Design and Optimization of Compressor Airfoils by

Using Class Function / Shape Function Methodology

Daniel Giesecke1 , Udo Stark2 , Rubén Harms Garcia1 , Jens Friedrichs1

AT
Abstract
ROT ING M
N This paper describes a new design method for high Reynolds number subsonic compressor blade
O
AC
SYMPOSIA

sections for industrial gas turbines and compressors. The focus is on the middle and end stages,
HINERY

where the Reynolds numbers are about 2 to 6 × 106 and the Mach numbers between 0.4 and
0.8. The new design method combines i) a parametric geometry definition method, ii) a fast
blade-to-blade flow solver, and iii) an optimization tool with a suitable objective function. The
development of a new blade section is based on a conventional NACA-65 design, subsequently
ISROMAC 2017 modified to an optimized CSM profile, where CSM means Class Function / Shape Function
International Methodology.
Symposium on The new profile shapes are obtained by superimposing a camber line and a thickness
Transport distribution. Both the camber line and the thickness distribution are prescribed as analytical
Phenomena and functions to cut down the CPU-time for geometry set up and to guarantee smooth geometries.
Dynamics of Numerical calculations are performed by applying the two-dimensional blade-to-blade solver
Rotating MISES. The optimization method used in this paper is the single-objective genetic algorithm
Machinery
(SOGA) from the DAKOTA library. The objective function consists of 5 components and takes
Hawaii, Maui into account the whole loss polar. The corresponding computing time is relatively short - that is
1 to 2 days.
December 16-21,
2017 At high Reynolds number, the new profiles show decreased design point losses and increased
operating limits compared to corresponding results using conventional NACA-65 profiles. In
addition, the present results show close agreement with those produced by so-called high
performance profiles of the relevant literature.
Keywords
Compressor — Design — Optimization
1 Institute
of Jet Propulsion and Turbomachinery, TU Braunschweig, Germany
2 Institute
of Fluid Mechanics, TU Braunschweig, Germany
*Corresponding author: d.giesecke@ifas.tu-braunschweig.de

INTRODUCTION number of geometric parameters (10 in [2] and 20 in [3])

which will all have to be optimized in a time-consuming
High Reynolds numbers and moderate subsonic Mach optimization process. A new method, described in this
numbers are typical for the compressor middle and end paper, has only three geometric parameters that have
stages of industrial gas turbines and compressors. The to be optimized. This, together with an analytical for-
Reynolds numbers of the order of 2 to 6 × 106 [1] are mulation of the thickness distribution and camber line,
much higher than the corresponding aeroengine numbers leads to relatively short computing times of two days
of about 0.6 to 1.2 × 106 [1]. When this was realized only. The paper concludes with the introduction and
some twenty years ago, several new methods were de- description of the two test cases - one of them a redesign
veloped taking into account the high Reynolds number of Profile No. 4 of the Design Case No. 4 in [2].
and high turbulence effects on boundary layer transition
[2, 3]. The new blade sections typically show a front
loaded pressure distribution, a boundary layer transi-
1. METHODS
tion (bypass transition) next to the pressure minimum The new design method combines i) a parametric ge-
at about 7 to 10% true chord, and a suction side deceler- ometry definition method, ii) a fast blade-to-blade flow
ation whose gradient becomes increasingly lower towards solver, and iii) an optimisation tool with a suitable ob-
the trailing edge. These features usually guarantee a jective function. In the present work, the development of
successful design with low design-point losses and low a new blade section is based on a conventional NACA-65
off-design losses and extremely wide loss polars. How- design subsequently modified to an optimized CSM pro-
ever, the existing methods have generally long computing file, where CSM means Class Function / Shape Function
times of up to two weeks [3] for a complete loss polar. Methodology [4]. The individual design steps may be
One of the reasons for this is probably the relative large summarized as follows: i) design of a conventional NACA-
 Design and Optimization of Compressor Airfoils by Using Class Function / Shape Function Methodology — 2/8

y y

t
2
f
γ ∆yT E

xf x xt x
c c

Figure 1. Camber line parameters Figure 3. Thickness parameters

Y [-] Y [-]

