3URFHHGLQJVRIWKH$60(-60(.

60(-RLQW)OXLGV(QJLQHHULQJ&RQIHUHQFH
$-.)OXLGV
-XO\+DPDPDWVX6KL]XRND-$3$1
Proceedings of ASME-JSME-KSME Joint Fluids Engineering Conference 2011
AJK2011-FED
July 24-29, 2011, Hamamatsu, Shizuoka, JAPAN

$-.
AJK2011-05011

THREE DIMENSIONAL OPTIMUM DESIGN OF A STEAM TURBINE STAGE WITH

NURBS CURVES

Tomohiro Tejima Yoshiki Niizeki

Toshiba Corporation Toshiba Corporation
Keihin Product Operations, Keihin Product Operations,
Yokohama 230-0045, JAPAN Yokohama 230-0045, JAPAN

Naoki Shibukawa Akihiro Onoda Satoru Watanabe

Toshiba Corporation Toshiba Corporation Toshiba Corporation
Yokohama 230-0045, JAPAN Yokohama 230-0045, JAPAN Yokohama 230-0045, JAPAN

Xin Yuan Zhirong Lin Xiaofeng Zhu

Key Laboratory for Thermal Key Laboratory for Thermal Key Laboratory for Thermal
Science and Power, Engineering Science and Power, Engineering Science and Power, Engineering
of Ministry of Education, of Ministry of Education, of Ministry of Education,
Tsinghua University, Beijing Tsinghua University, Beijing Tsinghua University, Beijing
100084, CHINA 100084, CHINA 100084, CHINA

ABSTRACT Stagger angle distribution in the radial direction was

Progress in the computer performance has enabled thought as one of the important design parameters of turbine
automatic optimization of the three dimensional shape of blades because it determines the flow pattern in radial
turbine blades with a large number of large-scale CFD direction. Then it was chosen as an optimized parameter with
(Computational Fluid Dynamics) calculations. This paper NURBS curves in this system.
presents an advanced aerodynamic optimization system for First, DOE was used for the human optimization, in which
turbine blades. The system can automatically find improved the parameter range for the advanced optimization was
blade shapes that give better aerodynamic performance in a estimated and the best shape obtained was used as the initial
turbine stage and reduces human efforts to generate blade shape shape for the evolutionary optimization to explore better blade
data, computational mesh, CFD input data etc. shape parameters.
The system consists of three parts; parameter updating Stage loss of an exhaust stage of IP (Intermediate Pressure)
part, blade shape generation part, and aerodynamic turbine which contains relatively high aspect ratio was chosen
performance evaluation part. In the parameter updating part, as the objective. In spite that such kind of stage was considered
users can choose DOE (Design of Experiment) or evolutionary not to be sensitive to three dimensional stacking, the results
optimization method such as GA (Genetic Algorithm), ASA showed good performance enhancement.
(Adaptive Simulated Annealing), etc. to define the parameters
in each step. The shape generation part changes the blade INTRODUCTION
shapes using NURBS curves whose control point parameters Among coal fired, combined cycle and nuclear power
are defined in the parameter updating part. Three-dimensional plants, steam turbine is taken as important component for
CFD grid is automatically generated for the changed blade electric power generation systems. Although its performance
shapes and steady CFD calculation is used to evaluate the seems to be at high and saturated level, they still require
aerodynamic performance of the changed blades in a turbine efficiency enhancement, because about 0.1% order
stage. improvement is large enough to achieve economical benefit and
carbon dioxide reduction. A steam path design, especially that

1 Copyright © 2011 by JSME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/ajk2011/69936/ on 04/05/2017 Terms of Use: http://www.asme.o

