Professional Documents
Culture Documents
JURY
i
.
ii
Remerciements
J’ai beaucoup appris au cours ces quatre années passées à l’Institut von Kar-
man. Je reste admiratif devant la qualité du travail scientifique effectué ici et devant
l’enseignement de haut niveau dispensé à des étudiants venus de tous les pays d’Europe
et des Etats Unis. Je remercie M. J.F. Wendt, le directeur de l’Institut, pour m’avoir
acceuilli au sein de cette organisation remarquable. Ma gratitude va également au consor-
tium Brite-Euram qui a fourni le support financier pour ce travail.
Mon travail a été supervisé par M. C.H. Sieverding, Professeur à l’Institut von
Karman. Je le remercie pour la confiance qu’il a bien voulu m’accorder en m’impliquant
dans ce programme de grande envergure dont il est un des initiateurs principaux. Je le
remercie également pour m’avoir apporté de manière constante son soutient et son intérêt.
J’ai grandement bénéficié de sa connaissance des turbomachines, de son esprit de rigueur
et d’opiniâtreté. Merci aussi pour les invitations aux concerts de musique classique.
iii
viendrai longtemps des premières montées en vitesse, des premiers tirs, de l’incident et de
la longue série d’essais.
Enfin, je remercie tous ceux qui m’ont accompagné au cours des ces années
et qui ont rendu notre séjour ici très agréable. Henri pour les cafés pris à Gand, Denis
pour l’organisation de nombreuses réjouissances, Nicolas et Nathalie, Sébastien et Pascale,
Hugues pour toutes les soirées passées ensemble. Je leur dis à bientôt.
iv
Contents
v
5 Scopes and objectives 71
vi
8 Fast response pressure measurements 145
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
8.2 Correction for sensitivity to temperature . . . . . . . . . . . . . . . . . . . . 145
8.3 Estimation of the temperature effect on the pressure sensors used at VKI . 146
8.4 Comparison of the effect of centrifugal forces and the effect of temperature
on the output of the pressure sensors . . . . . . . . . . . . . . . . . . . . . 147
8.5 A simple correction for centrifugal force and temperature effects . . . . . . 148
8.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
8.7 Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
III Description of the turbine stage, the test facility and the instru-
mentation 155
11 Instrumentation 189
11.1 Instrumentation location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
11.1.1 Convention on angular reference . . . . . . . . . . . . . . . . . . . . 189
11.1.2 Total upstream pressure and temperature (P01 , T01 ), inter-blade row
static pressure Ps2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
11.1.3 Instrumentation around the blade mid-section . . . . . . . . . . . . . 190
11.1.4 Total pressure and total temperature measurements in rotation:
P02r , T02r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
11.1.5 Position of instrumented blades on the rotor . . . . . . . . . . . . . 191
11.1.6 Stage exit static pressure Ps3 . . . . . . . . . . . . . . . . . . . . . . 191
11.1.7 Stage exit probe carriages . . . . . . . . . . . . . . . . . . . . . . . . 192
11.2 Data acquisition and calibrations . . . . . . . . . . . . . . . . . . . . . . . . 192
11.2.1 Data acquisition and conditioning . . . . . . . . . . . . . . . . . . . 192
11.2.2 Measurement of the rotational speed . . . . . . . . . . . . . . . . . . 193
11.2.3 Calibrations of the instrumentation measuring gas pressure and tem-
perature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
11.2.4 Calibration of the instrumentation measuring blade surface temper-
ature for heat flux measurements . . . . . . . . . . . . . . . . . . . . 195
vii
11.3 Work programme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
11.4 Tables, figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
viii
List of symbols
Roman
A Area [m2 ]
3
Gr Grashof number (free convection) glν 2β
H enthalpy [J/(kg K)]
I momentum of inertia [kgm2 ]
M Mach number va
Nu Nusselt number hl k
P pressure [Pa]
P power [W]
c µ
Pr Prandtl number pk = αν
Q heat [J]
i=n
RM S root mean square [ i=1 (x − x̄)2 ]/n
ρvl
Re Reynolds number µ
S entropy [J/(kg K)]
So Strouhal number fVl
St Stanton number ρchp u = RePNu
r
T temperature [K]
T torque [N.m]
V volume [m3 ] √
a speed of sound γrT [m/s]
c blade chord [m]
cp specific heat at constant pressure [J/(kg.K)]
d diameter [m]
f frequency [Hz]
fs sampling frequency [Hz]
g blade pitch [m]
g (Tw mean − Tp )/(Tg − Tp )
h blade height [m]
h heat exchange coefficient by convection [W/(m2 K)]
j complex variable : j*j = -1
k thermal conductivity [W/(m K)]
p Laplace variable p = α + jβ
q̇ heat flux [W/m2 ]
r gas constant (287.7 [J/(kg.K)] for air)
s curvilinear abcissa [m]
t time [s]
u peripheral speed [m/s]
v velocity [m/s]
w relative velocity [m/s]
ix
Greek
ρ density [kg/m3 ]
µ dynamic viscosity [kg/(ms)] [P l]
ν kinematic viscosity µρ [m2 /s]
α flow angle with respect to the machine axis [deg]
α sensitivity to temperature R10 ∂R
∂T 1/K]
α k 2
thermal diffusivity ρcp [m /s]
γ isentropic exponent (1.4 for air)
γ intermittency factor
η efficiency
ζ kinetic energy loss coefficient
τ time constant [s]
i=n
σ standard deviation [ i=1 (x − x̄)2 ]/(n − 1)
σ0−1 resistivity [Ωm]
ω pulsation 2πf [rad/s]
ϕ phase
Subscript
0 total conditions (no subscript when static conditions)
1 inlet of the stator blade row
2 outlet of stator, inlet of rotor
3 outlet of rotor blade row
ax axial, parallel to the machine axis
a air
c chord
co coolant
g gas
is state after an isentropic evolution
p prongs
r rotor
r relative conditions
s stator
s static quantity (sometimes without subscript)
t tangential
w wall
x
Part I
1
Chapter 1
The use of gas turbines is nowadays widely spread. The first gas turbine en-
gine was built in 1903 in Norway (Wilson [1]) and produced an effective power of 11 HP
(horse power). The same year, the Wrights brothers performed the first powered flight
using a piston engine producing 12 HP for a weight of 100 kg (0.12 HP/kg)(Anderson [2]).
Many problems had to be overcome in order to power a flight with a gas turbine engine,
especially to increase the power to engine weight ratio. However, the first flight powered
with a gas turbine engine took place in Germany in 1939 (Wilson [1]) with an engine
weighting 361 kg and producing 4.9kN thrust. Since then, thousands of researchers have
worked to improve the performance of those engines which are now powering the great
majority of the airplanes. If gas turbines are also widely used in electricity generation or
boat propulsion, intensive research is mainly carried out in the aircraft engine industry
where the problems of power and thrust to weight ratios, efficiency and safety are crucial
and where manufacturing costs are high.
Fig. 1.1 shows a picture and some sketches of an Airbus A340 aircraft powered
by 4 CFM56-5-C2 engines. When comparing the size of the aircraft and the size of the
engines, it is easy to appreciate how compact and powerful they are. The high power to
weight ratio is their major advantage with respect to piston engines: modern gas turbine
can produce more than 20 HP/kg. This is even more striking when looking at a few
figures concerning the Airbus A340-300 (see table 1.1). Maximum take-off weight is 267
tons whereas the four engines weigh about 10 tons.
Unfortunately, the figures also show that to cover 13 200 km, 135 000 litres
of fuel are needed at take-off. Covering this distance at an average Mach number of 0.8
leads to an average fuel consumption around 2000 litres of fuel per hour and per engine
(almost the weight of the engine). Fuel represents 37% of the maximum take-off weight
while the payload is only 15%. Obviously, fuel consumption reduction is a major challenge
for those engines. Moreover, any reduction in fuel consumption is not only a gain in fuel
cost, it is also a gain in payload. But let us look now deeper into the working principle of
gas turbine engines.
3
Airline operating weight empty 128 710 kg
Maximum payload 48 290 kg
Maximum take-off weight 267 000 kg
Maximum landing weight 189 000 kg
Weight of one engine 2500 kg
Number of passengers 303
Maximum fuel capacity 135 000 litres
Wing span 60.3 m
Wings area gross 361 m2
Overall length 63.3 m
Max diameter of fuselage 5.64 m
Maximum operating speed M=0.86
Range 13 200 km
Table 1.1: A few figures on the A340 extracted from Jane’s 1992-93 [3]
An alternative to the axial compressor is the radial compressor . The air enters
axially in the rotor wheel which pushes radially the air into a diffuser. The main pressure
rise results from the flow change of radius; an additional pressure rise is performed in the
diffuser which collects the air in the peripheral area. A high pressure ratio can be achieved
in a single stage (up to 7.0) but for a given rotational speed and inlet area, a radial com-
pressor achieves a lower mass flow. Moreover, if several successive stages are needed, the
flow has to be deviated back to axial direction between two stages which creates additional
pressure losses. This results in bigger sizes and lower power to weight ratios and explains
why the radial compressor is not very often used in aircraft engines.
Then, the pressurized air enters the combustion chamber. Fuel is mixed with
this air and a permanent combustion exists (ignition is needed only when starting the
engine). As the combustion chamber is open on both sides, the total pressure remains al-
most constant. Combustion brings the exhaust gases at high temperature (high enthalpy
4
level) between 750o C and 1700o C depending on applications.
This hot high pressure gas is expanded in the turbines. A stator blade row
accelerates the air which is discharged into the rotor. The power generated by this process
is transmitted to the compressor through the shaft. As the pressure decreases in a turbine
stage, the boundary layer tends to be stabilized and high pressure drops can be achieved
in a single axial turbine stage (up to 3.0) before reaching flow separation. Moreover, the
burned gas have a high level of enthalpy compared with the inlet air which explains that
there are fewer turbine stages than compressor stages in an engine. The turbines are also
divided in two groups, low pressure and high pressure which are respectively connected
through two distinct shafts to the groups of low pressure and high pressure compressors.
The low pressure and high pressure blade rows rotate at distinct rotational speeds allowing
to have adapted blade peripheral speeds. Some engines have three sets compressor-turbine
with three shafts (Rolls Royce engines).
The turbofan (Fig.1.2.c) is the most common engine used in civil aircraft
propulsion today. Propulsion results from both the jet exiting the core engine and the
fan thrust. The fan is located at the inlet of the engine; its diameter exceeds by far the
diameter of the core engine. Part of the mass flow is absorbed in the core engine and
another part is by-passed and discharged directly into the atmosphere generating thrust.
This part is quantified by the by-pass ratio. This large size low speed jet has a higher
propulsion efficiency and lower noise than a small high speed jet which explains the large
use of this type of engine on civil aircrafts. Additional turbine stages (with respect to
a turbojet engine) are required to drive the fan but the overall propulsive efficiency is
improved. Large by-pass ratio engines are usually limited to subsonic flights.
The turbopropeller (Fig.1.2.d) has also additional turbine stages called power
turbine which drive a propeller trough a reduction gear. The propeller provides the largest
part of the thrust. This engine is suitable for medium speed, low altitude flights.
5
The turboshaft (Fig.1.2.e) has also a power turbine to provide power for any
later process as electricity generation, boat propulsion, pumping... Usually, a gear is used
to provide power at rotational speeds lower than the one of the power turbine. As those
engines are designed for stationary applications, the residual thrust is kept as low as pos-
sible. Notice the use of a single stage radial compressor in the high pressure section and
the three independent shafts (all preceding engines had two shafts).
In order to illustrate the state of the art in terms of turbofan engines, the next
section presents a few figures about the CFM56-5-C2 engine.
Table 1.2: A few figures on CFM56-5-C engines extracted from Jane’s 1992-93 [3]
The fan, the low pressure compressor and the low pressure turbine are on the
same shaft rotating at 3000 to 4000 RPM which leads to peripheral speeds about 380
m/s at the tip of the fan blade (almost the speed of sound !). The compression ratio
of the fan is usually around 1.5, 1.6 at the blade tip. The low pressure compressor has
an overall compression ratio of 2.1 (1.2 per stage) whereas the high pressure compressor
has a compression ratio of 11.0 (1.3 per stage). The rotational speed of the core engine
(high pressure compressor, high pressure turbine) is around 10000 to 13000 RPM. Notice
that only one high pressure turbine stage drives 9 high pressure compressor stages. The
pressure ratio of the high pressure turbine stage is around 3.
6
These engines have to cope with largely varying inlet conditions: from ambi-
ent pressure, ambient temperature, low speed and full thrust at take-off (1 bar, 20o C, low
Mach number, 14 tons thrust per engine) to low pressure, low temperature, high speed,
low thrust during cruise (0.2 bar, -50o C, M=0.8, 3 tons thrust per engine). Depending on
the required thrust, the engine rotational speeds will also vary. The 4 first stators of the
high pressure compressor have variable incidence guide vanes in order to adapt the blade
to the varying inlet flow angle.
The engines must also satisfy a number of additional constraints like being
able to maintain a satisfactory level of thrust when ingesting birds, water (rain), ice or
when submitted to fan blade failure. The plane must also be able to fly for a certain time
if one of the engine completely fails (ETOPS time).
The actual tendency is to produce engines with large thrusts in order to power
large aircrafts with only two engines, even on long overseas distances. Engine control is
simplified, maintenance is less expensive and a significant gain in fuel consumption is ex-
pected. From the design point of view, the engine must have a large diameter leading
to high levels of centrifugal force, especially in the fan blades. Recently, three engine
manufacturers have started developing engines (General Electric:GE90, Pratt & Whitney:
P&W4084, Rolls Royce:Trent) able to deliver 50 tons of thrust in order to power the large
aircraft Boeing B777 with only two engines.
Coming back to the problem of engine performance, let us look more in details
what are the most important parameters influencing the efficiency of a gas turbine engine.
7
turbine shaft. In reality, entropy is added in the turbine and only ∆Ht is available. The
efficiency of the turbine is then defined by:
H03 − H04 ∆H0t
ηt = = (1.2)
H03 − H04,is ∆H0t,is
The gas is now in state 4. The kinetic energy contained in the exit jet (part
from 4 to 5) can be used for the propulsion in the case of an aircraft or will be lost in the
case of a stationary gas turbine. The gas turbine cycle is an open cycle, i.e. the gas at the
exhaust of the engine is in a different state than at the inlet.
There are two important characteristic parameters in a gas turbine cycle: the
specific power Ns , i.e. the net power output at the shaft (or propulsive work in the case
of an airplane) and the thermal efficiency ηth namely the ratio of the net work to the heat
added in the combustion chamber Qadd . Referring to Liess This can be expressed as:
T03 1 1
Ns = cp T1 ηt (1 − m ) − (π m − 1) [J/kg] (1.3)
T01 π ηc
T03 1
Qadd = cp T01 ( − 1) − (π m − 1) [J] (1.4)
T01 ηc
Ns ηt TT03
01
(1 − π1m ) − η1c (π m − 1)
ηth = = (1.5)
Qadd ( TT03
01
− 1) − η1c (π m − 1)
P02 γ−1
where π = P01 is the pressure ratio of the compressor, m = γ .
Fig. 1.5 compares the specific power and the thermal efficiency as a function
of pressure ratio for several turbine inlet temperatures (curves were computed with T01 =
293K, ηc = 0.85, ηt = 0.88, cp = 1003J/(kgK) constant, γ = 1.4). For given turbine inlet
temperature, the specific power and the thermal efficiency do not reach their maximum
for the same pressure ratio (thick dashed line) and a compromise must be found. As
the turbine inlet temperature increases, the thermal efficiency gets closer to the thermal
efficiency of the ideal Joule cycle (see Eq. 1.6) sketched as a plain line on Fig. 1.5.b.
1
ηth,Joule = 1 − (1.6)
πm
In order to reach very high turbine inlet temperature, the blades and the end-
walls of the first high pressure turbine stage have to be cooled. Temperature up to 2000 K
can be encountered in military aircraft engines whereas the most advanced single cristal
blade would not stand more than 1200 K.
8
force the flow several times from hub to tip and tip to hub. The design can also be such
that the cooling is performed by jets impinging normally to the surface which improves
heat exchange coefficients. This air is rejected into the main flow at the trailing edge or
at the blade tip. Examples of cores used for internally cooled blades manufacturing are
presented in Fig. 1.6.a. On Fig. 1.6.b, some blades with trailing edge slot ejection can be
seen.
-External cooling allows a further rise of the gas temperature because it protects the ex-
ternal blade surface from the main flow. A cold film is created thanks to coolant flow
ejection through rows of holes in the blade surface. Fig. 1.6.b presents some blades with
rows of cooling holes visible on the suction side.
The cooling air is usually taken at the compressor outlet (this cold air maybe
about 400o C!) and causes a small decrease in the cycle efficiency. In the case of external
cooling, the coolant flow is ejected into the main stream and interferes with it causing
aerodynamic penalties. This cold flow also lowers the thermal efficiency of the turbine
but the improvement of the cycle efficiency due to the higher turbine inlet temperature
compensates greatly this loss. Excess cooling decreases the overall efficiency whereas un-
sufficient cooling affects the blade life duration.
Specific power and thermal efficiency are also greatly dependant on compres-
sor and turbine efficiency. This is illustrated in Fig. 1.7 and Fig. 1.8. In a turbomachine,
the net power output is the difference between the turbine and the compressor work. Im-
proving the efficiency of one or another will lead to a higher gain of the overall cycle.
As far as aircraft engines are concerned, the increase in turbine inlet temper-
ature, compressor and turbine efficiency remain the main challenges. But those engines
being probably the most highly researched engines in history, it is not easy to obtain sub-
stantial improvements. The total to total efficiency of compressors and turbines is now
around 90% and the levels of turbine inlet temperature is currently around 2000 K in
military aircraft engines.
9
engine. In the following, one will focus exclusively on a small part of the engine: the first
high pressure turbine stage. A three-dimensional view of the stator and rotor blades is
presented in Fig. 1.10. This turbine stage is representative of modern aero-engines; details
on the geometry and characteristics of the blades will be given later. This stage was
developed in collaboration will several other organisations under a ”Brite-Euram” contract
with the EEC. For this reason, the turbine will be referred to as ”Brite” turbine stage in
the following. The stator has 43 cooled vanes; the hub and tip radii are constant. The rotor
has 63 blades with a meridional divergence of 10% of the blade height. Some stationnary
measurements performed behind the stator alone which are published in the literature
(Sieverding et al. [4], [5]) will be used for two purposes: illustration and knowledge of
the stator outlet flow without the presence of the rotor. As it is stated in the title of this
thesis, this works focuses on the rotor flow.
At first, the present understanding of the aerodynamic and heat transfer of
the flow across a turbine stage will be reviewed, beginning with a simplified bidimensional
approach of the flow across a turbine blade row.
10
1.6 Figures
11
a: Working principle
LP HP LP
HP
b: Turbojet engine
Fan
c: Turbofan engine
d: Turbopropeller
engine
Reduction gear
Power Turbine
e: Turboshaft engine
12
LP compressor HP compressor HP Turbine LP turbine
Fan
Booster
13
By-pass duct
n
ditio ber)
t ad 3
Turb
a
H e cham
u s t ion
omb
ine
(c
∆ Ht
∆ Ht,is
2
ressor
4
∆ Hc
Comp
∆ Hc,is
P01
5
S
Figure 1.4: Gas turbine thermodynamic cycle
500
400
T3=750 K
T3=1250 K
NS [kJ/kg]
300
T3=1750 K
200
100
-a-
0
0 10 20 30
0.8
Joule cycle
Location of maximum efficiency
0.6
ηth
0.4
0.2
-b-
0.0
0 10 20 30
π=P2/P1
Figure 1.5: Influence of turbine inlet temperature on specific power and thermal efficiency
14
Return channel Fins
15
500
ηc=0.70
400 ηt=0.88 ηc=0.80
NS [kJ/kg] ηc=0.90
300
200
100
0
0 10 20 30
0.8
Joule cycle
Location of maximum efficiency
0.6
ηth
0.4
0.2
0.0
0 10 20 30
π=P2/P1
Figure 1.7: Influence of compressor efficiency ηc on specific power and thermal efficiency
500
300 ητ=0.90
200
100
0
0 10 20 30
0.8
Joule cycle
Location of maximum efficiency
0.6
ηth
0.4
0.2
0.0
0 10 20 30
π=P2/P1
Figure 1.8: Influence of turbine efficiency ηt on specific power and thermal efficiency
16
Figure 1.9: Example of a machine combining reheating and intercooling
17
Rotor
Blade
Row
Stator
Blade
Row
Inlet Flow
18
Chapter 2
Turbine aerodynamics:
two-dimensional aspects
19
rive total pressure and flow angle before the blade row inside the channel and behind the
blade row; in the case of transonic flow, the blockage caused by a probe inside the channel
is so large that the flow would be greatly disturbed.
-Wall static pressure taps around the blade or on the endwall connected to pneumatic
pressure transducers allow to measure blade surface and outlet static pressure and derive
the corresponding Mach number distribution.
-Hot film gauges give an image of the wall shear stress; statistical data processing of this
information allows to derive the intermittency factor, hence the boundary layer state.
-Thin film gauges measure blade surface temperature history and allow to derive heat flux
distribution around the blade profile or on the endwalls of annular cascades in a transient
test.
-LASER velocimetry allows to determine the velocity field as well as the turbulence level.
-Schlieren photography permits shock wave visualization.
-Interferometric holography allows to derive flow density.
- ...
For example, the isentropic Mach number distribution around the Brite vane
is plotted in Fig. 2.4. The dots are derived from static pressure measurements performed
in a straight cascade composed of 8 Brite vane profiles at DLR (Amecke et al. [5]). The
continuous line is the result of a 2D Euler computation performed with a VKI in-house
code from Paillere [7]. The isentropic Mach number is plotted in function of the blade axis
in the ”mechanical” frame of reference (see Fig. 2.3). A good agreement between measure-
ments and calculation is observed. The inlet flow comes to the blade with a Mach number
M1 = 0.17 and is brought to rest on the stagnation point. Then the flow is accelerated on
each side of the blade surface up to an exit Mach number M2,is = 1.05. This acceleration
is performed in a different manner on the convex side and on the concave side:
-on the concave side, the flow is accelerated progressively due to the channel contraction
(bottom curve in Fig. 2.4),
-on the convex side, the acceleration in the nose area is mainly due to the large blade
curvature; later, it is due to the channel contraction; finally, after the throat, there is no
contraction anymore and the flow does not accelerate further.
The larger acceleration on the convex side causes a pressure which is lower than on the
concave side. For this reason, the convex side is called suction side while the concave side
20
is called pressure side. (see Fig. 2.3). A pitchwise pressure gradient results in the blade
to blade plane.
The discontinuity which can be seen on the suction side at xblade /c = 0.4 is
due to the impingment of the shock wave which emerge from trailing edge on the suction
side. Evidence of this is given in the Schlieren visualization (see Fig. 2.6) also performed
on the straight cascade at DLR (Amecke et al. [5]).
Of course, the Mach number distribution depends on blade geometry and for
a given geometry depends on the outlet Mach number level and on the flow incidence.
The isentropic Mach number derived from a 2D Euler computation on a cylindrical cut
of the Brite rotor blade at mid-span (see Fig. 2.7) is presented in Fig. 2.8. Coriolis forces
and centrifugal forces, characteristics for flows through rotating blade rows, cannot be
simulated in a 2D Euler code. The calculation is performed in a relative frame rotating
with the rotor, imposing uniform relative inlet conditions. The flow is accelerated from
M2,rel = 0.45 up to M3,is,rel = 0.93. There is a gradual acceleration on the pressure side
and the slight acceleration after the throat on the suction side in order to delay the sepa-
ration. No strong shock is observed in the trailing edge due to the subsonic Mach number
and the continuous acceleration up to trailing edge. region.
The pressure difference existing at a given axial position between pressure side
and suction side results in a tangential force which makes the blade row rotate in the case
of a rotor. Applying conservation laws, it can be shown that the tangential force 8vecU
applied on a blade is proportional to the circulation of the velocity around the blade
(M=v/a is an image of the velocity):
= ρv∞
U
v dS (2.2)
C
where C represents the blade contour. Assuming uniform inlet and outlet conditions:
= (Vinlet,t − Voutlet,t ) ∗ g
v dS (2.3)
C
In other words, the surface between the velocity curves of the suction side and pressure
side is proportional to the force applied on the blade. In the case of a rotating blade row
the enthalpy drop across the rotor can becomputed using the well-known Euler equation
for turbomachines:
∆H0 = u.(Voutlet,t − Vinlet,t ) (2.4)
In other words, the work recovered in a rotor depends on the tangential deviation of the
flow performed across the blade row. For this reason, in modern engines, the stator per-
forms a large deviation (typically 70 deg.) allowing the rotor to perform a large turning
(over 100 deg.). Then, the turbine stage is said ”highly loaded”. The exit angle is often
close to axial in order to limit the losses due to the swirl in the exit flow or in order to
allow the presence of a downstream stage. In the case of the Brite stage for example, the
outlet stator exit angle is 72 deg. and a turning of 107 deg. is performed in the rotor.
The stage exit flow angle is 12 deg.
21
The way the flow velocity is distributed on each side of the blade is an image
of how the load is distributed on the blade. A front-loaded loaded blades has a large
acceleration on the soon suction side whereas an aft-loaded blade exhibits the largest ac-
celeration on the late suction side.
The isentropic Mach number distribution can also provide information on the
boundary layer state. For example, the discontinuity on the suction side of the vane Mach
number is likely to cause transition.
In the laminar state the flow is organized in successives layers sliding on top
of each other without interaction between layers; the first layer is at null velocity (ad-
herence to the wall) whereas the last is at the velocity of the main flow; in this state,
the boundary layer is usually thin and generates low losses; low velocity fluctuations are
observed as well as low heat transfer rates because the layer constitutes a thermal insulator.
In the turbulent state the flow is not well organized anymore but submitted
to high velocity fluctuations in all directions characterized by the turbulence level; the
mixing of particles is very intense, the layer becomes thicker than the laminar one, high
losses are generated, high heat transfer rates are observed. The boundary layer thickens
faster than in the laminar case. Steep velocity gradients are observed at the wall. A tur-
bulent boundary layer generates high losses and high heat transfer rates which are both
not desirable. The solution is to avoid the transition from laminar state to turbulent state.
For this reason, many efforts were concentrated in understanding, modelling and trying
to predict transition.
In the transitional state the boundary layer oscillates between a laminar and
turbulent state; This state is usually characterized by the intermittency factor which is, in
a given point, the fraction of time during which the boundary layer is turbulent. Several
transitional modes were identified.
-The natural transition (see Fig. 2.9.a) also called Tollmein-Schlichting transition or attached-
flow transition occurs usually for low free stream turbulence levels (< 1%) and originates
in the boundary layer itself. For a critical value of the Reynolds number, small ampli-
tude two-dimensional Tollmein-Schlichting waves are observed inside the boundary layer;
they are progressively amplified while convected downstream and become 3D disturbances
erupting in turbulent spots which coalesce later to form a fully developed turbulent bound-
ary layer.
-The bypass transition occurs at high free stream turbulence levels (> 1%) or due to sur-
face roughness. The free-stream disturbances induce directly turbulent spots inside the
boundary layer; Tollmein-Schlichting waves and 3D disturbances are ”by-passed”.
-The separated-flow transition (see Fig. 2.9.b) occurs behind an obstacle (trip wire) or in
adverse pressure gradients. The boundary layer leaves the surface to form a free shear
layer in which transition takes place and may be followed by a turbulent reattachment.
22
A laminar-separation/turbulent-reattachment bubble is then observed in which the same
process than natural transition can take place.
-The shock induced transition is caused by the boundary layer shock interaction. Different
shock patterns can be observed depending if the interaction takes place with a laminar or
turbulent boundary layer.
-In the case of external cooling, transition can be induced due to coolant ejection across
the boundary layer.
23
On the rotor blade, the boundary layer remains laminar untill the very late suction side
where the transition is caused by a decceleration. This laminar state is justifed by the low
rotor Reynolds number (Rec,r = 0.5 106 ).
Due to the pitchwise velocity gradient, the outlet flow field is usually non-
uniform in terms of pitchwise static pressure distribution of pitchwise flow angle. This
non-uniformity increases with Mach number. In particular, sharp discontinuities are ob-
served in transonic flows due to the influence of shock waves. Behind the blade trailing
edge, the suction and pressure sides boundary layers are collected in the wake . This
wake is characterised by a total pressure drop concentrated in a small part of the pitch
in contrast with flow angle and static pressure which are varying all along the pitch. In
the case of cooled blade simulation, a thermal wake is also observed. The wake is also the
centre of high turbulence levels. High frequency non isotropic fluctuations are observed
due to the well known Von Karman vortex street but detailed informations on the time
varying wake characteristics is still needed (Cicatelli et al.[10]).
In the case of the Brite stator, pitchwise non-uniformities like total pressure
drop due to the wake, static pressure variation due to the shock and outlet flow angle vari-
ation were measured at DLR Göttingen in a two-dimensionnal cascade facility. Fig. 2.12
shows these non-uniform profiles at x/cax=0.813. At this axial position, the static pres-
sure varies by 10%; the downstream total pressure drop amounts to 8% of the inlet total
pressure. The outlet flow angle vcarise by 2.5 degrees across a pitch. The wake also ap-
pears on the Schlieren photography (Fig. 2.6). Of course, the wake spreads out when going
downstream. In order to quantify the wake shape evolution, each side of the wake can be
fitted by a gaussian curve and be characterized by the standard deviation σ (Amecke et
al. [5]). It was observed that the wake decays very fast in the near trailing edge region
and slower farther downstream as shown in Fig. 2.13. Similar observations were done by
Zaccaria et al. [11] who attributed the fast decay to the existence of high level of turbu-
lence intensity and large pressure gradients. The influence of coolant ejection through the
stator trailing edge on the wake appears to be small.
24
-downstream mixing losses due to the homogeneization of the flow when progressing down-
stream.
-cooling losses in the case of cooled blades if the coolant flow ejection at leading edge, on
pressure side or suction side affects the boundary layer state.
Profile losses constitute the main part of the overall amount of aerodynamic
losses. and can be estimated by measuring the total and static pressure downstream of a
cascade. Several loss coefficients can be defined. Here the kinetic energy loss coefficient
defined as the ratio of the kinetic energy leaving the cascade with the ideal kinetic energy
leaving the cascade will be used.
H02 − H2
η= (2.5)
H01 − H2,is
γ−1
Ts Ps γ
With h = cp T (ideal gas) and T0 = P0 (isentropic case) η can be ex-
pressed as :
(γ−1)
1− P2 γ
T02 P02
η=
(γ−1) (2.6)
T01 γ
1− P2
P01
Losses are often expressed by the local kinetic loss coefficient:
ζ =1−η (2.7)
An average value can be obtained by introducing in the formula the aver-
aged quantities instead of the local one. Several degrees of refinment can be reached in
computing the averaged quantities: area averaged, mass averaged, momentum averaged...
Pitchwise averaged values at 0.25*Cax and 0.54*Cax derived from measurements behind
the Brite stator at mid-span are respectively ζ = 2.79% and ζ = 3.13%. Of course, the
amount of kinetic losses increases with axial distance due to wake mixing.
A large number of cascade tests were performed for widely different geometries.
A number of experimental correlations were derived including influence of Reynolds num-
ber, Mach number trailing edge thickness, blade loading, pitch to chord ratio. A method
to evaluate the overall losses from those correlation is proposed in [12] by Sieverding. As it
is desirable to keep a laminar boundary layer over the largest part of the blade, the blade
designer will try avoid separation and even more flow separation without reattachment.
This is easily achieved on the pressure side of the blade where the favorable pressure
gradient helps to keep the laminar state. The problem is different on the suction side
where transition usually cannot be avoided. Recently, the design of highly loaded blade
makes the transition occur sooner on the blade. More and more tools are available to the
designer like inverse design codes which allow to find blades geometries corresponding to
a prescribed velocity distribution (Leonard [13]). The trailing edge thickness also plays
an important role and will be kept as small as possible in the limits of the mechanical
resistance. This is difficult to achieve especially in the case of blades where the coolant
is ejected through the trailing edge. In order to avoid this in the Brite stator case, the
coolant is ejected on pressure side just before the trailing edge.
25
2.6 Heat transfer
The design of a blade cooling system requires information on the convectional
heat transfer coefficient around the blade. In a first step, those coefficients are determined
around the uncooled blade. Then, the cooling system can be designed in order to min-
imise the mean blade temperature and obtain a blade temperature distribution as close
as possible to isothermal state.
The heat transfer around a blade is closely linked to the boundary layer state;
for this reason, it is sensitive to the same parameters than the boundary layer i.e. Reynolds
number, free stream turbulence level, curvature, pressure gradient, surface roughness, etc.
An example of heat transfer coefficient distribution along the blade curvilinear abcissa of
a higly loaded turbine blade is presented in Fig. 2.14 (see Arts et al. [14]).
In the leading edge area, high heat transfer rates are observed with a maximum
at the stagnation point. In this point, the velocity is zero and the boundary layer has not
yet developed leading to very high heat transfer rates. Under the influence of both the
large convex curvature and the favorable pressure gradient on the soon suction side and
pressure side, a laminar boundary layer develops insulating the blade from the hot gas.
Due to this, the heat transfer coefficient decreases very fast in this region.
Further on the pressure side, the curvature becomes concave. The boundary
layer state depends then on the Reynolds number, the free stream turbulence level and
on the importance of the destablising effect of the concave curvature with respect to the
stabilising effect of the favorable pressure gradient. However, in most of the cases, the
boundary layer remains in a laminar or a transitional state.
On the pressure side, the boundary layer remains laminar till the trailing edge
whereas on suction side transition to turbulence is observed at about s=75% (s: curvilinear
abcissa). In a turbulent boundary layer, high velocity fluctuations exist in all directions
which cancel the previous insulating effect. Heat transfer rates are even higher than at
leading edge.
An increase in the free stream turbulence level enhances the overall heat trans-
fer level mainly on the portions of the blade where the bondary layer is laminar (see
Fig. 2.14). The transition starts slightly sooner for the highest turbulent rate. At a higher
Reynolds number, the influence of free stream turbulence is even more striking. The tran-
sition on suction side with Re = 2106 and Tu=6% is detected at s=25(see Arts et al. [14]).
On pressure side, whereas the boundary layer is most probably laminar for Tu=1%, it is
fully turbulent for Tu=6%.
An overall increase of heat flux was also observed with increasing Reynolds
number. For the highest Mach number (1.1), the increase of Reynolds number also caused
the transition point to move toward leading edge.
The free stream Mach number does not cause any significant change in heat
transfer level.
26
the heat transfer rates to decrease. As expected, the pressure side boundary layer is less
sensitive to injection and comes back to laminar state very soon after the disturbance. The
increase in blowing ratio rate causes higher rates on all pressure side. On suction side, just
after the last injection row, the contrary is observed. This is explained by different type of
interaction between the coolant and the boundary layer. An increase in the density ratio
(CO2 instead of air) results in higher heat transfer levels (Shichuan et al. [15]).
