Design of high-efciency turbomachinery blades for energy conversion devices

with the three-dimensional prescribed surface curvature distribution blade

design (CIRCLE) method
T. Korakianitis

, I.A. Hamakhan, M.A. Rezaienia, A.P.S. Wheeler, E.J. Avital, J.J.R. Williams
School of Engineering and Materials Science, Queen Mary University of London, London E1 4NS, UK
a r t i c l e i n f o
Article history:
Received 1 April 2011
Received in revised form 29 June 2011
Accepted 1 July 2011
Available online 3 September 2011
Keywords:
Blade design
Turbine
Compressor
Airfoil
Curvature
Geometry
a b s t r a c t
The purpose of this paper is to present the advantages of the direct presCrIbed suRface Curvature distri-
bution bLade dEsign (CIRCLE) method for the design of high-efciency turbomachinery blades. These
advantages are illustrated by redesigning several examples of axial turbomachinery blades of interest
to energy conversion devices, and discussing in detail the aerodynamic performance and efciency
improvements of the redesigned blades over the original geometries. The two-dimensional (2D) method,
originally proposed for turbine blades, has been extended for use with 2D and three-dimensional (3D)
turbine, compressor and fan blades, and isolated airfoils. By specication, the method allows joining line
segments between the leading edge (LE) and trailing edge (TE) circles or ellipses so that the streamwise
distribution of surface curvature and slope of curvature are continuous everywhere from the LE stagna-
tion point to the TE stagnation point. The form of the line segments to prevent the wiggles of higher
order lines is presented. Also by specication, the CIRCLE method can be integrated with multi-objective
heuristic or evolutionary-algorithm optimization methods. The efcacy of the method is examined by:
redesigning two 2D turbine blades, one 2D compressor blade, and one 2D isolated airfoil; and by design-
ing one 3D compressor blade row and one 3D turbine blade row. The aerodynamic performance improve-
ments between the original and the sample redesigned blades are discussed in detail, resulting in higher-
efciency blades than the original geometries. Further extension of the method for centrifugal and
mixed-ow impeller geometries is a coordinate transformation. It is concluded that the CIRCLE method
is a new design environment enabling the original design (or redesign) of high-efciency 2D and 3D tur-
bomachine blades, with direct applications in a variety of energy conversion devices.
2011 Published by Elsevier Ltd.
1. Introduction
Axial, radial and mixed-ow turbomachine blades and airfoils
are used in all kinds of powerplants. On the large scale they are
used in airplane wings, and in gas turbines, steam turbines and
wind turbines for electricity generation; and in pumps, compres-
sors, fans and blowers for industrial processes. On the intermediate
scale they are used in auxiliary power turbines, automotive turbo-
chargers and water pumps, and even as pumps in household appli-
ances. On the small scale they are used in mechanical circulatory
support devices (ventricular assist devices), dentists drills, and
numerous other applications. In all cases the efciency of the over-
all process is highly dependent on the efciency of the turboma-
chine blades. Therefore design of high-efciency airfoils and
blades are essential for optimum aerodynamic, thermoeconomic,
and overall performance of turbomachinery-based powerplants
[13].
Gas turbine blades are three dimensional objects operating in
unsteady and complex ow elds. Designers decompose the three
dimensional (3D) blade-design problem to a series of two dimen-
sional (2D) problems in the streamwise direction. The 3D variation
of inlet and outlet ow angles vary from hub to tip and is deter-
mined by streamline curvature calculations [48]. The 2D blade
shapes are stacked to build the 3D geometry while meeting other
constraints. Turbomachinery manufactures have different se-
quences to design their blades, but all of them follow essentially
the above overall procedure. The goal of any blade design method
is to nd a geometry that satises ow requirements with mini-
mum loss, tolerable mechanical stresses, minimum disturbances
downstream and upstream, and in the case of compressors ade-
quate stall margin, among others. Compromises in performance
must be made to accommodate these three-dimensional con-
straints, material strength considerations, the location of cooling
passages and hollow sections, etc.
0306-2619/$ - see front matter 2011 Published by Elsevier Ltd.
doi:10.1016/j.apenergy.2011.07.004

Corresponding author. Tel.: +44 20 7882 5301.