0.10 0.05

0.08 0.04

0.06 0.03
Xf = 0.35
Xf = 0.40 KR = 0.5
0.04 0.02 KR = 1.0
Xf = 0.45
Xf = 0.50 KR = 1.5
0.02 Xf = 0.55 0.01 KR = 2.0
Xf = 0.60 KR = 2.5
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
X [-] X [-]

Figure 2. Camber lines of max. camber f /c = 0.1 at Figure 4. Thickness distributions of max. thickness
various chordwise positions t/c = 0.1 at various chordwise positions Xt (KR)

65 profile for a prescribed set of design parameters, ii) with parameters f /c as maximum ordinate (expressed as
design of a CSM reference profile based on modified pa- fraction of chord) and Xf as chordwise position of the
rameters of the conventional NACA-65 profile, and iii) maximum ordinate, see Fig. 1. As an example, Fig. 2
search of an optimized CSM profile. shows an evaluation of the above formula for f /c = 0.1
and various values of the chordwise position Xf of the
maximum ordinate.
1.1 Blade Section Geometry The thickness distributions, as seen in Fig. 3, are
The challenge for the geometry method is to produce a derived from the following formulas representing the
fast analytical procedure with a small number of param- Class Function / Shape Function Methodololgy [4] in its
eters for the design of subsonic, high Reynolds number simplest form (scaling factor omitted).
compressor blade sections. The section shapes are ob-
tained by superimposing a camber line and a thickness Yt (X) = C(X) · S(X) + X · ∆YT E . (2)
distribution. Both the camber line and the thickness dis-
tributions are prescribed by analytical functions to cut with Class Function
down the CPU-time for geometry set up and to guarantee √
smooth geometries. C(X) = X · (1 − X) (3)
The camber lines are so-called generalized parabolic
arc lines [5] of the form and Shape Function

X(1 − X) 1
Yc (X) = a · (1) S(X) = KR · (1 − X) + · X, (4)
1 + bX KR

with a and b representing the following abbreviations where KR is a shape parameter determining the leading
edge radius RLE
1 f 1 − 2Xf p
a= and b= S (0) = 2RLE , (5)
Xf2 c Xf2
 Design and Optimization of Compressor Airfoils by Using Class Function / Shape Function Methodology — 3/8

the trailing edge angle γ

ω
S (1) = tanγ + ∆YT E (6) ∆β1
∆β1,stall
and the maximum thickness location Xt
p 2 ωD
Yt0 (Xt ) = (Xt ) · (1 − Xt ) · S 0 (Xt ) (7)
 
p 1 − Xt
+ − (Xt ) + √ · S (Xt ) + ∆YT E (8) ω80
2 Xt
Y 0 (Xt ) = 0. (9) ωD
0.8 ∆β1
As an example, Fig. 4 shows an evaluation of the above
formulas for t/c = 0.1, ∆YT E = 0 and various values of β1,D β1
the shape parameter KR.
Altogether, the geometry model has finally only three Figure 5. Parameters of the objective function, adapted
parameters: two for the camber line and one for the from [3]
thickness distribution for each blade design.

and is defined as follows:

1.2 Blade-to-Blade Flow Solver
The blade-to-blade flow solver MISES 2.63 [6] has been OBF =
selected as the flow code for the optimization. This 
ωD
 
∆β1,ref

code describes the inviscid flow using the steady Euler C1 + C2 +
ωD,ref ∆β1
equations [7], while the viscous effects are modeled by
(10)
   
the integral boundary layer equations [8]. The coupled ∆β1,stall,ref ω80
C3 + C4 +
system of nonlinear equations is solved by a Newton ∆β1,stall ω80,ref
 
