International Journal of Fluid Machinery and Systems DOI: 10.5293/IJFMS.2010.3.3.

260
Vol. 3, No. 3, July-September 2010 ISSN (Online): 1882-9554

Original Paper

Study of the Flow in Centrifugal Compressor

Cheng Xu and Ryoichi Samuel Amano

Department of Mechanical Engineering, University of Wisconsin-Milwaukee

Department of Mechanical Engineering, Milwaukee, 53201, USA, amano@uwm.edu

Abstract

Reducing the losses of the tip clearance flow is one of the keys in an unshrouded centrifugal compressor design and
development because tip clearances are large in relation to the span of the blades and also centrifugal compressors
produce a sufficiently large pressure rise in single stage. This problem is more acute for a low flow high-pressure ratio
impeller design. The large tip clearance would cause flow separations, and as a result it would drop both the efficiency
and surge margin. Thus a design of a high efficiency and wide operation range low flow coefficient centrifugal
compressor is a great challenge. This paper describes a recent development of high efficiency and wide surge margin
low flow coefficient centrifugal compressor. A viscous turbomachinery optimal design method developed by the authors
for axial flow machine was further extended and used in the centrifugal compressor design. The compressor has three
main parts: impeller, a low solidity diffuser and volute. The tip clearance is under a special consideration in this design
to allow impeller insensitiveness to the clearance. A patented three-dimensional low solidity diffuser design method is
used and applied to this design. The compressor test results demonstrated to be successful to extend the low solidity
diffusers to high-pressure ratio compressor. The compressor stage performance showed the total to static efficiency of
the compressor being about 85% and stability range over 35%. The test results are in good agreement with the design.

Keywords: centrifugal compressor, flow separation, static efficiency, turbomachinery, impeller, surge.

1. Introduction
The flow enters a compressor axially with the action of the centrifugal force and it turns in the radial direction out from the
impeller. The flow then is directed to a radial annular vaned or vaneless diffuser. The flow exit from diffuser needs a volute or
collector to deliver the flow to the next stage or send to the next components [1-5]. Unlike an axial compressor or fan [6], the
work input for centrifugal compressor is almost independent of the nature of the flow. A centrifugal compressor can be designed
with much higher De Haller number than an axial compressor can achieve. Therefore, it is possible for a centrifugal compressor to
have a much higher stage pressure ratio than axial ones. Centrifugal compressors have wide applications for a mass flow rate less
than 10 kg/s [1-5, 7].
As is well recognized, low flow coefficient centrifugal compressors have wide applications in turbo shaft aircraft engines,
petrochemical plants, and manufacturing. However, being in a low-power class, these compressors need to be inexpensive to
manufacture and operate, requiring that the compressor having a simple design with less number of parts and a smaller relative
tolerance. Moreover, the five-axial machine is now a common tool for impeller machining, but most other parts should be
fabricated by using other type of machining to reduce the manufacturing costs.
Turbomachinery industries are interested in using optimization procedures that enable to enhance compressor efficiency and
wide operating ranges. Turbo machine design normally starts with a meanline program at each individual operating point on a
map, then throughflow calculation is performed, and finally, the impeller, diffuser and volute are designed. In this study, a
recently developed turbomachinery viscous optimal method [8,9] especially for axial machines was further extended to a
centrifugal compressor design. The main focus of this study lies in a development of a flow coefficient in the order of φ = 0.145
compressor. The design requirements for this compressor development are that the compressor stage pressure ratio is 3.65 and a
preferred flow rate is about 0.75 kg/s at the design condition with a total to static efficiency larger than 84% and the stability range
SB≥30%. The compressor design employs a viscous process for achieving efficiency and stability targets. Good surge margins
were achieved without use of a variable geometry for a steady-state operation. Special attention has been paid to a tip clearance
Received July 27 2010; accepted for publication September 7 2010: Review conducted by Prof. Zhang Yangjun (Paper number O10019C)
Corresponding author: Ryoichi Samuel Amano, Professor, amano@uwm.edu

260
during the impeller design. The compressor developed in this study consists of three major parts: an impeller, low solidity diffuser,
and volute. In this study, particular attention was paid mainly on impeller and low solidity diffuser development.

