AIAA 2005-4449

Towards Rocket Engine Components with Increased

Strength and Robust Operating Characteristics
Bogdan Marcu
Ali Hadid
Pei Lin
Daniel Balcazar
The Boeing Company
Rocketdyne Propulsion and Power
6633 Canoga Avenue
Canoga Park, California
Man Mohan Rai
NASA
Ames Research Center
Mofett Field, California
Daniel J. Dorney
NASA
Marshall Space Flight Center
Huntsville, Alabama
41
st
AIAA/ASME/SAE/ASEE Joint Propulsion
Conference & Exhibit
Abstract
High-energy rotating machines, powering liquid
propellant rocket engines, are subject to various
sources of high and low cycle fatigue generated
by unsteady fow phenomena. Given the
tremendous need for reliability in a sustainable
space exploration program, a fundamental
change in the design methodology for engine
components is required for both launch and
space based systems. A design optimization
system based on neural-networks has been
applied and demonstrated in the redesign of the
Space Shuttle Main Engine (SSME) Low
Pressure Oxidizer Turbo Pump (LPOTP) turbine
nozzle. One objective of the redesign efort was
to increase airfoil thickness and thus increase its
strength while at the same time detuning the
vane natural frequency modes from the vortex
shedding frequency. The second objective was
to reduce the vortex shedding amplitude. The
third objective was to maintain this low shedding
amplitude even in the presence of large
manufacturing tolerances. All of these objectives
were achieved without generating any
detrimental efects on the downstream fow through
the turbine, and without introducing any penalty in
performance. The airfoil redesign and preliminary
assessment was performed in the Exploration
Technology Directorate at NASA ARC.
Boeing/Rocketdyne and NASA MSFC
independently performed fnal CFD assessments of
the design. Four diferent CFD codes were used in
this process. They include WILDCAT/CORSAIR
(NASA), FLUENT (commercial), TIDAL (Boeing
Rocketdyne) and, a new family (Aardvark/Phantom)
of CFD analysis codes developed at NASA MSFC
employing LOX fuid properties and a Generalized
Equation Set formulation. Extensive aerodynamic
performance analysis and stress analysis carried
out at Boeing Rocketdyne and NASA MSFC
indicate that the redesign objectives have been fully
met. The paper presents the results of the
assessment analysis and discusses the future
potential of robust optimal design for rocket engine
components.
Copyrights etc Man Insert 1.
1.Introduction
One of the key requirements for sustainability of
Space Exploration is the operational reliability and
robustness of the systems deployed and used in
Towards Rocket Engine Components with Increased Strength and Robust
Operating Characteristics
Bogdan Marcu, Ali Hadid, Pei Lin, Daniel Balcazar
The Boeing Company
Rocketdyne Propulsion and Poer
!!"" Canoga A#enue
Canoga Park, Cali$ornia
Man Mohan Rai
%A&A
Ames Research Center
Mo$$ett 'ield, Cali$ornia
Daniel () Dorney
%A&A
Marshall &pace 'light Center
Hunts#ille, Ala*ama
space. In particular, the operation of propulsion
systems based exclusively in space essentially
requires systems which need minimal or no
maintenance. This requirement forces the
elimination of the current philosophy of operation
for reusable equipment, which is based on
overhaul and repair at intervals in time dictated
by the limits in the reliability of the system!s
components. In space there is limited or no
access to repair or replace worn or dama"ed
components. The absence of robust and reliable
components will necessitate propulsion systems
based on hi"h redundancy with the associated
cost and wei"ht penalties.
These shortcomin"s can be avoided by a
chan"e in desi"n paradi"m for the components
of space based propulsion systems. #ere we
propose a ma$or role for formal multiob$ective
optimi%ation methods in desi"n optimi%ation by
embeddin" them in the standard desi"n
process. In order to obtain component desi"ns
that are simultaneously characteri%ed by hi"h
performance, and hi"h stren"th and robustness,
the desi"ner must cope simultaneously with a
lar"e number of desi"n variables. The human
mind can only simultaneously address a limited
number of such variables in the context of a
conventional desi"n process. &ormal
optimi%ation al"orithms can handle a lar"e
number of desi"n parameters and meet
competin" desi"n requirements and ob$ectives
The particular pro$ect described in the present
report exemplifies how such an approach works
and demonstrates the ability of the proposed
desi"n paradi"m in providin" a revolutionary
increase in component stren"th, reliability,
performance and robustness. Such
improvements will be routinely required by future
space based systems.
2.General description
#i"h ener"y rotatin" machinery employed in
liquid propellant rocket en"ines are sub$ect to
many sources of hi"h cycle fati"ue. 'henomena
such as cavitation, manifold flow instabilities,
statorrotor interaction, and vortex sheddin"
"enerate unsteady forces of various frequencies
sub$ectin" the hardware to hi"h unsteady stress
levels.
'articular attention has been focused on the
phenomenon of vortex sheddin" at the trailin"
ed"e of turbine airfoils. (ost published studies
address cascade performance, with air or "as as a
workin" fluid. #an and )ox *+, present, in
quantitative detail, the sheddin" phenomenon on a
turbine no%%le airfoil. They measure a ran"e of
Strouhal numbers at successively increasin"
cascade dischar"e velocities. (ore and -dhye *., ,
Sieverdin"*/, 0, and 1rowand *2, showed that the
sheddin" frequency and the si%e of the vortices are
influenced by the nature of the boundary layer.
These vortices have a si"nificant effect on the
shape and depth of the wake. &urthermore, the
characteristics of vortex sheddin" have an impact
on the interaction with the blade downstream.
)ontini et al *3, and Sieverdin" *0, su""est that the
unsteady effects associated with the trailin" ed"e
vortices are insufficiently understood and
consequently not yet properly modeled. Sondak
and 4orney *5, 6, showed that for a typical turbine
cascade, the correct simulation of the flow
separation at the trailin" ed"e requires a "rid of
si"nificant density 7++8 points at the trailin" ed"e
re"ion9. -lso, in a sta"e confi"uration, the sheddin"
frequency may either lock onto an upper harmonic
of the blade passin" fundamental frequency, or split
into amplitudes at several harmonics.
