation back in the time domain, describes
a measuring point within the wake (P2) and the free-stream (P1) the velocity fluctuations without the periodic part. Accordingly, the
in measurement plane MP11. The values computed at both points following Eq. (12) can be used to determine the time-averaged ve-
and for both methods are summarized in table 5, normalized with locity fluctuation, using N = 200, 000 samples.
the area-averaged integral length-scale measured at the turbine inlet v
plane MPin (not shown here). u N
u1
With reference to figure 5, the autocorrelation function of mea- c = t ∑ (c′i )2
′ (12)
N i=1
MP11
310 The procedure described can be performed for all three spatial di-
300
L Tu,n[-]: 0.1 0.4 0.7 1 1.3 1.6 1.9 2.2 L Tu,E,n[-]: 0.1 0.4 0.7 1 1.3 1.6 1.9 2.2 rections, so allowing the 3D turbulence intensity Tu3D , given in
Eq. (6), to be calculated. The following illustrations (a) and (b)
290
in figure 6 show the comparison between the turbulence intensity
280
P1 P2 P1 Tu3D , determined using the FFT chopping method, subscript FFT,
y [mm]

270 P2
and the classic ensemble average method, subscript EA. In order to
evaluate the influence of the periodic parts removed, the third illus-
260
tration shows the turbulence intensity, including the blade passing
250 frequency and its harmonics (subscript FFT,wf ). The turbulence
240
intensity is likewise normalized by the area-averaged turbulence in-
0 10 20 30 40
x [mm]
50 60 70 80 0 10 20 30 40
x [mm]
50 60 70 80
tensity Tu3D at MPin .
The comparison clearly shows both evaluation methods - chopped
(a) (b)
1

P1 P2 MP11
P1 E P2 E
310
0.8

300 Tu3D,EA,n [-]: 0.05 0.25 0.45 Tu3D,FFT,n [-]: 0.05 0.25 0.45

0.6
290
Rxx[-]

0.4 280
y [mm]

270
0.2
260

0
250

0 0.5 1 1.5 2 0 0.2 0.4 0.6 0.8

T [ms] T [ms] 240
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
x [mm] x [mm]
(c) (d)
(a) (b)
Fig.5: Comparison of ACF and ACFE . (a) LTu,n , (b) LTu,E,n , (c) 310 0.4

ACF and ACFE at P1, (d) ACF and ACFE at P2 300

Tu3D,wf,n [-]: 0.05 0.25 0.45
EA
FFT

0.3
290

Tu3D,n[-]
280
y [mm]

0.2
Table 5: Integrated length-scales at P1 and P2 270

260
0.1
P1 P2 250

LTu /LTu,MPin 101 0.77 240

LTu,E /LTu,MPin 4 0.78 0 10 20 30 40

x [mm]
50 60 70 80 0 0.2 0.4 0.6
t/t S10.8
[-] 1 1.2 1.4
0

(c) (d)