 of blade shape for both stator and rotor is traditional, but still TABLE 1. STAGE SPECIFICATIONS
thought to be core technology which should be conducted with
computational fluid dynamics (CFD). Recently, computers and Blade Height Aspect Ratio
CFD codes have become more powerful and designers come to [mm]
be able to do huge amount of calculation. A coupling CFD HP 1st stator 62 1.0
codes with optimization tools is likely to be adopted for rotor 63 1.3
effective designs, which can reduce exploratory design iteration IP Exh. stator 344 5.4
cycles and enhance the design optimization scopes. There have rotor 360 3.4
been plenty of optimization examples for turbine blade design
[1] [2]. Lampart presented his numerical optimization results of
a high pressure turbine stage with stagger angle and compound 100%
lean stacking and showed some efficiency improvement [3].
Although 3D blade design should include altering blade 90%
profiles, which should be required more in very long blade
80%
design [4], it would be practical to change stacking
configurations without blade profile tuning in many cases, 70%
mainly because the profile changing will take more
computational resource and time than expected benefit. Major 60%
effects of three dimensional blade stacking are studied by 50%
several researchers [5-8].
Not only aerodynamic parameter like mass flow rate and 40%
exit flow angle, structural constraints are also important issue
which should be considered in any optimizations. 30%
Leaving Loss
Sivashanmugam et al. introduced computational structural 20% Leakage Loss
dynamics to a turbine stage optimization and presented a
Secondary Loss
solution which chose the best parameter combination of 10%
Profile Loss
efficiency and blade stress [9].
0%
Based on a turbomachinery aerodynamic optimization
system [10], Yuan et al. carried out a thoroughgoing CFD HP 1st IP Exh.
approach with an optimization system. The rotor blade and hub
end-wall surface of a typical high pressure steam turbine stage FIGURE 1. LOSS COMPONENTS
(LEFT: HP 1ST STAGE, RIGHT: IP EXHAUST STAGE)
were optimized [11]. The blade shape deformations include
blade profiles, lean stacking line and twist stacking line. An
optimization considering the effect of tip leakage was also
carried out. NOMENCLATURE
A typical trend in the optimization is that many researches NURBS Non-Uniform Rational B Spline
have been interested in relatively short height blades whose
aspect ratios are about 1.0 or less. It is mainly because A(u ) The position of a point on the curve
secondary losses occupies more portion in low aspect ratio u Parameter of NURBS curve
stages and stages with higher aspect ratio more than three are
thought to have relatively weaker loss dependence on blade
N i , p (u ) B-spline basis functions
stacking configurations, thus, a designer is apt to hesitate to ωi Respective weights
take much time to do stacking survey, as shown in Tab. 1 and
Fig. 1. Pi Control points coordinates
In this paper, however, authors focused on the leaving loss, h Height coordinate of a control point
which is the lost kinetic energy of the exhaust flow, of an
exhaust stage with fairly high aspect ratio and the combination δ Rotation angle
of blade loss and leaving loss was chosen as the objective. s/t Throat by pitch ratio
Stage performance enhancement with an evolutionary
optimization system was compared to the base result of human a Coefficient of quadratic function (h2)
design and the optimized combination of stagger angle and lean b Coefficient of quadratic function (h)
shape achieved good performance enhancement.
c Constant term of quadratic function

2 Copyright © 2011 by JSME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/ajk2011/69936/ on 04/05/2017 Terms of Use: http://www.asme.o

 MODEL In the parameter updating part, the system changes the
The steam turbine stage model for the present fluid parameters according to the optimization method. Users can
dynamic optimum design study is a typical IP turbine exhaust choose DOE (Design of Experiment) or searching type
stage for 660MW-class super-critical pressure steam turbines. optimization methods such as GA (Genetic Algorithm), ASA
The inner and outer wall diameter increase in the flow (Adaptive Simulated Annealing), etc. to define the parameters
direction. The slant angles to the axial direction are 4°in the in each step. Before starting the optimization, users choose
inner wall and 20°in the outer wall. which parameters to change and give the range of each
A typical package software for CFD and mesh generation parameter. Users can also give other constraints such as
is used for the numerical analysis. A three-dimensional steady strength properties to avoid unrealizable blade shapes.
RANS solver is used. The CFD domain is one stage, i.e. the In the blade shape generation part, blade shapes are
stator blades and the rotor blades, shown in Fig. 2. Only one generated using NURBS curves whose parameters are defined
pitch is modeled around both the stator blade and the rotor in the parameter updating part. NURBS curves are used not
blade. The meshes are re-generated by moving the grids every only for blade profiles but also for the stacking line
time the blade shapes are changed. A periodic condition is used distributions and the stagger angle distributions in the radial
at the circumferential periodic boundaries. A mixing-plane direction. The NURBS technique can generate smooth curves
treatment is used at the stator-rotor interface, where the with a small number of parameters.
variables are averaged at each spanwise section. The inlet total In the aerodynamic performance evaluation part, Three-
pressure, the inlet total temperature and the outlet static dimensional CFD grid is generated for the generated blade
pressure are given. shapes and steady CFD calculation is executed to evaluate the
aerodynamic performance of the generated blades in a turbine
stage.
The evaluation result in the aerodynamic performance
evaluation part is returned to the parameter updating part. The
result is used as the response to the parameters in each step of
the optimization method. Then, new parameters are defined
using this response when a searching type optimization method
is used.
Designers have only to define the parameters and the
optimization method. The system reduces their efforts to
generate blade shape data, computational mesh, CFD input data
etc.