:hile a lar"e body of published literature
addresses "as or air flow measurements, there is
little available material re"ardin" vortex sheddin"
on cascades of airfoils;hydrofoils operatin" in liquid
flows. <ee, #ah and <oellbach *8, performed a
=->S )&4 analysis of the unsteady flow
interaction inside an axial flow pump sta"e. :hile
blade load comparisons with experiments are
presented, quantitative data pertainin" to trailin"
ed"e vortex sheddin" is not included in this
investi"ation. 1usby et al *+?, present numerical
and experimental results for the same "eometry
with more details on the vortex sheddin" associated
with the rotor blade row of the axialflow pump.
Their study shows rotor trailin" ed"e sheddin"
frequencies varyin" with radius 7and blade
thickness9 and lockin" into the vane passin"
frequency harmonics. )iocan et al *++, published
detailed optical measurements 7<4@ and 'I@9 of
the flow throu"h a radial hydraulic turbine, however,
without addressin" the vortex sheddin"
phenomenon.
The present report addresses the issue of a very
ener"etic trailin" ed"e vortex sheddin"
phenomenon in an axial hydraulic turbine operatin"
in liquid oxy"en, namely the turbine powerin" the
<ow 'ressure Oxidi%er Turbo 'ump 7<'OT'9 of the
Space Shuttle (ain En"ine 7SS(E9. The analysis
presented here has been made in the very
conservative context of addressin" a fli"ht safety
issue for the Space Shuttle. In this study we
present the )&4 calculations carried out in
support of a revolutionary redesi"n of the
<'OT' turbine no%%le. Several )&4 codes have
been run simultaneously in order to determine
and compare the frequency and amplitudes of
sheddin" induced pressure fluctuations for the
nominal and redesi"ned airfoils.
Insert . (an
1ased on the )&4 calculations, structural
dynamics and stress calculations have been
made in order to determine the expected safety
factors and the expected life for the new
component,and the impact on the entire
turbopump..
&i"ure + shows a side view of the Space Shuttle
(ain En"ine 7SS(E9, and the location of the
<ow 'ressure Oxidi%er Turbo 'ump 7<'OT'9.
&i"ure . shows a simplified schematic of the
en"ine. - cross section of the <'OT' is shown
in &i". /. The pump side of the <'OT' consists
of a sin"le piece inducer powered by a six sta"e
hydraulic turbine. The turbine is fed via a tapoff
line from the dischar"e of the #i"h 'ressure
Oxidi%er Turbo 'ump 7#'OT'9. The dischar"e
of the <'OT' turbine is recirculated back to the
inlet of the #'OT'.
&i"ure 0 shows the exact location of the <'OT'
turbine no%%le. (etallur"ical inspections of the
no%%le parts have found evidence of hi"h cycle
fati"ue 7#)&9 at the no%%le trailin" ed"e near
the end walls. -nalysis of the known sources of
excitation of #)& points to vortex sheddin" as
the most probable cause. Stron" vortex
sheddin" can "enerate flappin" of the trailin"
ed"e as shown in the detail in &i".0. Indeed,
)&4 analyses which will be presented in the
next sections show that sheddin" frequencies
are close to the the blade trailin" ed"e flappin"
mode natural frequency..
The <'OT' turbine no%%le component is
currently bein" replaced at carefully monitored
time intervals ensurin" full safety for the Shuttle
fli"hts. If the same turbopump were to operate
within a space based system, the #)& wear of
the part would limit the operational life of the
overall system. )learly, it is imperative that
reliability and robustness are built into the
components for space based propulsion
systems.
3.CFD Analysis
The flow analyses presented in this study have
been performed usin" several )&4 codes.The
>-S- (S&) )orsair family of codes were used
durin" the initial phase of the pro$ect These codes
solve the timedependent, threedimensional
=eynoldsavera"ed >avierStokes equations. The
numerical al"orithm used in the computational
procedure consists of a timemarchin", implicit,
finitedifference scheme. The procedure is spacially
thirdorder accurate and temporally secondorder
accurate. The inviscid fluxes are discreti%ed
accordin" to the scheme developed by =oe *+.,.
The viscous fluxes are calculated usin" standard
central differences. -n approximatefactori%ation
technique is used to reduce and simplify the
matrices which need to be solved at every time
step. >ewton subiterations are used at each time
step to enhance stability and reduce lineari%ation
errors. The equations of motion are extended to
turbulent flows usin" an eddy viscosity formulation.
The turbulent viscosity is calculated usin" the two
layer 1aldwin<omax al"ebraic turbulence model
*+/,. The computational procedure uses O and #
type %onal "rids to discreti%e the flow field and
facilitate relative motion between rotor and stator
rows. The O"rids are bodyfitted to the surfaces of
the airfoils and "enerated usin" an elliptic "rid
"enerator. They are used to accurately resolve the
viscous flow in the blade passa"es and to easily
apply the al"ebraic turbulence model. The
al"ebraically"enerated #"rids are used to
discreti%e the remainder of the flow field. 4etails of
the al"orithms and the "riddin" methodolo"y can
be found in references *+0, +2 and +3 ,.
)alls to (essa"e 'assin" Interface 7('I9 and
Open(' parallel computation libraries have been
implemented in the code to reduce the computation
time for lar"escale threedimensional simulations.
The use of ('I allows the couplin" of different
"eometric components, such as a turbine cavity, in
a strai"htforward manner.
The )orsair family of codes has been exhaustively
validated for "as turbine analysis. In the present
study however, the codes have been used to
analy%e a turbine "eometry operatin" in liquid.
:hile the input parameters have been carefully
crafted to best approximate the actual flow
conditions, the speed of sound in the workin" fluid
could not be accurately scaled. 4ue to this
limitation, analyses of the <'OT' turbine have
initially been limited to the first turbine sta"eA further
expansion of the computational domain to include
more sta"es downstream would have led to density
effects that would result in a departure of the
computed solution from the actual physics.
:ith the above limitations in mind, a newer code
has been used for the present analysis. The
>-S- (S&) -ardvark code employs a
compressible;incompressible formulation based
on the Benerali%ed Equation Set *+5, +6,
formulation for the =eynolds -vera"ed >avier
Stokes equations. - preconditionin" al"orithm is
used for incompressible flows. The new
al"orithm utili%es the true thermodynamic
properties of the workin" fluid. - library of real
fluid properties has been implemented into the
code.