technique. The boundary conditions are defined by those σ80
C5
of the cascade to be designed and optimized. The design σ80,ref
inlet Mach number is kept constant while the inlet angle
is varied between positive and negative stall. Boundary Each component of the function has been normalized
layer transition is predicted using the criterion of Abu- using a corresponding reference value (index ref ) of a
Ghannam / Shaw [9] in a slightly modified version [10] to suitable reference profile. An illustration of all compo-
achieve a better modeling of the flow physics at transition. nents is shown in Fig. 5, where ωD is the design loss
The solver has been sufficiently validated for cascade coefficient at the design inlet angle β1,D . A loss coeffi-
flows at low and high Reynolds numbers, [11, 12] and cient of twice the value of ωD defines the operating range
[2, 13] respectively, and at various turbulence levels with ∆β1 on the loss characteristic. The difference between
and without turbulence grids [1]. AVDR effects are taken the (positive) stall point and the design point is generally
into account by a hyperbolic tangent variation of the known as stall margin ∆β1,stall or safety against stall.
streamtube thickness [14, 15]. For the inner 80 % of the operating range the average
value ω80 and the standard deviation σ80 are introduced
to achieve a loss characteristics as flat as possible. The
1.3 Optimization Method coefficients C are so-called weighting factors by which the
The optimization method used in this paper is the single- relative importance of a component can be changed. Any
objective genetic algorithm (SOGA) from the DAKOTA change has to be specified by the user and validated by
library [16]. The SOGA starts with a random generation test runs. The results of this paper have been produced
of profiles within the limits of the geometric parameters with the following weighting factors: C1 = 1, C2,4,5 = 0.5
(50 samples per generation). The generated individuals and C3 = 2.
will be evaluated by means of an objective function and
passed or not passed to the next generation (up to 19).
The accepted individuals will be promoted, the others
2. RESULTS
will be killed. The latter fail to produce the required exit 2.1 Test Case 1
angle β2 in specific limits. In the present work, the development of a new blade
The objective function of the present investigation section starts with a conventional NACA-65 profile which
consists of 5 components and takes into account the is subsequently modified to an optimized CSM profile.
whole loss characteristic. This function has previously This development is best documented in table form with
been used by Köller et al. [2], by Sieverding et al. [3] four columns for the different stages of the evolution,
 Design and Optimization of Compressor Airfoils by Using Class Function / Shape Function Methodology — 4/8

Table 1. Test Case 1

Design Specifications NACA65 CSM ref. CSM opt.

M a1 = 0.44 Xf = 0.5 −→ 0.45 −→ 0.554
Re1 = 2.5 × 106 f /c = 0.055 (C) −→ 0.055 (P) −→ 0.0451 (P)
β1 = 47◦ Xt = 0.4 −→ 0.2281 −→ 0.2290
β2 = 29◦ KR = 2.0 KR = 1.97
t/c = 0.093 λ = 34.7◦ −→ 34.2◦ −→ 34.2◦
s/c = 0.87
∆YT E = 0.005
AV DR = 0.99
T u1 = 3%
C: Circular Arc P: Parabolic Arc

Y [-]
NACA65 CSM ref. CSM opt.
0.2

0.1

-0.1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 X [-]
Figure 6. Test Case 1 profiles in comparison

Cp ∆β[◦ ] ω[−]
NACA65
NACA65
CSM ref. 35 CSM ref. 0.14
CSM opt. CSM opt.
-0.5
30 0.12

25 0.10
0 x
0.5 1.0 l
20 0.08

0.5 15 0.06

10 0.04
1.0
5 0.02

Figure 7. Test Case 1 pressure distributions at design 0 0.00

20 25 30 35 40 45 50 55 60
inlet angle β1 = 47◦
β1 [◦ ]

Figure 8. Test Case 1 overall performance

cf. Table 1 for Test Case 1. The first column lists the characteristics
specifications for the new blade section to be designed.
The following three columns present the main geometry
parameters (Xf , f /c and Xf (KR)) for the conventional in column 4. The estimated parameter values for the
NACA-65 profile in column 2, for the reference CSM CSM profile are based on the corresponding values of
profile in column 3 and for the optimized CSM profile the NACA-65 profile.
 Design and Optimization of Compressor Airfoils by Using Class Function / Shape Function Methodology — 5/8

