See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/252323242

Aerodynamic Analysis and Optimization of a

Transonic Axial Compressor with Casing
Grooves to Improve Operating Stability

ARTICLE in AEROSPACE SCIENCE AND TECHNOLOGY · JULY 2013

Impact Factor: 0.94 · DOI: 10.1016/j.ast.2013.01.010

CITATIONS READS

2 115

3 AUTHORS, INCLUDING:

Jin-Hyuk KIM Kwang-Yong Kim

Korea Institute of Industrial Technology Inha University
94 PUBLICATIONS 192 CITATIONS 470 PUBLICATIONS 2,414 CITATIONS

SEE PROFILE SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate, Available from: Jin-Hyuk KIM
letting you access and read them immediately. Retrieved on: 27 March 2016
 Aerospace Science and Technology 29 (2013) 81–91

Contents lists available at SciVerse ScienceDirect

Aerospace Science and Technology

www.elsevier.com/locate/aescte

Aerodynamic analysis and optimization of a transonic axial

compressor with casing grooves to improve operating stability
Jin-Hyuk Kim a , Kwang-Jin Choi a , Kwang-Yong Kim b,∗
a
Department of Mechanical Engineering, Graduate School, Inha University, 253 Yonghyun-Dong, Nam-Gu, Incheon 402-751, Republic of Korea
b
Department of Mechanical Engineering, Inha University, 253 Yonghyun-Dong, Nam-Gu, Incheon 402-751, Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history: A transonic axial compressor with circumferential casing grooves is optimized to improve operating
Received 9 May 2012 stability. Numerical analysis is conducted by solving three-dimensional Reynolds-averaged Navier–Stokes
Received in revised form 2 January 2013 equations using the shear stress transport turbulence model. An optimization process based on a
Accepted 29 January 2013
weighted-average surrogate model and steady flow analysis are performed with three design variables
Available online 8 February 2013
defining the tip clearance, blade tip angle, and depth of the grooves. The steady stall inception point
Keywords: is identified from the last converged point, and values of the stall margin as the objective function are
Axial compressor predicted using steady flow analysis at the design points sampled by Latin hypercube sampling in the
Circumferential casing grooves design space. The surrogate model is constructed based on these objective function values. Optimization
Stall margin of this model found the optimum design, which yields a considerable increase in the stall margin
Optimization compared to the smooth casing. To investigate unsteady behavior of the flow in the optimized compressor
Unsteady RANS analysis with casing grooves, an unsteady flow analysis is performed, and the stall inception point is re-predicted
using this analysis.
© 2013 Elsevier Masson SAS. All rights reserved.

1. Introduction experimentally measured the individual stage characteristics and

As a low aspect ratio turbomachine operating at high mass flow
rate, an axial compressor is an essential component of gas turbines
used in jet engines, marine engines, etc. The compressor may ex-
perience severe vibration due to surge and stall when it operates
below the designed mass flow rate. This phenomenon simultane-
ously causes problems of instability and reduced efficiency. Thus,
improving the stall margin is an important consideration in the
design of an axial compressor.
Tip leakage vortex is well known as one of the primary factors
in induced surge and stall in an axial compressor. The trajectory
of the tip leakage vortex, which can be affected by the geometry
near the tip region, has an important influence on the stability of
an axial compressor. To control this flow phenomenon, circumfer-
ential casing grooves (CCGs) have been applied as a casing treat-
ment method. This approach has been used in the design of axial
compressors for several decades [2]. Using experimental and nu-
merical methods, many researchers have examined techniques to
alleviate surge and stall, and to improve the operating stability
of axial compressors, using CCGs. Bailey [2] evaluated the effects
of depth, location, and the number of CCGs in a single-stage ax-
ial compressor through an experimental test. Wenzel et al. [34]
* Corresponding author. Tel.: +82 32 872 3096; fax: +82 32 868 1716.
E-mail address: kykim@inha.ac.kr (K.-Y. Kim).
overall performance of a multi-stage axial compressor with CCGs.
Shabbir and Adamczyk [32] numerically demonstrated the flow
mechanism for stall margin improvement of an axial compressor
with CCGs. Numerical and experimental investigations of a high-
speed small-scale compressor with CCGs were reported by Wu et
al. [36]. Houghton and Day [13] performed experimental and com-
putational studies to investigate the effect of the axial location of
a single casing groove on the stability and efficiency of a subsonic
axial compressor. Beheshti et al. [4] conducted a parametric study
to determine the performance and stability of a transonic axial
compressor with CCGs with respect to variations in tip clearance.
An evaluation of the stall margin and efficiency of a transonic ax-
ial compressor with various CCGs was presented by Kim et al. [20].
Huang et al. [14] studied the effects of the configuration, width,
and depth of CCGs on the stall margin and peak efficiency of a
compressor using 3-D numerical analysis.
To date, most CCG designs for the enhancement of axial com-
pressor stability have been realized through parametric studies.
However, with recent advances in computational fluid dynamics
(CFD) and computing power, several researchers [7,18,5] have car-
ried out optimizations of CCGs using systematic optimization tech-
niques. For example, optimization techniques combined with 3-D
Reynolds-averaged Navier–Stokes (RANS) analysis were applied to
the design of CCGs by Kim et al. [7,18]. Choi et al. [7] used de-
sign optimization to improve the stall margin of a transonic axial
compressor with two geometric design variables: the depth and