The results obtained with )orsair and -ardvark
are very similar, displayin" the same trends and
chan"es in pressure fluctuation frequencies and
amplitude. #owever there are small quantitative
differences in surface pressures, sheddin"
frequencies and amplitudes. The two codes are
to some extent numerically related to each other
7the airfoil topolo"y is modeled with the same
type of "rids, while the %onal boundary condition
implementation follows similar lo"ic, and the
same turbulence model is used9. #ence two
additional codes utili%in" different turbulence
models and based on unstructured "rids have
also been used for analysis.
The Tidal code, which stands for Time Iterative
4ensity;pressure based -l"orithm, is a =->S
code developed at =ocketdyne. The code
utili%es a finite volume, multi%one method, and
a steady;unsteady modulari%ed flow simulation
al"orithm. - unified approach is employed to
combine the density and pressurebased
methods to enable the computation of flow
fields ran"in" from incompressible to supersonic
flows. In the present analysis, a dual time
steppin" method is used to obtain a time
accurate solution. - central difference scheme
is applied to the convection terms and viscous
term. -n adaptive second order dissipation
method is employed for smoothin". The .
equation kC turbulence model is used.
'recondition is used in the analysis. 'arallel
processin" on a <inux cluster enables a rapid
turnaround.
The commercial code &<DE>T has also been
used here for .4 sin"le airfoil and .4 full sta"e
analysis. &<DE>T is a "eneral purpose
computer pro"ram for modelin" fluid flow and
heat transfer in complex "eometries. It provides
complete mesh flexibility in solvin" flow
problems with unstructured meshes in .4 and
/4 "eometries usin" trian"ular;tetrahedral,
quadrilateral;hexa"onal, or mixed 7hybrid9 "rids that
includes prisms and pyramids. The mesh can be
"enerated about complex "eometries with relative
ease usin" the preprocessor packa"e B-(1IT.
&<DE>T allows for multiple movin" reference
frames, includin" slidin" mesh interfaces and
mixin" planes for the modelin" of rotor;stator
interaction. The se"re"ated solver formulation was
used to solve the continuity, momentum and scalar
equations sequentially. Second order spatial
accuracy and implicit secondorder time steppin"
were used to reduce spatial and temporal
discreti%ation errors.
- refined unstructured mesh near the blade surface
and downstream of the trailin" ed"e to"ether with a
small time step si%e were used in order to
accurately capture the sheddin" frequency and
amplitude of the <'OT' turbine no%%le. Breat care
was exercised to ensure "rid and timestep
independent solution by refinin" the "rid and
choosin" a small time step of the order of +;2? of a
sheddin" period. The number of subiterations per
time step was chosen lar"e enou"h to ensure
conver"ence at each time step.
Turbulence effects were captured usin" the
unsteady =eynolds -vera"ed >avierStokes
7D=->S9 full =eynolds stress model 7=S(9 of
<aunder, =eece and =odi *+8,. In this model the
full transport equations for the turbulent =eynolds
stresses are solved to"ether with the continuity and
momentum equations. The slidin" mesh capability
of &<DE>T is used to predict the time dependent
flow throu"h a .4 rotorstator blade row. The time
varyin" rotorstator interaction is modeled by
allowin" the mesh associated with the movin" rotor
to translate 7slide9 relative to the stationary mesh
associated with the stator blade. Initially a steady
flow calculation with a stationary rotor was initiated
and the solution obtained was used as a startin"
solution for the time dependent slidin"mesh
calculation. Dnsteady lift forces on the rotor and
the stator blades were monitored to determine
when the unsteady flow predictions became time
periodic and independent of the initial condition.
4.Nominal Geometry Analysis. HCF
induced by vortex shedding
&i"ure 2 shows the flow field throu"h the <'OT'
no%%le cascade obtained usin" :ildcat. Entropy is
used to visuali%e the vortex sheddin" patterns at
the airfoil!s trailin" ed"e. The result was obtained
from a simulation of the flow throu"h a 0/ airfoil
cascade, at a pressure drop correspondin" to +?8E
=ated 'ower <evel 7='<9 of the SS(E. Dnlike
cylinder vortex sheddin", the sheddin" at the
trailin" ed"e of the no%%le creates alternatin"
vortices of unequal stren"thF the vortex released
from the pressure side of the vane is stron"er.
The effect of the pressure unsteadiness is felt
upstream of the trailin" ed"es, as seen on the
vane pressure loadin" envelopes in &i". 3.
Si"nificant unsteadiness at the trailin" ed"e
re"ion can be deduced from the lar"e
excursions in the pressure envelope. The effect
is felt on the suction side for almost two thirds of
the blade chord from the trailin" ed"e. The
frequency and amplitude of the vortex sheddin"
are shown on the &&T dia"ram in &i". 5. &ifty
points out of the /?+ "rid points around the
blade were sampled, and the location with the
maximum amplitude of pressure fluctuation was
selected 7the point is located at the tan"ency
point where the trailin" ed"e circle meets the
pressure side of the airfoil9. The results indicate
a point pressure fluctuation of G+6+ psi at a
frequency of 00,+52 #%. This lar"e amplitude
diminishes rapidly as one moves away from the
point of maximum amplitude 7the correspondin"
tan"ency point on the suction surface has a
sli"htly lower amplitude at the same frequency9.
#ence a more meanin"ful measure of unsteady
stress is the frequency and amplitude of
fluctuatin" pressure avera"ed over a ?.+?? inch
width section alon" the airfoil!s trailin" ed"e.
This avera"ed measure has an amplitude of G
25 psi at the same frequency of 00,+50 #%.
&i"ure 6 shows the flow field obtained from a .
4 )&4 :ildcat simulation of a sta"e
confi"uration. The first sta"e of the <'OT'
turbine is modeled with 00 no%%les and 33
blades 7.no%%les and /blades9 operatin" at
+?8E ='< of the SS(E. The vortex patterns
due to sheddin" are clearly visible in the wakes
all the way throu"h the rotor row inlet. There, the
vortices interact with the rotor blade, impin"in"
on the leadin" ed"es, and then bein" convected
downstream throu"h the rotor passa"e. The
spectral content of the pressure fluctuation at
the tan"ency point on the pressure side of the
no%%le is shown in &i"ure. 8. The results
obtained from the :ildcat and -ardvark
simulations are compared in this fi"ure. The data
show fluctuatin" pressure amplitudes of G.+.
psi at a frequency of 02065 #% from :ildcat and
G.?6 psi at 0/28. #% from -ardvark. The
frequency and amplitude values are very close
for the two codes.