Y [-]
opt. airfoil, [2] CSM opt.
0.2

0.1

-0.1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 X [-]
Figure 9. Optimized Test Case 1 profile in comparison with Test Case 4 profile of [2]

The new design method in this paper has been tested ∆β[◦ ] ω[−]
(Test Case 1) on the design of a high Reynolds number
CSM opt.
subsonic compressor blade section for industrial gas tur- 35 opt. airfoil - Exp., [2] 0.14
bines and compressors. The design specifications belong opt. airfoil - CFD, [2]
to the Test Case 4 of reference [2], and are shown here 30 0.12
in Table 1, first column. The Mach number, Reynolds
number and turbulence level are M a1 = 0.44, Re1 = 25 0.10
2.5 × 106 and T u1 = 3% respectively. For further details
see Table 1, columns 2, 3 and 4. 20 0.08
Figure 6 presents all three profiles - the conventional,
the reference and the optimized profile. The correspond- 15 0.06
ing blade pressure distributions are shown in Fig. 7 at
design air inlet angle (β1 = 47◦ ). An inspection of the 10 0.04
optimized pressure distribution led to the conclusion that
5 0.02
optimization at high Reynolds numbers and turbulence
levels of the present investigation inevitable ends up with
0 0.00
a front loaded profile. 20 25 30 35 40 45 50 55 60
As indicated in Fig. 7, boundary layer transition β1 [◦ ]
(bypass transition) starts right after the leading edge
on the suction side and somewhat downstream on the Figure 10. Optimized Test Case 1 performance in
pressure side for both the conventional NACA-65 and comparison with Test Case 4 profile of [2]
the optimized CSM profile. Deceleration of the turbu-
lent suction side boundary layer begins shortly after the
leading edge for the optimized profile and at about 25% particular close to the leading edge, where camber and
chord length for the conventional profile with a gradient curvature of the optimized CSM profile are relatively
becoming increasingly smaller or steeper respectively. low. Numerical performance curves for both profiles are
The controlled diffusion of the optimized CSM pro- shown in Fig. 10, where they demonstrate a remark-
file leads to a 0.8% reduction in design point losses. At able agreement in spite of considerably different profile
off-design, the controlled diffusion concept remains essen- shapes. In addition, Fig. 10 presents a comparison be-
tially valid, specially at the higher inlet angles. There, tween numerical and experimental Test Case 4 results.
the conventional profile is the first to reach the opera- The agreement, however, is not as good as before for
tional limit, while the optimized CSM profile generates a the numerical results. These differences were already ob-
significantly increased operating range (+45%) and stall served and described in [2], but a clear-cut explanation
margin (+18%), cf. Fig. 8. The increase in operating is still missing.
range and stall margin is only partly due to decreasing
deceleration gradients. The main cause turns out to be 2.2 Test Case 2
the reduced leading edge sharpness, which makes the The new design method of this paper has been devel-
profile less sensitive to off-design air inlet angles. oped for high Reynolds number applications. Neverthe-
Figure 9 shows the optimized CSM profile of this less, it has been assumed that the new method might
paper together with the corresponding profile 4 of refer- also be useful at lower Reynolds numbers (aeroengine
ence [2]. A comparison revealed noticeable differences, in Reynolds numbers or lower). This has been checked in
 Design and Optimization of Compressor Airfoils by Using Class Function / Shape Function Methodology — 6/8

Table 2. Test Case 2

Design Specifications NACA65 CSM ref. CSM opt.

M a1 = 0.12 Xf = 0.5 −→ 0.45 −→ 0.43
Re1 = 3.5 × 105 f /c = 0.1 (C) −→ 0.1 (P) −→ 0.1065 (P)
β1 = 30◦ Xt = 0.4 −→ 0.2281 −→ 0.2128
β2 = 0◦ KR = 2.0 KR = 2.84
t/c = 0.1 λ = 13◦ −→ 11.5◦ −→ 11.5◦
s/c = 1.0
∆YT E = 0.005
AV DR = 1.0
T u1 = 1%
C: Circular Arc P: Parabolic Arc