27
2.7 Figures
e
lan
a lp
ion
rid S3
Me
Orthogonal plane
S2
spanwise
θ ne
e pla
ise α lad
mw ob
a S1 et
tre ad
s Bl
pitchwise
0.14
0.12
0.10
0.08
[m]
0.06
0.04
0.02
0.00
-0.01 0.01 0.03 0.05
Figure 2.2: Straight cascade from stator blade cylindrical cut at mid-span
28
Inlet angle
α1
Gauging angle
(arccos (o/g))
Outlet angle
α2 Xblade
Throat (o) Chord (C)
Mechanical frame of reference
Pitch (g)
Stagger
Xcascade
Xblade
1.6
2D Euler computation
1.4
From measurements (DLR)
1.2
1.0
M2,is
0.8
0.6
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Xbl/C
Figure 2.4: Mach number distribution around the Brite stator blade computed with a 2D
Euler code; comparison with experimental data
29
γ Camber angle
ide
ns
c tio
Su
Pressur
e side
Leading edge
Mean
Camberline
Trailing edge
Figure 2.6: Schlieren photography of the flow around the Brite stator blade from Amecke
et al. [5] M2,is = 1.05, cm = 3%
30
0.14
0.12
0.10
0.08
[m]
0.06
0.04
0.02
0.00
-0.01 0.01 0.03 0.05
1.2
1.0
0.8
Mis
0.6
0.4
0.2
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Xax/cax
Figure 2.8: Computed Mach number distribution around the rotor blade mid section
31
Figure 2.9: Illustration of natural and separated flow transitions
500
400 0.25
Velocity [m/s]
0.20
300
0.15
200
0.10
100 0.05
0 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
s/ste s/ste
Momentum thickness [mm]
4
0.25
Shape factor
0.20 3
0.15
2
0.10
1
0.05
0.00 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
s/ste s/ste
Figure 2.10: Computed boundary layer characteristics for the stator mid section
32
Computation for 2% turb.
Velocity [m/s]
0.20
300
0.15
200
0.10
100 0.05
0 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
s/ste s/ste
Momentum thickness [mm]
4
0.25
Shape factor
0.20 3
0.15
2
0.10
1
0.05
0.00 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
s/ste s/ste
Figure 2.11: Computed boundary layer characteristics for the rotor mid section
Figure 2.12: Total pressure, flow angle and static pressure profile behind the stator M2,is =
1.05, cm = 0%, x/cax = 0.813 from Amecke et al. [5]
33
Figure 2.13: Wake shape evolution with axial distance behind the Brite stator from Amecke
et al. [5] M2,is = 1.05, cm = 3%
Figure 2.14: Heat transfer coefficient around a transonic vane from Arts [14] M2,is = 1.07,
Re2,is = 106
34
Figure 2.15: Heat transfer coefficient with and without film injection from Shichuan [15]
35
36
Chapter 3
37
Fig. 3.1 compares the experimental spanwise outlet isentropic Mach number
of the Brite stator with the solution of Eq. 3.1 using the gauging angle and the exit Mach
number at mid-span (M2,is = 1.05) as input and assuming constant enthalpy and entropy
radial gradients . The experimental isentropic Mach number was determined from the
stator inlet total pressure and from pitchwise averaged static pressure measurements per-
formed at hub and tip wall assuming a linear evolution across the channel using Eq. 2.1.
The simple modelling of Eq. 3.1 gives a good agreement with the measurements. The hub
part is supersonic whereas the top part is subsonic; for this reason the stator is called
”transonic”. The supersonic exit flow implies the appearance of a trailing edge shock with
variable shock strength from hub to tip.
Radial gradients also exist in the outlet angle and depend on the blade design
(vt in Eq. 3.1). But there exists additional non-uniformities in three-dimensional flow.
Other features that were not considered in the bidimensional approach are the
hub and tip endwall boundary layers. They are growing along the inlet section resulting
in non-uniform radial total pressure and total temperature profiles. Inlet total pressure
and temperature meridional profiles measured 1/3Cs,ax ahead of the Brite stator blade
leading edge in the VKI annular cascade (Sieverding et al. [5]) are presented in Fig. 3.2.
Other vortices appear behind the blade trailing edge which constitute the trail-
ing edge vortex sheet.
-The trailing filament vortices are due to the stretching of the inlet vortex filaments which
38
have different velocity and path length along suction side and pressure side.
-The trailing shed vorticity is due to the spanwise change of the flow circulation around
the blade.
Both trailing filament and trailing shed vorticies rotate opposite to the sense of rotation
of the passage vortex (Sieverding [22]).
Flow visualizations and secondary velocity vector plots behind blades with high turning
show those vortices (Sieverding [22], Yamamoto [20]).
The system passage-vortex, trailing edge vortex sheet depicted in Fig. 3.3 con-
stitutes the classical secondary flow model . The term secondary flows is used in contrast
with an ideal primary flow which would take place in a cascade with infinite aspect ratio.
In fact other vortices were identified and their interaction with the classical secondary flow
pattern is described below.
The pressure side leg of the horseshoe vortex (HP) travels towards the suction
side as it progresses in the streamwise direction due to the blade to blade pressure gradi-
ent. It meets the suction side near the point of minimum pressure. The sense of rotation is
the same as for the passage vortex and the HP vortex core merges with the passage vortex
reenforcing its vorticity. From this point on, the pressure side leg cannot be distinguished
from the passage vortex. The boundary layer is highly disturbed by the passage vortex
(which now includes HP): a part of it gets trapped into the vortex while another part is
swept toward suction side; separation lines are drawn on the endwalls (Sieverding [22]) as
shown in Fig. 3.4.b. Zone I shows the area in which the deepest part of the inlet boundary
layer is involved: it climbs on the suction side just before the point of minimum pressure
and appears at the exit between the two separation lines S1 and S2. Zone II is located
downstream of the separation line S2 which corresponds to the location where the pressure
leg separates from the wall. An oil flow visualization performed on the Brite stator clearly
shows the limiting streamlines of hub and tip passage vortices (see Fig. 3.5 (Sieverding
et al.[5]) The spanwise penetration of S2 gives an estimation of the passage vortex size.
In the case of Sieverding et al. [17] and Yamamoto [19] for example, the passage vortex
dimension is about a fourth of the blade span due to a moderated turning (about 70. deg).
In the case of Moustapha [24], the loss cores created by the passage vortices are seen at mid
span and are nearly merged in the exit plane due to a much larger turning angle (128. deg).
39
The suction side leg of the horseshoe vortex (HS) rotates in the opposite sense
with respect to the passage vortex and remains initially close to the suction side wall under
the influence of the blade to blade pressure gradient until its meets the passage vortex
(which includes the HP). At the point of minimum pressure in the cascade passage, the
HS meets the passage vortex. Then, the HS gets entrained in an helicoidal motion around
the passage vortex while contrarotating as it is shown in Fig. 3.4.a. For this reason, the
suction side leg of the horseshoe vortex can appear at different locations (mid span side,
endwall side) in the cascade exit plane depending on how fast the passage vortex rotates.
Depending on the blade turning, the passage vortex accomplishes typically one to two ro-
tations from leading edge to trailing edge. The rotational speed of the horseshoe vortices
is much higher.
The corner vortex is located in the endwall suction side corner. It rotates in
the opposite sense to the passage vortex and is difficult to visualize because of its small
size. However, its existence was deduced from an inversion of the overturning in the outlet
angle distribution near the endwall and from endwall flow visualizations. It appears mainly
in high turning cascades where a strong interaction between the limiting streamlines S1
and S2 takes place at the suction side wall (Sieverding [22]). In contrast with the passage
and trailing edge vortex sheet, those vortices can be predicted only by means of viscous
computations and are visible for example in a 3D Navier-Stokes calculation from Heider
et al. [21].
The vortices clearly influence the local flow angle distribution as well as the
pitchwise averaged flow angle. Fig. 3.6 shows the example of the influence of the passage
and the corner vortex on the pitchwise averaged flow angle.
The passage vortex is represented with secondary velocity vectors in the lower
half of an orthogonal plane. This orthogonal plane can be located inside the channel or
slightly downstream of the exit plane. Going from mid-span towards the hub, an under-
turning in the outlet angle is observed with respect to the mid-span value due to the top
part of the passage vortex. In the center of the passage vortex, secondary velocity vectors
are very small and the outlet angle returns to the mid-span value before facing a maximum
overturning when traversing the lower extremity of the vortex. Very close to the endwall,
the angle decreases again under the influence of the corner vortex. Experimental evidences
40
of such flow angle profiles profiles are given by Sieverding et al. [17] and by Yamamoto
[19], [20] In low aspect ratio high turning blades, the outlet angle distribution is affected
on the entire span due to the large extension of the passage vortex. This influence on flow
angle is also observed for a tip clearance vortex as observed by Heider et al. [21].
Meridional outlet flow angle distributions behind the Brite annular vane are
shown in Fig. 3.7. The gauging angle (arccos(throat/pitch)) used to solve the simple radial
equilibrium equation (Eq. 3.1) is also plotted. Due to the radial extension of the probe,
measurements in the endwall regions were not reported because they were affected by the
high pressure gradients existing in these regions. Only the underturning at hub is clearly
due to the passage vortex is visible. The passage vortex influence is limited to the lower
quarter of the channel due to a limited turning (72 deg.).
41
local total pressure (P0 = P02 ) in contrast with the isentropic Mach number plotted in
Fig. 3.1 where the upstream total pressure was used. The difference between the two
figures is the result of total pressure losses due to profile losses and secondary flow losses.
The boundary layer reduces the Mach number very fast in the close endwall region.
As secondary losses are linked to the passage vortex strength, they will in-
crease with thicker inlet boundary layer, lower aspect ratio, higher turning angle. They
are far from negligible and can contribute up to 30-50% to the the overall losses (Sharma
et al. [23]). Secondary losses reduction is not an easy task. From the mechanical point of
view, low aspect ratio blades with high loads (high turning) allow to minimize mechanical
constraints and rotational speed but this will increase the amount of secondary losses. Tip
endwall contouring can be used to redistribute transverse and radial pressure gradient in
order to minimize tip losses (Boletis [26]). Another possibility is to use design parameters
like blade lean or bow in order to redistribute and improve the endwall loads as well as
transverse and radial pressure gradients. Endwall profiling, lean and bow effects can be
studied with 3D-Euler codes and clear improvements were obtained compared to cylindri-
cal blades and constant channel height.
At the casing, the relative flow has a velocity opposite to the peripheral speed
whereas on top of the blade the relative velocity is zero. The tip clearance flow at mid-
clearance is schematically depicted in Fig. 3.13. On the front part, the flow is dominated
by the casing and the flow goes from suction side toward pressure side. in the back part,
the flow is dominated by the pressure gradient between the pressure side and the suction
side which pushes the flow through the tip clearance. A large pressure gradient is visible
on the pressure side where the leakage flow enters the clearance (Heider et al. [21], Ya-
mamoto [27], Yamamoto et al. [28]). Once the flow has traversed the tip of the blades,
it rolls itself into the so-called tip-clearance vortex. Notice also the presence of a smaller
vortex on the suction side of the blade due to the scraping of the boundary layer by the tip
of the blade (scrapping vortex) . The tip-clearance vortex size increases with increasing
tip-clearance. It tends to push the tip passage vortex towards mid-span and results in
a local wake broadening. The strength of the interaction between the tip-leakage vortex
and the passage vortex was found to depend on the gap height and the overall pattern is
sensitive to flow incidence (Yamamoto [27]).
The largest part of clearance losses is produced in the small tip-clearance gap
where the flow is submitted to strong shearing forces. The mixing of the tip clearance
vortex will result in additional losses but to a lesser extent. Tip-clearance losses increase
with growing clearance gap/blade height ratio. In the case of unshrouded blades, it seems
that the presence of a recessed endwall right after the stator does not reduce significantly
the losses (Heider et al. [21]). Moreover, the clearance depends on the operating point of
the engine (radial deformation of the rotor under the influence of the centrifugal force) as
42
well as some unsteady phenomena due to a different thermal expansion of the rotor and
of the casing. Active control tip clearance systems are used to reduce those effects.
Shrouded blades can be used to reduce the clearance (like on the high pressure turbine
of the Rolls Royce Trent 800) but at the expense oh higher centrifugal constraints and
additional manufacturing costs.
43
3.7 Tables, figures
1.0
0.25*Cax
0.54*Cax
0.8 ISRE
0.6
y/h
0.4
0.2
0.0
0.90 0.95 1.00 1.05 1.10 1.15
M2,is
44
1.0
0.8
0.6
T0/T0mid
Y/h
P0/P0mid
0.4
0.2
0.0
0.95 0.97 0.99 1.01
45
Figure 3.4: Endwall flow structure from Sharma [23]
Figure 3.5: Oil flow visualization behind the Brite stator blade row from Sieverding et al.
[5]
46
Blade to blade plane
ur ning
vert
O Flow angle
urning
Un dert
Orthogonal plane
Passage vortex
Channel height
Underturning
Overturning
Corner vortex
1 pitch Flow angle
47
1.0
X/Cax=0.25
0.8 X/Cax=0.54
Gauging angle
(arccos(o/g))
0.6
Y/h
0.4
0.2
0.0
69 70 71 72 73 74 75 76 77 78
Outlet angle α2 [deg.]
Figure 3.7: Meridional outlet angle distribution behing the Brite stator
48
Linear Contour Plot.
Density lines Step = 2.0000
.0 7.0
9.011 53.0
.0 1.0
0 9.0
13.0 15.017. 11.0 7.0
.017.0 3.0
5.0
50. 13.0
15 1.0 5.0
0
3.
7.0
7.0
5.0
5.0
1.0
1.0
3.0
1.0
7.0
9.0
3.0
y
9.0
25.
5.0
1.0
3.0
0
3.
7.0
1.
.0
0
1.0
5.0 .0 1317.0 15
1.0 11 .0
5.0
7.0
9.0 0
3.11
7.0
.0 5.013.0 7.0
7.0
15.0 17.0 15.0
3.0 11.0 5.01317.
.0 0 17.0
-25. 0. 25.
x
Figure 3.9: Kinetic loss carpet plot at 0.25 Cax downstream of stator trailing edge
11
1.0 7.0 .0 11
5.0
13.0 .0
13.0
0
5.
15
15.0
.0
1.0
5.0
9.0
1.0
3.0
7.0
7.0
5.0
3.0
9.0
3.0
y
9.0
7.0
25.
9.0
3.0
1.0
0
7.
0
5. 5.0
11.0
7.0
1.0
1.0
5.0
7. 15.0
13.0 15 101.0 .0 17.0 9.0
.0 13
0
.0
3.
0
3.
11
19.0
.0
-25. 0. 25.
x
49
1.0
0.8
X/Cax = 0.25
0.6 X/Cax = 0.54
Y/h
0.4
0.2
0.0
0 5 10 15 20 25 30 35 40
ζ (%)
Figure 3.11: Pitchwise average of kinetic losses
1.0
X/Cax=0.25
0.8 X/Cax=0.54
ISRE
0.6
Y/h
0.4
0.2
0.0
0.80 0.90 1.00 1.10 1.20 1.30
Mach number
Figure 3.12: Meridional Mach number profile behing the Brite stator
50
Figure 3.13: Schematic view of tip clearance flow at mid-clearance
51
52
Chapter 4
In the two previous chapters, stator blade row and rotor blade row were sim-
ulated with fixed blade rows in steady flow; in the bidimensional case, the inlet flow was
uniform whereas in the 3D aspects, the inlet was also uniform in the pitchwise direction
but had spanwise total pressure and temperature gradients due to endwall boundary lay-
ers. This last simulation is correct for an inlet guide vane but not for a stator preceded by
another stage. In this case, the inlet flow is unsteady and non-uniform in all directions.
Inlet flow angle and inlet total pressure profile are affected by the secondary flows of the
preceding stage as well as by the tip-leakage flow of the rotor and swirl of the entire flow
field (Boletis et al. in [30]). A similar remark can be done for the rotor inlet flow: com-
plex spanwise and pitchwise gradients exist at the stator outlet which the rotor crosses
periodically behind each vane. The steady static pressure field locked to an isolated blade
row is submitted to a rotor-stator interaction resulting in a periodic modulation of the po-
tential field. The wake, characterized by a total pressure drop, a total temperature drop,
a non-uniform angle and a high turbulence level is traversed by the rotor which results
operates then at off-design conditions. Radial gradients are encountered as well as strong
secondary flows in low aspect ratio vanes. In the case of a transonic stage, the rotor will
see in a relative frame the effect of shock waves emerging from the vane (total and static
pressure discontinuity). Finally, the rotor flow is submitted to large centrifugal forces and
Coriolis forces and is influenced by tip leakage flow.
As a result, instead of the steady flow assumed before, the rotor blade inlet
flow is a complex 3D non-uniform unsteady flow which results from the flow exiting the
vane and from the rotor-stator interaction. In order to study the amplitude and the
influence of those phenomenon, new types of facilities and instrumentation are needed.
53
Artificial generation of unsteadiness in cascades
The wake existing behind the trailing edge of a blade can be efficiently mod-
elled by the wake behind a cylindrical bar with a diameter equal to the trailing edge
diameter (Doorly [31]). Velocity deficit and shape as well as turbulence level are repro-
duced. The idea is then to simulate the stator blade row or the rotor blade row by a
fixed straight cascade and make the bars rotate in front of the cascade to obtain unsteady
periodic wakes or shock waves impinging on the blade.
The wake width can be varied by changing the bar diameter, the velocity
deficit and shape can be varied by changing the distance between the bars plane and the
blade leading edge plane, the Strouhal number can be changed by varying the displacement
speed or the number of bars. In the arrangement of Schoberei et al. [37], four different bar
spacings are used on the same belt allowing four different Strouhal numbers in a unique
test.
Studies were performed on flat plates, curved plates and blade profiles. There
exists a restriction for the instrumentation which can be used for this type of study linked
to the frequency response. The probes must be able to resolve phenomena at frequencies
higher than the bar passing frequency. For this reason, the instrumentation used is mainly
composed of hot films, thin films, fast response pressure transducers and hot wires. Some
figures and characteristics of the experiments selected in the literature to illustrate un-
steady flow in a blade row are reported in table 4.1.
However the resulting flow patterns do not include secondary flows. The in-
fluence of a rotor on the upstream stator flow cannot be studied, centrifugal and Coriolis
forces are not simulated and tip-clearance flow cannot be reproduced. In the case of shock
wave simulation, the resulting pattern and shock strength may be very different from the
one encountered in a real turbine stage.
54
Full turbine stage facility
In a turbine stage test rig, the 3D stage geometry can exactly be reproduced
and the blade interference problem silulated. It is cheaper to operate than a full engine
because the other components like additional turbine stages, combustion chamber, com-
pressors are not present. Moreover, as the simulation is performed through reproduction
of Mach number, Reynolds number and temperature ratios, level of pressure and temper-
ature are much lower than those encountered in real engines. This provides a considerable
gain in energy but also allows to use instrumentation under mild conditions. The conver-
sion of an annular cascade facility into a full turbine stage facility is not an easy task. The
presence of the rotor brings many complications due to large amounts of power involved,
high levels of centrifugal forces, vibrations control and oil lubrication problems which ex-
plains that only a few turbine stage facilities exist in the world.
For obvious reasons, blade interference problem can be better studied in full
turbine stage facility. Until recently, most investigations were carried out in large low
speed facilities. Because of the problem of data transmission associated with unsteady ro-
tor measurements, some researchers opted for for wake blade interference studies between
a rotor wake and a downstream blade row in a one and a half turbine stage configuration
like Yamamoto [38] [28]. The stator of the first stage is far away from the rotor so that the
wake are completely mixed out when the flow enters the rotor with the correct swirling.
If this arrangement allowed very detailed studies, it does not simulate properly rotor flow.
Low-speed multistage facilities are usually dedicated to low pressure turbines like at the
Whittle Laboratory (Cambridge,UK, Hodson et al. [39]), or at General Electric (Cincin-
nati, Ohio, Halstead et al. [40]). In this environment, mainly hot films were used due to
a difficult access of inner stages. The flow field complexity increases from stage to stage
and it is not easy to separate the differents effects in the resulting signal which includes
all kind of interactions.
High speed continuously running closed loop facilities exist at ABB Baden
(Switzerland), Onera Chatillon (France), NAL Bangalore (India) and the largest at DLR
Göttingen (Germany) with a mean diameter of 512 mm (see Amecke [6]). A huge amount
of power is usually need to run a turbine stage continuously (1MW is needed at DLR
Göttingen to operate the 500 kW turbine stage).
The limitation of cold flow facilities to aerodynamic investigation is a serious
drawback when the steadily rising turbine temperature requires more and more informa-
tions on the turbine heat transfer characteristics. The recent development of different
types of short duration blowdown facilities has opened the way of covering both aerody-
namic and heat transfer aspects. Those facilities are capable to simulate all important
engine characteristics like Reynolds number, Mach number, wall to gas temperature ratio,
coolant to gas temperature ratio and turbulence levels. They operate in the range of 1 to
3 MW with full engine dimension or even with scaled-up versions.
55
aphragm or a fast opening valve
- a dowstream dump tank separated from the test section by a throttling valve.
The various facilities differ by their air supply and type of power absorbtion
system. A large pressurised tank is used at MIT (Epstein et al. [41]) and Whright
Patterson Air Force Base (Haldeman et al. [42]). The tank is equiped with a double
shell in between which hot oil can be circulated in order to heat-up the gas inside the
reservoir. A shock tube is used at Calspan corp. Buffalo; the pressure and temperature
levels in the upstream tube are obtained with a shock travelling in the tube. This solution
is more economic and faster than preheating the upstream tank; on the other hand, testing
times are extremely short. Oxford (Ainsworth et al. [43]), DRA Pyestock (Hilditch et al.
[44]) and more recently VKI make use of the Light Piston Compression Tube principle
developped at Oxford. A light piston placed in the upstream cylindrical tank is pushed by
cold pressurized air and realizes downstream of the piston a quasi isentropic compression
rising at the same time pressure and temperature. Here also, compression and air heating
are realized in a fast and cheap way with test times above the one of the shock tube. The
three facilities are equipped with a full turbine stage. The VKI facility will be described
later with more details.
As regards the power absorption systems, an eddy current brake operating
during and after the blowdown is used at MIT, an axial turbo brake placed behind the
stage in the test section is used at Pyestock (Goodisman et al. [45]), inertia wheel and
turbine brake is used at VKI, whereas Oxford operates without power absorption system.
In order to obtain a γ closer to the one of burned gas, a gas mixture of Argon-
Freon was formerly used at the MIT turbine facility. The tests are now done with air for
environmental reasons.
Detailled information on those facilities can be found in Amecke [6]. Some fig-
ures concerning some of the above mentioned turbine stage facilities are reported table 4.2.
56
4.2 Unsteady inviscid flow field
Due to the velocity gradients in the blade passage, an upstream and down-
stream non-uniform static pressure field is associated with any blade row. In subsonic
flows, the pitchwise pressure gradients are rather smooth and decay exponentially with
axial distance in both direction with a typical lenght scale of the chord (Hodson [46]). The
axial distance between the stator and the rotor blade row is usually much smaller than
one chord. The static pressure field is then modulated by the rotor blade traverses wich
affects the flow field in both upstream and downstream blade row. This phenomenon has
a purely inviscid potential character. Some studies suggest that the effect on the static
pressure fluctuations of the upstream row is greatest than on the downstream row. All
authors agree to say that this influence decreases very fast as the axial gap between blade
row is increased. For stator-rotor spacing above 5% Cs,ax , which is the case in common
turbines, the wake interaction remains predominant (Doorly et al. [31]).
In the case of transonic or supersonic stator exit Mach numbers, the outlet
pressure field is characterized by strong static pressure gradient caused by the trailing
edge shock influence. Due to shock losses, the flow field looses its potential character. The
influence on the dowstream rotor is obviously much more important than the influence of
the rotor on the upstream stator. This phenomena will be adressed in a later section with
more details.
An influence of the rotor on the wake path was observed in experiments (Hod-
son [47]) as well as in computations (Hodson [48], Giles [49]). In the absolute frame, the
wake appears like a velocity deficit with respect to the main flow. In the relative frame,
the wake appears as a jet with a relative wake slip velocity which pushes the wake toward
the suction side. From the point of view of the relative flow, not only the velocity changes
across the wake but also the flow angle. This phenomena is illustrated in Fig. 4.1 from
Dietz et al. [50].
The periodic wake impingment on the rotor blade affects the rotor blade wall
static pressure distribution. In its model, Meyer (see [50]) separates the influence of the
wake on the rotor static pressure distribution in two contributions reffered to as wake
effect and wake distortion effect. This model concerns a single airfoil experiencing the
impingment of a single wake.
The wake wake effect is a pressure perturbation which travels at acoustic ve-
locity. As soon as the wake hits the leading edge, the blade sees a decrease in the flow angle
which modifies the circulation around the blade. This results in an increase of pressure
on the suction side and a decrease of pressure on the pressure side. These fluctuations are
57
propagated at acoustic speed.
The wake distortion effect is secondary with respect to the previous effect.
It is due to the local influence of the wake on the suction and pressure side during its
convection in the flow. For this reason, this perturbation travels at the local convective
speed.
Both wake effect and wake distortion effect were observed experimentally.
In reality, not one but successive wakes impinge periodically on not one but
several airfoils. As a result, the perturbation felt on suction side will propagate at acoustic
speed across the channel to appear on pressure side and vice versa. The interpretation
of pressure fluctuations is much more complex. Whereas Hodson [47] mentions pressure
perturbations travelling at 70% of convection speed on suction side, Dietz et al. [50] found
perturbations travelling at acoustic on pressure side as well as on suction side. Notice that
this last observation is more coherent with Meyer’s theory.
As for steady flow, the suction side is the most interesting and the most studied
side of the blade because its contribution to the profile losses is predominant. The transi-
tion on suction side plays a major role for the boundary layer momentum thickness at the
trailing edge. The boundary layer state along the blade suction side curvilinear abcissa at
a given time using the intermittency factor γ is illustrated in Fig. 4.2. The evolution of
γ in the case of natural transition is presented in Fig. 4.2.a whereas Fig. 4.2.b presents a
picture of γ distribution in the case of wake-induced transition. The presence of the wake
in the main stream perturbs the boundary layer. However, the interaction does not start
at the leading edge but a little after. Turbulent spots are emerging periodically inside the
boundary layer with a small lag with respect to wake passing events: this is called wake
induced transition . The emergence of those turbulent spots is attributed to the local
increase of free stream turbulence level caused by intense wake mixing (Schoberei [33],
Ladwig et al. [52]). Then, the turbulent spot travels inside the boundary layer (peak in
58
intermittency in Fig. 4.2.b) whereas the wake moves on away from the boundary layer and
is decoupled from the turbulent spot (Orth [32]).
The leading edge and trailing edge of the turbulent spot travel at different
speeds: 88% of the free stream velocity for leading edge and 50% of free stream velocity
for the trailing edge in the flat plate case with zero pressure gradient (Schubauer and
Klebanoff [53]). Similar convective speeds were found around turbine blade suction sides
(Doorly et al. [31], Addison et al. [54], Orth [32]). The spot leading edge speed being
higher than the spot trailing edge speed, the spot becomes wider and wider when trav-
elling downstream and they unavoidably merge into a turbulent boundary layer. This
phenomena is illustrated in Fig. 4.3 which represents the state of the boundary layer along
the suction side in function of time. In the case of a constant free stream velocity, the
boundaries of the turbulent spots are straight lines, their slope represent the travelling
speed of the spot leading edge and trailing edge. In between the two lines, the boundary
layer is in the wake-induced transitional state. Notice that in the case of a blade, the free
stream velocity is not constant and those lines are curved.
After the passage of the turbulent spot in the area which would remain laminar
without the presence of wake, a becalmed region of quasi-laminar flow appears. This
phenomena was observed by Schubauer and Klebanoff [53] behind turbulent spots on flat
plates and more recently by Orth [32], Addison et al. [54], Schoberei et al. [33], Halstead et
al. [40] on the suction side of blades. The trailing boundary of this region convects at 30 %
of the free stream velocity, faster than Tollmein-Schlichting waves which cannot reach this
region. This becalmed region is more resistant to flow separation because shear stress levels
are higher than in the laminar zone. For those reasons, only the by-pass transition mode
or an obstacle can cause a transition to turbulence. This region is depicted in Fig. 4.3
and is followed by a bypass transition zone which ends-up in a fully turbulent zone as
one could expect. This fully turbulent zone occurs sooner than without the presence of
wakes (Hodson et al. [55], Addison et al. [54]). Another becalmed region appears after
the turbulent spots initiating in the bypass transition zone. This becalmed region can
extend down to the trailing edge although the undisturbed boundary layer would be fully
turbulent. In [56] and [8], Mayle proposes a modelling of a parameter f of the boundary
layer using a time-averaged intermittency factor taking in account a normal transition
mode intermittency and a wake-induced intermittency with an experimental correlation
for each parameter:
f˜ = (1 − γ̃)fL + γ̃fT (4.1)
where ˜ means time-average, L stands for laminar, T for turbulent.
To take in account the multiple modes of transition, the intermittency factor
is modelled as
γ̃(x) = 1 − [1 − γn (x)][1 − γ˜w (x)] (4.2)
where n stands for normal transition and w for wake induced transition.
On the pressure side, fluctuations with lower amplitudes are observed and the
boundary layer remains mainly laminar (Hodson [46]), Hodson et al. [55]) due to the large
favorable pressure gradient dominating the entire pressure surface.
59
The suction side topology will be affected by varying the same parameters
that are influencing the boundary layer in steady state.
-A low Reynolds number enlarges the laminar zones; the transition occurs later and weaker
wake-induced transitional strips are observed. A very low Reynolds number promotes a
laminar separation without reattachment.
-A low free-stream turbulence level promotes laminar separation without reattachment.
A high level induces a turbulent reattachment instead of a separation. An increase of the
turbulence level inside the wake causes the transitional strip to move toward leading edge
as well as the turbulent wake-induced strip. The becalmed region is then more extensive
and stronger (Halstead et al [40]).
-A negative pressure gradient will shift the transition point downstream (10% of the plate
length for example in Schoberei’s experiment [33]) increasing the laminar area and de-
creasing the turbulent area. No influence of the pressure gradient is felt on the turbulent
spot propagation speed (Orth [32]).
-As the stator-rotor axial gap is reduced, the transitional area moves towards the leading
edge. For a spacing of 50% Cs,ax Hodson et al. [55] and Addison et al. [54] observed a
transition occurring 10% s sooner than for 0.75% Cs,ax and 20 % sooner than for 145%
Cs,ax .
-At high Strouhal number, an earlier mixing of the wake is performed which finally leads
on the suction side to a complete degeneration of the deterministic periodic wake flow into
a stochastic turbulent flow where becalmed region are absent (Schoberei et al. [37], Liu et
al. [34], Halstead et al. [40].
-The nozzle clocking effect was observed by Halstead et al. [40] with a succession vane-
blade-vane. When the upstream vane is placed with a shift of 40% of the pitch with respect
to the downstream vane, an additional transitional strip followed by a turbulent strip is
observed in the time distance diagram of the downstream vane. This strip appears peri-
odically in between the rotor wake-induced transitional strips caused by the convection of
the upstream vane wake across the rotor.
As in the steady case, losses are mainly due to boundary layer development
around the blade suction side and profile losses are higher in the presence of wakes (50%
higher in the case of Hodson [46]) due to the existence of turbulent spots, a sooner tran-
sition, a larger transitional area and turbulent area. Losses increase when increasing the
Strouhal number, the free stream or wake turbulence intensity and when decreasing the
axial spacing between rotor and stator. A diminution of losses is observed when increas-
ing the negative pressure gradient. The influence of Reynolds number on the losses is
more complex. In the cascade case, the profile losses decrease when the Reynolds number
increases (low pressure turbine) because high Reynolds number promotes turbulent reat-
tachment which generates less losses than laminar separation without reattachment. In
the presence of wakes at low Reynolds number the losses are lower than in the no-wake
case because the incoming wakes induce a turbulent reattachment which would not occur
otherwise (Hodson et al. [55]).
60
could clearly be associated with a travelling turbulent spot induced by the wake passage in
several cascade experiments (Doorly et al. [31], Ashworth et al. [35], Liu et al. [34]). The
amplitude of the fluctuations is very large at leading edge and tends to decrease on both
sides when progressing downstream. Notice that in the steady case, the heat flux decreases
very fast on either side of the stagnation point. In the unsteady case, the stagnation point
is likely to move; the gauge position and size must be carefully chosen. As for the steady
case, the mean heat transfer levels are closely related to the boundary layer state. As
it was found that the wake interaction causes larger transitional and turbulent regions
along the blade suction side, an enhancement of the overall heat transfer is observed.
The increase of the Strouhal number extends the surface of transitional and turbulent
boundary layer and causes a further enhancement (Dullenkopf et al. [57], Shichuan et al.
[15]). This is illustrated in Fig. 4.4 showing the time averaged Nusselt number for several
Strouhal numbers. Although the boundary layer remains laminar in a large part of the
blade suction side, the heat transfer level can be improved up to 60%. This is due to a
combined effect of the laminar boundary layer disturbed by turbulent spot and increase
of free stream turbulence due to wake travel and mixing. This effect is distinct from the
transition to turbulence characterized by the peak on the late suction side.
The increase in heat transfer level due to film injection is much larger than
the one provided by the presence of wakes. As expected the presence of both film injection
and wakes causes a further increase in the heat transfer level (Shichuan et al. [15]).
Hot streaks coming randomly out of the combustion chamber were also found
to influence the unsteady heat transfer. Those hot streaks are behaving like a jet (high ve-
locity due to high temperature) in the main flow. This gives them a tangential component
in the relative flows which directs them toward the pressure side whereas the wake, low
momentum fluid, has an opposite component and is directed toward suction side. This
segregation results in a hotter pressure side than the suction side (Sharma [58]). Similar
observations were made with unsteady 3D Euler code calculations (Saxer et al. [59]) which
revealed further details on radial migration of the gas. On the suction side of the rotor
blade, a small radial migration of the cooler gas from hub and tip toward mid-span takes
place; on the pressure side, the hot fluid is migrating from mid span toward hub and tip.
In the case of transonic exit Mach number, the vane trailing edge shock wave
constitutes an additional non-uniformity.
The shock waves travel across a rotor passage will be illustrated mainly with
61
an unsteady two-dimensional on a full turbine stage by Giles et al. [60] over one rotor
blade cycle in Fig. 4.5. The stator exit Mach number was 1.12.
On of the clearest phenomenon is the sweeping of the shock from the crown of the blade
toward the leading edge (from t=0.875 to t=0.375). This phenomena was also clearly
identified in an experimental simulation of Doorly et al. [31] and Ashworth et al. [35].
They performed experiments in a straight rotor blade cascade preceeded by a spoked
wheel type wake generator which was rotated at a sufficiently high speed to generate
shock waves. Successives Schlieren pictures clearly show the same sweeping mechanism of
the shock from the crown towards the leading edge.
No direct impact of the shock can reach the late suction side which is hidden.
However, at t=0.375, a reflected shock impinges shortly slightly after the crown.
On pressure side, the direct shock sweeps the entire surface due to the fact
that it becomes curved close to the wall (from t=0.50 to t¡0.875). Then, on the second
half a the period, a reflected shock is sweeping the pressure side from the trailing edge
towards the leading edge.
Another mechanism which is put in evidence in the calculation is the reflected
shock generated at t=0.375 which travels upstream and reaches the trailing edge of the
stator.
A three-dimensional inviscid calculations from Saxer et al. [59] confirmed this
unsteady shock pattern evolution This exemple is in no case universal but illustrates very
well the principal mechanism of the shock-rotor interaction. The pattern described here
is likely to change with rotor-stator spacing, Mach number...
In a first series of experiments, Doorly et al. [31] used a small number of bars
in their shock-wake generator in order to be able to distinguish the wake passing events
from the shock passing events. The unsteady pressure traces showed that the shock was
responsible for a sharp peak in pressure moving from the crown toward the leading edge.
A broader peak due to the wake effect is observed travelling downstream.
In a real turbine stage, it is more difficult to separate shock and wake effects.