- comparison of the results obtained usin" :ildcat
for the cascade and sta"e confi"urations 7&i"ure 5
and 89 show differences in sheddin" characteristics.
The sheddin" frequency for the sta"e confi"uration
is sli"htly different than the sheddin" frequency for
a vane in isolation because of a lockin effect. The
sheddin" frequency locks on to the closest upper
harmonic of the vane passin" fundamental
frequency, or to the closest upper harmonic and a
half *+?,. In this case the :ildcat simulation
indicates a lock into the +.
th
harmonic, while the
-ardvark simulation indicates a lock into the ++
th
H
I. It should be noted that the results obtained with
the two different codes are very close. The Strouhal
number based on local velocity at the trailin" ed"e
and the trailin" ed"e thickness is St J ?.../?../2.
The pressure amplitude avera"ed over the ?.+??
inch section at the trailin" ed"e is G5. psi 7-ardvark
simulation9.
&i"ure +? shows the no%%le vane response as a
function of frequency for the trailin" ed"e flap
mode, and the spectral content of the vortex
sheddin" phenomenon at +?8E en"ine =ated
'ower <evel 7='<9. The vane peak resonance
response value displayed corresponds to the
natural frequency of the nominal "eometry as
measured in the lab at 05 k#% without corrections
for <OK mass flow effects which can account for
.?E0?E reduction in frequency *.+,. -dditional
uncertainty comes from the actual hardware which
has lar"e deviations from the nominal airfoil
"eometry. Overall, with all the corrections and
variations due to hardware "eometry deviations, the
blade natural frequency associated with the trailin"
ed"e flap mode can ran"e anywhere between .0
k#% and 03 k#%. The vortex sheddin" frequency
70/.3 k#%9 in &i"ure +? represents the value
obtained with -ardvark for the nominal "eometry.
Several )&4 calculations have been performed for
no%%les with variations in "eometry representative
of hardware variations in vane trailin" ed"e
thicknesses. &urthermore, )&4 calculations for
cylinders have been performed *.?, at similar
=eynolds numbers 7based on local TE velocity and
TE diameter9 in order to calibrate the code output
for fluctuation frequencies. The results indicate an
overprediction of sheddin" frequency by +2.?E.
These uncertainties in the computed data and the
hardware "eometry result in a sheddin" frequency
ran"e between .6 k#% and 02 k#%. Thus the
sheddin" frequency ran"e and the vane natural
frequency ran"e overlap. This overlap and the lock
in mechanisms observed many times in operation
*5, 6, +?, are a stron" indication that vortex
sheddin" is the ma$or cause for #)& wear and
dama"e observed in the operation of the
<'OT'.

5.The traditional redesign approach
and its shortcomings
The initial approach taken to address the
resonance problem consisted of a LretrofitM of
the existin" shelf parts. The details are
discussed at len"th in reference *.+,. At the
expense of a small percentage of the overall
turbine performance, a shortening of the LPOTP
nozzle airfoil via machining was considered for
the purpose of decoupling the vortex shedding
from the fap mode of the vane TE. Shortening
of the blade, results in a thicker trailing edge,
and consequently, the vortex shedding
frequency decreases, while the stifness and
hence the trailing edge fap natural frequency
increase.
In the following we will briefy present the
shortcomings of this retroft approach.
After several trade-of analyses, a vane cut of
0.100 inch in length at the trailing edge was
found to provide a sufcient increase in trailing
edge thickness. Figure 11 shows the baseline
and retroft frst stage turbine geometries. The
cut cannot be accomplished uniformly over the
entire vane span because the cutting tool has no
access near the end walls. Consequently there
is a round cut towards the walls.
Figure 12 shows a snapshot of the fow feld
through the turbine stage for both the baseline
and retroft geometries at midspan. The thicker
trailing edge of the retroft nozzle and the sharp
edges left after the cut (only limited chamfer can
be achieved) produce a lower vortex shedding
frequency (evidenced by the lower density of
blobs convected downstream) but stronger
vortices (evidenced by the sharper color contrast
in the plot ###).
Figure 13 displays the quantitative efects of the
modifcation. On the one hand, the objective of
reducing the shedding frequency is achieved for the
nozzle vane, as shown in Figure 13a: the shedding
frequency has been lowered by about 17 kHz, from
44 kHz to 28 kHz. Modal analysis and lab tests for
the modifed nozzle vane indicate trailing edge fap
mode natural frequencies as high as 65-70 kHz,
which reduces to about 52 kHz in operation after
accounting for the LOX mass fow efect. Thus a
frequency separation of 24 kHz is achieved.
Paragraph required. (couple to ###)
Unfortunately, Figure 13b shows that the nozzle
problem has been solved at the expense of the
rotor blade downstream. Stronger vortices shed by
the cut back vane at lower frequency are convected
into the rotor passage. These vortices impinge on
the rotor with enough energy to create a strong
disturbance at a frequency of 29 kHz. The
frequency of this disturbance lies in the middle of
the frequency range for the rotor blade frst bending
mode natural frequency. This is not acceptable.
Therefore, the retroft solution has been rejected.
It is expected that the redesign of the baseline
airfoil using traditional design techniques will yield a
new and thicker airfoil that has a lower shedding
frequency but a larger shedding amplitude. This
amplitude may not be as large as that produced by
the cutback vane, but large enough to surface as a
disturbance amplitude at an undesired frequency
on the rotor blade downstream.
I3 insert Man
6.New, extended design requirements
The desi"n requirements for a new no%%le airfoil
should therefore contain additional provisions for
preventin" or limitin" the introduction of unintended
downstream perturbations. Obviously the overall
turbine performance must be preserved in order to
ensure the proper operation of the turbopump
component.The complete set of desi"n
requirements can be summari%ed as followsF
The airfoil should be as thick as possible at the
TE, thereby increasin" the natural flap mode
frequency, and decreasin" the sheddin"
frequency, thus decouplin" the two
frequencies in operation
Steady state 7time avera"ed9 flow conditions
at no%%le row dischar"e should be preservedF
N same no%%le throat area in order to
conserve <OK mass flow value.