1270-9638/$ – see front matter © 2013 Elsevier Masson SAS. All rights reserved.
http://dx.doi.org/10.1016/j.ast.2013.01.010
82 J.-H. Kim et al. / Aerospace Science and Technology 29 (2013) 81–91

Nomenclature

CCGs Circumferential casing grooves RSA Response surface approximation model

CFD Computational fluid dynamics SM Stall margin
D Depth of the circumferential casing grooves SST Shear stress transport
DOE Design-of-experiment T Tip clearance
EFFI Adiabatic efficiency TE Trailing edge
EXP Experiment Tt Total temperature
F Response URANS Unsteady Reynolds-averaged Navier–Stokes
FFT Fast Fourier transform w Weight
GGI General grid interface WTA Weighted-average surrogate model
GSME Generalized mean square cross-validation error X Design variable
KRG Kriging meta-model x Design point
LE Leading edge α, κ Constants of the PBA model
LES Large eddy simulation β Blade tip angle
LHS Latin hypercube sampling γ Specific heat ratio
m Mass flow rate η Adiabatic efficiency
N sm Number of basic surrogate models τ Period
PBA PRESS-based averaging surrogate model
Subscripts
PR Total pressure ratio
PRESS Predicted error sum of squares in Inlet
Pt Total pressure max Choking mass flow
P 1−5 Control points represented by Bezier curve out Outlet
RANS Reynolds-averaged Navier–Stokes peak Peak adiabatic efficiency point
RBNN Radial basis neural network model stall Near-stall point

width of the CCGs. Kim et al. [18] optimized the design of cir-
cumferential casing grooves to maximize the stall margin and peak
adiabatic efficiency of a transonic axial compressor by using a hy-
brid multi-objective evolutionary algorithm. Carnie et al. [5] used
optimization to enhance the stall margin of an axial compressor
with CCGs by applying a new meshing methodology, the zipper
layer meshing approach.
With the aforementioned recent trends, advanced numerical
methods such as unsteady RANS (URANS) analysis and large eddy
simulation (LES) have been used in the design of axial compres-
sors with CCGs to accurately identify the unsteady stall mecha-
nism. Experimental and numerical investigations of the unsteady
interaction of the rotor and circumferential grooves in a single-
stage transonic compressor were reported by Muller et al. [26].
Zhao et al. [38] investigated the effects of CCGs on the unsteadi-
ness of tip clearance flow to enhance compressor flow instability.
Legras et al. [21] demonstrated the mechanism of unsteady inter-
nal flows in a high-pressure multi-stage axial compressor equipped
with CCGs using RANS and URANS analyses. Various experimental
and numerical studies to understand the steady and unsteady flow
phenomena in transonic compressors with CCGs were reviewed by
Hah [11].
Among optimization strategies, surrogate modeling has recently
become a promising tool for high-performance turbomachinery
design [29,33,19,10,30]. Various surrogate modeling techniques,
with their intrinsic advantages, have been developed by many re-
searchers. In particular, Goel et al. [9] developed weighted-average
surrogate models consisting of a response surface approximation
model (RSA) [27], a Kriging meta-model (KRG) [22], and a ra-
dial basis neural network (RBNN) model [28]. They concluded that
weighted-averaging of surrogate models provides a more reliable
prediction method than individual surrogate models. Also, the re-
liability of one of these models was demonstrated based on the
design optimization of turbomachinery by Samad et al. [31] and
Kim and Kim [16].
This work presents an optimization procedure for an axial com-
pressor design with CCGs based on a surrogate modeling technique
coupled with 3-D steady RANS analysis. The optimization was per-
Table 1
Design specifications of the axial compressor with NASA Rotor 37.
Designed mass flow rate, kg/s 20.19
Rotational speed, rpm 17,188.7
Total pressure ratio 2.106
Inlet hub-tip ratio 0.7
Blade aspect ratio 1.19
Tip relative inlet Mach number 1.48
Hub relative inlet Mach number 1.13
Tip solidity 1.29
Number of rotor blades 36
formed to improve the stall margin of a grooved axial compressor
with three design variables defining the tip clearance, blade tip
angle, and depth of the CCGs. Moreover, URANS analysis was car-
ried out to investigate unsteady flow phenomena in the optimized
compressor with CCGs.
2. Numerical analysis
The compressor model investigated in this work is a transonic
axial compressor with NASA Rotor 37 [8]. The compressor rotor
operates at a speed of 17188.7 rpm. The total pressure ratio and
adiabatic efficiency are 2.106 and 88.9%, respectively, at the de-
signed mass flow rate of 20.19 kg/s. The blade section of NASA
Rotor 37 is defined by a multiple circular arc. The tip clearance is
0.356 mm (0.47% span), the choking mass flow is 20.93 kg/s, and
the near-stall point is 0.925 of the choke flow. Detailed specifica-
tions of the compressor model with NASA Rotor 37 are given in
Table 1 from the AGARD report by Dunham [8].
Flow through the axial compressor was analyzed by solving
3-D steady and unsteady RANS equations through a finite volume
solver, the commercial code ANSYS-CFX 11.0 [1]. Blade profile cre-
ation and computational mesh generation were performed using
ANSYS Blade-Gen and Turbo-Grid, respectively. CFX-Pre, CFX-solver,
and CFX-Post were used to define boundary conditions, solve
the governing equations, and postprocessing the results, respec-
tively. In addition, the creation of circumferential grooves and their
 J.-H. Kim et al. / Aerospace Science and Technology 29 (2013) 81–91 83

Fig. 3. Validation of the flow analysis [7].