In some particular cases, the shock impinges on the blade immediately after the wake or
even at the same time. The computed unsteady pressure field around the blade section
from Giles et al. [60] is presented in Fig. 4.6. The sweeping of the shock creates a steep
pressure rises of increasing amplitude as the pressure discontinuity travels from the crown
towards leading edge. A similar pattern is observed in the measured unsteady pressure
distribution of Rao et al. [61] in a transonic turbine stage. Fig. 4.6 also shows that no
pressure fluctuations are observed on suction side after the point of impact of the reflected
shock. On pressure side, two pressure waves can be identified: one of large amplitude go-
ing from the nose towards trailing edge associated with the direct shock impact, another
of smaller amplitude travelling from trailing edge towards leading edge associated with
the reflected shock.
Peak values up to 60% of the inlet total pressure can be observed at the lead-
ing edge due to shock impingment. Fluctuations are much bigger on the suction side than
62
on pressure side and are higher at the hub and tip than mid-span. In spite of those large
fluctuations, the time averaged computed solution is very close to the computed steady
solution (Saxer et al. [59]).
The pressure unsteadinesses travelling upstream of the rotor towards the stator
trailing edge does not propagate into the stator beyond the throat where the shock stops
them (Saxer et al. [59], Rao et al [61]).
To have a more realistic simulation, Doorly [31] performed a second test se-
ries with a larger number of bars. The combined effect of shock and wake clearly affects
the transition of the portion between the crown and the leading edge in a different way
than the wake effect alone. At high Strouhal numbers, the boundary layer becomes fully
turbulent similarly to what was observed with wakes (Doorly et al. [31]) leading to very
high heat transfer rates.
On the pressure side, the effect of the shock wave is often small. In Guenette’s
case [62], no fluctuation is observed, the pressure side being completely hidden from the
vane trailing edge.
Differences are observed between the heat transfer levels obtained in a cascade
and a full turbine stage on the same rotor geometry. The mean heat transfer level as well
as the amplitude of the fluctuations are somewhat higher in a cascade (Hilditch et al. [64],
63
Garside et al. [65]); this behaviour is not fully understood.
Significant level of unsteadiness on the stator heat transfer rates can be ob-
served on the late suction side and near trailing edge on the pressure side (Hilditch et al.
[66]). As for pressure, these fluctuations do not propagate beyond the throat.
64
4.9 Figures, tableaux
Figure 4.1: Illustration of wake distortion effect from Dietz et al. [50]
Intermittency Factor
0
Natural transition in steady flow
Intermittency Factor
0
Curvilinear abcissa
65
Wake induced
Turbulent
Stator blade passing period transition
Bypass
transition
Laminar
8
0V
0.5
8
V
0 V
0.3
8
0.88
Becalmed
LE Curvilinear abcissa TE
Figure 4.4: Influence of wake on heat transfer level from Dullenkopf et al. [57]
66
Figure 4.5: Shock pattern evolution during one rotor blade cycle from Giles et al. [60]
67
Figure 4.6: Computed unsteady surface pressures around the rotor blade section from
Giles et al. [60]
Figure 4.7: Measured unsteady Nusselt number distribution around the rotor blade section
from Guenette et al. [62]
68
BPF: bar passing frequency
Wake Generator: SQ squirrel cage, SW spoked wheel, B belt
Cascade : F flat plate Cu: curved plate SC:stator straight cascade RC:rotor straight cascade
Instrumentation : HT : heat transfer gauges, HW: hot wire anemometry, HF: constant temperature hot films, FPr : fast response
pressure transducer, LDV: Laser Doppler velocimetry, Sch: Schlieren
: curved plate arc length
69
1st Author
Ref.,
Tu in %
cs [mm]
cr [mm]
Rec2 ∗ 10−6
Recr3 ∗ 10−6
M2
M3r
NS
NR
spa /Cs,ax
So
RPM
WPF [Hz]
Dm [m]
Facility Instrumentation
Halstead [40] 0.44 0.61 108 72 610 1100 1.219 2 ST LS HF on Mylar HW
Sheldrake [67] 2.7 0.95 0.96 36 60 6500 3900 1 ST dual hot wire
Hilditcht [66] 27.1 2.6 0.94 9500 0.55 1 ST HT Pr on blade
Rao [61] 5.0 67.6 47.5 1.12 1.05 30 45 11400 5700 0.548 1 ST Fpr HT
Hodson [39] 0.13 0.60 4 ST LP HF on 3rd ST R
Dietz [50] 31.6 2.7 0.23 0.27 36 60 0.29 8500 5100 0.510 1 ST FPr on blade
Hilditch [64] 2.7 0.95 0.96 36 60 8434 1 ST HT enamel coat
Dunn [68] 5.0 1.0 23 35 0.19-0.50 27000 10000 1 ST HT pyrex insert
Guenette [62] 59.0 25.7 2.7 1.18 6190 3300 1 ST HT on polyimide
Hodson [55] 0.42 0.31 36 51 0.50 2.0 530 318 1.294 1 ST LS HF(R) visu P0 trav.
Binder [25] 0.82 1 ST L2F
Binder [69] 0.80 0.45 1 ST L2F
Hodson [47] 152.4 114.3 0.42 0.31 36 51 0.75 1.4 530 318 1.294 1 ST HW
70
Chapter 5
The three preceeding chapters showed the complexity of the flow across a
turbine blade row. This is well summarized in the following consideration from Lakshmi-
narayana [70].
”Turbomachinery flows are among the most complex flows encountered in fluid
dynamic practice. In most instances, they are three-dimensional, with laminar, transi-
tional and turbulent flow; separated flows are frequently encountered. The flow may be
incompressible, subsonic, transonic, or supersonic; some turbomachinery flows include all
these flow regimes. The viscous and turbulent regions encounter complex stress and strain
due to the three-dimensionality, appreciable pressure gradients in all directions, rotation,
curvature, shock-boundary layer interaction, heat transfer and interacting boundary lay-
ers. The flow is dominated by vortical flows: secondary, leakage, trailing, horseshoe and
scraping vortices. The absolute flow is always unsteady in a rotor, and both the relative
and absolute flows are unsteady in a multi stage environment. The free stream turbulence
is usually high. A large number of flow parameters are encountered: Reynolds number,
Mach number, rotation number, Richardson number, Prandtl number, Eckert’s number,
Thoma coefficient, flow incidence, etc. The geometrical parameters are also complex and
many: camber, blade and blade-row spacing, varying thickness form hub to tip and from
leading edge to trailing edge, stagger and skew, lean, twist, dihedral, flare, aspect ratio,
hub/tip ratio, annulus and hub wall shape, leading and trailing edge radii, tip clearance,
blades with cooling holes, tandem blades, part-span dampers, etc.”
71
to operate at off-design. From these variations of aerodynamic loading will originate vi-
brations (flutter) and noise. The boundary layer state is greatly affected as well as heat
transfer which results in higher blade thermal fatigue. Lower stage efficiency as well as
unexpected high levels of heat transfer are observed. Although numerical codes produce
predictions which are well in agreement with measurements, the modelling of transition
and turbulence is still not satisfactory.
A large part of the work will be concerned with the development of measure-
ments techniques able to resolve phenomenon fluctuating at wake passing frequency. For
the aerodynamic part, fast response piezo-resistive sensors are used. For the heat trans-
fer part, thin film gauges measuring blade surface temperature will be investigated. The
cold wire resistance thermometer will be used for total temperature measurements. For
all measuring techniques, the measuring principle will be presented and the frequency re-
sponse capacity will be proven. Indeed, all the instrumentation must be able to resolve
fluctuations at blade passing frequency (4600 Hz from the rotor point of view, 6900 Hz
when behind the rotor). The mechanical integrity of the sensors will also be tested to be
sure they can support the high centrifugal loads during measurements in rotation. Fast
response pressure sensors were not tested extensively due to their recognised performances
for this type of application. Large efforts were put on the cold wire and the thin film gauge
measurements techniques. For example, original data processing techniques were devel-
oppped like numerical frequency compensation for the cold wire probe, numerical heat
transfer derivation from the thin film surface temperature measurements.
Another part of this work focuses on the installation of the turbine stage in
72
the existing VKI CT3 Isentropic Light Piston Tunnel dedicated up to now to full annular
blade row testing. The aerodynamic caracteristics and performances of the newly designed
HP transonic turbine stage will be presented. The CT3 wind-tunnel will then be described
as well at the new turbine unit and its operating cycle, data transmission system... Then,
the instrumentation placed in front of the stator, around the rotor blade mid-span section
and behind the rotor will be commented.
The experimental test program on the highly loaded transonic turbine stage
consists in:
- measuring mean and unsteady relative and downstream total pressure, static pressure,
and total temperature,
- measuring the mean and unsteady pressure and Nusselt number distribution around the
rotor blade.
The influence of three parameters was tested: rotationnal speed, stator coolant
ejection rate, axial spacing between stator and rotor.
Finally, the steady and unsteady data is commented. The time-averaged value
are presented in order to control if the turbine stage operates at its design point. The
relative unsteady total pressure and total temperature fluctuations will be commented
with the help of stationnary measurements performed behind the stator alone. Unsteady
static pressure and Nusselt number distribution around the rotor mid-section profile will
be presented. A mechanism explaining how the stator trailing edge shock can influence
the static pressure and the Nusselt number will be proposed. Finally, the influence of
rotational speed, cooling and axial spacing will be studied.
73
74
Part II
75
Chapter 6
6.1 Introduction
For a given wire diameter, there exists a limit in frequency due to the time
that the wires needs to take the gas temperature (thermal inertia of the wire); this is
usually characterised by the wire transfer function. Several methods can be used in order
to determine this transfer function. LaRue et al. [73] used an electrical heating method
which consists in heating the wire internally using a sine current. Weeks et al. [76] and
Fournier [77] generated heat pulses on the wire by means of a chopped laser beam focused
on its active part. All results concluded that the response of the wire is similar to the
response of a first order system.
When placed in an unsteady temperature field, this probe suffers from a con-
duction phenomenon between the prongs on which the wire is fastened and the wire itself.
77
The prongs represent a large mass compared to the wire and are not able to follow the
temperature fluctuations, therefore imposing a temperature which is different from the gas
temperature at the wire extremities. The development of cold wire benefited greatly from
the development of constant current hot wire anemometry. Lag in the response of the
wire, compensation circuits and conduction with the prongs are all mentioned in a survey
on this probe by Schubauer [78] dealing with investigations prior to 1940. Experimental
and theoretical transfer function for constant current anemometers can be found in [79].
A modelisation was proposed by Maye [80] for the steady conduction case. Hojstrup et
al. [74], Millon et al. [75] and Fournier [77] gave an analytical expression of the transfer
function accounting for unsteady conduction with prongs. A more complete theoretical
study describing unsteady prong-stub-wire thermal interaction was presented by Tsuji et
al. [81]. Problems of thermal boundary forming on the prongs and flowing partially on the
active part of the wire may also influence the transfer function, especially at low speed
as mentioned by Paranthoen et al. [82]. An experimental determination of the transfer
function accounting for conduction was carried out by Hojstrup et al. [74] who generated
sinusoidal temperature variations with a loudspeaker. Paranthoen et al. [82] also derived
transfer functions by placing the probe in the wake of another wire heated by a sine current.
When the conduction effect is negligible (large l/d ratio), correction for the
roll-off of the wire at high frequency can be realised with an electronic compensation (RC
filter) (Paranthoen et al. [82], Weeks et al. [76]). LaRue et al. [73] applied a correction
on the signal spectrum. In a recent catalog, Dantec proposes a probe with two wires; one
measures the velocity in order to tune the compensation.
Finally, the electronic circuit driving the probe must not limit the frequency
of the whole system. For this purpose, Ji Ryong Cho et al. [83] developed a simple circuit
with a broad frequency bandwidth (13 kHz) and low signal to noise ratio.
6.2 Theory
6.2.1 Principle: constant current Wheatstone bridge
As it was said in the introduction, the sensing element is a very thin wire
usually made of tungsten or platinum mainly because the electrical resistance of this
metal changes broadly in function of temperature.
78
The resistance of the wire R0 at temperature T0 can be expressed in function
of its dimensions thanks to Eq. 6.1:
σ0−1 lw
R0 = πd2w
(6.1)
4
Rw = R0 (1 + αw (Tw − T0 )) (6.2)
The wire is placed in a Wheatstone bridge fed by a constant current intensity I0 (see
Fig. 6.1). An image of the value of the resistance R3 = Rw is given by the voltage Vout :
R3 (R2 + R4 ) − R4 (R1 + R3 )
Vout = I0 (6.3)
(R1 + R3 ) + (R2 + R4 )
The resitances R1 , R2 and R4 are not sensitive to temperature. Eq. 6.3 can be rewritten
as:
R3 R2 − R4 R1
Vout = I0 4 (6.4)
i=1 Ri
The value of the resistances R2 and R4 are large compared to those of R1 and R3 (usually
about 1000 times larger). As a consequence, the variation of the resistance R3 will not
affect significantly the denominator 4i=1 Ri which can be considered as constant. A linear
relationship links then the output voltage to R3 . Another consequence is that the current
intensity in the leg R3 is almost equal to the current at the inlet of the bridge I0 .
R2 R0 3 = R1 R4 (6.5)
The resulting output voltage is zero and Eq. 6.4 can be written: If the denominator is
assumed constant, Eq. 6.4 can be written:
The voltage output can then be amplified in order to obtain measurable values.
Eq. 6.6 shows that the voltage output will be large if the intensity is large.
Unfortunately, this intensity causes heat generation by Joule effect and the wire tempera-
ture would not result only from the gas temperature. For this reason, a compromise will
have to be found to have a negligible Joule effect but a measurable voltage output; this
will be addressed later.
79
6.2.2 Basic equations
The main heat transfer mode is the forced convection due to the air flowing
around the wire surface:
∂Tw
Qs = Vw ρw cw (6.8)
∂t
The wire is fastened onto prongs which have a large thermal inertia. The prongs will not
be able to follow high frequency fluctuations and a conduction phenomenon will take place
between the wire and the prongs; Tp is a boundary condition of the conduction equation
in the wire:
∂ 2 Tw
Qk = −kw Aw lw (6.9)
∂x2
In order to measure Vout , a current intensity I has to cross the wire; this brings
a heat source term due to the Joule effect:
In the above equations, it is assumed that the gradient in the radial direction
of the wire can be neglected because the wire diameter remains small compared its length.
∂Tw
τ + Tw = f (t) (6.12)
∂t
Vw ρw cw dw ρw cw
τ= = (6.13)
hSw 4h
f (t) = Tg (6.14)
80
In this case, τ is the time after which the response reaches (1 − e−1 ) or 63 % of the input
step level.
The transfer function of a first order system is well known and can be written
as:
1
H(ω) = (6.16)
1 + jωτ
1
|H(ω)| = (6.17)
1 + (ωτ )2
ˆ = atan(−ωτ )
H(ω) (6.18)
where |H(ω)| and H(ω)ˆ are respectively the gain and the phase of the transfer function
H(ω) and w = 2πf is the pulsation.
Another characteristic value for a√first order equation is the cut-off frequency
defined as the frequency for which |H(ω)| = 22 = 0.707. A straight forward relationship
exists between the cut-off frequency and the time constant, namely:
1
fcut = (6.19)
2πτ
In the present range of velocity, the free convection term does not apply
(Grashof). The Knudsen number K characterises the molecular effects which can hap-
pen in rarefied gases or with very small wire diameters (changes of boundary conditions
with respect to continuum flow) but is not predominant in the present application. As the
dependence with Prandtl number is small with the air temperature, a simplified relation
comes out:
−0.17
Tf
Nu = A + BRen (6.21)
Tg
where all the values should be evaluated at the film temperature and the reference length
is the diameter of the wire. A further simplification can be done in this application
81
T
because the wire is not heated; no correction is needed for the temperature loading Tfg .
An extensive series of measurements of the heat transfer coefficient around heated wires
led to the determination of the constants A, B and n. Other authors propose relations
of the same kind (Kings, Krammers, Andrews, Hilpert...) and each one comes up with
different values for A,B and n presented in table 6.1.
d2w ρw cw
τ= (6.23)
4ka N u
or developed
d2w ρ2w cw
τ=
(6.24)
ρg vg dw n
4kg A + B µg
Several trend can be extracted from this equation. The frequency response
will increase (the time constant will decrease) if:
-the gas velocity increases
-the gas density increases
-the gas viscosity decreases (smaller insulating boundary layer)
-the diameter decreases (the ratio exchange surface/mass to heat decreases)
To illustrate the influence of gas velocity and wire diameter on the wire fre-
quency response, a numerical application was performed with the correlation of Collis and
Williams (Eq. 6.22). In Reid [85] expressions for the conductivity and the viscosity of the
air in function of temperature are given at 1 and 10 bars. The difference between 1 and
10 bar is very small and in the next, one will use the laws at 1 bar namely:
82
where t is in Celsius and ka in [W/(cm.K)]
µa Ta 0.75
= (6.26)
µa0 Ta0
with µa0 = 1.85E − 05 Pl at Ta0 = 293K
The gas is air at Ta = 340K, the metal for the wire is tungsten. Physical characteristics
of tungsten (ρw , cw , αw ) are extracted from [86] and reported in table 6.2
Fig. 6.3 shows values of cut-off frequency derived from for several wire di-
ameters in function of gas velocity. As expected, small diameter wires allow to reach
high frequency responses but practical limitations are imposed in the choice by strength
as well as manufacturing considerations. This kind of graph is useful in order to choose
the diameter of the wire depending on the gas characteristics and the wanted frequency
response.
83
The only term depending on velocity in this equation is h. With f (v) = hπ 2 d3w Tg +
4σ0−1 I 2 (1 − T0 αw ) and g(v) = hπ 2 d3w − 4σ0−1 I 2 αw , the sensitivity to velocity can be com-
puted from:
∂f ∂g
∂Tw ∂v g − f ∂v
= (6.30)
∂v g2
where
∂f ∂h
= π 2 d3w Tg (6.31)
∂v ∂v
∂g ∂h
= π 2 d3w
∂v ∂v
h can be expressed in function of v by mean of a correlation like Eq. 6.21:
kg (A + BRen )
h= (6.32)
dw
n
ρg dw v kg
h= A+B
µg dw
The first graph shows the wire temperature in function of the current inten-
sity for several velocities and a wire diameter of 2.5µm. Of course, at a given velocity,
increasing the current intensity results in increasing the wire temperature. At a given
current intensity, an increase of velocity improves the convection around the wire whereas
the Joule effect remains the same; as a result, the wire is colder.
The corresponding sensitivity is plotted in the graph below. If Iw remains
below 1mA, the error induced by the Joule effect is less than 0.1 K at an air velocity
of 50 m/s and less than 0.05 K at 200 m/s. Oubviously, the current intensity must re-
main small and high amplification factors will have to be employed in the electronic board.
Unfortunately, the technique which will be used later to determine the fre-
quency response of the wire is based on an electrical heating of the wire using a sine
current (this technique will be described later in details). The current must be high
enough to heat the wire and its influence on the time constant determination has to be
84
estimated. The unsteady convection equation including Joule effect but neglecting con-
duction is Qs = Qh + Qj which results in another first order non homogeneous linear
equation:
∂Tw
Vw ρw cw = −hSw (Tw − Tg ) + Rw Iw2 (6.34)
∂t
which can be rewritten under the same form than Eq. 6.12:
∂Tw
τ + Tw = f (t) (6.35)
∂t
ρw cw Aw
τ= (6.36)
σ−1 α I 2
πhdw − 0 Aww w
σ0−1 Iw
2 (1+α T )
w 0
πN ukg Tg + Aw
f (t) = (6.37)
σ−1 α I 2
πN ukg − 0 Aww w
The influence of the current intensity on the cut-off frequency can be estimated
with Eq. 6.36 and is presented in Fig. 6.5 for a 2.5µm wire. Increasing the current intensity
reduces τ but calculations show that intensities up to 10 mA can be used without reducing
significantly the cut-off frequency.
∂ 2 Tw
= λ2 (Tw − Tg ) (6.38)
∂x2
where λ2 = kw4hdw . This is a classical equation for which an analytical solution exists (see
Kay & Nederman [87]) in the case the wire temperature profile is symmetric, namely both
prongs are at the same temperature. Boundary conditions and solution are given by:
T(w,x=0) = Tp (6.39)
Q(k,x= lw ) = 0 (6.40)
2
85
dθ l
Q(x=0) = −kw Aw = λkw θp tanh(λ ) (6.42)
dx x=0 2
In Fig. 6.6 profiles are shown for several l/d ratios for a 5µm wire in a 100 m/s
flow. The graph on the right indicates the mean temperature of the wire as a function of
l/d and gives an idea of the error done on the temperature. As expected, the larger the
l/d ratio, the closer the average wire temperature approaches the air temperature. On the
other hand, a bigger probe volume is obtained as well as a structurally weaker probe. For
a given l/d ratio, increasing the velocity results in a higher convectional exchange and in
a wire temperature closer to Tg .
This profile depends also on the ratio (h/k). As h depends on velocity and k
remains constant, the shape will change with velocity. This is shown in Fig. 6.7 where the
ratio (T(w,mean) − Tp )(Ta − Tp ) is presented in function of velocity for two wire diameters.
The l/d ratios considered here are small which result in a large error due to conduction.
Very large l/d ratios have to be considered if one wants to have a negligible conduction
error but the wire becomes then very weak. If this ratios have to be kept small, one has
to understand how conduction influences the probe output in unsteady temperature field.
∂Tw ∂ 2 Tw 1
=α 2
− (Tw − Tg ) (6.43)
∂t ∂x τ
where α = kw
ρw cw W/m2 and τ = dw ρ4hw cw s.
A numerical solution for the unsteady conduction equation can be obtained
using an implicit Crank-Nicholson scheme. The convection term is not included in the
original scheme but was added here without changing the implicit way of solving the
equation nor the stability of the scheme. The discretised equation is presented in Eq. 6.44;
for clarity, the index w referring to the wire was omitted.
Tin+1 −Tin
∆t
=
n+1
[η(Ti+1 −2Tin+1 +Ti−1
n+1 n −2T n +T n )
)+(1−η)(Ti+1 i i−1 ]
α ∆x2
− τ1 (ηTin+1 + (1 − η)Tin ) − Tan (6.44)
η is a parameter defining the contribution of terms at time (n+1) and at time n in the
different terms, i is the index for the distance along the wire.
All terms at time (n+1) can be brought on the left hand side; all the terms
on the right hand side are then known. If this equation is written for all nodes along
the wire, a tridiagonal matrix comes out which can be efficiently solved with algorithms
found in scientific numerical libraries. The scheme is unconditionally stable provided η is
larger than 12 namely the time step and the space step are decoupled. At each time step,
the temperature profile on the wire is solved. An average temperature along the length
of the wire is calculated and provides the output signal of the probe in function of time.
86
As a boundary condition, the prong temperature is imposed; in the following, the prong
behaviour was assumed to be similar to a first order equation (with a time constant much
below the wire one).
For example, Fig. 6.8 shows the response of the prongs and of a probe mounted
with a 2.5µm wire to a sine wave in gas temperature. For f=100 Hz, the prongs reproduce
only a small part of the fluctuation due to their big size. The wire response is much better
but is affected by the conduction; no phase shift is observed. For f=1000 Hz, the prongs
remain at a constant temperature; the wire response is affected by both conduction and a
lag in its own frequency response resulting in a phase shift and a small output amplitude.
This calculation can be repeated for a number of different frequencies; the space discreti-
sation can be kept identical for all frequencies whereas the sampling frequency is varied
because time step and space step are decoupled. The ratio (input amplitude / probe re-
sponse amplitude) provides the gain of the transfer function, and the time lag between the
sine waves is converted in phase. As a result, the probe transfer function is obtained as
shown in Fig. 6.9 for several gas velocities and a prong time constant of 2.s. The gain is
presented in dB (g = 20log10 Aout /Ain ). At very low frequency, both prongs and wire can
follow the fluctuations (g=1). Then, the prongs cannot follow them and the wire begins to
be affected by conduction. As the time response of the wire is much higher than the one
of the prongs, there is a plateau region in which the prongs are not responding. For high
frequencies, the wire itself cannot follow the fluctuations and the gain goes down quickly.
With increasing velocities, convection raises the average temperature of the wire (while
the conduction coefficient remains constant) and the gain of the plateau level. The cut-off
frequency also increases as one can expect. One will notice that the prongs time constant
was chosen to be invariant with the speed which is not the case in reality. The influence
of the prong time constant is presented in Fig. 6.10. A high time constant enlarges the
plateau level. Notice that if the probe is used only in the frequency domain where the
plateau is located, the correction for conduction can be achieved by multiplying the AC
component of the signal by a constant factor which is the inverse of the plateau level.
Similar results can be obtained with a much simpler numerical system based
on a simple discretisation of a first order equation.
6.2.8 Discrete first order system to simulate continuous first order sys-
tem
Let us consider the following first order equation
∂Tw
τ + Tw = Tg (6.45)
∂t
It can discretised as:
Twn+1 − Twn 1
= − (Twn − Tgn ) (6.46)
∆t τ
where n is such that t = n ∗ ∆t = n/fs . Terms at time n can be put on the left hand side
and give:
87
∆t
b0 = (6.48)
τ
∆t
a1 = −1 (6.49)
τ
The coefficients are linked by b0 − a1 = 1 and the transfer function of the
discrete system is (see Oppenheim [88] or Rabiner [89]):
b0
H(ω) = (6.50)
1 + a1 e−jω
b0
|H(ω)| = (6.51)
1 + 2a1 cosω + a21
ˆ = atan a1 sinω
H(ω) (6.52)
1 + a1 cosω
Twn+1 − Twn 1
= (T n − Tgn ) (6.53)
∆t τw w
Tpn+1 − Tpn 1
= (Tpn − Tgn ) (6.54)
∆t τp
Let us now consider the probe output T like a linear combination of both response:
where (1-g) is the ratio of heat transferred by conduction (plateau level) to the prong or
in other words, g corresponds to the ratio (Tw,mean − Tp )/(Tg − Tp ) plotted in the right
graph of Fig. 6.6 (steady conduction). One advantage with respect to the Crank-Nicholson
solution is that an analytical expression of the probe transfer function can be derived. It
is obtained by summing in the complex domain the transfer functions of the 2 first orders
(see Eq. 6.50,the Laplace transform is linear) weighted by their respective contributions
88
(g and (1-g)). An example is shown in Fig. 6.12. The frequency response of the wire with-
out conduction effect is plotted as a dotted line; the combination of the two first order
equations is plotted as a continuous line and matches exactly what was found using the
Crank-Nicholson scheme. At low frequency, both prongs and wire can follow the tempera-
ture fluctuations; then the prongs begin to lag behind until a plateau is reached where only
the wire responds; finally the wire also lags behind at high frequencies. If the transfer func-
tion of the prongs cannot be simulated with a single first order, this numerical analysis can
be extended to N first order systems. This allows a great freedom in the description of the
transfer function shape, unlike the simple analytical expression of Paranthoen et al. [82]
or the complex expression obtained by Tsuji et al. [81] which is limited to three first orders.
Another advantage is that the system constituted by Eq. 6.53, 6.54, 6.55 is
reversible. In other words, it can be used to simulate the probe response with the gas
temperature Tg as an input but it can also be reversed and provide the gas temperature
with the probe signal as an input. With this last configuration, a compensation system
correcting for prong-wire conduction and wire roll-off at high frequency is realised. Unlike
the electronic compensation (RC filter) described by Weeks et al. [76], the numerical
system allows to describe complicated transfer functions with an easy tuning through an
adequate choice of the various time constants. A simple Fortran program was written
including both sense of resolution (direct or reversed) for an arbitrary number of first
orders.
6.3 Experiments
6.3.1 Wire, probe, board, calibration
Different kind of wires can be purchased from Dantec, TSI or directly from
the manufacturer through Sigmund Cohn Corp. (USA). Three types of wires are mainly
used: platinum wires, tungsten wires, wollaston processed platinum wires. The wollaston
processed wire consists in a core of platinum embedded in a coat which can be removed by
chemical attack. This process is usually used for diameters below 1µm (down to 0.25µm!)
because there exists no other process. The main drawback of those wires is their mechani-
cal weakness. Wires with diameter larger than 1 µm can be drawn through dies. Platinum
wires have a good sensitivity to temperature but are weak. Tungsten have a lower sen-
sitivity to temperature but are much stronger. Usually, tungsten wires are plated with
platinum in order to protect them from corrosion. In the following, due to a necessary
mechanical strenght, only platinum plated wires are used with diameters of 5 or 2.5 µm.
For the same reason, the l/d ratio will be kept below 300 (l of the order of 0.6 mm). In
the litterature, l/d ratios up to 1000 are encountered but the flow velocity is then of the
order of 1 m/s; here the design flow velocity is 200 m/s.
The prongs are normal steel needles (sewing needles) that were burned in or-
der to be able to bend them (it also allows an easier soldering). The base diameter of the
needles is 0.4 mm, their height is of the order of 3 mm. They are embedded in a ceramic
insulating cylinder. Fastening the wire on the probe is realised or by electrowelding or by
tin solder.
89
The electronic board was designed after the recommendations of Ji Ryong Cho
et al. [83] which ensures a high frequency response of the board (13 kHz). A diagram is
presented in Fig. 6.14.
The effect of the current intensity on the wire temperature was checked for
both diameter 5 and 2.5µm. First, the wire sensitivity to temperature was determined by
placing the wire in a heated jet monitoring Rw in function of Tg . This resulted in a value of
αw = 0.00354K −1 which is close to the value given in the Hanbook [86](αw = 0.00318K −1 ).
Two methods can then be used to measure Rw in function of I. The voltage drop through
the wire in function of I can be measured and Rw = UIw is then easy to derive. The
disadvantage of this method is linked to the fact that when Iw goes to zero, Uw goes
also to zero and dividing Uw by Iw results in a large uncertainty. Another method is to
measure the variable resistance R4 needed to equilibrate the bridge for several values of I0
and then calculate Rw = RR1 R2 4 . Measurements with both methods are presented for the
5µm wire in Fig. 6.13. Only the first method was used for the 2.5µm. Those experiments
being done in still air, only natural convection cools the wire and the Joule effect (Rw I 2 )
remains predominant as shown by the shape of the curve. It was concluded that the cur-
rent intensity will be kept below 2 mA.
Static calibration of the cold wire probe was performed in a heated jet. As ex-
pected, a linear relation monitors the output voltage to the jet temperature (see Fig. 6.15.a).
A slight dependence of the slope on the velocity was observed. At 100 m/s, a slope of
21.43 K/V is obtained; slopes at 200 m/s and 150 m/s were respectively 4.7% and 2.2%
higher (in all cases, probe incidence with respect to the flow was 0o ). This effect can be
v2
attributed to the problem of recovery of the dynamic temperature (T0 = Ts + 2c p
) at high
velocity similar as for thermocouples. For the same reasons, the incidence of the prongs
with respect to the flow also has an influence.
Slopes at 90o and 45o incidence are respectively 7.2% and 4.9% higher than
the slope at 0o for a velocity of 100 m/s (see Fig.6.15.b).
90
remains below 5 mA, all data points fall onto the same line. The gain corresponding to
a first order transfer function (see Eq. 6.17) is plotted as a continuous line. The shape
matches well the experimental curve but the time constant τ used for this curve did
not correspond to the one predicted by Eq. 6.23 and Eq. 6.19. This problem will be
addressed further when discussing Fig. 6.17. Larger AC current led to a non negligible
residual output at high frequency due to the electrical response. In all following tests, a
3 mA AC current intensity was superimposed to a 10 mA DC current intensity. It was
shown in the theoretical part that this level of intensity does not affect significantly the
wire dynamic response. Notice that this method only provides information on the wire
response; conduction effects exist but the amount of heat transferred to the prongs is so
small than their temperature is not affected and remain constant (ambient temperature)
whatever is the frequency of the sine current.
However, the experimental data show the same trends as predicted by the
theory: higher velocity improves convection and increases the frequency response; a smaller
diameter reduces the thermal inertia of the wire and increases the frequency response.
A 2kHz cut-off frequency is achieved at 200 m/s with the 2.5µm wire at atmospheric
pressure. The value of the active length does not influence the cut-off frequency of the
wire as expected. The cut-off frequency level can change slightly from one wire to another
because of wire contamination and possibly small changes in the wire diameter. Smaller
diameters were not tested because platinum plated tungsten wires do not exist below
2.5µm. The influence of density on cut-off frequency was investigated by performing tests
with a 2.5µm wire in a variable density chamber. The density change was achieved by
setting different levels of static pressure in the measurement chamber; the results are
summarized in Fig. 6.18. Increasing the density improves the convection around the wire
and therefore the cut-off frequency. In order to take into account the influence of velocity
and density on the cut-off frequency and to establish a correlation for this wire, the Nusselt
number (computed by rearranging Eq. 6.23) was plotted against Re0.45 causing all points
to collapse onto a line as shown if Fig. 6.19. A linear regression allows to extract the
coefficients for the correlation:
91
for 4.0 ≤ Re ≤ 36.0. This correlation provides a straight forward way to compute the
cut-off frequency of the wire needed in the numerical compensation system.
92
plateau level because the contribution of convection rises in the overall exchange. A linear
regression on the plateau level in function of Re0.45 gave:
A fit with four first order functions was performed to simulate the prong behaviour. The
percentage of contribution of each first order to (1-g) was kept constant while the four
corresponding cut-off frequencies had a tendency to increase with flow velocity. In order
to characterise in a unique way the evolution of the 4 cut-off frequencies in function of
velocity, a fit was done in order to obtain a law of the kind
(780 ≤ Re ≤ 27800) and 4 equivalent diameters computed with Eq. 6.23. The contribu-
tions as well as the equivalent diameters are reported in table 6.4.
In this way, for a given flow velocity, the cut-off frequency corresponding to
the wire is computed from Eq. 6.56 and Eq. 6.23. The 4 cut-off frequencies for the prongs
can be computed identically using the diameters indicated in table 6.4 and Eq. 6.58.
The respectives contributions are found from Eq. 6.57 and table 6.4. Notice that those
correlations are only valid for this probe.
The effect of different probe head inclinations with respect to the flow was
also considered. Tests were run at 0o (probe is aligned with air stream) 45o and 90o . For
a velocity of 50 m/s, the prong response is slightly faster at 0o . With increasing velocity,
this difference becomes smaller and smaller; at 200 m/s, responses are almost identical.
As it can be observed from Fig. 6.23, the present domain of use is quite
different from what is encountered in the litterature. The probes were design to perform
measurements in a transonic turbine stage with 43 cooled guide vanes and 65 rotor blades
running at 6500 RPM. The probe would be fixed on a rotor blade in order to resolve the
thermal wakes which are coming out of the vanes at a blade passing frequency of 4.6kHz
and a velocity of 200m/s.
93
of approximately 1.2 mm on either side of the jet. The temperatures outside and inside
the jet are controlled by thermocouples. As for the step tests, the temperature difference
between the ambient air and the heated jet was typically 50 K. The electronic board as
well as an opto-electronic data transmission system (Sieverding et al. [90]) is installed in
the hollow shaft of the rotating wheel. The power supply for the in-shaft electronics is
provided through mechanical slip rings. The noise on the wire output signal in terms of
temperature (including CW board, transmitting and receiving board) did not exceed 0.8K
(RMS value) in still air. Spun at 3000 RPM at atmospheric pressure, the noise rose up to
1.05K (RMS value).
The tests were run at atmospheric pressure. The probe used for these tests was
of type C in Fig. 6.21. For all tests, the probe was aligned within ±10o with the relative
flow direction resulting from the jet velocity v in the axial direction and the peripheral
w2
speed u of the probe. The probe measures a total relative temperature: T0r = Ts + 2c
√ p
94
to 200 m/s but in any case the cut-off frequency is at least 2 orders of magnitude higher
than the jet passing frequency. Due to low peripheral speed, the incidence remains close
to 0o across the entire shear layer. For the high frequency case, the velocity changes from
140 m/s to 200 m/s which corresponds to a 10% change in the probe cut-off frequency. The
incidence changes from 45o to 0o which corresponds to a 5% variation in the slope of the
calibration curve (see Fig. 6.15.b). The maximum error on the compensated temperature
signal inside the shear layer was estimated to be 5%, accounting for the two phenomena.