surement point P1 reaches the abscissa after a long period of time, Fig.6: Comparison of the different evaluation methods. (a)
leading to non-physical integral time-scales and hence to large in- Tu3D,EA,n , (b) Tu3D,FFT,n , (c) Tu3D,w f ,n , (d) Circumferential tra-
tegral length-scales (s. table 5). In the free-stream region behind verse at rn = 50%
the blade, where the turbulence was generated far upstream, the
difference between both methods is significant. However, in re-
gions where turbulence was generated a moment ago (P2), the ACF FFT and ensemble average - to predict almost the same distribution
gives physical values and the approximated function ACFE lies in of turbulence intensity. The third illustration differs significantly:
the same order of magnitude. Since the difference between the in- this is due to the influence of the potential field of the rotor on the
tegrated length-scales obtained with ACF - denoted by LTu - and flow upstream. The turbulence intensity is not only predicted to be
ACFE - denoted by LTu,E - is at least an order of magnitude smaller too high, but the location of the maximum intensities also differs
than the length-scale itself, the approximated ACF is a reasonable from the actual position. In particular, the circumferential traverse
tool for estimating the change of the length-scales, due to different at rn = 50% shows clearly that no distinct differences can be ob-
inlet turbulence levels, and for investigating its behavior within a served.
gap or behind a turbulence generating grid.
Validation of hot-wire data
Turbulence Intensity TuFFT In order to validate the hot-wire data, distinct flow parameters
The proposed method can also be applied in order to calculate have been compared with 5-hole probe measurements. In addition
the turbulence intensity. In particular, this approach is favorable if to typical flow parameters, such as velocity and flow angles, the
no ensemble average (see Eq. (2)) can be performed (e.g. when streamwise vorticity ωsw is evaluated. In order to determine gra-
no trigger signal is available). Since the frequency chopping was dients in a streamwise direction, a method introduced by Gregory
carried out after subtraction of the time mean velocity, the result Smith et al. [19] for incompressible flow and modified for the use
in compressible flow by Niehuis et al. [20], is implemented. For de- various differences between the methods and the turbulence quan-
tailed derivation, the reader is referred to the references at the end tities will be discussed. The viewing direction for all 2D-plots is
of the paper. Following the calculation of the vorticity vector, the upstream.
streamwise vorticity component can be calculated.
MP01 - MP03
ω
~ ·~c
ωsw = (13) Turbulence Intensity Tu3D . Figure 8 shows the normalized
|~c|
MP01 P2 MP02 MP03
Comparing the assembled illustrations in figure 7, no distinct differ- 1

ences between both sets of probe data can be observed. Both mea-
surement techniques show the same flow characteristics and very 0.8
similar flow values. The greatest mean deviation between both mea- Tu3D,FFT,n[-]
surement techniques is found for the velocity with approximately 2.25
7 − 8m/s. The velocity calculated using the pneumatic probe shows 0.6 2.1
1.95
higher velocities. Despite the difference between the two, the im-

rn[-]
1.8
1.65
pact of the deviation observed will not influence the turbulence in- 0.4
1.5
1.35
tensity because of the division with c̄ in Eq. (6). 1.2
1.05
0.9
ωsw [1/s] 0.2 0.75

310

300 ωsw [1/s]: -27000 0 27000 ωsw [1/s]: -27000 0 27000 0

0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5
t/tgrid[-] [-] t/tgrid[-]
290
P1
280
y [mm]

Fig.8: Turbulence intensity Tu3D,FFT,n without removal of fvK

270

260

250 turbulence intensity Tu3D,FFT calculated on the basis of Eq. (6). To

240
eliminate any upstream influence from the rotor blades , the multi-
ples of the rotor blade passing frequency have been removed. All
c[m/s] illustrations, which have the same layout to enable quick compari-
310 son, show a clear decline in the turbulence intensity when moving
300 c [m/s]: 10 50 90 130 170 c [m/s]: 10 50 90 130 170 downstream. The observed decay within the rods’ wake is con-
sistent with correlations found by Roach [18]. Whereas in mea-
290
surement plane MP01 the wakes of three grid rods can be identi-
280
fied, they are hardly apparent in the downstream measurement plane
y [mm]

270 MP02 and completely mixed out in the last one. Additionally, it
260
seems as though the periodicity of the flow-field changes and the
three wakes merge into one with periodic borders on both sides.
250
A striking feature is observed at approximately rn = 30% behind
240 each rod in MP01. Even the subsequent downstream measurement
plane shows a region of high-velocity fluctuations. It would seem
α [◦ ] as though the high-velocity fluctuations, generated at rn = 30%,
310
merge when moving downstream, thus forming one large area of
300 α [°]: 65 67 69 71 73 75 α [°]: 65 67 69 71 73 75
high turbulence intensity. Investigations of all three stochastic ve-
290
locity components revealed the circumferential component c′u to be
mainly responsible for the strong fluctuations.
280
To obtain more information about the frequencies of which the
y [mm]