A k–ω -based shear stress transport (SST) turbulence model [25]

Fig. 1. Structure of the grid system.
was used to account for transport of the turbulent shear stress
and to produce accurate results for flow separation under an ad-
verse pressure gradient [3]. This model uses a k–ω model for the
near-wall region and a k–ε model in the bulk domain. A blending
function is applied to ensure a smooth transition between these
two regions. The accuracy of the numerical scheme for turbulent
flow depends strongly on the treatment of the wall shear stress. In
this work, to benefit from the SST model, resolution of the bound-
ary layer with more than ten mesh points was applied. In addition,
the first grids near the wall were adjusted to keep y +  2 to ac-
curately capture wall shear stress, and also to implement a low
Reynolds number SST model [25].
The computations for the steady and unsteady RANS analyses
were performed using an Intel® Core I7 CPU 2.67 GHz PC. The
computational run times were approximately 6 to 7 h and 21 to
22 h, respectively.

Fig. 2. Effects of computational grids on the performance curves [20]. 3. Validation of numerical results

mesh generation were performed using ANSYS Design-Modeler and The results of the steady RANS analysis were validated by com-
parison with experimental data [8] in a previous study [7]. The
ICEM-CFD, respectively.
compressor model used in this test did not employ the casing
The computational domain, which consists of a single passage
grooves. Fig. 3 shows the validation results for the performance
between blades, is shown in Fig. 1. The working fluid was consid-
curves of the total pressure ratio PR and adiabatic efficiency η ,
ered to be an ideal gas. The total pressure and total temperature at
which are defined as follows:
the inlet were set to 101,325 Pa and 288.15 K, respectively. The
P t ,out
mass flow rate was set at the outlet. The solid surfaces in the PR = (1)
computational domain were considered to be hydraulically smooth P t ,in
with adiabatic and no-slip conditions. The periodic boundary was γ −1
P t ,out
( P t ,in
) γ −1
set at the blade passage interface, and the tip clearance was con-
η= T
(2)
sidered along with the passage. The general grid interface (GGI) ( Ttt,,out
in
)−1
method was used for the connection between the passage and the
circumferential grooves where the grids on either side of the two where γ , P t , and T t indicate the specific heat ratio, total pres-
connected surfaces do not match [1]. sure, and total temperature, respectively, and the subscripts in and
As shown in Fig. 1, a structured grid system was created in the out refer to the inlet and outlet, respectively. These results show
computational domain with O-type grids near the blade surface some uniform underestimations of the total pressure ratio and adi-
abatic efficiency through the entire mass flow range, but are in
and H/J/C/L-type grids in the other regions. In previous work [20],
good qualitative agreement with the experimental data.
a grid-dependency test was performed with various numbers of
The numerical stall inception point was identified from the last
grids; Fig. 2 shows the performance curve results for the pressure
converged point by reducing the normalized mass flow by 0.002.
ratio and adiabatic efficiency. As shown in this figure, the values of
The peak adiabatic efficiency point was also found by reducing the
the pressure ratio and adiabatic efficiency vary only slightly with normalized mass flow by 0.002 in the high mass flow region. This
the variation of grid size from approximately 480,000 to 740,000. work used the following convergence criteria, which was proposed
The predicted near-stall normalized mass flows for grid numbers by Chen et al. [6]:
of 270,000, 480,000, and 740,000, are 0.939, 0.921, and 0.920, re-
spectively. Thus, this work used a total of 480,000 grids to define 1. The inlet mass flow rate variation is less than 0.001 kg/s for
the main flow passage of the axial compressor. Additionally, each 300 steps.
groove was constructed with approximately 24,000 grids. Fig. 1 2. The difference between the inlet and outlet mass flow rate is
shows a typical example of the grid system used in this work. less than 0.5%.
84 J.-H. Kim et al. / Aerospace Science and Technology 29 (2013) 81–91

3. The adiabatic efficiency variation is less than 0.3% per 100

steps.

In Fig. 3, the near-stall point predicted based on these con-

vergence criteria is 0.921. The near-stall point measured experi-
mentally is 0.925, which represents the mass flow rate normalized
by the choking mass flow rate. Therefore, it is assured that the
steady RANS results obtained based on these convergence criteria
are valid and reliable.

4. Objective function and design variables

The purpose of present optimization was to maximize the stall

margin of a transonic axial compressor with CCGs. The objective
function is the stall margin SM, which is defined as follows:
 