2. Outline Of The Design Process

The design of modern industrial centrifugal compressors requires the aerodynamic optimization of each component. Typically,
the detailed design cycle starts with a parametric study involving the vector diagram by means of a one-dimensional mean-line
study. Normally, after the mean-line study, two-dimensional analyses are conducted, which is followed by the designs of impeller,
diffuser vane, and volute. The design parts were analyzed by employing a three-dimensional blade-to-blade flow analysis. The
volute design is based on the conservation of the momentum and conservation of mass in this study [10]. The attention of the
volute design was paid to the tongue shape [11]. After each iteration of the optimization study, the impeller or diffuser
performance was estimated. Then, the optimal design of impeller and diffuser were attained based on the design condition, overall
computational fluid dynamics (CFD) analysis for each component of the machine was conducted. The stage parameter
optimization study was performed before the final stage was designed. This paper focuses the impeller design as well as an
incorporation of a diffuser design.
In general, there are three types of design processes for impeller and diffuser blades: inverse, direct and hybrid direct-inverse
designs [8,9]. The inverse design is faster because it is based on the two-dimensional flow analysis to form the blade or airfoil
according to a pressure profile. However, the location of the profile control points is difficult to control and numbers of arcs are
generally large. This makes it difficult to use the same curve definition routine for design and manufacturing. At the same time,
direct design may require more time in the aerodynamic design process, but one can generate more accurate designs in the
manufacturing process. The hybrid direct-inverse design method is still not mature enough to use in the airfoil design. The present
study focuses on the development of the direct design process. The direct design process is to design the blade and airfoil profile
by representing the blade and airfoil as curves and to check the performance by a fast computing two-dimensional code [8,9].
After each basic impeller blade and diffuser profile is designed, a fast performance evaluation code is used to check the airfoil
performance [8,9]. Then, the impeller or diffuser is designed, a quasi-three dimensional code is developed to more accurately
check the airfoil performance in order to judge them whether it needs redesigning or not.
Impeller blade and diffuser airfoil shapes can be represented in different ways. One of them is to represent as curves in the
different parts. Because blade and airfoil shape are favored to have a smooth curve to avoid unexpected flow circulation and
separation, it requires a second-order continuity property. The Bezier curves have this property of continuity. The Bezier curves
may be represented by using control points Pi (i=0 to m). Then the curves can be defined parametrically as follows:
m

B(t) = ∑ Pb
i =0
i i ,m (t ) (1)

where B(t) is a point on the curve for the parametric value t and can be represented as

⎡ x(t ) ⎤
B(t ) = ⎢ ⎥ (2)
⎣ y (t )⎦
Pi (i=1 to m) is the control point,

⎡x ⎤
Pi = ⎢ i ⎥ (3)
⎣ yi ⎦
and the Bernstein basis function is

m!t i (1 − t ) m −i
bi , m (t ) = (4)
i!(m − i )!
Numerical optimization techniques are used in different steps for different components of aerodynamic design [12-15].
However, due to a complex three-dimensional nature of flow, Computational Fluid Dynamics (CFD) calculation is still a time
consuming process. Most of the optimizations [13] were performed mainly in impeller design. Although optimization of the single
component is very important, the overall optimization is more critical.
The optimization design method has been developed in the recent decades for a complex system and its subsystems while
observing all design requirements and constraints, objectives and goals. In the mean-line optimization, a design of experimental
method was used to optimize the machine efficiency. Recently, a Six Sigma based probabilistic design optimization formulation
[16] has been used to optimize the engineering designs. Variability is incorporated within all elements of probabilistic
optimization formulation. In the design optimization process, a deterministic optimization is applied first before the resulting
design is evaluated for quality by measuring reliability and robustness. The quality of the design is then improved by
implementing Six Sigma probabilistic optimization and the resulting tradeoff is evaluated. This method was used in the impeller
and diffuser blade design optimization.
It is known that the flow behavior inside any component of compressor stage not only determines the aerodynamic efficiency
of that component itself, but also strongly affects the flow of the downstream and upstream components. The flowchart of the
design process is shown in Fig. 1. The design procedure includes mean-line analysis and parameter optimization, impeller,
diffuser, and volute design and optimization, and CFD analysis and optimization steps.