E-mail address: korakianitis@alum.mit.edu (T. Korakianitis).
Applied Energy 89 (2012) 215227
Contents lists available at SciVerse ScienceDirect
Applied Energy
j our nal homepage: www. el sevi er . com/ l ocat e/ apener gy
Various investigators use different denitions for blade design
methods: direct; inverse, semi-inverse, full-inverse or full-optimi-
zation methods [9]; analysis and design modes [10]; optimization
and design methods [11]. In our work, the direct design has been
dened as a process in which the blade geometry is specied and
the resultant aerodynamic performance on that geometry is calcu-
lated; and the inverse method has been dened as a method in
which the desired blade performance is specied and the geometry
that would accomplish such performance is calculated.
In the direct method it is easy to fulll mechanical and geomet-
ric constraints, but it is usually laborious to obtain the desired dis-
tribution of pressure or velocity along the prole. On the other
hand, it is difcult to obtain an acceptable geometry with an in-
verse method [1215] because it requires multiple variations of
the velocity distribution until an acceptable prole geometry is ob-
tained. The inverse design method also has difculties in both the
leading edge (LE) and trailing edge (TE), due to mathematical sin-
gularity (zero velocity) at the stagnation points [15,16]. The inverse
method ends up with blades with zero thickness at the trailing
edge, which are impossible to manufacture; or with other adapta-
tions made at the trailing edge introducing uncertainties. This last
difculty makes the non-heuristic inverse method acceptable for
some compressor blade geometries of thin trailing edge, but unac-
ceptable for turbine geometries that have thicker trailing edges.
This paper presents a method for the design 2D and 3D blade
shapes with continuity in surface curvature and slope of surface
curvature from LE to TE, resulting in blades of inherently good
aerodynamic performance. The method can be used to provide
nished blade designs of high efciency, as illustrated in later
examples. Alternatively, it can also be used to provide initial geom-
etries for other direct and inverse design methods, or to provide
geometries for optimization methods with genetic and heuristic
algorithms.
The CIRCLE method is based on modications to the earlier 2D
blade-design method [1727] and its earlier 3D extensions [8] that
allow the designer to include 3D LE and TE circles or ellipses, while
maintaining continuous slope of curvature everywhere (2D and
3D) on the blade surfaces. The CIRCLE method starts from the TE
shapeanddesigns the 2Dbladeshapeinthree linesegments: y1near
the LE; y2 in the middle part of the surface; and y3 near the TE. By
specication the method ensures blade-surface curvature and
slope-of-curvature continuity from the LE stagnation point to the
TE stagnation point. The streamwise blade-surface curvature distri-
bution is manipulated to optimize the aerodynamic performance of
2DSections 2D shapes are designed near hub, mean and tip regions,
and the 3D blade shape is designed by smoothly (again with curva-
ture continuity) varying the 2D parameters from hub to tip.
The CIRCLE method decouples the traditional maximum thick-
ness and maximum camber discussions (used in early airfoil de-
signs) from blade design. Similarly to inverse design methods,
the CIRCLE method is guided by the surface pressure and surface
Mach number distributions with their relation to surface-curva-
ture distribution, and the output is the blade shape. The design se-
quence shapes the surface curvature and with it the location of
maximum loading, forwards or backwards, on the blade surface.
The advantages of the CIRCLE blade design method are illustrated
with several examples: redesign of two 2D turbine sections; rede-
sign of two 2D compressor blades; design of one 3D turbine blade-
row; design of one 3D compressor bladerow; and redesign of an
isolated airfoil used in wind turbines. The computed aerodynamic
improvements of the redesign example cases are discussed.
2. Effect of streamwise blade-surface curvature
The theoretical and experimental evidence that both curvature
and slope of curvature affect boundary-layer development and
Nomenclature
b axial chord (nondimensionally b = 1)
c blade chord, leading to trailing edge
c
0
, c
1
, . . . thickness coefcients (Eqs. (3), (5))
C1, C2, . . . Bezier control points (Fig. 1d)
C = 1/r curvature (Eq. (1) and Fig. 1d)
C
D
drag coefcient
C
f
skin-friction coefcient
C
L
tangential-loading (lift) coefcient
C
p
pressure coefcient
i incidence
k
1
, k
2
, . . . exponential thickness polynomials (Eqs. (3), (5))
L streamwise length along blade surface
M Mach number
o throat circle (Fig. 1a)
p pressure
P points or nodes on the blade surfaces
r local radius of curvature (Eq. (1))
Rey Reynolds number
R
n
stage reaction
S tangential pitch of the 2D blades (Fig. 1)
_
S rate of entropy generation along the blade surface
(x, y) Cartesian coordinates
(X, Y) nondimensionalized coordinates (with b) y1, y2, y3
blade segments: leading edge; main CIRCLE part; and
trailing edge (Figs. 1a, e and f)
Y
L
(p
o,in
p
o,ot
)/(p
o,in
p
st,ot
), pressure loss coefcient
z length along 3D blade height
z
0
(length along 3D blade height)/(blade height)
Z
L
(p
o,in
p
o,ot
)/p
o,in
, stagnation pressure loss factor
Greek
a ow angle
b blade-surface angle
k stagger angle of the blades
/ angle of throat diameter (Figs. 1a and e)
u stage ow coefcient
n1, w1 to n5, w5 parameters to specify 3D distribution of 2D-sec-
tion parameters (FigS. 1g and 6a)
w
L
stage loading coefcient (stage work coefcient)
Subscripts
cmb camber line
crd chord line
in blade inlet region
is isentropic (M
is
in Fig. 7)
o stagnation
ot blade outlet region
p pressure side
p2 pressure-side TE circle to y1 segment (Fig.1af)
pm pressure-side y1 to y2 segments (Fig. 1af)
pk pressure-side y2 to y3 segments (Fig. 1af)
p1 pressure-side y3 segment to LE circle (Fig. 1af)
s suction side
s2 suction-side TE circle to y1 segment (Fig. 1af)
sm suction-side y1 to y2 segments (Fig. 1af)
sk suction-side y2 to y3 segments (Fig. 1af)
s1 suction-side y3 segment to LE circle (Fig. 1af)
st static
216 T. Korakianitis et al. / Applied Energy 89 (2012) 215227
aerodynamic performance has been presented in [2527]. For a gi-
ven Reynolds number and with other factors being the same,
smooth streamwise surface curvature distribution results in lower
losses and therefore higher efciency, as is further justied in the
blade re-design examples in the following. One possible exception
to this general rule, where laminar ow about to separate may be
tripped to turbulent ow by introducing a curvature disturbance
on the blade surface before a separation bubble, is mentioned later
in this paper. One must distinguish here between: surface rough-
ness and fouling (with which turbomachine blades must operate);
and the slope-of-curvature discontinuities in the as-designed
shape at the junctions of the splines. The latter are invisible in
the 2D blade shapes and the blade looks very smooth; but they
may produce unusually-loaded blades, higher losses, and thicker
wakes.
Smooth streamwise blade-surface pressure distributions
(avoiding local accelerations and decelerations) require smooth
surface-curvature distributions (continuous slopes of pressure
and curvature along the blade surface). Continuous slope of curva-
ture requires continuous third derivatives at the splines or surface
patches used to design the blades, as illustrated in the following
two equations for curvature C and slope of curvature C
0
for
blade-surface line segments y = f(x), y
0
= df(x)/dx, y
00
= d
2
f(x)/dx
2
and y
000
= d
3
f(x)/dx
3
.
C
1
r

y
00
1 y
02

3=2
1
C
0

dC
dx

y
000
1 y
02

3y
0
y
02
1 y
02

5=2
2
The geometry of some blades presented in the literature exhibit
slope of curvature disturbances at spline knots along the main part
of the blades, affecting boundary layer development, the point of
transition, etc. Even more blades present a slope of curvature dis-
continuity where the LE circle or other shape joins the main part
of the blade, causing local ow accelerations and decelerations,
disturbances that in many cases cause LE separation bubbles, and
which always affect aerodynamic performance.
Overall, this local slope of curvature disturbance or discontinu-
ity has resulted in test and production airfoils (isolated and in turb-
omachines) that exhibit spikes or dips of various magnitudes in
isentropic surface Mach number and pressure-coefcient distribu-
tion, which occasionally result in unexpected loading distributions
along the blade length and in local separation bubbles. These ef-
fects are visible as small local kinks in surface pressure or isen-
tropic Mach number distributions in some of the computational
and experimental data published, for example, in [2831], and
with local separation bubbles in [32], and [3335]. In the following
we present several cases, where the CIRCLE method is used to re-
design blades exhibiting such ow disturbances caused by slope
of curvature discontinuities, providing similar-looking redesigned
blades of higher computed efciency.
3. 2D and 3D Blade and airfoil design
Fig. 1 illustrates the CIRCLE method, originally developed for 2D
and 3D turbine blades, and its modications for compressor blades
and isolated airfoils [17,18,2527,8]. A summary is included here
for completeness, and to place the discussion of results in context.
The suction and pressure sides are each divided in three segments,
y1, y2 and y3, which are joined to the LE and TE shapes. Key inputs
are the LE and TE circle radii (or ellipses, or other analytic shapes),
and the inlet and outlet ow angles a
1
and a
2
. Blade pitch S is a key
parameter for compressors and turbines, related to tangential load-
ing coefcient C
L
[2527], while it is not used in isolated airfoils.
Compressors and turbine blades are set at stagger angle k
(Figs. 1a and e), while for isolated airfoils this is set to k = 0
(Fig. 1f). Compressor and turbine geometries are dened on the
suction side by the distribution of the area available to the ow
along the passage, and by the minimum area in the passage (e.g.,
throat circle diameter o and angle / in Figs. 1a and e). In subsonic
turbine cases the area distribution is monotonically increasing
from inlet to the throat (Fig. 1a). In subsonic compressor cases
the area is increasing from the throat to the outlet (Figs. 1e). This
paper illustrates the design of subsonic cases, but the CIRCLE meth-
od allows for supersonic designs in which this area change from in-
let to outlet is not monotonic.
The throat circle o and its angle / from the axial direction deter-
mine the suction-surface blade control point P
sm
and correspond-
ing blade angle b
sm
. The corresponding input values on the
pressure side for turbines, compressors and airfoils are arbitrary
blade point P
pm
and blade angle b
pm
. For isolated airfoils one major
difference from compressors and turbines is that, on the suction
side, P
sm
and b
sm
are inputs unrelated to area passage.
The CIRCLE method presented in this paper illustrates the use of
LE and TE circles. These are the hardest shapes to join to the blade
surfaces as there is a transition from the constant curvature of the
circle region to the locally varying curvature of the remaining air-
foil surface. Therefore the method presents the most difcult case
of joining the LE and TE edge shapes to the rest of the airfoil sur-
face; and all other LE and TE shapes are easier variations of the
main methodology presented.
The TE radius and stagger angle k locate the TE circle (Figs. 1a, b
and e). The suctionand pressure surfaces detach fromthe TE circle
at points P
s2
and P
p2
(Fig. 1b). These specied by input parameters
b
s2
and b
p2
respectively (local airfoil-surface angles, determined
by the wedge angle of the trailing edge, and related to the outlet
ow angle a
ot
). The trailing edge region (line segment y3) from P
s2
to P
sm
on the suction surface is specied by an analytic polynomial
y = f(x) of the exponential form originally introduced in [8]:
y3f x c
0
c
1
xc
2
x
2
c
3
x
3
c
4
k
1
xxP
s2
c
5
k
2
xxP
s2