mpeak PRstall
SM = × − 1 × 100% (3)
mstall PRpeak

where m represents the mass flow rate and the subscripts peak
and stall mean the peak adiabatic efficiency point and the near-
stall point, respectively.
The present work selected three design variables, including the
variable related to blade shape. These are the tip clearance T and
blade tip angle β , and the depth D of the CCGs (D affected the
objective function more sensitively than the width in our previous
work [7]).
Fig. 4(a) shows the meridional plane of the compressor blade
with the CCGs to define the design variables T and D. The five
CCGs are evenly installed from the leading edge (LE) to the trailing
edge (TE). The width of each groove is 16% of the tip axial chord,
and the gap is 5% of the tip axial chord of the blade.
Blade tip angle β is defined as the angle between the axis of
rotation and a tangent of the camber line at the tip, as shown
in Fig. 4(b). In this work, the entire β distribution was changed
along the blade tip with the fixed meridional geometry by control
points represented by a fourth-order Bezier curve [17] as shown in
Fig. 4(b). When the β distribution at the blade tip was changed,
the blade profiles at other locations were interpolated using a
B-spline curve from hub to tip, with the fixed blade profile at the
mid-span. The advantage of using a Bezier curve for shape param-
eterization is that only control points located along the curves can
control the curves. When one control point in the Bezier curve
is moved vertically, the others are kept fixed. Thus, each control
point is controlled independently, and all these points can be con-
sidered as design variables. In this work, to limit the number of
design variables, all control points were fixed except the control
point P 3 , which gives the most sensitive results for the curve
variation among the control points. Consequentially, through the
minute variation of the control point P 3 located approximately
Fig. 4. Definition of design variables. (a) Definitions of the tip clearance and the
60% chord from the LE of the blade in the chordwise direction, the groove depth. (b) Definition of the blade tip angle. (c) Example of a changed blade
entire β distribution at the blade tip can be changed, as shown shape with variation of β at blade tip.
in Fig. 4(c). Fig. 4(c) shows a typical example of a changed blade
shape with the β variation. Table 2
Ranges of design variables.
5. Optimization techniques Variables D (mm) β (deg) T (mm)
Lower bound 0.214 60.04 0.178
For design optimization, it is important to find the feasible de- Upper bound 0.641 69.62 0.534
sign space formed by ranges of design variables. The ranges of
design variables were specified based on a preliminary parametric
study, as shown in Table 2. Twenty-five design points within the design points 3, 5, 10, 15, and 20; this is probably due to the cou-
design space were selected using Latin hypercube sampling (LHS) pling effects between the independent variables. Viewed in this
as the design-of-experiment (DOE) [24]. light, the high objective function values were generally obtained
Table 3 shows the results of design variable and objective func- by bigger blade tip angles and narrower tip clearances compared
tion values at 25 design points sampled by LHS. As tabulated in to the reference design. Thus, the objective function value can be
Table 3, relatively high objective function values were found at maximized through design optimization based on these results.
 J.-H. Kim et al. / Aerospace Science and Technology 29 (2013) 81–91 85

Table 3 weight, and thus the lower contribution toward the final PBA sur-
Design variables and objective function values at design points sampled by LHS. rogate model and vice versa. In this work, the weights were cal-
Design Design variables Objective function values culated using the generalized mean square cross-validation error
points D (mm) β (degrees) T (mm) SM (%) (GMSE) or the PRESS. The weighting scheme for the PBA model is
1 0.623 67.54 0.341 13.148
given as follows:
2 0.267 63.37 0.504 11.243
( E i + α E avg )κ
3 0.249 65.04 0.223 15.487 wi = 
i ( E i + α E avg )
4 0.214 61.29 0.312 12.602 κ
5 0.498 69.20 0.267 14.969
6 0.605 61.70 0.237 11.160 
N SM
Ei
7 0.481 60.87 0.356 12.481 E avg = , α < 1, β < 0
8 0.338 60.04 0.430 10.178 N SM
i =1
9 0.463 62.54 0.178 12.486 
10 0.587 66.70 0.193 15.500 Ei = GMSEi , i = 1, 2, . . . , N SM (6)
11 0.641 62.95 0.371 11.396
12 0.445 68.37 0.386 13.299 This weighting scheme requires the user to specify two param-
13 0.303 67.12 0.534 10.913 eters, α and κ , which control the importance of averaging and
14 0.392 65.45 0.445 13.207 of the individual surrogate, respectively. In the present work, two
15 0.409 66.29 0.208 16.251
constants, α and κ , were chosen as α = 0.05 and κ = −1 [9]. Af-
16 0.516 64.62 0.297 12.508
17 0.427 62.12 0.519 12.165 ter constructing the PBA surrogate model, the sequential quadratic
18 0.285 69.62 0.415 12.865 programming (SQP) [23] algorithm was used to search for the
19 0.320 68.79 0.282 14.414 optimum point from the PBA model. The SQP is dependent on
20 0.374 59.62 0.252 14.895 the initial guess of the optimum point; hence, a series of trials
21 0.552 67.95 0.490 13.746
22 0.231 65.87 0.401 11.563
was conducted before obtaining the final optimum point from any
23 0.570 60.45 0.475 10.872 surrogate. The optimization procedure employed in this work is
24 0.534 64.20 0.460 11.924 shown in Fig. 5.
25 0.356 63.79 0.326 13.279
6. Results and discussion
In the present work, a weighted-average surrogate model
6.1. Shape optimization
(WTA3 model) developed by Goel et al. [9] was adopted. This
model was later called the predicted error sum of squares (PRESS)-
In this work, shape optimization based on the PBA surrogate
based averaging (PBA) model by Samad et al. [31]. The predicted
model combined with 3-D steady RANS analysis has been per-
response surface of the PBA model was defined based on the PRESS
formed to improve the stall margin of an axial compressor with
as follows:
CCGs. For optimization, the PBA model with the basic surrogate

