Sie sind auf Seite 1von 14

Whole-assembly Flutter Analysis of

a Low Pressure Turbine Blade

A.I. Sayma M. Vahdati J.S. Green & M. Imregun


Imperial College, MED
Exhibition Road, London SW7 2BX, UK

Rolls-Royce plc, PO Box 31, Derby DE2 8BJ, UK

Number of pages : 14
Number of gures : 10
Number of tables : 1

Whole-assembly Flutter Analysis of a Low


Pressure Turbine Blade
A.I. Sayma M. Vahdati J.S. Green & M. Imregun

Abstract
This paper reports the ndings of a utter investigation on a low pressure turbine blade using a 3D non-linear integrated aeroelasticity method. The approach
has two important features: (i) the calculations are performed in a time-accurate
and integrated fashion whereby the structural and uid domains are linked via an
exchange of boundary conditions at each time step, and (ii) the analysis is performed on the entire bladed-disk assembly, thus removing the need to assume a
critical vibration mode shape. Although such calculations are both CPU and incore memory intensive, they do not require pre-knowledge of the utter mode and
hence they allow a better understanding of the aeroelasticity phenomena involved.
The ow is modelled inviscidly but the steady-state viscous e ects are accounted
for using a distributed loss model. The structural model was obtained from a
standard FE representation and a large number of assembly modes were included
in the calculations. The study was focused on three part-speed conditions at
which a number of unstable modes were known to exist from the available experimental data. The whole assembly was modelled using about 664,000 mesh points
and predictions were made of aeroelastic modal time histories. From those time
histories, it was possible to identify the forward and backward traveling waves
and to deduce the unstable modes of vibration. The theoretical predictions were
found to be in very good agreement with the experimental ndings.

1 Introduction
Modern turbomachines operate under very complex ow regimes where a mixture
of subsonic, transonic and supersonic regions coexist. Shock wave interactions
and shock wave boundary layer interactions lead to a complex ow structure, the
modelling of which needs to be both computationally-ecient and numerically
accurate. The accuracy requirement becomes even more important in fan utter
calculations, where the position of the shock is the most single important factor
(Goldstein et al. 1977, Bendiksen 1986).
A literature survey on utter prediction methods is beyond the scope of this
paper and the interested reader should consult Marshall & Imregun (1996). However, a very brief overview will be given here for sake of completeness. Most
utter computations consider a typical sector vibrating in some given assembly
mode (or interblade phase angle) for which utter is expected to occur. In such
2

analyses, the interblade phase angle must be prescribed at the periodic boundaries (Lane 1955, Erdos et al. 1977, He 1992). Although such approaches have
relatively modest CPU and in-core memory requirements, they also have the following serious shortcomings: (i) the utter frequency of the aeroelastic system
must be assumed to be identical to the natural frequency of the structural system,
(ii) no interaction can take place between the various assembly modes, (iii) the
approach is valid only when the aeroelastic system has reached a periodic steadystate condition1, and (iv) if the utter mode is not known in advance, all possible
interblade phase angles need to be tried as a series of separate computations.
In the present work, it is proposed to undertake a whole-assembly analysis
in order to overcome the problems above. The non-linear aeroelasticity method
described by Vahdati & Imregun (1996) will also be used here. Such an approach
allows the inclusion of all assembly modes and hence removes the need to assume
a critical vibration mode shape, characterised by an interblade phase angle or a
nodal diameter pattern. Also, it is worth noting that previous utter analyses of
turbomachinery blades are focused on fan blades, though it is well known that
large turbine blades also su er from the same instability.

2 The ow model
The ow model is described in detail by Vahdati & Imregun (1996) but a brief
summary will be given here for the sake of completeness.
2.1

Governing equations

The ow equations will be given for a three dimensional Cartesian co-ordinate


system which is xed to a rotating blade. In this case, the conservation form of
the unsteady, Reynolds-averaged Navier-Stokes equations is given by:

U + @F

@
@t

@x

G + I;
= @@x

(1)

where U is the solution vector, F and G are the inviscid and viscous uxes
respectively and I contains the terms due to the rotation of the coordinate system.
These terms can be written as:
2
3
2
3
2
3
0

(u w )
6
7
U = 64 u 75 ; F = 64 u (u w ) + p 75 ; G = 664  @T 775 ; (2)