270
signal is composed, the velocity fluctuations in the high region of
260 turbulence (P1) and in the lower region of turbulence in the wake
250
(P2) are investigated by means of a fast Fourier transformation. The
result is shown in figure 9. The illustration clearly shows large am-
240
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
plitudes at multiples of the von Karman vortex street frequency fvK .
x [mm] x [mm]
The characteristic frequency can be computed using the Strouhal
(a) (b) number 0.2 - 0.21. In order to identify the effect of the charac-
Fig.7: Validation of hot wire data at MP11. (a) 5-hole probe data, teristic frequencies on the turbulence intensity, the fvK have been
(b) Hot wire probe data removed by applying the moving average filter introduced to the
velocity fluctuation signal. Since no single characteristic peak for
the frequency is evident here (s. figure 9 (a)), the neighboring high
amplitudes were also removed. With reference to the plots given in
figure 10, the striking feature seen previously is no longer present.
RESULTS As a result of removing the von Karman frequencies, the turbulence
This section is divided into two subsections. In the first, the tur- intensity is greatly reduced in comparison to the previous figure 8.
bulence quantities downstream of grid G2 will be discussed using It is evident that the fluctuations generated by the von Karman fre-
the evaluation methods presented. Following the discussion of the quencies contribute nearly 30% to the overall velocity fluctuations
flow in front of the first stator, the results for the measurement and are additionally responsible for the concentrated bulk seen in
planes MP10-MP12 introduced will be shown, while turbulence MP01.
grid G1 is mounted. Based on a number of selected examples, the The most striking difference between both evaluation points is
the high amplitude at approximately 3500 Hz, which is only visible Table 6: Influence of fvK removal on turbulence intensities
for P1. In order to confirm the overwhelming influence of the cen-
tral high amplitude on the overall turbulence intensity, only the cen- Location All fvK removed Peak fvK removed
tral peak has been removed. The outcome of this removal is sum- P1 −24% −12%
marized in table 6. The results show the percentage decrease in tur- P2 −14% −1.4%
bulence intensity in relation to the turbulence when the von Karman
frequencies are not deleted. The table 6 clearly shows that the main
contributor to the high turbulence intensity is the peak amplitude.
Removing only that characteristic frequency, the turbulence inten-
be taken into consideration. By way of contrast, it is necessary to
sity decreases by 12% at P1, Tu3D at P2 however decreases by only
disregard the characteristic frequencies for determination of the in-
1.4% due to the absence of the frequency. Unfortunately, the reason
tegrated time and length-scale respectively. As presented in [16]
for the strong influence of this specific frequency at rn = 30% is not
and shown in figure 5, periodic fractions will lead to the ACF cross-
as yet apparent.
ing zero too early and so leading to integrated time-scales which are
too small. In order to estimate the integrated length-scales correctly,
P1 the frequencies fvK and their neighboring augmented amplitudes (as
shown in figure 9) have been removed before post-processing the
time signal. For the contour plots which follow, the approximated
Velocity Amplitudechopped [m/s]
Velocity Amplitude [m/s]

autocorrelation function ACFE has been used. But since the turbu-
lence was generated not too far upstream, the fairly good agreement
between the integrated ACF and ACFE will be shown by means of
circumferential, area-averaged radial distributions (see figure 12) .
Figure 11 illustrates the development of the integrated length-
scale downstream of the turbulence generator. A constant scale for
0 5000 10000 15000 0 5000 10000 15000 all contour plots has been used to allow good comparison between
f [Hz] f [Hz]
the different measurement planes. The integrated length-scale is
(a) (b) largest in the centre of the rods’ wake, and decreases towards the
P2 boundaries of the wake. The same observation can be made for
the second measurement plane MP02. In contrast to the turbulence
intensity, the rods’ wakes can still be identified. The main reason
Velocity Amplitudechopped [m/s]
Velocity Amplitude [m/s]

for this behavior is the removal of the characteristic frequencies.