261
 Fig. 1 Centrifugal design process.

Fig. 2. Cut away view

3. Impeller Design
A centrifugal compressor stage consists of three main components: a rotating impeller, diffusers, and a volute as shown in
Fig.2. The low flow coefficient impeller design is critical in terms of aerodynamic efficiency and operating range of the
compressor stage. Impeller aerodynamics also impacts the performance of the diffuser and volute. It is very important to achieve
an optimized configuration of the impeller. The traditional impeller design has been performed by using a one-dimensional
analysis; then, the blade and endwall geometries were selected according to the standard criteria so that the new impeller has a
scale of an exiting one. This design method could not provide the predication of the impeller flows in order to improve the design;
therefore, the flow separation in the impeller is a necessary consequence of the stipulation of the impeller geometry in most cases.
In this study, one-dimensional studies were initially performed to do parameters studies. It was shown that the rotational speed
at design condition of 65,000 rpm was necessary to achieve high compressor efficiency and to maintain overall diameter
objectives. Due to the limitations of motor and structure, the design rotational speed was set to 59,000 rpm. The blade tip to hub
diameter ratio is about 0.34 and an inlet absolute average Mach number is about 0.4. The inlet relative tip Mach number is about
0.9. The impeller was designed without splitter due to the size of the impeller and manufacturing restrictions. A high impeller
backward exit angle was used in this design to improve the efficiency through the reduction of the impeller loading and to
improve a surge margin by providing a steeper energy addition characteristic. The high impeller exit blade angles required a
coordinated aeromechanical design effort.
The detailed flow structure and impeller performance were analyzed by using the Computational Fluid Dynamic (CFD)
analysis to help a parametric study and an optimization. In order to perform the optimization, a number of geometrical parameters
were identified. The design modifications were made by varying the values of such parameters. The final impeller was designed
by using a fast three-dimensional viscous code [14].
H-type of mesh [14,15], which is suitable for obtaining low-skewed grids around long and thin blades, was employed to
compute the flow in a single blade passage. Grid sizes were chosen in order to accomplish a grid independent state to have enough

262
accuracy and fast enough for design and optimization. After several preliminary studies it was shown that the computational grid
size of 43×43×153 was grid-independent; and thus was used for all of the design simulations through this study. The typical
meridional mesh is shown in Fig. 3.
The computation was carried out based on the single blade passage during the impeller design. After each calculation was
done, a Design of Experiments (DOE) Optimization study was performed [12,13]. For every calculation, the convergence was set
when the residual RMS level with at least a fifth order of magnitude reduction with respect to its initial value. The typical
convergence history is shown in Fig. 4.

Fig. 3. Mesh in meridional plane.

(Half of the mesh points are made visible)

Fig. 4 Convergence history.

With several iterations of optimizations, the final impeller was designed. The meridional channel shape has characteristics at
both inlet and exit with a straight channel and middle has relatively large curvature. The impeller hub and shroud relative-velocity
distributions at design pressure ratio are shown in Fig. 5. The final blade has a backswept angle β=56o. The final blade angle
distributions at blade hub and shroud are shown in Fig. 6. The final blade angle was adjusted after the optimization study at design
point. The impeller blade angle was set to about 3o incidence at both tip and root sections in order to increase the choke margin.
The blade angles at both shroud and hub combined with meridional channel shape provide the reasonable blade loading as shown
in Fig. 6. The static pressure contour at mid meridional plane between two blades is shown in Fig. 7. The velocity vectors for the
flow near the blade pressure-side and suction-side are shown in Figs. 8 and 9, where one can note that the flows are well
organized inside the impeller with only the small separation region appearing near the tip of the blade. Fig. 10 shows the relative
velocity vectors along the shroud surface near the impeller exit. It is depicted in this figure that a flow separation appears near the
shroud tip region. Flow separations were induced by the large tip clearance (2% of the exit width) at the tip due to the blade-
trailing wake and pressure gradient. Several efforts were made to eliminate flow separations. The variable tip clearance was used
to reduce the separation. The reduction of the shroud loading and reduction of the blade trailing edge thickness showed the
improvement of flows near the shroud tip ranges. However, the reduction of the shroud loading increased the hub loading which
caused the flow separation at the hub. The reduction of the trailing edge thickness caused the mechanical stress and vibration
frequency problems. Due to the limitation of the development time, current design was used as the final design. Variable clearance
can reduce the affects of efficiency with the change in the axial tip clearance as shown in Fig. 10. As shown in Fig. 11, the low
clearance sensitivity allows the impeller to operate at the clearance up to 6% of clearance to exit blade height ratio without
dropping efficiency 3%. The variable tip clearance allows the impeller manufacturing having a relatively large tolerance and also
renders impeller operational work in an extremely severe condition such as a fast rapid acceleration like turbocharger and a
helicopter engine. The absolute flow angle contours at the rotor exit are shown in Fig. 12. As seen in this figure, only the flow
region close to the tip entails a high flow angle. This is because the meridional velocity is very low in this region.