E (u w ) + P u
u  +
@x
j

ij

ij

ik

1 In

most cases, extra computations are needed to impose the required interblade phase angle

pattern over several cycles

were  is the density,  is the speci c total energy, u is the absolute velocity vector,
w is the contra variant velocity vector,  is the summation of the molecular and
eddy di usivity. The viscous stress tensor  is given by:
i

ij

2  @u
(3)
3 @x
where  is the summation of the molecular and eddy viscosities.
The term I of equation 1 is given by:
2
3
0
66 0 77
I = 6666 
u3 7777
(4)


u
4
2 5
0
where
is the rotational speed. The above system of equations is complemented
by the perfect gas equation :
@u
 =
@x

ij

p = (

+ @u
@x

ij

1)

u2

(5)
2
where denotes the ratio of the uid speci c heats. The Eddy viscosity is
calculated using the one-equation turbulence model of Baldwin and Barth (1991).
2.2

Numerical methodology

The spatial solution domain is discretized using linear tetrahedral elements. The
equations are discretized by rst obtaining a variational weak formulation. The
variables are here represented over the discrete domain via piecewise linear interpolation shape functions. The resulting central di erence scheme is numerically stabilised using a fourth order dissipation scheme for smooth ows. The
matrix dissipation scheme used follows that proposed by Swanson and Turkel
(1992). A more dissipative scheme is required near discontinuities such that unwanted numerical oscillations can be prevented. This is achieved by switching to
a second-order dissipation scheme in such situations. The switching between the
two schemes is determined using a pressure-based sensor, Vahdati and Imregun
(1996).
The semi-discrete system of equations is advanced in time using an implicit
time integration. The resulting system of non-linear equations is solved using
Jacobi iterations and their nal form is given by:

K(U)U = RHS + RHS


i

(6)

where K(U) is given in Vahdati et al (1993), RHS is the right hand side evaluated
using the latest values of the solution vector, subscripts i and v referring to the
inviscid and viscous contributions respectively.
2.3

The loss model

A viscous simulation of a full assembly will have very substantial computational


requirements so that detailed parametric studies are not yet feasible. Consequently, a distributed loss model is used to represent the steady-state viscous
uxes and the unsteady viscous e ects are inherently assumed to be negligible.
Despite this limitation, the method is usually considered to be applicable to a
wide range of turbomachinery operating conditions as the variation of the unsteady losses is expected to be small. In general, a loss model assumes that the
e ect of the shear stresses on the motion can be represented by a distributed friction force, the resulting ow representation being still of inviscid nature, but not
isentropic (Hirsch 1995). The loss model is de ned here numerically using data
from a previous viscous analysis of a single blade-passage on a much ner mesh.
The viscous solution is then interpolated onto the coarser mesh which is used
for an inviscid analysis of the whole-assembly. The viscous stresses of the rst
solution are thus saved on the second mesh and used throughout the unsteady
inviscid solution as a source term.

3 The structural model


The structural model is based on a linear modal model, the mode shapes and natural frequencies being obtained via standard FE analysis techniques. The mode
shapes are interpolated from the structural analysis mesh onto the aerodynamic
mesh as the two discretization levels are unlikely to be coincident.
The structural part of the aeroelastic equations of motion are uncoupled by
using the mode shape matrix .
d2 
dt2

+ !2 =
r

X
N

 :F
i;r

=1

(7)

where r is the mode index, i is the node number,  is the modal de ection, N is
the number of surface nodes and ! is the natural frequency. F is the aerodynamic
force which is obtained using:
F = p A n
(8)
where p denotes the static pressure, A denotes the elemental area and n denotes
the unit normal vector, all at node i.
Equation (7) is advanced in time using the Newmark- method which is unconditionally stable. Boundary conditions from the structural and aerodynamic
i

domains are exchanged at each time step and the aerodynamic mesh is moved to
follow the structural motion.
A study of the vibration characteristics of bladed disks is beyond the scope of
this paper (see for instance Ewins 1973) but a very brief summary will be given
here. Bladed disk assemblies vibrate in terms of nodal diameters and nodal circles,
though the discussion will focus on nodal diameters only as they are of direct
relevance to utter instability. For instance, a 2-nodal diameter vibration mode
is characterised by the existence of 2 motionless lines (or diameters) positioned
in such a way that the rim of the disk will describe a cos(2) (or sin(2)) pattern,
 denoting the angular position along the rim. Because of circular symmetry, for
a non-rotating bladed-disk, there will be two such modes with identical natural
frequencies but orthogonal shapes, the rst one being described by cos(n) and
the other by sin(n), n being the actual number of nodal diameters. Under the
e ect of rotation, these two modes will split into forward and backward travelling
nodal diameters, the de nition being associated with the direction of the shaft
rotation. Fan utter is known to occur in forward nodal diameters while turbine
utter occurs in backward nodal diameters. For each individual cantilevered
blade mode, e. g. rst ap, rst torsion, etc., there will be N assembly modes,
N being the total number of blades. For an even number of blades, the assembly
modes will consist of 0; 1; :::::; n=2 1; N=2 nodal diameter modes, 1 to (N=2 1)
nodal diameters being the double modes giving rise to forward and backward
travelling waves.