FIGURE 2. CFD DOMAIN

(LEFT: STATOR BLADE, RIGHT: ROTOR BLADE) Parameter update
(GA,ASA,etc.)

Blade shape
generation
METHOD
An advanced aerodynamic optimization system for turbine
blades is presented. The system can automatically find
Aerodynamic performance
improved three dimensional shapes of both stator and rotor
evaluation
blades that give better aerodynamic performance in a turbine (Mesh generation, CFD)
stage.
A commercial optimization package software is used for
the optimization and the sequential operation. A Non-Uniform
candidate
Rational B-Splines (NURBS) is used for parametric
representations of blade shapes. A commercial Navier-Stokes
CFD solver is used to evaluate turbine performance.
FIGURE 3. FLOW DIAGRAM OF AUTOMATIC BLADE
OPTIMIZATION SYSTEM
Automatic Blade Optimization System
The system consists of three parts; parameter updating
part, blade shape generation part, and aerodynamic
performance evaluation part. The flow diagram of the system is
shown in Fig.3.

3 Copyright © 2011 by JSME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/ajk2011/69936/ on 04/05/2017 Terms of Use: http://www.asme.o

 Blade Design With NURBS Curves the lean shape of the stator blades. The blade cross-sectional
NURBS curves are used for the stagger angle distributions profiles are not changed throughout all cases.
and the stacking line distributions in this paper. The optimization objective is the stage loss. Total to static
A NURBS curve is given by a weighted average of all loss is used as stage loss, which is composed of the total to total
control points Pi . The weight function is B-spline basis loss (blade loss) and the leaving loss. These three losses are
used as indices of the performance evaluation of the turbine
function N i , p (u ) times a weight ωi . So the stagger angle stage.
distribution or the stacking line distribution is determined once
the control points and the corresponding weights are specified.
A NURBS curve is defined such that BASE RESULT: DOE HUMAN OPTIMIZATION
n DOE is used in the human optimization. Stagger angle
∑W i, p (u ) Pi
(1) distributions in radial direction of both the stator and the rotor
A(u ) = i =0
n blades are chosen as optimization parameters, which are
∑W
i =0
i, p (u ) changed within a designer’s concept.
s/t distributions of both the stator and the rotor blades are
With expressed in a quadratic function form of height, such that
Wi , p = Ni , p (u )ωi (2)
s/t(h) = a h2 + b h + c (4)
Where Pi are control points coordinates, ωi their respective where s/t(h) is the throat by pitch ratio at height h. If “a” is
weights, N i , p (u ) the pth-degree B-Spline basis functions, and positive, the s/t in the middle span is larger than those of both
ends, so it is called A-shape in the paper. On the contrary, if “a”
A(u ) the position of a point on the curve. The basis function is negative, the s/t is called U-shape in the paper. The
are obtained through a knot vector, which defines the functions’ coefficients a and b are chosen as optimization parameters and
break points, of the form the constant term c is calculated from a and b so that the throat
area keeps the designed value.
1K
{0, , 0, u p +1 ,K, um− p +1 ,1,
23 { K,1} (3) The optimization parameters are the coefficients of the
p +1 p +1 stator as and bs, and the coefficients of the rotor ar and br. The
number of the parameters is four.
Third degree NURBS curves with four control points are Two L9 orthogonal arrays (3 x 3 factorial experiment
used to represent the stagger angle distribution and the stacking design tables) are used in the DOE; one is to optimize the
line distribution. Positions of NURBS control points excluding combination of as and ar, and the other is to optimize the
their corresponding weights are given to optimization combination of bs and br. All the parameters are set three-level.
programs. The coefficients bs and br are kept constant in the as and ar
To represent the stagger angle distributions in the radial DOE, and as and ar are kept constant in the bs and br DOE.
direction with NURBS curves, the control point vector is The stage efficiency is calculated using CFD for 18
defined in the form of (h, δ), where h is the height in the designs. Two Second-order RSMs (Response Surface Models)
radial direction and δ is the rotation angle from the given are generated from each L9 DOE result to approximate the
original shape. efficiency response to the parameters as and ar, and bs and br.
The efficiency contour maps of the RSMs are shown in Fig. 4.
Optimization Methods
Three optimizations are carried out and compared. The
first one is optimization by human and the others are
optimizations by the system. DOE is used in the human
optimization and ASA is used in the optimizations by the
system.
In the first and second optimizations, stagger angle
distributions in radial direction of both the stator and the rotor
blades are chosen as optimization parameters, which are
thought as important design parameters of turbine blades
because they determine the flow pattern in radial direction. In
this paper, throat by pitch ratio (s/t) distributions are used as
corresponding variables to stagger angle distributions. In the
third optimization, stacking line distribution of the stator blades
is added to the optimization parameters. The circumferential
coordinates of the stacking line are changed, which determine