N same dischar"e flow an"le in order to
conserve downstreamrotor blade work.
-dditional flow disturbances should not be
introduced downstreamF
N control of flow separation at the no%%le TE
N diminished the sheddin" amplitude
-irfoil sheddin" should be insensitive to
"eometry discrepancies due to manufacturin"
or wear and tear
N Expected manufacturin" methodF castin"
with G ?.??3 inch tolerance with lar"e
variations
The last requirement has been added as a result
of lar"e variations in the current hardware
"eometry with the intention of partially removin"
the operational performance sensitivity to airfoil
"eometry deviations.
This set of requirements is extremely complex,
and difficult, if not impossible to achieve within
the context of traditional desi"n techniques.
Thus, it constitutes an excellent case for formal
desi"n optimi%ation methodolo"y. 4rawin" on
previous results from collaborations between
>-S- -=) Exploration Technolo"y 4irectorate
and =ocketdyne, the SS(E <'OT' >o%%le
redesi"n team decided to address the redesi"n
task by utili%in" desi"n optimi%ation technolo"y
developed at >-S- -=). The airfoil redesi"n
and preliminary assessment was performed in
the Exploration Technolo"y 4irectorate at >-S-
-=). 1oein";=ocketdyne and >-S- (S&)
independently performed final )&4
assessments of the desi"n
7.Optimization methodology. The new
airfoil geometry
*(an, insert 0 H fi"s,O
Figure 14 shows the new design overlaid on the
baseline nozzle design for comparison. This
geometry is the result of fve design iterations
and is referred to as O5 in the rest of the text. The
airfoil is very thick, and in particular, the trailing
edge thickness is increased by about 60%
Intuitively one would expect such an increase in
thickness to introduce a signifcant change in the
wake profle and consequently a change in the fow
downstream of the nozzle row. A signifcant amount
of analysis, presented in the following sections, was
performed to determine if the new airfoil produced
any detrimental changes in the fow downstream.
8.CFD Analysis of performance
8.1Nozzle Performance
&i"ure +2 presents a comparison of the no%%le
vane pressure loadin", for the baseline and
optimi%ed airfoils. One observes that while the
baseline airfoil loadin" occurs mostly on the last
two thirds of the axial chord, the optimi%ed airfoil
distributes the load more uniformly. - vertical line
marks the proximity of the trailin" ed"e in both
plots. - si"nificant difference exist in the pressure
difference between the pressure side and the
suction side at the marked locationA a lar"er p for
the baseline "eometry and a diminished p for O2.
&i"ure +3 is a carpet plot comparin" the level of
unsteadiness in pressure oscillations at different
locations on the airfoils. The plot should be
interpreted as followsF the x axis represents
frequency, in #%, the y axis represents locations
around the airfoil, startin" and endin" at the leadin"
ed"e, and the % axis represents the amplitude of
the pressure fluctuation at each point. The plot
shows a reduction in sheddin" frequency from 0/.+
k#% to /..5 k#%, and a remarkable reduction in the
overall amount of unsteadiness for the O2 profile.
The new reduction provides a reduction of 53E in
peak amplitudes near the trailin" ed"e.
The data in fi"ures +2 and +3 were obtained from
-ardvark
&i"ure +5 shows a similar comparison but obtained
from a /4 Tidal simulation at midspan. In this
computation pressure fluctuations were recorded at
a location near the vane trailin" ed"e, where
-ardvark and :ildcat showed maximum amplitudes
for the baseline and O2 airfoils. The plot shows the
fluctuations for O2 are about 2?E less than the
fluctuations for the baseline, with a correspondin"
frequency decrease from /2 k#% to .2 k#%.
&i"ure +6 shows results obtained from .4
&<DE>T computations, where data has been
sampled as in the Tidal computation, i.e. at a
location near the vane trailin" ed"e, where
-ardvark and :ildcat showed maximum
amplitudes for the baseline and O2 airfoils. This
plot shows a 2.E reduction in amplitude for the
O2 vane with a correspondin" reduction in
frequency from 23 k#% to 05 k#%. &<DE>T
yields sheddin" frequencies which are clearly
hi"her than physical, "iven the calibration runs
made for cylinders usin" -ardvark 7no
calibration runs were made for cylinders usin"
TI4-< or &<DE>T9.
The important feature in figures 16 through
18 is a consistent trend obtained with a
variety of CFD codes. The O5 vane reduces
pressure fuctuation ampitudes by 5!"#5$
whie reducing the shedding fre%uency by
about 1! &'(.
8.2Overall Turbine Performance
&i"ures +8 and .? compare the +
st
sta"e rotor
blade loadin" and .
nd
sta"e stator loadin" at
+?8E en"ine ='<. Each plot overlays the
pressure loadin" for the confi"urations usin" the
baseline and the O2 no%%les. 1oth plots
indicate insi"nificant chan"es in rotor blade and
stator vane loadin"s downstream. It is
reasonable to assume that if only small chan"es
can be observed in the two downstream rows
followin" the replaced no%%le, even smaller
chan"es in performance are to be expected
further downstream and therefore the no%%le
replacement produces little or no chan"e in
turbine performance
8.3Downstream Unsteady Flow Analysis
&i"ure .+ shows a comparison of the unsteady
tan"ential force on the +
st
sta"e rotor blade
obtained from -ardvark. The &&T analysis for
the time varyin" tan"ential force is shown in this
fi"ure. The fundamental frequency for the sta"e
confi"uration is the vane passin" frequency of
/.58 k#% correspondin" to 00 no%%le vanes in
the simulation. The use of O2 instead of
baseline no%%le airfoil decreases the amplitude
at this frequency from G6.6E of the mean value
to G3.6E. -dditionally, O2 does not introduce
additional disturbances other than the small
amplitude disturbances similar to those
observed for the baseline desi"n. This is a
si"nificant accomplishment "iven that the TE
thickness of O2 is si"nificantly lar"er that that of the
baseline airfoil.
Biven the importance of limitin" or perhaps even
reducin" the disturbances introduced downstream
compared to the baseline case, the Tidal and
&<DE>T codes were also used in this analysis. The
results of these analyses are "iven belowF
&i"ure .. shows the &&T analysis of the unsteady
tan"ential force on the rotor blade obtained from
:ildcat. The plot indicates a reduction in amplitude
from G+8E to G+.E at fundamental frequency and
diminished amplitudes at hi"her frequencies with
the use of O2.