4 Numerical results
The above method was used for the utter analysis of an early-developmentstandard low-pressure turbine for which some experimental observations were
available at 88%, 94% and 99% of maximum shaft speed conditions. A steadystate viscous solution was obtained rst for a single blade passage. The unstructured mesh used containing 85,275 points is shown in gures 1 and 2. Figures
3 and 4 show the solution at midsection for 99% speed case. The predicted
ow results were compared to available measured data and good agreement was
observed in all cases.
The viscous solution was interpolated onto an inviscid single-blade-passage
mesh of about 8,000 points. The single passage mesh was then expanded to a
whole-assembly mesh of about 664,000 points. A view of a sector of this mesh is
shown in g 5.
A total of 15 assembly modes, namely nodal diameters 3-12, 15, 20, 25, 33
and 42 for the rst ap mode, were included in the utter calculations. Because
of computational considerations, the calculations were started by giving initial
modal velocities to some of the assembly modes, a procedure analogous to hitting
a blade and monitoring whether the response is decaying or growing in time.
6

Typical unsteady pressure contours are shown in gure 6 and examples of stable
and unstable traveling wave modal time histories given in gures 7 and 8.
For each time history, it is possible to calculate a logarithmic decrement
(logdec) value which is a linear measure of the decay rate between two given
cycles in the time history. In general, the time histories do not show a linear
behaviour and care should be taken in interpreting the results obtained using the
linear logdec tool.
The results obtained are summarised in g 9 where the negative x-axis values
indicate backward travelling nodal diameters. The instability is seen to be in the
(backward travelling) 5 to 15 nodal diameter range, the 94% speed case being
worst. At this particular speed, the 8ND mode is the least stable. These ndings
are in very good agreement with experimental observations which are based at
monitoring the stress levels at a given speed. Provided an adequate number of
blades are instrumented, it is possible to obtain the vibration mode shape by a
Fourier transform of individual blade responses. The same information can also
be inferred from the resonant frequencies, albeit with less certainty, if the modes
are well separated, which is usually the case for turbine stages with relatively
exible disks. In any case, the measured maximum stress values indicated that
utter instability occurred for the backward travelling 7, 8 & 9 nodal modes, the
situation getting worse with increasing speed. The predicted trend is very similar
but a large number of modes are seen to be unstable. This feature is attributed
to the fact that the calculations assume zero structural damping. The inclusion
of some structural damping will almost certainly stabilize some of the modes with
small negative damping.
Finally, the actual displacement time history at a point near a blade's tip
is shown in g 10. The instability can easily be inferred from the beating-like
motion whose time history is showing an overall growing trend.

5 Concluding Remarks
1. A 3D integrated non-linear aeroelasticity method has been presented and
applied to the utter analysis a low-pressure turbine. Numerical results
have been found to be in very good agreement with the experimental observations.
2. All modes of interest were included in one computation and no assumptions
about critical vibration modes were necessary. The results showed some
signi cant non-linear time histories, highlighting the diculties of using
equivalent linear damping for stability purposes. Interactions between the
various assembly modes were also observed.
3. The computational cost for such 3D non-linear calculations is high, both in
CPU and memory requirements. Therefore, standard whole-assembly cal7

Nodal Diameter

Frequency(Hz)
88%
94%
99%
3
171.095 171.096 174.774
4
190.831 187.528 191.253
5
220.147 213.492 217.241
6
250.463 240.562 244.393
7
281.761 268.919 272.861
8
316.007 300.771 304.807
9
354.185 337.317 341.404
10
396.407 378.855 382.942
11
441.981 424.893 428.927
12
489.259 474.066 478.001
15
607.398 606.689 610.314
20
670.106 672.203 676.120
25
690.428 689.416 693.351
33
706.216 703.028 706.899
42
711.670 707.913 711.749
Table 1: Structural frequencies at % of maximum shaft speed
culations are likely to be limited to inviscid ow modelling for the next few
years, though the rst whole-assembly viscous calculations should appear
well before the end of the Century.