4 Copyright © 2011 by JSME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/ajk2011/69936/ on 04/05/2017 Terms of Use: http://www.asme.o

 Best Point 1.4
STATOR
-0.10 1.2

1.0

NORMALIZED s/t
rotor a

-0.15 0.8

0.6
-0.20
0.4 FV
DOE Stag.
0.2
-0.25
0.0
-0.32 -0.24 -0.16 -0.08 0.00 0.08 0.0 0.2 0.4 0.6 0.8 1.0
stator a NORMALIZED HEIGHT

-0.08
Best Point 1.4
-0.09 ROTOR
1.2
-0.10
NORMALIZED s/t
1.0
rotor b

-0.11
0.8
-0.12 0.6

-0.13 0.4 FV
0.2 DOE Stag.
-0.14
0.0
0.03 0.06 0.09 0.12 0.15 0.0 0.2 0.4 0.6 0.8 1.0
stator b NORMALIZED HEIGHT

FIGURE 4. CONTOUR MAPS OF EFFICIENCY

(UPPER: as VS ar, LOWER: bs VS br) FIGURE 5. THROAT BY PITCH RATIO DISTRIBUTIONS
(UPPER: STATOR BLADE, LOWER:ROTOR BLADE)

The parameters are optimized using these RSMs. The s/t TABLE 2. NORMALIZED LOSSES
distributions of the optimized stator and rotor blades are shown
in Fig. 5. The results are compared with the free-vortex design FV DOE
stage. The solid line labeled ‘FV’ shows the s/t distribution of Stag.
the free-vortex design and the broken line labeled ‘DOE Stag.’ Stage Loss 1.000 0.991
shows the s/t distribution of the optimized design by DOE. Both Leaving Loss 0.525 0.518
the stator and the rotor blades have A-shape distributions. Blade Loss 0.475 0.473
The performance of the optimized design is calculated
using CFD. The normalized losses of the optimized design are
shown in Tab. 2. The losses are normalized by the stage loss of
the free-vortex design. The stage loss of the optimized design is It is mainly because the leaving loss is reduced. Generally,
reduced to 0.991. a flat distribution of the outlet velocity in the radial direction
with 0 swirl and pitch angle reduces the leaving loss. Leaving
loss increases in proportion to the square of the outlet velocity,
so a flat distribution has smaller leaving loss. The velocity
distributions in the radial direction are shown in Fig. 6. The
distribution of FV design is U-shape, while that of the
optimized design by DOE is almost flat from the hub wall to
0.8 normalized height. This change is caused by the A-shape

5 Copyright © 2011 by JSME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/ajk2011/69936/ on 04/05/2017 Terms of Use: http://www.asme.o

 distributions of s/t of the stator and the rotor blades. With an A-
15
shape distribution, the mass flow decreases near the hub and tip SWIRL
walls and increases in the middle span, so the velocity
10
distribution becomes flat.

SWIRL ANGLE[deg]
The swirl and pitch angle distributions are shown in Fig. 7.
5
The swirl angle is close to 0°everywhere in height both in the
distribution of the optimized design and the FV design, so the 0
change in the swirl angle does not affect the leaving loss much.
The velocity increases in proportion to 1/cos of the angle to the -5
axial direction, so the leaving loss does not change much when
the angle is close to 0°. On the contrary, the pitch angle is -10 FV
close to 20°near the tip wall because of the slant angle of the DOE Stag.
wall, so the smaller pitch angle near the tip wall contributes to -15
the leaving loss reduction in the DOE. With A-shape 0.0 0.2 0.4 0.6 0.8 1.0
distribution of the rotor blades, the fluid near the hub and tip NORMALIZED HEIGHT
walls moves towards the middle span and the positive pitch
angle near the tip wall becomes smaller. 30
The blade loss is slightly reduced in the DOE design. Fig.8 PITCH
FV
and Fig. 9 show the blade loss distributions and the mass flow
distributions respectively. Total to total loss is described as DOE Stag.
PITCH ANGLE [deg]
20
blade loss in Fig.8. Although the loss is increased near the hub
and tip walls in the optimized design with DOE, the blade loss
in the stage is reduced because of less mass flow near the hub 10
and the tip end walls, where the blade loss is high, and more
mass flow in the middle span, where the loss is lower than the
FV design. 0

