Proceedings of ASME Turbo Expo 2018

Turbomachinery Technical Conference and Exposition

GT2018
June 11-15, 2018, Oslo, Norway

GT2018-75192

DESIGN OPTIMIZATION OF SPLITTER BLADES FOR ROCKET ENGINE

TURBOPUMP

Francesco Torre Shinichi Konno

Delft University of Technology Osaka Institute of Technology
Delft, The Netherlands Osaka, Japan

Claudio Lettieri Matteo Pini Yutaka Kawata

Delft University of Technology Delft University of Technology Osaka Institute of Technology
Delft, The Netherlands Delft, The Netherlands Osaka, Japan

ABSTRACT Nomenclature
In this paper we present and validate a shape optimization c Absolute velocity
framework for the design of splitter blades that extends the
p Static pressure [Pa]
operative range under cavitation while maintaining the wetted
performance of rocket engine turbopumps. For a target P Total pressure [Pa]
turbopump application, the optimization framework allows for Circumferential velocity [m/s]
U
independent changes to the blade angle distributions across the
R Impeller tip radius
span and to the pitchwise position of the splitter blades while
preserving the thickness distributions. W Weighting vector NPR formulation
The optimization is conducted with a surrogate-based X Input sample vector NPR formulation
gradient method. The geometry is optimized at a fixed b Constant NPR formulation
cavitation number corresponding to a 5% head coefficient drop-
off, while constraints are imposed on the wet pump Greek symbols
performance. It is found that this approach, coupled with the φ Flow coefficient
optimal design points distribution provided by the Design of Head coefficient
Experiment method, reduces the computational cost of the ψ
optimization process by minimizing the number of multiphase ρ Density [kg/m3]
calculations. Cavitation number
σ
The numerical results suggest that the optimized splitter
blades successfully increase the pump operative range by 2.2% γ Slip factor
and increase the head coefficient by 5.3% compared to the Subscripts/superscripts
baseline case with non-optimized splitters. These results are
corroborated by experiments conducted in a closed-loop water 2 Tip
test facility. Several pump geometries are tested through rapid cav Cavitation condition
prototyping using additive manufacturing. The experimental
data validate the optimization framework, demonstrating a in Pump inlet
4.7% increase of pump operative range and a 7.6% increase in ind Inducer inlet tip
head coefficient. The calculations are used to gain insight in the
physical mechanisms for the performance improvement. The out Impeller outlet
analysis of the results indicates that the improved performance tip Impeller tip at outlet
is due to the optimized position and shape of the splitter blades vap Vapor
which increase the pump slip factor.
w Wet condition
ideal Ideal condition