N SM models was constituted as follows:
F̂ PBA (x) = w i (x) F̂ i (x) (4)
i F PBA = 0.566F RSA + 0.414F KRG + 0.021F RBNN (7)
where N SM is the number of basic surrogate models used to con- The PRESS values were 0.669, 1.132, and 33.617 for the RSA,
struct the PBA model. The ith surrogate model at design point x KRG, and RBNN, respectively. In order to prevent the designer from
produces weight w i (x), and F̂ i (x) is the response predicted by the a poor surrogate, the PBA model assigned the highest weight to
ith surrogate model. To simplify the function, Eq. (4) can be writ- the RSA, which produced the lowest error, and the lowest weight
ten as follows: to the RBNN, which produced the highest error.
The design variables for the optimized compressor predicted by
F PBA = w RSA F RSA + w KRG F KRG + w RBNN F RBNN the PBA model are listed in Table 4(a). In comparison with the
X lj < x j < X uj , j = 1, 2, 3, . . . , 6 (5) smooth casing, the optimum design has a bigger blade tip angle
and a narrower tip clearance. The stall margin of the smooth cas-
where F RSA , F KRG , and F RBNN are the responses predicted by the ing obtained through the RANS calculation was 11.63%, as shown
basic surrogates RSA [27], KRG [22], and RBNN [28], respectively. in Table 4(b). The stall margin for the optimum design was pre-
X l and X u are the lower and upper bounds of each design variable, dicted to be 16.75% based on the optimization, and was calculated
respectively. to be 16.76% by the RANS analysis. The PBA model produced a
The weight for each surrogate model is decided such that good prediction with a relative error of 0.01% compared to the
the basic surrogate that produces the higher error has the lower RANS calculation. Consequently, the stall margin of the optimum

Fig. 5. Optimization procedure.

86 J.-H. Kim et al. / Aerospace Science and Technology 29 (2013) 81–91

Table 4 steady RANS analysis for the smooth casing and optimum design
Results of design optimization. at the near-stall points were almost the same (2.073 and 2.074,
(a) Design variables
respectively). Also, the peak adiabatic efficiencies are 85.12% and
Designs D (mm) β (deg) T (mm) 84.94%, respectively. In comparison with the smooth casing, the
Smooth casing – 64.62 0.356 CCGs reduced significantly the near-stall mass flow, and conse-
Optimum design 0.371 67.17 0.193
quently improved the stall margin. However, the peak adiabatic
efficiency slightly decreased when the CCGs were installed. These
(b) Objective function values
results have been demonstrated by many other researchers [13,14,
Designs Prediction RANS Prediction error Increment
12,15].
(%) (%) (%) (%)
Fig. 7 shows the trajectories of the tip leakage vortices at the
Smooth casing – 11.63 – –
Optimum design 16.75 16.76 0.011 5.13
near-stall point of the smooth casing (m/mmax = 0.921) for the
smooth casing and optimum design. The tip leakage vortex is pro-
duced at the LE near the blade tip by the interaction between
tip leakage flow and incoming flow. In the compressor without
CCGs, the tip leakage vortex is mainly driven by tip leakage flow,
and proceeds along the pressure surface of the blade as shown
in Fig. 7(a). On the other hand, the pressure gradient over the
blade tip is reduced due to the broader tip clearance caused by
the CCGs. Hence, as shown in Fig. 7(b), the trajectory of the tip
leakage vortex is mainly driven by the incoming flow, resulting in
the trajectory of the tip leakage vortex close to the passage cen-
ter. The decreased pressure gradient over the blade tip reduces the
strength of the tip leakage vortex. Thus, the optimum design shows
that the trajectory of the tip leakage vortex becomes closer to the
passage center since it is mainly driven by the incoming flow. This
is due to the decrease in the velocity of the tip leakage vortex
compared to the smooth casing, as shown in Fig. 7(b).
The static pressure contours at the near-stall point for the
smooth casing and optimum design at 98% span are shown in
Fig. 8. A stagnation zone was observed near the LE of the suc-
tion surface in the case of the smooth casing resulting from high
pressure occurring near the LE due to the deceleration of the in-
flow. These results were also demonstrated by Wilke et al. [35].
They reported that the angle of attack of the inflow was very
steep, especially near the blade tip. The inflow cannot completely
follow the direction given by the blade profile, which results in
an extended stagnation zone at the blade suction surface and a
significant deceleration of the inflow at the upstream part of the
following pressure surface. On the other hand, in the optimum de-
sign with CCGs, the reduced stagnation zone was observed near
the LE of the suction surface as shown in Fig. 8(b). It is thought
that the incoming flow can follow the direction of the blade pro-
file as the pressure near the LE of the pressure surface is reduced
by the variation of the angle of attack with the optimized angle
distribution at the blade tip and grooves.
Fig. 9 shows the static pressure distributions on the pressure
and suction surfaces of the blade at 98% span for the smooth cas-
ing and optimum design. With the CCGs, large variation of the
Fig. 6. Performance curves. (a) Total pressure ratio. (b) Adiabatic efficiency.
pressure difference between the pressure and suction surfaces was
observed at each groove location. A significant reduction in the
design represents a 5.13% improvement compared to the smooth
pressure difference occurred near the LE region due to a decrease
casing.
in blade loading near the LE by a variation of the angle of attack,
while the pressure difference at the location between the second
6.2. Steady flow analysis
and third grooves increased.
To find the main factors responsible for the improved stall Fig. 10 shows static entropy contours for the smooth casing and
margin of the optimized compressor with CCGs, the performance optimum design on a meridional plane at each peak efficiency
curves and internal flow fields for the smooth casing and opti- point. The entropy values were averaged in the circumferential
mum design were compared and analyzed as shown in Figs. 6–10. direction. As shown in Fig. 10, the application of the CCGs had
Fig. 6 shows performance curves of the total pressure ratio and the negative effects on the efficiency of the compressor. Higher entropy
adiabatic efficiency for compressors with smooth casing and opti- generation was observed near the casing of the axial compressor
mum design. As shown in Figs. 6(a) and (b), the near-stall points with the CCGs (the optimum design) compared to the smooth cas-
for the smooth casing and optimum design were predicted to be ing. This explains the inevitable decreases in the efficiency of the
0.921 and 0.892, respectively. The total pressure ratios predicted by compressor.
 J.-H. Kim et al. / Aerospace Science and Technology 29 (2013) 81–91 87