1.2 -10
0.0 0.2 0.4 0.6 0.8 1.0
1.0 NORMALIZED HEIGHT
NORMALIZED VELOCITY

0.8 FIGURE 7. SWIRL AND PITCH ANGLE DISTRIBUTIONS

AFTER ROTOR BLADES
0.6 (UPPER: SWIRL ANGLE, LOWER: PITCH ANGLE)

0.4
FV 3.0
0.2 DOE Stag.
2.5
NORMALIZED LOSS

0.0 FV
0.0 0.2 0.4 0.6 0.8 1.0 2.0 DOE Stag.
NORMALIZED HEIGHT
1.5

FIGURE 6. VELOCITY DISTRIBUTION 1.0

0.5

0.0
0.0 0.2 0.4 0.6 0.8 1.0
NORMALIZED HEIGHT

FIGURE 8. BLADE LOSS DISTRIBUTIONS

6 Copyright © 2011 by JSME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/ajk2011/69936/ on 04/05/2017 Terms of Use: http://www.asme.o

 1.4 1.2
STATOR
NORMALIZED MASS-FLOW

1.2 1

NORMALIZED s/t
1.0 0.8
0.8
0.6
0.6
0.4
DOE Stag.
0.4
FV 0.2 ASA Stag.
0.2
DOE Stag.
0
0.0
0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.2 0.4 0.6 0.8 1.0
NORMALIZED HEIGHT NORMALIZED HEIGHT

FIGURE 9. MASS FLOW DISTRIBUTIONS

1.2
ROTOR
1.0
STAGGER ANGLE OPTIMIZATION WITH
NORMALIZED s/t

EVOLUTIONARY METHOD 0.8

Stagger angle distributions are optimized using the system.
ASA is used as an optimization method. The optimized design 0.6
by DOE is used as an initial design, and better designs are 0.4 DOE Stag.
explored.
ASA Stag.
NURBS curves are used to determine the stagger angle 0.2
distributions in the radial direction. The distributions are
changed by the control point vectors of the NURBS curves, as 0.0
0.0 0.2 0.4 0.6 0.8 1.0
explained in the previous section. Four control points are
equally placed in height and their height is kept constant. Only NORMALIZED HEIGHT
the rotation angles are changed. All the control points except
the one at the hub wall are chosen as optimization parameters,
so there are six parameters, three control points of the stator FIGURE 10. THROAT BY PITCH RATIO DISTRIBUTIONS
blades and three control points of the rotor blades. The rotation (UPPER: STATOR BLADE, LOWER:ROTOR
angle coordinates of the control points at the hub wall are kept BLADE)
0 at first, and then once a stagger angle distribution is defined, a
certain rotation angle is added to or subtracted from all the TABLE 3. NORMALIZED LOSSES
rotation angle coordinates so that the throat area keeps the
designed value. DOE ASA
The optimized s/t distributions of the stator and the rotor FV
Stag. Stag.
blades are shown in Fig. 10. Both the stator and the rotor blades Stage Loss 1.000 0.991 0.988
of the ASA design have A-shape distribution, but some changes Leaving Loss 0.525 0.518 0.520
from the DOE design can be seen. In the ASA design case, the Blade Loss 0.475 0.473 0.468
s/t of the stator blades is larger around the 0.7 normalized
height. The s/t of the rotor blades is also changed; smaller near
the hub wall and larger near the tip wall. Regarding the leaving loss, the loss of the ASA design is
The normalized losses of the optimized designs by DOE increased from the DOE design. The velocity distributions are
and ASA are shown in Tab.3. The stage loss is at a close level, shown in Fig. 11. The flat distribution of the DOE design is
but slightly reduced to 0.988. Although the leaving loss is changed to A-shape distribution. Because the ASA design has
increased from the DOE design, the reduction in the blade loss wider throat at the tip side for both the stator and the rotor
is large enough to reduce the loss in the stage. blades, the stage exit velocity is larger there, which would cause
more loss than the DOE design. Less loss in the hub region
cannot compensate the increase at the tip.
As shown in Fig. 12, the blade loss is reduced from the
DOE design. The mass flow distributions are shown in Fig. 13.