Refinement of the contour plot in figure 10 of MP02 would likewise
show, in diluted form, the wake of the cylindrical rods.
The final measurement plane, MP03, shows a very clear mixed-
out flow state. The three 2D-traverses evidently show the growth of
the integrated length-scale when moving away from the grid. The
0 5000 10000 15000 0 5000 10000 15000
correlation found by [18] reveal the same trend. As suggested by
f [Hz] f [Hz]
Pope [3], the growth of the integrated length-scales should be inter-
(c) (d) preted in such a way that the smaller and faster turbulent motions
Fig.9: Fast Fourier Transformation. (a) FFT at P1, (b) Chopped dissipate at a higher rate than the larger and hence slower turbu-
FFT at P1, (c) FFT at P2, (d) Chopped FFT at P2 lence structures. Consequently, the autocorrelation of the time sig-
nal increases, leading to greater integral time-scales and integrated
length-scales respectively.
To compare the approximated integrated length-scales LTu,E with
the length-scales determined with the original autocorrelation func-
MP01 MP02 MP03
tion, figure 12 shows the circumferential, area-averaged values for
both evaluation methods. The proposed approximation method
matches the magnitude and the qualitative development of the in-
tegrated length-scale determined by ACF reasonably well.
Tu3D,FFT,n[-]
1.85
1.7 MP01 MP02 MP03
1.55 1
1.4
1.25
1.1
0.95 0.8
0.8
0.65 LTu,E,n [-]
0.5
0.6
1.38
1.22
rn[-]

1.08
0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5 0.4
0.93
t/tgrid[-] t/tgrid[-] t/tgrid[-] 0.77
0.63
0.47
Fig.10: Turbulence intensity Tu3D,FFT,n after removal of all fvK 0.2 0.33

0
0 0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5
Integral Length-scale. Before examining the results in de- t/tgrid[-] t/tgrid[-] t/tgrid[-]

tail, the importance of the von Karman vortex street frequencies on

the integrated length-scale will be analyzed. In order to measure Fig.11: Integrated length-scale LTu,E,n in MP01 - MP03
the turbulence intensity correctly, the von Karman frequencies must
 MP01 MP02 MP03 MP10 MP11
1 310

300
Tu3D,n [-]: 0.05 0.25 0.45 0.65 0.85 1.05 Tu3D,n [-]: 0.05 0.25 0.45 0.65 0.85 1.05
0.8

290
0.6
rn [-]

280

y [mm]
0.4
270

0.2 260

L Tu,n L Tu,n L Tu,n

L Tu,n,E L Tu,n,E L Tu,n,E
250
0
0.6 0.8 1 1.2 1.4 0.6 0.8 1 1.2 1.4 0.6 0.8 1 1.2 1.4
L Tu,n [-] L Tu,n [-] L Tu,n [-]
240
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
x [mm] x [mm]
Fig.12: Circumferential area-averaged integral length-scales
MP12 Circumferential average
310 1
Tu3D,n [-]: 0.05 0.25 0.45 0.65 0.85 1.05
300
0.8
290
MP10 - MP12
In this subsection, the development of the turbulence quantities 280 0.6
MP10FFT

y [mm]

rn [-]
MP11FFT
behind the first stator will be shown and discussed by means of three 270
MP12FFT
MP12EA
0.4
equidistant measurement planes MP10-MP12. In contrast to the 260
previous results, the following turbulence quantities were measured
0.2
while the smallest turbulence generating grid was being mounted. 250

