C557/067/99
Forced response prediction within the design
process
J S GREEN and J G MARSHALL
Rols-Royes pk, Dery, UK
SYNOPSIS
High eyele fatigue caused by vibration is a serious problem for turbomachinety blading, The
raditional method for assessing vibration acceptability is to rely on past experience and
ich obviously has its
designs are only revealed late in the develop
roblems, as unacceptable
ent programme
his paper presents the application of a numerical method for the prediction of forced
response in turbomachinary blade rows, The unsteady aerodynamic model ses the three
dimensional linearised Filer equations. Unsteady pressures are passed to a separate transicn
Jynamies program to predict the blade resonant responses, including non-linear mechanical
effects, Computational results are presented for atypical turbine stage and compared to i:
engine strain gauge measurements of resonant frequency and amplitude
The emphasis here isnot the detail of sophisti
unsteady CFD tools, but instead the focus
ison the application and key elements of a system which ean influence component design at
mn carly stage in the process, The method models an acceleration of the rotor using sealing
nd interpolation rather than a single fixed point. Good agreement was found with
‘experimen results in terns of amplitude and frequency shif.
1 NOMENCLATURE
5 critical damping ratio k sper element contact stiffness
Q —eynamie magnification factor ‘ction coefficient2, INTRODUCTION
essive vibration can cause a component to fail due to high cycle fatigue. in aero-engines
this must be avoided because it could lead to sudden engine failure, risking the safety of the
ff, Airworthiness regulations therefore require in-situ engine ests to prove the vibration
integrity. takes place late in the development program and vibration
problems which come to light at this stay
The traditional approach for avoids
of vibration problems during the design process isto
caloulate the modes of vibration using a Finite Element approach and to plot these on a
‘Campbell diagram, This shows the engine shaft speeds at which resonances occur but itis not
clear whether vibrations will b
eptable, The designer relies on previous experience and
may put unnecessary constraints on the design due to vibration concems, A method for
predicting the vibration response is rogue if fevelopment costs and time scales are
to be reduced, Indeed it may also give the designer freedom to study unconventional airfoils
which may improve efficiency, life oF weight
A forced response method mist include models of both the aerodynamic and mechanical
properties of the structure, Reviews by Verdon (1) and Marshall (2) show a large ran
possible approaches, highlighting advantages and limitations. Fully coupled
viscous methiods have been developed (3), but these models are computationally expensive
Lineatised CFD methods an ue to the speed advantago and there is
growing body of evidence that linear mo
sachinary bladerows (4) & (5)
can be used successfully for unsteady flow i
turbos
The majority ofthe foreed response approaches reported calculate the vibration response at a
specific aerodynamic condition. However, damping devices such as under platform dampers
vency by as much as 20% . The resonance will thea occur at
different operating point i.e. the shafl spoed and aerodynamic boundary condition will both
change.
Te purpose of the work reported here
could be used to predict the vibration chara
0 investigate the requirements of a system which
ristcs ofa turbine blade with an under platform
damper. The method is requited for use within the design process and must the
to have an impact on the design. One method for achieving this is presented and
results are compared against experimental data
3, METHOD OVERVIEW
1od adopted uses existing methods forthe aerodynamic and mechanical parts of the
calealation, but combines them in a new way. A 3D linearised Euler method developed by
Giles (4), known as SLIQ was used for the unsteady Jeulation, The
blade surface forcing function produced by SLiQ is applied to a reduced finite element mode!
calculated using a transient dynamics method
‘nd a non-lineer model of the
odynamie parts of th
of the structure. The vibration response is thet
of Thomas & Gladwell (6) and includes both linear dampi
tunder platform damper, developed by Sanliturk (7). The novel aspect is the foreing function,‘mechanical properties and damping al vary with time using a simple scaling approach so that
fan engine acceleration ean be modelled, A flow chart shown in figure | indicates how each of
these elements interact,
Aerodynamic Method
SLiQ performs the unsteady aerodynamic calculation in wo stags
talculates the steady flow in a single
ge using non-linear Buler equations with a standard
cell vertex finite volume scheme on a structured hexahedral mesh. The solution is time
marched using a 4-stage Runge-Kutta integrator, with standard second and fourth order
smoothing.