3
where k
1
and k
2
are exponential functions resulting in terms of
increasing importance as we approach point P
s2
, and of negligible
importance away from P
s2
. Thus Eq. (3) is a cubic equation near
point P
sm
; and the basic cubic equation has exponential modica-
tions as it approaches the TE circle at point P
s2
. The six coefcients
c
0
to c
5
are evaluated from the conditions of point, rst, second and
third derivative continuity (four conditions) of the airfoil surface
line at P
s2
; and prescribing the point and slope of the airfoil surface
at P
sm
(two additional conditions). This approach enforces slope of
curvature continuity in the vicinity of the TE circle, while concur-
rently avoiding the wiggles of higher-order polynomials. As the
TE circle is usually small, the changes in streamwise curvature in
this vicinity are usually large; but this approach enforces curvature
and slope of curvature continuity through these locally-large
streamwise changes of curvature.
The design of line segment y2 between points P
sm
and P
sk
is
accomplished by mapping the curvature distribution for the
shape of the blade surface in that region from the C vs. X plane
to the Y vs. X plane using four-point to six-point Bezier splines in
curvature (Fig. 1d). For illustration purposes Fig. 1d shows a six-
point Bezier spline, though in principle any n-point Bezier spline
and slope-continuous NURBS can be used; and usually four Bezier
control points are sufcient. The curvature segment corresponding
from P
s2
to P
sm
is evaluated from analytic polynomial y1 (using Eq.
(3)) and plotted on the C vs. X plane starting from the TE at X = 1.0
and ending in point C6s in Fig. 1d. The slope of the curvature C
s
(x)
at point C6
s
(corresponding to blade point P
sm
) is computed from
Eq. (3) and becomes an input to further calculations. On the
T. Korakianitis et al. / Applied Energy 89 (2012) 215227 217
C2p
Y
Y
Fig. 1. 2D and 3D blade geometry denition (adapted from [26,8]).
218 T. Korakianitis et al. / Applied Energy 89 (2012) 215227
curvature of the suction surface we specify points C1
s
, to C5
s
. Point
C1
s
is specied at an x location corresponding to P
sk
. Since the slope
of the Bezier curve is tangent to the line of knots at its ends, the
tangency condition at point C6
s
ensures slope-of-curvature conti-
nuity from C1
s
to C6
s
(from P
sk
to P
s2
).
Using central differences Eq. (1) is written for curvature at air-
foil point i as a function of (x, y) coordinates of points i 1, 1 and
i + 1 [26]. Given (x
i1
, y
i1
), (x
i
, y
i
), x
i+1
and C
i
we can compute y
i+1
starting from blade points P
sm
and progressing explicitly point by
point towards the leading edge to points P
sk
. The Bezier spline is
iteratively manipulated until the slope and the y location of the air-
foil surface at points P
sk
, and the shape of the curvature distribu-
tion, are acceptable.
In the LE area we implement a hybrid method based on modi-
cations of the earlier implementations [18,26,8]. First we introduce
the LE shape, such as a circle or ellipse (Fig. 1c). The suction and
pressure blade surfaces detach from the leading edge circle at
points P
s1
and P
p1
, specied by input parameters b
s1
and b
p1
respec-
tively. These are local blade-surface angles, determining the
wedge blade angle at the LE. Then a parabolic construction line
is dened, and a thickness distribution is added perpendicularly
to the construction line (as in [18,26]). The construction line starts
from a key geometric point such as the origin, the leading edge of
the blade, or the center of the leading edge circle. The thickness
distribution discussed below is added orthogonally to this para-
bolic construction line in a manner that the thickness distribution
(and therefore also the blade surface) have continuous point, rst,
second and third derivative (continuous y, y
0
, y
00
, y
000
and therefore
continuous C
0
) at both points P
s1
and P
sk
. P
s1
is where the LE thick-
ness distribution joins the LE shape (circle). P
sk
is where the thick-
ness distribution and resultant line y1 join the main part of the
blade y2, corresponding to curvature point C
1s
(Fig. 1 c and d). This
is analogous to the circle-joining work of [8] with exponentials in
the polynomials; and to the above trailing-edge region subsection
on joining the TE circle to y3.
The suction-side construction line can be (for instance) of the
form:
yx Ax
2
Bx C 4
and the thickness distribution y
t
added orthogonally to the con-
struction line (in order to subsequently arrive at the coordinates
of the leading edge segment y1) is of the form
y
t
c
0
c
1
x c
2
x
2
c
3
x
3
c
4
k
11
x xP
s1
c
5
k
12
x
xP
sk
c
6
k
13
x xP
s1
c
7
k
14
x xP
sk
5
where functions k
11
, k
12
, k
13
and k
14
are exponential polynomials.
These exponential functions acquire increasing importance as we
approach points P
s1
and P
sk
on the blade surface, so that Eq. (5) is
a cubic polynomial away from these two end points. The eight
parameters of the thickness function c
0
to c
7
are derived from the
conditions to match: y, y
0
, y
00
and y
000
(and thus C
0
) at point P
s1
;
and at point P
sk
respectively.
This approach ensures continuity of curvature and slope of cur-
vature from the TE circle to the main part of the blade surface
through the leading-edge thickness distribution and into the LE cir-
cle. The procedure is similar for the pressure side of the blade.
In the 2D isolated airfoil P
sm
and P
pm
, and blade angles at these
locations b
sm
and b
pm
are user specied. This feature can be used by
designers preferring to specify these points at the maximum airfoil
thickness thus relating this design aspect to the usual maximum
thickness and location of maximum thickness specications of
the usual isolated airfoil design methods. However, this is not nec-
essary (and in most cases it is not desirable) because in the CIRCLE
method the blade design is accomplished by comparing the
required surface pressure or isentropic Mach number distribution
with the surface curvature distribution, as originally illustrated in
[27]. This interdependence of surface pressure and surface curva-
ture distributions can be used to front-load, mid-load or aft-load
compressor, turbine or isolated airfoil blades; and as a result the
CIRCLE design method has the combined advantages of direct
and inverse design methods.
3.1. Extension from 2D to 3D blade and airfoil design
Fig. 1g, h and i illustrate the extension of the 2D blade design
method to 3D. The 3D blade designs are obtained as 2D sections
smoothly varying along the blade span. This is accomplished by
prescribing the smooth variation of 2D blade-design parameters
at any arbitrary blade-row radius along the blade span. Key inputs
for the 3D CIRCLE method are: the blade design parameters of
three key 2D blade sections (near the hub, mean and tip of the
3D blade); and the additional 3D parameters that specify the var-
iation of these 2D parameters along the blade span. For example
Fig. 1g illustrates the variation of one of these 2D blade design
parameters along the non-dimensional blade-row height z
0
as
described in further detail below. The parameter illustrated in this
case is C2
p
(the value of Bezier control point C2
p
on the pressure
side of the 2D method).
Many 3D parameters are functions of blade height (z) from the
throughow calculation, and they are transferred to fractional
blade height (z
0
). Flow angles a
in
(z) and a
ot
(z) are outputs of the
throughow calculation and become inputs to the 3D blade design.
The hub and tip diameter of the blade row and the number of
blades specify the blade pitch in each 2D section S(z).
Additional inputs specied by the user are o(z), k(z), and b(z).
For instance o(z) and k(z) can be specied at the hub, mean and
tip radii, and these key values can be used to provide smoothly
varying distributions of o and k along z and z
0
using Bezier curves
as described for parameter C2
p
below. Similarly b(z) may be
constant from hub to tip, or it may vary along z, thus providing
an additional input to control C
L
(z) in each 2D section.
For subsonic designs the values of a
in
and a
ot
usually vary
smoothly along z from hub to tip, and the resulting radial varia-
tions in these blade-design input parameters are also smooth and
relatively easy to specify. In each 2D section the values of points
C
s
and C
p
specifying the 2D blade section along X (Fig. 1d) can be
used to manipulate the streamwise curvature distribution of the
blade. Variations in curvature distribution can be used to front,
mid or aft load the pressure distribution in each 2D blade section
as described in [26,27]. Increasing the value of k in each 2D section
results in thinner and more front-loaded blades [26].
Typically we obtain the mean blade design (at z
0
= 0.5) as a 2D
section, and with either big or small changes in the blade parame-
ters (a
in
, a
ot
, k, etc.) along the turbine radius we also obtain the
near-hub (at z
0
0.0) and near-tip (at z
0
1.0) 2D blade geome-
tries, so that each of these three dominant 2D blade sections
has the desirable 2D aerodynamic performance at design and off-
design incidence, in the manner described in the previous section
and as illustrated in the blade re-design cases below. This gives
values for each one of the 2D blade design parameters at z
0
= 0.0,
0.5, and 1.0. Next, we prescribe the 3D variation of each 2D blade
design parameter with Bezier curves in the radial direction in the
manner illustrated in Fig. 1g. For instance for the blade design
parameter denoting the value of point C2
p
(Fig. 1g), we have input
values: at z
0
= 0.0 corresponding to ((n1, w1)); at z
0
= 0.5 corre-
sponding to ((n3, w3)); and at z
0
= 1.0 corresponding to ((n5, w5)).
We provide as additional inputs for the radial variation of C2
p
Bezier-curve control points (n2, w2) and n4. Then the Bezier curve
shown in Fig. 1g gives as an output the value of the control point,
w4. The resultant Bezier curve provides an overall description of
C2
p
from hub to tip of the blade. The Bezier curve specifying the
T. Korakianitis et al. / Applied Energy 89 (2012) 215227 219
3D variation of any blade design parameter may be: convex; con-
cave; or nearly linear. Note this approach requires only three
control-point inputs per 2D varying parameter like C2
p
.
The centers of gravity of the 2D sections can be stacked in any
radial orientation, such as along the radius, to reduce the bending
moment experienced by the spinning blades; or axial sweep, or lean
(dihedral, leaning of the blade perpendicular to the stagger angle or
in the tangential direction) may be introduced in the 3D shape.
The 3D CIRCLE method can provide sharper local variations of
3D parameters than those shown in Fig. 1g for transonic and super-
sonic bladerows. The method can be extended to radial and mixed-
ow turbomachines by a coordinate transformation ((x, y) along
the streamlines and z perpendicular to streamlines. Using this
method a variety of 3D turbine, compressor, and isolated airfoil
geometries have been designed. Sample 3D compressor and
turbine geometries are illustrated in Figs. 1h and i.
4. Circle-method use with blade optimization methods
The exact location of the n points controlling 2D curvature, such
as C1, C2, C3 and C4 (Fig. 1) in each 2D section, is not as critical as
the resulting shape of the curvature distribution; but these input
parameters are also specied as smoothly varying along r with
the Bezier curves in Fig. 1g. The resultant shapes can be stacked,
for instance along the center of gravity of the sections, resulting
in 3D blade shapes like the one illustrated in Figs.1i and h. Desired
changes in 3D surface pressure or 3D streamlines are compared
with changes in 3D curvature distributions and the location of
the 3D blade surfaces. After the rst iteration (rst geometric de-
sign and analysis) the user examines the resulting 2D or 3D blade
loading distributions, and decides, where to increase and decrease
local curvature (and local loading). After the second iteration the
user gains an appreciation of the magnitude of the required
changes in curvature to cause the desired 2D or 3D changes in
Mach number or pressure distribution; or other aspects, such as
the 3D passage vortex and ows near endwall regions. The proce-
dure is repeated until a desirable 2D 3D blade geometry and aero-
dynamic performance are obtained. The above procedure can be
automated with used-dened optimization functions and simple
or complex, visual or codied multi-objective heuristic or evolu-
tionary-algorithm optimization methods in order to optimize
various aspects of blade geometry or performance, e.g. [36,37].
C
S
M
is