7 Copyright © 2011 by JSME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/ajk2011/69936/ on 04/05/2017 Terms of Use: http://www.asme.o

 The blade loss is smaller in the middle span and near the tip
wall, where the mass flow is increased, and the blade loss is 1.4
larger near the hub wall, where the mass flow is reduced. The

NORMALIZED MASS-FLOW
blade loss of the ASA design is reduced mainly because the 1.2
secondary loss is largely reduced near the tip wall. Although 1.0
more mass flow in the middle span and less mass flow near the
walls reduce the blade loss in a stage, the reduction has a limit 0.8
because the secondary loss greatly increases if the s/t near the
0.6
wall is too small.
0.4
DOE Stag.
0.2
1.2 ASA Stag.
0.0
NORMALIZED VELOCITY

1.0 0.0 0.2 0.4 0.6 0.8 1.0

NORMALIZED HEIGHT
0.8
FIGURE 13. MASS FLOW DISTRIBUTIONS
0.6

0.4
DOE Stag. By changing only the stagger angle distributions, the
0.2 ASA Stag. system can reach as high as the human design. However, the
system automatically explores different shapes and finds good
0.0 shapes without human efforts. It is difficult to judge whether
0.0 0.2 0.4 0.6 0.8 1.0 the blade loss is the smallest or not, because the secondary loss
NORMALIZED HEIGHT greatly increases if the s/t is too small near the walls, although
the blade loss decreases and the mass flow increases in the
FIGURE 11. VELOCITY DISTRIBUTIONS middle span. It is also difficult to judge the smallest stage loss,
the best combination of the blade loss and the leaving loss. The
system is a useful tool to find the parameters to minimize the
loss.
3.0
DOE Stag.
2.5 STAGGER ANGLE AND STACKING LINE
NORMALIZED LOSS

ASA Stag.
OPTIMIZATION
2.0
To test the ability of the system to improve a turbine stage,
1.5 stacking line distribution of the stator blades is added to the
optimization parameters. The circumferential coordinates of the
1.0 stacking line is changed and the lean shape of the stator blades
is changed. Lean shape of stator blades also affects the flow
0.5 pattern in radial direction.
A NURBS curve is used to define the stacking line
0.0
distribution. The stacking line distribution is expressed by the
0.0 0.2 0.4 0.6 0.8 1.0
circumferential displacement distribution from the radial axis.
NORMALIZED HEIGHT Four control points are used to express a NURBS curve. Four
control points are equally placed in height and the height
FIGURE 12. BLADE LOSS DISTRIBUTIONS coordinates are kept constant. All the control points except the
one at the hub wall are changed. The one at the hub wall is kept
0 and defines the base position of the stacking line. There are
nine parameters altogether, three control points to define the
stacking line distribution of the stator blades, three control
points to define the stagger angle distribution of the stator
blades and three control points to define the stagger angle
distribution of the rotor blades. The stagger angle distributions
are changed in the same process as the previous section.

8 Copyright © 2011 by JSME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/ajk2011/69936/ on 04/05/2017 Terms of Use: http://www.asme.o

 ASA is used to optimize the parameters. The optimized
design by DOE in the previous section is used as an initial 1.2
design. STATOR
1
The optimized circumferential displacement distribution

NORMALIZED s/t
and the s/t distributions are shown in Fig. 14 and Fig. 15
0.8
respectively. The circumferential displacement is normalized by
the height of the stator blade. The positive displacement in 0.6
Fig.14 means the cross-sectional profile is moved to the
direction of the rotation. The stacking line is so-called 0.4
compound lean distribution, i.e. the blade is leaned towards the DOE Stag.
pressure side at the both ends. Regarding the s/t distributions, 0.2 ASA Stag.
both the stator and the rotor blades have A-shape curves. In ASA Stag.+Stack.
particular the s/t distribution of the rotor blades is almost the 0
same as the DOE design, so the s/t near the tip wall becomes 0.0 0.2 0.4 0.6 0.8 1.0
small again. NORMALIZED HEIGHT

0.025 1.2
ASA Stag.+Stack. ROTOR
NORMALIZED DISPLACEMENT

1.0
0.020
NORMALIZED s/t

0.8
0.015
0.6

0.010 0.4 DOE Stag.