&i"ure ./ shows the same analysis usin" data
produced by TI4-<. The amplitude at the
fundamental frequency is practically unchan"ed in
this calculation 7a sli"ht reduction from G0..E to
G/.6E9 while the hi"h frequency amplitudes remain
very small 7less than ?.2E of avera"e9 with the use
of O2.
In &i"ure .0, the analysis is repeated for data
produced &<DE>T. &<DE>T shows no chan"e in
the amplitude at the fundamental frequency
7G+8.2E9 and no additional disturbances introduced
at hi"her frequencies with the use of O2.
The simulation results provided by codes with
different numerical schemes, different "rid
structures and different turbulence models all
display a consistent trendF there is a modest
benefit or no change downstream of the no((e
row when the current baseine no((e geometry
is repaced by the O5 airfoi. These results have
built confidence in the capabilities of the new
desi"n.
8.4Robustness in operation
7(an insert 29
9.Structural dynamics analysis.
-n exhaustive amount of analysis for assessin" the
structural dynamics of the new desi"n has been
performed. Only the most important aspects are
included in this paper.
&i"ure .2 shows the TE flap mode response,
corrected for a /?E <OK mass flow effect, for both
the baseline and the O2 airfoils. The frequency
ran"es associated with the vortex sheddin" at
+?8E en"ine ='< for the baseline and O2 airfoils
are also shown in the fi"ure. &low data was
obtained from -ardvark for airfoils at the extremes
of hardware "eometry variation and ad$usted by
a .?E frequency shift based on calibration usin"
cylinder simulations. The O2 performance is
treated in a conservative manner, in the sense
that althou"h this "eometry is characteri%ed by
exceptional robustness, and minimum variation
in flow performance due to "eometry variation,
the same percent statistical bounds are applied
on vortex sheddin" frequencies as for the
baseline confi"uration. The plot clearly shows
the "ood separation in sheddin" and flap mode
natural frequencies for the O2 "eometry at
+?8E en"ine ='< re"ime. - comparison of
fi"ures +? and .2 shows the si"nificant increase
in frequency separation with the use of O2.
&i"ure .3 is an elaboration of the information in
&i"ure .2 throu"hout the throttlin" ran"e of the
SS(E en"ine. The plot shows the ran"e of
frequencies associated with the baseline
"eometry in blue fields and the ones associated
with the O2 "eometry in purple. )&4 simulations
have been performed for the flow re"imes
correspondin" to 30.2E, 6?E, +?0E and +?8E
en"ine ='<, with intermediate ='< point
performance obtained via a cubic spline fit. The
data processed for the chart contains all the
corrections for the uncertainties related to the
flow conditions, <OK mass flow effects and
possible alterations in the manufactured
"eometry of the hardware, and thus the lar"e
ran"e of frequency values for both "eometries.
-"ain, the data is processed takin" the
conservative approach typical for analysis of
manrated propulsion hardware such as SS(E.
One can observe that the baseline desi"n
"eometry sheddin" frequency ran"es interfere
si"nificantly with the natural vane frequency
associated with the TE flap mode at all en"ine
='< re"imes. In spite of the conservative
estimate of the performance for O2, a lar"e
mar"in of .5E is obtained as separation
between frequencies at the hi"hest ='< re"ime
considered 7+?8E9. :ith this mar"in, the
frequencies are considered completely detuned.
10.Stress analysis.
Similar to the structural dynamics section, only
some principal results are presented here from
the lar"e amount of stress analysis done for this
pro$ect. 1esides pressure loads induced by the
steady state and unsteady aerodynamic
phenomena, additional loads "enerated by the
en"ine vibration, static loads transmitted throu"h
the turbopump support structure and mechanical
and thermal loads durin" transients have been
included in the analysis.
Table + summari%es some of the main results for
the no%%le vane, showin" the overall chan"e in
stress resultin" from the replacement of the
baseline "eometry with the O2 "eometry. The
various cate"ories of stress considered are reduced
by +8E to 3??E "eneratin" an overall increase in
the safety factor from /.2 to 3./ and thus "ivin" the
part fitted with the O2 airfoil an essentially infinite
life in operation.
&i"ure .5 shows the locations of peak cumulative
stress 7steady state H alternate9 on the rotor blade.
The locations are near the leadin" ed"e 7&i"ure .5
a9 and trailin" ed"e 7&i"ure .5 b9. The maximum
occurs in the leadin" ed"e re"ion where the
calculated factor of safety is 0.8+ for the baseline
desi"n confi"uration and 3./+ for the O2
confi"uration, an increase by .6E in safety.
These results constitute an essential element of
support for the desi"n methodolo"y for components
of space based propulsion systems. :ithout any
penalty in performance, a component desi"ned via
multiob$ective optimi%ation methodolo"y has
resulted in almost a +??E increase in stren"th,
robust operation and practically infinite life.

11.Contribution potential for Space
Exlploration Missions
The methodolo"y presented here can be applied to
airfoil desi"n for all turbomachinery equipment
developed for space operations. 1ased on detailed
analysis identifyin" all sources of intense stress,
<)& and #)&, includin" provisions necessary for
deep throttlin" of rocket en"ines, desi"n
requirements similar to the extended desi"n
requirements presented in section 3 can be
formulated. Subsequent airfoil desi"n usin"
optimi%ation methodolo"y as presented here will
ensure increased stren"th and robust operatin"
characteristics for the components desi"ned.
Biven the powerful al"orithms which form the
foundation of the optimi%ation methodolo"y
demonstrated here, one can extend its application
to other components of space based systems and
subsystems. 'rincipal candidates are those
components deemed the most fra"ile in operation,
for example, seals of various kinds, bearin"s,
in$ectors, or valves.
The same methodolo"y can be applied on a lar"er
scale, and earlier in the evolution of a mission
conceptual desi"n, for tradeoff studies. (ission
formulation, vehicle;platform architecture and
space fli"ht procedures can be parameteri%ed
usin" lar"e sets of variables modelin" all
necessary aspects. (ultiob$ective optimi%ation
can be performed in hi"hdimensional desi"n
spaces that include all the necessary variables.
Optimize-not-compromise should become the
motto of conceptual thinking for all future space
exploration architectures.
12.Conclusions.