Fig. 2. Comparison of original Kiock blade (from [38]) with redesigned S1 blade.
220 T. Korakianitis et al. / Applied Energy 89 (2012) 215227
5. Sample 2D turbine blade redesigns
Mesh generator GAMBIT and ow solver FLUENT have been
used in the RANS computations throughout this paper. Fig. 2 shows
RANS computations and comparison with the experimental results
of the turbine blade tested by Kiock et al. [38], and RANS computa-
tions of blade S1, redesigned with the CIRCLE method. For the
experiments M
in
= 0.260, M
ot
= 0.782, a
in
= 30, and a
ot
= 67.33.
The mesh elements used for the Kiock and S1 blades are: 20,108
quadrilateral cells; 39,253 2D interior faces; and 20,650 nodes
for all zones. A 15-layer structured O-mesh with y
+
< 3 was used
around the blades, and a pave mesh consisting of structured and
unstructured regions was used in the passage. The k x and
k turbulence models have been used with SST for transition.
The results for both turbulence models are approximately the
same, and the results shown are those for the k x SST-transition
model.
The experimental and computational data for the original Kiock
blade show disturbances on the suction surface M
is
at X 0.1,
X 0.5 and X 0.8; and on the pressure surface an acceleration-
deceleration region at X 0.05 (Fig. 2c). Fig. 2a shows the surface
curvature distributions for the original blade (jagged lines, evalu-
ated numerically from the original data points) and the curvature
distributions for the redesigned S1 blade (smooth lines). Compared
to the original blade the redesigned blade S1 on the suction surface
lowers curvature near X 0.6 (slightly raising the y location of the
surface and corresponding M
is
). To compensate for this geometry
change and still bring the leading edge to the geometric origin,
S1 lowers curvature near X 0.4 and raises it again near
X 0.15. As a result the y location of the S1 surface is lower near
X 0.4 and higher near X 0.15. Blade S1 has much smoother
computed surface M
is
than the original blade. Lines y1 join
smoothly the LE circle on both the suction and pressure sides, as
well as lines y2, removing the small ow disturbances of the origi-
nal blade in these regions. The small acceleration-deceleration
region on the pressure surface at X 0.05 has been removed.
Figs. 2d, e and Fig. 2f further illustrate the comparison of com-
puted surface M
is
, boundary layer displacement thickness d