Fig. 7. Trajectory of tip leakage vortex at near-stall point of smooth casing, m/mmax = 0.921. (a) Smooth casing. (b) Optimum design.

Fig. 8. Static pressure contours at near-stall point of smooth casing (98% span, m/mmax = 0.921) (unit: kPa). (a) Smooth casing. (b) Optimum design.
88 J.-H. Kim et al. / Aerospace Science and Technology 29 (2013) 81–91
Fig. 9. Static pressure distributions at near-stall point of smooth casing, m/mmax =
0.921 (98% span).
Fig. 10. Static entropy contours on meridional plane at each peak adiabatic efficiency
point (unit: J/kg K). (a) Smooth casing. (b) Optimum design.
6.3. Unsteady flow analysis
URANS analysis was performed to investigate the difference in
the unsteady flow structure between the smooth casing and op-
timum design with the CCGs. On the basis of the stall inception
point predicted by steady RANS analysis, the stall inception point
was re-predicted using URANS analysis. The unsteady stall incep-
tion point was identified by the unsteady static pressure signals at
a selected local point on the blade. Fig. 11(a) shows a typical ex-
ample of the static pressure signal at a local point on the LE near
Fig. 11. Typical examples of unsteady static pressure signal at a local point on LE
near the tip of the optimum design with CCGs near the stall point. (a) Converged
static pressure signal (m/mmax = 0.889). (b) Non-converged static pressure signal
(m/mmax = 0.882).
Fig. 12. Unsteady static pressure signals at local point on LE near blade tip at the
near-stall point of the smooth casing (m/mmax = 0.919).
the tip of the optimum blade with CCGs at the near-stall point of
the optimum design (m/mmax = 0.889). This unsteady static pres-
sure signal is shown to be established, which implies a converged
URANS solution. On the other hand, a typical example of a non-
converged static pressure signal at a mass flow rate slightly less
than that of the near-stall point (m/mmax = 0.882) is shown in
 J.-H. Kim et al. / Aerospace Science and Technology 29 (2013) 81–91 89

Fig. 13. Amplitude values of pressure signal with frequency characteristics detected
by FFT analysis.
Fig. 11(b). Our convergence criterion for the URANS analysis was
also suggested by Zhang et al. [37].
The stall inception points for the smooth casing and optimum
design predicted by unsteady RANS (URANS) analysis with the
aforementioned criterion are shown in Fig. 6. The figure shows
that the normalized mass flow rates at the stall inception points
predicted by URANS analysis for the smooth casing and optimum
design are 0.919 and 0.889, respectively. These values are smaller
than those predicted by steady RANS analysis. For the smooth cas-
ing, the stall inception point predicted by steady RANS analysis is
closer to the experimental measurement of 0.925, as indicated in
Fig. 3, than to the point predicted by URANS analysis. Therefore,
the prediction of the stall inception point using URANS analysis
with the convergence criterion is less accurate than the prediction
using steady RANS analysis. And, the time-averaged total pressure
ratio and adiabatic efficiency values at these points are lower than
those at the stall inception points predicted by steady RANS anal-
ysis.