240
Turbulence Intensity. Figure 13 illustrates the turbulent in- 0 10 20 30 40 50 60 70 80
0
0.1 0.2 0.3 0.4 0.5
x [mm]
Tu 3D,n [-]
tensity Tu3D,FFT and its development downstream of the first sta-
tor. The reduction in turbulence within the secondary flow areas, as
well as the broadening of the wake can be quite clearly observed. In Fig.13: Turbulence intensity development in MP10 - MP12
particular, the upstream influence of the rotor leads to a significant
broadening in the wake. Within the free-stream there are no distinct
differences noted between the first two measurement planes. In con-
trast to this, the turbulence intensity within the free-stream increases the circumferential direction, leading to the longest period of time
in MP12. With reference to the radial distribution in figure 13, the in which the probe is exposed to the stator wake and consequently
development of the turbulence becomes even more evident. After a measures a higher turbulence intensity compared to the undistorted
clear decrease in turbulence kinetic energy, the turbulence intensity wakes in MP10 and MP11.
rises again in the measurement plane right in front of the rotor. In or- Integrated Length-scale. As was explained initially, only the
der to verify the behavior observed, for the circumferential average, approximated integrated length-scales LTu,E will be shown. As pre-
both calculation methods are shown. Since no differences occur for viously, the value will be normalized by the integrated length-scale
MP10 and MP11, only the last measurement plane is presented in measured on the inlet plane of the turbine. In contrast to the inte-
figure 13. There is a slight offset between both curves. This is possi- grated length-scale upstream of the first row, an expansive area of
bly due to the difference in the number of samples for each method. large integral length-scale is apparent in figure 15. With respect to
For the FFT-based turbulence intensity Tu3D,FFT , almost twice as [3], the turbulence dissipation rate is highest within high-velocity
many samples have been used, causing a slightly higher turbulence flows leaving behind the large scale motions which are character-
intensity. ized by large integrated length-scales. Due to the strong influence
One of the main reasons for the increase in turbulence is the prox- of the downstream rotor blades, the integral length-scale changes
imity of MP12 in relation to the rotor’s leading edge. As is known, and reduces significantly in the free-stream area. In the first in-
the first stator’s wake is bent around the leading edge of the rotor, stance, this may be the effect of the wake bending due to the rotor
which in turn causes the wake to change its orientation. The wake leading edge.
is no longer aligned with the main direction of flow, but - depending In the next section, the development of the integral quantity
on the rotor’s position - is temporarily parallel to the circumferential within the wake of the first stator will be discussed. As seen previ-
velocity component. The procedure described is only visible when ously in figures 11 and 12, the size of the energy-containing eddies
looking at the time-resolved turbulence intensity. In order to be able increases within the wake when moving downstream. But unlike
to perform the unsteady analysis, the investigator must switch to within the rods’ wake, the integral length-scale increases towards
the ensemble average method. Because only this method will allow the boundaries of the wake. Similar observations have been made
a turbulence intensity to be saved for each resolved rotor position by Camp and Shin [16]. In the gap of the third stator and the fourth
(here 487 rotor positions have been analyzed and at each position rotor of a multistage compressor, they likewise observed that the in-
the turbulence intensity was determined by performing an average tegral length-scale was smallest within the blade wake and increases
over 233 rotor revolutions). Instead of time-averaging, as shown towards the borders. Despite the small gap of only 50% in the rotor
in Eq. (4), the turbulence intensities were saved in 487 separate axial chord length, an increase in the integral length-scale is visible
files. The result of the post-processing described briefly is shown despite the small distance between the measurement planes of only
in figure 14. The picture sequence presented at different time-steps 20% in the axial gap (see table 4).
reveals the bending of the first stator wake due to the rotor. The
first time-step t1 shows the first stator wake and the upstream influ- CONCLUSION
ence of the rotor blades, which is indicated by the slight increase in Detailed hot-wire measurements were taken in the 1.5 stage ax-
turbulence intensity to the right and left of the wake. With respect ial flow turbine rig on three measurement planes in front of the first
to the time sequence shown, the rotor moves counterclockwise and stator, allowing a detailed evaluation of the turbulence quantities be-
it can be observed, that the wake is shifted in a circumferential di- hind a turbulence generating grid. In order to analyze the turbulence
rection at time-step t2 . Moving the rotor further in a circumferential intensity and integral length-scales within the turbine, 2D-traverses
direction, the wake begins to bend around the leading edge. Be- were carried out on three planes between the first and second row. A
tween time-steps t3 and t5 the wake is oriented almost parallel to method for calculating the integral length-scale, proposed by Camp
 t1 t2 MP10 MP11
310 310

300
Tu 3D,n [-] 0.05 0.25 0.45 0.65 0.85 1.05 Tu 3D,n [-] 0.05 0.25 0.45 0.65 0.85 1.05 L Tu,n,E [-]: 0.1 0.4 0.7 1 1.3 1.6 1.9 2.2 L Tu,n,E [-]: 0.1 0.4 0.7 1 1.3 1.6 1.9 2.2
300

290 290

280 280
y [mm]

y [mm]
270
270

260
260

250
250

240
240
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
x [mm] x [mm] 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
x [mm] x [mm]

t3 t4 MP12
310
310

300
Tu 3D,n [-] 0.05 0.25 0.45 0.65 0.85 1.05 Tu 3D,n [-] 0.05 0.25 0.45 0.65 0.85 1.05

300
L Tu,n,E [-]: 0.1 0.4 0.7 1 1.3 1.6 1.9 2.2

290
290

280
y [mm]