and
entropy generated along the blade surface
_
S between the original
and redesigned blades.
The removal of the small M
is
disturbances at X 0.05 on the
suction and pressure sides of the blades has resulted in smoother
distributions of d

and lower local entropy generation

_
S in thee
regions. As a result of the removal of the deceleration-acceleration
region at 0.05 < X < 0.50 and the prolonged delay in ow accelera-
tion in the same region on the pressure surface, the growth of d

in
that region is more rapid in blade S1; and also as a result of the
higher acceleration at 0.50 < X < 1.00, d

is reducing much faster

in that region in blade S1. The smoother M
is
along the suction sur-
face of S1 has resulted in thinner d

and lower
_
S everywhere along
C
Fig. 3. Comparison of surface curvature, and of computed and tested isentropic surface Mach numbers, of original HD blade (from [3335]) with redesigned I1 and I9 blades
(adapted from [39]).
T. Korakianitis et al. / Applied Energy 89 (2012) 215227 221
Y
M
is
M
is
M
is
M
is
M
is

S
C
f
Fig. 4. Comparison of MAN GHH 1-S1 (Steinert, from [12]) with C1 and C2 compressor blades at various incidences.
222 T. Korakianitis et al. / Applied Energy 89 (2012) 215227
the suction surface, indicating lower losses than the original Kiock
blade. As a result the computed mass-weighted average stagnation
pressure loss for the Kiock blade is Z
L
= 0.0134; and for the S1 blade
is Z
L
= 0.00967.
Fig. 3 shows aspects of the geometry and aerodynamic perfor-
mance of blade HD [3335] and blades I1, I4 and I9 designed with
the CIRCLE method. The HD blade prole is a thin, hollow, castable
root section from the rotor of a low-pressure turbine. It was de-
signed to operate at air inlet ow angle 38.8 relative to the axial
direction and to provide approximately 93 of ow turning. The
test Rey = 2.3 10
5
. Further experimental details can be found in
[3335]. Joining the leading edge circle with the blade surfaces
causes local ow discontinuities and a suction side laminar separa-
tion bubble, after which the ow re-attaches and becomes turbu-
lent further downstream.
The RANS computations shown for the HD, I1, I4 and I9 blades
have used: 19,705 quadrilateral cells; 38,967 2D interior faces; and
20,148 nodes for all zones. Other aspects of the copmutations are
similar to those for Kiock and S1 blades. Fig. 3a shows the curva-
ture distribution of the original HD blade (jagged line, evaluated
numerically from the original blade data points) and of blades I1,
I4 and I9. The surface curvature distributions of blades I1, I4 and
I9 are smoother lines, as these blades have been designed with
the CIRCLE method. The gure also shows the curvature of blade
I1 trying to follow the curvature of the HD blade in the vicinity
of the leading edge with a curvature spike on the suction side.
This spike in the surface curvature of blade I1 (which we would
not normally use in this region of a blade design) is now required
in order to reproduce the ow spike in the LE region of the HD
blade in blade I1. The spike is not prescribed in the curvature
distributions of blades I4 and I9, which are smooth by specication
and design. The resultant computed isentropic Mach number sur-
face distributions are shown in Fig. 3b and c. The sharp local accel-
eration-deceleration region on the pressure side of the leading
edge of blade I9 has also been smoothed. The separation and re-
attachment points have been predicted accurately for the HD
blade. The mass-averaged stagnation pressure loss computed for
the HD blade is Z
L
= 0.00316 and for the I9 blade Z
L
= 0.00220.
Different methods to control the differences in the surface cur-
vature, especially between the LE shape and the rest of the blade,
have been proposed, for instance [4044]. CIRCLE is the most suc-
cessful blade-design method in the open literature in controlling
this LE spike difculty between the LE shape and the rest of
the blade.
6. Sample 2D Compressor blade redesigns
Fig. 4 shows a comparison of the geometry and aerodynamic
performance of the high-subsonic Mach number MAN GHH 1-S1
compressor blade, tested by Steinert et al. [12], with redesigned
blades C1 and C2, at design and off-design incidences 4 and
+5. C1 is the rst and C2 is the second redesign attempt. The solid
lines in Figs. 4b to f are viscous computations of the original blade
shape, and the dashed lines are the viscous computations of the
redesigned C1 and C2 blades. The mesh elements used for the com-
putations are: 30,520 quadrilateral cells; 60,558 2D interior faces;
and about 31,000 nodes for all zones. A 2D O-mesh and a
Pave-unstructured mesh consisting of a combination of structured
and unstructured regions have been used. The mesh around the
airfoil consisted of 21 structured clustered O-grid layers with wall
boundary parameter y
+
< 1. The remaining ow eld was discret-
ized with quadrilateral and a small numbers of triangular cells.
The k x SST-transition model has been used. The experimental
boundary conditions of this blade at design conditions are: p
o,in
=
101,325 Pa; inlet stagnation temperature 287.15 K; turbulence
intensity 1.5%, turbulence length scale l
m
/chord = 0.0476;
M
in
= 0.618; and pressure ratio 1.1021. Further details can be found
in [12]. The original blade exhibits LE spikes on the pressure side
at negative incidence, and on the suction side at positive incidence.
Blade C1 exhibits an acceleration-deceleration regime on the pres-
sure side near the LE. This has been largely removed in the second
redesign attempt, in blade C2.
Fig. 4g shows the computed d

along the suction and pressure

surfaces of the original and re-designed blade at several incidences.
The differences in the trends in thickness of d