Fig. 14. Mach number contours at 98% span at the near-stall point of the smooth casing (m/mmax = 0.919). (a) T = 1/6τ . (b) T = 2/6τ . (c) T = 3/6τ . (d) T = 4/6τ .
(e) T = 5/6τ . (f) T = 6/6τ .
90 J.-H. Kim et al. / Aerospace Science and Technology 29 (2013) 81–91
Fig. 14. (continued)
Fig. 12 shows the unsteady static pressure signals at a lo-
cal point on the LE near the blade tip for the smooth casing
and optimum design at the near-stall point of the smooth cas-
ing (m/mmax = 0.919). As shown in Fig. 12, the amplitudes of the
signal for the optimum design are substantially lower than those
for the smooth casing. As previously discussed, the broader tip
clearance caused by the CCGs induced a decrease in the pressure
gradient over the blade tip that resulted in a decrease in the ve-
locity of the tip leakage vortex.
Fig. 13 shows the amplitude values of the pressure signal with
frequency characteristics detected by fast Fourier transform (FFT)
analysis based on unsteady static pressure signals predicted at a
local point on the LE near the blade tip for the smooth casing and
optimum design. The harmonic frequencies were observed over all
the frequency ranges in both the smooth casing and optimum de-
sign. The frequency of the peak amplitude for the optimum design
was found to be shifted to a higher frequency compared to those
for the smooth casing. The amplitude values for the optimum de-
sign were significantly reduced in most of the frequency range,
especially at the harmonic frequencies, compared to the smooth
casing.
Fig. 14 shows the time history of instantaneous Mach number
contours at 98% span for the smooth casing and optimum design
at the near-stall point of the smooth casing. Both of the instan-
taneous Mach number contours are compared in one period of
the pressure signal for the smooth casing. The period was divided
into six time steps to clarify the change in the flow structure. For
the smooth casing, the separation occurred on the suction surface
of the blade as shown in Fig. 14(a), and a large low-speed zone
caused by the interaction between the tip leakage vortex and the
passage shock developed near the LE of the pressure surface of
the blade as shown in Fig. 14(b). Consequently, the surging phe-
nomenon arose in the entire blade passage, as shown in Fig. 14(c).
In the optimum design with the CCGs, the surging phenomenon
was restrained and delayed, as shown in Fig. 14(e), compared to
the smooth casing (Fig. 14(c)). These results illustrate the enhance-
 J.-H. Kim et al. / Aerospace Science and Technology 29 (2013) 81–91
ment and extension of the stall margin for the optimum design
with CCGs.
7. Conclusion
An axial compressor with CCGs was optimized using a PBA sur-
rogate model coupled with 3-D steady RANS analysis. Optimization
was carried out to enhance the stall margin of a grooved axial
compressor with three design variables defining the tip clearance,
blade tip angle, and depth of the CCGs. The results of shape opti-
mization show that the stall margin for the optimum design with
the CCGs was improved by 5.13%, while the peak adiabatic effi-
ciency decreased by only 0.18% with nearly the same total pres-
sure ratio compared to the smooth casing. In addition, unsteady
RANS analysis was performed to clarify the unsteady flow struc-
ture caused by the CCGs. The stall inception points re-predicted
through this unsteady analysis, the time-averaged total pressure
ratio, and the adiabatic efficiency values at these points were gen-
erally less than those predicted by the steady analysis. The steady
RANS analysis resulted in a better prediction of the stall inception
point for the compressor with a smooth casing than the URANS
analysis with the convergence criteria. For the optimum design us-
ing CCGs, the time history of the unsteady flow field indicated that
the surging phenomenon was clearly restrained and delayed com-
pared to the smooth casing.
Acknowledgements
This work was supported by the National Research Foundation
of Korea (NRF) grant funded by the Korean government (MEST)
(No. 2012-0005159). The authors gratefully acknowledge this sup-
port.
References
[1] ANSYS CFX-11.0, ANSYS CFX-Solver Theory Guide, ANSYS Inc., 2006.
[2] E.E. Bailey, Effect of grooved casing treatment on the flow range capability of
a single-stage axial-flow compressor, NASA Technical Memorandum, National
Aeronautics and Space Administration, Washington, DC, 1972, NASA TM X-
2459.
[3] J.E. Bardina, P.G. Huang, T.J. Coakley, Turbulence modeling validation, in: 28th
AIAA Fluid Dynamics Conference, Snowmass Village, USA, 1997, AIAA-1997-
2121.
[4] B.H. Beheshti, J.A. Teixeira, P.C. Ivey, K. Ghorbanian, B. Farhanieh, Parametric
study of tip clearance—casing treatment on performance and stability of a
transonic axial compressor, ASME Journal of Turbomachinery 126 (4) (2004)
527–535.
[5] G. Carnie, Y. Wang, N. Qin, Design optimization of casing grooves using the
Zipper layer meshing method, in: ASME Turbo Expo 2011, Vancouver, Canada,
2011, GT2011-45483.
[6] H. Chen, X.D. Huang, S. Fu, CFD investigation on stall mechanisms and cas-
ing treatment of a transonic compressor, in: 42nd AIAA/ASME/SAE/ASEE Joint
Propulsion Conference & Exhibit, Sacramento, USA, 2006, AIAA-2006-4799.
[7] K.J. Choi, J.H. Kim, K.Y. Kim, Design optimization of circumferential casing
grooves for a transonic axial compressor to enhance stall margin, in: ASME
Turbo Expo 2010, Glasgow, UK, 2010, GT2010-22396.
[8] J. Dunham, CFD validation for propulsion system components, AGARD Advisory
Report 355, 1998, ISBN 92-836-1075-X.