A design methodology based on formal multi-
objective optimization technology has been
applied and demonstrated for the redesign of the
SSME LPOTP turbine nozzle. The new nozzle
design has achieved about 100% increase in
strength, signifcantly extended life in operation
and an elevated robustness in operation while
the overall turbine performance has been
maintained the same as for original design.
The design methodology is proposed as a
standard design procedure for components of
space based systems as it provides a means to
design such components with signifcantly
improved strength, reliability and robust
operating characteristics.
13.References
+ #an, <. S. and )ox, :. =., L- visual Study of
Turbine 1lade 'ressureSide 1oundary <ayersM,
ASME Journal of Engineering for o!er, @ol.
+?2, +86/, pp. 052..
. (oore, P. and -dhye, =., LSecondary flows
and losses downstream of a turbine cascadeM,
Journal of Engineering for "as #urbines and
o!er, @ol. +?8, +862, pp. 83+838.
/ Sieverdin", ). #., L=ecent pro"ress in
understandin" of basics aspect of secondary
flows in turbine blade passa"esM, ASME$
Journal of Eengineering for #urbines and o!er,
@ol. +?5, +862, .06.25.
0 Sieverdin", ). #., and #einemann, #., LThe
influence of boundary layer state on vortex
sheddin" from flat plates and turbine cascadesM,
ASME paper 68BT.83, +868.
2 1rowand, &. Q., L&actors affectin" vortex
sheddin" on airfoilsM, 'rivate communication, .??..
3 )ontini, 4., (anfrida, B., (ichelassi, @. and
=iccio, B., L(easurements of @ortex Sheddin" and
:ake 4ecay 4ownstream of a Turbine Inlet Buide
@aneM, %lo! #urbulence and &ombustion, @ol. 30,
.???, pp. ./2.56.
5 Sondak, 4. <., and 4orney, 4. P., RR@ortex
Sheddin" in a Turbine )ascade,SS 'nternational
Journal of #urbo and Jet Engines$ @ol. +3, >o. .,
+888, pp. +?5+.3.
6 Sondak, 4. <., and 4orney, 4. P., RRSimulation of
@ortex Sheddin" in a Turbine Sta"e,SS ASME
Journal of #urbomachiner(, @ol. +.+, Puly, +888,
pp. 0.60/2.
8 <ee, TT., #ah, ). and <oellbach, P., LDnsteady
&low Interaction Inside a #i"h=eynolds>umber
-xial&low 'ump Sta"eM, -I-- 'aper 86?85?,
+885.
+? 1usby, P. -., Taylor, <. Q., Pian", (. and
:hitfield, 4. <., LDnsteady /4 Incompressible &low
Interaction in (ultiple 1lade=ow TurbomachineryM,
A'AA aper )*-+,-., +886.
++)iocan, B. 4., -vellan, &. and Queny, P<.,
LOptical (easurement Techniques for Experimental
-nalysis of #ydraulic Turbine =otorStator
InteractionM, 'roceedin"s of the -S(E .??? &luids
En"ineerin" 4ivision Summer (eetin", Pune +++2,
.???, 1oston, (assachussetts.
+. =oe, '. <., RR-pproximate =iemann Solvers,
'arameter @ectors, and 4ifference Schemes,!!
Journal of &omputational h(sics, @ol. 0/, +86+,
pp. /25/5..
+/ 1aldwin, 1. S., and <omax, #., RRThin <ayer
-pproximation and -l"ebraic (odel for Separated
Turbulent &low,!! -I-- 'aper 56.25, #untsville, -<,
Panuary, +856.
+0 =ai, (. (., U>avierStokes Simulations of
=otorStator Interaction Dsin" 'atched and
Overlaid Brids,U -I-- Pournal of 'ropulsion and
'ower, @ol./, >o.2, pp./65/83, +865.
+2 =ai, (. (., UThree4imensional >avier
Stokes Simulations of Turbine =otorStator
InteractionA 'art I (ethodolo"y,U -I-- Pournal
of 'ropulsion, @ol. 2, >o. /, +868, pp. /?2/++.
+3 4.P.4orney and =.<. 4avis, L>umerical
Simulations of Dnsteady Transonic &lows in
TurbomachinesM, -I-- paper 80.6//.
+5 ). <. (erkle, @. Sankaran, 4. P. 4orney, and
4. <. Sondak, L- Benerali%ed &luid &ormulation
for Turbomachinery )omputationsM, -I-- 'aper
.??//888
+6 Sondak, 4. <., and 4orney, 4. P., LBeneral
Equation Set Solver for )ompressible and
Incompressible Turbomachinery &lowsM, -I--
paper .??/00.?
+8 <aunder 1. E., =eece B. I. and =odi :.,
V'ro"ress in the development of a =eynolds
stress closure!, P. &luid (ech., vol. 36, +852
.? 'eu"eot, P.:. and 4orney, P. 4.,
LInvesti"ation of @ortex Sheddin" behind a
)ylinder Dsin" :ildcat and -ardvark )&4
)odes.M, 'rivate )ommunication, .??/.
.+ (arcu, 1., 1alca%ar, 4., Wdenek, ).4.,
<unde, Q. P., 4orney, 4. P., L@ane @ortex
Sheddin" Effects on SS(E <'OT' TurbineM,
Proceedings of the 52
nd
JANNAF Interagency
Propulsion Committee Meeting, Las Vegas, NV,
10-13 May, 2004
&i"ure +F Side view of the Space Shuttle (ain
En"ine 7SS(E9, with the location of the <ow
'ressure Oxidi%er Turbopump 7<'OT'9
&i"ure .. SS(E Schematic.
&i"ure /. )rosssection of the <'OT'. The
turbine is fed by tappin" hi"h pressure liquid
oxy"en from the main oxidi%er pump dischar"e
and routin" it into the turbine inlet via channels
drilled throu"h the mounts. &eed pressure
durin" operation is around 0??? psi.
&i"ure 0. Schematic of the SS(E <'OT' with
the location of the turbine no%%le. Insert shows
the airfoil trailin" ed"e flappin" mode.
Figure 5: Vortex shedding on the SSME LPOTP
turbine nozzle, baseline nominal geometry.
Shown is the entropy feld. Wildcat cascade
analysis.
Figure 6: Nozzle pressure loading: time
averaged, minimum and maximum pressure
envelope. Wildcat cascade analysis.