between the Steinert

and C2 blades are as expected: d

is thicker in regions of higher

diffusion, and thinner in regions of lower diffusion, following the
M
is
trends in Fig. 4cf. Fig. 4h shows the computed
_
S along the suc-
tion surface of the original and re-designed blade. The redesigned
blade shows decreased entropy production everywhere on the sur-
face, and decreased skin friction coefcient, as shown in g.Fig. 4i.
The computed boundary layers are thinner, tolerance to inci-
dence is increased, and the rates of entropy generation are lower
Fig. 5. Comparison of pressure loss coefcient Y
L
, and stagnation pressure loss
factor Z
L
, between the MAN GHH 1-S1 (Steinert, from [12]) and C2 compressor
blades, at various incidences.
Table 1
Hub to tip blade-section ow parameters for the 3D compressor stator design shown
in Fig. 1h and Fig. 6.
a
in
a
ot
u
in
u
ot
w
L
R
n
C
L
Hub 46.57 30.88 0.560.6 0 0.36 0.58 0.853
1 46.25 30.74 0.560.59 0.34 0.59 0.850
2 45.96 30.61 0.550.58 0.32 0.59 0.850
3 45.72 30.47 0.550.57 0.31 0.59 0.852
4 45.53 30.34 0.540.56 0.29 0.59 0.858
Mean 45.39 30.19 0.540.55 0.28 0.60 0.868
6 45.25 30.04 0.530.54 0.27 0.60 0.878
7 45.14 29.89 0.520.53 0.25 0.60 0.890
8 45.06 29.74 0.520.53 0.24 0.60 0.904
9 45.01 29.60 0.520.52 0.23 0.60 0.920
Tip 45.00 29.46 0.510.51 0.22 0.60 0.938
Table 2
Hub to tip blade-section ow parameters for the 3D turbine stator design shown in
Fig. 1i and Fig. 7
a
in
a
ot
u
in
u
ot
w
L
R
n
C
L
Hub 0.00 68.29 0.820.75 2.55 0.39 0.837
1 0.00 66.6 0 0.8 00.77 2.41 0.42 0.904
2 0.00 65.94 0.780.77 2.31 0.44 0.940
3 0.00 65.45 0.760.75 2.21 0.46 0.972
4 0.00 64.88 0.750.74 2.13 0.48 1.005
Mean 0.00 64.30 0.730.73 2.08 0.50 1.039
6 0.00 64.00 0.720.72 2.00 0.51 1.064
7 0.00 63.59 0.7 00.70 1.93 0.53 1.092
8 0.00 63.22 0.690.69 1.86 0.55 1.120
9 0.00 62.87 0.680.68 1.74 0.56 1.146
Tip 0.00 62.47 0.670.67 1.71 0.58 1.174
T. Korakianitis et al. / Applied Energy 89 (2012) 215227 223
along the surfaces of the C2 than the original blade. As a result the
losses are lower, as shown in Fig. 5.
7. Sample 3D Compressor and turbine blade designs
The throughow calculation method described in [8] has been
used for the design of a sample 6-stage compressor and 3-stage
low-pressure turbine. The hub-to-tip results for the 4th-stage
compressor stator are shown in Table 1; and for the rst-stage
low-pressure turbine stator are shown in Table 2. These have
become inputs for the design of hub, mean and tip sections for
the 3D compressor and turbine blade rows. The turbine section
has been specically chosen with a
in
= 0 in order to further illus-
trate the capability of the CIRCLE method to design 3D blades
resisting LE disturbances at design and off-design point where
the LE circle (or other shape, less demanding in terms of curvature
change) joins the blade surfaces.
The design parameters for the 2D blade sections along the blade
span have been prescribed as described before: for instance the
variation of C2
p
for the turbine blade row is shown in Fig. 1g;
and the variation for stagger angle for the compressor blade row
is shown in Fig. 6a. Additional variations in sweep and dihedral
can be used to account for the passage vortex, ows in endwall re-
gions, and other 3D effects. With inputs like these, the compressor
and turbine sections shown in Fig. 1h and i have been obtained.
The 3D sections of the compressor stacked along the center of grav-
ity are illustrated in Fig. 6b. The aerodynamic performance of the
turbine stator at r
0
= 0.1, 05 and 0.9 at design incidence and at inci-
dence i
cmb
= 5 is shown in Fig. 7. These have been computed with
RANS computations of similar detail to those described in the pre-
vious blade redesigns. It is evident that, by specication, the
CIRCLE method can be used to eliminate the LE disturbance regions
at design and off-design incidence.
8. Sample 2D Isolated airfoil redesign
The Eppler 387 airfoil is used for some wind turbine blades. Like
many other airfoils, the original geometry exhibits slope of curva-
ture discontinuities in the region where the LE circle joins the
blade surfaces. The CIRCLE method has been used to obtain a series
of redesigned airfoils of geometry very similar to that of the origi-
nal airfoil, with successive aerodynamic improvements in each
successive redesign. In this paper we compare the performance
of the original Eppler 387 airfoil with that airfoils A2 and A4.
Fig. 8a shows a comparison of the geometries of the three air-
foils. The experimental data on the original Eppler airfoil are:
Rey = 10
5
; turbulence intensity 0.5%; and i
crd
= +4. Further experi-
mental details can be found in [45]. The RANS computations (k x
SST-transition model) used a 2D structured C grid (for the pointed
TE of the Eppler airfoil) and a structured O grid (for the circular TE
of the A2 and A4 airfoils). In these boundary-layer meshes we have
used 50 y points and clustering around the LE region with y + 61.2.
The boundary layer grids were surrounded by unstructured Pave C
meshes extending 12 chords upstream and 20 chords downstream
of the airfoils, with a total number of about 350,000 grid points in
each case.
Fig. 8b illustrates the computed LE ow-disturbance region of
the original Eppler geometry. The CIRCLE method removed this
disturbance region in the A2 and A4 airfoils (Fig. 8c). Such ow
disturbances near the LE region are frequently caused by surface-
curvature disturbances in the same region, and may cause local
separation bubbles, as in the HD turbine blade discussed earlier.
0.2 0 0.2 0.4 0.6 0.8 1 1.2
0.5
0
0.5
1
1.5
2
2.5
X
Hub
10%
20%
30%
40%
Mean
60%
70%
80%
90%
Tip
C.G. stacking
Hub
Y