Inst
about the steady flow. The sources of unsteadiness that be anal
hati is a harmonic perturbation
are: prescribed upstream
wakes, upidownstream potential fields and prescribed blade motion. The unsteady
perturbation equations are formulated in a similar form to th
non-finear steady equations but
one particular frequency. They can, therefore, be solved by the standard numerical scheme
sing psuedo time
used for the steady flow abo marching. The details of this procedure are
in),
In the approach presented here, two soure
of unsteadiness are treated separately and
superposed later in the analysis, Firstly for each
rodynamic damping is calculat
vibration mode in turn, This is done by interpolating the modeshape amplitude and phase
fiom a eyclc fi nent solution to the grid points atthe solid boundary of the SLIQ
mesh, Since SLiQ calculates the flow in a passage between two blades, a phase lag is applied
bet
fn inter-blade
phase angle. The flow perturbations are calculated in response to this prescribed motion, and
the aerodynamic damping is then the work done on the fluid, integrated over the solid
boundary for one vibration cycle, Thus a value of damping may be calculated for each mod
The second calculation assumes a sinusoidal disturbance at the inflow boundary as the
unsteadiness and again Uhe perturbation to the steady flow is calculated, The upstream flow
disturbance is caused by the blade rotating through a non-uniform flow field produced by the
ines, The steady flow in the nozzle is caleulated using a standard
Navier-Stokes CFD code, from which both the potential and vortical components may b
extracted. The pertrbation atthe inlet i caleulated by
a Foutier decomposition of the nozzle
steady exit flow to give magnitude and phase of each harmonic. The unsteady flow for each
but itis usually only the fundamental that is required
ney i. off resonant
as real and imaginary parts ata g
harmonie must
alewlation is a flow perturbation exp
higher harmonies would be ata differest freq esult of the
from which the forcing on the blade surface is determined,
Mechani
1 Model
A standard Finite Element method is used to model th
weture (8). The model contains &
full description of the geometry, temperature, material properties ete. and is first used to
alculate the undamped frequencies and modeshapes of the structure, When plotted on athe shaft speed and
Campbell diagram this de 1 boundary conditions for the CFD
The model is roduced using a method fi
disturbances are interpolated onto the structural model and converted into equivalent reduced
forces for each of the reduced structural variables,
Pl
posed by Guyan (9) to significantly reduce the
perturbations caused by upstream flow
A proprictary transient dynamics program (SI18) calculates the response of a structure to
applied forces using the implicit time marching method of Thomas and Gladwell (6), and can
include non-linear elements. Figure | shows that many inputs are required to SM18: line
aerodynamic and mechanical damping, non-linear mechanical e
ents, foreing function and
the definition ofthe model
The linear me
snical damping represents the inherent damping in the structure such as
material damping and damping due to friction between the blade root and the disk ete.
Dampi
but can differ widely for different
ales can be measured inthe laboratory or extrapolated from previous experienc:
mponents and different modes on the same component.
‘are based an under platform damper element developed by
surface contact parameters such as friction
coefficient and surface stiffness, which must be measured for the material combination in
ee)
The non-linear damping
Sanliturk (7). The model
ures geometric a
‘question, a representative operating conditions (temperatute, roational
‘Transient Model
12 Approach
‘The under platform dampers can cause a change in the resonant frequency, that
the resonant amplitude and fr jeted, the system is used to simulate an engine |
acceleration through the resonance, All input parameters vary with time: fore frequency and
amplitude, damper
would be typical for an engine stain gauge test. The stresses can be recovered and used in a
o to ensure
eight, material properties ele. to mimic a true engine acceleration as
blade life assessment
The most difficult paran
an aerofoil is highly depe
e. The method allows the use of unsteady pressures ats
sto scale are the unsteady pressures, In general, the flow around
dant on the operati
point and sealing should enly be used with
points
extreme c: eral operating
ye the turbin
point due tothe way it interacts with the e
is designed fo stay at @ constant non-dimensional operating
mpressor. The high pressure rotor in particul
stationed between two choked nozzles and therefore the pressure rato remains constant and
Mach umber distribution changes very litle berween idle and maximum thrust. Figu
shows a graph of HP sta
zero altitude) running conditions. The scatter is mainly due to variation in ambien
atmospheric conditions and taking this into consideration a single best fit curve can be
calculated. This curve is then used as the basis for scaling the unsteady pressures on the blade
let pressure against shaft speed for a selection of ground (4.4, CASE STUDY
A High Pressure Turbine blade, typical of modem design, was chosen to. demonstrate
essential feaures ofthe approach and highlight implication to any forced response prediction
method,
The HP stage consists of 40 nozzle guide vanes (NGVs) and 92 development standard blades.