Figure 7 FFT analysis of the pressure fuctuation at
the point of maximum fuctuation amplitude. Wildcat
cascade analysis.


Figure 8. Vortex shedding on the SSME LPOTP
turbine nozzle, baseline nominal geometry, stage
confguration. Shown is the entropy feld. Wildcat
stage unsteady analysis.
&i"ure 8. &&T analysis of the pressure
fluctuation at the point of maximum fluctuation
amplitude on the no%%le in the sta"e
confi"uration. :ildcat vs. -ardvark solution
comparison.
&i"ure +?. @ane vortex sheddin" versus vane
trailin" ed"e flap mode natural frequency. The
TE flap mode frequency value shown 705 k#%9
was measured in air, no correction is applied for
lox mass flow effect 7which would reduce the
frequency by .?E0?E9. @ortex sheddin"
frequency shown is -ardvark .4 )&4 solution,
without correction 7cylinder vortex sheddin" runs
indicate +2E.?E overprediction9.
Figure 11. Nozzle retroft design: 0.100 cut-back at
vane trailing edge. The round shape of the cut is
dictated by lack of access for the cutting tool near
the end-walls. A portion of frst turbine stage is
shown here in a model.
45487 !
" #1# psi
$il%c&t
4'118 !
" #(8 psi
A&r%)&r*
+&seline ,eo-etr.
/e%esi,n
Figure 12. Comparison between the fow through
the baseline frst stage turbine geometry (a) and
the retroft modifed geometry (b). Shown is 2-D
Aardvark CFD solution.

vane passin"
frequency
disturbance
introduced by
intensified
sheddin"
amplitude
.60/? #% G 06 psi
02 620 #% G 5. psi
vortex
sheddin"
pressure
fluctuation
&
b
Figure 13. Vane cut-back efects: a) Shedding
frequency is reduced by 14 kHz on a 0.100 band at
the trailing edge of the vane but b) a new and
strong disturbance is introduced downstream
evidenced by the large amplitude at 29 kHz on the
blade tangential force spectral analysis. 2-D
Aardvark CFD solution.
&
b
Baseline
+,
Figure 14. Vane redesign via optimization: over-
plot of baseline geometry versus optimized
geometry. Trailing edge thickness is increased
by almost 100%.
&
b
&
b
Figure 15. Vane pressure loading at 109% engine
RPL a) baseline geometry b) O5
- Baseline
- O5
psi
53E reduction
in point pressure
amplitude
comparin" max to
max amplitudes
Figure 16. Comparison of unsteady pressure
fuctuation during operation on the nozzle vane.
Data has been sampled at 50 points around the
vane from Aardvark 2-D CFD solution. O5 vane
displays signifcantly reduced unsteadiness.
/?E
TI4-<
Figure 17. Comparison of TE pressure
fuctuation due to vortex shedding obtained from
Tidal 3-D CFD. Amplitude is reduced by 30%
while frequency is reduced by 10 kHz.
2.E
&<DE>T
Figure 18. Comparison of TE pressure
fuctuation due to vortex shedding obtained from
FLUENT 2-D CFD. Amplitude is reduced by 50%
while frequency is reduced by 9 kHz.
Figure 19. 1
st
stage rotor blade time averaged
pressure load comparison between baseline and
O5 nozzle confguration. Shown is Aardvark 2-D
CFD solution.
Figure 20. 2
nd
stage stator vane time averaged
pressure load comparison between baseline and
O5 nozzle confguration. Shown is Aardvark 2-D
CFD solution.
+&seline
05
Figure 21. 1
st
stage rotor blade time unsteady
tangential force FFT analysis, Aardvark 2-D
CFD solution .
+&seline
05
Figure 22. 1
st
stage rotor blade time unsteady
tangential force FFT analysis, Wildcat 2-D CFD
solution .
+&seline
05
Figure 23. 1
st
stage rotor blade time unsteady
tangential force FFT analysis, Tidal 3-D CFD
solution .
+&seline
05
Figure 24. 1
st
stage rotor blade time unsteady
tangential force FFT analysis, FLUENT 2-D
CFD solution .
Figure 25. Vane natural frequencies associated
with the trailing edge fap mode, baseline versus
O5 comparison. The values are corrected for
LOX mass fow efect by 30%. Shown are also
the ranges for vortex shedding frequencies,
based on Aardvark 2-D CFD with 20% correction
based on calibration calculations using
cylinders.
Figure 26. Campbell diagram for the nozzle vane.
Shown are the ranges for vane natural frequencies
associated with the TE fap mode and the shedding
frequencies range and variation with engine RPL.
While baseline geometry shows possible
interference at all RPLs, redesigned O5 still has a
margin of 27% (marked with accolade) at 109%
RPL.
1e&%in, E%,e 2r&ilin, E%,e 1e&%in, E%,e 2r&ilin, E%,e
Figure 27. Maximum stress locations on the frst
stage rotor blade occur at LE and TE. Loading for
the calculations has been provided from Aardvark
2-D CFD calculations. The larger stress from the
two local maxima occurs at the LE where in O5 the
factor of safety for the rotor blade increases from
4.91 (baseline) to 6.31.
@ortex sheddin" frequency
ran"e 1aseline
@ortex sheddin" frequency
ran"e O2
>atural &requency
1aseline
>atural &requency O2
MEM+/A3E 4*si5 +E36I37 4*si5 202A1 S2/ESS 4*si5
lo&% step +1 0P 5
8
CA37E +1 0P 5
8
CA37E +1 0P 5
8
CA37E
MA9: S2/ESS 6;E 20 /A6IA1 & A9IA1 +C 7:58 5:7< ='18 ':44 ':5( #8 11:(# <:#< =1<8
MA9: S2/ESS 6;E 20 S2A20/ +10$0>> 10A6 4:(5 (:8' ='<(8 1:85 (:(? =#7778 5:8< (:8< =5?#8
MA9: S2/ESS 6;E 20 @A3E P/ESS;/E 031A 11:5# 7:84 =478 ?:?8 1:#8 =4#'8 18:#( <:11 =1((8
MA9 0PE/A2I03 SI1@E/ SEA1 +0220M A2 06 #<:5( ##:## =''8 ##:4# 15:78 =4#8 51:<# '8:(( ='78
TABL. /