263
 Fig. 5 Blade surface velocity distributions

-10 Hub
Shroud
-20

-30
β

-40

-50

-60

-70
0 0.2 0.4 0.6 0.8 1
M

Fig. 6 Blade angle distribution.

Fig. 7 Static pressure distributions at mid-plane between two blades.

264
 Fig. 8 Velocity vectors near pressure surface.

Fig. 9 Velocity vectors near suction surface.

Fig. 10 Relative velocity vectors near shroud surface.

265
 Fig. 11 Tip clearance sensitivity.

Fig. 12 Absolute flow angle contour at rotor exit.

4. Diffuser Design
A vaned diffuser has become more popular because of its high efficiency feature compared with a vaneless diffuser. However,
the problem with a conventional vaned diffuser design compressor lies in the fact that the operating range is often relatively small
compared with a vaneless diffuser [2]. In order to overcome this problem, a low solidity vaned diffuser was proposed [2,5,17].
The low solidity diffuser research showed that a low solidity diffuser could provide not only a higher efficiency but also a
reasonable operating flow range. As a remedy, different types of vane shape were tried for a low solidity diffuser. However most
of the diffuser vane shapes were either derived from a NACA airfoil or just a flat plate [2,5] and the applications of the low
solidity vaned diffuser were limited in low pressure ratio machines. In the present study, in order to meet the higher efficiency and
wide flow operating range requirements, a diffuser was designed to apply to this high-pressure ratio machine. The diffuser was
generated by optimizing sections and stacking them up by using three-dimensional analyses [8,9].
In this part of the designs, the solidity of the diffuser is σ = 0.7. Given the solidity and inlet conditions, change of the vane
number will cause to alter the length of the vane and the turning angle of the vane. The blade loading was thus changed. After the
optimization study, the number of the diffuser vanes was set to nine, which gives a high efficiency and a better flow control.
Three diffuser vane sections were designed and optimized for each section [9]. The diffuser vane airfoil sections were designed
based on the inlet condition of impeller CFD analysis and the exit static pressure condition of the design requirements. After
completion of the design of each section, the diffuser blade was stacked-up based on three-dimensional analyses. The final
diffuser was obtained after several optimization analyses. The typical mesh of the diffuser analysis at mid-span is shown in Fig. 13.
The H-mesh was used and generated by employing different blocks. The design condition at mid-span pressure contour (Fig. 14)
shows that there is a little incidence at the diffuser vane. This small incidence allows the compressor to have a wide operating
range and have a relatively high efficiency at off-design conditions. The velocity vectors, Fig. 15, show that the flow field is well
organized. There is no separation found in the design condition.

Fig. 13 Mesh at the mid-span of the diffuser.

266
 Fig. 14 Static pressure contour at mid-span.

Fig. 15 Velocity vector at mid-span.