tions and it aerodynamic behaviour changes very litle
ng speed range, T
across its useful w Yolo is shrouded (but not interlocked), and also
damper which introduces non-lineae fe
has an underplatfon 1 damping. The resonances
of interest are Ist Torsion (IT) and second flap (2F) vibration modes which are excited by
wake passing and occur at low speeds. The results are presented with and without ffetion
auge results from development engines.
Normal Vibration Modes
A 3D finite element model of a blade and dise sector was used, (See figure 3). Cyclie
(ND). The first
First torsion (IT) at $1.0% and
further investigation
suined at the dise sector boundaries for vibration with 40 nodal diameters
lated and plotted on the Campbell diagram (figure 4)
ond flap (2F) at 67.4% shaft speed were identitied for
ve modes were
Aero Forcing
Based on the Campbell diagram above, an appropriate
wamic operating point was
identified and a through flow caleulation was performed at 69.2% speed to provide boundary
conditions. The flow through the upstream nozzle guide vane was calculated using a 3D
Navier-Stokes steady code and this provides the definition of the inlet disturbance for SLiQ
Figure 5 shows the flow distortion that is prociced by the nozzle at the inlet pl
(th
SLIQ mesh, The nozzle guide vane is designed to produce a very curved wake so that Ihe
Phase relationship betwee
1 different radial positions tend to minimise vibration
A single operating point was used because previous iad shown that fora similar
3 the error in scaling from an ope
o 84% was less than 10% in
ters of response. This ercor is considerabl
variables, such as
have been us:
stud
smaller than the scatter in some of the other
Jamping and friction parameters. A larger database of results could
‘ith interpolation, but i{ was not thought necessary for the purpose of this
The unsteady lineatised part of SLiQ was used to determine the unsteady flow in the rotor
Jue to the Ist harmonic of nonuniform flow in the nozzle, The incoming "wake was
evaluated due to the combined potential and vortcal field. The unsteady flow results may be
sitown inthe form of raw real and imaginary parts ~ but these results can he hard to interpret
so the results are plotted over a number of blade pass
animated in time. The uns
sults for 4 time poinis at mid- height are plotted in figure 6a, Here the static pressure is
pte
eady static
pressure and the envelope of unsteadiness around the aerofbil at midES
hat thete is significant level of unsteadiness tn the rotor row (equivalent to about 10% of
the steady lift)
Acro Damping
The aerodynamic damping for the rotor has been
condition. This was done by interpolating the vib
‘one of the blades onto the aerodynamic m
all modes is very small a etitc
a dynamic m
alculated for the sume steady flow
tion modes from the cyclic FE model of
It was found that the aerodynamic damping in
damping ratio, typically of 0.03%, which is equivalent to
nifiation factor, Q, of 1500. These levels of damping are dwarfed by the
werent” damping which ranges from ¢0.25% to C=1.0%. These values of
‘mechanical dari ments of the blade, held in a block
whose fire profile matches that ofthe blade frtree, This experimental set-up is not identical
to the engine configuration, but exper reement with
‘measurements obiained by engine run
are based ot laboratory mea
ence has shown that they are in broad
Model Reduct
The full FE model was reduced using Guyan reduction. The full model has approximately
50000 degrees of freedom and the reduced model has just 44, but even with this huge
reduction the frequency error is less than 0.25% for the first § modes. Guyan master nodes
Were chosen to be in areas of mechanical non-linearity and additional degrees of freedom
Were included to improve frequency and modeshape match. The modeshape match was
compared visually and showed no discemable difference.