Y [-]
NACA65 CSM ref. CSM opt.
0.3

0.2

0.1

-0.1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 X [-]
Figure 11. Test Case 2 profiles in comparison

a second test case, cf. Table 2 for Test Case 2. The compared to those for the conventional NACA-65 profile
design specifications appear in the first column, specially with relatively long portions of laminar boundary layers,
the Mach number M a1 = 0.12, the Reynolds number cf. Fig. 13. However, thanks to the smaller deceleration
Re1 = 3.5 × 105 and the turbulence level T u1 = 1%. gradient and the reduced leading edge sharpness, the
The following three columns 2, 3 and 4 present, as before predicted operating range and stall margin increased by
in Test Case 1, the main geometric parameters (Xf , f /c more than 55 and 65% respectively.
and Xf (KR)) for the conventional NACA-65 profile, the
reference CSM profile and the optimized CSM profile.
All three profiles are shown in Fig. 11, the corresponding
3. CONCLUSION
design point (β1 = 30◦ ) pressure distribution in Fig. 12
and the complete loss and turning characteristics in Fig. In summary, a new method has been presented for the
13, together with experimental results for the NACA-65 design of high Reynolds number, subsonic compressor
profiles. blade sections for industrial gas turbines and compres-
The most noticeable difference between the conven- sors. By carefully selecting the number of design and
tional NACA-65 and the optimized CSM profile is again optimization parameters, the combination with a genetic
the upstream moved location of the maximum thickness algorithm led to a competitive design method.
for the optimized CSM profile, cf. Fig. 11. Transition of At high Reynolds numbers, the new profiles show de-
the laminar boundary layer under Test Case 2 condition creased design point losses and increased operating limits
(low Reynolds number, low turbulence level) happens compared to corresponding results using conventional
to occur via laminar separation bubbles downstream NACA-65 profiles. In addition, the presented results
the pressure minimum on both sides of the profiles, cf. show close agreement with those produced by so-called
Fig. 12. In spite of a smaller deceleration gradient and high performance profiles of the relevant literature. At
a reduced leading edge sharpness, it is the optimized low Reynolds numbers, the new profiles show slightly
CSM profile that shows 13% higher design point losses increased design point losses but again considerably in-
 Design and Optimization of Compressor Airfoils by Using Class Function / Shape Function Methodology — 7/8

Cp NOMENCLATURE
NASA65 Symbols
CSM ref. a, b parabolic mean arc line parameter
-1.0 CSM opt.
c chord
C Class Function
C1 - C5 weighting factors
-0.5 f maximum camber
KR Shape Function factor
Ma mach number
0 x
RLE leading edge radius
0.5 1.0 l Re Reynolds number
s spacing
S Shape Function
0.5
t thickness
Tu turbulence level
xf position of maximum camber
1.0 xt position of maximum thickness
Figure 12. Test Case 2 pressure distributions at design X dimensionless x value
inlet angle β1 = 30◦ Xf dimensionless position of maximum camber
Xt dimensionless position of maximum thickness
Y dimensionless y value
∆β[◦ ] ω[−] Yc dimensionless y value for camber line
NACA65 Exp. Yt dimensionless y value for thickness distribution
40 0.16
NACA65 CFD β flow angle
CSM ref.
35 CSM opt. 0.14 ∆β1 operating range
∆yT E trailing edge thickness
30 0.12 ∆YT E dimensionless trailing edge thickness
γ trailing edge angle
25 0.10
λ stagger angle
20 0.08 ω total pressure loss
σ standard deviation
15 0.06
Subscripts
10 0.04 ref reference
5 0.02 stall stall margin
D design
0 0.00 1 inlet
5 10 15 20 25 30 35 40 45 50 2 outlet
β1 [ ◦ ] 80 80 % of operating range