1 Copyright © 2018 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 09/06/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 INTRODUCTION cavitation drop-off. Zhang et al. [5] investigated different
impeller geometries to assess the impact of splitter blades on
Turbopumps are used in large-scale liquid rockets to inject the pump performance. The addition of splitter blades leads to a
the cryogenic propellants into the combustor at the required reduction of the flow blockage derived from an increase in
pressures and mass flow rate. Turbopumps operate at high throat area at the leading edge (LE) of the main blade. This led
rotational speed, which in turn leads to cavitation at the inlet of to increased head coefficients and extended operational range.
the impeller. Cavitation can cause performance issues, Yang et al. [6] compared the cavitating performance of
instabilities and mechanical damage of the blades. impellers with and without splitter blades. The authors noticed
The typical cavitating performance of a turbopump is a reduction of cavitation volume in the pump with splitter
presented in the form of the so-called “knee-curve” in Figure 1. blades due to increased pressure and reduced flow velocity at
The curve shows the head coefficient of the Space Shuttle Main the LE of main blade. Other works in literature [7-13] found
Engine liquid oxygen low pressure turbopump. The data are improvements caused partly by a reduction of flow separation
obtained from Brennen [1] and the head coefficient is within the blade passage and partly to an increase in the
normalized with the non-cavitating head coefficient. During a average flow angle at the trailing edge. Yuan et al. [14] reported
cavitation ramp test the pressure of the system is reduced that centrifugal impellers with splitter blades yield reduced
slowly and continuously, decreasing the cavitation number blade loading and increased absolute circumferential velocity
! and total pressure at the outlet. This indicates that the
𝜎 = (𝑝!" − 𝑝!"# )/(0.5𝜌𝑈!"# ) while maintaining fixed flow
coefficient. As the pressure is reduced cavitation in the circumferential position of splitters plays a crucial role on
turbopump increases, causing head drop and ultimately leading pump performance.
to cavitation surge. The head drop-off point is often defined as While a large body of work exists on the effect of splitter
the point for which the head coefficient is 3-5% lower than the blades on pump performance [4,5,6,7], our observation is that a
head coefficient without cavitation, as indicated by Brennen systematic exploration of the design space has not been
[1], Onoue et al. [4] and Gulich [15]. The drop-off signifies the performed. Splitter blades are often designed by simply adding
loss of performance due to cavitation and the limit of the stable short blades with the same shape as the main blades, greatly
range of operation of the turbopump. limiting their effectiveness. Design guidelines have been
proposed by Gülich [15], Cavazzini et al. [16] and Japikse [2],
including indications on how to select the total number of
blades, the meridional position of splitters and the impeller
design. Several works in literature [7-13] indicate that the
circumferential position, the length and the blade angle
distributions are the most critical design parameters for splitter
blades. Cavazzini et al. [16] suggests that splitters impact is
constrained to a change in the incidence angle caused by a
modified blade thickness blockage of the main blades. The
benefits of splitters are often limited to certain ranges of flow
coefficients for which the presence of the splitter blades can
significantly impact the main blade incidence angle. In
particular, Cavazzini [16] observes that only at high flow rates
splitter blades reduce the incidence angle at the leading edge of
the blades thus leading to improved suction performance, while
the opposite is noticed for low flow rates.
Gradient-free and gradient-based shape optimization methods
have seen increased application to splitter blades/vanes design
Figure 1: Cavitating performance of the Space Shuttle Main Engine in recent years. Clark et al. [17] used multi-objective
low pressure oxygen pump on a knee curve. Data from Brennen [1]. optimization to the design of splitters for low aspect ratio vanes
of turbines. The analysis led to the identification of the critical
flow feature and proposed design recommendations for vane
Several strategies have been demonstrated to improve the
rows employing splitters. While several works [10,13,16,18]
performance of cavitating turbopumps. These include the
conducted the optimization of the number and length of splitter
addition of an inducer upstream of the impeller, casing
blades, shape optimization is rarely applied in rocket
treatment, leading edge shaping and splitter blades. Japkise [2]
turbopumps due to the challenges associated with the
and Furst [3] report that splitter blades enable higher mass flow
multiphase flow. As of today, there is no clear indication of how
through the impeller because of the reduction of cavity
to select the best design of splitter vanes and the physical
blockage in the turbopump. Onoue et al. [4] proved that splitter
mechanism responsible for performance improvement is not
blades can be applied in conjunction with inducers to improve
completely understood [2][13][16].
the pump cavitating performance and delay the onset of

2 Copyright © 2018 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 09/06/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 SCOPE OF PAPER TECHNICAL APPROACH
This paper aims at identifying the physical mechanism for
cavitating performance improvement of turbopumps with The baseline pump consists of a three-bladed unshrouded
splitter blades. To this end we devise an optimization inducer, a three-bladed shrouded impeller and a volute. The
framework to find the optimal splitter blade shape for a target details of the geometry are given in Table 1. Figure 3 shows a
turbopump application. The framework is validated through meridional view of the pump and an isometric view and cut of
comparison with experimental measurements. The combination the volute. The inducer is located 2 diameters upstream of the
of optimizer-driven numerical calculations, and experimental impeller to avoid any coupling between the components.
testing is used to examine the flow features that lead to the
performance improvement. The optimization expands the stable
operating range of the turbopump under cavitation while
maintaining the pump wetted performance. Shape-optimization Table 1: Overview of the main parameters of the baseline pump.
is conducted at design flow coefficient and at cavitation number
corresponding to 5% head drop-off. The optimum design is
manufactured through rapid prototyping and tested in a closed- Details of Baseline Pump
loop water test facility. Measurements indicate an extension of
the pump operating range of 4.7%. Previous works on Design flow coefficient 0.207
optimization of cavitating turbopumps has mainly focused on Blades 3
reducing the cavity volume. In this work, no appreciable effect
is found on the volume of cavities and the analysis of the flow Tip radius 66.6 mm
field shows that a reduction of the flow deviation angle is the Inlet hub to tip ratio 0.405
key mechanism leading to the improvement of performance.
The slip factor is improved mainly by removing a large region Outlet hub to tip ratio 1
of separated flow downstream of the cavitating region on the Inlet tip angle 18.3°
main blade pressure side, as shown in Figure 2. This separated
flow region is due to the destabilization of the boundary layer Outlet tip angle 14°
due to of rotating effects. Decisive parameters of the optimized Rotational speed 3600 rpm
splitter are the reduced circumferential position and the blade
angle distribution at hub. Together, they determine increased Blade thickness 3 mm
velocity and reduced flow turning between the main blade Inducer Tip gap 2%
pressure side and the splitter suction side. This suppresses the
separation region, enabling the streamlines to better follow the
main blade.