280

y [mm]
270
270

260
260

250
250

240
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80 240
x [mm]
0 10 20 30 40 50 60 70 80
x [mm]

t5 t1
310
Fig.15: Integrated length-scale LTu,n,E development in MP10 -
Tu 3D,n [-] 0.05 0.25 0.45 0.65 0.85 1.05 Tu 3D,n [-] 0.05 0.25 0.45 0.65 0.85 1.05
300 MP12
290

280
y [mm]

270
has a strong upstream effect and causes the stator wake to bend
260
around the leading edge leading, in turn, to higher turbulence
250 intensities being measured.
240
0 10 20 30 40
x [mm]
50 60 70 80 0 10 20 30 40
x [mm]
50 60 70 80 5. Approximated integral length-scales within the wake of the
first stator reveal, as observed in MP01 to MP03, an in-
crease in size when moving downstream. In contrast to the
Fig.14: Bending of the first stator wake rods’ wake, the integral length-scale reaches a minimum value
within the wake’s core.

In future, research activities will apply the methods proposed to

and Shin [16], was applied and additionally used to determine the measurement planes behind the rotor and the second stator of the
turbulence intensity. Instead of integrating the autocorrelation func- turbine geometry shown. Additionally, the quantities measured will
tion up to its first zero crossing, a simple e-function approach was be compared to different numerical models in order to validate and
used and integrated. improve the models’ prediction of turbulence quantities.
Additionally, with respect to the different post-processing meth-
1. The comparison of both methods for computing the turbu-
ods, an alternative to the autocorrelation function and its approxi-
lence intensity showed good agreement along the complete
mation is investigated. In [21], Hinze derived a mathematical rela-
flow field. A slight difference was observed in measurement
tionship between the autocorrelation function and the energy den-
plane MP12, which is the consequence of differences in the
sity spectrum E1 (n), so enabling the investigator to determine the
number of sample points.
integral time-scale using the energy density spectrum. Referring to
2. An approximation of the autocorrelation function has been [21], the integral length-scale can be calculated as follows.
proposed and applied. In the measurements shown, the ap-  
proximated integral length-scale is always smaller than the c̄
LTu = lim E 1 (n) (14)
original one and a difference of 10 to 20% is present. De- n→0 4 · (c′ )2
spite the offset, the method can be used in flow areas in which
the original autocorrelation function does not perform well. In Eq. (14) n denotes the frequency. This formula has been used
by El-Gabry et al. [22], Thurman et al. [23] and others and will be
3. Behind the grid, the turbulence intensity is mainly a conse- applied to experimental data in the future as well.
quence of the von Karman vortex street, as could be shown by
deleting the characteristic frequencies. The integral length- ACKNOWLEDGMENT
scale is largest in the rods’ wakes and reduces towards the The investigations were conducted as part of the COORETEC-
wakes’ borders. The length-scale rises the greater the distance Turbo 2020 joint research program in the framework of AG Turbo.
from the grid. The work was supported by the Bundesministerium fuer Wirtschaft
und Technologie (BMWi) as per resolution of the German Federal
4. Turbulence measurements behind the first stator clearly show Parliament under grant number 03ET2011F. The authors gratefully
the broadening of the wakes, as well as the diffusion of turbu- acknowledge AG Turbo and MTU Aero Engines GmbH for their
lence kinetic energy in the neighboring free-stream. As was support and permission to publish this paper. The responsibility for
shown with time-resolved turbulence measurements, the rotor the content lies solely with its authors.
References [17] O’Neill, P., Nicolaides, D., Honnery, D., and Soria, J., 2004.
[1] Menter, F. R., 1994. “Two-equation eddy-viscosity turbulence “Autocorrelation functions and the determination of integral
models for engineering applications”. AIAA journal, 32(8), length with reference to experimental and numerical data”. In
pp. 1598–1605. 15th Australasian Fluid Mechanics Conference The Univer-
sity of Sydney, Sydney, Australia, pp. 13–17.
[2] Wilcox, D. C., 1988. “Reassessment of the scale-determining
equation for advanced turbulence models”. AIAA journal, [18] Roach, P., 1987. “The generation of nearly isotropic turbu-
26(11), pp. 1299–1310. lence by means of grids”. International journal of heat and
fluid flow, 8(2), pp. 82–92.
[3] Pope, S. B., 2000. Turbulent flows. Cambridge university
press. [19] Gregory-Smith, D., Graves, C., and Walsh, J., 1988. “Growth
of secondary losses and vorticity in an axial turbine cascade”.
[4] Schlichting, H., and Das, A., 1970. “On the influence of Journal of turbomachinery, 110(1), pp. 1–8.
turbulence level on the aerodynamic losses of axial turboma-
chines”. From Flow Research on Blading, LS Dzung, ed., El- [20] Niehuis, R., Lücking, P., and Stubert, B., 1990. “Experimental
sevier Publ. Co. and numerical study on basic phenomena of secondary flows
in turbines”. AGARD, Secondary Flows in Turbomachines 17
[5] Suder, K. L., Obrien, J. E., and Reshotko, E., 1988. “Exper- p(SEE N 90-21009 14-07).
imental study of bypass transition in a boundary layer”. M.S.
Thesis National Aeronautics and Space Administration. Lewis [21] Hinze, J., 1975. Turbulence (2nd edn). McGraw-Hill, New
Research Center, Cleveland, OH. York.
[6] Butler, R. J., Byerley, A. R., VanTreuren, K., and Baughn, [22] El-Gabry, L. A., Thurman, D. R., and Poinsatte, P. E., 2014.
J. W., 2001. “The effect of turbulence intensity and length “Procedure for determining turbulence length scales using
scale on low-pressure turbine blade aerodynamics”. Interna- hotwire anemometry”.
tional journal of heat and fluid flow, 22(2), pp. 123–133.
[23] Thurman, D., Flegel, A., and Giel, P., 2014. “Inlet turbulence
[7] Carullo, J., Nasir, S., Cress, R., Ng, W., Thole, K., Zhang, L., and length scale measurements in a large scale transonic tur-
and Moon, H., 2011. “The effects of freestream turbulence, bine cascade”.
turbulence length scale, and exit reynolds number on turbine
blade heat transfer in a transonic cascade”. Journal of Turbo-
machinery, 133(1), p. 011030.