Acknowledgments
The authors would like to thank Rolls-Royce plc for both sponsoring the work
and allowing the publication of this paper.

References
[1] Baldwin, B. S. and Barth, J. 1991 , A one equation turbulence transport
model for high Reynolds number wall-bounded ows, AIAA Paper 91-0610.
[2] Erdos, J. I., Altzner. E. & McNally, W. 1977 Numerical solution of periodic
transonic ow through a fan stage. AIAA Journal, 15, 165-186
[3] Ewins, D. J. 1973 Vibration modes of bladed disk assemblies. I. Mech. E. J.
of Mech. Eng. Sci. 15, 165-186
[4] Goldstein, M. E., Braun, W. & J. J. Adamczyk, J. J. 1977 Unsteady ow in
supersonic cascade with strong in-passage shocks, J. Fluid Mech. 83, 569604.
8

[5] He, L. 1992 Method of simulating unsteady turbomachinery ows with multiple perturbations. AIAA Journal, 30, 2730-2735
[6] Lane, F. 1956 System mode shapes in the utter of compressor blade rows.
J. of Aeronautical Sciences, 23, 54-66
[7] Marshall, J. G. & Imregun, M. 1996 , A survey of aeroelasticity methods
with emphasis on turbomachinery, J. of Fluids and Structures, 10, 237-267
[8] Hirsch, C. 1995, Numerical computation of Internal and external ows Volume I, John Wiley & Sons.
[9] Vahdati, M. & Imregun, M. 1996 A Non-linear Integrated Aeroelasticity
Analysis of a Fan Blade Using Unstructured Dynamic Meshes. Journal of
Mech. Eng. Sci, Part C, 210, 549-563
[10] Vahdati, M., Morgan, K. and Peraire, J. 1993 The computations of
viscous compressible ows using an upwind algorithm and unstructured
meshes, Computational Non-linear Mechanics in Aerospace engineering,
AIAA Progress in Aeronautics and Astronautics series.

Figure 1: Single passage viscous mesh used for determining the loss model

Figure 2: Zoom view of g 1 of the hub section. The mesh is unstructured in the
tangential/axial direction and structured in the radial direction
10

0.63
8
0.6

0.34
0.39
0.44
0.49 0.53
0.58

0.06
5
0.1 0.20

0.72

0.25

0.3
0.39 4
0.44
0.49
0.68
0.53
0.58
0.63
8
5
0.
0.68
0.63

0.34

0.6

0.6
0.6 3
3

0.29

Figure 3: Steady-state Mach number contours at mid-section


2.72

2.66

2.61

2.55

2.50
2.39
2.28

2.72

2.
12

2.2

2.17

8
2.8
2.83
2.77
2.72
2.66

2.23

2.72

2.55
2.44
2.34

2.28
2.23

Figure 4: Normalised steady-state pressure contours at mid-section


11

Figure 5: A view of a sector of the whole assembly mesh containing about 650,000
points

Figure 6: Unsteady pressure contours obtained for the mesh of gure 5


12

8 ND (99 % speed), backward travelling wave

Modal displacement m/sqrt(kg)

x 10

5
0

0.05

0.1

0.15
0.2
Time (seconds)

0.25

0.3

0.35

Figure 7: Modal displacement for an unstable mode

8 ND (99 % speed), forward travelling wave

Modal displacement m/sqrt(kg)

x 10

4
0

0.05

0.1

0.15
0.2
Time (seconds)

0.25

0.3

Figure 8: Modal displacement for a stable mode


13

0.35

88% speed

1.2

94% speed
99% speed

Damping %

0.8

0.6

0.4

0.2

0.2
40

30

20

10

0
10
Nodal diameter

20

30

40

Figure 9: Damping vs. nodal diameter for the three speeds

0.0001
8e-05

Tip trailing edge Displacment (m)

6e-05
4e-05
2e-05
0
-2e-05
-4e-05
-6e-05
-8e-05
-0.0001
0

0.05

0.1

0.15
0.2
Time (seconds)

0.25

0.3

0.35

Figure 10: Actual dispalcement at the tip trailing edge


14

Das könnte Ihnen auch gefallen