The optimization framework is developed using Ansys 17.1

[19]. Figure 4 shows a schematic representation of the
framework. The parametric CAD software Design Model takes
the geometric input parameters from the Parameters Manager,
creating the impeller geometry. The adaptive mesh generator
TurboGrid, creates the computational domain and Ansys CFX
is used for the multiphase calculations. The parameters
manager block creates a link between the outputs of the
calculations and the inputs. The combinations of input and
output from the completed calculations of each design point are
sent to the optimizer block.
The first part of the optimizer block is the Design of
Experiments tab, which selects the algorithm to generate a
combination of the input parameters. The Response Surface
optimizer generates the surface which interpolates the outputs
from the calculations. The response surface is transferred to the
actual Optimizer that identifies the optimum directly without
Figure 2: Schematic representation of relative velocity streamlines. running any extra simulations.
The optimized splitter blade shape at hub avoids flow separation on
main blade pressure side.

3 Copyright © 2018 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 09/06/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Figure 3: a) Meridional view of the inducer and impeller. b) Isometric representation of the volute used in the computations and cut-away view.

Figure 4: Block diagram of the optimization framework.

Experimental Facility: positioned 3 diameters upstream of the pump inlet, and

differential pressure transducers measuring at both inlet and 4
The experimental tests are performed in the closed-loop test diameters downstream of the outlet. Sampling time can be
facility of the Fluid Machinery Laboratory at the Osaka regulated in order to avoid excessive data dispersion especially
Institute of Technology, shown in Figure 5. The test facility in the region of cavitation breakdown.
provides steady and unsteady performance data for turbopumps The baseline and optimized pump components are
over a wide range of operating conditions. Tests are conducted manufactured using 3D printers for the ABS (Acrylonitrile
using water as medium. Water flows out of the tank into a flow butadiene styrene) polymer. During the printing process, the
conditioning section, through the test section, and through a height of the set down layers, the density of the supporting
throttling valve before returning to the tank. structure and the thermal control of the environment are
A vacuum pump is used ahead of each test run to deaerate selected in order to obtain an acceptable level of surface
the water and separate circuit allows for filtration and removal roughness and dimension tolerance. The performance of the 3D
of any contamination. The inlet line supplying the pump printed components are tested experimentally and compared
provides 11 diameters of development length between the tank with the metal counterparts, with no appreciable difference
and pump inlet. This provides for a well-characterized, uniform observed in the measured wetted and cavitating performance.
inlet flow. The water flow is measured using an electromagnetic
flowmeter. An optical encoder measures the pump rotational
speed. The data acquisition system consists of static and
dynamic pressure transducers in various locations, one

4 Copyright © 2018 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 09/06/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 0.382

Sta0c Head Coefficient

0.382

0.381

0.380
0 1 2 3 4 5 6 7 8 9
Number of elements (millions)
Figure 6: Mesh convergence study indicates 2.5 million elements are
Figure 5: Test facility at the Osaka Institute of Technology. sufficient to capture pump performance.