[8] Restemeier, M., Jeschke, P., Guendogdu, Y., and Gier, J.,
2013. “Numerical and experimental analysis of the effect of
variable blade row spacing in a subsonic axial turbine”. Jour-
nal of Turbomachinery, GT2011-45637, 135(2), p. 021031.

[9] Niewoehner, J., Poehler, T., Jeschke, P., and Guendogdu, Y.,
2015. “Investigation of nonaxisymmetric endwall contouring
and three-dimensional airfoil design in a 1.5 stage axial tur-
bine - part ii: Experimental validation”. Journal of Turboma-
chinery, GT2014-26785, 137(8), p. 081010.

[10] Bruun, H. H., 1996. Hot-wire anemometry: principles and

signal analysis, Vol. 7. IOP Publishing.

[11] Poensgen, C., 1989. Ein Verfahren zur Vermessung der insta-
tionären dreidimensionalen Geschwindigkeitsvektoren in Tur-
bomaschinen. Institut für Strahlantriebe und Turboarbeits-
maschinen, Rheinisch-Westfälische Technische Hochschule
Aachen.

[12] Bearman, P., 1971. “Corrections for the effect of ambient

temperature drift on hot-wire measurements in incompressible
flow”. DISA information, 11(1), pp. 25–30.

[13] Jørgensen, F. E., 2001. How to Measure Turbulence with Hot-

Wire Anemometers: A Practical Guide. Dantec dynamics.

[14] Gostelow, J., 1977. “A new approach to the experimental

study of turbomachinery flow phenomena”. Journal of En-
gineering for Gas Turbines and Power, 99(1), pp. 97–105.

[15] Taylor, G. I., 1938. “The spectrum of turbulence”. In Pro-

ceedings of the Royal Society of London A: Mathematical,
Physical and Engineering Sciences, Vol. 164, The Royal So-
ciety, pp. 476–490.

[16] Camp, T., and Shin, H.-W., 1995. “Turbulence intensity and
length scale measurements in multistage compressors”. Jour-
nal of turbomachinery, 117(1), pp. 38–46.