5. Performance Prediction and Test

The presently considered compressor was designed by using a viscous design method by optimizing both efficiency and surge
margin. In this design, the compressor performance was calculated by using the mean-line calculation and combination of the
blockage and loss results from three-dimensional viscous flow analyses. The full compressor stage CFD analyses only run for
three points, i.e. near surge, design and near choke conditions. The loss and blockage between three points were obtained using a
linear relationship.
In CFD analysis, the structure hexahedral meshes are generated to define inlet pipe (541056 cells), impeller (541056 cells) and
diffuser zones (541056 cells), while the unstructured tetrahedral cells are used to define the volute (470,614 cells) as shown in Fig.
16. The impeller stability operating range (SB) was defined from choke to a surge point. The choke point is simulated by reducing
the back-pressure with an increasing strength of the shock at either impeller inducer or diffuser vane throat. Stall is an unsteady
phenomenon, which is impossible to simulate by using a steady flow analysis. However, the investigation of the stall inception by
using the steady code can provide important information of the surge margin. In this analysis, the surge point is defined when the
computation is not converged if the flow rate is reduced. The computational results are shown in Fig. 18 for the case of tip
clearance of 2%.
The designed compressor stage was built and tested in an open loop test rig (Fig. 17). Air was drawn from ambient and
discharged to ambient. The impeller was driven by an electric motor coupled with a gearbox. The test was performed at design
impeller rotational speed and two different tip clearance settings were used, i.e. 2% and 5% of blade height. The mass flow rate of
the compressor was measured at the discharge by a calibrated ASME nozzle [17]. The flow rate measurement uncertainty was
within 2.0% and the speed was measured with a one pulse per revolution signal sensed by a magnetic probe, which looks to be a
notch in the shaft. The measurement uncertainty of the speed was less than 0.01%, which has a very insignificant effect on the
stage performance measurement. The temperatures were measured by using Type E half-shielded thermocouple. The uncertainty
of the temperature measurement is about 0.25%. The measurement uncertainty for total pressure and static pressure are 0.25% and
0.15%, respectively. The inlet upstream and stage exit conditions were measured by employing five Kiel probes and five
thermocouples at upstream of three time of impeller inlet pipe and discharge. Ten wall static pressure taps were used for the
measurement of the inlet and discharge static pressure. The measurement results for the tip clearance effect are shown in Fig. 11.
It is shown that variable tip clearance has a better effect on the rate of efficiency loss with increasing the tip clearance. The
performance measurement is shown in Fig. 18. The experimental results show a good agreement with the calculations. Due to the
stall phenomena during the test when the compressor is close to the stall, the vibration level increases. For the safety purpose of
the bearing system, the test machine did not run to the machine surge. It was shown that except the flow close to choke range, the
CFD provided a good estimate of the compressor performance. For the compressor flow close to choke, CFD provided more flow
range than performance test.
It is showed that the design analyses and performance predications agree well with tests. The design procedures developed
here produce a very competitive design for both peak efficiency and operating range. It is shown that the centrifugal compressor
design procedure developed by this research provided a fairly successful design.

267
 Fig. 16 The computational mesh for compressor.

Fig. 17 Test rig of a single stage compressor

Fig. 18 Compressor performance.

5.1 Volute
Fig. 19 shows the static pressure distribution at mid span of the volute and diffuser. It can be seen that, for all the flow cases,
the static pressure distributions near the vane diffuser inlet are symmetric and static pressures near volute inlet are distorted. The
flow near choke presented the highest static pressure distortion at the inlet of the volute. The unsymmetrical pressure distributions
inside the diffuser also cause the variations of the Mach number as shown in Fig. 20. The isentropic Mach number distributions
show that, by increasing the compressor flow rate, the highest Mach number location moves downstream from vaneless diffuser to
vaned diffusers. When the flow rate approaches closer to choke, the position of the highest Mach number moves toward the
leading edge of the diffuser vanes. The diffuser vane inlet throat limited the maximum compressor flow, which causes a flow
choke.

268
 a. φ/φ0 =1.08 b. φ/φ0=0.7 c. φ/φ0=1.25

Fig. 19 Static pressure contours at mid-span.

a. φ/φ0 =1.08 c. φ/φ0=1.25

b. φ/φ0=0.7

Fig. 20 Mach number contours at mid-span.

6. Conclusions
A low flow coefficient high-pressure ratio single stage centrifugal compressor was designed by using a viscous optimization
method. This compressor is capable of a high performance. A wide operating range was achieved without resorting to variable
geometry and flow controls. The experimental results showed that the design is in good agreement with the test results. The
applications of CFD provide the reliable design. It is proven that the design system developed in this study is feasible; the design
thus provides a useful tool for a future centrifugal compressor development. The future development for this machine will be
focused on the three-dimensional optimization of the impeller, and will introduce better material in order to build the impeller
blades thinner and flow control investigation of impeller tip region.