6.3.6 Accuracy
As the compensation amplifies the high frequency components of the signal
(including the noise), the accuracy of the compensated signal decreases with increasing
frequency. It is not possible to give a unique value of accuracy for the entire frequency
domain. Therefore, the transfer function was decomposed in four zones of different accura-
cies. The first zone is the domain where both prongs and wire can follow the temperatures
fluctuations. Here the accuracy is of the same order as for the thermocouple used for
the calibration. This uncertainty is estimated to ±0.5K. The second zone is the domain
where the wire responds well but where the prong response lags behind. The uncertainty
is assumed to be a linear transition between the first and the third zone. The third zone
is the plateau level. The dispersion on the plateau level was determined through repeat
step tests, leading to an uncertainty of ±1.75K (20:1). In the fourth zone, the wire does
not follow anymore the temperature variations. Repeat tests with the electrical heating
method led to an uncertainty of ±1.85K (20:1).
Using the compensation amplifies the error especially in the fourth zone. This
is illustrated in Fig. 6.27 where the error on the gain of the transfer function is plotted as
well as the resulting error for the compensation (notice that the scale for the gain is not
in dB but linear).
6.4 Conclusions
95
6.5 Figures
Constant
Current
Source
I0
R2 R1
Vout
R3=Rw
R4
Tungsten wire
coated with
platinum
96
5
3
Nu
0
0 10 20 30 40
Re
6000
1.0 µm
5000
1.5 µm
Cut-off frequency (Hz) (G=- 3dB)
2.5 µm
5.0 µm
4000 7.0 µm
9.0 µm
3000
2000
1000
0
0 50 100 150 200 250
Flow velocity (m/s)
Figure 6.3: Calculation of cut-off frequency in function of velocity for several diameters
97
Wire temperature [K]
342
v= 50 m/s
v=100 m/s
341 v=200 m/s
340
0 1 2 3 4 5
Sensitivity to velocity [K/(m/s)]
0.000
-0.005
v= 50 m/s
-0.010 v=100 m/s
v=200 m/s
-0.015
-0.020
0 1 2 3 4 5
Intensity (mA)
4000
I0 = 0 mA
I0 = 10 mA
3000 I0 = 20 mA
I0 = 30 mA
Cut-off frequency
2000
1000
0
0 50 100 150 200 250
V (m/s)
98
mean (Tw-Tp)/(Ta-Tp)
1.0
0.8
(Tw-Tp)/(Ta-Tp)
0.6
0.4
l/d=100
l/d=200
l/d=400
0.2
l/d=800
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0 500 1000
x/l l/d
0.7
(T(w,mean)-Tp)/(Ta-Tp)
0.6
d = 5 µm l/d=120
d=2.5 µm l/d=240
0.5
0.4
0 50 100 150 200 250
Velocity [m/s]
99
f=100 Hz f=1000 Hz
360 360
Gas
Probe
350 350
Prongs
340 340
Temperature [K]
330 330
320 320
310 310
300 300
0 5 10 15 0.5 1.0 1.5 2.0
Time [ms] Time [ms]
(Prong time constant τ = 2s, correlation for Nu: Collis & Williams)
0
-5
Gain (dB)
50 m/s
100 m/s
150 m/s
200 m/s
-10
-15
0.01 0.1 1 10 100 1000 10000
Frequency (Hz)
100
τ is indicated in the legend (v=100 m/s)
0
-5
Gain (dB)
20.0 s
2.0 s
0.2 s
-10
-15
0.01 0.1 1 10 100 1000 10000
Frequency (Hz)
-20
Gain [dB]
-60
0.01 0.1 1 10 100
0.0
-0.5
Phase [Rad]
-1.0
-π/2
-1.5
-2.0
0.01 0.1 1 10 100
f/fs
101
0
-5
Gain(dB)
-15
0.01 0.1 1 10 100 1000 10000
Frequency (Hz)
2.5 µm diameter
(αω=0.00354 (1/Κ)) 5 µm diameter wire
100 100
From Rw=Uw/Iw
From Rw=(R1 R4)/R2
80 80
Temperature [C]
60 60
40 40
20 20
0 5 10 15 20 0 10 20 30 40 50
I0 [mA] I0 [mA]
102
Figure 6.14: Electronic board
103
70 Velocity Slope(K/V)
Temperature (C)
100 m/s 21.43
60 150 m/s 21.91
200 m/s 22.43
50
40
30 Inclination 0 deg.
-a-
70 Inclination Slope(K/V)
Temperature (C)
0 deg. 21.43
60 45 deg. 22.47
90 deg. 22.97
50
40
30 Velocity 100 m/s
-b-
-3 dB level
-5 Amplitude of AC component
0.25 mA
Gain (dB)
0.50 mA
1.50 mA
3.00 mA
first order
-10
DC component 10 mA
Air velocity 25 m/s
Wire diameter 2.5 µm
Active length 0.6 mm
-15
10 100 1000 10000
Frequency (Hz)
Figure 6.16: Gain obtained by electrical heating for several current amplitudes
104
2500
1500
Active length:
0.4 mm
1.5 mm
1000 0.6 mm
y 5 µm
Theor
500
D ifferent wires
0
0 50 100 150 200
Axial velocity (m/s)
ps=2.00 bar
2000
1500
1000
0 50 100 150 200
Velocity (m/s)
Figure 6.18: Cut-off frequency in function in function of air velocity for several static
pressure
105
ps=1.00 bar
ps=1.25 bar
2.0 ps=1.75 bar
ps=2.00 bar
Nu
1.5
Nu=0.458+0.345*Re**0.45
1.0
2.0 3.0 4.0 5.0
Re**(0.45)
0
1.0
Reconstructed step
Numerical response
Gain (dB)
0.6
-4
0.4
106
0
200 m/s
-5 0.6
Stubs
0.6 0.4
Gain(dB)
0.4
3.0
3.0
2.5
-10
A B C
0.4
All dimensions are in mm
-15
0.01 0.1 1 10 100 1000 10000
Frequency (Hz)
0.8
(T-Tcold)/(Thot-Tcold)
-4
0.6
Gain (dB)
-6
0.4 -8 50 m/s
50 m/s
100 m/s 100 m/s
200 m/s -10 200 m/s
0.2
-12
0.0
-14
0 5 10 15 0.1 10 1000
Time (s) Frequency (Hz)
107
12000
d=0.25 µm Weeks-Beck-Joshi
l/d=1600 LaRue-Deaton-Gibson
10000 Disa manual
Paranthoen
Fournier
Cut-off frequency [Hz]
8000 Hojstrup
Denos
Denos
6000 d=0.625 µm
l/d=1000
d=0.5 µm
4000
d=1µm
l/d=400
2000 d=0.7µm d=1 µm d=2.5µm
l/d=1700 l/d=400 l/d=240
d=4 µm
l/d=300 d=5µm
l/d=240
0 d=6.3 µm
l/d=200
Data
Balancing probe Acquisition
Infrared receiver
System
380 mm
Trigger
Wheel
In shaft Air
Thot Tcold electronic Turbine
Infrared emitter
Nozzle
Heater
108
0
1.0
(T-Tcold)/(Thot-Tcold)
0.8
Gain (dB)
-5
0.6 3.2 Hz
0.0
-15
0.1 10 1000 0 100 200 300
Frequency (Hz) Time (ms)
1.0 1.0
(T-Tcold)/(Thot-Tcold)
(T-Tcold)/(Thot-Tcold)
0.8 Jet temperature 0.8
profile 980 Hz
0.6 Recorded 0.6
signal
0.4 0.4
0.0 0.0
2 3 4 5 6 7 0.8 1.2 1.6 2.0
Time (ms) Time(ms)
1.0 1.0
(T-Tcold)/(Thot-Tcold)
(T-Tcold)/(Thot-Tcold)
0.8 0.8
0.4 0.4
0.2 0.2
0.0 0.0
0.05 0.15 0.25 0.35 0.05 0.15 0.25
Time (ms) Time (ms)
109
15
1.0
W=200 m/s
(T-Tcold)/(Thot-Tcold)
10
Compensator transfer function 0.8
Gain (dB)
5 3.2 Hz
0.6
0
0.4
-5
Probe transfer function 0.2
-10
0.0
-15
0.1 10 1000 0 100 200 300
Frequency (Hz) Time (ms)
(T-Tcold)/(Thot-Tcold)
profile
0.8 0.8
Compensated 980 Hz
0.6 signal 0.6
Recorded
0.4 signal 0.4
0.0 0.0
2 3 4 5 6 0.8 1.2 1.6 2.0
Time (ms) Time(ms)
1.0 1.0
(T-Tcold)/(Thot-Tcold)
(T-Tcold)/(Thot-Tcold)
0.8 0.8
4600 Hz 6000 Hz
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
0.05 0.15 0.25 0.35 0.05 0.15 0.25
Time (ms) Time (ms)
110
2.0
1.5
Gain (dB)
1.0
0.5
0.0
0.01 0.1 1.0 10 100 1000 10000
Frequency (Hz)
111
112
Chapter 7
7.1 Introduction
As it was mentioned earlier, levels of turbine inlet temperature are nowadays
very high in order to obtain a high thermal efficiency of the engine thermodynamic cycle.
For this reason, the first or the two first turbine stages are cooled (both vane and rotor
blades) and the cooling efficiency is a crucial point for the engine operation.
The two main parameters to control in order to ensure an optimum blade life
duration are:
- a moderated mean temperature level of the blade with respect to the fusion temperature
of the metal
- limited thermal gradients in order to avoid thermal fatigue and creep.
In this perspective, the designer has to be able to predict the levels of temper-
ature at the inner and outer surface of the blades as well as throughout the walls. This
information can be obtained by solving the conduction equation with an iterative proce-
dure where the two boundary conditions, inner and outer surface temperature, are updated
after each iteration. The equilibrium is reached when the heat fluxes due to internal and
external convection and the heat flux due to conduction across the walls are balanced. The
estimation of convectional heat fluxes at the wall is provided by experimental correlations
of the convectional heat transfer coefficient h [W/m2 ]. Indeed, if the gas temperature and
the wall temperature are known (or assumed for the first guess of the iterative procedure)
and if the heat transfer coefficient can be determined from correlations, the convectional
heat flux can be determined with Eq. 7.1.
For similarity purposes, the heat transfer coefficient is often reported in terms of Nusselt
number:
q̇ c hc
Nu = = (7.2)
kg (Tg − Tw ) kg
113
or Stanton number
h Nu
St = = (7.3)
ρcp u ReP r
Measurements on internal cooling channels provide the knowledge for the in-
ner convectional heat transfer coefficient. Measurements around the external blade profile
provide the outer convectional coefficient and are the purpose of this chapter.
2) The wall temperature is derived from a very thin metallic strip laying on
the blade surface called thin film gauge . In a given temperature range, the change of
resistance of the gauge with temperature can be approximated by a linear law (Eq. 7.4)
R = R0 (1 + αT (T − T0 )) (7.4)
where R0 is the wire resistance for a reference temperature T0 and αT is the temperature
coefficient [K −1 ]. It is not the temperature which is measured directly but the resistance
of the gauge. This is realized by placing the gauge in a Wheatstone bridge fed with a
constant current which delivers a voltage output proportional to the resistance change. A
calibration allows to convert this voltage into a temperature.
3)The thin film gauges measuring technique was originally developed to de-
rive heat fluxes from experiments in short duration (typically between 1 and 50 ms in
a shock tube) high enthalpy hypersonic tunnels. They are nowadays commonly used in
short duration blowdown wind tunnels. Prior to the blowdown, the blade is under iso-
thermal state at ambient temperature. During the blowdown, the gas temperature rises
like a step and lasts only during a short time. The blade surface temperature rises by a
few degrees whereas the center of the blade remains at initial temperature because the
flux has not enough time to reach this point. The surface temperature history during the
blowdown and the knowledge of the constant temperature at the center of the blade allow
to solve the unsteady heat conduction equation. The flux at the wall is obtained as a result.
The conversion of surface temperature history into heat flux history is not
straight forward and needs to be explained with more details. Analytical solutions as well
as analogue and numerical methods will also be presented and compared. Then, several
114
types of gauges will be described and a choice will be made for the current application.
A calibration technique for the determination of the physical characteristics of the sub-
strate will be explained and tested. Finally, preliminary measurements in rotation will be
performed in order to prepare the measurements in the turbine stage.
The wall heat flux can be derived from the wall temperature history and the
initial temperature of the substrate under several assumptions:
- the film thickness is negligible compared with the substrate thickness,
- the thermal conductivity of the film is much higher than the one of the substrate,
2
- the characteristic time of the film ( e kρc ) is small compared with the characteristic time
of the experiment,
- the gauge is located in a place where the conduction in the substrate in can be treated
as a one-dimensional problem.
With those assumptions, the thin film has a negligible effect on the heat con-
duction process and does not need to be taken in account in the modelisation.
115
where p = eσ+jω is the Laplace variable , σ and ω being both reals. At the wall, Eq. 7.9
reduces to:
q̇ (x=0,p)
T (x=0,p) = √ √ (7.10)
ρck p
or for the flux in function of temperature:
√
q̇ (x=0,p) = T (x=0,p) ρck p (7.11)
If p reduces to p = ejω namely the domain is restricted to the Fourier domain
of periodic signals, Eq. 7.11 reduces to:
q̇(x=0,ω) = T(x=0,ω) ρck jω
ˆ
q̇(x=0,ω) = π/4 (7.12)
√
|q̇(x=0,ω) | = ρck ωT(x=0,ω)
Back in the time domain using the inverse Laplace transform, Eq. 7.10 is
expressed as:
t
1 q̇(x=0,t) (τ )
T(x=0,t) = √ √ dτ (7.13)
π ρck 0 t−τ
or for the wall heat flux as a function of the wall temperature:
t
ρck T (t) 1 T (t) − T (τ )
q̇(x=0,t) = √ + 3 dτ (7.14)
π t 2 0 (t − τ ) 2 x=0
T(x,t=0) = T0 = cte
q̇(x=0,t) = q0 = cte (7.15)
T(x=∞,t=t∞ ) = T0 or q̇(x=∞,t=tf ) = 0
116
Eq. 7.13 reduces then to:
2q̇(x=0,t) t
T(x=0,t) − T0 = √ (7.16)
π ρck
For this case, temperature and heat flux profiles across the substrate can be
derived analytically:
η
q̇(x,t) 2 2
=1− √ e−ζ dζ = erf c(η) (7.17)
q̇(x=0,t) π 0
T(x,t) 2 √
= e−η − π η erf c(η) (7.18)
T(x=0,t)
x
where η = √4αt and erfc is the complementary error function (erf c(η) = 1 − erf (η)). An
approximate solution for the error function can be computed from:
2
erf (η) = 1 − (a1 S + a2 S 2 + a3 S 3 )e−η (7.19)
1
with a1 = 0.3480242, a2 = −0.0958789, a3 = 0.7478556, S = 1+pη and p = 47047. Both
function are plotted in Fig. 7.3.
With those relations, the flux profile and the temperature profile inside the
media can be computed at any time as illustrated in Fig. 7.2.c. The calculations was
performed with the physical characteristics of a machinable glass ceramic substrate and a
contant wall heat flux of 5360 [W/m2 ]. This value of flux is representative of regions on a
blade with low heat transfer coefficient. One can see how the temperature profile inside
the substrate changes as a function of time as well as the corresponding wall temperature
and wall flux history (Fig. 7.2.a and b).
In practice, the substrate has not an infinite thickness and the experiment
has not an infinite duration. Given the duration of the experiment, Eq. 7.17 allows to
determine the error done on the flux for a given thickness of the substrate using the
semi-infinite substrate assumption. For example, the ratio q̇(x) /q̇(x=0) falls below 1% for
x
η = √4αt = 1.824 (see Fig. 7.3) which gives the relation:
√
x = 3.648 αt (7.20)
In fact, it is not the flux which changes like a step function during a blowdown
test but the gas temperature. Although the surface temperature can rise significantly
during a test, h is quite insensitive to the ratio Twall /Tgas and can be assumed constant.
As a result q̇wall = h(Tgas − Twall ) decreases slightly. There exists an analytical solution of
Eq. 7.7 for the case of a step function in the fluid temperature, a constant heat exchange
coefficient h and a non-constant flux as described in Whitaker [92] or Schneider [93]. The
117
boundary conditions are:
T(x,t=0) = T0
q̇(x=0,t) = h(Tgas − T(x=0,t) ) (7.21)
T(x→∞,t=tf ) = T0 or q(x→∞,t=tf ) = 0
The solution for the flux and the temperature inside the media are:
q̇(x,t)
= erf c(η) (7.22)
q̇(x=0,t)
T(x,t) − T0
= erf c(η) − eβ(2η+β) erf c(η + β) (7.23)
Tgas − T0
√
where β = √
h t
which gives at the wall:
ρck
T(x=0,t) − T0 2
= 1 − eβ erf c(β) (7.24)
Tgas − T0
Example of temperature and flux distribution in the substrate are presented
in Fig. 7.4 for a step in the fluid temperature of 40o C, a convection coefficient h =
134.0W/(m2 K) (similar to the one in the constant heat flux solution) and an initial wall
heat flux of 5360 [W/m2 ]. The corresponding wall temperature and wall heat flux are
presented in Fig. 7.5. This last solution is closer to a blowdown test than the constant
heat flux solution.
118
where R and C are a resistance and a capacitance per unit lenght. A voltage proportional
to the surface temperature is introduced in the circuit (Vin in Fig. 7.6) and the voltage
Vout is the image of the flux. There exists other possible combinations of resistances and
capacitances. More details can be found in Schultz and Jones [91].
119
perfect square root of time due to smoothing. Let x be the theoretical input (square root
of time), xs the smoothed input, y the theoretical output (step function), ys the output
of xs and Ein , Eout the errors on the input and output signal defined as:
xs(t) − x(t)
Ein = (7.29)
x(t = tf )
ys(t) − y(t)
Eout = (7.30)
y(t = tf )
When the error on the input falls below 0.1%, the error on the output falls below 0.8%.
Errors are above those levels for only 12 points (the input signal has 1024 points) in the
region of the transient. After this region, the error on input and output continue decreases
as time increases. At t=0.120 s, the error on output is below 0.01%.
It can be concluded that the Cook and Felderman integration gives a reason-
able accuracy on the solution. One drawback is that this algorithm is quite time consuming
for large time series because to compute the solution at the n-th point, n-1 terms need
to be summed-up. An interesting alternative could be the use of the discrete Fourier
transform.
Fourier transform
The Fourier transform can be used to convert surface temperature history in
flux history (Doorly [94]). A FFT is performed on the time signal T(t) resulting in an
equivalent series in the frequency domain T (ω). A complex product is performed with
the complex transfer function H(jω) giving the flux solution in the complex domain. An
inverse FFT provides then the flux in the time domain.
T (t) → F F T → T (ω)
120
In the case of non-periodic signals, the FFT technique is quite heavy because
it requires a large extension of the signal with zero and the advantage of this fast al-
gorithm is balanced by the time lost to compute the zero padd extension. However, in
the case of periodic fluctuations like a long series of blade passing events, the problem
of the transient is of less amplitude and is confined to the extremities of the signal. The
advantage of the FFT technique is its ability to fit any transfer function. Unlike the Cook
and Felderman approach, it can perform the conversion (surface temperature -¿ heat flux)
for multi-layered gauges. An alternative to the FFT technique is to solve the unsteady
conduction equation numerically with the surface temperature as a boundary condition.
The temperature profile inside the medium is obtained and the flux at the wall is derived
from the derivative of the temperature profile at the wall. This can be achieved with the
Crank-Nicholson scheme.
Crank-Nicholson scheme
The temperature profile inside the medium can be computed at each time step
by solving Eq. 7.7 with an implicit Crank-Nicholson scheme (Eq.7.31).
Tin+1 −Tin
∆t
=
n+1
[η(Ti+1 −2Tin+1 +Ti−1
n+1 n −2T n +T n )
)+(1−η)(Ti+1 i i−1 ]
α ∆x2
(7.31)
η is a parameter defining the contribution of terms at time (n+1) and at time n in the
different terms, i is the index for the depth in the substrate.
All terms at time (n+1) can be brought on the left hand side; all the terms on
the right hand side are then known. If this equation is written for all nodes used to dis-
cretise the substrate, a tridiagonal matrix comes out which can efficiently be solved with
algorithms found in scientific numerical libraries. The scheme is unconditionally stable
provided η is larger than 12 i.e. the time step and the space step are decoupled.
121
the wall. The scheme was applied on the same test case as the previous algorithm using
1000 points to digitise the medium and a non-constant spacing (see Fig. 7.10). When the
error on the input falls below 0.1%, the error on output falls below 0.4%. Only 12 points
in the step region area have larger error levels. At t=0.120 s, the error on the output is
still 0.02% and does not decrease below 0.01% afterwards.
The Crank-Nicholson scheme is much faster than the Cook and Felderman so-
lution for large time series because a constant number of operation is involved whereas the
Cook and Felderman solution requires an increasing number of operations with increasing
number of points in the input time serie. Non-periodic signals can be handled without the
problems of transient response at the end of the signal experienced with the FFT. As the
FFT technique, it could be extended to several media for the data reduction of gauges on
polyimide substrate or enamel coating. Another advantage of this scheme is that the flux
can be introduced as a boundary condition and the surface temperature rise is obtained
as a result.
The three techniques presented above can replace the analogue circuits with
a reasonable accuracy provided oversampling is performed on the input signal followed by
a numerical smoothing. An analogue or a numerical filter would produce similar results.
Usual materials are nickel and platinum. They both have a good sensitivity to
temperature and a large resistivity (see table 7.3) so that measurable resistance changes
can be achieved with usual gauge dimensions. Due to their poor resistivity, copper or
gold are used to make connecting pathes from the ends of the platinum gauge towards
the substrate extremities. Indeed, for a given temperature change, a negligible change of
resistance will be observed with gold and copper compared to nickel or platinum. The
pathes are also very thin in order to preserve the substrate surface quality.
122
placed in a Wheatstone bridge), the substrate must be an electrical insulator.
-If it is also a good thermal insulator (low thermal diffusivity) the heat will enter the
substrate slower and a larger surface temperature rise (easier to measure) will result.
Moreover, the semi-infinite assumption can be held for a larger duration.
Because of their high thermal diffusivity, metals are not used as substrates;
moreover, they are not electrical insulators. Fortunately, thermal insulators are often
electrical insulators. The four first materials presented in table 7.4 fit those criteria and
are usually used. Macor is a trademark of machinable glass ceramic, Kapton and Mylar
are trademarks from Dupont de Nemours for polyimide sheets. Nickel bounds well onto
polyimide substrates, platinum bounds well onto Macor or quartz substrates.
It is clear that, although the material used will be different from a real metallic
blade, the simulation of convection around a body does not depend on the material.
However, the surface roughness should be identical in both cases.
Platinum gauges on Macor substrate have been used for a long time at VKI
and their manufacturing is well documented (Ligrani [96]). The platinum is under a liquid
from and is hand-painted with a pen onto the substrate. The substrate is then fired in
an oven at a temperature of the order of 700o C. If the resistance of the platinum strip
is too high, another layer of platinum is painted on top and the substrate is fired again.
This process is repeated until a value of resistance between 50 and 100 Ω is achieved
for dimensions of the order of 1 mm times 0.5 mm. Gold leads are painted onto the
substrate surface with a liquid gold solution from both ends of the platinum strips to the
point where a connection will be made with wires. The same firing process is used and
eventually several layers are painted in order to obtain resistances of the order of 1 Ω
so that their change of resistance with temperature can be neglected. Because it is very
difficult to solder wires with tin directly onto gold, a silver pad is painted at the end of the
gold paths and fired. Unfortunately, a rotor blade made entirely out of Macor or Pyrex
would not withstand the centrifugal force in a high speed turbine rotor. Another solution
has to be found for the current application.
123
gauge or the insert is glued into the slot. This is a quite simple solution which allows the
use of the same substrates as in cascade tests. Haldeman et al. (1992) [42] instrumented
metallic blades with platinum thin film gauges deposed onto pyrex inserts. Dunn [95] uses
also very small pyrex inserts glued on metallic vanes or blades. Depending on the insert
thickness and test duration, an appropriate data reduction technique is used to derive the
heat flux from the surface temperature measurements (semi infinite model or multi layered
models).
Dunn [95] mentions that, in theory, inserting a button type gauge in a metallic
blade would introduce a discontinuity in the thermal boundary boundary layer and, as
the substrate is a thermal insulator, create a heat island . Experiments were done around
a blade in order to compare the heat fluxes obtained with successive button type gauges
and with a large insert starting from the leading edge of the blade so that there is no
discontinuity in the thermal boundary layer. The comparison was done at two different
Reynolds numbers and no significant difference was found between the two types of gauges.
However, in the case of cooled rotating blades, the blade walls are thin and
the machining of slots would weaken considerably the blade. Thin film on thin polyimide
sheet substrates are well suited for this case.
The use of thin films deposited on a polyimide sheet is recent. Only a few man-
ufacturing details are reported in the litterature. The bounding of the metallic layer onto
the polyimide substrate is done under vacuum by sputtering techniques. The polyimide
sheets are made of Kapton or Mylar (registered trademarks from Dupont de Nemours).
The surface preparation of the sheet is very important. The bounding of the metallic layer
onto the sheet as well as the absence of craks depends on it to a large extent. The thick-
ness of the gauge is not easy to control but it is more constant along the gauge than with
hand-painting technique. The instrumented sheet is bounded to the blade surface with a
very thin glue layer. The thickness of the polyimide sheet must remain small enough in
order to preserve the blade geometry.
The semi-infinite medium assumption cannot be used to compute the flux for
this type of technique. Provided the glue layer can be neglected, a double layer model
is required: one layer has finite thickness (polyimide sheet), the other layer is considered
with semi-infinite thickness (metallic substrate). Special calibration techniques are also
needed to determine the thermal product of each layer (see Epstein et al. [97]).
Another alternative to this solution is to coat the entire blade with an insu-
124
lating material such as enamel and paint the gauge on the coating.
The metallic blade is coated with insulating material and gauges are deposited
on the coating
Hilditch et al. [64] (1990) used a metallic blade covered with vitreous enamel
on which the platinum thin films were painted. The thickness of the enamel layer is main-
tained small enough so that the blade surface dimensions are not significantly affected.
The deposition process of the enamel layer on a metallic blade is complex and the man-
ufacturing process is not fully reported in the litterature. The enamel is initially under
liquid form and is fired at high temperature. A metal without carbone must be chosen in
order to avoid the migration of the carbone towards the surface during firing (high surface
roughness would result). Moreover, the dilatation of enamel and metal must be closely
matched in order to avoid craks when cooling the blade down or heating it up.
Like for the previous type of gauges, a multilayer data reduction technique is
required as well as a special calibration technique for the substrate thermal products (see
Hilditch et al. [64]).
σ0−1 lg
R0 = (7.32)
sg
where lg is the gauge length and sg = wg tg is the cross sectional area (product of width
by thickness). Taking as an example a 60 Ω platinum gauge, 1.44 mm long and 0.44 mm
large, the calculated film thickness is tg = 0.048µm which ensures a frequency response
125
which is far above what is required (values for the resistivity can be found in table 7.3).
The gauge is one of the 4 legs of a Wheatstone bridge; the working principle
of the Wheastone bridge was described previously in section 6.2.1. The bridge output
voltage is proportional to the gauge resistance change:
126
of points near the wall. Both measured surface temperature and computed flux histories
are plotted in Fig. 7.11. The flux from the analogue circuit is very similar although it has
a phase lag of 9.(1/fs). After a phase shift, the analogue signal was subtracted from the
computed flux. As a result there is a mean difference of 862.35 W/m2 which corresponds
to 0.84% of the value of the flux extrapolated at the beginning of the temperature rise
(102000 W/m2). The dispersion on the difference between the computed flux and the
flux derived from the analogue circuit is ±2.6% (20 : 1) of this value. There is a very
good agreement between the two time traces as shown in Fig. 7.12. In this graph, the
analogue trace is corrected for the phase shift and the computed flux ordinate was shifted
for lisibility purpose.
This proves the reliability of the numerical scheme in terms of mean values as
well as fluctuations in time. Moreover, in contrast with the analogue circuit, the numerical
conversion does not introduce a phase lag.
R ∗ I2
q̇(x=0) = ζ (7.35)
a
where a is the gauge area and ζ a correction factor for non-uniform heating. Non uniform
heating is mainly due to non uniform film thickness. This correction is usually far from
negligible as stated in [91].
Ideally, the test is performed under a vacuum in order to ensure that all the
heat flux goes into the substrate. In practice, the test can be performed in air because the
gauge dimensions are small and the heat flux extracted by natural convection is negligible.
127
While the current intensity can be measured accurately, this is not the case
for the gauge area and the correction factor. In order to eliminate√ those error, a second
calibration can be done in a liquid of known thermal product ( ρck )l . The heat pulse
is diffused by conduction into the substrate as well as into the liquid; the respective heat
fluxes can be expressed as:
√
ρck
q̇v(x=0,t) = √ fv (T, t) (7.36)
π
√ √
ρck ( ρck )l
q̇l(x=0,t) = √ fl (T, t) + √ fl (T, t) (7.37)
π π
(l refers to the liquid, v to vacuum). The magnitude of these heat fluxes can be estimated
with:
Rv ∗ Iv2
q̇v(x=0,t) = ζ (7.38)
a
Rl ∗ Il2
q̇l(x=0,t) = ζ (7.39)
a
√
and ρck can be computed with Eq. 7.40
√
( ρck )l
ρck = fv (T,t)Il2 Rl
(7.40)
fl (T,t)Iv2 Rv −1
The gauge area and the correction factor do not appear in this expression. If the initial
temperature level (usually ambient temperature) is identical in both experiments (Rl =
Rv ), as well as the constant current intensity, Eq. 7.40 reduces to:
√
( ρck )l
ρck = f (T,t) (7.41)
fl (T,t) − 1
v
Of course, due to the Joule effect, the film heats up, the resistance increases as
does the flux. In practice, the resistance change is small with respect to the mean resistance
of the gauge and the testing time is short. For those reason, the flux can be considered as
constant. Provided that the 1D semi-infinite substrate hypothesis is satisfied, the surface √
temperature rise will be similar to a square root in function of time f (T, t) = T (t)/ √t.
For both experiments, the surface temperature history can be fitted by a law T = aT t
and the thermal product is then simply determined by:
√
( ρck )l
ρck = aTv (7.42)
aTl − 1
Experimental determination
The gauge is connected to a specific electronic circuit which delivers a step
function in current intensity. The voltage output of the board can be related to the gauge
resistance through:
Vout
R = R0 +
I0 G
128
and to the gauge temperature through
R − R0
Tgauge = ∆R
∆T
.
The sensitivity of the board to resistance was calibrated with two different
current intensities (40 and 20 mA) leading both to a bridge sensitivity G = 19.90Ω/(V A).
The gauge was calibrated in an controlled temperature oil bath leading to a gauge sensi-
tivity of ∆R
∆T = 0.1465Ω/C (α = 0.0022361/K).
For the tests performed in air and vacuum, an attempt was done to use the
gauge area measured under microscope (1.81 mm2 ). A correction factor ζ of 1 was assumed
2
to compute the flux at the wall q̇wall = RaI ζ. The thermal product was then determined
using the result of the fit of the surface temperature history (aT ) and Eq. 7.43:
√
q̇wall π
ρck = (7.43)
aT
where q̇wall is an averaged value. In the range of intensity used for these tests, no in-
fluence of intensity on the resulting thermal product was observed. The mean values of
129
thermal product were 1633 [J/(m2 Ks0.5 )] in vacuum and 1665 [J/(m2 Ks0.5 )] in air. It
can be concluded from these values that there is no significant difference between the tests
performed in vacuum and in air. However, these values are unrealistic compared with the
two first values of table 7.1. This proves the necessity of doing the correction mentioned
in the previous section.
√
Material ρck [J/(m2 Ks0.5 )]
Macor with thin film (VKI [99], 293 K ) 2090
Macor with thin film (Calspan [100], 293 K ) 2008
Macor non instrumented (Manufacturer data, 293 K ) 1780
Macor non instrumented (Nasa [100] 298 K ) 1765
Glycerin 925
Water 1625
The correction was applied as specified in Eq. 7.42. The thermal products
of the liquids used in the calibrations are reported in table 7.1. For each intensity level,
three tests were performed: one in air, one in glycerine and one in distilled water. Two
values can be computed from these three tests: a value computed from a test in vacuum
(or air) corrected with a test in glycerine, a value computed from a test in vacuum (or air)
corrected with a test in water. The mean corrected values are reported in table 7.2.
130
7.3.6 Heat flux measurements in rotation
Modulation of the signal with an electronic board for the measurement of heat
flux fluctuations at high frequency
Because of the transfer function (flux/surface temperature), see Eq. 7.13, prob-
lems will arise in terms of the amplitude of the surface temperature change due to heat
flux fluctations at high frequency. Fig. 7.15 compares the surface temperature rise of a
Macor substrate submitted to a sinusoidal flux variation of 4600 [W/m2 ] for a frequency
of 1 Hz and a frequency of 4660 Hz (blade-passing frequency). The surface temperature
variation at 4660 Hz is almost a hundred times smaller although the flux fluctuation has
the same amplitude than at 1Hz. In order to be able to measure both the large surface
temperature change at the time scale of the blowdown test and the small surface temper-
ature fluctuations at blade passing frequency, an analogue circuit with an amplification
increasing with frequency can be used (Hilditch et al. [64]). Ideally, the amplification
would change inversely to the transfer function (surface temperature/heat flux) namely
√
A(ω) = jω.
For time and cost reasons, it was decided to implement on an existing board a
variable amplification with frequency. Unfortunately, the maximum amplification allowed
by the existing amplifier did not exceed 10 times the actual amplification. For this reason,
the amplification was done over a small frequency range according to the transfer func-
tion depicted in Fig. 7.16. The dot-dashed line represents the ”ideal” amplification. The
modification was done at the University of Limerick upon VKI request. Up to 100 Hz, the
amplification is quasi-constant and equal to 50. Above 100 Hz, the amplification increases
with frequency and reaches a maximum at blade passing frequency i.e. about 5 kHz. At
this point, the amplification is 500. Above this frequency, the gain starts to decrease in
order not to amplify the noise at high frequencies. At 50 kHz, the amplification is about
100. The shape of the transfer function could not be chosen freely due to the constraint
of the existing board but this solution is still better than working with a board with a
constant gain.
131
of the modulation function.
For the demodulation of the steady state component, a built-in Matlab func-
tion using a M-th order linear system was employed. The transfer function of the linear
system is described by two polynomials a and b of order na and nb:
nb
na
y(n) = bi x(n − i) − ai y(n − i) (7.45)
i=0 i=1
Given the demodulation transfer function as an input, the Matlab algorithm adjusts the
coefficients of the two polynomials a and b in order to obtain the best fit. A correct fit
could be obtained only up to 100 Hz. In order to avoid an incorrect demodulation above
this frequency, a numerical low-pass filter at 100 Hz was applied prior to demodulation.
The difference between the modulated and demodulated signal is very small because the
gain of the transfer function is almost constant below 100 Hz.
With these two data reduction techniques, the entire frequency domain is cov-
ered and adapted solutions are found in order to process both steady-state and fluctuating
component.
132
adapted for each rotational speed as well as the jet absolute velocity in order to have a
constant jet relative velocity. The signal was phase-locked averaged over 20 traverses and
is shown in Fig. 7.18 as well as the flux computed with the Crank-Nicholson algorithm.