Numerical Methodology:
OPTIMIZATION FRAMEWORK
ANSYS CFX 17.1 is used to run steady state, single
passage Reynolds Average Navier-Stokes (RANS) calculations. The optimization is conducted changing the blade
A detailed description of the solver can be found in [19]. The circumferential position, the blade length and the blade angle
computational approach is based on a finite-volume method distribution at the hub and shroud. The design is optimized at
using an implicit incompressible formulation with a second cavitation number corresponding to 5% head drop-off and
order spatial discretization. RANS calculations are closed constraints are imposed to maintain the pump wetted
through the two-equation k-ω shear stress transport turbulence performance. The blade angle distributions at the hub and
model. The total pressure boundary condition is selected at the shroud are parametrized using Bezier curves for the camber line
inlet and the mass flow rate is defined at the outlet. Multiphase as indicated in Figure 7. The blade thickness distribution is
flow simulations are conducted with a Eulerian formulation fixed and selected based on best practices for radial
using a homogenous mixture of water and vapor. The mass turbomachinery design. Six parameters are used: five control
transfer rate due to cavitation is evaluated with the Rayleigh- points define the Bezier curves, of which two control points are
Plesset equation [19]. at the hub and three control points at the shroud. The additional
The computational domain is solved in the relative frame and control point at the shroud defines the position of the leading
comprises three regions: inlet, impeller and outlet/volute. The edge, thus the blade length. One parameter is used to adjust the
impeller mesh is constructed with the adaptive mesh generator circumferential position of the splitter with respect to the main
TurboGrid, allowing for changes of the geometry. The upstream blades as indicated in Figure 8.
influence of the volute on the impeller is carefully assessed
through comparison of the results of calculations with the β
volute
and those conducted without the volute, using simplified outlet.
The computations show no appreciable impact of the volute on
the impeller performance, suggesting that there is no
appreciable effect of the volute on the flow behavior in the
region affected by the splitter blades. The optimizer-based
calculations are therefore conducted without the volute to %M
0 50 100 =
reduce the computational cost of the simulations. hub
The maximum value of the y+ is 30 in all the calculations and shroud
limited to a small area at the impeller leading edge, while on
average y+ is about 10. A mesh convergence study has been Figure 7: Illustration of the Bezier curves parametrization. The blade
conducted to ensure mesh-independent results for the angle distribution is parametrized with three control points of the
Bezier curves.
computations. Figure 6 shows the static head coefficient versus
mesh size. The variation of static head coefficient is below 1%
for the meshes with number of elements above 2 million,
indicating mesh convergence.

5 Copyright © 2018 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 09/06/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 The accuracy of the refined response surface at the optimum
point is of 0.13% for the wet head coefficient and 0.12% for the
𝑃=0 cavitating head coefficient, corroborating the response surface
predictions. The Nonlinear Programming by Quadratic
Lagrangian (NLPQL) algorithm is applied to find the global
optimum and the best candidate is simulated to confirm the
𝑃 = 0.5 improvement.

𝑀′
𝑃=1

𝜃
Figure 8: The circumferential or pitch position parameter determines
the proximity of the splitter to the suction side (p=1) or to the pressure
side (p=0) of the main blade.

To create the Design of Experiments, the Optimal Space-Filling

Design Method is chosen. This is a version of a Latin-
Hypercube Sampling (LHS) method where iterations maximize
the distance between sample points. Given n parameters and s
design points, the LHS algorithm divides the range in 𝑠 !
“intervals. The points are randomly generated inside a section
so that for every one-dimensional projection of the s samples,
there is only one sample in each box [20]. Upper and lower
bounds are also imposed in the Design of Experiments tab
based on geometry and meshing limitations. The limits of the
design space are carefully assessed through computations to
confirm that those represent unsuitable design configurations.
This indicates that the limitations imposed by geometry and
mesh do not influence the optimization process. The Non-
Figure 9: a) Visualization example of the response surface for two
Parametric Regression (NPR) model is selected to construct the input parameters. The squares represent the computed output of each
multi-dimensional response surface because of its capability to design point. b) Response surface accuracy assessment. Values of the
handle non-linearity between input and output parameters. The head coefficient predicted by the surrogate model are compared to
NPR method is a variant of the well-known support vector those calculated from CFD.
machine algorithm commonly used for data classification. It
determines the function of the surface, using the following
formulation [19]:
Optimized Design Assessment
𝑌 =< 𝑊, 𝑋 > +𝑏
where W represents a weighting vector and X the input sample The optimal shape of the splitter blades is obtained after testing
vector from the Design of Experiments. The equation is 25 design points. The percentage variations of the parameters
determined as a minimization problem of the vector W such as with respect to baseline for the optimum splitters are presented
all or most of the sample points lie within a tolerance area. The in Table 2. The baseline geometry has been designed through
response surface for only two of the six parameters is shown as coupled experimental and numerical analysis as indicated in
an example in Figure 9. Onoue et al. [4].
The Screening method is used to find the global optima on the Figure 10 shows the wetted pump performance of baseline and
response surface. The candidate points are then run to compare optimized designs. Numerical calculations show about 18%
the predicted outputs with the computed results. The response larger head coefficient than the experiment. The overestimation
surface predictions are retained when the relative discrepancy is likely due to the fact that the calculations do not include the
with the calculated output falls below 1%. Additional design leakage path and therefore do not include leakage and windage
points are simulated and then added to the surrogate model if loss. Nevertheless, the difference in performance between the
the required accuracy of the response surface is not satisfied.