The static pressure i mapped onto the structure and reduced to apply to the generalised
ariables. This model now consists of « mass and stiffness matrix with 44 degrees ef freedom
and 44 force
wilh independent magnitude and pase
Transient Dynamie Simulation
The under platform dampers were modelled as 3 elements on each side of the platform with
the mass distributed. Each damper applies forees to the strueture based upon
geometric properties, instantaneous shaft speed and the relative displacements at its poims of
application. The relationship between displacements and force are based on a macroslip
‘model which required knowledge of the fiction coefficient, j, and surface stiffness, k, as
seen in figure 7. The hysteresis loop shown
can be considered as a sifness part (linear
relation between fore
and displacement), which causes the rise in frequeney; and a damping
part (the area within the loop) which causes the reduetion in amplitude. The damper elements
fare connected between two points one of which is under the blade platform, atthe true point
‘of application, and the other is connected to earth, In the eal configuration the damper would
be connected to the adjacent blade, but its amplitude and phase are unknown. Mistuning
effects in these high nodal diameter modes are significant and itis likely that the adjacent
blade has a slightly different resonant frequency and is therefore not at resonance atthe same
time. This situation is more akin to the earth attachunent condition than the assuming the
adjacent blade hasthe same amplitude asthe blade of interest but at differemt phase4 graph of stage inlet pressure was plotted fora variety of engine manoeuvres (
ind is then normalised to the operating point ofthe CFD calculations, Tis was then v
factor to seale the reduced forces to other conditions. This is equivalent to scaling th
unsteady pressure amplitude on the blade surface based on the inlet total pressure, whilst
keeping the pl
A rotor a
period of 4 seconds, The excitation frequen
is therefore ramped accordingly to produce the
49 engine order excitation line and other parameters are scaled based on the instantaneous
shalt spez
5, RESPONSE RESULTS
An example of the damped results in the form of "shot for the friction damped case is
shown in figure 8, This form of post-processi
Campbell diagram and i
intezpretation, The central plot, labelled Envelope, shows the envelope of the entire time
ooks rather complicated but i i basicaly a
similar to the form sed for engine strain gauge result
signal, This time signal is broken into small blocks and converted to the frequency domain
analysis and plots the amplitude at each freq
The main part of the figure (top-left) uses the results of the Fourier
logarithmic prey scale, The
other two views, labelled Peak-hold and Component are different views on the same data
The peakchold graph shows the maximum amplitude along a horizontal line (constant
raph show the maxinsun for vertical Jie (constant time
frequency), and the compone
e order line ean be clearly seen. The modes of vibration can also be seen
(almost) horizontal lines, and wher
be clearly seen inthe C
these eross the 40 engine order excitation line a peak
k-hold views,
ponent and P
The predicted amplitudes fom the transient dynamic analysis are shown in table 1
Hose [Swan Gomer] Sneed] Free
ampituse fmm] rd | eh
[Drcamped oa | soe pseore
Pamper Doe] Ese |,
0.008 | ra3% |ss963
cosa | 115.7% Jerezo
Table 1: Results fron
forced response ansAmplitude Correlation,
‘Strain gauge engines tests on Test_A and Test_B gave measurements of vibration amplitude
for First Torsion(1T) & Second Flap (2F) modes. These 2 tests are nominally identical in al
major respects. Data from both engine tests is compared against the prediction in table 2, and
this data is shown graphically in figure
Sood ArpTde PAT
TREO] OED
Mex] 0004 | ores
Mesn| 0,005 8.0099,
Ma] 0012 cores,
Mean] o008e | 0.0108
Freaces
undamped) 0.012 7
damped] 0.010 0.008
Table 2: Mensured Amplitudes compared to Predictions
When the damped pres ed with the mean measured response a good
agreement can be s ween the undamped and daraped predictions shows the
effectiveness ofthe damper for each mode. In this case the damper is much more effective for
the second flap (2F) mode than fr the first torsion mode (IT) and this agrees with previous
ence from development engines. The scatter in the experimental results is not typical
but itis certainly not unusual and highlights the difficulties of method validation,
‘The prediction for the undamped case appears to agree well with the measured maximum. It
is possible thatthe worst blade in an engine set has the highest amplitude because the damper
is not working effectively but this conclusion is flawed because it neglects the amplitude
scatter caused by aerodynamic and mechanical mistuning
Frequency Shift Correlation
The effect of the under-platform damper on fiequency is also of key importance. The
predicted frequeney shift due to the damper is taken from the above tables and compared
against experimental results,
Taae Fraguerey Sa
Predicted —[— Expertmontal
7 O13
F 6. 18.37%
2 0 ata
Predicion undaned ve, damped resonant fea
Experimental - damped resonant req (in engine) vs clita flab)
able 3: Measured Amplitudes compared to Predictions
571067190