As expected, the amplitude of the surface temperature decreases clearly with increasing
rotational speed. At 1540 RPM, the maximum surface temperature rise amounts to 0.1o C
only. However, the computed heat flux has a similar shape and magnitude for the three
different rotational speed. Above 2000 RPM, the shape of the heat flux enlarges and lower
amplitudes are observed. This was attributed to the fact that the stagnation point on the
blade is moving due to a change of relative inlet angle on the blade with rotational speed.
The jet does not impinge directly on the gauge located at the leading edge which explains
that the heat flux fluctuation shape is different.
7.4 Conclusions
A short theoretical study was undertaken in order to understand better the link
between surface temperature and heat flux at the wall. Analytical solutions were presented
which proved to be very useful for the estimation of the error on the flux when using the
semi-infinite substrate hypothesis. One of the analytical solutions was also used for the
determination of the thermal product of the substrate. Several numerical methods were
tested successfully in order to replace the conventional VKI analogue conversion system.
An original and efficient solution was developped using the Crank-Nicholson scheme. Its
quality was proven by comparison between the output of a standard analogue circuit and
the flux computed from the surface temperature history during a blowdown test. Finally,
the thermal product of the Macor substrate was determined experimentally. The measured
value compares very well with the results of other calibrations on the same material. An
existing electronic board was modified in order to improve the signal to noise ratio in
the high frequency range. Adapted solutions were found for the demodulation of both
steady state and fluctuating components of the board output. Measurements in rotation
were performed and demonstrated that the blade can withstand high centrifugal forces
in psite of the implementation of large inserts. Phase-locked average surface temperature
measurements across a jet were presented as well as numerically converted heat-fluxes.
This proves the feasability of heat flux measurements at high frequency.
133
7.5 Tables, Figures
Material Price(1995)
α ρR orσ0−1 (high purity metal)
−1
K ∗ 1E3 Ωm ∗ 1E8 FB
Gold 3.41 2.44 1g 5550
Nickel 5.40 11.1 250g 1950
Copper 3.90 1.70 500g 1750
Platinum 3.01 90.0 1g 6000
134
100000
10000
Magnitude
1000
100
1.0
0.8
Phase [Rad]
0.6
0.4
0.2
0.0
1 10 100 1000 10000 100000
Frequency [Hz]
24
23
22
21
20
0.0 0.2 0.4 0.6 0.8 1.0
Time [s] -a-
6000
Flux [W/m2]
4000
2000
0
0.0 0.2 0.4 0.6 0.8 1.0
Time [s] -b-
Substrate Temperature [C]
23
t=0.00 s
22
t=0.25 s
21 t=0.50 s
t=1.00 s
20
19
0 1 2 3 4
Substrate Depth [mm] -c-
Figure 7.2: Constant heat flux case: surface temperature history (a), wall heat flux history
(b), temperature profile inside the media at several successive times (c)
135
η=x/ sqrt(4 α t)
1.0
Qx / Qx=0 = erfc(η)
2
-η
Tx / Tx=0 = e - sqrt(π) η erfc(η)
0.8
0.6
0.4
0.2
0.0
0.0 0.5 1.0 1.5 2.0 2.5
η
Figure 7.3: Constant heat flux case: Tx /Tx=0 and q̇x /q̇x=0 as a function of η
Flux in substrate with q0 initial =5357 W/m2 from erfc α=0.88E-6 W/m2
26
0.01 s
Temperature [C]
0.50 s
24
1.00 s
2.00 s
22
20
0 1 2 3 4
Substrate depth [mm]
6000
0.01 s
Flux [W/m2]
0.50 s
4000
1.00 s
2.00 s
2000
0
0 1 2 3 4
Substrate depth [mm]
Figure 7.4: Constant gas temperature case: temperature and flux profiles in the substrate
136
q0 initial =5357 W/m2 from erfc α=0.88E-6 W/m2
25
Temperature (C)
24
23
22
21
20
0.0 0.5 1.0 1.5 2.0
Time [s]
6000
Flux (W/m2)
4000
2000
0
0.0 0.5 1.0 1.5 2.0
Time [s]
Figure 7.5: Constant gas temperature case: example of wall temperature and wall flux
history
Vout
R R
R/2
Vin C C
Unit length
Figure 7.6: RC network for analogue resolution of 1D heat conduction equation with
semi-infinite substrate assumption
137
0.0006
Input signal
0.0004
0.0002
0.0000
1.0
0.8
Output signal
0.6
0.4
0.2
0.0
0.0004
0.0002
0.0000
0.0
-0.5 0.5
-
-1.0
0.0
-1.5
-2.0
0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8 1.0
Time [s] Time [s]
138
23
Crank-Nicholson
Substrate Temperature [C]
22 Exact solution
t=0.00 s
t=0.25 s
t=0.50 s
t=1.00 s
21
20
0 1 2 3 4
Substrate Depth [mm]
0.0006
Input signal
0.0004
0.0002
0.0000
1.0
0.8
Output signal
0.6
0.4
0.2
0.0
139
40
30
10
100000
Flux [W/m2]
50000
Figure 7.11: Flux computation with Crank-Nicholson scheme for a real test
140000
Analogue circuit
Crank-Nicholson
130000
Flux [W/m2]
120000
110000
100000
90000
0.06 0.08 0.10 0.12 0.14
Time [s]
140
Tests in air (rac)
80
Intensity [mA]
60
40
20
0.30
Flux [W]
0.20
0.10
0.00
0.0 0.5 1.0 1.5 2.0 2.5
Time [ms]
65.5
65.0
299
Temperature [K]
297
295
293
0.0 0.5 1.0 1.5 2.0 2.5
Time [ms]
141
1 Hz 4660 Hz
6000 6000
4000 4000
Flux [W/m ]
2
2000 2000
0 0
-2000 -2000
-4000 -4000
-6000 -6000
0 1 2 3 4 0.0 0.5 1.0
Time [s] Time [ms]
Surf. temp. change [K]
0.02
1.0
0.5 0.01
0.0 0.00
-0.5 -0.01
-1.0
-0.02
0 1 2 3 4 0.0 0.5 1.0
Time [s] Time [ms]
1000
Amplifier gain
100
10
0 1 10 100 1000 10000
2.0
1.0
Phase [Rad]
0.0
-1.0
-2.0
0 1 10 100 1000 10000
Frequency [Hz]
142
Figure 7.17: Blade with Macor insert in the leading edge
Atmospheric conditions
1.5
1.0
Ts rise [C]
0.5
0.0
15000
173 m/s (∆T=48 C) 9 RPM
175 m/s (∆T=44 C) 310 RPM
186 m/s (∆T=40 C) 1540 RPM
Flux [W/m ]
10000
2
5000
0
-0.5 0.0 0.5 1.0 1.5
Reduced distance
Figure 7.18: Surface temperature and heat fluxes measured in rotation across a heated jet
143
144
Chapter 8
8.1 Introduction
With the application of piezo-resistive semi-conductor sensors to pressure mea-
surements, a new tool became available to resolve pressure fluctuations at high frequency.
The topology and schematic of such pressure sensor are sketched in Fig. 8.1. It consists of
a thin silicon membrane on top of which a Wheastone bridge of piezo-resistors is diffused.
The property of a piezo-resistor is to exhibit a change of resistance which is proportional
to the strain applied on it. The resitors are positionned on the membrane in such a way
that, when a pressure is applied on the membrane, two of them are under tension whereas
the two others are under compression. The voltage output of the Wheastone bridge gives
an image of the strain of the membrane which is proportional to the pressure applied on
its surface. A carefull positioning and matching of the resistors allows to have a volt-
age output which is proportional to the pressure applied (Ainsworth [101]). The thin
diaphragm has a very small mass and is able to follow very fast fluctuations. Actual chips
have natural frequencies up to 400 kHz.
1
P = (V − O) (8.1)
S
In a limited range of temperature, linear laws fit very well the dependance of
dS
the slope as a function of tempe!rature dT as well as the dependance of the origin as a
dO
function of temperature dT . With the knowledge of these sensitivities, the “true” pressure
145
can be derived from Eq. 8.2
1 dO
P = V − (O0 + (T − T0 )) (8.2)
S0 + dT (T − T0 )
dS dT
dS
dT is positive whereas dO
dT is negative. Under a given pressure, if the gauge
output is not corrected, a temperature which is higher than the reference temperature
causes:
- an increase in the offset 0 [V] corresponding to an underestimation of the true pressure,
- an increase in the slope S [V/Pa] leading to an overestimation of the true pressure.
Obviously, one needs to know the temperature of the chip during the measure-
ment in order to be able to apply this correction. The temperature of the chip is measured
thanks to the fact that the overall resistance Rb of the pressure sensor varies linearly with
temperature which can be modelled by Eq. 8.3.
The coefficient needed for the temperature correction of each sensor (Eq. 8.2)
are reported in [102] and are plotted in Fig. 8.4. The slope, the slope sensitivity to tem-
perature and the temperature coefficient αb are of the same order for all sensors. It is
not the case for the offset sensitivity to temperature which is about 100 times higher for
two sensors (no 18 and 22) (a logaritmic scale was used for this graph). Such differences
can be entirely attributed to the manufacturing process of the sensors. The mean of each
coefficient was computed (excluding the two sensors mentioned above) in order to estimate
the errors resulting from the temperature dependance for an “average” sensor.
For a given constant voltage, the difference between the uncorrected pressure
and the corrected pressure was computed for a temperature increase up to 30 K. This was
done in three steps:
146
dS
1) pressure difference computed with dT = 0 (only the sensitivity of the offset to temper-
ature is taken in account),
d0
2) pressure difference computed with dT = 0 (only the sensitivity of the slope to temper-
ature is taken in account),
d0 dS
3) pressure difference resulting from both dT and dT = 0.
The results are reported in Fig. 8.5 For this set of sensors, it appears that the
effect of temperature on the slope is bigger than on the offset and the overall effect of
temperature is an underestimation of the pressure.
One blade instrumented with 8 sensors was tested in rotation in the CT3
windtunnel in order to observe the effects of the centrifugal force as well as the effect of
temperature increase due to windage losses.
Fig. 8.6 shows the evolution of pressure and temperature for three sensors dur-
ing a run-up from 0 to 6000 RPM. The curves of all the others sensors are in between the
one of sensor 12 and the one of sensor 19. The absolute pressure (Ps3 ) in the test section
is measured with a pneumatic static pressure tapping located in the casing downstream
of the rotor. The temperature (T01 ) is measured with a thermocouple placed upstream of
the stator (thick lines in Fig. 8.6). The initial conditions for this test were T01 = 19.5 o C
and Ps3 = 45 mbar. The dotted, dashed and dash-dotted lines are the uncorrected pres-
sures and temperatures of the three sensors; the continuous line next to each pressure line
represents the corresponding corrected pressure.
The pressure and the temperature measured by the sensors on the blade differ
considerably from those given by the pressure tapping and the thermocouple. This dif-
ference increases as the rotational speed increases. Even if the windage flow around the
blade causes a local pressure which is different from the pressure in the test section, it
is unlikely that at 6000 RPM, one sensor sees a pressure which is half of the pressure in
the test section whereas the other sees a pressure which is twice as high. As regards to
temperatures, it is surprising to note that the sensors 8 and 19, which are directly affected
by the heating of the blade surface due to windage losses, exhibit lower temperatures than
the thermocouple located upstream of the stator i.e. at a large distance from the rotor.
147
corrected pressure levels are still not realistic. The most likely explanation for this effect
is the influence of the centrifugal forces on the sensor membrane. Depending on the incli-
nation of the membrane with respect to the radial direction (direction of the centrifugal
force), the drift can be positive or negative. It is likely that the centrifugal forces also
cause an error on the sensor temperature signal.
If the discrepancy is entirely caused by 2), the error is only ±3% of the initial
test section pressure. As the initial pressure never exceeds 70 mbars, the error would not
exceed ±2.1mbar which is negligible.
8.6 Conclusions
The working principle of the piezo-resistive fast pressure sensor was explained
as well as a method used to correct the output signal for a temperature effect. Tests
148
showed that the drift to centrifugal force is predominant compared with the drift due to
temperature effect. A simple method was proposed in order to correct for both effects
prior to blowdown. The problem of the correction during the blowdown (rotational speed
varies and blade surface temperature increases) will be adressed later.
149
8.7 Figures
Pressure
Piezoresistors
Silicon
membrane
Figure 8.1: Topology and schematic of semi-conductor sensor chip from Gossweiler et al.
[103]
Rp=Rp0 ( 1 + αp (Tp-Tp0))
Piezo-resistif sensor
Vpressure
Rp
Wheatstone bridge
Vchip Temperature
150
Figure 8.3: Rotor blade instrumented with flush-mounted fast response pressure sensors
cal/kulox.set
1e-02 1e-06
8e-07
dOdT *-1 [V/C]
1e-03
SO [V/Pa]
6e-07
1e-04
4e-07
1e-05
2e-07
1e-06 0e+00
5 10 15 20 25 5 10 15 20 25
1e-03
1e-03
1/S0 dS/dT [1/C]
0.0019
9e-04
α[1/K]
8e-04 0.0018
7e-04
0.0017
6e-04
5e-04 0.0016
5 10 15 20 25 5 10 15 20 25
Gauge no Gauge no
Figure 8.4: Coefficients needed for the correction of the fast response pressure sensors
output for each sensor
151
S=4.5e-7 [V/Pa] dO/dT=-1.5 e-5 [V/C] dS/dT=3.6e-10 [V/(Pa C)]
kulcorr.set True pressure drift for a voltage of 1V (0.8888 bar)
8
6 ∆P due to 0
[mbar] 4
2
-10
[mbar]
∆P due to S
-20
-5
[mbar]
-15
0 5 10 15 20 25 30
Temperature increase [K]
Figure 8.5: Error resulting from temperature effect for a typical sensor
Pressure [bar] Rotational speed [RPM]
6000
4000
2000
no 19
Corrected for
no 8 temperature
0.08
no 12
0.06 Ps3
0.04
0.02
Gauge temperature [C]
no 19
30 no 8
no 12
25 T01
20
15
0 50 100 150 200 250 300
Time [s]
152
ratset2.005
6000
4000
2000
no 19
0.30 no 8
no 12
0.20 Ps3
0.10
Gauge temperature [C]
no 19
65 no 8
55 no 12
45 T01
35
25
15
370 390 410 430
Time [s]
Figure 8.7: Sensors pressure and temperature evolution during dump tank pressure rise
ratset6.005
0.04
12
0.02
8
P(kul,corr)-Ps3
0.00
19
-0.02
-0.04
0e+00 1e+07 2e+07 3e+07
2
RPM
153
154
Part III
155
Chapter 9
The design of the high pressure transonic turbine stage was carried out by Alfa-
Avio. Although it was designed for research, the main characteristics are representative
for advanced aero engines. The blade design is a long iterative procedure starting with
1D calculation and ending with 3D Navier-Stokes calculations. The procedure will not be
described here but is fully reported in Ref. [5] for the stator and Ref. [104] for the rotor.
In this chapter is given a summary of the rotor and stator blade operating conditions and
geometries as well as the main stage characteristics at mid-span.
The first high pressure turbine stage is located right after the combustion
chamber. For this reason, the flow enters the stator blade row axially: α1 = 0. Modern
turbine stages are highly loaded i.e. stator and rotor blade perform both high turning
angles, above 70.3 degrees for the stator and above 100 degrees for the rotor. For this rea-
son, a stator outlet angle α2 = 72.3o and a deflection of 105 degrees across the rotor were
chosen. A transonic stator outlet Mach number M2,is = 1.05 was adopted and reflects
also the actual design tendency. In this Mach number range, mass flow is maximum and
losses due to shock waves are still low. The Reynolds number based on chord and outlet
velocity was chosen as Re2,c = 1.0E06 and is a compromise between values encountered
in large jet engines and helicopter engines. The natural turbulence level in the VKI wind
tunnel is of the order of 1 to 2 % (as in many other tunnels) and is not representative of
high turbulence levels encountered in real engines. Turbulence grid could be installed but
one of the objectives of the experimental programme is to provide a test case for unsteady
computations, it was decided to stay at a low turbulence level.
157
temperature ratio, and coolant to wall temperature ratio) but under lower pressure and
temperature: P01 = 1.620 bar and T01 = 440 K. This choice will be justified later when
describing the working principle of the test facility.
The inlet velocity V1 is computed from the design mass flow ṁ = 10.857 kg/s
and inlet area S1 = 0.118 m2 with Eq. 9.1.
ṁ = ρV S (9.1)
In order to simplify the following calculations, the expansion across the stator
will be assumed adiabatic namely the blade and endwalls are at adiabatic wall tempera-
ture and no coolant is ejected from the blade. With this hypothesis, the total enthalpy
across the stator remains constant: H01 = H02 which is equivalent to T01 = T02 with a
constant cp (H = cp T ).
Ideally, an isentropic process would take place inside the blade row and one
would have: P02 = P01 .
The design outlet isentropic Mach number being M2,is = 1.05, the isentropic outlet condi-
tions P2,is can be computed using Eq. 9.4 and T2,is can be derived from Eq. 9.3 or Eq. 9.5
valid for any isentropic evolution.
γ−1
γ
Ti Pi
= (9.5)
Tf Pf
158
(The indices i and f refer to an initial and final state of the isentropic evolution; notice
that Eq. 9.5 is valid for total quantities as well as for static quantities).
As the process is not isentropic there is a total pressure loss across the stator.
An estimation of the overall losses was carried out by the designer resulting in a mean total
downstream pressure P02 = 1.582 bar which corresponds to a Mach number M2 = 1.03
(M2 is computed by reversing Eq. 9.4). This Mach number allows to compute the static
temperature T2 using Eq. 9.3. Even in the non-isentropic case, T02 = T01 in a stationary
blade row with adiabatic evolution. The downstream static pressure P2 is an imposed
condition and is the same in the isentropic and non-isentropic case. Outlet velocity is
computed with Eq. 9.6.
V = M.a = M γrT (9.6)
The design outlet angle being α2 = 72.3o , axial and tangential components of the outlet
velocity can be computed. Outlet conditions are summarised in table 9.2.
Rotor blade
The turbine diameter was chosen to fit the existing test section dimensions.
Given the diameter and the circumferentiel speed, the rotational speed of the rotor is
6500 RPM at design. The design Reynolds number based on chord and relative outlet
159
conditions is 0.5 106 and results from a compromise between what is encountered in high
pressure stages of turbofan, turboprop and turboshaft. The radius at mid span is fixed by
the wind tunnel geometry and the blade aspect ratio. The mid-span radius at rotor inlet
is 369.7 mm corresponding to a peripheral speed u2 = 251.6 m/s.
Rotor relative inlet conditions are derived from state 2 after the non-isentropic
evolution. The relative inlet velocity vector is obtained by combining the absolute out-
let velocity vector V
2 and the peripheral speed u
2 as sketched on the velocity triangle in
Fig. 9.1:
V2ra = V2a
V2rt = V2t − u2
v2rt
α2r = atan( )
v2ra
Static conditions are identical in the relative and the absolute frame of reference.
√ The
relative Mach number is computed with the relation M2r = V2r ∗ a2 = V2r ∗ γrT2 . Rela-
tive total pressure and temperature result from Eq. 9.4 and Eq. 9.3. Relative rotor inlet
conditions are summarised in table 9.3.
u23 u2
H03r − = H02r − 2
2 2
For a constant radius and a constant cp , this relation is equivalent to:
T03r = T02r
160
All outlet isentropic conditions can be computed from the isentropic Mach number (M3r,is =
0.93). The resulting exit static pressure is identical for the isentropic and non-isentropic
evolution (it is an imposed condition) which allows to compute all other quantities in the
same way that it was done for the stator. The results are presented in table 9.4.
Isentropic and non isentropic evolution across the rotor are sketched in the
(T-S) diagram in Fig. 9.2.
Outlet absolute conditions are computed from the non isentropic relative con-
ditions with a peripheral speed u3 = 251.6m/s. The peripheral velocity vector is added to
the relative outlet velocity vector to give the absolute velocity vector (see Fig. 9.1). The
static conditions being identical for the relative and, the absolute frame, the outlet Mach
number, the total pressure and temperature can be computed. Results are presented in
table 9.5.
The isentropic static temperature presented in table 9.4 was computed with
respect to state 02. For the estimation of the efficiency, the isentropic evolution from state
01 will be considered. The resulting total temperature is derived from Eq. 9.5 and is equal
to T03,is,01 = 331.6 K.
161
9.2 Geometrical blade and stage characteristics
Stator blade
The design of the blade shape was performed by means of an inverse 2D Euler
code. From an initial blade geometry, the code finds the modifications to apply in order
to obtain a prescribed Mach number distribution around the blade. On the suction side,
a smooth acceleration up to the shock, minimum shock strenght and smooth deceleration
behind the shock were prescribed. On the pressure side, the aim was to avoid the usual
velocity peak in the nose region.
This code was used in an iterative procedure together with a boundary layer
code and a 3D Euler code. An additional constraint was to be able to realise the inter-
nal cooling of the blade with coolant ejection on the pressure side of the trailing edge,
namely a trailing edge which is thick enough to maintain mechanical integrity in presence
of coolant slots. Fig. 9.3 shows the uncooled blade and the blade with a cut in the trailing
edge for coolant ejection from the pressure side. Notice that, the throat will be different
in the cooled and uncooled case.
The 3D blade results from the 2D section stacked along a radius passing
through the throat center of the uncooled blade. As the leading edge is located right
on the same axis as the throat center, the 3D blade can be obtained by keeping a constant
blade section along the z axis (the leading edge is radial). This rather simple design pro-
162
vide considerable gain in time for the design and costs when manufacturing the blades.
Below are given the throat, pitch and gauging angle at hub, mid and tip section.
Table 9.7: Throat, pitch and gauging angle at hub, mid and tip section for the uncooled
vane
Rotor blade
In the rotor blade design, attention was paid to avoid the influence of sec-
ondary flows at mid-span. The trailing edge thickness is such that the blade could be
transformed later on in a coolable blade with ejection at trailing edge from the pressure
side. The number of blades is 64 and the pitch to chord ratio of 0.75 is a compromise
between the actual tendency to have a large pitch to chord ratio (less blades means lower
costs, less inertia and less rotor disc stresses) and maintaining a reasonable aspect ratio
in order to avoid large secondary flows.
The design procedure was similar to the one applied for the stator. Geomet-
rical characteristics at mid span are presented in table 9.8.
163
The geometry of the blade at hub, mid span and tip is shown in Fig. 9.4. The
change in blade section as a function of the radius reflects the change of inlet conditions
along the radius due to non-uniform radial distribution of peripheral speed, static pressure
and stator outlet angle.
The rotor was designed with a meridional flow channel divergence of 10%
in order to reduce the axial outlet velocity and to minimise the secondary losses. The
projection of the blade suction side along the y axis (as defined in Fig. 9.4) is plotted
in Fig. 9.5 and gives an idea of the shape of the divergence. The stage geometry at mid
span is depicted in Fig. 9.6
Fig. 9.7 shows a 3D view of an arrangment of two stator blades and three
number of rotor blades
rotor blades. Notice that the ratio number of stator vanes = 1.48 is close to 1.5. This allows
to perform the numerical simulation with a domain of 2 vanes and 3 blades rather than
43 vanes and 64 blades.
Blade row
Total pressure loss coefficient
Expresses the total pressure loss non-dimensionalised by the dynamic pressure. For the
stator, this coefficient can be written
P01 − P02
ω= (9.8)
P02 − P2
ω = 4.9% for the stator, ω = 11.2% for the rotor. The higher value in the rotor case is
justified by a larger achieved flow deviation.
(γ−1)
1− P2 γ
H02 − H2 V22 T02 P02
η= = 2 =
(γ−1) (9.9)
H01 − H2,is V2,is T01 γ
1− P2
P01
For the present case, T02 = T01 , ηstator = 0.9687. The corresponding kinetic energy loss
coefficient ) is :
ζ =1−η (9.10)
164
Stage
Total to total pressure ratio
Characterises the total pressure drop performed across the stage
P01
= 2.69
P03
Specific work
Expresses the work performed per unit mass of flow in the stage. It can be expressed as
It could also be computed from the velocities using the Euler equation for turbomachines
[105]: ∆H0 = u3 V3,t − u2 V2,t , Vt being an algebraic value. Notice that in the case where
u2 = u3 the work is directly linked to the flow turning between stator exit and rotor exit.
The quantity ∆H0 is depicted in Fig. 9.2
Power
P = ṁ ∆H0 = 10.857 ∗ 103.8 103 = 1038 kW
Torque
T = Pω = P
2πf = 1038 103 /(2 ∗ π ∗ 6500/60) = 1525 N.m
The same expression is obtained with temperatures assuming a constant cp which leads
to ηT T = 0.94. The quantities ∆H0 and ∆H0,is are depicted in Fig. 9.2.
Va
φ=
u
φ = 0.605 based on rotor exit.
Degree of reaction
The degree of reaction of a turbine is defined as:
165
Stage loading
∆His
Ψ=
u2
Ψ = 1.64
Strouhal number
A typical non dimensional number in unsteady phenomena is the Strouhal number defined
as:
l fl
St = = (9.11)
tV V
l is a characteristic length, f the frequency of the phenomena, V the main stream velocity.
This number compares the time scale of a typical perturbations (1/f) with the time scale
of the main flow convected around a body (l/V). In the case of this stage, a rotor blade
(l=c=48.42 mm) sees perturbations emerging from the 43 stator vanes at a frequency
f=6500/60*43 = 4660 Hz. The relative inlet rotor speed being V=162 m/s, the Strouhal
number is St=1.39
166
9.4 Figures
V3r=322
u3=252
α2 r= 45.9
α2= 72.3
Stator blade row
α3 r= 61.9
α3= 12.0
V2r=172
u2=252
V2=394
167
460
P01
01 P02
T01
440
P1 02
420
∆H/2Cp
V 2/2Cp
/2Cp
W 2/2Cp
Temperature (K)
400 2
2,is
P02r
2
∆His/Cp
380 P03r
03,r
P2
/2Cp
W 3/2Cp
360
2,is 2
3,is
2
2
W
T03 P03
340
03
03,is from 01 03,is from 02 P3
320 3
300
-10 -5 0 5 10 15 20 25 30
Entropy [J/(kg K)]
168
Figure 9.3: Stator blade cylindrical cut at mid section with and without pressure side
ejection slot
20
hub
mid
10 tip
0
y [mm]
-10
-20
-30
-20 -10 0 10 20 30
x [mm]
Figure 9.4: Hub, mid and tip section of the rotor blade
169
390
370
z (mm)
350
330
-35 -15 5 25
x(mm)
Figure 9.5: Projection of the rotor blade suction side along the y direction
140
120
100
r θ [mm]
80
60
40
20
0
-20 0 20 40 60 80 100 120
x [mm]
170
Rotor
Blade
Row
Stator
Blade
Row
Inlet Flow
171
172
Chapter 10
The compression tube turbine test rig is based on the large VKI compression
tube annular cascade facility constructed in the late 80’ies and operated routinely since
1990 (see Sieverding and Arts (1992) [106]).
The principle of this type of facility was developped at Oxford in the mid-
seventies (Schultz et al. [107]). The facility is of the blowdown type. The test section
is supplied with air from a large cylinder in which the air is compressed through a free
moving light-weight piston. After passing through the test section the air is discharged
into a dump tank. The testing time depends on the ratio of the test section size and the
cylinder volume and is typically of the order of 0.5 to 1 s.
173
10.2 General lay-out of the compression tube facility
The main elements of the facility are shown in Fig 10.1:
- a 1.6 m diameter and 8 m long compression tube containing a light-weight piston driven
by air from the Institute’s high pressure air supply system (250 bars),
- a vertical oriented fast opening shutter valve closing the 280 mm diameter central vent
hole in the end plate of the tube,
- a radial diffuser discharging the compressed air into an annular settling chamber of 500
mm inner diameter and 900 mm outer diameter,
- the test section,
- a 15 m3 vacuum tank separated from the test section through a variable area sonic throat.
The photograph of the facility (Fig 10.2), taken in 1991, gives an impression
of the size of the facility. A schematic view of the meridional plane of the test section with
the annular cascade configuration is shown in Fig. 10.3. The guide vane is preceeded by an
inlet contraction simulating the exit of a combustion chamber. The contraction protudes
into the much larger settling chamber. Downstream of the guide vane, the air is guided via
a short constant area duct to a discharge chamber. The outlet duct connects this chamber
with the far-dowstream dump tank. The outlet duct is mounted on a large carriage. The
attachment of the outlet duct to the guide vane housing on one side and the dump tank
on the other side is secured by hydraulic clamps. After opening of the hydraulic closure
the outlet duct is pushed backwards to provide access to the test section.
The test section of the turbine test rig comprises now 5 major modules (see
Fig. 10.4):
- a stator housing with guide vane (apart minor modifications, it is the same as before),
- a rotor housing with a rotor shaft, a rotor disc and an inertia wheel,
- a sonic throat module,
- a data transmission shaft module,
- an air motor drive.
The rotor shaft is supported at the front by one free roller bearing and at the
rear by two fixed ball bearings with oblique contact in order to support the large axial
load during blowdown. The bearing housing is attached to the rotor housing by 6 airfoil
shaped struts. The bearing housing carcase is shaped to the form of the hub end-wall of
the stage outlet flow duct. The rotor is mounted in overhang position. Because of the long
overhang section of the existing guide-vane housing, the rotor housing had to be designed
with the rotor protruding far from the rotor housing.
174
Since the turbine is not equipped with a brake to absorbe the power the rotor
speed will change during the test run. The acceleration depends on the power generated
by the turbine and the inertia of the rotating parts. The overall change of the rotational
speed depends on the duration of the test run. The inertia of the rotor shaft, disc and
blades of the BRITE turbine amounts to about 9.7 kg m2 . Assuming an initial rotor speed
of 6500 RPM and a running time of 0.5 s, the rotor will accelerate from 6500 to 7200
RPM during the run (1400 RPM/s). To limit this acceleration the inertia of the rotor was
augmented by adding an inertia wheel mounted on the shaft between the front and the
rear bearings. The overal inertia is of the order of 19.4 kg m2 leading to an acceleration
rate of about 700 RPM/s, i.e. the rotationnal speed varies by only 5% during a 0.5 s test
run. The overall rotating mass is of the order of 300 kg.
The right end of the rotor shaft is coupled to the data transmission shaft which
serves also as link between the rotor shaft and the air motor drive. The air motor (from
Tech. Development Inc., Model 50A, used by the aircraft companies as engine air starter)
delivers about 43 kW at 6000 RPM with a 10 bar air supply and a mass flow rate of 1kg/s .
The accelerating time of the rotor to near nominal conditions before the test run depends
on the total inertia of all rotating parts, the bearing losses and the windage losses. The
latter are by far the most important at high rotational speeds and the test section has to
be depressurized to a very low level to reach the desired speed level within a reasonable
time period. For tests at design speed (6500 RPM) the run-up time is about 300 seconds
at a test section pressure level of about 50 mbar.
The power of the air motor is transmitted through a Thomas miniature flexible disc cou-
pling. A similar coupling is used between the driver shaft and the data transmission shaft
while a semi-rigid large disc coupling between this shaft and the turbine rotor shaft.
The variable sonic throat designed to set the stage pressure ratio and ensure a
constant stage outlet static pressure during the test run is positioned at a short distance
downstream of the rotor bearing housing. The space between the bearing house and the
throttle valve serves as discharge chamber for the flow exiting from the stage outlet duct.
The throat opening is actuated manually by a perpetual screw. An encoder indicates its
exact position.
Similar as for the annular cascade configuration, the outlet duct is fixed on a
hydraulic actuated displacement carriage. However the attachment of the rotor housing,
the sonic throat module and the data transmission shaft module to the outlet duct results
in an overhang weight for the carriage of 2200 kg compared to 300 kg before, requiring a
complete redesign of the carriage and the carriage support structure. Fig. 10.5 shows a
photograph of the facility in its opened position. To avoid damage of the facility in the
case of strong rotor vibrations due to the loss of rotor blade instrumentation or worse,
loss of a blade, the test section is anchored very solidly to the new foundations of 12 m3
of concrete (the centrifugal force on a blade at 6500 RPM is equivalent to 6 tons).
10.4 Aero-brake
After a blowdown test, with an initial speed of 6400 RPM, the rotor speed
would typically reach 6900 RPM due to the 1MW power delivery during blowdown. With-
175
out any brake, the deceleration of the rotor would be very slow. The lack of any active
control of the rotor speed in case of an emergency let to the decision to design on top
of the inertia wheel a small aero brake which could be actuated when ever needed. The
power of this aero-brake would be too small to significantly reduce the acceleration of the
turbine rotor during the blowdown but should significantly increase the deceleration rate
of the rotor at any other time during the test cycle if required so.
The aero-brake is designed as partial admission impulse type rotor. The airflow
enters axially (there is no stator) from 4 individual chambers covering each two rotor
passages. The air is fed to these chambers via the struts supporting the rotor bearing
housing. The mean radius of the rotor blades is 0.300 m. The turbine is designed for a
nominal rotational speed of 4000 RPM. At this speed, the peripheral speed is 125 m/s
and the rotor relative inlet angle is 51.3 deg. The thermodynamic conditions at design
are reported in table 10.2. The corresponding velocity triangle is shown in Fig. 10.6. The
design was performed assuming ηrotor = 0.7. The specific work is 45 kJ/kg. With an
overall mass flow of 1Kg/s, the power is 45 kW.
Quantity 2 2r 3r 3
P0 [bar] 2.09 2.30 1.755 2.678
T0 [K] 280.0 287.7 287.7 324.7
M 0.30 0.484 0.94 1.284
P [bar] 1.967 1.967 1.0 1.0
T [K] 275.0 275.0 245.0 245
α[deg] 0. 51.3 53.3 64.1
v [m/s] 100 161 295.3 403.2
The design of the blade was carried out with a 2D Euler inverse code (devel-
opped at VKI by Léonard [13]). The blade section is shown in Fig. 10.7. 70 blades of 13
mm height were machined on top of the inertia wheel.
The wires from the instrumented blade are soldered onto a 40 tracks flat cable
glued onto the rotor surface. The flat cable extends from the tip of the disc down to its
root. At the root, wires take the signal to a set of female connectors arranged annularly so
that the centrifugal force pushes the male part of the connector towards the female part.
The standard 4 pins male connectors are connected to the in-shaft instrumentation board
(see Fig. 10.4) through a 32 pins standard connector. The shaft allows the presence of
2 instrumentation boards. Each intrumentation board is connected to a 8 channel data
transmission board through a 32 pins connector. The instrumentation boards can be eas-
ily changed depending on the type of measurements (pressure, heat transfer, temperature).
The data tranmission from the rotating frame to the fixed frame is performed
with an in-house developped opto-electronic system (see Sieverding et al. [90]). The
176
transmission board converts the information carried by the amplitude of the voltage into
the frequency domain. There is a linear relationship between voltage and frequency. If
the input voltage is 0 V, the board delivers a 500 kHz square wave; if the input voltage
is 5 V, the boards delivers a 1Mhz square wave. The square wave signal drives a ring of
5 infra-red emittors. The 16 rings of the 16 transmision channels are located between the
rotor shaft and the air motor (see Fig. 10.4 and Fig. 10.8). A fixed receiver faces each
ring. Whatever is the angular position of the rotor, the receiver will always see light from
at least one emitting diode. An alumunium disc isolates two succesive rings so that the
radiations do not interfere. The frequency response of the transmission system is 78 kHz.
The linearity error is below 1 %.