6 Copyright © 2018 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 09/06/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 experimental and numerical data of both baseline and the Figure 12 illustrates a visual representation of the splitters
optimized geometries is well captured in the calculations. The blades profiles at hub and shroud before and after the
results show a consistent delta between experimental and optimization. The blades are shifted towards the main blade
numerical data for both cases. The trends reveal that the pressure side and the shroud profiles in Figure 12b feature a
numerical overestimation is a constant off-set along the shorter chord.
considered flow coefficients regime, with only a minor increase
of ∆𝜓 = 0.02 for the optimized case around 𝜑 ⁄ 𝜑!"#$%& = 1.4.

Table 2: Parameters variation to obtain the optimal shape

Variation w.r.t.
Parameter
baseline
Circumferential position -5.1%

Hub
Control point 1
+7.68%
(midpoint)
Control point 2 (TE) +65.8%

Shroud
Figure 11: Cavitating performance comparison between experiments
Control point 1 (LE) -6% and computations. Experimental measurements indicate a larger
improvement of operating range than computations.
Control point 2
+14%
(midpoint) 1.5

Control point 3 (TE) +35% Baseline hub

1.2 Op0mized hub
y/ymin,baseline

0.9

0.6

0.3
0.8 2.0 3.2 4.4 5.6
x/xmin,baseline
a)
1.2
Baseline Shroud
1.0 Op0mized Shroud
y/ymin,baseline

0.8

0.6

Figure 10: Wet performance comparison between experiments and 0.4

computations. A constant offset is present between the baseline and
optimized design, validating the optimization framework. 0.2
0.8 1.4 2.0 2.6 3.2 3.8 4.4
x/xmin,baseline
Figure 11 shows the measured and computed pump cavitating
performance. The calculations capture the head drop-off point b)
but underestimate the performance improvement. Nevertheless, Figure 12: Comparison of the hub and shroud profiles of the baseline
the experimental measurements indicate an improvement of the and optimized splitter blades.
pump operating range under cavitation of 4.7 %.

7 Copyright © 2018 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 09/06/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 The change of trailing edge angle near the hub of the splitter
blades has negligible impact on the work coefficient. Figure 13
compares the numerical cavitating performance of the baseline
1.1
and the optimized pumps. The optimized pump yields a 5.3%
increase in head rise coefficient at the optimized cavitation
number which eventually leads to a 2.2% reduced drop-off
cavitation number or, alternatively, to a 2.2% extension of the

PTOT/PTOT,max
operative range with cavitation. 0.8

0.5

Splitter LE
0.2
Main Blade LE
Streamwise Loca0on TE

Figure 14: Streamwise distribution of circumferential and spanwise

averaged total pressure. The optimized impeller yields improved
performance compared to the baseline.
Figure 13: Simulated cavitating performance, comparison between 1.1
optimized and baseline pumps. The head coefficient increase (blue
arrow) at the optimized cavitation number generates an extension of Baseline Hub
the operative range (green arrow).
Op0mized Hub
0.8
PTOT/PTOT,max

CHARACTERIZATION OF THE CAVITATING

PERFORMANCE GAIN

The analysis of the flow field reveals that the circumferential 0.5
position and the blade angle distributions of the splitter have no
beneficial effect on main blade cavitation. The cavitation
volume of the optimized impeller is nearly identical to the Splitter LE
baseline design. The optimized splitter shape removes the 0.2
separation in the hub-region on the main blade pressure side. Main Blade LE
Streamwise Loca0on TE
Flow separation downstream of the cavitating region is
enhanced by the increased pressure gradient due to the cavity Figure 15: Total pressure distribution along a streamline close to hub.
collapse and reduced blockage. The trend is similar to the streamwise averaged of Figure 14.
The streamwise distribution of total pressure is shown in Figure
14. Total pressure losses are larger for the optimized design in 1.1
the first part of the passage and up to about 60% of streamwise
Baseline Shroud
coordinate. However, the optimized splitter blades reduce the
losses due to separation leading to a net increase in total
0.8 Op0mized Shroud
pressure of 5 % at the exit of the component. Figure 15 presents
PTOT/PTOT,max

the total pressure distribution along a streamline placed close to

the hub between main blade pressure side and splitter blade
suction side, while Figure 16 compares the total pressure along 0.5
a streamline at the shroud. The performance gain is 1% higher
near the hub, indicating that the mechanism leading to the
performance improvement is in a region close to hub. Splitter LE
0.2
Main Blade LE Streamwise Loca0on TE

Figure 16: Total pressure distribution along a streamline close to

shroud. The optimized impeller yields performance similar to the
baseline between 50% and 80% of the streamwise coordinate.