Fig. 6. Sample 3D compressor design. Left: variation of stagger angle k along non-dimensional blade height z
0
for the compressor rotor shown in Fig. 1h. Right: 3D stacking of
the sections of the same blade (the x and y scales of the sections are unequal).
Fig. 7. Isentropic surface Mach number distributions of the bladerow of Fig. 1i at z
0
= 0.1, 0.5, and 0.9 at design point a
in
= 0 and at incidence 5.
224 T. Korakianitis et al. / Applied Energy 89 (2012) 215227
For the Eppler airfoil the computations show that the disturbance
does not cause locally reverse (separated) ow in that region, but a
local acceleration-deceleration in the surface-pressure distribu-
tion, as shown by the differences in C
p
distributions between the
Eppler and A4 airfoils on the suction side near the LE in Fig. 8d.
The circle points are the experimental data on the original airfoil
(from [45]); the solid line is the RANS computation of the original
blade shape; and the dashed line is the RANS computation of the
redesigned A4 airfoil. The removal of the ow-disturbance region
near the LE is caused by the smoothing of both the A2 and A4
curvature distributions between 0.01 < X < 0.02, as illustrated in
Fig. 8e. The curvatures of blades A2 and A4 are practically identical
except near the LE, as seen on the right side of Fig. 8e.
The computations indicate that, when these three airfoils
operate at Rey = 10
5
, the boundary layer remains laminar until
about 0.6 < X < 0.7. After that the momentum of the boundary layer
near the surface is insufcient to carry the ow, and there is a
laminar separation bubble in that region. The RANS computations
indicate that the ow re-attaches turbulent; but the feature of the
laminar separation in that region is a characteristic of the Reynolds
number of the ow and the basic diffusion required by the airfoils
(Eppler, A2 and A4). Both blades A2 and A4 have removed the small
change in the slope of curvature that occurs at X 0.6 in the Eppler
airfoil (Fig. 8e); but the smooth curvatures of A2 and A4 cannot
change the nature of the ow, and cannot prevent the laminar
separation in that region. However, the smoothing of the curvature
from the LE region and throughout the airfoil surface has a bene-
cial effect on performance. At i
crd
= 4 (Fig. 8d), the computations
indicate that the transition point of the Eppler airfoil is at
X = 0.677; of A2 is at X = 0.682; of A4 is at X = 0.688; and the corre-
sponding airfoil wakes are progressively thinner, resulting in lower
values of C
D
, higher values of C
L
/C
D
, and higher efciencies.
Airfoils A2 and A4 have small differences in geometry except
near the LE. The wedge angle on the suction side is 35 for A2
and 25 for A4; resulting in a more rounded LE for A4 compared
to A2, and the differences in curvature near the LE shown in Fig. 8e.
This curvature difference has an effect on the aerodynamic perfor-
mance of the blades, shown in the polars of the airfoils. These have
been obtained with XFOIL [46] and the computed results are
shown in Fig. 8fh. The combined effect is shown in the C
L
/C
D
distributions in Fig. 8h. The values of C
L
/C
D
for airfoil A4 have
higher values and wider plateau than both A2 and the original
y/c
C
p
C
L
C
L
C
D
C
D
C
C
Fig. 8. Comparison of Eppler 387 and redesigned A2 and A4 airfoils.
T. Korakianitis et al. / Applied Energy 89 (2012) 215227 225
Eppler airfoils, throughout the i
crd
range, except very near i
crd
= 6.5
(where A2 is slightly better than A4).
9. Conclusions
This paper presents the CIRCLE method for the design of 2D and
3D subsonic, transonic or supersonic blades for axial compressors
and turbines, and isolated blades or airfoils. The method is illus-
trated with: two 2D turbine-blade redesigns; one 2D compres-
sor-blade redesign; one new 3D compressor blade design; one
new 3D turbine blade design; the redesign of the isolated Eppler
387 to the A2 and A4 airfoils. This details of the aerodynamic
improvements are discussed in each redesign case.
The CIRCLE method is based on prescribing streamwise 2D
suction- and pressure-surface smoothly-varying curvatures from
leading to trailing edge of the blades and airfoils. This curvature
and slope of curvature continuity includes the locations where the
suction and pressure surfaces join the leading and trailing edge cir-
cles, ellipses, or other shapes, so that curvature and slope of curva-
ture are smooth and continuous everywhere along the blade
surfaces fromLE stagnation point to TE stagnation point. The CIRCLE
method can be easily coupled to multi-objective heuristic or evolu-
tionary-algorithmoptimizationmethods for blade andairfoil design.
In the 3D method the 2D sections of the hub, mean and tip (or
near hub, mean and tip) are designed rst. As a rst step the 2D
blade surface curvature distributions in these three key sections
is manipulated until a desirable aerodynamic performance at
design point ow as well as at incidence ow is obtained, avoiding
local ow acceleration-deceleration regions and other ow distur-
bances. Once completed, the blade geometries of these three key
sections are kept constant. The blade-design parameters for any
other intermediate 2D section are smoothly varied from hub to
tip with Bezier curves in the radial direction, providing a smooth
variation of 2D blade sections from hub to tip. The 3D variation
of these blade design parameters is iteratively manipulated until
a desirable aerodynamic performance from hub to tip is obtained
with 2D or 3D ow computations. The 2D blade sections are
stacked from hub to tip along the centers of gravity, or the leading,
or the trailing edge, or with another stacking strategy. The resul-
tant 2D and 3D blades exhibit superior aerodynamic characteris-
tics, while concurrently the designer has full control of blade
structural characteristics.
The CIRCLE method is a new design environment that attaches
greater signicance to the streamwise surface curvature distribu-
tion rather than the exact location of (x, y) points on the blade, even
though the designer has direct control of the blade surface as in di-
rect methods. Similarly to inverse design methods, the CIRCLE
method is guided by the surface pressure and surface Mach num-
ber distributions (via their relation to surface-curvature distribu-
tions), and the output is the blade shape. The design sequence
shapes the surface curvature and with it the location of maximum
loading, forwards or backwards, on the blade surface. Therefore
this method combines the best advantages of direct and inverse
blade design methods. It is concluded that the method is a new
design environment enabling design of higher-efciency turboma-
chine blades.
Acknowledgments
The authors acknowledge the contributions of MSc and PhD
students, and postdocs, who over two decades have contributed
to coding various aspects of the CIRCLE blade design method in
FORTRAN, C++ and MATLAB, and on various platforms and operat-
ing systems: George Pantazopoulos; Nick Vlachopoulos; Paschalis
Papagiannidis; Dequan Zou; Richard Binzley; Sean Spicer; Brandon
Wegge; Yan Tan; Mingyu Shi; and Akbar Rahideh. The PhD
research of Idres Hamakhan was sponsored by the Ministry of
Education of Kurdistan; and the PhD research of M. Amin Rezaienia
is currently sponsored by Queen Mary, University of London.
References
[1] Ghigliazza F, Traverso A, Massardo AF. Thermoeconomic impact on combined