Because of the short tunnel running time, the blade temperature can be con-
sidered to stay at ambient temperature Twall = 293K To reproduce a temperature ratio
Tgaz
Twall = 1.5, the upstream total gas temperature must be T01 = 293 ∗ 1.5 = 440K. The
ratio Tcoolant /Tgaz is equal to 1 for this application. For other applications, the cooling air
is refrigerated or heated in order to achieve the correct temperature ratio.
The stage pressure ratio determines the Mach number at the vane exit. For a
stage pressure ratio of 2.69 (total to total), the design vane exit isentropic Mach number is
M2,is = 1.05. The isentropic outlet Mach number fixes the ratio between static and total
quantities at the vane exit thanks to the relations:
−1
T γ−1 2
= 1+ M
T0 2
− γ
P γ−1 2 γ−1
= 1+ M
P0 2
For M2,is = 1.05, these ratio are respectively 0.819 and 0.498.
Rec µ
ρ=
v.C
√
The isentropic exit velocity can be determined with v = M.a where a = γrT
is the speed of sound. Across a stator, for an adiabatic process, one has: T02 = T01 . The
static temperature for the isentropic évolution is then T2,is = 0.498 ∗ 440 = 360.5 K. The
corresponding speed of sound is a2,is = 1.4 ∗ 287.7 ∗ 360.5 = 381.0 m/s. The exit velocity
for an isentropic evolution is then
177
The viscosity µ2 is a function of temperature; at 360 K, it can be estimated
to µ2 = 2.242 10−5 [kg/(m s)] (the result is very much dependant on the value of the
viscosity). The chord of the stator being Cs = 72 10−3 m, the density at the outlet is
ρ2 = 0.778 kg/m3 .
The outlet static pressure is then simply P = ρrT i.e. Ps2 = 0.807 bar. The
pressure ratio at vane outlet being 0.819, the total downstream pressure P02 is 1.620 bar.
For an isentropic process, one has P01 = P02 .
T01 = 440. K
where i refers to the initial state (before compression) and f to the final state (after
compression). The final level Tf = T01 and Pf = P01 are imposed. The initial temperature
level is the ambient temperature .i.e. Ti = 293 K. As a result, Pi = 0.390 bar. In practice,
the compression is not isentropic. By experience, the total pressure losses and the total
temperature losses between the downstream of the piston and the inlet of the vane are of
the order of 10%. The computation of the initial conditions is then performed for final
levels which are 10% higher than the wanted levels in the test section inlet, i.e. typically
484 K and 1.782 bar leading to Pi = 0.307 bar.
Prior to a blowdown, the pressure level in the upstream tube is set to the
pressure level Pi , computed in the previous section, by means of an ejector or a vaccuum
pump (about 20 mn). The pressure level in the dump tank is lowered down to its minimum
(40 mbar) by a large vaccum pump (about 10 mn). The coolant reservoir is loaded with
pressurised air (3.2 bar). Once all the initial conditions are set, the 10 bar supply of the
air motor is opened in order to speed-up the rotor in the evacuated test section. All the
bearings are lubricated continuously; the overall oil flow rate is typically 355 l/h. The level
of vibrations is controlled continuously with two accelerometers placed, one next to the
front bearing, the other next to the rear bearings. Fig. 10.9 shows a [run-up, run-down]
test without blowdown. The evolution of the rotationnal speed and vibration level are
plotted as a function of time. In this case, the vibration level did not exceed 1.5 mm/s
due to a very good dynamic balancing. For some other tests, the level encountered during
a run-up could reach 10 mm/s. Once the rotor speed is stabilized at 6400 RPM, the
blowdown can be performed.
178
At time t=0 (see Fig. 10.10) the gate valve between the compression tube
and the test section is closed. At time t1 cold high pressure air from the Institutes high
pressure air storage tanks enters the compression tube behind the piston, see Fig. 10.1.
The piston moves forwards and starts to compress and heat-up the air in a quasi-isentropic
way. At time t2, the final tube pressure Pf and final temperature Tf are reached and the
gate valve is opened with an opening time of the order of 70 msec. During this time
the test section total pressure rises rapidly to its nominal value. The downstream static
pressure rise depends on the opening of the downstream throttle valve. The total pressure
remains constant for about 500 msec from time t3 to t5.
The downstream conditions remains constant at least until the continously in-
creasing dump tank pressure rises beyond the critical pressure ratio for choking conditions
in the throttle valve at time t4. As soon as the throat becomes unchoked, perturbations
from the dump tank can travel upstream of the throat. At t5, the piston has reached
the end of the tube: the test run is finished. Fig. 10.11 presents a typical example of the
evolution of all pressures. After the initial rise, all pressures remain remarkably constant
until the end of the test, indicated by the sudden dip in the total pressure. This behaviour
is in contrast to all other light piston compression tube tunnels which are characterized by
important pressure oscillations during the test run. These oscillations are ascribed usually
to the sudden opening of the gate valve. In the VKI tunnel two factors seem to play an
important role in eliminating these fluctuations:
- the relative long opening time of the gate valve,
- probably more importantly, the use of a settling chamber which acts as a damping devise
because of its pressure losses.
At the opening of the main gate valve the rotor undergoes a rapid acceleration.
At the end of the blowdown the rotor starts to decelerate sous l’effet des pertes. Then,
the aero brake is switched on in order to decelerate more rapidly the rotor. In the case
of Fig. 10.9 the rotor is decelerated from 7000 RPM down to 1000 RPM in 2mn04s. The
brake is more efficient at high speed than at low speed because it was designed to be
optimum for 4000 RPM.
179
10.8 Figures
180
Figure 10.2: Photograph of the compression tube facility (1991)
181
Figure 10.3: Meridional view of the test section in the annular cascade configuration
182
Figure 10.4: Sketch of the new turbine facility
183
Figure 10.5: Opened facility with the rotor in overhang position
184
u=125
403
v3=
5
29
r=
v3
53
v2=100
51.3
v2
40.0
30.0
20.0
y (mm)
10.0
0.0
70 blades
13 mm high
-10.0 293.25 mm mean radius
26.32 mm pitch
48.29 mm chord
(0.545 pitch to chord ratio)
-20.0
-10.0 0.0 10.0 20.0 30.0 40.0 50.0 60.0
x(mm)
185
Figure 10.8: Photographs of the shaft equipped with 16 emittor rings (top) and receiver
unit (bottom)
186
8000 3.0
7000
5000 2.0
RPM
4000
3000 1.5
2000
1.0
1000
0 0.5
0 100 200 300 400 500 600
Time [s] Vib7000.set
187
6500 RPM 3% injection
2.0
1.8
P01
1.6
Averaging
1.4
Pressure [bar]
1.2
1.0
Ps2
0.8
P03
0.6
Ps3
0.4
0.2 Pdump
0.0
-0.3 -0.2 -0.1 0.0 0.1 0.2
Time [s] rot.043
188
Chapter 11
Instrumentation
As a convention, 0 deg is defined as the vertical axis of the tunnel. The leading
edge and the axis of a vane are paralell to this axis; this vane will be referred to as vane
1. The angular position becomes positive when rotating in the trigonometric sense when
looking at the turbine from downstream. The rotor rotates in this same trigonometric
sense. All the angular positions given below will be given with respect to this reference.
In the configuration with the smallest axial spacing between stator and rotor,
the distance between the measurement plane of P01 and T01 and the stator leading edge
is 14.37 mm (0.33*Cax,stator).
rotor stator
Chord 48.42 72.0
Axial Chord 39.59 43.2
189
11.1.3 Instrumentation around the blade mid-section
In order to determine the location of the measurement points along the blade
mid-span, a 3D Navier-Stokes calculation was performed by Alfa Avio [104]. The choice
of the gauge location is based on the relative isentropic Mach number at 50% span shown
in Fig. 11.1. 24 measurement points were selected: 13 on suction side, numbered from 1
to 13, and 11 on pressure side, numbered from 14 to 24. The values of axial coordinates,
curvilinear abcissa and isentropic Mach number from the 3D Navier&Stokes calculation
are reported in table 11.4. The location of the points around the blade profile is shown in
Fig. 11.2. The 3D coordinates of each gauge is given in table 11.5.
The pressure sensors are distributed over three blades as indicated in table
11.6. The blades were instrumented at the University of Oxford. One of the instrumented
blade is depicted in Fig. 11.3
The intrumentation of the blades with heat transfer gauges was performed at
VKI. Slots were machined in the blade in order to host Macor inserts shaped to the blade
surface. The locations were traced on the inserts by means of a 4-axis numerical milling
machine. To this end, the blade is mounted in a special support (Fig. 11.4). This support
allows an easy alignment of the blade axis with the machine vertical axis. A fine pen is
placed in the tool holder and for each point, the blade is rotated around its axis so that
the pen would impinge the blade surface perpendicular to the tangent to the surface at the
point of interest. The pen is then carefully driven to each of the required position in (x,y,z).
The gauges are hand-painted on the inserts with typically 4 to 6 gauges per
insert. The 24 measurement points are distributed on 6 different inserts shown in Fig. 11.5.
For mechanical stress reasons, there is only one insert per blade; two of the six blades are
shown in Fig. 11.6. The distribution of the gauges on the different blades can be found in
table 11.7; locations 21 and 6 were instrumented twice.
190
11.1.4 Total pressure and total temperature measurements in rotation:
P02r , T02r
Total pressure and total temperature measurements are performed slightly
ahead of the stagnation point of the blade by means of a fast response pressure probe
(Kulite) and dual hot wire aspirated probe. The probes are inserted in the nose so that
they are aligned with the relative design inlet angle of 45.5 deg (see Fig. 11.7). Due to a
problem of angular sensitivity of the fast response pressure probe, a short pipe is placed
around the probe head in order to create a cavity in which the stagnation pressure is
reached.
The temperature measurements are performed with a cold wire probe. The
measurement point is located along the perpendicular to the surface at location of the
pressure sensor no 4, 2 mm away from the wall
Plane no Dist. with rotor stacking axis [mm] dist. with rotor TE mid span [mm] [*Cax,rotor]
1 41.86 20.0 0.50
2 61.86 40.0 1.00
3 81.86 60.0 1.50
In the first plane, the downstream static pressure Ps3 can be measured at the
tip endwall with two sets of 10 pressure tappings located in two different sectors as illus-
trated in Fig. 11.9. Individual angular positions are listed in table 11.9 (the numbering
191
is done in the sense of rotation of the rotor). Holes were drilled radially. For the two
last planes, only one tap is available in each of the two sectors, the angular position being
identical to tap no 5 in table 11.9.
Ps3 can also be measured at the hub in the first axial position with 10 taps but
only in one sector. The ten holes were drilled parallel to the radius located at the angular
position 82.5 deg and are symetrically distributed around this axis with a spacing between
taps of 5.5 mm corresponding to 8.372/(10-1)=0.93 deg in angle. The corresponding
angular positions are given in table 11.10. The two last planes are not instrumented at
the hub endwall.
Total pressure P03 , total temperature T03 and flow angle al3 measurements
will be carried out with these carriage systems.
For this reason, the signals of the high frequency response probes were shared
in a low-pass filtered component and a high-pass (fpass = 100Hz) filtered component. In
order to obtain a good resolution of the fluctuating component on the data acquisition
system, the high-passed fluctuations were amplified by a factor of 10 prior to acquisition.
The data acquisition was performed with two time bases:
-the steady-state component was sampled at 1171.875 Hz with 1200 samples (1 s duration)
-the fluctuating component was sampled at 256*1171.875=300000 Hz with 35000 samples
(0.116 s duration).
192
In order to avoid aliasing of the steady state component, the low-pass fil-
ter cut-off frequency was set to 750 Hz. The fluctuating component from the rotating
instrumentation is naturally low-pass filtered at 78kHz which is the higher limit of the
transmission system. For the stationary instrumentation, the fluctuating component is
low-passed filtered at 50 kHz.
The blade passing frequencies and the mean number of samples per traverse
for each of the three rotational speeds are reported in table 11.3.
A periodmeter with a time base of 5 Mhz determines the time necessary for
one revolution (64 pulses) and delivers a voltage proportional to this time period. This
voltage is refreshed at each revolution namely about every 0.01 s and allows accurate trig-
gering to start the fast acquisition when the rotor reaches the wanted rotational speed .
The output voltage of the photodiode is also sampled at 300 kHz. Because one
of the 64 pulse is larger than the other, the analysis of this signal alows to determine when
the infrared beam enters the large hole. The rotor position with respect to the stator
at this moment was determined previously. The processing of this signal is performed
by software. It provides a more accurate measurement of the rotational speed than the
periodmeter (better than +-3 RPM below 7000 RPM) but cannot be used for triggering
because the analysis is performed a posteriori. Once per revolution, it also provides the
time at which the beam enters the large hole i.e. it locates the rotor position with respect
193
to the stator at this time. The error on this time is only ±1/300000 s. The corresponding
error on the spatial location at 6800 RPM is ±1.6% of a stator pitch or ±2.4% of a rotor
pitch. This time will be very usefull for the phase locked averaging process.
Prior to a test, two acquistions are performed with all the intrumentation
connected: one at atmospheric pressure and ambient temperature in the test section, the
other at the initial pressure prior to the blowdown, Pdump (typically 70 mbar). Those
files are used to calibrate all the pressure sensors (pneumatic or fast response) as well as
to control the day to day drifts of the voltage outputs of the cold wire probe, the thermo-
couples, the measurement of R0 of the thin film gauges, etc.
The pneumatic pressure sensors (Sensym and N.S.) exhibit a very good sta-
bility in terms of slope and origin. In order to estimate the dispersion between sensors, a
number of successive measurements at P=Pdump was performed with a dozen of pressure
sensors. The resulting dispersion (20:1) is ±8mbar.
The fast response pressure gauges (gauges on the rotor blades or probe be-
hind the rotor) are calibrated at ambient temperature in the same way as the pneumatic
pressure sensors. A good repetitivity is obtained as shown in Fig. 11.11. Excluding tem-
perature effect and centrifugal force effects, the accuracy of the fast response pressure
sensors is as good as the sensors used with the pneumatic probes.
The voltage Vsense (image of gauge temperature for the gauges located on the
rotor blades) was also measured at least once a day at ambient temperature to have the
offset of the law DV/DT. The slope (DV/DR) of the board relating the voltage output to
the gauge resistance was calibrated only once. The temperature coefficient αgauge of each
gauge was provided by the University of Oxford [102].
The cold wire probes were also calibrated only once. The expected error on
cold wire measurements excluding frequency response problems is identical to the one of
the thermocouples.
194
11.2.4 Calibration of the instrumentation measuring blade surface tem-
perature for heat flux measurements
The blades with the inserts were put into an oil bath in order to perform the
calibration for the sensitivity to temperature for each gauge. Prior to calibration, three
thermal cycles were performed (heating up and cooling down the blade) in order to check
that there were no hysteresis effect. For all gauges, the resistances always came back
to their initial values at ambient temperature within ±0.02Ω. The average ∆R/∆T is
0.106Ω/o C corresponding to a sensitivity to temperature of α = 0.00189.
The board calibration was performed in situ for each channel and both R0
and ∆R modes. In the R0 measuring mode, the ratio ∆V /∆T is about 1mV /C which
means that a drift of only 10 mV would cause an error of 10 degrees. In order to avoid
the day-to-day drift, the origin of the law was computed before each test by taking a
measurement at ambient temperature.
195
The different configurations will be tested in two parts:
Part 2: maximum inter-blade row gap at 6500 RPM and 3% coolant flow rate.
The axial gap at mid-span is defined as the distance between stator trailing
edge and rotor leading edge. Spacer rings allow the displacement of the stator with respect
to the fixed axial position of the rotor from a minimum axial gap at mid-span of 15 mm
to a maximum of 21 mm corresponding respectively to 0.347 and 0.486 times the axial
chord of the stator.
In the following some results of part 1 of the investigation program are pre-
sented.
196
11.4 Tables, figures
Table 11.4: Relative isentropic Mach number distribution from 3D Navier&Stokes com-
putation
197
Position no x [mm] y [mm] r[mm] z[mm]
1 0.057 1.084 368.859 368.857
2 0.569 2.930 368.780 368.768
3 2.145 5.960 368.629 368.581
4 5.577 9.655 368.577 368.451
5 10.203 11.874 368.507 368.315
6 17.365 11.172 368.327 368.157
7 23.291 6.773 367.932 367.870
8 27.358 1.418 367.619 367.616
9 29.887 -2.897 367.434 367.422
10 31.996 -7.030 367.299 367.231
11 34.470 -12.322 367.174 366.967
12 36.914 -17.900 367.097 366.660
13 39.001 -23.034 367.074 366.350
14 0.340 -0.602 368.815 368.814
15 1.698 -1.485 368.653 368.650
16 3.836 -1.397 368.580 368.577
17 6.785 -1.015 368.573 368.572
18 14.155 -1.540 368.451 368.448
19 22.360 -5.321 368.002 367.964
20 27.956 -10.232 367.574 367.431
21 32.227 -15.494 367.285 366.958
22 35.270 -20.039 367.143 366.596
23 36.697 -22.349 367.101 366.420
24 37.976 -24.501 367.079 366.260
198
Blade 1 Blade 2 Blade 3 Blade 4 Blade 5 Blade 6
21 18 14 1 6 11
22 19 15 2 7 12
23 20 16 3 8 13
24 21 17 4 9
18 5 10
6
Table 11.7: Distribution of the heat transfer gauges on the 6 different blades
Table 11.8: Angular position of static pressure taps for ps2 measurements
Table 11.9: Angular position of static pressure taps for ps3 measurements in plane 1 at
tip downstream of the rotor
199
no position [deg]
1 78.31
2 79.24
3 80.18
4 81.11
5 82.03
6 82.96
7 83.89
8 84.82
9 85.75
10 86.69
Table 11.10: Angular position of static pressure taps for ps3 measurements at hub in plane
1 downstream of the rotor
Misloc.set
1.5
Relative Isentropic Mach Number
9
8 10
1.0 11 12
7
6 24
5 23
4 22
21
3
0.5 20
2
19
1 16 17 18
0.0
0 10 20 30 40
xax [mm]
200
instrfig.set
6
4
10
7
3
2
8
14 17
0 15 16 18
9
19
y [mm]
10
20
-10 11
21
12
22
-20
23 13
24
0 10 20 30 40
xax [mm]
201
Figure 11.4: Blade support for machining
202
Figure 11.6: 2 blades instrumented with Macor inserts carrying thin film gauges
203
0 deg
Sense of rotation
11 taps with
0.8372 deg spacing
(tip only)
α1
α2
α1 -23.93 deg
α3
rhu α2 -33.42 deg
b=
34 α3 -42.91 deg
4.5
rtip=3
mm
9
5.2 mm
0 deg
10 taps with α1 25.12 deg (3 stator pitches)
0.8372 deg spacing
α2 200.93 deg (24 stator pitches)
(tip only)
1
m
m
.4
39
α1
=3
ub
rtip=395.2 mm
rh
α2
2
Sense of rotation 10 taps with
0.8372 deg spacing
(tip only)
Figure 11.9: Static pressure taps for ps3 at tip in plane 1 downstream of the rotor
204
0 deg
1 o
2
8.5 o
5.65
o o
51 58.6
o
o 8.5
51
Sense of rotation
o
8.5 : ~ a stator pitch
o
5.65 :~ a rotor pitch
Axial position: 0.5 Cax,rotor downstream of rotor trailing edge
(view looking from downstream)
1.00
slope [V/bar]
0.90
0.80
0.70
0.60
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
0.2
0.1
Origin [bar]
0.0
-0.1
-0.2
-0.3
-0.4
-0.5
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Gauge position no
205
206
Part IV
207
Chapter 12
Time-averaged values
Although P01 is quite constant, Ps2 , P03 and Ps3 have a slight tendency to
increase during the test. An attempt was made to reduce the pressure increase but it
resulted in a significant decrease in the mean level of Ps3 ; as a result, the turbine was op-
erating below the design pressure ratio (Ps3 /P01 =0.327). Thus, it seems that imposing the
design pressure ratio results in a mass flow which is slightly too high resulting in a slight
increase of the pressure levels in the stage during the blowdown. In order to quantify this
increase, each point of the curves Ps2 , P03 and Ps3 as a function of time was first reduced
by the corresponding point in P01 ; it resulted that, on a 200 ms duration, Ps2 , P03 and
Ps3 increase respectively by 1.2%, 2.7% and 4.3% of their mean value. The portion of the
signal used for time-averaging and phase locked averaging does not exceed 100 ms. Over
this duration, the increase are then limited to 0.6%, 1.4% and 2.2%.
209
The upstream total temperature evolution T01 measured with a thermocouple
is presented in Fig. 12.2. The high initial temperature (30o C) prior to blowdown is due
to windage losses which heat-up the gas temperature in the test section. Due to a low
frequency response of the thermocouple, it is not possible to quantify accurately the tem-
perature variation during a test.
The coolant mass flow rate is controlled by a sonic throat placed between the
coolant reservoir and the vanes. Total pressure histories upstream and downstream of this
throat are reported in Fig. 12.3 together with the evolution of P01 and Ps2 . The coolant
flow ejection starts before the blowdown so that the coolant flow is well established during
the test. In the reservoir, the coolant flow is at ambient temperature but when it exits in
the test section, it is colder due to a sudden expansion.
Imposed upstream and downstream conditions as well as coolant flow rate were
maintained within a reasonable tolerance for all tests as shown. Time-average values were
computed on a 100 ms duration in the zone indicated by arrows in Fig. 12.1. This zone
coincides with the zone where the data acquisition at high sampling rate was performed
to record fluctuations at blade passing frequency. Table 12.1 reports the means of the
time-averaged values of 35 tests. The dispersion (20:1) is indicated after each value as a
percentage of the mean value. This dispersion expresses the bandwidth containing 95 %
of the values assuming the values are distributed according to a Gaussian law. Design
conditions are also reported in table 12.1.
P01 =1.620 bar, T01 =440 K, Ps3 =0.530 bar, ṁc =0.315 kg/s (3% of 10.5 kg/s).
All tests are very close to the imposed design conditions.
Case T01 [K] ±% P01 [bar] ±% Ps3 [bar] ±% ṁc [kg/s] ±% no of tests
6000 442.4 2.1 1.6271 2.0 0.5405 2.4 0.315 6.1 8
6500 441.6 2.0 1.6278 1.3 0.5325 3.6 0.316 4.9 10
6800 438.3 2.1 1.6301 1.6 0.5287 4.6 0.322 5.2 8
6500 no coolant 442.2 2.8 1.6237 1.8 0.5254 2.8 - - 10
design at 6500 440.0 - 1.620 - 0.530 - 0.315 - -
210
torque), the inertial momentum of the rotor is I[kgm2 ] = θ̇[rad/s2 ]/T [N.m] = 18.1kg.m2 .
With this momentum, the torque for the other rotational speeds can be computed as well
as the corresponding power P [W ] = T [N.m] ∗ θ[rad/s]. These are reported in table 12.2.
Although there is a slight increase of power with rotational speed, the change of power
does not exceed +-2% of the power at 6500 RPM whereas the rotational was varied from
-10% to +5% around the design value.
A noticeable increase of the mean level of Ps2 is observed when increasing the
rotational speed as shown in table 12.3. This can be explained as follows. There exists an
approximately linear relationship between the stage loading coefficient Ψ = ∆H u2
is
and the
∆his,rotor
degree of reaction r = ∆his,stage . An increase of Ψ causes a decrease of r and vice-versa.
As it was said previously, the power estimated from the rotor acceleration rate is quasi
constant therefore ∆his,stage is constant. Increasing the rotational speed decreases the
loading coefficient and increases the enthalpy drop across the rotor. To keep the overall
enthalpy drop constant, the exit Mach number in the stator decreases causing an increase
of the static pressure Ps2 .
Table 12.3: Intermediate static pressure Ps2 increase with rotational speed
211
12.1.4 Stage exit total pressure and temperature: P03 , T03
The downstream total pressure P03 was measured both by means of a pneu-
matic probe and a fast response probe. Two different locations were used but in both
cases, the probe axis was aligned horizontally with a vane leading edge. Measurements
were performed at mid-span only (r=367.3 mm). For each rotational speed, the probe an-
gle was set at the values indicated in table 12.4. Those values are derived from the velocity
triangle at design point. It was assumed that the stator exit velocity does not change with
rotational speed thus, the change of the rotor exit flow angle is only dependant on the
peripheral speed. Because the probe measuring location is distant from the probe axis by
7 mm, the measuring point is aligned with the probe axis and therefore the vane leading
edge only if the setting angle is 0. The circumferential shift of the measuring point as a
function of probe setting angle is reported in table 12.4. The variation of angles reported
do not exceed a variation of ±3% with respect to the value at 6500 RPM.
There is an excellent agreement between the pressure history of the two pneu-
matic probe and the fast response probe as shown in Fig. 12.6. The fast response probe
signal was low-pass filtered at 750 Hz prior to data acquisition in order to avoid aliasing
and the curve presented in the graph was smoothed for clarity. The frequency response of
the pneumatic probe is satisfactory. Mean values of the downstream total pressure mea-
sured with the pneumatic probe and the fast response probe are reported in table 12.5.
The exit Mach number computed from the measured P03 and Ps3 is also reported in the
same table as well as the design conditions at 6500 RPM.
A traverse across one pitch was performed with the fast response probe at
6500 RPM without coolant ejection. A reference level was given by the pneumatic probe
in a fixed position. The signal of the fast response probe was divided by the pneumatic
probe level for each data sample in order to eliminate pressure oscillations due to the
piston displacement and the slight pressure increase with time. Results are presented in
Fig. 12.7 and show a maximum variation of the mean value across a pitch of 5% (30 mbar).
212
The downstream total temperature T03 was measured by means of a thermo-
couple and a cold wire probe. The cold wire probe output was corrected for conduction
effect. The probe locations were the same as for P03 measurements. The cold wire probe
signal was low-pass filtered at 750 Hz prior to data acquisition. It is presented in Fig. 12.8
after smoothing for clarity. There is a good agreement between the probes time-averaged
values. The comparison of temperature history from the cold-wire probe and from the
thermocouple puts into evidence the low frequency response of the thermocouple. The
cold wire captures temperature fluctuations which can be attributed to oscillations of the
piston during blowdown. The temperature decrease which can be seen prior to blowdown
is due to the start of the coolant ejection. In this case, the temperature reaches a minimum
of 286 K. This low temperature level can be explained by the expansion of the coolant gas
in the evacuated test section. Mean time-averaged values of the ratio T03 /T01 computed
from the thermocouple and cold wire probe are presented in table 12.6; the design value
at 6500 RPM is 0.768 (without coolant).
Table 12.6: Mean values of downstream temperature T03 /T01 (design without coolant at
6500 RPM is T03 /T01 =0.768)
No clear tendency can be observed. Moreover, the coolant flow should be taken
into account when comparing design and measured values.. The coolant flow temperature
during blowdown cannot be measured in the test section for obvious reasons. It is diffi-
cult to estimate due to two successives expansions of the coolant flow, a first in the sonic
throat controlling the coolant mass flow and a second across the ejection slot. Assuming
a coolant temperature of 293 K and a coolant flow rate of 3%, the mean temperature of
the flow is T01 = 0.97 ∗ 440 + 0.03 ∗ 293 = 435.6K. This does not account for the cooling
of the main stream by the cold blades.
Fig. 12.10 compares the pressure rise with the temperature rise The ordinates
of the signals were reduced between 0 and 1, 0 being the level prior to blowdown, 1 during
blowdown. It is interesting to note that the temperature is established much faster than
the pressure. This is not surprising since the pressure is established only once the test
213
section is filled with the main stream and the downstream throat is choked.
214
12.2.3 Rotor blade surface pressures; influence of rotational speed, coolant
ejection and spacing
As it was mentioned before, the centrifugal force causes large drift in the gauge
outputs. The drift prior to blowdown is corrected by setting the pressure level at the pres-
sure level in the test section. However, during blowdown, both surface temperature and
centrifugal force increase and additional corrections are required.
The correction for temperature was applied only for two gauges (18 and 22) for
which the sensitivity of the origin of their calibration law is hundred times higher than for
the others. For the other gauges, the correction for temperature during a test is negligible
as shown in Fig. 12.11.
Regarding the effect of the centrifugal force during at test, one has to keep
in mind that the rotational speed increases typically from 6400 to 6800 RPM during a
blowdown. Using the correction law Pkul,corr − Ps3 = f (RP M 2 ) (see section on fast re-
sponse pressure measurements in the report concerning instrumentation) the drift due to
centrifugal force for gauge 12 (one of the most sensitive), amounts to 5.6 mbar namely 0.6
% of the design relative total pressure (0.927 bar). It can be concluded that for all gauges,
the correction due to a change of the centrifugal force during a test is negligible.
Measured relative pressure profiles at 6000, 6500 and 6800 RPM are shown
in Fig. 12.12. Gauge no 18 was eliminated for all tests because although the shape of
the signal seems correct and the calibration is repeatible, the mean value was too far
above the other gauges. Gauge 19 exhibited a strange steady-state response, although
the fluctuating component seems correct and could be used only in one test (6500 RPM
without cooling). Gauge 2 broke after 3 tests and is not available at 6800 RPM. Finally,
no output was measured out of gauge 21 probably due to a connection problem; 17 had
also connection problems but gave an output for two tests.
A cause for this discrepancy could be that the inlet relative flow angle is
far above the design. A 2.5 D Euler calculation showed that the relative inlet angle
should be at least 50.0 deg instead of 45.0 deg at 6500 RPM to justify such a difference.
215
The calculation also showed that only the front suction side is affected by the change of
incidence.
Two reasons can cause a higher relative inlet angle (due to the velocity trian-
gle):
- an absolute Mach number which is higher than design by 0.08 (M2,is 1.05− > 1.13)
- an absolute outlet angle which is higher than design by 2 deg (α2 72.3− > 74.3 deg)
Steady measurements behind the stator alone showed that for an absolute isentropic Mach
number of 1.05 at mid-span, the absolute isentropic Mach number at tip was 0.96 and that
the measured absolute outlet angle was close to the design angle. The same level of Mach
number being measured at tip in presence of the rotor, it is unlikely to have 1.13 at mid-
span. A measurement error as large as 2 deg. in the outlet angle is unlikely to occur.
The only possibility would be that the rotor influences the absolute exit flow angle. An-
other possibility which could explain this discrepancy will be presented when discussing
the fluctuating component of the pressure signals.
The blade pressure distributions at 6500 RPM with and without coolant ejec-
tion are shown in Fig. 12.14. The corresponding Mach number distribution is shown
Fig. 12.15. The acceleration on the front suction side seems to be slightly larger without
coolant ejection.
The blade pressure distributions at 6500 RPM with coolant ejection and spac-
ing of 0.35cs,ax and 0.50cs,ax are shown in Fig. 12.16. No clear influence of the spacing on
the distribution can be observed.
q̇ Crotor
Nu = (12.1)
(T0r − Twall )ka
In order to compute this Nusselt number, there are three physical quantities
to determine: relative total gas temperature T0r , surface temperature Twall and heat flux q̇.
w2 − v 2
T0r = T01 + (12.2)
2cp
216
T0r should be computed from a mixed-out stator exit total temperature in order to compare
the Nusselt number distribution in the cooled and uncooled case. Because this tempera-
ture is not known with accuracy, the relative total temperature is computed with T01 in
all cases. T01 is measured for each test and the relative inlet velocity is determined from
the velocity triangle with the design exit velocity and the peripheral speed corresponding
to the test. At design, (T01 = 440K, 6500 RPM) this temperature should be 377 K.
The mean temperature level of the blade just before the test depends on:
- the initial blade temperature level before run-up (it will be high higher than ambient if
a test was performed just before because the blade did not cool down completely).
-the level of rotational speed ( the windage losses increase with the square of the rotational
speed).
-the time needed to reach this level (if the run-up is longer, the blade temperature will
increase).
For those reasons, it is difficult to show a picture of the initial surface temper-
ature distribution (different levels depending on tests). However, a shape of this temper-
ature profile was interpolated by superimposing all the tests and is shown in Fig. 12.17.
A typical value for the mean blade surface temperature prior to the blowdown and the
injection is 45o C. Noticeable higher temperature levels can be seen in the nose region on
the pressure side and on both sides of the trailing egde. Those regions are directly exposed
to strong shear flows and recirculations.
Surface temperature histories of selected gauges are shown in Fig. 12.19. Prior
to a blowdown, the coolant ejection from the vane causes a temperature decrease which
is different for each gauge. The temperature decrease depending on gauge location is
shown in Fig. 12.18. The largest temperature decrease is observed at the leading egdge,
as expected. For the test without coolant ejection, the initial temperature level is directly
given by Fig. 12.17 whereas for the others, one need to substract the values in Fig. 12.18
to the values in Fig. 12.17.
Coming back to Fig. 12.19, a temperature rise of 25 K is observed at the
leading edge (gauge 14). It drops suddenly to a much smaller increase at gauge 13.
The flux is computed at each time step from the derivative of the inner blade
temperature profile computed with an implicit Crank-Nicholson scheme. One boundary
condition is the measured surface temperature rise, the inner boundary condition being a
constant
√ temperature at the inner side of the insert during the test. The thermal product is
ρck = 2073[J/(m2 Ks0.5 )]. The substrate thickness used in the calculation was adapted
to the duration of the acquisition (0.116 s instead of 1 s for the steady state component)
in order to gain accuracy. The heat fluxes computed from the surface temperature history
shown in Fig. 12.19 are plotted in Fig. 12.20. In order to minimise the error due to
the surface temperature increase during the blowdown (up to 25 K for gauge 14) in the
Nusselt number calculation, the flux is extrapolated at the start of the blowdown. A linear
regression is performed in a part of the heat flux history located after the transient and
before the end of the blowdown. The coefficients of the regression are used to compute
the value of the heat flux at the start of the blowdown. This is illustrated in Fig. 12.20
for gauge 2.
217
Nusselt number distribution along the blade mid-flow path; influence of rota-
tional speed, coolant ejection and spacing
Gauge 23 and 1 did not yield any output and are not reported. For the gauges
which were doubled (gauges 6 and 21), an average was done. The dispersion of the Nus-
selt number differences between the two measurements is ± 70 (20 : 1) for gauge 6 and
± 208 (20 : 1) for gauge 21.
Nusselt number distributions at the three different rotational speeds are shown
in Fig. 12.21. The highest level of heat transfer is observed at the leading edge as expected.
As the laminar boundary layer develops under a large favorable pressure gradient in the
nose region, the heat transfer rate decreases.
On the suction side, the dominant parameter seems to be the Mach number
distribution especially in the nose region. At 6000 RPM, a sudden increase of Nusselt
number is observed between gauge 4 and 5. This corresponds with a local deceleration in
the corresponding Mach number distribution (see Fig. 12.13) i.e. a local adverse pressure
gradient which promotes transition. As the flow reaccelerates (gauge 6, 7 and 8), the
boundary layer has a tendency to reverse back to a laminar state which explains that
the Nusselt number decreases again. A second local increase is observed between gauge
8 and 9 also corresponding to a local decceleration in the Mach number profile. The dif-
ferences in the Mach number distribution at 6000, 6500 and 6800 RPM were explained
by an incidence effect. At 6500 RPM, the decceleration occurs later than at 6000 RPM
(gauges 5 and 6) and is less steep. A corresponding increase in Nusselt number is observed
between gauge 5 and 6 and it is of less amplitude than at 6000 RPM. At 6800 RPM, there
is a quasi-continuous acceleration and no increase in the Nusselt number is observed. In
contrast, the Mach number distribution shows a decceleration between gauge 9 and 10
for all rotational speeds and the corresponding increase in Nusselt number for the three
rotational speeds.