8 Copyright © 2018 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 09/06/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 The analysis of the flow field reveals that the optimized splitter
blades reduce the flow turning near the main blade pressure
side and the splitter blade suction side. Flow separation is found
in this region caused by the destabilizing effect of the Coriolis
acceleration on the low momentum region of the boundary
layer. This induces secondary flow as illustrated in Figures 17
and 18. In the Figures the shroud is not shown for clarity. The
optimized splitters are closer to the main blade pressure side
reducing the throat area and increasing the meridional velocity
component. The increased momentum counteracts the adverse
pressure gradient and the boundary layer destabilization due to
the Coriolis acceleration, leading to the suppression of the
separation region. Figures 19 and 20 show velocity streamlines
at midspan and 20% span to highlight the difference between
baseline and optimized case on the flow path in the proximity
of the main blade trailing edge.

Figure 19: Streamlines of relative velocity at midspan. Between main

blade pressure and splitter suction side, the reduction in flow turning is
appreciable.

Figure 17: View from the impeller inlet of the baseline case.
Streamlines departing from the leading edge separate from the wall
after few fractions of the main blade meridional length.

Figure 20: Streamlines of relative velocity at 20% span. Separation on

the pressure side of the main blade is caused by Coriolis induced
secondary flow.

The key contribution of the splitters is the reduction of the

Figure 18: View from the impeller inlet of the optimized case. absolute value of the slip factor γ, defined as 𝑐!!!"#$% − 𝑐!! =
Streamlines departing from the leading edge, smoothly follow the 1 − 𝛾 𝑢! according to [15], where 𝑐!!!"#$% is the ideal
main blade pressure surface. circumferential component of the absolute velocity at the
impeller trailing edge. This means that the flow follows more
closely the main blade near the trailing edge in the optimized
impeller. Figure 21 shows the spanwise distribution of the
circumferential velocity component at the impeller trailing

9 Copyright © 2018 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 09/06/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 edge. The plots are taken at r/R=1.02. The circumferential CONCLUSIONS
velocity is about 10% larger in the optimized case. This paper presents a shape optimization framework for the
design of splitter blades that improves cavitating performance
2*11
while maintaining the wetted performance of a target pump
1*64
1*69

application. The main outcomes can be summarized as follows:
1*91

1. An effective framework for the optimization of cavitating

'

!'

,

performance has been devised and validated against

1*71
experiments. The framework used the cavitating condition


,

at 5% drop-off to obtain the optimization of pump
cavitating performance, minimizing the number of
1*51
!'

,


$

parameters.


,


1*31
$
2. The optimized splitter blades design yield 4.7% extension
of the operative range under cavitation while maintaining
the head-capacity curve within 2% of the baseline
1*11
geometry.
1*39
1*44
1*49
1*54
1*59
1*64
1*69
1*74
1*79
1*84
1*89

#3+3

Figure 21:
Spanwise distribution of the averaged circumferential 3. The optimized splitter geometry removes the flow
component of the absolute velocity at the impeller trailing edge. separation downstream of the cavitating region, leading to
reduced flow slip at the trailing edge of the main blade.
The increased blade outlet angles for the optimal splitter shape,
presented in Figure 12, require an additional investigation about
the contributions to the performance improvement. REFERENCES
By carrying an analysis of the velocity triangles at different
span locations in an ideal no-slip condition, it emerges that the
blade angle modification contributes less than 1% to the [1] Brennen., C., “Hydrodynamics of Pumps”, Concepts ETI Inc.
increase of the circumferential component in the optimized case and Oxford University Press, Oxford, 1994, pp. 123-163.
as indicated to Figure 22. Therefore, the main contributor to the [2] D. Japikse, W. M. Marcher, R.B. Furst. “Centrifugal Pump
pump performance improvement is the reduction of the slip
factor. Design and Performance”. White River Junction, VT: Concepts
ETI, 1997
[3] R.B. Furst. “Liquid rocket engine centrifugal flow
turbopumps”. NASA SP-8109, NTIS N74-28961, 1973
[4] J. Onoue, S.Konno, A. Ueda, K. Morita, Y. Kinoshita, M.
Hayakawa, Y. Kawata. “Improvement of two phase
performance of the centrifugal pump”. AICFM The 13th Asian
International Conference on Fluid Machinery, Tokyo 7th-10th
September 2015
[5] Zhang Y.L., Yuan S.Q., Zhang J.F., Feng Y.N., and Lu
J.X. “Numerical investigation of the effects of splitterblades
on the cavitation performance of a centrifugal pump”. IOP
Conference Series: Earth and Environmental Science v22,
2014
[6] Yang W., Xiao R., Wang F., and Wu Y. “Influence of
Splitter Blades on the Cavitation Performance of a Double