ASA Stag.
0.2
0.005 ASA Stag.+Stack.
0.0
0.000 0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.2 0.4 0.6 0.8 1.0 NORMALIZED HEIGHT
NORMALIZED HEIGHT
FIGURE 15. THROAT BY PITCH RATIO DISTRIBUTIONS
FIGURE 14. CIRCUMFERENTIAL DISPLACEMENT (UPPER: STATOR BLADE, LOWER:ROTOR BLADE)
DISTRIBUTION (NORMALIZED BY HEIGHT)

TABLE 4. NORMALIZED LOSSES

The normalized losses are shown in Tab.4 and Fig. 16. The
stage loss is reduced to 0.954 by optimizing the stagger angle DOE ASA ASA
FV
and the stacking line distributions with ASA. Stag. Stag. Stag.+Stack
The leaving loss is at a close level to the optimized designs Stage Loss 1.000 0.991 0.988 0.954
in the previous section. The velocity distributions are shown in Leaving Loss 0.525 0.518 0.520 0.519
Fig. 17. The velocity distribution of the optimized design turns Blade Loss 0.475 0.473 0.468 0.435
to be similar to that of the optimized design with DOE.

9 Copyright © 2011 by JSME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/ajk2011/69936/ on 04/05/2017 Terms of Use: http://www.asme.o

 pushes the secondary flow, which rises from the wall, to the
1 walls and reduces the growth of the secondary loss.
0.9 In the present exhaust stage, changing only the stagger
0.8
angle distributions has a limit of reduction of losses because the
secondary loss at the tip side increases if the s/t is small near the
NORMALIZED LOSS

0.7 tip wall, but the mass flow and the leaving loss increase if the
0.6 s/t is large. By controlling both the stagger angle and the
0.5 stacking line distributions, the secondary loss can be reduced
without increasing the mass flow and the leaving loss near the
0.4 tip wall.
0.3
0.2 Leaving Loss
0.1 Blade Loss 3.0
0 DOE Stag.
FV DOE ASA ASA 2.5 ASA Stag.

NORMALIZED LOSS
Stag. Stag. Stag.+Stack. ASA Stag. + Stack.
2.0

FIGURE 16. NORMALIZED LOSSES 1.5

1.0

1.2 0.5
NORMALIZED VELOCITY

1.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0
0.8 NORMALIZED HEIGHT

0.6 FIGURE 18. BLADE LOSS DISTRIBUTIONS

0.4 DOE Stag.

ASA Stag.
0.2 ASA Stag. + Stack.
1.4
NORMALIZED MASS-FLOW

0.0 1.2
0.0 0.2 0.4 0.6 0.8 1.0
NORMALIZED HEIGHT 1.0
0.8
FIGURE 17. VELOCITY DISTRIBUTIONS
0.6

The blade loss is greatly reduced in the optimized design. 0.4 DOE Stag.
The blade loss and the mass flow distributions are sown in Fig. ASA Stag.
0.2 ASA Stag. + Stack.
18 and Fig. 19 respectively. The blade loss near the hub wall is
reduced from the stagger angle optimum design with ASA, 0.0
while the blade loss near the tip wall is increased, but not as 0.0 0.2 0.4 0.6 0.8 1.0
much as the DOE design. The mass flow distribution is almost NORMALIZED HEIGHT
the same as the DOE design, so the mass flow near the tip wall
is reduced to the same level as the DOE design. The blade loss FIGURE 19. MASS FLOW DISTRIBUTIONS
in a stage is reduced because the mass flow near the tip wall is
reduced without great increase in the secondary loss.
The result shows that the system can automatically
The main cause of the small increase in the secondary loss
improve a turbine stage. The system finds better combination of
is the pitch angle distribution of the flow after the stator blades
many parameters without human efforts. It is difficult to
with compound lean. The distributions are shown in Fig. 20.
optimize many parameters by human resource. Designers
The optimized design has smaller pitch angle near the hub wall
sometimes ignore the influence of some parameters and cannot
and larger near the tip wall. This means that the direction of the
reach the highest performance. The system solves this problem.
flow is changed towards the walls. The flow towards the walls

10 Copyright © 2011 by JSME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/ajk2011/69936/ on 04/05/2017 Terms of Use: http://www.asme.o