[9] T. Goel, R.T. Haftka, W. Shyy, N.V. Queipo, Ensemble of surrogates, Structural
and Multidisciplinary Optimization 33 (3) (2007) 199–216.
[10] T. Goel, D.J. Dorney, R.T. Haftka, W. Shyy, Improving the hydrodynamic per-
formance of diffuser vanes via shape optimization, Computers & Fluids 37 (6)
(2008) 705–723.
[11] C. Hah, Toward optimum configuration of circumferential groove casing treat-
ment in transonic compressor rotors, in: ASME–JSME–KSME Joint Fluids Engi-
neering Conference 2011, Hamamatsu, Japan, 2011, AJK2011-K05020.
[12] M. Hembera, H.P. Kau, E. Johann, Simulation of casing treatments of a transonic
compressor stage, International Journal of Rotating Machinery 2008 (2008)
657202.
91
[13] T. Houghton, I. Day, Enhancing the stability of subsonic compressors us-
ing casing grooves, ASME Journal of Turbomachinery 133 (2) (2011) 021007
(11 pages).
[14] X. Huang, H. Chen, S. Fu, CFD investigation on the circumferential grooves
casing treatment of transonic compressor, ASME Turbo Expo 2008, Berlin, Ger-
many, 2008, GT2008-51107.
[15] H. Jian, W. Hu, Numerical investigation of inlet distortion on an axial flow com-
pressor rotor with circumferential groove casing treatment, Chinese Journal of
Aeronautics 21 (8) (2008) 496–505.
[16] J.H. Kim, K.Y. Kim, Optimization of Vane diffuser in a mixed-flow pump
for high efficiency design, International Journal of Fluid Machinery and Sys-
tems 4 (1) (2011) 172–178.
[17] J.H. Kim, J.H. Choi, K.Y. Kim, Surrogate modeling for optimization of a cen-
trifugal compressor impeller, International Journal of Fluid Machinery and Sys-
tems 3 (1) (2010) 29–38.
[18] J.H. Kim, K.J. Choi, A. Husain, K.Y. Kim, Multiobjective optimization of cir-
cumferential casing grooves for a transonic axial compressor, AIAA Journal of
Propulsion and Power 27 (3) (2011) 730–733.
[19] J.H. Kim, J.W. Kim, K.Y. Kim, Axial-flow ventilation fan design through multi-
objective optimization to enhance aerodynamic performance, ASME Journal of
Fluids Engineering 133 (10) (2011) 101101 (12 pages).
[20] J.H. Kim, K.J. Choi, K.Y. Kim, Performance evaluation of a transonic axial com-
pressor with circumferential casing grooves, Proceedings of the Institution of
Mechanical Engineers, Part A: Journal of Power and Energy 226 (2) (2012) 218–
230.
[21] G. Legras, N. Gourdain, I. Trebinjac, X. Ottavy, Analysis of unsteadiness on cas-
ing treatment mechanisms in an axial compressor, in: ASME Turbo Expo 2011,
Vancouver, Canada, 2011, GT2011-45806.
[22] J.D. Martin, T.W. Simpson, Use of kriging models to approximate deterministic
computer models, AIAA Journal 43 (4) (2005) 853–863.
[23] MATLAB® , The language of technical computing, release 14, The Math. Works
Inc., 2004.
[24] M.D. McKay, R.J. Beckman, W.J. Conover, A comparison of three methods for
selecting values of input variables in the analysis of output from a computer
code, Technometrics 42 (1) (2000) 55–61.
[25] F.R. Menter, Two-equation eddy-viscosity turbulence models for engineering
application, AIAA Journal 32 (8) (1994) 1598–1605.
[26] M.W. Muller, C. Biela, H.-P. Schiffer, C. Hah, Interaction of rotor and casing
treatment flow in an axial single-stage transonic compressor with circumferen-
tial grooves, in: ASME Turbo Expo 2008, Berlin, Germany, 2008, GT2008-50135.
[27] R.H. Myers, D.C. Montgomery, Response Surface Methodology-Process and
Product Optimization Using Designed Experiments, John Wiley & Sons, Inc.,
New York, 1995.
[28] M.J.L. Orr, Introduction to Radial Basis Neural Networks, Center for Cognitive
Science, Edinburgh University, Scotland, UK, 1996, http://anc.ed.ac.uk/rbf/.
[29] S. Pierret, R.A. Van Den Braembussche, Turbomachinery blade design using a
Navier–Stokes solver and artificial neural network, ASME Journal of Turboma-
chinery 121 (2) (1999) 326–332.
[30] M.M. Rai, N.K. Madyavan, Aerodynamic design using neural networks, AIAA
Journal 38 (1) (2000) 173–182.
[31] A. Samad, K.Y. Kim, T. Goel, R.T. Haftka, W. Shyy, Multiple surrogate model-
ing for axial compressor blade shape optimization, Journal of Propulsion and
Power 24 (2) (2008) 302–310.
[32] A. Shabbir, J.J. Adamczyk, Flow mechanism for stall margin improvement due
to circumferential casing grooves on axial compressors, ASME Journal of Tur-
bomachinery 127 (4) (2005) 708–717.
[33] X.F. Wang, G. Xi, Z.H. Wang, Aerodynamic optimization design of centrifugal
compressor’s impeller with kriging model, in: Proceedings of the Institution
of Mechanical Engineers, Part A, Journal of Power and Energy 220 (6) (2006)
589–597.
[34] L.M. Wenzel, J.E. Moss, C.M. Mehalic, Effect of casing treatment on performance
of a multistage compressor, NASA Technical Memorandum, National Aeronau-
tics and Space Administration, Washington, DC, 1975, NASA TM X-3175.
[35] I. Wilke, H.P. Kau, G.. Brignole, Numerically aided design of a high-efficient cas-
ing treatment for a transonic compressor, in: ASME Turbo Expo 2005, Nevada,
USA, 2005, GT2005-68993.
[36] Y. Wu, W. Chu, H. Zhang, Q. Li, Parametric investigation of circumferential
grooves on compressor rotor performance, ASME Journal of Fluids Engineer-
ing 132 (12) (2011) 121103 (10 pages).
[37] Y. Zhang, X. Lu, W. Chu, J. Zhu, Numerical investigation of the unsteady tip
leakage flow and rotating stall inception in a transonic compressor, Journal of
Thermal Science 19 (4) (2010) 310–317.
[38] S. Zhao, X. Lu, J. Zhu, H. Zhang, Investigation for the effects of circumferential
grooves on the unsteadiness of tip clearance flow to enhance compressor flow
instability, in: ASME Turbo Expo 2010, Glasgow, UK, 2010, GT2010-22652.