On the pressure side, the level of Nusselt number is globally higher than on
the suction side. In contrast with the suction side, the pressure side boundary layer does
not benefit from the stabilising effect of the convex curvature. The effect of insulation of
a laminar boundary layer is not as good as on suctin side. A slight increase of the Nusselt
number with rotationnal speed can be observed. The pressure side Mach number distribu-
tion is usually less sensitive to incidence than the soon suction side. Thus, the increase of
the Nusselt number level with rotationnal speed cannot be attributed to a Mach number
effect. It can be better attributed to a Strouhal number effect which tends to increase the
free stream turbulence and enhance the overall level of heat transfer. This increase of free
stream turbulence is caused by the increasing frequency of the perturbations due to wakes
and shocks in the main flow.
At the trailing edge, the Nusselt number decreases on both side of the blade.
This can be attributed to a faster growth of the laminar boudary layer in this region due
to less curvature on suction side, less channel convergence on pressure side.
Although the shock and wake ingestion constitutes large destabilizing factors,
it seems that the boundary layer remains mainly laminar on the suction and pressure side
218
probably under the influence of a quite low Reynold number Rec = 0.5 106 .
The Nusselt number distributions with and without coolant at 6500 RPM are
presented in Fig. 12.22. The profiles are similar but the mean Nusselt number is much
lower in the case of no coolant. An explanation will be proposed when describing the
pressure fluctuating component.
The Nusselt number distribution at 6500 RPM with coolant ejection at 0.35 Cs,ax
and 0.50 Cs,ax is presented in Fig. 12.23. Some points are doubled or tripled for spacing
0.50 Cs,ax due to repeat tests and overlapping positions. The repetitivity of the measure-
ments is satifactory. The dotted line passes through the mean value of the repeat points.
The two distributions are quite close to each other.
219
12.3 Figures
6500 RPM 3% injection
2.0
1.8
P01
1.6
Averaging
1.4
Pressure [bar]
1.2
1.0
Ps2
0.8
P03
0.6
Ps3
0.4
0.2 Pdump
0.0
-0.3 -0.2 -0.1 0.0 0.1 0.2
Time [s] rot.043
460
440
420
400
Temperature [C]
380
360
340
320
300
280
260
-0.3 -0.2 -0.1 0.0 0.1 0.2
Time [ms]
Figure 12.2: Upstream total temperature evolution
220
6500 RPM 3% injection
3
Pressure [bar]
2
P01
Downstream of the
sonic throat
1 Ps2
0
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2
Time [s] rot.043
RPM=805.20*t[s]+6487.1
6700
6600
Rotational speed [RPM]
6500
6400
6300
-0.3 -0.2 -0.1 0.0 0.1 0.2
Time [s]
Figure 12.4: Evolution of rotational speed during a test
221
rot.016
0.58 pitch 1
pitch 2
pitch 3
Behind stator alone
0.56
Ps2/P01
0.54
0.52
0.50
0.0 0.2 0.4 0.6 0.8 1.0
Stator pitch
Figure 12.5: Static pressure at tip behind the stator
0.8
0.6
Pressure [bar]
0.4
0.2
0.0
-0.3 -0.2 -0.1 0.0 0.1 0.2
Time [s]
Figure 12.6: Comparison fast response probe, pneumatic probe for P03 measurement
222
rot.051 rot.052
1.020
1.010
P03 traverse / P03 fixed
1.000
0.990
0.980
0.970
0.960
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Stator pitch
Figure 12.7: Fast response pressure probe traverses behind the rotor
390
Thermocouple output
Compensated CW signal
370
350
Temperature [K]
330
310
290
270
250
-0.6 -0.4 -0.2 0.0 0.2
Time [s] cwcompset.032
223
cwr.051
500
450
Temperature [K]
400
350
cw.034 p03.034
1.6
1.4
1.2
1.0
0.8
0.6
0.0
-0.2
-0.3 -0.2 -0.1 0.0 0.1 0.2
Time [s]
224
rot.031 Tcorr.set
6
0.6
0.4
0.2
-1
-2
-3
-0.5 -0.3 -0.1 0.1 0.3 0.5
Time [s]
Figure 12.11: Temperature correction
kulp1.set
0.55
0.40
0.35
0.30
0.25
0.20
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s /s(13)
Figure 12.12: Pressure distribution at 6000, 6500 and 6800 RPM
225
mach.set
1.4
1.2 8 10
5 12
4 7
6
Isentropic Mach number
1.0 3
2
0.8
22
0.6
0.4 20
3D NS computation
16 6000 RPM
0.2 6500 RPM
6800 RPM
0.0
0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 12.13: Isentropic Mach number distribution at 6000, 6500 and 6800 RPM
kulp2.set
0.55
0.40
0.35
0.30
0.25
0.20
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 12.14: Pressure distribution at 6500 RPM with and without cooling
226
machc.set
1.4
1.2
Isentropic Mach number
1.0
0.8
0.6
0.4
3D NS computation
0.2 6500 RPM
6500 RPM without coolant
0.0
0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 12.15: Isentropic Mach number distribution at 6500 RPM with and without cooling
kulp3.set
0.55
0.50
0.45
ps/p01
0.40
0.35
0.30
6500 RPM spa=0.35*Cs,ax
0.25 6500 RPM spa=0.50*Cs,ax
0.20
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 12.16: Pressure distribution at 6500 RPM with cooling at 0.35 cs,ax and 0.50 cs,ax
axial spacing
227
55
Initial surface temperature [C]
50
45
40
35
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 12.17: Typical shape of initial temperature profile prior to blowdown and prior to
coolant ejection
6000 RPM
6500 RPM
4 6800 RPM
Temperature decrease [K]
1
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 12.18: Temperature decrease due to coolant ejection
228
6500 RPM
30
24
22
12
25 20
16 18 10
14 14
Surface temperature rise [C]
20 2 8
4
6
15
10 2
6
5
13
0
-5
-10
-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2
Time [s]
Figure 12.19: Surface temperature history (6500 RPM)
140000
120000
100000 14
Flux [W/m2]
80000
60000
2
40000
6
20000 13
-20000
-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2
Time [s]
Figure 12.20: Computed heat flux history (6500 RPM)
229
14
3000
6000 RPM
6500 RPM
6800 RPM
2000
Nu
2
1000 17
21 19 3
22 9
4 11
8
24 5
6 7
0
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 12.21: Nusselt number distribution at 6000, 6500 and 6800
2000
Nu
1000
0
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 12.22: Nusselt number distribution at 6500 RPM with and without coolant
230
14
3000 6500 RPM 0.35 Cs,ax
6500 RPM 0.50 Cs,ax
2000
Nu
16 2
20 18
10
1000 24
4 6 8
12
22
0
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 12.23: Nusselt number distribution at 6500 RPM with coolant ejection for spacings
0.35 Cs,ax and 0.50 Cs,ax
231
232
Chapter 13
For a given instrumented blade, the phase reference is chosen to be zero when
the stacking axis of the instrumented blade is aligned with the stacking axis of a vane.
A value of 1 means that the instrumented blade has completed an angular displacement
equal to 1 stator pitch. In this way, phase locked average signals coming from instru-
mented blades sitting at different positions on the rotor can be compared directly with
each other. In the litterature, the phase reference is sometimes chosen when the rotor
blade trailing edge is tangent to the stator leading edge. This sytem was not adopted in
prevision of full 3D surveys. With such a convention, the phase reference would depend
on the height of the instrumentation. In the proposed system, the stacking axis of blade
and vane are both radial, hence the phase reference will be the same at whatever radius
the instrumentation is located.
233
The phase reference for a stationary probe standing behind the rotor was cho-
sen to be zero when the rotor stacking axis is aligned with the nose of the probe at a probe
setting angle of 12 deg. (exit flow angle for nominal operating conditions). A value of 1
means that the probe has seen a complete rotor passage. With this system, phase locked
average signals coming from probes located at different circumferential positions can be
compared.
The electronic boards introduce a time delay in the signal. Regarding the
circuit which drives the infra-red diode giving the rotational speed, the time delay is neg-
ligible. Regarding the in-shaft boards for instrumentation, the time delay introduced by
the board driving the fast response pressure transducers does not exceed 0.5% of a blade
passing period and no correction will be applied for this delay. The time delay introduced
by the heat transfer board is corrected in the numerical demodulation.
The phase locked average routine takes into account the evolution of the ro-
tational speed. As it was said previously, the evolution of the rotational speed is fitted
by a linear law during the 0.116 s of acquisition at high sampling frequency. During one
stator passage traverse, it is assumed that the rotational speed is constant. Given an
initial time, the mean value of the rotational speed on the traverse is used to determine
the final time of the traverse. This is repeated for N sucessive traverses. Then, for each
traverse, the time axis is reduced between 0 (initial time) and 1 (final time or initial time
of the next traverse). The phase locked average time axis is defined in NCLA classes
between 0 and 1. The number of classes is usually chosen as the mean number of sampled
points per vane passage. Finally, every point of every traverse is classified into one of
the classes. Average and RMS can be computed in each class. The advantage of this
method compared to methods which interpolate the signal is that the signal is kept as
it was sampled; in this way, more confidence can be attributed to the RMS values. On
the other hand, the points which are in a class were not sampled exactly at the same phase.
The time axis can be related to the space axis through the radius and the
rotational speed. The period at which a blade passing event repeats itself can be con-
verted into one stator or rotor pitch. In the following, pitch or time period will be used
234
indeferently.
Another interesting output of the phase locked average routine is the correla-
tion coefficient. This correlation coefficient is computed as it is usually done for a linear
regression. The abcissa is the phased-locked average signal periodically repeated for the
N successive traverses and the ordinate is the raw signal. It measures the divergence of
the raw signal from the mean period.
235
236
Chapter 14
There exist a quite good correlation between the raw measurements and the
phase locked average as shown in Fig. 14.2. The correlation coefficient corresponding to
this case is 0.91.
The mean period is not changing very much depending on the number of peri-
ods used in the phase locked average as shown in Fig. 14.3. Notice that small oscillations
which at a first glance look like noise are remaining in the phase locked average signal
showing that they are real features of the signal. In the following, 129 traverses (3 rotor
full revolutions) will be used in all phase-locked averaged signals.
The sharp increase in the phased-locked averaged signal can most probably be
attributed to the stator trailing edge shock. In the pitchwise RMS distribution, a spike
associated with the pressure discontinuity can be observed (see Fig. 14.6). The spike is em-
bedded in a larger but lower amplitude fluctuation. The location of the middle of the sharp
rise is at 0.966 of a stator pitch. Fig. 14.7 shows a 2D cut of the stage at mid-span. Lines
were plotted to represent the wake centerline and the shock system. The wake centerline
is drawn at an angle of -72.3 deg with respect to the axial direction (flow angle measured
behind the stator see [5]). The shock which goes towards the rotor blade row is drawn with
237
an angle of +20 deg. This angle was measured on a Schlieren picture from [5]. The lines
stop at 0.35.Cstator,axial which corresponds to the axial spacing between the stator and
the rotor blades. Of course, this sketch assumes that there is no influence of the rotor on
the stator flow field measured previously behind the stator without the downstream rotor.
It is interesting to note that for this particular spacing, shock and wake converge into a
same point in the rotor leading edge plane. The large low amplitude fluctuation in the
RMS distribution could be associated with the wake. The rotor blade carrying number 1
is the instrumented blade. It is shown in position 0 with respect to the stator; the stacking
axis of vane 1 and blade 1 are aligned. When blade 1 axis will be at 0.910 of a pitch, it
will meet the shock. This estimation coincides quite well with the measured position 0.966.
As for the other measurements, the number of periods in the phase locked
average does not change significantly its shape above 100 periods.
The probe has a rather limited frequency response and requires the use of a
numerical compensation which consists in amplifying the high frequencies. For the given
flow conditions the probe natural cut-off frequency is about 2 kHz. As the roll-off is similar
to a first order, after the cut-off, the amplitude of the response decreases by -20dB per
decade, i.e. at 20 kHz, the amplitude of the response is one tenth of what it was at 2
kHz for a same input signal. For this reason, the signal was low-pass filtered at 20kHz
so that the compensation does not amplify the signal by a factor bigger than 10. The
correlation coefficient computed with the phase-locked average of the non-compensated
signal low-pass filtered at 20kHz is 0.6. This rather poor coefficient can be explained by
the fact the the signal to noise ratio is not very large due to the low frequency response
of the probe. The correlation coefficient for the phase-locked average of the compensated
signal is 0.5. This decrease can be attributed to the amplification of the high frequency
components by the compensation. With increasing frequency, the signal to noise ratio be-
238
comes poorer and poorer. The compensation amplifies both noise and signal. The quality
of the electronic boards and the pick-up of external noise by the wire are very important
factors which influence the correlation coefficient.
Nevertheless, the test repetitivity is quite good as shown in Fig. 14.11 where
the phase locked average signal of three different tests under the same conditions are pre-
sented.
The phase-locked averaged signals for the different rotational speeds are plot-
ted in Fig. 14.12. Tests aith and without coolant ejection are presented in Fig. 14.13. A
slightly larger fluctuation is observed without coolant ejection.
239
the calculation.
The computations of the relative pressure at 6800 RPM taking into account
only the pitchwise distribution of static pressure (P02 , T02 , α2 are left constant in a first
step) are presented in Fig. 14.17 for two cases: 1) static pressure distribution from the
Euler computation, 2) static pressure interpolated from measurements at the endwalls. A
comparison is established with the measured profile at 6800 RPM. A very good agreement
is obtained for the shape of the distribution as well as for for the location of the shock when
using the computed static pressure distribution. Using the measured distribution lead to
a smeared out profile. Fig. 14.18 shows the computed profiles including sucessively the
pitchwise distribution of total pressure, total temperature and angle. It appears clearly
that the overall shape of the total pressure profile results mainly from the static pressure
distribution. The next most influencing profile is the total pressure. Total temperature
and angle distribution have negligeable effects.
240
-It seems that the shock position is little affected by the rotor (computed and measured
location coincide).
-The wake is probably distorded and deflected in the rotor inlet flow and its is very difficult
to identify its position and shape in the relative total pressure profile.
14.3.3 Computed relative inlet total temperature and relative inlet an-
gle
The computed relative inlet total temperature profiles at 6800 RPM are presented in
Fig. 14.21 including successively the pitchwise profiles of absolute total pressure, total
temperature and angle in their original angular position. It is mainly the inclusion of the
total temperature profile in the calculation which modifies the relative total temperature.
The total pressure contributes to the resulting profile to a lesser extent. The influence of
the distribution of angle is negligeable. A comparison between the measured relative total
temperature profile and the computed one shifting the abcissa of the absolute total tem-
perature and pressure by the same empirical amount than in the previous section (-0.55
of a pitch) is presented in Fig. 14.22. The phase of the two traces cannot be compared
because the relative total temperature was measured on the suction side of the blade (at
position 4) while the calculation is performed at the nose of the blade. However, the two
traces show similarities in their shapes as well as in the order of magnitude of the fluctu-
ation (of the order of 25 K!).
The computed relative inlet angle profiles at 6800 RPM are presented in
Fig. 14.23 including successively the pitchwise profiles of absolute total pressure, total
temperature and angle in their original angular position. Absolute temperature profile
has little effect on the computed relative inlet angle. The contribution of absolute total
pressure and absolute angle is more significant.
241
14.4 Figures FFT of P0r/P01 from probe in the nose
-1 rot.018
10
-2
10
Amplitude
-3
10
-4
10
-5
10
0 20000 40000 60000 80000
Frequency [Hz]
Figure 14.1: FFT of pressure signal P0r/P01 from probe in the blade nose
rot.016
0.06
0.04
0.02
P0r/P01
0.00
-0.02
-0.04
Phase locked average
Raw measurements
-0.06
0.4015 0.4020 0.4025 0.4030
Time [s]
242
rot.016 planper.016
43 passages (1 revolution)
129 passages (3 revolutions)
258 passages (6 revolutions)
0.04
0.02
P0r/P01
0.00
-0.02
-0.04
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.3: Influence of the number of periods in the phase locked average
0.02
P0r/P01
0.00
-0.02
-0.04
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.4: Repetitivity of measurements
243
6800 RPM
rot.031 rot.018 repp0r14.set
0.08
Gauge at position no 14
Fast response probe in the nose
0.06
0.04
P0r/P01
0.02
0.00
-0.02
-0.04
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.5: Total and static pressure at stagnation point
6500 RPM
rot.016 plarms.set
0.04
0.02
P0r/P01
0.00
-0.02
244
Difference between blade axis (1-1): 0.0
-0.5
+20
Peripheral direction [stator pitch]
0.0
-72.3 1
1
0.5
24
22
1.0
20 11
16 18
1 9
3
1.5 7
5
2.0
0.0 0.5 1.0 1.5 2.0
Axial direction [Stator axial chord]
Figure 14.7: Schematic of shock-wake system at vane TE
rot.015 16 18 plap0ra.set
0.08
6000 RPM
6500 RPM
0.06 6800 RPM
0.04
P0r/P01
0.02
0.00
-0.02
-0.04
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.8: Influence of rotational speed
245
plap0rc.set
0.08
6500 RPM
6500 RPM no coolant
0.06
0.04
P0r/P01
0.02
0.00
-0.02
-0.04
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.9: Influence of coolant
p0r3.set
0.08
Probe in the nose
Gauge at position 14
0.06 RMS from probe
P02r/P01 (fluctuation)
0.04
0.02
0.00
-0.02
-0.04
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.10: Phase locked averaged for the spacing 0.50 Cs,,ax
246
All tests 6500 RPM without coolant
cwrnoc.set
rot.052
0.04 rot.047
rot.049
0.02
T02r/T01
0.00
-0.02
-0.04
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.11: Test repetitivity for cold wire probe
cwrall.set
6000 RPM
6500 RPM
6800 RPM
0.04
T02r /T01 (fluctuations)
0.02
0.00
-0.02
-0.04
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.12: Phase-locked average signals of cold wire probe at 6000, 6500 and 6800 RPM
247
cwrallc.set
6500 RPM
6500 RPM no coolant
0.04
Fluctuations of T02r /T01
0.02
0.00
-0.02
-0.04
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.13: Phase-locked average signals of cold wire probe at 6500 RPM with and
without coolant ejection 0.25 C s,ax
1.00
0.95 P02/P01
T02/T01
SS PS -a-
0.90
0.55
Ps2/P01
0.50
0.45
-b-
0.40
α2 [deg]
73
72
-c-
71
0.0 0.5 1.0 1.5 2.0
Angular position [pitch]
Figure 14.14: Total pressure, flow angle and static pressure profile behind the stator at
mid span M2,is = 1.05, cm = 3%, x/cax = 0.25
248
0.25 Cax
1.04
0.65 Euler computation 1.02
Measurements
0.60 1.00
P0/P01
Ps/P01 0.55 0.98
0.96
0.50
0.94
0.45 0.92
0.40 0.90
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
Angular Position [pitch] Angular Position [pitch]
1.30 74.0
1.10 72.0
1.00 71.0
0.90 70.0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
Angular Position [pitch] Angular Position [pitch]
Y/h=0.5 X/Cax=0.25
8
1.00
7
6 P02/P01
0.95
5
P02/P01
0.90
Tu(%)
4
Tu(%)
3 0.85
2
0.80
1 Continuous traverse
Probe in fixed position 0.75
0
0.0 0.5 1.0 1.5 2.0
Angular position [pitch]
Figure 14.16: Turbulence level in the wake behind the Brite stator at 0.25 Cax
249
6800 RPM
abre0.set
0.10
0.05
0.00
P0r/P01
-0.05
-0.15
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.17: Computation of relative total pressure from stationnary measurements at
6800 RPM; comparison with measured P0r
abre1.set
0.10
0.05
P0r/P01 (fluctuation)
0.00
-0.05
250
6800 RPM
0.10
0.05
0.00
P0r/P01
-0.05
Measured distribution
-0.10 Computed using original P02 profile
Computed using P02 profile with a phase delay of 0.20
Computed using P02 profile with a phase delay of 0.45
-0.15
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.19: Computation of relative total pressure from stationnary measurements at
6800 RPM: translation the wake position with respect to static pressure by -0.55 of a pitch
Velocity Vector Plot
251
abre3.set
0.04
0.02
T02r/T01 (fluctuation)
0.00
-0.02
-0.04
Ps2 variable
Ps2, P02 variables
-0.06 Ps2, P02, T02 variables
Ps2, P02, T02, α2 variables
-0.08
0.0 0.5 1.0 1.5 2.0
Phase
252
abre4.set
0.04
0.02
T02r/T01 (fluctuation)
0.00
-0.02
-0.04
-0.08
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.22: Computation of relative total temperature from stationnary measurements
at 6800 RPM: translation of the wake position with respect to static pressure by -0.55 of
a pitch
abre5.set
10.0
5.0
α2r (fluctuation) [deg]
0.0
-5.0
-10.0
Ps2 variable
-15.0 Ps2, P02 variables
Ps2, P02, T02 variables
Ps2, P02, T02, α2 variables
-20.0
0.0 0.5 1.0 1.5 2.0
Phase
Figure 14.23: Computation of relative inlet angle from stationnary measurements at 6800
RPM including pitchwise variation of pressure, temperature and angle
253
254
Chapter 15
The minimum and maximum of the phase-locked average, the mean RMS over
one pitch and the correlation coefficient at 6500 RPM are plotted in Fig. 15.2 for all gauges
as a function of the curvilinear abcissa s/s(13).
Phase locked averaged static pressure fluctuations at 6500 RPM around the
blade mid-section are presented under the form of a carpet plot in Fig. 15.3. The value
presented is Ps/P01. The abcissa is the curvilinear coordinates s/s(13), the ordinate is a
time period or a pitch. The dotted lines indicate negative values whereas plain lines indi-
cate positive values. For clarity, the spacing between the isolines is not constant. Bewteen
±0.025 , the step is 0.010; after the step is 0.020. This allows to show on the same plot
fluctuations of large amplitude in the nose region and fluctuations with smaller amplitudes
elsewhere.
In order to analyse the signal in more detail, several graphs of the type of
Fig. 15.4 will be presented in order to describe the entire blade surface. The gauge sig-
nal will be described successively starting on the pressure side at the trailing edge going
towards the leading edge (gauges 24 to 14) and then on the suction side from the leading
edge to the trailing edge (gauges 1 to 13). Fig. 15.4 shows the time traces of gauges 24 to
16 over two periods. The mean value of the fluctuations calculated from the phase-locked
average routine is zero. For clarity a difference of mean level between two traces was in-
troduced. This difference is proportionnal to the curvilinear distance between the gauges;
this was done in the same way for all the graphs which will be presented later.
255
amplitude of the fluctuations decreases slightly from gauge 24 to gauge 19 as well as the
RMS level and the correlation level (see Fig. 15.2). The correlation level stays quite high
for those gauges compared with gauges of the late suction side for example. Gauge 18
displays very small fluctuations, small RMS and local mimimum in correlation. There is
a clear change of the shape of the fluctuations across this gauge which is located in the
concave part, probably hidden from a direct shock influence. In one period, gauge 16
presents two double spikes patterns embedded in a larger fluctuation.
Time traces of gauges 16 to 5 (front pressure and suction side) are plotted
in Fig. 15.5. The pattern of gauge 15 is very similar and is almost in phase with gauge
16. Across one pitch, a pair of double spikes embedded in a larger fluctuation can be
observed. The pattern of the signals from gauge 14 to 4 is different. They all exhibit large
levels of fluctuation. The amplitude of the fluctuation is maximum at gauge 4 as well as
the RMS mean level and the correlation correlation coefficient which is very close to 1.
The amplitude of the static pressure fluctuation (peak to peak) amounts to 40% of the
theoretical relative inlet total pressure (0.927 bar). The amplitude of the fluctuation is
divided by more than two when going from gauge 4 to gauge 14. For all those gauge, the
pattern is quite similar: a large fluctuation per period beginning with a steep increase,
immediatly followed by a short double-peak patterns. Then, after a short decrease, a
plateau can be observed. This plateau gets larger and larger when going from gauge 4 to
gauge 14. At gauge 5, the pattern has changed noticeably: the sharp pressure increase
has given way to a smoother and moderated pressure rise. Gauge 6 (see Fig. 15.6) has a
similar shape and amplitude but seems to be in phase opposition with gauge 5. For both
gauges, not only the amplitude has decreased but also the RMS level and the correlation
coefficient. The patterns of gauges 7 and 8 are similar to the ones observed for gauges 24
to 19 on the pressure side. Two fluctuations of similar amplitudes are observed over one
pitch. These two gauges show a minimum in RMS whereas the correlation coefficient is
still high; this was also the case for gauges 24 to 19 on the pressure side. For gauges 9 to
13, the amplitude continue to decrease. The double fluctuation observed in gauges 7 and
8 is not so evident anymore and only a single fluctuation of decreasing amplitude can be
observed. At trailing edge, the fluctuation of pressure can hardly be seen. Whereas the
RMS level increases, the correlation decreases and becomes very poor at the trailing edge
(0.3 only at gauge 13). Albeit the mean RMS levels are similar as for gauges 1 to 4, the
pitchwise distribution is completely different. For gauge 1 to 4, there is a sharp spike in
RMS whereas for gauges 9 to 13, the RMS distribution is quasi constant.
256
observed. If there is an influence without impact, no spike in the RMS is observed. The
pitchwise RMS distribution corresponding to Fig. 15.4, 15.5, 15.6 are respectively plotted
in Fig. 15.9, 15.10, 15.11. The term crown will be often used; it is defined as the point of
the blade surface which is tangent to the supposed direction of the shock. A 2.5 D Euler
unsteady calculation performed by Giles [60] on a transonic turbine is used as an help
for the interpretation of the measured unsteady pressure traces. The calculation was run
with quite similar aerodynamic conditions, with the same (stator pitch)/(rotor pitch) ratio
(2/3), a similar axial spacing (0.31 ∗ Cs,ax ) instead of 0.35) but on a different geometry.
A schematisation of the shock position at different times resulting from this calculation
is presented Fig. 15.12. At t=0.0, the position of the rotor with respect to the stator
is slightly different in Giles case and in this case but an equivalence can be established:
tGiles = tBrite + 0.35 (referred to as tG and tB in the following).
The shock hits first the crown of the blade (tG=0.875 in Fig. 15.12). It im-
pinges almost parallel to the surface about the position of gauge 5 at tB=0.55 Fig. 15.5.
No spike can be observed in the pitchwise RMS distribution (Fig. 15.10) probably due to
the incidence of the shock with respect to the surface. Then the shock sweeps the front
suction side from the crown down to the leading edge i.e. in the a sense opposite to the
convection of the flow along the blade surface.
From gauge 4 to gauge 14, the pitchwise RMS distribution suggests that the
shock impinges directly on the blade surface and causes a steep gradient in the pressure
distribution. The steep rise is always followed by a double peak features which is not fully
understood. It could be attributed to the shock reflection and the existence of a recircu-
lation bubble. The same double peak was observed by Dietz et al. [50] but in their cas,
it seems to be located in the steep rise. The amplitude of the steep rise is maximum at
gauge 4 and decreases when going toward leading edge. An inverse phenomena is observed
in the calculation of Giles (see [60]) and in measurements from Rao et al. [61] (in both
case, stator exit Mach number was 1.12 which is slightly haiger than here). For these
gauges, the raw signals exhibits a very good correlation with the phase-locked averaged
signal (see Fig. 15.2) probably because the shock imposes directly a very strong, large
periodic fluctuations. The large amplitude of the fluctuation justifies the high amplitude
of the mean RMS level.
The path of the shock in the soon suction side can be guessed by locating the
abcissa of the end of the steep rise successively from gauge 4 to 14 (it could also be done
with the spike in the pitchwise RMS distribution). This path is compared in Fig. 15.8
with a path computed from results of the Brite phase 1. Assuming the shock at an an-
gle of about 20 deg. with respect to axial direction and given the rotational speed and
the location of each gauge, the position of the rotor with respect to the stator can be
determined each time the shock impinges on a gauge (see Fig. 15.7). Of course, this path
assumes that there is no influence of the rotor on the shock position which is probably not
the case. However, the comparison of the estimated path with the measured path gives a
reasonnable agreement.
Coming back to Fig. 15.5, one can observe that the double peak which is ob-
served only once in a pitch at gauge 1 is observed clearly twice per pitch for gauges 14,
257
15 and 16. Corresponding spikes in the pitchwise RMS distribution can be observed (see
Fig. 15.10). The first spike can be attributed to the direct impact of the shock at tB=0.95,
the second spike is distant by 0.6 of a period. The calculation of Giles [60] shows that the
leading edge is influenced alternatively by the direct impact of the shock and the reflected
shock of an adjacent blade (see t=0.375 in Fig. 15.12). From this graph, one can see that
the blade which feel the direct impact in the nose at tG=0.375 will feel the reflected shock
a rotor pitch later, namely 0.66 of a period which corresponds quite well with what is
observed here.
Gauge 18 exhibit almost no fluctuation (Fig. 15.4) and the pitchwise RMS
profile is flat Fig. 15.9. A local minimum is observed in amplitude, mean RMS and corre-
lation (Fig. 15.2). This can be explained by the fact that the gauge is located in a concave
part where the shock cannot access. This is not the case in Giles representation where the
shock is curved due to the local flow velocity and reaches the concave part (tG=0.625 -¿
tG=0.750).
In Giles case, the shock sweeps the pressure side from the leading edge down to
almost the trailing edge between tG=0.50 and tG ¡ 0.875. Then, the reflected shock sweeps
the pressure side from downstream to upstream (tG=0.0 -¿ tG ¡ 0.50). The pressure plot
of Giles shows two pressure waves, one of large amplitude due to the shock, one of smaller
amplitude due to the reflected shock, which cross each other. Here the downstream gauges
(22 to 24), exhibit a double fluctuation per period. The first fluctuation seem in phase
for all gauge whereas the second fluctuation has a tendency to get closer and closer to
the first when going downstream. This mechanism is not yet understood. An unsteady
computation on this geometry would probably help in understanding this phenomena.
Coming back to the suction side, one can notice that gauge 5 and 6 are in
oppostion of phase. The fluctuation at gauge 5 was explained by the effect of the shock.
The fluctuation of gauge 6 could be explained by the effect of the reflected shock drawn in
Fig. 15.12 at tG=0.375. The direct shock hitted the crown at tG=0.875 i.e. half a period
before which can justify the phase oppostion of the signals 5 and 6. Downstream of this
position, Giles does not observe fluctuations, justified by the fact that this portion of the
blade is hidden from both direct and reflected shock.
In contrast, gauges 7, 8 and 9 exhibit two fluctuations per period. Their levels
of amplitude, correlation and mean RMS (for gauge 7 and 8 only) (Fig. 15.2) is similar
to the one observed for gauges 19 to 24. Moreover, the phase of the these fluctuations is
identical to the phase of gauges 19 and 20 which are almost at the same axial position on
258
pressure side. This suggest that the same phenomena is affecting these gauges and gauges
19 to 24 on pressure side.
259
good. The shock generated unsteadiness could be the cause of this discrepancy. In terms
of time-averaged distribution, there exists a large favorable pressure gradient in the nose
area of the rotor blade. This type of gradient is well known to be a stabilizing factor of
the boundary layer. When the shock passes, a strong adverse pressure gradient which
causes the discontinuity in the Mach number distribution like at t=1/4 Fig. 15.13. Such
discontinuity is likely to cause separation with a recirculation bubble attached to the shock
wave. This recirculation bubble would sweep the front suction side attached to the shock
motion and modify quite strongly the flow path. The “mean blade geometry” seen by
the flow would be modified with respect to the real blade geometry resulting in a larger
curvature in the nose. This larger curvature could justify an overacceleration of the flow
with respect to the prediction.
Fig. 15.14 compares the time traces of selected gauges at the three rotation-
nal speed 6000, 6500 and 6800 RPM. Nor the shapes not the amplitude of the pressure
fluctuations seems to be significantly affected by the rotational speed for these gauges.
Comparison of the minimum-maximum, mean RMS and correlation coefficient at 6000,
6500 and 6800 RPM can be found in Fig. 15.17. The amplitude of the fluctuations has
a slight tendency to increse with the rotational speed, particularly on the late pressure
side. The carpet plots corresponding to 6000 RPM 6500 RPM and 6800 RPM are shown
respectively in Fig. 15.15, 15.3, 15.16, A concentration of isolines appears clearly on the
front suction side due to the shock sweeping. By plotting a line across this region of
concentration of isolines for the three rotationnal speed, one can show that the slope of
the line (d(period)/d(curvilinear abcissa) decreases namely the front suction side is swept
faster (more frequently) at 6800 RPM than at 6000 RPM.
The carpet plots corresponding to 6500 RPM without coolant ejection is shown
in Fig. 15.18. Comparison of the minimum-maximum, mean RMS and correlation coeffi-
cient at 6500 RPM with and without injection can be found in Fig. 15.19. The general
tendency is to have slightly lower amplitudes, lower mean RMS and lower correlation co-
efficient without coolant. This probably causes a lower free stream turbulence which could
explain partially that the Nusselt number distribution is much lower without coolant than
with coolant (see Fig. 12.21). The lower amplitude of the fluctuations is more striking in
the leading edge area (gauges 1,2,3,4) where the wake (which contains or not the coolant)
hits first.
The carpet plots corresponding to 6500 RPM with coolant ejection and a
spacing of 0.50 Cs,ax is shown in Fig. 15.18. Comparison of the minimum-maximum, mean
RMS and correlation coefficient at 6500 RPM with coolant ejection at 0.35 Cs,ax and
0.50 Cs,ax can be found in Fig. 15.21. The amplitude of the fluctuations clearly decreases
on the soon suction side (gauges 2,3,4). This maybe due to the smearing of the static
pressure profile with the axial distance. On pressure side, the fluctuations remain at the
same level.