cycle performance of advanced blade cooling systems. Appl Energy
2009;86(10):213040.
[2] Agazzani A, Massardo AF, Frangopoulos CA. Environmental inuence on the
thermoeconomic optimization of a combined plant with NOx abatement. Trans
ASME, J Eng Gas Turb Power 1998;120:55765.
[3] Fast M, Assadi M, De S. Development and multi-utility of an ann model for an
industrial gas turbine. Appl Energy 2009;86(1):917.
[4] Massardo A, Satta A. Axial-ow compressor design optimization. Part 1.
Pitchline analysis and multivariable objective function inuence. Trans ASME, J
Turbomach 1990;112(3):399404.
[5] Massardo A, Satta A, Marini M. Axial-ow compressor design optimization.
Part 2. Throughow analysis. Trans ASME, J Turbomach 1990;112(3):40510.
[6] Massardo AF, Scial M. Thermoeconomic analysis of gas turbine based cycles.
Trans ASME, J Eng Gas Turb Power 2000;122:66471.
[7] Pachidis V, Pilidis P, Talhouarn F, Kalfas A, Templalexis I. A fully integrated
approach to component zooming using computational uid dynamics. Trans
ASME, J Eng Gas Turb Power 2006;128(3):57984.
[8] Korakianitis T, Wegge BH. Three dimensional direct turbine blade design
method. AIAA paper 2002-3347. In: AIAA 32nd uid dynamics conference and
exhibit, St. Louis, Missouri; June 2002.
[9] Meauze G. Overview on blading design methods. In: Blading design for axial
turbomachines, AGARD Lecture Series 167, AGARD-LS-167. AGARD; May 1989.
[10] Stow P. Blading design for multi-stage hp compressors. In: Blading design for
axial turbomachines, AGARD Lecture Series 167, AGARD-LS-167. AGARD; May
1989.
[11] Bry PF. Blading design for cooled high-pressure turbines. In: Blading design for
axial turbomachines, AGARD Lecture Series 167, AGARD-LS-167. AGARD; May
1989.
[12] Steinert W, Eisenberg B, Starken H. Design and testing of a controlled diffusion
airfoil cascade for industrial axial ow compressor application. Trans ASME, J
Turbomach 1991;113:58390.
[13] Selig MS. Multipoint inverse design of an innite cascade of airfoils. AIAA J
1994;32(4):77482.
[14] Dang T, Damle S, Qiu X. Euler-based inverse method for turbomachine blades.
Part 2. Three-dimensional ows. AIAA J 2000;38(11):200713.
[15] Phillipsen B. A simple inverse cascade design method. ASME paper 2005-GT-
68575; 2005.
[16] Liu G-L. A new generation of inverse shape design problem in aerodynamics
and aero-thermoelasticity: concepts, theory and methods. Int J Aircr Eng
Aerosp Technol 2000;72(4):33444.
[17] Korakianitis T. A design method for the prediction of unsteady forces on
subsonic, axial gas-turbine blades. Doctoral dissertation (Sc.D., MIT Ph.D.) in
Mechanical Engineering, Massachusetts Institute of Technology, Cambridge,
MA, USA; September 1987.
[18] Korakianitis T. Design of airfoils and cascades of airfoils. AIAA J
1989;27(4):45561.
[19] Korakianitis T. On the prediction of unsteady forces on gas turbine blades. Part
1. Description of the approach. J Turbomach, Trans ASME 1992;114(1):11422.
[20] Korakianitis T. On the prediction of unsteady forces on gas turbine blades. Part
2. Analysis of the results. J Turbomach, Trans ASME 1992;114(1):12331.
[21] Korakianitis T. Blade-loading effects on the propagation of unsteady ows and
on forcing functions in axial-turbine cascades. J Phys III 1992;2(4):50225.
[22] Korakianitis T. On the propagation of viscous wakes and potential-ow in
axial-turbine cascades. J Turbomach, Trans ASME 1993;115(1):11827.
[23] Korakianitis T, Pantazopoulos G. Improved turbine-blade design techniques
using fourth-order parametric spline segments. Comput Aided Des (CAD)
1993;25(5):28999.
[24] Korakianitis T. Inuence of statorrotor gap on axial-turbine unsteady forcing
functions. AIAA J 1993;31(7):125664.
[25] Korakianitis T. Hierarchical development of three direct-design methods for
two-dimensional axial-turbomachinery cascades. J Turbomach, Trans ASME
1993;115(2):31424.
[26] Korakianitis T. Prescribed-curvature distribution airfoils for the preliminary
geometric design of axial turbomachinery cascades. J Turbomach, Trans ASME
1993;115(2):32533.
[27] Korakianitis T, Papagiannidis P. Surface-curvature-distribution effects on
turbine-cascade performance. J Turbomach, Trans ASME 1993;115(2):33441.
[28] Okapuu U. Some results from tests on a high work axial gas generator turbine.
ASME paper 74-GT-81; March 1974.
[29] Gostelow JP. A new approach to the experimental study of turbomachinery
ow phenomena. ASME paper 76-GT-47; 1976.
[30] Wagner JH, Dring RP, Joslyn HD. Inlet boundary layer effects in an axial
compressor rotor. Part 1. Blade-to-blade effects. ASME paper 84-GT-84;
1984.
[31] Sharma OP, Pickett GF, Ni RH. Assessment of unsteady ows in turbines. ASME
paper 90-GT-150; 1990.
226 T. Korakianitis et al. / Applied Energy 89 (2012) 215227
[32] Hourmouziadis J, Buckl F, Bergmann P. The development of the prole
boundary layer in a turbine environment. Trans ASME, J Turbomach
1987;109(2):28695 [ASME paper 86-GT-244, 1986].
[33] Hodson HP, Dominy RG. Three-dimensional ow in a low pressure turbine
cascade at its design condition. Trans ASME, J Turbomach 1987;109(2):17785
[ASME paper 86-GT-106].
[34] Hodson HP, Dominy RG. The off-design performance of a low-pressure turbine
cascade. Trans ASME, J Turbomach 1987;109(2):2019 [ASME paper 86-GT-
188].
[35] Hodson HP. Boundary-layer transition and separation near the leading edge of
a high-speed turbine blade. Trans ASME, J Eng Gas Turb Power
1985;107:12734 [ASME paper 84-GT-179].
[36] Kim HJ, Koc S, Nakahashi K. Surface modication method for aerodynamic
design optimization. AIAA J 2005;43(4):72740.
[37] Samad A, Kim KY. Shape optimization of an axial compressor blade by multi-
objective genetic algorithm. Proc Inst Mech Eng. Part A. J Power Energy
2008;222(A6):599611.
[38] Kiock R, Lehthaus F, Baines NC, Sieverding CH. The transonic ow through a
turbine cascade as measured in four European wind tunnels. Trans ASME, J Eng
Gas Turb Power 1986;108(2):27784 [ASME paper 85-IGT-44, 1985].
[39] Hamakhan IA, Korakianitis T. Aerodynamic performance effects of leading
edge geometry in gas turbine blades. Appl Energy 2010;87(5):1591601.
[40] Buche D, Guidati G, Stoll P. Automated design optimization of compressor
blades for stationary, large-scale turbomachinery. In: Proceedings of IGTI03
ASME Turbo Expo 2003: Power for Land, Sea and Air, Atlanta, Georgia, USA;
June 1316, 2003.
[41] Sieverding F, Ribi B, Casey M, Meyer M. Design of industrial axial compressor
blade sections for optimal range and performance. Trans ASME, J Turbomach
2004;126(2):32331.
[42] Li H-D, He L, Li YS, Wells R. Blading aerodynamics design optimization with
mechanical and aeromechanical constraints. In: Proceedings of ASME Turbo
Expo, GT2006-90503: Power for Land, Sea and Air, Barcelona, Spain; May 8
11, 2006.
[43] Lee KS, Kim KY, Samad A. Design optimization of low-speed axial ow fan
blade with three-dimensional RANS analysis. J Mech Sci Technol
2008;22:18649.
[44] Chen B, Yuan X. Advanced aerodynamic optimization system for
turbomachinery. Trans ASME, J Turbomach 2008;130:0210050210017.
[45] McGhee RJ, Walker BS. Experimental results for the Eppler 387 airfoil at low
Renolds numbers in the Langley Low Pressure Turbine Tunnel. NASA-TM-
4062; 1988.
[46] Drela M, Giles MB. Viscous-inviscid analysis of transonic and low Reynolds-
number airfoils. AIAA J 1987;25(10):134755.
T. Korakianitis et al. / Applied Energy 89 (2012) 215227 227