Suction Centrifugal Pump”. Advances in Mechanical
Figure 22:
Comparison of the two contributions to the increase of cu2.
The outlet blade angle modification adds only a minor contribution. Engineering V2014, 2014
[7] Gölcü M. and Pancar Y. “Investigation of performance
characteristics in a pump impeller with low blade discharge

10 Copyright © 2018 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 09/06/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 angle”. World Pumps2005; 2005(468): 32–40 Modern Design of Experiments Methods for Computational
[8] Gölcü M., Pancar Y., and Sekmen Y. “Energy saving in a Simulations (Invited)". 41st Aerospace Sciences Meeting and
deep well pump with splitter blade”. Energy Convers Manage, Exhibit, Aerospace Sciences Meeting.s
2005; 47(5): 638–651
[9] Gölcü M., Usta N. and Pancar Y. “Effects of splitter
blades on deep well pump performance”. J Energy
Resour Technol-ASME2007; 129(3): 169–176
[10] B.L. Cui, Z.C. Zhu, J.C. Zhang, Y. Chen. “Flow
simulation and experimental study of low-specific-speed high-
speed complex centrifugal impellers”. Chinese Journal of
Chemical Engineering,vol.14,no.4,pp.435–441,2006
[11] Cui B., Lin Y. and Jin Y. “Numerical simulation of flow
in centrifugal pump with complex impeller”. J Therm Sci
2011; 20(1): 47–52
[12] Chen H., Liu W., Jian W., et al. “Impellers of low specific
speed centrifugal pump based on the draughting technology”.
25th IAHR Symposium on Hydraulic Machinery and Systems,
Timisoara, Romania, 20–24 September 2010, ID 012018
[13] G. Kergourlay, M. Younsi, F. Bakir, R. Rey. “Influence
of splitter blades on the flow field of a centrifugal pump: Test-
analysis comparison”. Int J Rot Mach 2007; 2007:ID 85034
[14] Yuan S., Zhang J., Tang Y., Yuan J., Fu Y. “Research on
the Design Method of the Centrifugal Pump With Splitter
Blades”. ASME. Fluids Engineering Division Summer
Meeting, Volume 1: Symposia, Parts A, B and C ():107-120.
doi:10.1115/FEDSM2009-78101
[15] J.F. Gülich. “Centrifugal Pumps”. Springer-Verlag Berlin
Heidelberg 2008, 2010. Second Edition
[16] G. Cavazzini, G. Pavesi, A. Santolin, G. Ardizzon, R.
Lorenzi. “Using splitter blades to improve suction
performance of centrifugal impeller pumps”. Proc IMechE
Part A: J Power and Energy 2015, Vol. 229(3) 309–32
[17] Clark C., Pullan G., Curtis E., Goenaga F. “Secondary
Flow Control in Low Aspect Ratio Vanes Using Splitters”.
ASME. Turbo Expo: Power for Land, Sea, and Air, Volume
2B: Turbomachinery ():V02BT38A023. doi:10.1115/GT2016-
56625.
[18] M. Gölcü. “Artificial neural network based modeling of
performance characteristics of deep well pumps with splitter
blade”. Energy Conversion and Management, vol. 47, no. 18-
19, pp. 3333–3343, 2006
[19] Ansys Academic Research, Release 17.1, Theory
Manual, ANSYS, Inc, 2011, Canonsburg, PA
[20] Giunta A., Wojtkiewicz S., and Eldred M. . "Overview of

11 Copyright © 2018 ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 09/06/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use