Nomenclature
b = Bernstein basis function PRd = pressure ratio at design point
C = absolute velocity at impeller outlet Q = volumetric flow
D = diameter of impeller SB = ( Qc-Qs)/Qc
i = point number SM = (PRs-PRd)/PRd
L= chord length t = parameter
M = non-dimensional meridional curve length U = peripheral velocity
N = impeller rotation speed (rad/s) W = relative velocity
p = diffuser leading edge pitch = 2πr3/Z x, y = coordinates
P = control point Z = number of vane or impeller blade
PRs = pressure ratio at surge

269
 Subscripts
1 = impeller inlet s =surge point
2 = impeller exit o = operation point
3 = diffuser inlet p = polytropic
c = choke point r =radial direction
d = design point

Greeks
β = blade angle from radial direction σ = solidity ( = L/p)
φ = flow coefficient based on compressor inlet condition = ϕ = head coefficient =H/ u22
Q/(ND3)
η = efficiency

References
[1] Boyce, M.P, 2003 “Centrifugal compressor: a basic guide,” PennWell Corparation, Tulsa, OK, USA.
[2] Japikse, D., 1996, “Centrifugal compressor design and performance,” Concepts ETI, VT, USA.
[3] Zangeneh, M., Vogt, D. and Roduner, C., 2002, “Improving a vaned diffuser for a given centrifugal impeller by 3D inverse
design,” Presented at the ASME TURBO EXPO 2002, Amsterdam, Netherlands, June 3-6 2002. ASME paper no. GT-2002-30621.
[4] Amineni, N. K., and Engeda, A., 1997, “Pressure recovery in low solidity vaned diffusers for centrifugal compressors,” 97-GT-
472.
[5] Senoo, Y., Hayami, H. and Ueki, H., 1983, “Low solidity trandem cascade diffusers for wide flow range centrifugal blowers,”
83-GT-3.
[6] Xu, C, Amano, R. S. and E. K. Lee, 2004, “ Investigation of an axial fan—Blade stress and vibration due to aerodynamic
pressure field and centrifugal effects,” JSME International Journal, series b, vol. 47, No. 1, pp. 75-90.
[7] Kenny, D. P. 1984, “The history and future of the centrifugal compressor in aviation gas turbine,” 1st Cliff Garrett
Turbomachinery Award Lecture, Society of Automotive Engineers, SAE/SP-804/602.
[8] Xu, C. and Amano, R.S., 2009, “On the Development of Turbomachine Blade Aerodynamic Design System,” International
Journal for Computational Methods in Engineering Science & Mechanics Vol. 10, Issue 3, pp. 186-196.
[9] Xu, C and Amano, R. S., 2002, “Turbomachinery Blade Aerodynamic Design and Optimization,” GT-2002-30541, 2002.
[10] Weber C R, and Koronowski, M E, 1987, “Meanline performance prediction of volutes in Centrifugal compressors,” ASME
31st Gas Turbine Conference and Exhibit, Dusseldorf, Germany, 86-GT-216.
[11] Dong, R., Chu, S., and Katz, J., 1997, “Effect of Modification to Tongue and impeller Geometry on Unsteady Flow, Pressure
Fluctuations and Noise in a Centrifugal Pump,” ASME J. Turbomach., 119, PP. 506-515.
[12] Arora, J.S., 1998, Introduction to Optimum Design, NY, MC Graw-Hill.
[13] D. Bonaiuti, A. Arnone, M. Ermini, and L. Baldassarre, “Analysis and Optimization of transonic centrifugal Compressor
Impellers Using the Design of Experiments Technique,” G T-2002-30619.
[14] Xu, C. and Amano, R.S., “A Hybrid Numerical Procedure for Cascade Flow Analysis,” Numerical Heat Transfer, Part B,
2000, Vol. 37, No. 2, pp. 141-164.
[15] Xu, C. and Amano, R.S., “Computational Analysis of Pitch-Width Effects on the Secondary Flows Of Turbine Blades,”
Computational Mechanics, Vol. 34, No. 2, 2004, pp. 111-120.
[16] Harry, M. J., 1997, “The nature of Six Sigma quality,” Motorola Univ. Press., Shaumburg, Illinois.
[17] Xu, C. and Valentine D, 1997, “Diffuser for a centrifugal compressor,” US patent # 7581925.
[18] CFX Ltd, 2003, “CFX5, version 5.6,” UK.
[19] ASME, Performance Test Code on Compressors and Exhausters, PTC 10-97, 1997.

270