260
15.5 Figures
-2
10
Amplitude
-3
10
-4
10
-5
10
0 20000 40000 60000 80000
Frequency [Hz]
261
kmrc65.set
4
Min Max [P/P01]
2
0.10
6
22 20 16 8 10
18 12
0.00
-0.10
0.008
mean RMS
0.006
0.004
1.0
0.8
Correlation
0.6
0.4
0.2
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 15.2: Minimum, maximum of mean period, mean RMS and correlation coefficient
of pressure at 6500 RPM with coolant ejection and a spacing of 0.35 Cs,ax
262
Figure 15.3: Carpet plot of pressure fluctuations on the blade surface as a function of time
at 6500 RPM
kxm165.set
0.4
16
0.3 18
24
19 22
12
20
P/P01
20 16 18 10
0.2 14
2 8
4
6
22
0.1
23
24
0.0
263
kxm265.set
0.75
5
0.65
4
3 2
1 24
22
0.55 12
14 20
P/P01
16 18 10
14
2 8
15
0.45 4
6
16
0.35
0.25
0.0 0.5 1.0 1.5 2.0
Phase
kxm365.set
13
0.85 12
11
10 24
22
0.75 9 12
20
P/P01
8 14
16 18 10
7 2 8
0.65 6 4
6
5
0.55
0.45
0.0 0.5 1.0 1.5 2.0
Phase
264
Difference between blade axis (1-1): 0.0
-0.5
+20
0.5
24
22
1.0
20 11
16 18
1 9
3
1.5 7
5
2.0
0.0 0.5 1.0 1.5 2.0
Axial direction [Stator axial chord]
mancorr.set
1.0
Phase
0.8
0.6
0.4
-0.1 0.0 0.1 0.2 0.3 0.4
s/s(13)
Figure 15.8: Shock path from estimated from unsteady measurements and from steady
shock pattern
265
kxmr165.set
0.10
16
0.08
18
0.06
19
RMS
20
0.04
22
0.02 23
24
0.00
0.0 0.5 1.0 1.5 2.0
Phase
kxmr265.set
0.16
0.14
0.12 4
3
RMS
2
0.10 1
14
15
0.08 16
0.06
0.0 0.5 1.0 1.5 2.0
Phase
266
kxmr365.set
0.19
13
12
0.17 11
10
RMS
9
0.15 8
7
6
0.13 5
0.11
0.0 0.5 1.0 1.5 2.0
Phase
267
Figure 15.12: Shock pattern evolution during one rotor blade cycle from Giles et al. [60]
268
ϕ=0 or ϕ=1 ϕ=1/8 ϕ=1/4 ϕ=3/8
1.2 8
12
dependant
4
6
0.8 22
Mach number
0.4 20
18
269
1.2
0.8
Mach number
0.4
0.0
0.0 0.5 0.0 0.5 0.0 0.5 0.0 0.5
Figure 15.13: Unsteady Mach number distribution; plain line: mean, dotted line: time
ϕ=7/8 ϕ=3/4 ϕ=5/8 ϕ=1/2
Gauge 2 Gauge 6
0.15
0.10
0.05
P/P01
0.00
-0.05
-0.10 +@
Gauge 4 Gauge 12
0.15
0.00
-0.05
-0.10
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
Phase Phase
270
Figure 15.15: Pressure fluctuations on the blade surface as a function of time at 6000
RPM
Figure 15.16: Pressure fluctuations on the blade surface as a function of time at 6800
RPM
271
kmrc.set
4
2
Min Max [P/P01]
0.10
20 6
22 18 16 8 10 12
0.00
-0.10
0.008
mean RMS
0.006
0.004
1.0
0.8
Correlation
0.6
6000 RPM
6500 RPM
0.4 6800 RPM
0.2
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 15.17: Minimum, maximum of mean period, mean RMS and correlation coefficient
of pressure at 6000, 6500 and 6800 RPM with coolant ejection
272
Figure 15.18: Pressure fluctuations on the blade surface as a function of time at 6500
RPM without coolant ejection
273
kmrc65s.set
4
Min Max [P/P01]
2
0.10
6
22 20 8 10
18 16 12
0.00
-0.10
0.008
mean RMS
0.006
0.004
1.0
0.8
Correlation
0.6
0.2
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 15.19: Minimum, maximum of mean period, mean RMS and correlation coefficient
of pressure at 6500 RPM with and without coolant ejection
274
Figure 15.20: Pressure fluctuations on the blade surface as a function of time at 6500
RPM with coolant ejection and a spacing of 0.50 Cs,ax
275
kmrc652.set
4
Min Max [P/P01]
2
0.10
6
8
22 20 10
18 12
0.00
-0.10
0.008
mean RMS
0.006
0.004
1.0
0.8
Correlation
0.6
0.2
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 15.21: Minimum, maximum of mean period, mean RMS and correlation coefficient
of pressure at 6500 RPM with coolant ejection for spacings 0.35 Cs,ax and 0.50 Cs,ax
276
Chapter 16
277
16.2 Influence of the shock wave on the Nusselt number
distribution
The carpet plot of the Nusselt number fluctuations around the blade mid-
section is presented in Fig. 16.5. The time traces are presented individually for all gauges
in Fig. 16.6 16.7 16.8 in a similar way than it was done for the pressure traces.
Gauges 24 to 19 (Fig. 16.6) exhibit two fluctuations per pitch as for pressure.
For these gauges, the signals seem to be in phase. There is a change of behaviour from
gauge 18 on. Only one fluctuation per period can be observed for gauges 18, 17, 16. Levels
of correlation and RMS are less significant than in the case. Gauges 16 to 14 (Fig 16.7) also
exhibit a single flucutation per period and have similar shape. A plateau is maintained
on most of the pitch.
The shape changes for gauges 2 to 4. The coincidence of the location of the
steep rise in pressure and Nusselt is even more striking for these gauges. A comparison
between pitchwise fluctuations of Nusselt number and surface pressure is shown Fig. 16.9
for gauge 2. The double spike feature is present in both signals. In this region, the direct
shock influence appears clearly with a steep rise which occurs at the same position than
for the pressure. The amplitude of the fluctuation is maximum for gauge 4 (Fig. 16.4)
and represents 44% (peak to peak) with respect to the mean value at the stagnation point
(gauge 14). This region of high amplitude fluctuation can be clearly identified on the
carpet plot as well as the path of the shock on the blade surface from the crown to the
stagnation point as a function of time. In the simulation of Doorly [31] (rotating bars in
front of a straight cascade, exit Mach number 0.96) “the direct impingment of the shock
wave on the blade produces little effect” on the heat transfer rate. A small perturbation
travelling with the shock from the crown down to the leading edge can be seen and in some
cases, a reciculation bubble visualised. However, the heat transfer rate is mostly affected
by a large recirculation bubble created close to the stagnation point which is convected
downstream the suction side while the shock clears out the nose region. This difference
maybe explained by the high Reynolds number in Doorly’s experiment: 2.0 106 instead of
0.5 106 here. The shape of the fluctuation is also quite different from some other experi-
ments in the litterature. In the case of Guenette for example, an intense spike embedded
in a fluctuation of lower amplitude is seen at leading edge. In his case, the stator exit
Mach number is much higher (1.18).
278
16.3 figures
htsurf.set
0.04
Gauge 2
Gauge 4
0.03
Gauge 10
Gauge 13
0.02
Surface temperature [K]
0.01
0.00
-0.01
-0.02
-0.03
-0.04
0.0 0.5 1.0 1.5 2.0
Phase
Figure 16.1: Phase locked averages of surface temperature from selected gauges
279
gauge 2
1500
43 passages
129 passages
258 passages
1000
Nu
500
-500
0.0 0.5 1.0 1.5 2.0
Phase
200
Nu
-200
rot.034 gauge 406
rot.039 gauge 506
-400
400
200
Nu
-200
rot.039 gauge 221
rot.043 gauge 121
-400
0.0 0.5 1.0 1.5 2.0
Phase
Figure 16.3: Repetitivity of phase locked average for heat transfer measurements
280
htmrc65.set
1500
1000
500 22 6 8
10
14 12
16
24 20 18
0
-500
800
mean RMS
600
400
0.5
Correlation
0.4
0.3
0.2
0.1
0.0
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 16.4: Minimum, maximum of mean period, mean RMS and correlation coefficient
of Nusselt number at 6500 RPM
Figure 16.5: Carpet plot of Nusselt number fluctuations as a function of time at 6500
RPM
281
htxm165.set
4000
16
17
3000
18
19 24
22
12
20
2000 20 16 18 10
Nu
14
21 2 8
4
6
1000 22
24
0
htxm265.set
6500
5
4
5500 3 24
22
12
2 20
16 18 10
Nu
14
4500 2 8
14
4
6
15
16
3500
2500
0.0 0.5 1.0 1.5 2.0
Phase
282
htxm365.set
8500
13
7500 12 24
22
10 11 12
9 20
16 18 10
Nu
8 14
6500 2 8
7 4
6
6
5500 5
4500
0.0 0.5 1.0 1.5 2.0
Phase
Gauge 2
rot.034 rot.022 6500 RPM htk2.set
0.15
900
700 0.10
Pressure
500
0.05
P/P01
100 0.00
-100
-0.05
-300
-500 -0.10
0.0 0.5 1.0 1.5 2.0
Phase
Figure 16.9: Comparison of heat transfer and pressure phase locked averaged profiles in
position no 2
283
ϕ=0 or ϕ=1 ϕ=1/8 ϕ=1/4 ϕ=3/8
14
3000
Nusselt
2000
16 2
20
10
1000 22
18
4 12
6 8
3000
Nusselt
2000
1000
0
-0.8 0.0 0.8 0.0 0.0 0.0
ϕ=7/8 ϕ=3/4 ϕ=5/8 ϕ=1/2
Figure 16.10: Unsteady Nusselt number distribution; plain line: mean, dotted line: time
dependant
284
Figure 16.11: Carpet plot of Nusselt number fluctuations as a function of time at 6000
RPM
Figure 16.12: Carpet plot of Nusselt number fluctuations as a function of time at 6800
RPM
285
Figure 16.13: Carpet plot of Nusselt number fluctuations as a function of time at 6500
RPM without coolant ejection
1500
6000 RPM
6500 RPM
6800 RPM
1000
500
Nu
-500
-1000
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
s/s(13)
Figure 16.14: Heat transfer: minimum and maximum of phased locked average at 6000,
6500 and 6800 RPM
286
Bibliography
[1] Wilson D.G. The design of High-efficiency turbomachinery and gas turbines. The
MIT Press, Cambridge, Massachussets, 1985.
[2] Anderson J.D. Introduction to flight. Mac Graw-Hill book company, 1978.
[4] Sieverding C.H., Arts T., and Dénos R. Investigation of the flow field downstream
of a turbine trailing edge cooled nozzle guide vane. Journal of Turbomachinery, also
ASME-94-GT-209, pages 291–300, April 1996. VKI 01-123(1) paper 13, 12 pages.
[5] Sieverding C.H., Arts T., Dénos R., Amecke J., Kapteijn C., Martelli F., Michelassi
F., Coluantoni S., Santoriello G., Schröder T., Bernard J., and Lapidus Y. Advances
in engine technology, chapter 3: Investigation of the wake mixing process behind a
transonic turbine inlet guide vanes with trailing edge coolant flow ejection. John
Wiley & Sons Ltd., 1995. Editor: R. Dunker.
[6] AGARDograph 328. Advanced method for cascade testing. AGARD-AG-328, 1993.
p11-60: Linear cascades, p73-85 Annular and rotating cascades, p103-152: Transient
cascade testing.
[7] Paillere Henri. Multidimensionnal upwind residual distribution schemes for the Euler
and Navier-Stokes equations on unstructured grids. PhD thesis, Université Libre de
Bruxelles, Von Karman Institute, June 1995.
[8] Mayle R.E. The role of laminar turbulent transition in gas turbines engines. Journal
of Turbomachinery, 113:509–531, october 1991.
[9] van den Braembussche R. Calcul des couches limites compressibles par méthode
intégrale. Novembre 1977. Méthode de le Foll, IVK CR 1978-3.
[10] Cicatelli G. and Sieverding C.H. A review of the research on unsteady turbine
blade wake characteristics. 85th propulsion and energetics panel symposium on loss
mechanisms and unsteady flows in turbomachines, Derby, UK, 1995.
[12] Sieverding C.H. Axial turbine performance prediction method. NATO advanced
study Institute on Thermodynamics and fluid mechanics on turbomachinery, Izmir
Turkey, 1984.
287
[13] Léonard Olivier. Conception et développement d’une méthode inverse de type Euler
et application á la génération de grilles d’aubes transsoniques. PhD thesis, Faculté
Polytechnique de Mons, Institut von Karman, 1992.
[15] Shichuan Ou, Je-Chin Han, Mehendale A.B., and Pang Lee C. Unsteady wake over a
linear turbine blade cascade with air and co2 film injection. ASME paper 93-GT-210,
1993.
[16] Vavra M.H. Aerothermodynamics and flow in turbomachines. R.E. Krieger Publish-
ing Company, 1974.
[17] Sieverding C.H., Van Hove W., and Boletis E. Experimental study of the three
dimensional flow field in an annular turbine nozzle guide vane . Journal of Engi-
neering for Gas Turbines and Power, (ASME-paper 83-GT-120), 106:No2 437–444,
April 1984.
[18] Moore J. and Adhye R.Y. Secondary flos and losses downstream of a turbine cas-
cade. Journal of Engineering for Gas Turbines and Power, (ASME-paper 85-GT-64),
107:961,, 968, October 1985.
[19] Yamamoto A. Production and development od secondary flows and losses in two
types of straight turbine cascades: part1-a stator case. Journal of Turbomachinery,
109:186–193, April 1987.
[20] Yamamoto A. Production and development od secondary flows and losses in two
types of straight turbine cascades: part2-a rotor case. Journal of Turbomachinery,
109:194–200, April 1987.
[21] Heider R., Duboue J.M., Petot B., Billonet G., Couaillier V., and Liamis N. Three
dimensionnal analysis of turbine rotor flow including tip clearance. ASME-93-GT-
111, 1993.
[22] Sieverding C.H. Recent progress in the understanding of basic aspects of secondary
flows in turbine blade passages. Journal of Engineering for Gas Turbines and Power,
(ASME-paper 84-GT-078), 107:No2 248–257, April 1985.
[23] Sharma O.P. and Butler T.L. Predictions of endwall losses and secondary flows in
axial flow turbine cascades. Journal of Turbomachinery, 109:No2 229–235, April
1987.
[24] Moustapha S.H., Paron G.J., and Wade J.H.T. Secondary flows in cascades of
highly loaded turbine blades. Journal of Engineering for Gas Turbines and Power,
(ASME-paper 85-GT-135), 107:1031, October 1985.
[25] Binder A., Förster W., Mach K., and Rogge H. Unsteady flow interaction caused by
stator secondary vortices in a turbine rotor. Journal of Turbomachinery, (ASME-
paper 86-GT-302), 109:pp251–257, April 1987.
288
[26] Boletis E. Effect of tip endwall countouring on the three dimensionnal flow field in
an annular turbine nozzle guide vane: Part1-Experimental investigation. Journal of
Engineering for Gas Turbines and Power, (ASME-paper 85-GT-71), 107:983–990,
October 1985.
[27] Yamamoto A. Interaction mechanisms between tip leakage flow and the passage
vortex in a linear turbine rotor cascade. Journal of Turbomachinery, 110:329–338,
July 1988.
[28] Kenichiro Ikeuchi Yamamoto A., Takayuki Matsunuma and Eisuke Outa. Un-
steady endwall/tip-clearance flows and losses due to turbine rotor-stator interaction.
ASME-paper 94-GT-461, 1994.
[29] Denton J.D. Loss mechanisms in turbomachines. ASME paper 93-GT-435, 1993.
[30] Boletis E. and Sieverding C.H. Experimental study of the three dimensionnal flow
field in a turbine stator preceded by a full stage. Journal of Turbomachinery, 113:1–9,
January 1991.
[31] Doorly D.J. and Oldfield M.L.G. Simulation of the effect of shock wave passing on
a turbine rotor blade. Journal of Engineering for Gas Turbines and Power, ASME
paper 85-GT-112, 107:998–1006, 1985.
[32] Orth U. Unsteady boundary layer transition in flow periodically disturbed by wakes.
ASME paper 92-GT-283, 1992.
[33] Schoberei M.T. and Radke R.E. Effects of periodic unsteady wake flow and pressure
gradient on boundary layer transition along the concave surface of a curved plate.
ASME-paper 94-GT-327, 1994.
[34] Liu X. and Rodi W. Measurement of unsteady flow and heat transfer in a linear
turbine cascade. ASME paper 92-GT-323, 1992.
[35] Ashworth D.A., LaGraff J.E., Schultz D.L., and Grindrod K.J. Unsteady aerody-
namic and heat transfer processes in a transonic turbine stage. Journal of Engineer-
ing for gas turbines and Power, (ASME paper 85-GT-128), 107:1022–1030, 1985.
[36] McFarland V.E. and Tiederman W.G. Viscous interaction upstream and downstream
of a turbine stator cascade with a periodic wake field . ASME paper 92-GT-162,
1992.
[37] Schoberei M.T., Pappu K., and Wright L. Experimental study of the unsteady
boundary layer behaviour on a turbine cascade. ASME-paper 95-GT-435, 1995.
[38] Yamamoto A., Minura F., Tomihisa S. Tominaga J., Outa E., and Matusuki M.
Unsteady three-dimensional flow behaviour due to rotor-stator interaction in an
axial flow turbine. ASME-paper 93-GT-404, 1993.
[39] Hodson H.P., Huntsman I., and Steele A.B. An investigation of boundary layer
development in a multistage lp turbine. ASME paper 93-GT-310, 1993.
289
[40] Halstead D.E., Wisler D.C., Okiishi T.H., Walker J.G., Hodson H.P., and Hyoun-
Woo Shin. Boundary layer development in axial compressors and turbines Part1:
composite picture. ASME-paper 95-GT-461, 1995.
[41] Epstein A.H., Guenette G.R., and Norton R.J.G. The MIT blowdown facility. ASME
paper 84-GT-116, 1984.
[42] Haldeman C.W., Dunn M.G., MacArtur C.D., and Murawski C.G. The USAF
advanced turbine aerothermal research rig. AGARD Conference Proceedings No.527:
Heat transfer and cooling in gas turbine, 1992.
[43] Ainsworth R.W.and Schultz D.L., Davies M.R.D., Forth C.P.J., Hilditch M.A.,
Oldfield M.L.G., and Sheard A.G. A transient flow facility for the study of the
thermofluid-dynamics of a full stage turbine under engine representative conditions.
ASME paper 88-GT-144, 1988.
[44] Hilditch M.A., Fowler A., Jones T.V., Chana K.S., Oldfield M.L.G., Ainsworth
R.W., Hogg S.I., Anderson S.J., and Smith G.C. Installation of a turbine stage in
the pyestock isentropic light piston facility. ASME paper 94-GT-277, 1994.
[45] Goodisman M.I., Oldfield M.L.G., Kingcombe R.C., Jones T.V., and Ainsworth
R.W.and Brooks A.J. An axial turbobrake. ASME paper 91-GT-1, 1991.
[46] Hodson H.P. Boundary layer and loss measurements on the rotor of an axial-flow
turbine. ASME paper 83-GT-4, 1983.
[50] Dietz A.J. and Ainsworth R.W. Unsteady pressure measurements on the rotor of a
model turbine stage in a transient flow facility. ASME paper 92-GT-156, 1992.
[51] Dring R.P., Joslyn H.D., Hardin L.W., and Wagner J.H. Turbine rotor-stator interac-
tion. Journal of Engineering for Power, ASME paper 82-GT-3, 104:No2, p729–742,
1982.
[53] Schubauer G.B. and Klebanoff P.S. Contributions on the mechanics of boundary
layer transition. NACA report 1289, 1956.
[54] Addison J.S. and Hodson H.P. Unsteady transition in axial flow turbine. ASME
paper 89-GT-289, 89-GT-290, 1989.
290
[55] Hodson H.P. and Addison J.S. Wake boundary layer interaction in an axial flow
turbine rotor at off-design conditions. ASME paper 88-GT-233, 1988.
[56] Mayle R.E. and Dullenkopf K. A theory for wake induced transition. ASME paper
89-GT-57, 1989.
[57] Dullenkopf K. and Mayle R.E. The effect of incident turbulence and moving wakes
on laminar heat transfer in gas turbines. ASME paper 92-GT-377, 1992.
[58] Sharma O.P., Pickett G.F., and Ni R.H. Assessment of unsteady flows in turbines.
Journal of Turbomachinery, (ASME-paper 90-GT-150), 114:79–90, January 1992.
[59] Saxer A.P. and Giles M.B. Predictions of 3-D steady and unsteady inviscid transonic
stator/rotor interaction with inlet radial temperature non-uniformity. ASME 93-
GT-10, 1993.
[60] Giles M.B. Stator/rotor interaction in a transonic turbine. AIAA-88-3093, 1988.
[61] Rao K.V., Delaney R.A., and Dunn M.G. Vane-blade interaction in a transonic
turbine stage, part I: aerodynamics. Journal of Propulsion and Power, 10:305–311,
May-June 1994.
[62] Guenette G.R., Epstein A.H., Giles M.B., Haimes R., and Norton R.J.G. Fully scaled
transonic turbine rotor heat transfer measurements . Journal of Turbomachinery,
ASME paper 88-GT-171, 1989.
[63] Rao K.V., Delaney R.A., and Dunn M.G. Vane-blade interaction in a transonic
turbine stage, part II: heat transfer. Journal of Propulsion and Power, 10:312–317,
May-June 1994.
[64] Hilditch M.A. and Ainsworth R.W. Unsteady heat transfer measurements on a
rotating gas turbine blade. ASME paper 90-GT-175, 1990.
[65] Garside T., Moss R.W., Ainsworth R.W., Dancer S.N., and Rose M.G. Heat transfer
to rotating turbine blades in a flow undisturbed by wakes. ASME-paper 94-GT-94,
1994.
[66] Hilditch M.A., Smith G.C., Anderson S.J., and Chana K.S. Unsteady measurements
in an axial flow turbine. 85th propulsion and energetics panel symposium on loss
mechanisms and unsteady flows in turbomachines, Derby, UK, 1995.
[67] Sheldrake C.D. and R.W. Ainsworth. The use of hot-wires applied to aerodynamic
measurements in a model turbine stage. Proceedings of First European conference
on turbomachines, 1995.
[68] Dunn M.G., Seymour P.J., Woodward S.H., George W.K., and Chupp R.E. Phase
resolved heat-flux measurements on the blade of a full scale rotating turbine. Journal
of Turbomachinery, (ASME-paper 88-GT-173), 111:08–19, January 1989.
[69] Binder A., Förster W., Kruse H., and Rogge H. An experimental investigation
into the effect of wakes on the unsteady turbine rotor flow. Journal of Engineering
for Gas Turbines and Power, (ASME-paper 84-GT-178), 107:No2 pp458–466, April
1985.
291
[70] Lakshminarayana B. An assessment of computational fluid dynamic techniques in
the analysis and design of turbomachinery. Journal of Fluids Engineering, September
1991.
[71] Sheldrake C.D. and Ainsworth R.W. Hot-wire anemometry technique on rotating
turbine experiments. Proceedings of the 12th Symposium on Measuring Techniques
for Transonic and Supersonic Flow in Cascades and Turbomachines, 1994.
[72] Pfriem H. Zur Messung veränderlicher Temperaturen von gasen und Flüssigkeiten.
Forschung Ing.-Wesen,Bd.7,Heft 2, 1936. (Measurement of temperature fluctuations
in fluids).
[73] La Rue J.C., Deaton T., and Gibson C.H. Measurement of high-frequency turbulent
temperature. Review of scientific instruments, 46 no 6:757,764, 1974.
[74] Hojstrup J., Rasmussen K., and Larsen S.E. Dynamic calibration of temperature
wires in still air. DISA Information no. 20, pages 22–30, september 1976.
[75] Millon F., Paranthoen P., and Trinite M. Influence des échanges thermiques entre
le capteur et ses supports sur la mesure des fluctuations de température dans un
écoulement turbulent. Journal of Heat and Mass Transfer, 21:1–6, 1978.
[76] Weeks A.R., Beck J.K., and Joshi M.L. Response and compensation of temperature
sensors. Journal of Physics and Scientific Instrumentation, 1988.
[78] Schubauer G.B. Early developments in hot-wire anemometry at nbs and look at
some elsewhere. Advances in hot wire anemometry, 1968.
[79] Olivari D. Effects of the wire supports on the frequency response of hot wire
anemometers. ref. Von Karman Institute, TN 118, 1976.
[80] Maye J.P. Error due to thermal conduction between the sensing wire and its sup-
ports when measuring temperatures with a wire used as a resistance thermometer.
Electronic Measurements of Mechanical Events Disa information, 1970.
[81] Tsuji T., Nagano Y., and Tagawa M. Frequency response and instantaneous tem-
perature profile of cold-wire sensors for fluid temperature fluctuation measurements.
Experiments in Fluids, 1992.
[82] Parantheon P., Petit C., and Lecordier J.C. The effect of thermal prong-wire inter-
action on the response of a cold wire in gaseous flow. Journal of Fluid Mechanics,
124:457–473, 1982.
[83] Ji Ryong Cho and Kyung Chun Kim. A simple high performance cold-wire ther-
mometer. Meas. Sci. Technol., 4:1346–1349, 1993.
[84] Collis D.C and Williams M.J. Two dimensionnal convection from heated wires at
low Reynolds number. Journal of fluid mechanics, 6:357,384, 1959.
292
[85] Reid R.C, Prausnitz J.M., and Poling B.E. The properties of gases and liquids. Mc
Graw and Hill, New York, 1987.
[86] Lide D.R. Chemistry and physics handbook, volume 71 st Edition. 1990-91.
[87] Kay J.M. and Nedderman R.M. An introduction to Fluid Mechanics and Heat trans-
fer, volume Third Edition. Cambridge University Press, 1974.
[88] Oppenheim A.V., Willsky A.S., and Young I.T. Signals and systems. Prentice/Hall
International editions, 1983. (description of first order linear systems).
[89] Rabiner L. and Gold B. Theory and application of digital processing. Prentice-Hall,
1975.
[90] Sieverding C.H., Vanhaeverbeek C., and Schulze G. An opto-electronic data trans-
mission system for measurements on rotating turbomachinery components. ASME
paper 92-GT-337, 1992.
[91] Schultz D.L. and Jones T.V. Heat transfer measurements in short-duration facility.
AGARDograph No 165, 1973.
[92] Whitaker S. Elementary heat transfer analysis. Pergamon unified engineering series,
Rober Maxwell, 1976.
[93] Schneider P.J. Conduction heat transfer. Addison-Wesley publishing company Inc,
Reading, Massachusetts, 1957.
[94] Doorly J.E. Procedures for determining surface heat flux using thin film gauges on a
coated metal model in a transient facility. Journal of Tubomachinery, 110:242–250,
April 1988.
[95] Dunn M.G. The thin film Gauge. Lectures Series of the von Karman Institute on
Measurements techniques, 1995.
[96] Ligrani P.M., Camci C., and Grady M.S. Thin film Heat transfer gages construction
and measurements details. 1982.
[97] Epstein A.H., Guenette G.R., Norton R.J.G., and Cao Yuzhang. High frequency
response heat flux gauge for metal blading. AGARD Conference Proceedings No.390:
Heat transfer and cooling in gas turbines, 1985.
[98] Guo S.M., Spencer M.C., Lock G.D., Jones T.V., and Harvey N.W. Heat transfer
measurements using thin film gauges . Proceedings of the 12th symposium on mea-
suring techniques for transonic and supersonic flow in cascades and turbomachines
held in Praag (Tchequie) 12-13th september 1994, 1994.
[99] Rutherford A.W. and Arts T. Aero thermal performance of a 2d, highly loaded,
transonic turbine nozzle guide vane. 1990.
[100] Miller C.G. Comparison of thin-film resistance heat-transfer gauges with thin-skin
transient calorimeter gages in conventional hypersonic wind tunnels. NASA Tech-
nical Memorandum 83197, 1981.
293
[101] Ainsworth R. Recent development in fast response aerodynamic technology. Von
Karman Institute Lecture Series 1995-01 on Measurement Techniques, 1995.
[102] Batt J.J.M., Ainsworth R.W., and Allen J.L. Temperature compensation applied
to surface mounted kulite pressure sensors. Internal Report, University of Oxford,
1995.
[103] Gossweiler C., Humm H.J., and Kupferscmied P. The use of piezo resistive semi-
conductor pressure transducers for fast response probe-measurements in turboma-
chinery. Proceedings of the 10thsymposium on measuring techniques for transonic
and supersonic flow in cascades and turbomachines held in Bruxelles in september
1990, 1990.
[104] Santoriello G. and Colella A. Rotor blade aerodynamic design. Alfa Romeo Avio,
Brite Euram area 3 technical report, October 1993.
[105] Sieverding C.H. Advanced course in turbines. Von Karman Institute Course Note.
[106] Sieverding C.H. and Arts T. The VKI Compression Tube Annular Cascade Facility
CT3. ASME paper 92-GT-336, 1992.
[107] Schultz D.L., Jones T.V., Oldfield M.L.G., and Daniels L.C. A new transient cascade
facility for the measurements of heat transfer rates. AGARD CP No 229, 1977.
294
List of Figures
295
3.6 Influence of passage vortex on flow angle and velocity . . . . . . . . . . . . 47
3.7 Meridional outlet angle distribution behing the Brite stator . . . . . . . . . 48
3.8 Secondary velocity vector plot from Yamamoto [19] . . . . . . . . . . . . . . 48
3.9 Kinetic loss carpet plot at 0.25 Cax downstream of stator trailing edge . . 49
3.10 Kinetic losses carpet plot at 0.54*Cax . . . . . . . . . . . . . . . . . . . . . 49
3.11 Pitchwise average of kinetic losses . . . . . . . . . . . . . . . . . . . . . . . . 50
3.12 Meridional Mach number profile behing the Brite stator . . . . . . . . . . . 50
3.13 Schematic view of tip clearance flow at mid-clearance . . . . . . . . . . . . . 51
296
7.1 Transfer function of 1D conduction equation . . . . . . . . . . . . . . . . . . 135
7.2 Constant heat flux case: surface temperature history (a), wall heat flux
history (b), temperature profile inside the media at several successive times
(c) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.3 Constant heat flux case: Tx /Tx=0 and q̇x /q̇x=0 as a function of η . . . . . . 136
7.4 Constant gas temperature case: temperature and flux profiles in the sub-
strate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
7.5 Constant gas temperature case: example of wall temperature and wall flux
history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.6 RC network for analogue resolution of 1D heat conduction equation with
semi-infinite substrate assumption . . . . . . . . . . . . . . . . . . . . . . . 137
7.7 Flux computation using Cook and Felderman algorithm . . . . . . . . . . . 138
7.8 Flux computation using FFT technique . . . . . . . . . . . . . . . . . . . . 138
7.9 Wall temperature profiles comparison: Crank-Nicholson, exact solution . . 139
7.10 Flux computation using Crank-Nicholson implicit scheme . . . . . . . . . . 139
7.11 Flux computation with Crank-Nicholson scheme for a real test . . . . . . . 140
7.12 Comparison Crank-Nicholson, analogue circuit . . . . . . . . . . . . . . . . 140
7.13 Current step generated by the circuit, corresponding flux . . . . . . . . . . 141
7.14 Resistance and surface temperature history . . . . . . . . . . . . . . . . . . 141
7.15 Illustration of flux/surface temperature ratio evolution with frequency . . . 142
7.16 Transfer function of the heat transfer board . . . . . . . . . . . . . . . . . . 142
7.17 Blade with Macor insert in the leading edge . . . . . . . . . . . . . . . . . . 143
7.18 Surface temperature and heat fluxes measured in rotation across a heated jet143
297
10.5 Opened facility with the rotor in overhang position . . . . . . . . . . . . . . 184
10.6 Brake blade velocity triangle . . . . . . . . . . . . . . . . . . . . . . . . . . 185
10.7 Brake blade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
10.8 Photographs of the shaft equipped with 16 emittor rings (top) and receiver
unit (bottom) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
10.9 Run-up and run-down of the rotor, vibration level . . . . . . . . . . . . . . 187
10.10Operation cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
10.11Evolution of pressures during a test . . . . . . . . . . . . . . . . . . . . . . . 188
298
14.1 FFT of pressure signal P0r/P01 from probe in the blade nose . . . . . . . . 242
14.2 Phase locked average and raw data . . . . . . . . . . . . . . . . . . . . . . . 242
14.3 Influence of the number of periods in the phase locked average . . . . . . . 243
14.4 Repetitivity of measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 243
14.5 Total and static pressure at stagnation point . . . . . . . . . . . . . . . . . 244
14.6 Phase locked average and RMS at 6500 RPM . . . . . . . . . . . . . . . . . 244
14.7 Schematic of shock-wake system at vane TE . . . . . . . . . . . . . . . . . . 245
14.8 Influence of rotational speed . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
14.9 Influence of coolant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
14.10Phase locked averaged for the spacing 0.50 Cs,,ax . . . . . . . . . . . . . . . 246
14.11Test repetitivity for cold wire probe . . . . . . . . . . . . . . . . . . . . . . 247
14.12Phase-locked average signals of cold wire probe at 6000, 6500 and 6800 RPM247
14.13Phase-locked average signals of cold wire probe at 6500 RPM with and
without coolant ejection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
14.14Total pressure, flow angle and static pressure profile behind the stator at
mid span M2,is = 1.05, cm = 3%, x/cax = 0.25 . . . . . . . . . . . . . . . . 248
14.15Comparison between 2D Euler prediction and measurements . . . . . . . . . 249
14.16Turbulence level in the wake behind the Brite stator at 0.25 Cax . . . . . . 249
14.17Computation of relative total pressure from stationnary measurements at
6800 RPM; comparison with measured P0r . . . . . . . . . . . . . . . . . . 250
14.18Computation of relative total pressure from stationnary measurements at
6800 RPM including pitchwise variation of pressure, temperature and angle 250
14.19Computation of relative total pressure from stationnary measurements at
6800 RPM: translation the wake position with respect to static pressure by
-0.55 of a pitch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
14.20Computed unsteady velocity field . . . . . . . . . . . . . . . . . . . . . . . . 251
14.21Computation of relative total temperature from stationnary measurements
at 6800 RPM including pitchwise variation of pressure, temperature and
angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
14.22Computation of relative total temperature from stationnary measurements
at 6800 RPM: translation of the wake position with respect to static pressure
by -0.55 of a pitch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
14.23Computation of relative inlet angle from stationnary measurements at 6800
RPM including pitchwise variation of pressure, temperature and angle . . . 253
299
15.10RMS traces along one pitch for gauges 16 to 5 . . . . . . . . . . . . . . . . . 266
15.11RMS traces along one pitch for gauges 5 to 13 . . . . . . . . . . . . . . . . . 267
15.12Shock pattern evolution during one rotor blade cycle from Giles et al. [60] . 268
15.13Unsteady Mach number distribution; plain line: mean, dotted line: time
dependant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
15.14Chosen phase-locked averaged signals at 3 different RPM . . . . . . . . . . 270
15.15Pressure fluctuations on the blade surface as a function of time at 6000 RPM271
15.16Pressure fluctuations on the blade surface as a function of time at 6800 RPM271
15.17Minimum, maximum of mean period, mean RMS and correlation coefficient
of pressure at 6000, 6500 and 6800 RPM with coolant ejection . . . . . . . 272
15.18Pressure fluctuations on the blade surface as a function of time at 6500
RPM without coolant ejection . . . . . . . . . . . . . . . . . . . . . . . . . . 273
15.19Minimum, maximum of mean period, mean RMS and correlation coefficient
of pressure at 6500 RPM with and without coolant ejection . . . . . . . . . 274
15.20Pressure fluctuations on the blade surface as a function of time at 6500
RPM with coolant ejection and a spacing of 0.50 Cs,ax . . . . . . . . . . . . 275
15.21Minimum, maximum of mean period, mean RMS and correlation coefficient
of pressure at 6500 RPM with coolant ejection for spacings 0.35 Cs,ax and
0.50 Cs,ax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
16.1 Phase locked averages of surface temperature from selected gauges . . . . . 279
16.2 Influence of number of periods in phase-locked averaged Nusselt number . . 280
16.3 Repetitivity of phase locked average for heat transfer measurements . . . . 280
16.4 Minimum, maximum of mean period, mean RMS and correlation coefficient
of Nusselt number at 6500 RPM . . . . . . . . . . . . . . . . . . . . . . . . 281
16.5 Carpet plot of Nusselt number fluctuations as a function of time at 6500
RPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
16.6 Nusselt number time traces: gauges 24 to 16 (6500 RPM) . . . . . . . . . . 282
16.7 Nusselt number time traces: gauges 16 to 5 (6500 RPM) . . . . . . . . . . . 282
16.8 Nusselt number time traces: gauges 5 to 13 (6500 RPM) . . . . . . . . . . . 283
16.9 Comparison of heat transfer and pressure phase locked averaged profiles in
position no 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
16.10Unsteady Nusselt number distribution; plain line: mean, dotted line: time
dependant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
16.11Carpet plot of Nusselt number fluctuations as a function of time at 6000
RPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
16.12Carpet plot of Nusselt number fluctuations as a function of time at 6800
RPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
16.13Carpet plot of Nusselt number fluctuations as a function of time at 6500
RPM without coolant ejection . . . . . . . . . . . . . . . . . . . . . . . . . . 286
16.14Heat transfer: minimum and maximum of phased locked average at 6000,
6500 and 6800 RPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
300
Liste des publications
301
Index
302
corner, 40
horseshoe, 39
passage, 38
scrapping, 42
tip clearance, 42
trailing edge sheet, 38
wake, 24
wake
distortion effect, 58
effect, 57
Wheatstone bridge, 78
303