Figure 13. Test Case 2 overall performance Abbreviations

characteristics AVDR axial velocity density ratio
CPU central processing unit
CSM Class Function / Shape Function methodology
creased operation limits compared to conventional airfoil NACA National Advisory Committee for Aeronautics
results. OBF objective function
However, for the chosen design approach further re- SOGA single-objective genetic algorithm
search is required. Applying it to airfoils for aeroengine
application will be part of future research. Furthermore,
cascade experiments would validate the new method and
numerical results.
 Design and Optimization of Compressor Airfoils by Using Class Function / Shape Function Methodology — 8/8
[14]
REFERENCES U. Stark and H. Hoheisel. The combined effect of
[1] axial velocity density ratio and aspect ratio on com-
Heinz-Adolf Schreiber, Wolfgang Steinert, and Bern- pressor cascade performance. Journal of Engineering
hard Kuesters. Effects of reynolds number and free- for Power, 103(1):247–255, January 1981.
stream turbulence on boundary layer transition in
[15]
a compressor cascade. Journal of Turbomachinery, Udo Stark. Ebene Verdichtergitter in quasi zweidi-
124(1):1–9, February 2000. mensionaler Unterschallströmung. VDI Forschung-
[2] sheft 641/87, pages pp. 247–255, 1987.
Ulf Koeller. Entwicklung einer fortschrittlichen Pro-
[16]
filsystematik fuer stationaere Gasturbinenverdichter. B.M. Adams, L.E. Bauman, W.J. Bohnhoff, K.R.
PhD thesis, Ruhr-Universität Bochum, DLR- Dalbey, M.S. Ebeida, J.P. Eddy, M.S. Eldred, P.D.
Forschungsbericht 1999-20, 1998. Hough, K.T. Hu, J.D. Jakeman, J.A. Stephens, L.P.
[3] Swiler, D.M. Vigil, and T.M. Wildey. Dakota, A
Frank Sieverding, Beat Ribi, Michael Casey, and
Multilevel Parallel Object-Oriented Framework for
Michael Meyer. Design of industrial axial compressor
Design Optimization, Parameter Estimation, Un-
blade sections for optimal range and performance.
certainty Quantification, and Sensitivity Analysis:
Journal of Turbomachinery, 122(2):323–331, April
Version 6.4 User’s Manual, Sandia Technical Report
2004.
SAND2014-4633 edition, Updated May 9 2014.
[4]
Brenda M. Kulfan and John E. Bussoletti. Fun-
damental parametric geometry representations for
aircraft component shapes. AIAA Paper 2006-6948,
pages 1–45, 2006.
[5]
Hermann Schlichting and Erich Truckenbrodt. Aero-
dynamics of the Airplane (Translated by H. J.
Ramm). McGraw-Hill International Book Company,
Berlin, Germany, 2. edition, 1979.
[6]
Mark Drela and Harold Youngren. A User’s Guide to
MISES 2.63. MIT Aerospace Computational Design
Laboratory, February 2008.
[7]
M. B. Giles. Newton solution of steady two-
dimensional transonic flow. GTL Report No. 186,
1985.
[8]
M. Drela. Two-dimensional transonic aerodynamic
design and analysis using the euler equations. GTL
Report No. 187, 1986.
[9]
B. J. Abu-Ghannam and R. Shaw. Natural transition
of boundary layers–the effects of turbulence, pressure
gradient, and flow history. Journal of Mechanical
Engineering Science, 22(5):213–228, October 1980.
[10]
Mark Drela. MISES Implementation of Modified
Abu-Ghannam / Shaw Transition Criterion (Second
Revision). MIT Aero-Astro, 1998.
[11]
Heinz-Adolf Schreiber, Wolfgang Steinert, Toyotaka
Sonoda, and Toshiyuki Arima. Advanced high-
turning compressor airfoils for low reynolds number
condition - part II: Experimental and numerical anal-
ysis. Journal of Turbomachinery, 126(4):482–492,
December 2004.
[12]
Toyotaka Sonoda and Heinz-Adolf Schreiber. Aero-
dynamic characteristics of supercritical outlet guide
vanes at low reynolds number conditions. Journal
of Turbomachinery, 129(4):694–704, August 2007.
[13]
Ernesto Benini and Andrea Toffolo. Development
of high-performance airfoils for axial flow compres-
sors using evolutionary computation. Journal of
Propulsion and Power, 18(3):544–554, May 2002.