 This can be conceptually thought but difficult to realize the
30
blade shape, which the optimization system seems to carry out.
For further improvement of a turbine stage, the authors are
25 planning to add blade profile parameters to the optimization
PITCH ANGLE[deg]

parameters.
20
ACKNOWLEDGMENTS
15 This study is supported by Tsinghua-Toshiba Energy &
Environment Research Center.
10
DOE Stag.
5 ASA Stag. REFERENCES
ASA Stag. + Stack. [1] Cravero, C., Satta, A., 2001, “A Navier-Stokes Based
0 Strategy For the Aerodynamic Optimization of a Turbine
0.0 0.2 0.4 0.6 0.8 1.0 Cascade Using a Genetic Algorithm”, ASME Paper No.
NORMALIZED HEIGHT 2001-GT-0508.
[2] Dornberger, R., Stoll, P., Büche, D., Neu, A., 2000,
FIGURE 20. PITCH ANGLE DISTRIBUTIONS AFTER “Multidisciplinary Turbomachinery Blade Design
STATOR BLADES Optimization”, AIAA Paper No. 2000-0838.
[3] Lampart, P., 2004, “Numerical Optimization of a High
Pressure Steam Turbine Stage”, J. Computational and
Applied Mechanics, 5(2), pp. 311-321.
CONCLUSIONS
[4] Senoo S., Ogata K., 2010, “Three-Dimensional-Design
An advanced aerodynamic design optimization system for
Method for Long Blades of Steam Turbines Using Fourth-
a steam turbine stage is developed.
Degree NURBS Surface”, ASME Paper No. GT2010-
In the stagger angle optimizations, the normalized stage
22312.
losses are 0.991 by human DOE and 0.988 by the system. By
[5] Harrison, S., 1992, “The Influence of Blade Lean on
changing only the stagger angle distributions, the system can
Turbine Losses”, ASME J. Turbomachinery, 114(1), pp.
reach as high as the human design, but the result shows that the
184–190.
system automatically explores different shapes and finds good
[6] Denton, J. D. and Xu, L., 1999, “The Exploitation of 3D
shapes without human efforts.
Flow in Turbomachinery Design”, VKI LS 1999-02,
By adding stacking line distribution (the lean shape) of the
Rhode-St-Genese, Belgjum.
stator blades to the optimization parameters, the normalized
[7] Arabnia M., Ghaly W., 2009, “A Strategy for Multi-Point
stage loss is reduced to 0.954. The pitch angle distribution of
Shape Optimization of Turbine Stages in Three-
the flow after the stator blades with compound lean is changed
Dimensional Flow”, ASME Paper No. GT2009-59708.
towards the both end walls, which reduces the growth of the
[8] Arabnia M., Ghaly W., 2010, “On the Use of Blades
secondary loss from the end walls. By controlling both the
Stagger and Stacking in Turbine Stage Optimization”,
stagger angle and the stacking line distributions, the secondary
ASME Paper No. GT2010-23399.
loss can be reduced without increasing the mass flow near the
[9] Sivashanmugam V., Arabnia M., Ghaly W., 2010, “Aero-
tip wall.
Structural Optimization of an Axial Turbine Stage in
The result shows that the system can automatically
Three-Dimensional Flow”, ASME Paper No. GT2010-
improve a turbine stage of high aspect ratio by optimizing the
23406.
stagger angle and the stacking line distributions. The system
[10] Chen, B., and Yuan, X., 2009, “Advanced Aerodynamic
finds better combination of many parameters without human
Optimization System for Turbomachinery”, ASME J.
efforts. It is difficult to optimize many parameters by human
Turbomach. 130, 021005.
resource. Designers sometimes ignore the influence of some
[11] Yuan, X., Tanuma, T., et al., 2010, “A CFD Approach to
parameters and cannot reach the highest performance. The
Fluid Dynamic Optimum Design of Steam Turbine Stages
system solves this problem.
with Stator and Rotor Blades”, ASME Paper No. GT2010-
Although it is applied for the stage with relatively high
22477.
aspect ratio of more than three, a best combination of stagger
angle and leaned stacking can achieve total loss reduction. It
turns to be found out that blade loss reduction without leaving
loss increase is one of the point for an exhaust stage design.

11 Copyright © 2011 by JSME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/ajk2011/69936/ on 04/05/2017 Terms of Use: http://www.asme.o